Gesammelte Abhandlungen IV
 3662442892, 9783662442890

  • Commentary
  • No attempt at file size reduction
  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

NUNC

COGNOSCO

é—3sP

TRENT

EX PARTE

UNIVERSITY LIBRARY

Digitized by the Internet Archive in 2019 with funding from Kahle/Austin Foundation

https://archive.org/details/gesammelteabhand0000wey!

HERMANN GESAMMELTE

WEYL

ABHANDLUNGEN

BAND

IV

Flerausgegeben von

K. Chandrasekharan

BERLIN

SPRINGER-VERLAG -: HEIDELBERG: NEW YORK

1968

Alle Rechte vorbehalten. Kein Teil dieses Buches darf ohne schriftliche Genchmigung des Springer-Verlages ibersetzt oder in irgendeiner Form vervielfaltigt werden. © by Springer-Verlag Berlin - Heidelberg 1968 Library of Congress Catalog Card Number 68-19815 Printed in Germany Titel-Nr. 1488

Inhaltsverzeichnis Band IV 122,

On the use of indeterminates in the theory of the orthogonal and symplecticigcOups igre) wre -=: aces Mees oe Neier, hte ome SY! 123, Concerning the differential equations of some boundary layer problems 124. Concerning the differential equations of some boundary layer problems. II 125; On the differential equations of the simplest boundary layer problems. . . 126. Theory of reduction for arithmetical equivalence. II... ... 0... 1k Onpeometty of nunibers jr .ah ens. oetau Meme mene yee 128. Elementary note on prime number problems of Vinogradoff’s type (R. D.

a) Ayes and EL W/ ny)

120; 130. 1315 132, 1935

Sie

ieee

ele

sees

wesc

ead ey

eo

ena

On the theory of analytic curves (H. Weyi and J. Weyt) ....... OmlHodpes theory ofsharmonicanteprals |.) .90 529s jt) ee ba os

Obituaty: DAvip Hirperr 1862—1943,

Davin Hirserr and his mathematical Concerning a classical problem in the differential equations 200i meat wm 134, Comparison of a degenerate fess of Fe alristoleme yee en ae rey y

2.

4

work . theory of 20 ce) Einstein’s Orleans ie

a

ee

2

1 9 15

18

46 d5

97 111 115

121

. 6 2 6 6. ss 130 singular points of ordinary i ee Boe ole a es 173 with Birkhoff’s theory of ABR” A e ec een 213

135, How far can one get with a linear field theory of gravitation in flat spaceRCH rare os caer bales Se cent ears ci sys cat rneewen ec oY 218 136. Fundamental domains for lattice groups in division algebras. I, II. . . . 232 137. Baconmm (WOLEGANG PAULI) (ees csrotr su. Wn er sk Ste sees 265 138. Mathematics and logic. A brief survey serving as a preface to a review of “The Philosophy of BERTRAND RusseLE”

139; 140. 141. 142. 143.

Comment ona paper by LEVINSON. A remark on A remark on Wissenschaft Elementary

2

7

7.

4.

55

5)

5 ss

2

2 ee

268

280

the coupling of gravitation and electron (left out, see p. 285) the coupling of gravitation and electron. ........286 als symbolische Konstruktion des Menschen... ... . 289 algebraic treatment of the quantum mechanical symmetry

elim fF Goon e aOR Gass co Goo 7 ao too 6 U8 346 144. Supplementary note (anschliefiend an: S. Minakshisundaram: A generalization ot Epstein zeta fuactions)§.y.0-

145, 146. 147. 148.

= y-meuen

fess

lle

oy:

360

Almost periodic invariant vector sets in a metric vector space... . « 362 Inequalities between the two kinds of eigenvalues of a linear transformation 390 Relativity theory as a stimulus in mathematical research...

Shock waves it atbittaty, Huis

meen

eeu

eet)

lite

as

....

ele

111008

394

401

149. 150. 1510 152. 153: 154. 155; 156. 1570 158.

ASS, 160. 161. 162. 163.

50 Jahre Relativitatstheorie Ramifications, old and new, of the eigenvalue problem Elementary proof of a minimax theorem due to VON NEUMANN. . . A half-century of mathematics... ....... SO Hab Sa a 5% Radiation capacity . . . . Kapazitaét von Strahlungsfeldern . . . ....... Die natiirlichen Randwertaufgaben im AuBenraum fir Steahlungsfelder beliebiger Dimension und beliebigen Ranges. Uber den Symbolismus der Mathematik und mathematischen Phy sik Universities andScience in Germany =. s) 2) 2 see eee A simple example for the legitimate passage from complex numbers to numbers of an arbitrary field Uber die kombinatorische und kontinuumsmaBige Definition der Uberschneidungszahl zweier geschlossener Kurven auf einer Fliche . Bauer on Theory of Groups Gartan on Groups and) Diserential'Geometry 292 32 2 ee Courant and Hizpert on Partial Differential Equations. . . .... . Review: The philosophy of BERTRAND RussELL. . . . . 2...

Verschiedenes 164. Address of the President of the Fields Medal Committee 1954 . . . . . 165. Address on the Unity of knowledge delivered at the Bicentennial Con-

ference of Columbia, Watversityaeaes

See) sent

eee

166. Erkenntnis und Besinnung 167. Riickblick auf Ziirich aus dem Jahted 93022 ae see) oe HERMANN WEYL (1885—1955) par C. CHEVALLEY et A. Wer, Vollstandige Liste aller Titel

erene eee Bri tats

122. On the use of indeterminates in the theory of the orthogonal and symplectic groups American Journal of Mathematics 63, 777—784 (1941) 1.

These lines contain a supplement, concerning the use of indeterminates, to my book, The Classical Groups, Princeton (1939), which I cite in

the following as CG.

The general infinitesimal operation of the orthogonal

group,

(1)

de — on,

is described by a skew-symmetric matrix S = | six | of which the n(n—1)/2

elements si, (i

maintain

CMe. i

Turormm combinations

degree nf of the para-

II.

The

3rApCp

with

linear closure of (10), t.¢., the set Gy of all linear numerical

coefficients

rp, is an

algebra

containing

oo =1.

Equation

the unit matrix. The

unit matrix is the coefficient Cy of the monomial

(2) entails

B—s”

ps

B89

m (Fe5) —™ (gee) ™ (SFr)

or, because of (8),

A

= Cror = DY CpCqop'oq”. Da

After the substitution of y;(s’,s”) ¢r(s’, 8”) divided by Av’. Hence

(11)

for sj; each o, turns into a polynomial

ACUTE. S) OpCgap'ag” = > Cror(s’, 8”). Dd

3

We now apply the following trivial algebraic lemma:

6

of some variables 7,,° °°, 21,

be a given polynomial

¢=1-+--~

Let

with the constant term 1. Then the coefficients of an arbitrary polynomial F of degree m may be linearly expressed by the coefficients of ¢F' = G.

Arranging the terms of F in ascending lexicographic order one obtains coefficients

unknown

the

for

equations

linear

recursive

nomial

m

enter).

# =F,+F,+---.

(1—.)+,

Putting

of degree

terms

by Fy the

Denote

with

the

(only terms of @ of a degree

coefficients b of G as the known right members not exceeding

a of F,

and

¢—=1—o

using

» in any

poly-

the power

series

one finds more explicitly

R=

(#20;

m2lwZl-

Cioanae seta

tet:

Sm).

If ¢ has integral coefficients, then the coefficients of the linear combinations expressing the a in terms of the 6 are likewise integers. This lemma is of immediate application to the equation (11) in which A ton are monomials no two begins with the constant term 1. The products op’oq” of which are equal, and thus we arrive at equations C,0q = X y"naCr

with rational integers y%pg. They prove our statement. The same equations hold for the matrices Cp‘), and thus the linear combinations 3A,C,‘‘) form

an algebra ©),

The stage for the whole

drama

is the field x of rational

numbers (or, as the arithmetician might care to observe, the even narrower ring of rational integers). 4.

We go on operating in x.

Turorem

III.

The next step is

The algebra ©; in x is fully reducible.

Suppose that a subspace independent vectors

{a}

of P;, is spanned by a number of linearly

@ = (41y,° * +, Gyn),

(N =nt).

The orthogonal subspace {ay} is spanned by a complete set of linearly inde=a, of the simultaneous equations pendent solutions

N

Dd anzi =

é=1

0.

(The ranges of the two indices A and p are disjunct.)

In x the two subspaces

have no vector in common except 0 and hence span the whole space; or the matrix U whose columns are the a, and dy is a non-singular square matrix. By construction the symmetric matrix

U*U

decomposes into a (A, A)- and a

7

(u,»)-part

whereas

the

rectangles

(A,)

and

(,A)

are

empty.

We

now

This circumstance

assume the first subspace {a,} to be invariant under ©;. is expressed by an equation

R(s)-U =U -Q(s)

where Q(s) is reduced in the sense that its (A, »)-part is empty.

We propose

to prove that its (4, A)-part is also empty.

To that end, multiply by U* to the left:

U*R(s)U = (U*U)Q(s). We then see that the (A, )-part of U*R(s)U is empty,

(12)

[U*R(s) Tam =,

and that it is sufficient to establish the same fact for the

(y,A)-part of the

same matrix.

If § is an S-matrix, so is its transpose S*. involutorial substitution of the parameters

(13)

(4B) > s(Bx),

changes the indeterminate

s(%B)

The following simple linear

s;,

>s(aB),

s(a/B) > s(a6’),

S = L(s) into 8*, hence A, (9), into A*, ;A

(11;A)*, and since | E+ S| =|#+ S*|, also R(s) into R¥(s) back into R(s). Taking the transpose of the equation (12),

into

and R*(s)

[U*R*(s)T Jin.» = 0, and then carrying out the substitution (13), one arrives at the desired result

[U*R(s)T]cu,x) = 9. (13)

induces a simple involutorial permutation among

op,op—> op.

the monomials

Our argument shows that 2O*y op = BCp ope = ACps ps

C%y =

Cys,

Thus the algebra ©; is a set € of matrices C which coincides with the set ©* of its transposed

proposition:

elements

C*.

The

method

employed

yields

this

general

For any set © of matrices in a real field k with the property

€* = € reduction results in full reduction if projection modulo the invariant

subspace is carried out as projection upon the orthogonal subspace.

This is

in line with older investigations by E. Fischer who studied groups © of linear transformations in a real field & enjoying the property ©* — € and showed that

their invariants depending on a variable form have a finite integrity basis.?

Once in possession of our three theorems we put the same machinery into

1 Journal

fiir reine u. ang. Mathematik,

vol.

140

(1911), pp.

48-81.

8

We still operate in the field x.

play as employed on pp. 141-142 of CG.

An

(9) is necessarily an infini-

f(x, y,°°*) which is invariant under the generic

tesimal invariant. Apply the hypothesis to tS instead of S and take merely the first power of the parameter ¢ into account. Consequently Theorem I shows that the algebra described as 2‘) on p. 174 is the commutator algebra

of the commutator algebra of I‘ [(#—S)/(Z-+8)]

(Theorem

is an algebra containing the unit matrix

©

ducible (Theorem III), the algebra ‘7

or of ©.

II)

is not wider than ©‘?

Hence as

and fully re(R. Brauer’s

principle). In view of the definition of 2‘ this fact remains true in any field of characteristic zero, Any polynomial (A) of degree f depending on the ai proceeds from a linear form

matrix A = | ai | with variable components

y(A) of an arbitrary bisymmetric A‘ by the substitution A‘? —II'? (A). If (A) is annulled by the substitution (9) we must have y(Op‘?) =0, and therefore y(A‘)) vanishes for all matrices A‘) of 9{'. Thus results Theorem In restating it I propose the following terminology

(6.3. B) on p. 174 of CG.

for ideals of polynomials $(21,---,a:). Elements ¢1,-- -,¢m of a given ideal 3, whose degrees are fi,* - +, fm respectively, are said to constitute a form basis of % if any element ¢ of § of degree f may be obtained in the form & =hidi

where

h; is a polynomial

vanishing

of

of degree

+: + > + hndm

f—f;

f—f; 0 and

= 0 plays an important part:

w"+2ww"+ 2A (k?—w'?)=0

for

220;

|

w(0) = w'(0) =0,

for

z->0o.

|

w'(z) >k

(A)

We consider J as a given constant, but # asa variable parameter. A mathematically satisfactory proof of its solvability has never been given, although various numerical devices, including Busu’s differential analyzer, have been set at work on it. We shall here give a complete solution of the problem), first for the

two special values A = 0 and 4 = 1/2 by a process of alternating approxima-

tions, rapidly

converging

(§§ 2, 4, 5), and

and

then approach

thus well suited for numerical the general

case

computations

(§§ 6, 7) by the method

of

fixed points of transformations in a functional space,—which is considerably less amenable to calculation. In between (§ 3) the first method will be applied to certain boundary-value problems closely related to (A,). There are available two hydrodynamic interpretations of (A,). Consider first the steady flow of an incompressible viscous fluid of constant density @ and kinematic viscosity ¢? filling the half z > 0 of an m-dimensional Euclidean space with the Cartesian coordinates *,, ..., ¥, and the cylindrical coordinates

PS (Ghee

ee

ye)

re

If cylindrical symmetry prevails and hence the radial (7) and vertical (z) com-

1) See the author’s preliminary notes in Proc. Nat. Acad. Sci. 27, 578-583 (1941), and 28, 100-102 (1942).

19

ponents w, v of velocity as well as the pressure 9p depend on r, z only, then the following differential equations obtain for z > 0:

Fe

aye ou

Ou

dv

dv

Re th

op

ge

te

+o

et (du

m—2

7

u),

(1)

op

Pay ae au

where the Laplace operator A is defined by Ap=

ra

“Oy!

aes

tiar

eo

fay culaee

These equations are to be combined with the boundary conditions u+>0,

v+>0

for

z>0.

For an ideal fluid, e = 0, we have this simple solution:

Uy (7, 2) = Pe 7!

(7,2) = — 2kz,

L

2

Us

bo = const — 5 (uy + v2) = const — 2k’ ta =F +

ath

arising from the harmonic velocity potential

oa

aie

ey 8

Pe)

Se)

(a

and involving an arbitrary positive constant k. As is necessary, the vertical (though not the radial) component velocity vanishes along the boundary z = 0. The Navier-Stokes equations (1) for the viscous fluid possess a solution of the form mT

1F'(2),

v=—2F(2),

p=const — 2n{ (—* 7) + L(a)|

which approaches the solution w#9, v9, Pp for z > oo. The first equation (1) is

identically satisfied, the second and third yield

etF"" + 2FF+" and

kL’

—*_ (ht

= 2h Fy

F2) =0

et

(2)

respectively. Setting F(ez) = e+ w(z) we obtain the equations (A,) with 2 = 1/(m—1) from which the viscosity constant has disappeared, so that w is independent of e. Equation (2) in integrated form gives

A + L(e 2) = e%{w'(2) + w%(2)}.

20

Our solution describes approximately the flow of a viscous fluid around an obstacle with a blunt nose in the neighborhood of the forward stagnation point. A=1 and 1/2, have been The cases of physical interest, m= 2 and 3, i.e. treated by Hirmenz and Homann respectively?). Let the subscript « in u,, v,, p, indicate dependence of our flow on the viscosity constant «. Certainly u,, v,, Pe tend to up, v9, Po with e > 0 in the

region z > 0, but the convergence cannot be uniform at the boundary because

the viscous fluid adheres, the ideal glides along the wall. Hence we have the phenomenon of a boundary layer of thickness ~ ¢ in which the velocity rises from 0 at the surface to the external value

4 = u(r, 0) = Indeed

u,(7, €2) ,

+ +0,(r,€2),

,(7, €2)

(3)

tend with ¢ > 0 to the values

U(r, 2) =

2

rw'(z),

m—1

V(r, z) = —2 (2),

P(r) = po(r, 0) = const — 2 (Get): [As a matter of fact, the first two quantities (3) are independent of e, the last

differs from f(r) by the term 2 ?{w'(z) + w?(z)} of order e%.] According to L.

PRANDTL, similar circumstances with regard to convergence for e > 0 are to be expected along the surface of any obstacle immersed in a fluid of viscosity e*. We propose to formulate the two-dimensional boundary layer problem in terms of conformal coordinates &,, &, which arise from the Cartesian coordinates

%1, %_ by a conformal transformation. Let ,, “2 be the covariant components

of velocity with respect to these coordinates &,, & and

ds? = ane + daz = e(aé? + aé3)

the square of the line element. The Navier-Stokes equations assume the form 7

Byes

Ge

:

where

om ‘

=

11

22 58

0e

2,

08;

: (aut

OE,

bY 5 nS

Au,

=0,

oP

OE, ou eon

a (oe 0?u;

(4)

Ou;\ ees) 0

Ee) a

li Gia

2

(5)

07,

1) Hremenz, Dincter’s Polytechn. J. 326, 321-326 (1911). — F. Homann, Z. angew. Math. Mech. 16, 153 (1936).

21

We suppose that “,, %», p with e > 0 converge to the flow .,of an ideal fluid arising from a harmonic potential p: te = ae:

Along with serviceable, trarily fixed. potential of

po = const pe

ii

const =o

Se

y any multiple ky with a positive constant factor k is which means that the total strength of the stream may Choose ¢,, €, so that a multiple k¢ of € = &, + 7€, is the the limiting flow ~% and let the stream line &,=0 be

equally be arbicomplex the one

which forms the boundary. We have good reasons to believe, and this belief

is the basis of the boundary-layer theory, that u,, u,/e and #, when expressed in terms of the arguments

= &,, 7 = ,/e, tend to limiting functions

U(E, 7),

V(é, n), P(§, 7) which satisfy the equations arising from (4), (5) by the same passage to the limit. The second equation (5) then shows that 0P/0n = 0, and

P(E, n) is therefore independent of 7 and has the value 2

B(é) = dol, 0) = const — 2 e(&) ’

e(&) = e(E, 0),

throughout the boundary layer. Thereafter the two other equations give

(B)

“ar tay?

So + ng (a 0) = Fe

u Sev

b

where

ne) = a d tog HB)

One has to add the boundary conditions U—~0,

for

V—O0

and

7>0

U-k

for

n>oo.

(B)

A full justification of the basic hypothesis of boundary-layer theory will hardly be possible without changing its differential form as given by these equations into a suitable integral form and without proving the existence of a unique

solution of the problem (B, B)}).

Because of the first equation (B), the flow (U, V) derives from a stream

function y,

au gon

a

=

satisfying the formidable differential equation Oy

\2

Op

dy

Oty

oy

_

0%y

— ($5) | + aga aq 7 ant OE Tae wee

(By)

1) Experience shows that in general the assumptions of the theory are fulfilled only along a certain frontal part of the surface of the solid. For the whole theory see S. Go-pstein, Modern Developments in Fluid Dynamics, vol. I (Oxford, 1938).

22 and the boundary conditions

yo,

7-0,

for

0

for

Sook

Suppose now the obstacle is an angle of 7A (0

(B,)

40°00.

0,

6

z

2/g2(2) dl we get G,(z) = c(z — 2)? °

forse 23,

frele) dz — Bi10

converge, the asymptotic behavior of

is indicated by

and that of w(z) by

f(z)~f (@ — ¢) g() ao = Bz — Br

(17)

w(2)~ kz — Pre,

(18)

As g(z) = g,(z) = e~*8 implies

= fe-?8 gz — B2Bafe*aea3

8. {ey p (4

we find the numerical coefficient « in (11) to be < 0-684. According to the most reliable computations!) « = 0-664. Hence the very first approximate value for « which can be derived from our method misses the mark by not more than 3 per cent.

Given

large z,

any positive constant

c 0 for z> 0. 0

This shows (1) that #, never vanishes, therefore never changes sign and, because of (0) = 1, stays positive throughout; and (2) that # has the same sign as #,.

In the same manner we find that 77, 7, 7, (in this order) are all positive, CeO Next

consider

e059

a finite interval

7 0

iy

0 < z

0

fore 0) and

tion @ for which g(a) = 0, y,(a) = —1. Again we see from

rn=—| that

g, is negative

and

(Lof+

» positive

10" determine

that

solu-

saps et) ae 0. for a

> oo. The solution

(2) = (2) — 2+ O(2) is the one we wish to construct?). It has the properties

o(0)=1;

w()>0,

a,(z2) $0,

(44)

and is characterized by the fact that the condition

y(z) = w(z) — m- Bz) 20 cannot be satisfied throughout the interval 0 < z < co for any positive con-

stant m.

except this w satisfies (40). Indeed, It remains to show that no solution according to what has just been stated, any such solution would have to be of the form m> 0. G2) = w(z) + m- Bz),

z=20. This r then imply

$(t2-+(1—t) 2") >

for any number ¢ between 0 and 1. A function ¢ having this property in our n-dimensional space preserves it on any lincar subspace. The “terrace” {r J[®]

J[®)

(5.2)

--- Eps aa

x, Ge Mea

os)

sa)

Thus, by integrating (5.2) with respect to 241, ..., T,, We get

I [bg] > [9]. The

substitution

(5.3)

etd

et,

,

See

performed on the last »—k variables resulis in the equation

IM[b.) =a". JMB)

Hence

JOE

IS".

Jd]

for

g>1.

For our Riemann integrals the performance of the integration in two steps %, ... | Xp41 --- Zp» and of the substitution (5.3) offers no difficulty. But maybe it is preferable first to argue for the step function ¢,, thus avoiding the difficulty about the maximum of ¢ mentioned above, and then to combine

with

the ensuing inequality

Jb) >a". JM hs] J] > JP(p]—He.

85

For k=0

we define 4=¢

and therefore

FS I=a.| $e) de. Any k-dimensional discrete lattice &, may serve here instead of (5.1), since the first k coordinate vectors ¢,, ..., ¢, may be so chosen as to span the pre-assigned lattice (that is, so that the lattice consists of the vectors @,¢,+...+@,¢, With integral components a). THEorEM IV.

By means of a k-dimensional lattice &, we define

#%(2) = o(e) | Tea, a), 1

Og

the product extending to all vectors a 40

of the lattice %,, and

Jeg=| sade, Then

Jyl= q) JU.

J,(q)/q"-* is an increasing function of q.

In particular,

Joalar=|_ sayae. The application to the situation considered by Minkowski is immediate.

¢ is now the characteristic function of the convex solid : %, consist of the lattice vectors in

we have

Indeed, outside

$f = bq",

[,,..., 6,].

So

long

f(w)2, k-1

%

i=1

Mije» (Har) /b (Gr) ) > (HG) /$(9*)) (9).

(4) We

For any positive

O(h) notice

although

that

[Sh

©(k;a,,---,a,r)

it should

not

be

as,

ar)

depends

forgotten

that

oo

— => x(q).

symmetrically

k may

take

on

the

k; a1,

value

- -.ar,

0.

The

quantitative factor R may be given the following form. Using the abbreviation e?’tiz — e(x) we introduce the following functions of a real variable z:

(5)

SiN Ji(z)= | _ (e(aizr) flog x) da

Then &(N,~2)

is the Fourier transform of their product,

(Sale

VN

(6)

ages

R(N,2) = fi 2@- - +JIq(2) + ¢(— az) de. -00

The proof depends on Vinogradoff’s sums extending over the primes p,

5)

= HCN 2) = Bein); PSN

851)

= 8( A No),

and starts with the exact formula

(7)

A(W,k) = f° Si(q) > + 8,(9)e(— kn) dy.

Following Vinogradoff’s exposition step by step we call attention to a few points where we deviate from him. [1] The lower limit in (5) or in the typical function

J (2) = fies (e(za) /log x) dx has been

set at

VN

rather

than

at 2.

Looking

at the

context

where

this

function makes its first appearance, one realizes that the results are not affected by this change. One immediate advantage is that the estimates

(8)

| J(z)|S2BN/n,

— | J(z)| S2/(an | z])

99 hold for all z, hence

(9)

| Je(z SC: 8(2) )|

throughout where

N/n tor |z|= N+, (2) =} I/(a@|¢ |) dor [2 | 2X. A more important advantage will presently become evident.

[2] The integra-

tion in (6) which Vinogradoff restricts within the bounds + n**/N has been extended to + oo. Because of the second inequality (8) the value of R is not

changed thereby by more than CN?n-**(r-),

Tt ig exactly this step which

makes the main term in the formula (I) independent of Vinogradoff’s exponent h and gives a result from which h has vanished altogether. [3] It is not hard to see why in the expression of x(q) the numbers q:,° - -,q, occur. A rational point is of the form » = m/q where m

is an element of the set Gz of all $(q)

prime residues mod qg. In the reduced form the fraction ayy = a,m/q has the denominator q; and hence e(aiyp) depends only on the residue class mod q, to which p belongs. The following elementary arithmetical observations are to

the point. q’ being a divisor of g, g = q‘d, the number of elements m of Gq which are congruent mod q’ to a given element m’ of Gq has the same value v for each m’. Clearly v Sd and v = $(q)/$(q’). In case a’/q’ is the reduced form of a fraction a/q we therefore have v S (a, q), @ fortiori

(10)

S| 4|-)(7) ¢(q

0, and the sum

if a

e(am/q) =v

meGq

The last remark

m’e Gq!

e(a’m’/q') = (6(9)/0(9)) Hw (7)-

finds application for a——k.

Because

of (10), G(k)

is

majorized by the convergent sum

ACA

ae

[4] One crucial point in the estimate for the contribution to the integral (7)

of part IL of the circumference 0 S y= 1

is the fact that

a

‘5 ° | S(n)| 2dy7 =2(W) SN. An equally good result is obtained in our case where all the factors Si(n)

1

are

different, by observing that if ° | Si(n)| >| S2(m)| dy does not exceed the square

root of

[21 Suey] tay 1 S| by nC. BP)

Baa NP

100

By (9) we realize that | R(N,2)|

0+ Near,

Hence we set

R(N,2) = (N*/n’) - R*(N, 2) and then find not only | R*(N,x)| SC

(II)

| R*(N, x)

— R*(N,7)|

which indicates that R*(N,2x)

variable x if N is large.

but also

SC: (|r—2a’ |/N),

varies very slowly over wide stretches of the

Of course C now means constant with respect to V

and x (or N and z/N).

It seems reasonable to measure x against VN.

therefore write R*(N,xN)

—Ry(x).

Then

Let us

(6) and the inequality (9) prove

that the first r— 2 derivatives of Ry(a) with respect to « are continuous and satisfy relations

(aa)

| Ry (x)| 2;. The asymptotic development of R(N,x) may be obtained

in the

same

way

as for 11.)

We

think

the

parallel

is fairly

instructive: the difference in the singular series is due to the fact that the

residues of primes mod q are necessarily relatively prime to ¢ (apart from the

“few” primes going into q) whereas integers leave all residues mod g. The difference in the quantitative factor comes from the fact that 1 is the density

with which the integers x are distributed over the x-axis while 1/log w is the density of the primes (a fact which has to be combined with the wniform equidistribution over the respective residues modulo any integer q). More irregular problems that can be tackled at all will conform to the same scheme. The main terms of the asymptotic formulas (29) are of the order of magnitude W"-! and N\-1/n" respectively. Of course we may claim their validity only for a sufficiently large number r of terms. The limit is set by the simpler of the two problems. Indeed, the efficiency of the method depends on two circumstances:

1) the contribution to (7)

of part IT of the circum-

71 must be negligible; 2) the singular series must converge ference 0 sufficiently well. The first is the main point. For our present prime number

problem we have to set

110

Si(n) = & e(anp') Suppose

that by

analyzing

Waring’s

(pS BW).

problem

we

have

ascertained

an

even

number s = 2u = 4 such that

(30)

A, (N,0)

SG)? by — Aaa (WV, 0)

ow? (W, 0)

where A refers to the signature

(

$= 2Uu; OE

Sys

"BursBu

ig

oe ——ie

By Bu ):

and then by Schwarz’s inequality and (30) we find the relation

J, 1 S8eG0) = Buln)

Suan) © Sau(n) | dy SC BHInK

which forms the base for settling the first point The accessory point, namely convergence of the singular series, is taken

care of by the following estimate for o due to Loo-keng Hua °:

(31)

|o(a/q)| SCe

for (a,q) =1 where « is any positive number < 1% and C, depends neither on anor q. The requirement of “ sufficiently good ” convergence of the series 3 x(q) means that it is majorized by a series (’ - ¥q-* with some exponent « > 1. Because of (31) this is obviously the case as soon as r= 5. UNIVERSITY

AND

INSTITUTE

OF SASKATCHEWAN, FoR ADVANCED

SASKATOON,

SASKATCHEWAN,

Stupy,

PRINCETON,

N. J.

(1938),

pp. 335-346, Lemma

® See Vinogradoff, l.c.*. ° Math.

Zeitschr., vol. 44

1. 3.

129. On the theory of analytic curves

(H. Weyl and J. Weyl)

Proceeding of the National Academy of Sciences of the United States of America 28, 417—421 (1942)

In a paper of great importance and beauty,! L. V. Ahlfors has simplified and vastly expanded the theory of meromorphic curves inaugurated by the authors of this note.?

In the meantime the younger of us had generalized

the first and second main

theorems

curves whose parameter p varies following observations will show method this more general theory completion as the special case of As in reference 3 let G denote

of our theory to arbitrary analytic

over a given Riemann surface 8. The how by a proper adaptation of Ahlfors’ may be brought up to the same degree of the meromorphic curves. any compact region of # which surrounds

a given nucleus Go, and G its complement. We speak of the condenser G whose outer conductor is G and whose inner conductor Gp carries unit Let ¢ = ¢gq be the potential, which vanishes in G and assumes a charge. RinG). To be precise, the space of the condenser is G* — Go* constant value where Go*, G* are defined by gy = R, ¢ > 0, respectively; for the sake

The constant K is called the

of simplicity we identify G)*, G* with Go, G.

Ahlfors’ formula potential (or the reciprocal capacity) of the condenser. Let x(p) be our (19) for the order function T, is generalized as follows. curve and ¢a local uniformizing parameter at any point of the Riemann surface. Form

X} = S}

=

[x, dx/dt,....@—'x/d#-], 2 | x7

| ay, |X#1|

The integral element S, dt di is invariant;

eal

if one uses any analytic differen-

tial dz instead of dt one gets an S, connected with S, by the equation

S, | dz/dt|? = S: We arrive at this expression for 7,

Ti = SS Si

dt dt,

(1)

the integral extending over the entire Riemann surface (or over G only).

112

Our basic idea is to decompose the space of the condenser into thin layers by the equi-potential lines ¢ = yg. Fora fixed value g; = R — r the

inequality g, S ¢ S R defines a region G,, and the condenser G, has the potential R — g, = 7. The harmonic function ¢ is the real part of an analytic function ¢ — ic = f in G — G) which, to be sure, is not single-valued;

but the differential df is, and this is all that counts.

in that part of the integral (1) which extends over index / we form the integral

Q(r) = SS;+do

Use df instead of dt

G — Gp.

Omitting the

(do = 0)

along the line C;; ¢ = R-—~r. (Here the differentiation d/df in S, really amounts to d/dc.) The flux through the line, dc, is the total charge 1 (not 27; being mathematicians, we use the Heaviside units). The formula (1) now reads

T = AR + fo®e- Q(R — ¢)-de =AR+ where Ap

of G.

fo(R — 1) -Q(r) dr,

is the integral of S,dt di over the nucleus G, and hence independent

Applying the formula to the condenser G, we obtain for T(G,):

T(r) = Aor + So(r — p) Q(p) dp. This proves at once that 7((r) is a (positive increasing) convex function of the variable 7 and that

@T/dr? = Q(r).

(2)

The second main theorem has been formulated in reference 3 in two different forms, first on the basis of a given “‘meromorphic’’ function z on #,

and then by means of the intrinsic non-Euclidean metric on # to which the theory of uniformization leads. In the first form the meromorphic

character of the differential dz in G is sufficient, and it is not necessary to use the same dz for each domain G. from the choice dz = df.

The really important relation arises

This differential is not defined inside G.

one must apply the fundamental formula to ment of Gp brings in a topological moment.

G — G, and the separate treatIn the notation used loc. cit.,

we arrive at the Plicker formulas for analytic curves

Vi + (Tita — 27, + Ti-1) = Qi(r) | 0® + 2a in which the compensating term ©,(r) is the integral QX(r) = 1/2S log S/+doe

Hence

(do = 0)

taken along the line C, and

7 = Zpe() — mR.

113 The last sum extends over the ‘‘critical points’’ ¥, i.e., the zeros of df inside G — G (or, what is the same, the zeros of the electric field strength — grad ¢).

yo is an integer, namely the Euler characteristic of Gj. It is clear that n depends on G, but neither on the index / nor on the curve x(p). It corresponds to the term 2p — 2 in Pliicker’s formulas for an algebraic curve of genus p. The value of 7 is little influenced by whether or not one includes in the sum critical points

0.

near the outer wall of the condenser where ¢ =

But it seems to violate the law of continuity at the inner wall.

How-

ever, according to M. Morse’s now classical relation, Zpl

where

—y

=

vm +

»,

is the Euler characteristic of G.

Hence one can write

1 = —2y(R — g(w)) + rR, a formula in which the critical points near the inner wall

little.

¢ = R count very

Notice the inequality

—wR

Sn

SR.

Again dropping the index / we have 20(r) The potential R(G)

S log Q(r).

of the condenser G increases if G is enlarged,

and

hence converges either to ~ or to a finite limit Ro under exhaustion of B by

Choose a number « > 1. ConG (case of infinite or finite total potential). sidering that T’(r) = Ao and thus T(r) 2 Av as soon as r 2 1 we find in

familiar fashion from (2) that an inequality log Q(r)

> «2 - log T(r) + (x + 1) log GC

with a given positive constant C cannot hold throughout a subinterval of 1 1/2 this fact

a(G) = b- log TG) from holding for all sufficiently large domains G;

but it says a good deal

ect to log T(G). more about the behavior of Q(G) with resp

114

In the case of finite total potential Ry the function ¢g¢ tends to a limit ¢ with the exhaustion of # by G, and it is then natural to use only the regions

So S Ro. Almost everywhere, i.e., with G,(0 Sr < Ro) defined by Ro — 1 the exception of an r-set over which the integral of (Ry — r)~ is finite, the inequality 22(r).


1 in which [X', E”] takes the place of [X', c] can also be treated with fixed exponents. Ahlfors’ second set of defect relations arises from application to the dual curve of the first set thus obtained. A more detailed account will probably be published in a planned monograph on analytic curves in the Annals of Mathematics Studies. 1 Ahlfors,

L. V.,

“The

Theory

Fennicae, Nova Series A, Tom.

of Meromorphic

III, No. 4 (1941).

Curves,”’

2 Weyl, H. and J., Ann. Math., 39, 516-538 (1938). * Weyl, J., Ibid., 42, 371408 (1941).

Acta

Societatis

Scientiarum

130. On Hodge’s theory of harmonic integrals Annals of Mathematics 44, 1—6

(1943)

The attempt which HopcE made in Chapter III of his beautiful book!) to

establish the existence of harmonic integrals with preassigned periods has not been entirely successful because the proof is partly based on a false statement (p. 136) concerning the behavior of the solution of a non-homogeneous integral equation when the spectrum parameter approaches an eigen value. In a Princeton seminar on the subject, BOHNENBLUsT pointed out that counter examples are readily available even for linear equations with a finite number of unknowns. For instance the equation Ax + Ax =c with

1-63) Ch 6)

is solvable for 2 = 0 (x, arbitrary, x, = 1) and yet the solution for 2 + 0, =

+ »

%%=0

does not tend to a limit with 2 > 0.

In his book Honce uses the parametrix method first developed for a single elliptic differential equation by Levi and Hirsert?). Building on the formal foundations laid by Honce, I will show here how the argument can be made conclusive. H1LBERT’s procedure served me as a model.

Let7 be the dimensionality of our Riemannian manifold. I denote by *u, Du

the dual form and the derivative of any (linear differential) form w and use the

abbreviation A for the operator D*D.

For two forms u,v of rank p,n—p

respectively (v, u) designates the integral of the product v - w over the whole

manifold. (*#, «) is positive unless w = 0. An immediate consequence is Lemma 1. Au = 0 implies Du = 0. Indeed (Dx Du, u) = 0 leads to («Du, Du) = 0, hence Du=0.

1) W.V.D. Hopes, The Theory and Applications of Harmonic Integrals (Cambridge, 1941). See also Proc. London Math. Soc. [2] 41, 483-496 (1936), where Honce ascribes the idea of using Hirpert’s parametrix method to H. Kneser. I find it hard to judge whether a previous proof along different lines (Proc. London Math. Soc. [2] 38, 72 (1933)) is complete, or rather how much effort is needed to make it complete. For the Euclidean case, see W. V. D. Hope, Proc. London Math. Soc. [2] 36, 257 (1932), and H. Wevt, Duke Math. J. 7, 411-444 (1940). 2) B.E, Levr, Mem. Soc. ital. Sci. [8a] 16 (1909); D. Hivperr, Grundziige einer allgemeinen Theorie der linearen Integralgleichungen (Leipzig, 1912), pp. 223-231.

116

In the following, f, u, ,

n are forms of rank # and g, v, y, # forms of rank

is fixed; no induction with respect to p takes place. The n—p. The rank goal is to prove the following Theorem I. For any given null form g, g ~ 0, the equation

Au=g

has a solution u.

(1)

I copy Hopce’s two basic formulas (3) and (4) on pp. 132, 133 of his book,

replacing ~—1 by K be the operator into y [ Ky(x, y) with the kernels y

p and using the abbreviation 1/y = (—1)""(n—2)a,. Let with the kernel y - K,(x, y) which carries any form u(x) u(y), and K’ its transpose. The ‘parametrix’ operators Q, P wp_1(%, y) are symmetric, - w,(x, y) and y

(Qu, 8) = (Qg, »)Finally, I set DPD = IT. Hopce’s formulas read Ku—u=QdAu

+ (— 1)"

K'v—v=AQv + (—1)"*

u,

(2)

Tv.

(2’)

The solutions of the equations Ku-—u=0,

K’v—v=0

will be called the eigen forms of the kernels K and K’

value 1’). We try to solve our problem integral equation suggested by (2),

by means

(scilicet ‘for the eigen

of the non-homogeneous

Ku—u=Qg.

(E)

It is essential to study this equation not only for null forms g but in a wider set G; the success of the method depends on the proper choice of that linear space G. Here is my definition: g belongs to Gwhenever

Every form of the type

PDg is closed,

DPDg— Ile — 0s f=I1v_

is said to belong toA Evidently

(v arbitrary) G contains all closed forms g whereas

elements { of F are null forms. “and

G are orthogonal:

Lemma2. (g,/) = 0 forge G feA Indeed, if PDg is closed, then

(PDg, Dv) = 0 = (Dg, PDv), an equation which may at once be changed into

(g, DPD») = 0.

all

117

I take over Hopce’s Lemma I on p. 142:

Lemma

3. I} y is any eigen form of K' then Qy is closed.

For the sake of completeness I repeat the simple proof. Equation (2’) yields for§ = Qy:

AE = (-1)"4« Ty,

hence D*A§ = DxDxDE

= 0 and then by double

D*eDE=0, Incidentally

(3)

we learn from

application of Lemma 13),

DE=0.

(3) and the intermediate

ITy = 0, or that the eigen forms y of K' lie in G.

equation

Aé = 0 that

We analyze the eigen forms of K and K’ as follows. Within the linear space

of all eigen forms @ of K we consider the subspace f of the closed eigen forms y and choose our basis Pry

Pir

Pry so

Pn

for all eigen forms accordingly, i.e. g,, ..., p, span f. Equation (2) yields

Q4g = (— Ir tT

@.

(4)

This proves on the one hand that each closed eigen form

condition I7xp = 0,

of K satisfies the

Lemma 4. «pé G for every pEf. It shows on the other hand that y = A@ satisfies the conditions

AQp=0, because the operators AJJ and JA

Ip~=0

annihilate. It then follows from (2') that

p is an eigen form of K’. The m forms D@,, ..., Dg, are linearly independent by construction, and hence by Lemma 1 Yr = AG,

aia

the same is true for the forms Ym

= APn-

The transposed kernel K’ has the same number /+m eigen forms as K. We determine a basis CR Gcn Vins

of linearly independent

Uhyoor rch

(5)

of which the y’s are a part.

The integral equation (E) is solvable if and only if

(Qs, y) =9= @ Oy) for every eigen form y of K’, or with the notation § = Qy, if

(g, 6) = 9. 1) One differentiation may

(6)

be saved here by applying the formula (Ds, Dt) = 0 holding for

any two forms s, ¢ with continuous first derivatives of rank

p — 1 and n — p — 1 (see H. Wevt,

Duke Math. J. 7, 426) 1940); these Setecra p. 469) to s = PDy and t= *DE with the result

(ITw, AE) = 0 = (+A, AE) whenceAE = 0 = IT.

118

forms yj, .--, Pm “ Let us say that y is of the first kind when & = Qy¢ The are of the first kind, on account of the equation (4). We choose our basis (5) so that

Pro

Pms

Poor Pp

span the linear manifold of all eigen forms of K’ of the first kind. By Lemma 2 the relation (6) holds good for any g€ Gin case y is of the first kind, and thus the m+/ conditions (6) reduce to the last /—y of them, (8 Op)

(€, OY,43) =9,---.

(7)

= 0.

Let G, denote the set of those forms g€ G which satisfy the conditions (7). We have found that under the assumption g€ GQ, the integral equation (E) has a solution u. For this solution « we obtain from (2):

Q(g — Au) = (— 1)" TT,

(8)

hence AQ(g — Au) = 0. Combining this with IJ(g — Au) = 0 and applying (2’) to v = g — Au one finds

6 — AU = PHA

ten im tatoo

tay

to be an eigen form of K’. More precisely, because of (8), Qy¢.F

form of the first kind,

which

forces c,,,,

..., ¢, to vanish.

U + CP, +++: + CmPm We arrive at the following

Intermediary

Proposition.

y

is aneigen

Writing

For any g¢&, there exists a form

constants c,, ..., ¢, such that

g—Au=

4 yt

uw for

u and »

+6, 4,-

(9)

We know from Lemma 4 that the dual form *@ of any element ¢ of f lies

in G. That subspace of f the elements

of which satisfy the conditions

(*P, O41) = 9, «+s (HP, OY) = is of a dimensionality ~ = y. Let the basis y,, ..., p; of f be so chosen that Pr» +++» Py Span this subspace. From (9) we obtain for the » unknowns cg the p linear equations

2, ap “= (8,P,) (#=1,...,4;8=1,...,>) where

I maintain:

(10)

Fag = (Pp Pa) -

Lemma 5. | H,,|| is a non-singular square matrix.

Once this is established we have reached the goal. For then the v conditions

(9) =90

(«=1,...,»)

119

imply ¢,=0 whereby satisfies the relations

(9) reduces

© O41) = 9.

(8,0)

to g—Au=0.

=0;

In other words,

if geG

(a) =0...69,)=0

(11)

then the equation (1) is solvable. A null form g fulfills all our requirements, because the w,; and Qy; are closed, the first by construction, the others by Lemma 3.

Proof of Lemma 5. We have found the equations (10)

g€G,. For

PHU

+4,

to be solvable if

Y,

(12)

the integral («@p, p) is a positive definite quadratic form of a,,..., a,. Hence we can determine the coefficients a; in (12) so as to assign arbitrary values 0, to the integrals (40; 0,)) (2— 1,22, 2) But g= «y€ G,. Hence we see that the equations

Er Hayy = be

(@=1,...,4;B=1,...,9)

have a solution c, for arbitrary b,. In view of 4 = y this statement is equivalent to our lemma. In proving Theorem I we actually showed that the equation Au = g is solvable if g € G satisfies the conditions (11). Hence each such g is a null form, and the linear space is of finite dimensionality < / modulo the space of null forms. As G contains all closed forms of rank —#, we find a fortiori that the number #,,_, of linearly independent closed forms of rank »— modulo null

is finite and 2x'= U(s)x, whereby J=J(x1

- - - x,) changes into a

new form J(xf +++ xn )=J*(x1+++x,). J is an invariant if Jeo=J for every s. (The restriction to unimodular transformations s enables

one to avoid the more involved concept of relative invariants and to remove the restriction to homogeneous polynomials, with the convenient consequence that the invariants form a ring.) The classical problem is a special case of the general problem of invariants in which 5 ranges over an arbitrary given abstract group I’ and s—>U(s) is any representation of that group (that is, a law according to which every element s of I’ induces a linear transformation U(s) of the » variables 1, ° ++, %, such that the composition of group elements is reflected in composition of the induced transformations). The development before Hilbert had led up to two main theorems, which however had been proved in very special cases only. The first states that the invariants have a finite integrity basis, or that we can pick a finite number among

them,

say 41, - « - , ¢m, such that every invariant J is expressi-

ble as a polynomial in 41, - + - , 7. An identical relation between the

basic invariants 11, ++: , tm is a polynomial F(z: - - - 2m) of m independent variables 2, - - - , 2m, which vanishes identically by virtue of the substitution B=

(x1 ~ + Xn), > - 2

Bm = Im(X1 °° + Xn).

The second main theorem asserts that the relations have a finite ideal basis, or that one can pick a finite number among them, say F,, -- +, Fs, such that every relation F is expressible in the form

r ay; ion OH enoGEO

(1)

the Q; being polynomials of the variables 21, ++ +, 2m. I venture the guess that Hilbert first succeeded in proving the sec-

ond theorem. The relations F form a subset within the ring k[z: +++ 2m] of all polynomials of 2,+-+, 2m the coefficients of which lie in a given field k. When Hilbert found his simple proof he could not fail to notice that it applied to any set of polynomials 2 whatsoever and he thus discovered one of the most fundamental theorems of algebra, which was instrumental in ushering in our modern abstract approach, namely:

(A) Every subset & of the polynomial ring k[a: + + » 2m] has a finite

ideal basis.

Is it bad metaphysics to add that his proof turned out so simple because the proposition holds in this generality? The proof proceeds by the adjoining of one variable 2; after the other, the individual step

138

being taken care of by the statement: If a given ring r satisfies the condition (P): that every subset of r has a finite ideal basis, then the ring r[z] of polynomials of a single variable z with coefficients in r satisfies the same condition (P). Once this is established one gets

not only (A) but also an arithmetic refinement discussed by Hilbert in which the field & of rational numbers is replaced by the ring of rational integers. The subset 2 of relations to which Hilbert applies his theorem (A)

is itself an zdeal, and thus the ideal {m, egy } , that is, the totality of all elements of the form (1), Q; © R[z - - + gm], not only contains, but coincides with,

2. The proof, however,

works even if 2 is not an

ideal, and yields at one stroke (1) the enveloping ideal {=} (2) the reduction of that ideal to a finite basis,

{>} = {F,

of 2 and

oo,

F,}.

Construction of a full set of relations Fi, - - - , F, would finish the

investigation of the algebraic structure of the ring of invariants were it true that any relation F can be represented in the form (1) in one way only. But since, generally speaking, this is not so, we must ask for

the

“vectors

of

polynomials”

M=(Mi,---,

M,)

for

which

MiFi+ --++M,.F) vanishes identically in 2 (syzygy of first order). These linear relations M between Fi, -- - , F, again form an ideal to which Theorem (A) is applicable, the basis of the M thus obtained giving rise to syzygies of the second order. To the first two main theorems Hilbert adds a third to the effect that if redundance is avoided, the chain of syzygies breaks off after at most m steps. All this hangs in the air unless we can establish the first main theorem, which is of an altogether different character because it asks for an integrity, not an ideal basis. Discussing invariants we operate in the ring kz=k[x1 - + - x,] of polynomials of x1, +--+, x, in a given field k. Hilbert applies his Theorem (A) to the totality 5 of all invariants J for which J(0, + ++, 0)=0 (a subring of kz which, by the way, is not an ideal!) and me determines an ideal basis a1, - + + , im of 3. Each of the invariants i=i, may be decomposed into a sum 1=i%+47@4 +++ of homogeneous forms of degree 1, 2,---, and

as the summands

are themselves

invariants,

the i, may

be assumed

to be homogeneous forms of degrees v,=1. Hilbert then claims that the 1, - + + , ¢m constitute an integrity basis for all invariants. I use a finite group I’ consisting of N elements s (although this case of the general problem of invariants was never envisaged by Hilbert himself) in order to illustrate the idea by which the transition is made. Every invariant J is representable in the form

(2)

JS =6+

Litt

ey

Lintin

(L, € kz)

139

where ¢ is the constant J(0). If J is of degree v one may lop off in L, all terms of higher degree than y—», without destroying the equation. If it were possible by some process to change the coefficients L in (2)

into invariants, the desired result would follow by induction with re-

spect to the degree of J. In the case of a finite group such a process is readily found: the process of averaging. The linear transformation U(s) of the variables «1, + - - , x, induced by s carries (2) into

J = 6+ Lyinte + Livin e Summation over s and subsequent division by the number N yields the relation where

Vemiceta Lc 1 Lr= yd

7 :

It is of the same nature as (2), except for the decisive fact that according to their formation the new coefficients L* are invariants.” Actually Hilbert had to do, not with a finite group but with the classical problem in which the group I consists of all linear trans-

formations s of g variables m, - - - , 7,, and instead of the averaging

process he had to resort to a differentiation process invented by Cayley, the so-called Cayley Q-process, which he skillfully adapted

for his end. (It is essential in Cayley’s process that the g? components

of the matrix s are independent

variables; instead of the absolutely

invariant polynomials J one has to consider relatively invariant homogeneous forms each of which has a definite degree and weight.) Hilbert’s theorem (A) is the foundation stone of the general theory of algebraic manifolds. Let us now think of k more specifically as the field of all complex numbers. It seems natural to define an algebraic manifold in the space of » coordinates x1, +++, %, by a number of simultaneous algebraic equations fi:=0, --- , fa=O0 (fi © kz). According to Theorem (A), nothing would be gained by admitting an infinite number of equations. Let us denote by Z(fi, - - - , fn) the set of points

x= (x1, +++, %n) wherefi, - + - , fx and hence all elements of the ideal j= {fy cae ifn} vanish simultaneously. g vanishes on Z(f1, «+ + , fa)

whenever

ge {fy - ++, fa}, but the converse is not generally true. For

2 The example of finite groups is used here as an illustration only. Indeed, a direct elementary proof of the first main theorem for finite groups that makes no use of Hilbert’s principle (A) has been given by E. Noether, Math. Ann. vol. 77 (1916) p. 89. In dividing by N we have assumed the field & to be of characteristic zero.

140

instance x; vanishes wherever x? does, and yet x is not of the form x-q(a1 + + + x,). The language of the algebraic geometers distinguishes here between the simple plane «1=0 and the triple plane, although the point set is the same in both cases. Hence what they actually mean

by an algebraic manifold

is the polynomial

ideal and not the

point set of its zeros. But even if one cannot expect that every polynomial g vanishing on Z(fi, + - + , fr) =Z(F) is contained in the ideal ¥={f,-+-, fx} one hopes that at least some power of g will be. Hilbert’s “Nullstellensatz”

states that this is true, at least if k is the

field of complex numbers. It holds in an arbitrary coefficient field k provided one admits points x the coordinates x; of which are taken from k or any algebraic extension of k. Clearly this Nullstellensatz goes to the root of the very concept of algebraic manifolds.* Actually Hilbert conceived it as a tool for the investigation of invariants. As we are now dealing with the full linear group let us consider homogeneous invariants only and drop the adjective homogene-

ous. Exclude the constants (the invariants of degree 0). Suppose we have

ascertained

w non-constant

invariants

J;,---,

J, such that

every non-constant invariant vanishes wherever they vanish simultaneously. An ideal basis of the set 3 of all non-constant invariants certainly meets the demand, but a system Ji, - ++, J, may be had much more cheaply. Indeed, by a beautiful combination of ideas Hilbert proves that if for a given point x=x° there exists at all an invariant which neither vanishes for « =x° nor reduces to a constant, then there exists such an invariant whose weight does not exceed a certain a priori limit W (for example, W=9n(3n-+-1)8 for a ternary

ground form of order 7). Hence the Ji, - - - , J, may be chosen from the invariants of weight not greater than W, and they thus come

within the grasp of explicit algebraic construction. When Hilbert published his proof for the existence of a finite ideal basis, Gordan

invariants, Hilbert

the formalist, at that time looked upon as the king of

cried

out:

remonstrated

“This

then,

is not

mathematics,

as he did

it is theology!”

all his life, against

the

dis-

paragement of existential arguments as “theology,” but we see how, by digging deeper, he was able to meet Gordan’s constructive demands. By combining the Nullstellensatz with the Cayley process he further showed that every invariant J is an integral algebraic (though not an integral rational) function of Ji, - - - , J,, satisfying an equation

3B. L. van der Waerden’s book Moderne Algebra, vol. 2, 2nd ed., 1940, gives on pp. 1-72 an excellent account of the general algebraic concepts and facts with which

we are here concerned.

141

JO Go eG,

10

in which the G’s are polynomials of Ji, - +--+, J,. Hence it must be possible by suitable algebraic extensions to pass from Ji, - ++, J, to a full integrity basis. From there on familiar algebraic patterns such as were developed by Kronecker and as are amenable to explicit construction may be followed. After the formal investigations from Cayley and Sylvester to Gordan, Hilbert inaugurated a new epoch in the theory of invariants. Indeed, by discovering new ideas and introducing new powerful methods he not only brought the subject up to the new level set for algebra by Kronecker and Dedekind, but made such a thorough job of it that he all but finished it, at least as far as the full linear group is concerned. With justifiable pride he concludes his paper, Ueber die vollen Invariantensysteme, with the words: “Thus I believe the most important goals of the theory of the function fields generated by invariants have been attained,” and therewith quits the scene. Of later developments which took place after Hilbert quit, two main lines seem to me the most important: (1) The averaging process, which we applied above to finite groups, carries over to continuous compact groups. By this transcendental process of integration over the group manifold, Adolf Hurwitz treated the real orthogonal group. The method has been of great fertility. The simple remark that invariants for the real orthogonal group are eo ipso also invariant under the full complex orthogonal group indicates how the results can be transferred even to non-compact groups, in particular, as it turns out, to all semi-simple Lie groups. (2) Today the theory of invariants for arbitrary groups has taken its natural place within the frame of the theory of representations of groups by linear substitutions, a development which owes its greatest impulse to G. Frobenius. Although the first main theorem has been proved for wide classes of groups I. we do not yet know whether it holds for every group. Such attempts as have been made to establish it in this generality were soon discovered to have failed. A promising line for an algebraic attack is outlined in item 14 of Hilbert’s Paris list of Mathematical Problems. Having dwelt in such detail on Hilbert’s theory of invariants, we must be brief with regard to his other, more isolated, contributions to algebra. The first paper in which the young algebraist showed his real 41 recommend to the reader’s attention a brief résumé of his invariant-theoretic work which Hilbert himself wrote for the International Mathematical Congress held at Chicago in conjunction with the World Fair in 1893; Collected Papers, vol. 2, item 23.

142

mettle concerns the conditions under which a form with real coefficients is representable as a sum of squares of such forms, in particular with the question whether the obviously necessary condition that the form be positive for all real values of its arguments is sufficient. By ingenious continuity arguments and algebraic constructions Hilbert finds three special cases for which the answer is affirmative, among them of course the positive definite quadratic forms, but counterexamples for all other cases. Similar methods recur in two papers dealing with the attractive problem of the maximum number of real ovals of an algebraic curve or surface and their mutual position. Hilbert conjectured that, irrespective of the number of variables, every rational function with real (or rational) coefficients is a sum of squares of such functions provided its values are positive for real

values

of

the

arguments;

and

in his

Grundlagen

der

Geometrie

he pointed out the role of this fact for the geometric constructions with ruler and “Eichmass.” Later O. Veblen conceived, as the basis of the distinction between positive and negative in any field, the axiom that no square sum equals zero. Independently of him, E. Artin and O. Schreier developed a detailed theory of such “real fields,” and by means of it Artin succeeded in proving Hilbert’s conjecture.® In passing I mention Hilbert’s irreducibility theorem according to which one may substitute in an irreducible polynomial suitable integers for all of the variables but one without destroying the irreducibility of the polynomial, and his paper on the solution of the equation of ninth degree by functions with a minimum number of arguments. They became points of departure for much recent algebraic work (E. Noether, N. Tschebotareff and others). Finally, it ought to be recorded that on the foundations laid by Hilbert a detailed theory of polynomial ideals was erected by E. Lasker and F. S. Macaulay which in turn gave rise to Emmy Noether’s general axiomatic theory of

ideals. Thus in the field of algebra, as in all other fields, Hilbert’s con-

ceptions proved of great consequence for the further development. ALGEBRAIC NUMBER FIELDS When

Hilbert, after finishing off the invariants, turned to the the-

ory of algebraic number fields, the ground had been laid by Dirichlet’s analysis of the group of units more than forty years before, and by

Kummer’s, Dedekind’s and Kronecker’s introduction of ideal divisors. The theory deals with an algebraic field x over the field & 50. Veblen, Trans. Amer. Math. Soc. vol. 7 (1906) pp. 197-199. E. Artin and O. Schreier, Abh. Math. Sem. Hamburgischen Univ. vol. 5 (1926) pp. 85-99; E. Artin,

ibid. pp. 100-115.

143

of rational numbers. One of the most important general results beyond the foundations had been discovered by Dedekind, who showed that the rational prime divisors of the discriminant of « are at the same time those primes whose ideal prime factors in x are not all distinct (ramified primes). J being a rational prime, the adjunction to « of the /th root of a number a in x yields a relative cyclic field K=x(a') of degree / over x, provided x contains the Ith root of unity §=e?"*!! (and according to Lagrange, the most general relative cyclic field of degree / over x is obtained in this fashion). It may be said that it was this circumstance which forced Kummer,

as he tried to prove

Fermat’s theorem of the impossibility of the equation a’+'=~’', to

pass from the rational ground field k to the cyclotomic field x,=k()

and

then

whether

to conceive

the number

his ideal

numbers

in

of classes of equivalent

x, and

to investigate

ideal numbers

in

«; is

prime to J. Hilbert sharpened his tools in resuming Kummer’s study of the

relative

cyclic

fields of degree

/ over

k:, which

he

christened

“Kummer fields.” His own first important contribution was a theory of relative Galois fields K over a given algebraic number field x. His main concern is the relation of the Galois group T of K/x to the way in which the prime ideals of x decompose in K. Given a prime ideal 8 in K of relative degree f, those substitutions s of I for which sG=§8 form the splitting group. As always in Galois theory one constructs the corresponding subfield of K/« (splitting field), to which a number of K belongs if it is invariant under all substitutions of the splitting group. The substitutions ¢ which carry every integer A in K into one, tA, that is congruent to A mod § form an invariant subgroup of the splitting group of index f, called the inertial group, and the corresponding field (inertial field) is sandwiched in between the splitting field and K. Let p be the prime ideal in x into which $ goes, and $¢ the exact power of $ by which p is divisible. I indicate the nature of Hilbert’s results by the following central theorem of his: In the splitting field of ® the prime ideal p in x splits off the prime factor p* = 8" of degree 1 (therefore the name !); in passing from the splitting to the inertial field p* stays prime but its degree increases to f; in

passing from the inertial to the full field K, p* breaks up into e equal prime factors $8 of the same degree f. For later application I add the following remarks. If $ goes into p in the first power only, e=1 (which is necessarily so provided p is not a divisor of the relative discriminant of K/x), then the inertial group consists of the identity only. In that case the theory of Galois’s strictly finite fields shows that the splitting

group is cyclic of order f and that its elements 1, s, s?,- +--+, sf! are

144

uniquely determined by the congruences sA = Ay,

s?A

= AP’,---

(mod$)

holding for every integer A. Here P is the number of residues in k modulo p and thus P/ the number of residues in K modulo $. Today we call s=o(§) the Frobenius substitution of $B; it is of paramount importance that one particular generating substitution of the splitting group may thus be distinguished among all others. One readily sees that for any substitution

wu of the Galois group o(wf) =u-1-o($) -u.

Thus if the Galois field K/x is Abelian, the substitution o(%) =o(w)

depends on p only and may be denoted by (&).

In 1893 the Deutsche Mathematiker-Vereinigung asked Hilbert and Minkowski to submit within two years a report on number theory. Minkowski dropped out after a while. Hilbert’s monumental report Die Theorie der algebraischen Zahlkérper appeared in the Jahresberichte of 1896 (the preface is dated April 1897). What Hilbert accomplished is infinitely more than the Vereinigung could have expected. Indeed, his report is a jewel of mathematical literature. Even today, after almost fifty years, a study of this book is indispensable for anybody who wishes to master the theory of algebraic numbers. Filling the gaps by a number of original investigations, Hilbert welded the theory into an imposing unified body. The proofs of all known theorems he weighed carefully before he decided in favor of those “the principles of which are capable of generalization and the most useful for further research.” But before such a selection could be made that “further research” had to be carried out! Meticulous care was given to the notations, with the result that they have been

universally adopted (including, to the American printer’s dismay, the German letters for ideals!) He greatly simplified Kummer’s theory, which rested on very complicated calculations, and he introduced those concepts and proved a number of those theorems in which we see today the foundations of a general theory of relative Abelian fields. The most important concept is the norm residue symbol, a pivotal theorem on relative cyclic fields, his famous Satz 90 (Collected Works, vol. I, p. 149). From the preface in which he describes the general character of number theory, and the topics covered by his report in particular, let me quote one paragraph: “The theory of number fields is an edifice of rare beauty and harmony. The most richly executed part of this building, as it appears to me, is the theory of Abelian fields which Kummer by his work on

the higher laws of reciprocity, and Kronecker by his investigations

145

on the complex multiplication of elliptic functions, have opened up to us. The deep glimpses into the theory which the work of these two mathematicians affords reveals at the same time that there still lies an abundance of priceless treasures hidden in this domain, beckoning as a rich reward to the explorer who knows the value of such treasures and with love pursues the art to win them.” Hilbert himself was the miner who during the following two years brought to light much of the hidden ore. The analogy with the corresponding problems in the realm of algebraic functions of one variable where Riemann’s powerful instruments of topology and Abelian integrals are available was for him a guiding principle throughout (cf. his remarks in item 12 of his Paris Problems). It is a great pleasure to watch how, step by step, advancing from the special to the general, Hilbert evolves the adequate concepts and methods, and the essential conclusions emerge. I mention his great paper dealing with the relative quadratic fields, and his last and most important Ueber die Theorte der relativ Abelschen Zahlkérper. On the basis of the examples he carried through in detail, he conceived as by divination and formulated the basic facts about the so-called class fields. Whereas Hilbert’s work on invariants was an end, his work on algebraic numbers was

a beginning. Most of the labor of such number theorists of the last decades, as Furtwangler, Takagi, Hasse, Artin, Chevalley, has been devoted to proving the results anticipated by Hilbert. By deriving from the ¢-function the existence of certain auxiliary prime ideals, Hilbert had leaned heavily on transcendental arguments. The subsequent development has gradually eliminated these transcendental methods and shown that though they are the fitting and powerful tool for the investigation of the distribution of prime ideals they are alien to the problem of class fields. In attempting to describe the main issues I shall not ignore the progress and simplification due to this

later development. Hilbert’s theory of norm residues is based on the following discov-

eries of his own: (1) he conceived the basic idea and defined the norm residue symbol for all non-exceptional prime spots; (2) he realized the necessity of introducing infinite prime spots; (3) he formulated the general law of reciprocity in terms of the norm symbol; (4) he saw that by means of that law one can extend the definition of the norm symbol

to the exceptional

prime

spots where

the really interesting

things happen.—It was an essential progress when E. Artin later (5) replaced the roots of unity by the elements of the Galois group as values of the residue symbol. In sketching Hilbert’s problems I shall make use of this idea of Artin’s and also of the abbreviating language

146

of (6) Hensel’s p-adic numbers and (7) Chevalley’s idéles.* As everybody knows an integer @ indivisible by the prime p#2 is said to be a quadratic residue if the congruence x*=a (mod ) is solvable. Gauss introduced the symbol (§) which has the value +1 or —1 according to whether a is a quadratic residue or non-residue mod p, and observed that it isa character, (4) (+) =($*). Indeed, the # residues modulo p—as whose representatives one may take 0, 1,---, (p—1)—form a field, and after exclusion of 0 a group in which the quadratic residues form a subgroup of index 2. Let K=k(b!/?) be a quadratic field which arises from the rational ground field & by adjunction of the square root of the rational number b. An integer a#0 is called by Hilbert a@ p-adic norm in K if modulo any given power of p it is congruent to the norm of a suitable integer in K. He sets “K)=+41 if a is p-adic norm, —1 in the opposite case, and finds that this p-adic norm symbol again is a character. The investigation of numbers modulo arbitrarily high powers of p was systematized by K. Hensel

in the form

of his p-adic numbers,

and

I repeat Hilbert’s

definition in this language: “The rational number a0, or more generally the p-adic number a,+0, is a p-adic norm in K if the equation a, = Nm (x + yb'/?) = x? — by? has a p-adic solution x=x,, y=y,; +1

or

—1

according

to whether

the norm symbol

or not a, is (p-adic)

(ay, K) equals norm

in K.”

The p-adic numbers form a field k(p) and after exclusion of 0 a multiplicative group G, in which, according to Hilbert’s result, the p-adic

norms in K form a subgroup of index 2 or 1. The cyclic nature of the

factor group is the salient point. One easily finds that the p-adic squares form a subgroup G; of index 4 if p#2, of index 8 if p=2, and thus the factor group G,/G; is not cyclic and could not be described by a single character. Of course every p-adic square is a p-adic norm in K. Both steps, the substitution of K-norms for squares and the passage from the modulus pf to arbitrarily high powers of p; the first step amounting to a relaxation, the second to a sharpening of Gauss’s condition for quadratic residues; are equally significant for the success of Hilbert’s definition. Every p-adic number a,+0 is of the form p*-e, where e, is a p-adic unit, and thus a, is of a definite order h (at p). An ordinary rational number a coincides with a definite p-adic number J,(a) =a,. Here I, symbolizes a homomorphic projection of k into k(p):

T,(a+ a’) = 1,(a) + 1,(a'),

— T,(aa’) = Ip(a)-,(a').

6 The latest account of the theory is C. Chevalley’s paper La théorie du corps de

classes, Ann, of Math. vol. 41 (1940) pp. 394-418.

147

The character (%5*) is identical with (I,(a), K). We come to Hilbert’s second discovery: he realized that simple laws will not result unless one adds to the “finite prime spots” p one infinite prime spot g. By definition the g-adic numbers are the real numbers and I,(a) is the real number with which the rational number a coincides. Hence the real number a, is a g-adic norm in K if the equation a,=x*—by? has a solution in real numbers x, y. Clearly if 6>0 or K is real, this is the case for every a,; if however

b A(p2A(x)).

It is immaterial whether we fix the representative in the unique manner described above; our specific rule would not fit anyhow unless x ranges over the numbers 0, 1, 2, - - - . Instead we imagine a quantifier pz of universal applicability which, as it were, selects the representative for us. Zermelo’s axiom of choice is thus woven into the principle of the excluded

middie.

It is a bold step; but the bolder

the better,

as long as it can be shown that we keep within the bounds of consistency! In the formalism, propositional functions are replaced by formulas the handling of which must be described without reference to their meaning. In general, variables x, y, - + - will occur among the symbols of a formula %. We say that the symbol pz binds the variable x in the formula % which follows the symbol"? and that x occurs free in a formula wherever it is not bound by a quantifier with index x. x, ¥, —, pz are symbols entering into the formulas; the German letters are no such symbols, but are used for communication. It is more natural to describe our critical axiom (8) as a rule for the formation of axioms. It says: take any formula YY in which only the variable x occurs free, and any formula

6 without free variables, and by means of

(9)

f(b) > A(p A).

them build the formula

Here (6) stands for the formula derived from % by putting in the entire formula b for the variable « wherever x occurs free. In this way formulas may be obtained as axioms according to certain rules. Deduction proceeds by the rule of syllogism: From two formulas a and a—>b previously obtained, in the second of which the first formula reappears at the left of the symbol —, one obtains the formula b. How does Hilbert propose to show that the game of deduction will never lead to the formula 00? Here is the basic idea of his procedure. As long as one deals with “finite” formulas only, formulas from which the quantifiers pz, py, *- + are absent, one can decide whether they are true or false by merely looking at them. With the entrance of p such a descriptive valuation of formulas becomes impossible: evidence ceases to work. But a concretely given deduction is a sequence of 1 If we wish the rule that pz binds x in all that comes after to be taken literally,

we must write ab in the form—>{§. The formulas will then look like genealogical trees.

161

formulas in which only a limited number of instances of the axiomatic rule (9) will turn up. Let us assume that the only quantifier which occurs is p, and wherever it occurs it is followed by the same finite formula %f, so that the instances of (9) are of the form (10)

W(b1) > A(p2M), +++ , N(b) >

Assume,

moreover,

A(p A).

by, - - - , 6, to be finite. We

then carry out a

re-

duction, replacing pz by a certain finite r wherever it occurs as part of a formula in our sequence. In particular, the formulas (10) will change into

(11) We

(bi) now

(61),

see how

- - +, W(6,)

> A(x),

to choose

++» , ACbs)

> A(x).

r: if by examining

the finite formulas

one after the other, we find one that is true, say

(bs), then we take 6; for r. If every one of them turns out to be false, we choose r at random. Then the hk reduced formulas (11) are “true” and our hypothesis that the deduction

leads to the false for-

mula 00 is carried ad absurdum. The salient point is that a concretely given deduction makes use of a limited number of explicitly exhibited

individuals

b:, ---,

b, only.

If we make

a wrong

choice,

will do no harm

as long

for example, by choosing Alcibiades rather than Aristides as the rep-

resentative of incorruptibility,

our

mistake

as the few people (out of the infinite Athenian crowd) with whom we actually deal are all corruptible. A slightly more complicated case arises when we permit the - +, 6, tocontain pz, but always followed by the same %. Then bi, we first make a fentative reduction replacing p.% by the number 0, say. The formulas bi, ---, 6, are thus changed into reduced finite formulas

6°,

---, 6; and

(10) into

M(t) —> (0), - = = , (B,) > (0). This reduction will do unless (0)

is false and at the same time one

of the (69), - ++, (6), say 1(6§), is true. But then we have in oS a perfectly legitimate representative of 2%, and a second reduction which replaces p:% by 63 will work out all right. However, this is only a modest beginning of the complications ++ with different variables and apawaiting us. Quantifiers pz, py, plied to different formulas will be piled one upon the other. We make a tentative reduction; it will go wrong in certain places and from that failure we learn how to correct it. But the corrected reduction will probably go wrong at other places. We seem to be driven around in a vicious circle, and the problem is to direct our consecutive corrections

162

in such a manner as to obtain assurance that finally a reduction will result that makes good at all places in our given sequence of formulas. Nothing has contributed more to revealing the circle-like character of the usual transfinite arguments of mathematics than these attempts to make sure of consistency in spite of all circles. The symbolism for the formalization of mathematics as well as the general layout and first steps of the proof of consistency are due to Hilbert himself. The program was further advanced by younger collaborators, P. Bernays, W. Ackermann, J. von Neumann. The last two proved the consistency of “arithmetics,” of that part in which the dangerous axiom about the conversion of predicates into sets is not yet admitted. A gap remained which seemed harmless at the time, but already detailed plans were drawn up for the invasion of analysis. Then came a catastrophe: assuming that consistency is established, K. Gédel showed how to construct arithmetical propositions which are evidently true and yet not deducible within the formalism. His method applies to Hilbert’s as well as any other not too limited formalism. Of the two fields, the field of formulas obtainable in Hilbert’s formalism and the field of real propositions that are evidently true, neither contains the other (provided consistency of the formalism can be made evident). Obviously completeness of a formalism in the absolute sense in which Hilbert had envisaged it was now out of the question. When G. Gentzen later closed the gap in the consistency proof for arithmetics, which Gébel’s discovery had revealed to be serious indeed, he succeeded in doing so only by substantially lowering Hilbert’s standard of evidence.'* The boundary line of what is intuitively trustworthy once more became vague. As all hands were needed to defend the homeland of arithmetics, the invasion of analysis never came off, to say nothing of general set theory. This is where the problem now stands; no final solution is in sight. But whatever the future may bring, there is no doubt that Brouwer and Hilbert raised the problem of the foundations of mathematics to a new level. A return to the standpoint of Russell-Whitehead’s Principia Mathematica is unthinkable. Hilbert is the champion of axiomatics. The axiomatic attitude seemed to him one of universal significance, not only for mathematics, but for all sciences. His investigations in the field of physics are conceived in the axiomatic spirit. In his lectures he liked to illustrate the method by examples taken from biology, economics, and so on. The modern epistemological interpretation of science has been profoundly influenced by him. Sometimes when he praised the axiomatic method 18 G, Gentzen, Math. Ann. vol. 112 (1936) pp. 493-565.

163

he seemed to imply that it was destined to obliterate completely the constructive or genetic method. I am certain that, at least in later life, this was not his true opinion. For whereas he deals with the primary mathematical objects by means of the axioms of his symbolic system, the formulas are constructed in the most explicit and finite manner. In recent times the axiomatic method has spread from the roots to all branches of the mathematical tree. Algebra, for one, is permeated from top to bottom by the axiomatic spirit. One may describe the role of axioms here as the subservient one of fixing the range of variables entering into the explicit constructions. But it would not be too difficult to retouch the picture so as to make the axioms appear as the masters. An impartial attitude will do justice to both sides; not a little of the attractiveness of modern mathematical research is due to a happy blending of axiomatic and genetic procedures. INTEGRAL EQUATIONS Between the two periods during which Hilbert’s efforts were concentrated on the foundations, first of geometry, then of mathematics in general, there lie twenty long years devoted to analysis and physics. In the winter of 1900-1901 the Swedish mathematician E. Holmgren reported in Hilbert’s seminar on Fredholm’s first publications on integral equations, and it seems that Hilbert caught fire at once. The subject has a long and tortuous history, beginning with Daniel Bernoulli. The mathematicians’ efforts to solve the (mechanical, acoustical, optical, electromagnetical) problem of the oscillations of a continuum and the related boundary value problems of potential theory span a period of two centuries. Fourier’s Théorie analytique de la chaleur (1822) is a landmark. H. A. Schwarz proved for the first time (1885) the existence of a proper oscillation in two and more dimensions by constructing the fundamental frequency of a membrane. The last decade of the nineteenth century saw Poincaré on his way to the development of powerful function-theoretic methods; C. Neumann and he came to grips with the harmonic boundary problem; Volterra studied that type of integral equations which now bears his name, and for linear equations with infinitely many unknowns Helge von Koch developed the infinite determinants. Most scientific discoveries are made when “their time is fulfilled”; sometimes, but seldom, a genius lifts the veil decades earlier than could have been expected. Fredholm’s discovery has always seemed to me one that was long overdue when it came. What could be more natural than the idea that a set of linear equations connected with a discrete set of mass points gives way to an integral equation when one passes to

164

the limit of a continuum? But the fact that in the simpler cases a differential rather than an integral equation results in the limit riveted the mathematicians’ attention for two hundred years on differential equations! It must be said, however, that the simplicity of Fredholm’s results is due to the particular form of his equation, on which it was hard to hit without the guidance of the problems of mathematical physics to which he applied it: a(s) — f

x6

)x(t)dt = f(s)

(0; Sts1S1)3

Indeed the linear operator which in the left member operates on the unknown function x producing a given f, (E—K)x=f, consists of two parts, the identity E and the integral operator K, which in a certain sense is weak compared to E. Fredholm proved that for this type of integral equation the two main facts about » linear equations with the same number 2 of unknowns hold: (1) The homogeneous equation

[f(s)=0]

has

a finite

x(s)=¢1(s), > ++, a(s), transposed kernel K’(s,

number

of linearly

independent

solutions

and the homogeneous equation with the t)=K(#, s) has the same number of solu-

tions, ¥a(s),--- , Wa(s). (2) The nonhomogeneous equation is solv-

able if and only if the given f satisfies the # linear conditions

f soveis = 0

Geeimeceas

Following an artifice used by Poincaré, Fredholm introduces a parameter \ replacing K by AK and obtains a solution in the form familiar from finite linear equations, namely as a quotient of two determinants of H. v. Koch’s type, either of which is an entire function of the parameter \.

Hilbert saw two things: (1) after having constructed Green’s function K for a given region G and for the potential equation Au=0 by means of a Fredholm equation on the boundary, the differential equation of the oscillating membrane A®+A¢=0 changes into a homogeneous integral equation

$(s) — rf K(s, )o(t)dt = 0 with the symmetric kernel K, K(¢, s) =K(s, ) (in which the parameter

d is no longer artificial but of the very essence of the problem); (2) the problem of ascertaining the “eigen values” \ and “eigen func-

165

tions” ¢(s) of this integral equation is the analogue for integrals of the transformation of a quadratic form of variables onto principal axes. Hence the corresponding theorem for the quadratic integral form

ia ff K(s, t)«(s) «(d)dsdt

(12)

0

)

with an arbitrary symmetric kernel K must provide the general foundation for the theory of oscillations of a continuous medium. If others

saw the same, Hilbert saw it at least that much more clearly that he bent all his energy on proving that proposition, and he succeeded by the same direct method which about 1730 Bernoulli had applied to the oscillations of a string: passage to the limit from the algebraic problem. In carrying out the limiting process he had to make use of the Koch-Fredholm determinant. He finds that there is a sequence of eigen values hi, A2, - + - tending to infinity, \,>

for n>,

an orthonormal set of corresponding eigen functions ¢,(s),

and

n(s) — Xn ff we Don(Adt = 0, 0

fF ba(s)6a(6)ds = dan such that

f ‘ f "K(s, t)x(s)x(t)dsdt = D> Ex/ dns 0

0

£, being the Fourier coefficient S,x(s)bn(s)ds. The theory implies that

every function of the form

y(s) -f

K(s, t) x(t)dt

may be expanded into a uniformly convergent Fourier series in terms of the eigen functions ¢n,

ys) = Labal)s

= f y(s)oa(s)ds.

Hilbert’s passage to the limit is laborious. Soon afterwards E. Schmidt in a Gottingen thesis found a simpler and more constructive proof for these results by adapting H. A. Schwarz’s method invented twenty years before to the needs of integral equations. From finite forms the road leads either to integrals or to infinite

166

series. Therefore Hilbert considered the same problem of orthogonal transformation of a given quadratic form

DE netae

(13)

into a form of the special type (14)

aki Stee

2

ap

OS

(kn =

1/dn> 0)

also for infinitely many (real) variables (x1, x2, - - - ) or vectors x ina

space of a denumerable infinity of dimensions. Only such vectors are admitted as have a finite length |x| i 2

2

coher they constitute what we now call the Hilbert space. The advantage of Hilbert space over the “space” of all continuous functions «(s) lies in a certain property of completeness, and due to this property one can establish “complete continuity” as the necessary and sufficient condition for the transformability of a given quadratic form K, (13), into (14), by following an argument well known in the algebraic case: one determines

1, k2,-~--

as the consecutive

maxima

of K on the

“sphere” |x| ?=1. As suggested by the theorem concerning a quadratic integral form, the link between the space of functions x(s) and the Hilbert space of vectors (x1, ¥2, : + - ) is provided by an arbitrary complete orthonormal system 1(s), w2(s),

+: + and expressed

by the equations

Ln = ar x(5)Un(s)ds.

Bessel’s inequality states that the square sum of the Fourier coefficients x, is less than or equal to the square integral of x(s). The relation of completeness, first introduced by A. Hurwitz and studied in detail by W. Stekloff, requires that in this inequality the equality sign prevail. Thus the theorem on quadratic forms of infinitely many variables at once gives the corresponding results about the eigen values and eigen functions of symmetric kernels K(s, 1)—or would do so if one could count on the uniform convergence of > x,w,(s) for any given vector (x1, 2, -- +) in Hilbert space. In the special case of an eigen vector of that quadratic form (13) which corresponds to the integral form (12), w=

De Kenan,

m

167

Hilbert settles this point by forming the uniformly convergent series

Ny eal

8

K(s, #)um(t)dt

which indeed yields a continuous function ¢(s) with the nth Fourier coefficient

XD Kamin

= ny

and thus obtains the eigen function of K(s, #) for the eigen value }. Soon afterwards, under the stimulus of Hilbert’s investigations, E. Fischer and F. Riesz proved their well known theorem that the space of all functions x(s) the square of which has a finite Lebesgue integral enjoys the same property of completeness as Hilbert space, and hence one is mapped isomorphically upon the other in a one-toone fashion by means of a complete orthonormal system w,(s). I mention these details because the historic order of events may have fallen into oblivion with many of our younger mathematicians, for whom Hilbert space has assumed that abstract connotation which no longer distinguishes between the two realizations, the square integrable functions x(s) and the square summable sequences (x1, x2,

+--+).

I think

Hilbert

was

wise

to keep

within

the

bounds

of continuous functions when there was no actual need for introducing Lebesgue’s general concepts. Perhaps Hilbert’s greatest accomplishment in the field of integral equations is his extension of the theory of spectral decomposition from the completely continuous to the so-called bounded quadratic forms. He finds that then the point spectrum will in general have condensation points and a continuous spectrum will appear beside the point spectrum. Again he proceeds by directly carrying out the transition to the limit, letting the number of variables x, x2, +++ increase ad infinitum. Again, not long afterwards, simpler proofs for his results were found. While thus advancing the boundaries of the general theory, he did not lose sight of the ordinary and partial differential equations from which it had sprung. Simultaneously with the young Italian mathematician Eugenio Elia Levi he developed the parametrix method as a bridge between differential and integral equations. For a given elliptic differential operator A* of the second order, the parametrix K(s, ?) is a sort of qualitative approximation of Green’s function, depending like the latter on an argument point s and a parameter point ¢. Itis supposed to possess the right kind of singularity for s=¢ so that the nonhomogeneous equation A*u=f for

168

u = Kp,

u(s) = ff x6

t)p(t)dt

gives rise to the integral equation p+Lp=f for the density p, with a kernel L(s, t) =A*K(s, ¢) regular enough at s =¢ for Fredholm’s theory to be applicable. It is important to give up the assumption that K satisfies the equation A*K =0, because in general a fundamental solution will not be known for the given differential operator A*. In order not to be bothered by boundary conditions, Hilbert assumes the domain of integration to be a compact manifold, like the surface of a sphere,

having

and

finds that

the method

works

the right kind of singularity,

argument

if the parametrix,

is symmetric

with

besides

respect to

and parameter.

What has been said should be enough to make clear that in the terrain of analysis a rich vein of gold had been struck, comparatively easy to exploit and not soon to be exhausted. The linear equations of infinitely many unknowns had to be investigated further (E. Schmidt, F. Riesz, O. Toeplitz, E. Hellinger, and others); the continuous spectrum and its appearance in integral equations with “singular” kernels awaited closer analysis (E. Hellinger, T. Carleman); ordinary differential equations, with regular or singular boundaries, of second or of higher

order,

received

their

due

share

of

attention

(A.

Kneser,

E. Hilb, G. D. Birkhoff, M. Bécher, J. D. Tamarkin, and many others).!4 It became possible to develop such asymptotic laws for the distribution of eigen values as were required by the thermodynamics of radiation (H. Weyl, R. Courant). Expansions in terms of orthogonal functions were studied independently of their origin in differential or integral equations. New light fell upon Stieltjes’s continued fractions and the problem of momentum. The most ambitious began to attack nonlinear integral equations. A large international school of young mathematicians gathered around Hilbert and integral equations became the fashion of the day, not only in Germany, but also in France where great masters like E. Picard and Goursat paid their tributes, in Italy and on this side of the Atlantic. Many good papers were written, and many mediocre ones. But the total effect was an appreciable change in the aspect of analysis. Remarkable are the applications of integral equations outside the field for which they were invented. Among them I mention the following three: (1) Riemann’s problem of determining 7 analytic func\ For later literature and systems of differential equations see Axel Schur, Math. Ann. vol. 82 (1921) pp. 213-239; G. A. Bliss, Trans. Amer. Math. Soc. vol. 28 (1926)

pp. 561-584; W. T. Reid, ibid. vol. 44 (1938) pp. 508-521.

169

tions fi(z),---, fn(z), regular except at a finite number of points, which by analytic continuation around these points suffer preassigned linear transformations. The problem was solved by Hilbert himself, and subsequently in a simpler and more complete form by J. Plemelj. (A very special case of it is the existence of algebraic functions on a Riemann surface if that surface is given as a covering surface of the complex z-plane.) Investigations by G. D. Birkhoff on matrices of analytic functions lie in the same line. (2) Proof for the completeness of the irreducible representations of a compact continuous group. This is an indispensable tool for the approach to the general theory of invariants by means of Adolf Hurwitz’s integration method, and with its refinements and generalizations plays an important role in modern group-theoretic research, including H. Bohr’s theory of almost periodic functions.*

Contact

is thus made

with

Hilbert’s old friend, the

theory of invariants. (3) Quite recently Hilbert’s parametrix method has served to establish the central existence theorem in W. V. D. Hodge’s theory of harmonic integrals in compact Riemannian spaces.'* The story would be dramatic enough had it ended here. But then a sort of miracle happened: the spectrum theory in Hilbert space was discovered to be the adequate mathematical instrument of the new quantum physics inaugurated by Heisenberg and Schrédinger in 1923. This latter impulse led to a reexamination of the entire complex of problems with refined means (J. von Neumann, A. Wintner, M. H. Stone, K. Friedrichs). As J. von Neumann was Hilbert’s collaborator toward the close of that epoch when his interest was divided between quantum physics and foundations, the historic continuity with Hilbert’s own scientific activities is unbroken, even for this later phase. What has become of the theory of abstract spaces and their linear operators in our times lies beyond the bounds of this report. A picture of Hilbert’s “analytic” period would be incomplete without mentioning a second motif, calculus of variations, which crossed the dominating one of integral equations. The “theorem of independence” with which he concludes his Paris survey of mathematical problems (1900) is an important contribution to the formal apparatus of that calculus. But of much greater consequence was his audacious direct assault on the functional maxima and minima problems. The whole finely wrought machinery of the calculus of variations is here 15 H. Weyl and F. Peter, Math. Ann. vol. 97 (1927) pp. 737-755. A. Haar, Ann. of Math. vol. 34 (1933) pp. 147-169. J. von Neumann, Trans. Amer. Math. Soc. vol. 36 (1934) pp. 445-492. Cf. also L. Pontrjagin, Topological groups, Princeton, 1939. 16 W. V. D. Hodge, The theory and applications of harmonic integrals, Cambridge, 1941. H. Weyl, Ann. of Math. vol. 44 (1943) pp. 1-6.

170

consciously set aside. He proposes instead to construct the minimizing

function as the limit of a sequence of functions for which the value of the integral under investigation tends to its minimum value. The classical example is Dirichlet’s integral in a two-dimensional region G,

nl fA) + CO}

Admitted are all functions u with continuous derivatives which have given boundary values. d being the lower limit of D [u] for admissible u, one can ascertain a sequence of admissible functions ~, such that D{u,]—d with n~. One cannot expect the u, themselves to converge;

rather

they

have

to

be

prepared

for

convergence

by

the

smoothing process of integration. As the limit function will be harmonic and the value of the harmonic function at any point P equals its mean value over any circle K around P, it seems best to replace

u,(P) by its mean value in K, with the expectation that this mean value will converge toward a number u(P) which is independent of the circle and in its dependence on P solves the minimum problem. Besides integration Hilbert uses the process of sifting a suitable subsequence from the w, before passing to the limit. Owing to the simple inequality

[Dlvm — a]} 2 S {Dlum] ~ a}? + {Dlun] = dr

discovered by S. Zaremba this second step is Hilbert’s method is even better suited for boundary does not figure so prominently as problems. By a slight modification one is able

unnecessary. problems in which the in the boundary value to include point singu-

larities, and Hilbert thus solved the fundamental problem for flows on Riemann surfaces, providing thereby the necessary foundation for

Riemann’s own approach to the theory of Abelian integrals, and he further showed that Poincaré’s and Koebe’s fundamental theorems on uniformization could be established in the same way. We should be much better off in number theory if methods were known which are as powerful for the construction of relative Abelian and Galois fields over given algebraic number fields as the Riemann-Hilbert transcendental method proves to be for the analogous problems in the fields of algebraic functions! Its wide application in the theory of conformal mapping and of minimal surfaces is revealed by the work of the man who was Hilbert’s closest collaborator in the direction of mathematical affairs at Géttingen for many years, Richard Courant.” 1 A book by Courant on the Dirichlet principle is in preparation.

171

Of a more indirect character, but of considerable vigor, is the influence

of Hilbert’s ideas upon the whole trend of the modern development of the

calculus

of variations;

in

Europe

Carathéodory,

Lebesgue,

Tonelli could be mentioned among others, in this country the chain reaches from O. Bolza’s early to M. Morse’s most recent work. Puysics

Already

systematic

before study

Minkowski’s

of theoretical

death in 1909, physics,

Hilbert had begun

in close collaboration

a

with

his friend who had always kept in touch with the neighboring science. Minkowski’s

work

on

relativity

theory

was

the first fruit of these

joint studies. Hilbert continued them through the years, and between 1910 and 1930 often lectured and conducted seminars on topics of physics. He greatly enjoyed this widening of his horizon and his con-

tact with physicists, whom he could meet on their own ground. The harvest however can hardly be compared with his achievements in pure mathematics. The maze of experimental facts which the physi-

cist has to take into account is too manifold, their expansion too fast,

and their aspect and relative weight too changeable for the axiomatic method to find a firm enough foothold, except in the thoroughly consolidated parts of our physical knowledge.

Men like Einstein or Niels

Bohr grope their way in the dark toward their conceptions of general

relativity

or atomic

structure

by

another

type

of experience

and

imagination than those of the mathematician, although no doubt mathematics is an essential ingredient. Thus Hilbert’s vast plans in physics never matured. But his application of integral equations to kinetic gas theory and to the elementary theory of radiation were notable contributions. In particular, his asymptotic solution of Maxwell-Boltzmann’s fundamental equation in kinetic gas theory, which is an integral equation of the second order, clearly separated the two layers of phenomenolog-

ical physical

laws to which

the theory leads; it has been carried out

in more detail by the physicists and applied to several concrete problems. In his investigations on general relativity Hilbert combined Einstein’s theory of gravitation with G. Mie’s program of pure field physics. For the development of the theory of general relativity at that stage, Einstein’s more sober procedure, which did not couple the theory with Mie’s highly speculative program, proved the more fertile. Hilbert’s endeavors must be looked upon as a forerunner of a unified field theory of gravitation and electromagnetism. However, there was still much too much arbitrariness involved in Hilbert’s Hamiltonian function; subsequent attempts (by Weyl, Eddington,

172

Einstein himself, and others) aimed to reduce it. Hopes in the Hilbert

circle ran high at that time; the dream of a universal law accounting both for the structure of the cosmos as a whole, and of all the atomic

nuclei, seemed near fulfillment. But the problem of a unified field theory stands to this day as an unsolved problem; it is almost certain that a satisfactory solution will have to include the material waves (the Schrédinger-Dirac y for the electron, and similar field quantities for the other nuclear particles) besides gravitation and electromagnetism, and that its mathematical frame will not be a simple enlargement of that of Einstein’s now classical theory of gravitation. Hilbert was not only a great scholar, but also a great teacher. Wit-

nesses are his many pupils and assistants, whom he taught the handicraft of mathematical research by letting them share in his own work

and its overflow, and then his lectures, the notes of many of which have found their way from Géttingen into public and private mathematical libraries. They covered an extremely wide range. The book he published with S. Cohn-Vossen on Anschauliche Geometrie is an outgrowth of his teaching activities. Going over the impressive list attached to his Collected Papers (vol. 3, p. 430) one is struck by the considerable number of courses on general topics like “Knowledge and Thinking,”

“On

the Infinite,” “Nature

and

Mathematics.”

His

speech was fairly fluent, not as hesitant as Minkowski’s, and far from monotonous. He had no difficulty in finding the pregnant words, and liked to emphasize short pivotal phrases by repeating them several times. On the whole, his lectures were a faithful reflection of his spirit; direct, intense; how could they fail to be inspiring?

133.

Concerning a classical problem in the theory of singular points of ordinary differential equations Actas de la Academia Nacional de Ciencias Exactas, Fisicas y Naturales de Lima 7, 21—60 (1944)

1.

Introduction:

Aim,

Hypotheses,

Qualitative Propositions

In connection with some hydrodynamical investigations I recently had to make use of the classical theory of singular points of an ordinary differential equation of first order. The older form of the theory, going back to the days of Briot and Bouquet, depends on the assumption of analyticity. However, later treatments,

by Bendixson

and

above

all by

Perron,

have

relaxed

the

conditions enormously.! In particular, Perron goes as far as one can possibly go, but his arguments are subtle and to a large extent of non-constructive character. I wish to return here to the problem in its parametric form

(1)

dx

= 1.) > dt

dy

oe) dt

I, Bendixson, Acta Mathematica 24, 1901, 1-88. O. Perron, Math. Zeitschr. 15, 1922, 121-146; 16,

1923,

- 295. 273

174

under assumptions sufficiently wide for all applications, yet so devised as to make possible a direct and straightforward approach that leads to remarkably sharp results. Theorem IX is the most

striking instance. The stress lies on the explicit and precise estimates. The constructions we give are of a nature that alone

satisfies the practical needs: they can actually be carried through with any degree of accuracy that is wanted; i.e. the error in the

results will be less than «= 10°", » being a preassigned integer, provided the data are known with an uncertainty not exceeding a corresponding

amount

6.

Every

by an explicit estimate of the error. The

right members

limit equation

of the equations

(1) are supposed

real functions of the real variables x, y defined hood of, and vanishing at the origin O:

F(0,0)

= 0,

~myx

We assume that the eigenvalues, the roots

p = k,1

equation

| are real and distinct,

mn,

,

in the neighbor-

transformation

Gle,y)

p+my

to be

G(0,0) = 0.

In first approximation the infinitesimal x, y -plane described by (1) is linear,

Fay ~ mux ) t+mpy,

is supported

mix p+ Mtn

=0,

of

the

+ my. of the secular

175

(2)

(my, - m2)? + 42m After

a suitable

suppose

that

affine

F (x,y) ~-

change kx,

> 0.

of coordinates

G(x,y)~-

we

ly.

may

We

then

therefore

set

F@,9) =-kx+ Let us use

the

max(|x|,

|y|).

consistently

by

f,9),

abbreviation

in which

|x],

for

|y@)|< Ro

Se

origin

not by

f(p),

and

will

be

|p| 0, Then |G) = (tra) toe

Bropi Gl 1.6. Let G>

such that |G|

tl

aye

7D: fh) > 0,

0 (and g = 2). Suppose we

= c+ (g-1- tr G)’. Then

have a

positive constant ¢ ? implies r*/r < e. One may

Lidia geste rn at,

assume = 1.

(3)

The g numbers (7 + 7*)/2, (y + *)/2, 7”, ..., “9-) have the arithmetical mean = 1. Hence their geometric mean

(3) and (4) imply

cannot exceed unity:

(oe re gi A

4ry*

were

:

re —7\2

(Fey) Ste

r*(L—Vi—c) 0 and suppose we have a number ¢ and a positive constant ¢

1 such that

Bett, Ges tye |G ct)

Then we can ascertain two positive numbers } and B S g which depend on c

only such that

bis

Bt.

(5)

We have to derive the inequalities (5) from the fact that the g positive numbers /“ satisfy the relations oat,

7.77

2c,

235

In doing so we may assume ¢ = 1. The assumption x 7”)... * stands for 7’ implies

Bs

pee

Lo

Because of

aud

gi

(9)

)

\9-1

(a)

Sy=

= 00

=

fp = py) —

=

lh

this implies

> c, in which

(fat x(£=5)

ge — a,

SiG,

The logarithmic derivative

AG

ee

x

L—

4

g-—

2%

shows that /(x) increases monotonely from 0 to 1 and then decreases mono-

tonely from 1 to 0, while x travels first from 0 to 1 and then from 1 to g. Therefore /(x) = c requires x to lie between the two roots x = b and x= B

the equation f(x) = c has between 0 and g; b ) = #° is one for which

ee

5a ei)

(eox) ay

whatever the vectors x, x, and the element

« in §.

After interchanging &

and y, (7) may also be interpreted as the expression of such a linear mapping.

237

It is non-singular

and has an inverse if

x= 0 is the only vector mapped into

=0.

By extension of the rational field & to the field K of all real numbers § becomes an algebra § of rank g over K. All that has been said about § remains valid in §,. However §, will, in general, have ceased to be a division

algebra. Indeed, only if |D| = Nm 6 + 0, has the element 6 of §, a reciprocal

6-1 and the equation = 67 for 7 the solution 7 = 6-1. (This is merely the special case » = 1 of our remarks concerning linear substitutions.) In S"/§ the following principle of the step-by-step construction of a basis of a linear subspace holds: If m—1 vectors d,, ..., Dm, are linearly independent and d does not lie in (1, ..., Dm 4] then d,, ..., Dm 1, d are linearly independent. It goes without saying that this principle breaks down in S"/¥,. In terms of N vectors a,, ..., ay of S"/%§ which are linearly independent in k, every vector x of S”/ may be expressed as a linear combination #,a,+ +++ + UyQy with rational coefficients «,. The x with integral u, form a lattice U, ‘spanned’ by the vectors a,, ..., ay. Let d,, ..., D, be any ” vectors of UM which are linearly independent in §: we speak of them as constituting a semi-basis

of 2 (in %). (6) will lie in W whenever (7,, ..., 7») lies in a certain lattice 2, the ‘representation of WU in terms of the semi-basis D,, ..., D,,’. As d, is repre-

sented by e,, & and oy, .-., DE carried into one sense equivalent

contains the unit vectors e,. Any of YW lead to two representations another bya linear substitution ae or belong to the same class. Hence

two semi-bases D,, ..., Dy, 2 and L* of YW which are = Ys %yyM, and are in this all possible representations

2 of Y in terms of semi-bases satisfy two requirements: they belong to the same class and they contain the unit vectors e, (‘admissible lattices’ 2).

The element f of % is called a multiplier of U if xf lies in XU whenever x does.

The multipliers of a given lattice form an order {3}. The multipliers of 2 are identical with those of the representing lattice 2; even without regard to Wf it is obvious that equivalent lattices 2 and * have the same multipliers. With

e,,..., €, every vector

© = 161 bs

ae On Sn — (61---» Sn)

the components €,, of which are in {%} lies in 2. All these vectors form a lattice

5, the ‘unit lattice’ of {F}. If @,, -.-, @, is a minimal basis of {S}, 3 is spanned

by the N vectors e,,«; (Gat Ey = Xp @1 + -°* + Xpg Wy is replaced

eee) and if each component t= by the column x, of its coordinates

(%y1» «++» %yg), then § consists of those vectors x for which all N coordinates

%,; are rational integers. The index j = [@:3] is the number of vectors in AU

which are incongruent mod 3. On limiting ourselves to the vectors of the form (So

cop

Bh

on @)

3m and a corin 2 and % we obtain two (mg)-dimensional lattices L,, responding index jm = [2m:%m]- In the series j,, ..., 7, each term is a divisor

238

of the subsequent one, and the last coincides with 7. There is only a finite £D § with a given index j, a fortiori with a given series number of lattices of indices j,, ..., 7,. That number is further limited by the restriction to lattices 2 of a given class. Let /(x) be a gauge function in MINKOwWsSKI's sense, so that f(x) < 1 defines a symmetric convex body in S"/§,. Denote by V its volume computed in terms of the above described coordinates x,;. The fundamental parallelotope of 3, Lk

02435 has

the

volume

1 in these

leo)

Seach

coordinates,

hence

that

of

2

the

volume

j-!=

[£:3]}-4. Assume now that the convex body contains no vector z of £ besides

the origin, in other words that x =o Then

is the only solution of f(z) < 1 in 2.

the basic principle of Minkowski’s

inequality 2-%- V #°, is called a Jattice transformation if it maps the lattice & in one-to-one fashion into itself. The ‘lattice group’ of all such transformations is the main

object of our study. s carries a semi-basis

},, ..., 0,

of YW into a semi-basis df, ..., Df; vice versa a semi-basis oT, ..., D* arises from

d,, ..., D, by a lattice substitution s, dj = d%,Hs if, and only sentations 2 and {* of Y in terms of the dD and the d* coincide.

B. THE

BASIC

CONCEPT

§ 3.

The

decomposition

of the

OF

Conjugate

matrix

QUADRATIC

if, the

repre-

FORM

Complex

algebra

(§,)

yields

a certain

basis

@), ..., @j—we call it normal basis—such that with every matrix A in (¥,)

the transpose A’ also belongs to (§,). To prove this, use for a moment the following notations: For an abstract algebra a let Mj, a be the algebra of all

h-rowed matrices \ a,,||, the coefficients «,, of which are elements of a. For a

matrix algebra YW let M;, MU and I), U be the algebras consisting of the matrices

| Au Ais Arn | A, SOF #8) Ay, Aog, ---, Ag, | (Ae€ YM), | 0, A, ..., An»

|

» Ann

An»

while 21+ 8 consists of the matrices

40| ras

|

(AcU,

Be B).

0,

0,

0 0] A

(A € , & rows)

239 After extension to K, §, is still a semi-simple algebra and hence, according

to WEDDERBURN, of the form

M,, 8 + My, 82+ °° where §,, G2, ... are division algebras over K and the + sign indicates a direct

sum. Therefore the regular representations nected by the relation (Sx)

= I, M,,

(%)

(8) ae is. M,,

and (%,),

(&.) te

(Fs), ... are con-

aie

(10)

This decomposition refers to an appropriate basis. There are only three possibilities for §, (p = 1, 2, ...): it is either the field K itself, basis 1; or the complex field over K, basis 1, 7; or the quaternion algebra, basis 1, 7, ;, & in the usual

notation. In the three cases the generic matrix of (§,) is

a,

| a, —4, %,

|

a

|

@, —G1, —as, —Gz

|

Gres Go Gs, 43, —@,,

|,

hy Go, Gy,

EN = Ai a

Hence 4, lies in (§,) if A, does, and then (10) shows at once that A’ lies in (¥x) if A does. Our result means two things: (1) Every element « of §,% has a conjugate « which arises from it by an anti-automorphic involution; i.e. the conjugate of a is « and, x being any real number, the conjugates of xa, «+f, « B are

xa,a%+B, Ba.

(2) If A is the matrix representing « in terms of the normal

basis, then « is represented by A’. Since A’ has the same trace and determinant as A, one finds at once

tra=tra,

Nma=Nma«.

Because we keep this involution & >ێ and the normal basis w/ fixed, we need

not discuss how far they are uniquely determined.

The transition

-> € is a linear operation and hence in terms of any basis

@,, ..., @, expressed by a matrix

J=(tull,

*#=J*

where

o,=

Xin

Oj.

(11)

As J? is unity the determinant | /| = +1. In the following we shall have to make use of two bases of §, only: the

fixed

order

normal

{3}.

argument

basis wo,

fei Qa. and a

For the remainder

refers to the normal

ments of Fx.

fixed minimal basis @,, ..., w, of the

of this section and for the next

basis,

and small

Greek

section our

letters denote ele-

240

a is symmetric if a =a,

form linear subspaces of §,,

skew if

« = —a. The symmetric

and the skew «

the dimensionalities of which we denote by gt

and g-. Since every element « is the unique sum of a symmetric and a skew

element,

g++ g- =g. gt is at least 1 since ¢ is symmetric.

y > 0 and y = 0 indicate for a symmetric y that the representing symmetric matrix G(= G’) is > 0 or = 0 respectively. tr (En) = x’T) y is a symmetric bilinear form because the conjugate of

Enis n€.

Lemma 3. 1. The quadratic form T,[x] = x’ T, x = tr (& €) is positive. This follows at once from 1.1. We write tr (£ &) = (abs &)? and suppose y to be symmetric. The other lemmas in §1 then give rise to the following propositions:

Lemma 3.2.

abs (§ + ) S abs é + abs n, abs (§ 9) S abs € - abs n.

Lemma 3.3. y =0

except for € = 0. Let 7’ 0

be the eigenvalues of the symmetric matrix

senting y, 7 = 7’ the lowest, 7* = y Lemma

3.4,

the highest.

tr (Ey &) lies between 7: tr (Gq €) and r* - tr iG é).

Lemma 3.5. Let y > 0. Then Nmy S Lemma 3.6. such that

(g-!: try).

Let y > 0 and suppose we have a positive constant

Nmy 2c (p>

3.7.

c 0 and suppose we have a number ¢ and a positive

constant c < 1 such that

Then

trySpgt,

Nmyac??.

bis) = BE

On... Ay aa

(=

Ass Oe

(=)

where } = b(c) and B = B(c) are determined as described in 1.7. In the interest of reasonably low estimates we add the following remark. Let /? be the rank of the division algebra § over its center. The decomposition of the matric algebra (%) in the field of all complex numbers shows that its generic element is a string of f equal matrices A diagonally arranged,

|

;

241

Hence the sequence of the eigenvalues 7’, ..., 7 of the matrix G representing a symmetric y consists of 4 sections each containing / equal numbers, g=fh. If this is taken into account, one may replace c by cll and g by h in the defi-

nitions of the constants e, b and B entering into Lemmas

§ 4.

Quadratic

3.6 and 3.7.

Forms. Jacobi Transformation

A quadratic form (in &,) is defined by an n-rowed square matrix

(12)

These I” form a linear space H of

Bn dimensions.

The

B= (By, ss En),

n(n — 1)

n(n+

et net

value

of a given

1)

2

gt

(n=) i

form

J” for

quadratic

j

(13) a given

TE] = MeDEY &

vector

(14)

is a symmetric element of §,. By a linear substitution

AL x UtesDes)

O67,

(6)

the form (12) passes into another form A’I'/. I’ is said to be positive if the symmetric matrix of degree N

G,, being the matrix of degree g representing y,, in terms of the normal basis w?. (This definition is suggested by the fact that it is the non-vanishing of the determinant (8) which characterizes a non-singular linear substitution.) In other

words

numbers

we

%,; (“=

assume

that

1,...,”;4=

for any

7 columns

x, = (aa)

ey

Ng)

of real

1, ..., g)

Giz] = > «,G,,%, > 0

(15)

242

except if (x1, ..., %») = (0, ..., 0). The positive I’ form an open convex cone H+ in H which is certainly not empty since the unit matrix ||¢,,|| lies in it. It would not do to define the positive character of the form I” by the

condition J[z] > 0 for x + 0. For then there would exist no positive forms at all (unless x is, like % itself, a division algebra). Indeed the inequality &y&> 0 implies Nm y- (Nm )? > 0, hence Nm & + 0, and thus the relation Evyé > 0 will never be satisfied for elements + 0 of &, the norm of which

vanishes.

Suppose J’ is positive in our sense. Then necessarily G,, >0, hence |G,,| +0

and %, = yy, > 0 has a reciprocal y,,’. Write

O12 = Min Yaa +) On = Mat Yan 1

= 81 + Oyoko

4

=X

to + Onda,

+ Dyoty +

+ Diy Xp-

We then have the parallel formulas {z]

= oy m4o+

DL 5

wy

n

21s;

Giz] =2K,z+

2

wy

ie ay,

yates

2

44 Ge X,-

This is the first step of JAcoB1’s transformation. The (7—1)-rowed

Ne

||

matrix

na)

is symmetric in the same sense as J” is. If one chooses x5, ..., x, arbitrarily but determines x, by

2 = % + Dyk, + ++ + DinXn = 0 one sees that the new form

n

DiziGa ae

Hy

2

is positive and hence by our definition J*®) > 0. This makes the inductive

continuance of JAcosi’s process possible, and we end up with a formula

Px}

=

mt

+

Cy mn on

(16)

where each x is symmetric and positive and

Note the parallel formulae

fu Sut 2 OS, ve

Giz] =z K,z,+-++2/ feared

Dp

Bee

(17) K, 2,

243 If Nm I’ designates the determinant |G|, we therefore have

|G| =|K,|...|K,|

or

NmI'=Nmx,...Nm xp.

It follows from (16) that I’[x] = 0. But more important for us is the fact that the number

tp[z] = tr (P[z]) = Y tr Er §)

(18)

is positive except for x = 0, as the same expression (16) and Lemma 3.3 show. t,{x] is thus a positive quadratic form in the ordinary sense in the vector space S"/%q or SX/K. It is on this form and not on G[x] that we must base the theory of reduction. The relation

proves by Lemma 3.3 that thoes

Line

(19)

All positive quadratic forms J’ in § are equivalent to the unit form E: Lemma 4.1. J" is positive if and only if it is expressible as A’A by means of a non-singular

(For n= 1: y > 0 if and only if y= aa, Nma + 0.) Proof. It is clear that A’A = I is positive provided A is non-singular; for then G = A’A where

(20) Vice versa, let I” be positive and denote the (positive) eigenvalues of G by

7, .... 7) and their diagonal matrix by R. Then there exists an orthogonal U such that G=U'RU=U"RU. Let R1/? be the diagonal matrix of the positive square roots Fue yale...

A=

Then A is symmetric and

UlR!? U—

Ue Rieu.

G=A4°=A'A.

and

244 Suppose

” the sequence 7’, 7”, ... contains m distinct numbers

7,, 7%, -.. LA-

degree

GRANGE’s formula of interpolation gives a polynomial f(x) of formal with real coefficients such that

m—1

tr) = 12", .-

‘Hn)an?,

But then /(R) = R12, (G) = A. The matrix A = /(I’) solves our problem.

The lemma shows that the question whether J” is symmetric and positive

can

be

decided

once

the

is given;

§ >é

conjugation

the

answer

does

not

depend on the choice of the normal basis in terms of which this process appears as transposition X > X’, We can now finally dismiss the normal basis, and

from here on it is the minimal basis w, ..., @, of {§} by which we determine the representing column x and matrix X of any element & of § or Gx.

Cc. AN OLD SOME

(WITH

STORY

MINOR

§ 5. Minkowski’s

RETOLD

ADAPTATIONS)

Fundamental

Inequality

Express the symmetric bilinear form

tr (E 9) = x Toy in terms of that basis. We #9, = tr (@; @,) as follows.

compute

the determinant

d of its coefficients

If J represents the operation & > & in terms of the

basis w,, (11), then

tr (@; @,) = SY fy tr (1 ox)

7

or hence

d is necessarily

| tr (@; @,) | =f’:

| tr (@; ox) l ,

d= + |tr (w, @,)|. positive.

tr (w,; w,)

are the coefficients of the bilinear form

tr (& y), which is also symmetric because tr (XY)

depends symmetrically on

the two matrices X, Y. Thus we find that d is the absolute value of the deter-

minant |tr (w; @,)|. This absolute value is independent of the choice of the minimal basis w,, ..., w, of {%} and is therefore known as the discriminant

of {3}- Computation of tr (w; w,) by means of the basis «, itself shows that these coefficients and therefore d are rational integers. The non-degeneracy of

the symmetric bilinear form tr (6 4) implied by d + 0 is an important fact which concerns the division algebra § over & (though our proof passes through &x by means of the conjugation & > ).

245

Lemma 5.1. The determinant d of the quadratic form

tr (€4)= 2’ Ty x= 7, |x] equals the discriminant of {i &} and is a positive rational integer.

Since the conjugate of fy 7 is ny €, provided y is symmetric, tr (Ey ) =

x'Ty is a symmetric bilinear form. Its coefficients are #,, = tr (@; y @,). have

We

YO = DL) bie Or,

T = J) tr (0; @1) * ox» 7

tr (@, yw) Thus: Lemma 5.2. equals d- Nmy.

| tr (@;y The

,)| = | tr (@; @,)| -|GI.

determinant

of

the quadratic

form

tr Ey &) = T,[x]

In terms of the fixed minimal basis w,, ..., @, of the order {%} we express

each component €, of a vector x = (&,, ..., €,) by the column of its coordinates

Suir

i =1,

&4 = 2; X,40;,

and

now

..., g) as coordinates

use the N = ng quantities x,;

of x;

(u=1,

..., n;

they follow one another in the order wi =

11, ..., 1g; 21, ..., 2g; ... The Jacosi transformation (17) then appears as a linear transformation 4,=%,+ 3; D,,%, (21) vp

which connects the coordinates z,; with x,; and has the triangular matrix

|:

(22)

Hence (16) and Lemma 5.2 prove that the determinant of the quadratic form t(x] = t,[x] of the variables x,; equals d"-Nm

x,...Nm

x,.

A lattice 2 is given, and {§} with the minimal basis w,, ..., w, is the order of its multipliers. Since for any positive ¢ there is only a finite number of lattice vectors x for which [x] «, & a» is the identity, therefore

Oy &p = € OF &, = a. Hence all elements of g,, are of the form {a, ..., a} and & +o &« is the identity. But then (42) is the identity.—In its influence upon

Vig the group Q, is actually g,/g,. We endow the g-dimensional space of the variable ‘point’ = y,. with a metric by means of the positive form tr (G Se).

The operations of g,/g, are linear metric-preserving of the &-space.

In familiar

fashion

this finite group as follows. We

h+1

we

construct

(‘orthogonal’)

a fundamental

mappings

domain

for

choose a point € = m, which is carried into

distinct points a, 2, ..., 2, by the h+1 operations of g,/g, and set up

the A linear inequalities expressing that the variable point & lies at least as

near to m) as to My, ..., Mp:

1(é) = tr{(%—%,) SO

(Y=1,...,2).

By adding these h inequalities /(y,,) = 0 to the ones L(I’) = 0 defining Z,, we obtain a convex part Z{ of Z which is invariant under the whose i+ 1 images generated by the operations of g,/g cover Z) and overlappings. We then carry out the same construction respect to the group g,/g3, -.-, for y,,, with respect to the group gp Thus by a number of linear inequalities

By

= 0,

My, Ol

group g, but without gaps for y,; with _3/Gn=Gn-1-

=A ah

each concerning only one coefficient of I’, a convex part Z¥ of Z, is constructed,

the images of which by the mappings s of g cover Z, without gaps and over-

lappings. Denote the corresponding part of Z(d,, ..., Dn) by Z*(d,,

...,D,):

255 an assembly of such Z°(d,,

..., D,,) in which each color is represented by one

member constitutes a fundamental domain for the lattice group. (In the case n = 1, one has of course to proceed in the same manner in the g+-dimensional space of a symmetric variable y = y,, rather than in the g-dimensional spaces OF Vigy -++, Yin) § 8. The Third

Theorem

of Finiteness

To make sure that the pattern of cells shows no inner clustering in H+ it

is not sufficient, cell borders

as MINKOWSKI

on not more

seems

than a

to have

finite number

believed, to prove that each

of neighbors.

Rather,

one has

to introduce a variable subregion H, of H+ depending on a real parameter t > 0 in such manner that it grows as ¢ increases and sweeps over the whole region H+ as ¢ tends to infinity, and then to prove that there is only a finite number of lattice substitutions s carrying a given cell Z = 7(d;, ..., D,) into cells Z* which have points in common with H,. In analyzing Minkowskt's proof I came to adopt the following definition of the expanding subregion. Given p = 1, w > 0 and a semi-basis q, ..., ¢,

of Ul, we say that the positive form J’ lies in Ae

if

t,[z]

2p:

ty[c,]

n and every vector x that is in U but not in [¢,..., Cm—al,

for every m = 1,..., and if moreover

for

wooy aly Oh)

tr[G_ — 0) = trl(m] — @- tle]

4 < m whenever »

is in Y and in [¢,, ..., ¢,]. While

and w increase to

infinity, the set Z(c,, ..., ¢,|p, w) grows, and any given point I’ of H* will

finally come to lie in it. Instead of p and w one could of course introduce a lot more parameters, but could also reduce the two parameters to one, for instance by setting p = exp w; it makes little difference either way. The

of Discontinuity.

Theorem

substitutions s carrying a given cell

points in common with

i

p,w

Z(G

eG

There is only a finite number

Z=Z(d,,

..., D,)

2) WZ

Pp, w

of lattice

into cells Z* that have

gO

From the beginning we may assume d, = e,. Then % coincides with 2 and Z(b,, ..., D,) with Z). The image Z5 has points in common with H, » if

HE wy + 0. Let s-} carry G, ..., Cy into ef, ..., ef, and I” be a common Zo point of Z, and Z(e%, ..., e%|p,w). Then the ef are vectors in 2. Write tr[€m] = tm, Er[e*] = tm. The form I is £-reduced whereas

t,[z] 2 P1- trlen]

whenever x is in 2 and outside [ef, ..., @m-1], and t,[e*, — ] = t,[e%] — w- t,[er]

(43)

256

whenever 1 is in 2 and in [e*, ..., e*] (u K

G[z] = x Chee: Given a positive number #, let us say that G belongs to the set §, if

261

We have constructed a number a which depends on & only, such that for every L-reduced form I’ the corresponding ty belongs to R,. The

quadratic forms of N real variables

x,, ..., %y with real coefficients

form a linear space ® in which the positive ones form an open convex cone Rt.

In SIEGEL’s conception the theorem of discontinuity deals with this space ® of dimensionality N(N +1)/2 rather than with the space of quadratic forms I’ in §,x of dimensionality g,,. It is clear that with ¢ increasing to infinity, , will exhaust R+, and the set H,-of all positive "for which the corresponding tp lies in R, will exhaust H+. A lattice transformation

S (Gy Sy)

(54)

6, = 3) O15,

> (G1, Ca)



when expressed in terms of the N coordinates x,; appears as the linear transformation A, (20); it has the property that the coefficients both of A and its inverse A~? are rational numbers with the common denominator 7. A general principle of SIEGEL’s!) asserts that, given a > 0 and ¢ > a, there exists but a finite number of transformations A of this character which carry R, into sets that have points in common with ®,. This principle, which is a very powerful

tool in all investigations concerning

quadratic forms, including the

indefinite ones, permits him to transfer the problem of the discontinuity of the lattice group from H to 8. Compared to MinKowsk1’s approach which the previous section followed in its outline, this method has the disadvantage of yielding undesirably high estimates for the number of such images Z} of Z, as may be expected to have points in common with H;. But it recommends itself by the generality of the underlying principle. The lattice group consists of certain linear substitutions (54) and is, therefore,

contained

in the

continuous

group

W

of all non-singular

linear

sub-

stitutions A = ||«,,||, ,, © &x- Consider continuous representations of W which

become discontinuous under restriction of the variable element s of W to the lattice group, in the sense that in the representation space no set of points which are equivalent under this group has an accumulation point. As the Theorem of Discontinuity in either of its two forms proves, the representation

[+I

= I'(A-1] in the space H* of all positive quadratic forms I” is of this

discontinuous nature. S1rGEL devotes the major part of his paper?) to developing a general principle from which it follows that among all representations of

such nature our J’ > J is the ‘most compact’ and therefore of least dimension.

I shall not deal here with this side of the problem of reduction. § 10.

Let

Volume

of the Fundamental

us consider that portion

of the pyramid

I’ =||y,,|| whose points I’ satisfy the condition

NmI'S1.

Domain

Z, of the L-reduced

positive

(55)

1) C. L, Stecex, Abh. Math. Sem. Hansischen Univ. 13, 217 (1940), Satz 3. 2) C. L. StecEL, Discontinuous Groups, Ann. Math. 44, 674-684 (1943). Cf. also M. E1curer, Comm. Math. Helv. 11, 253-272 (1938/39).

262

Using the Jacosi transformation of the coefficients y,, into x,, dy, (“ < »)

SIEGEL proved?) that this portion of Z, has a finite volume V in the g,-dimensional linear space H. I describe here an alternative procedure which operates directly with the y,, and leads to simpler estimates. Dealing first with the lateral y,, (u < v) set for a moment — yj) y,,=

By, (uw < v) and take By, = 8 as an example. Write y,,,, = y,. Choose any element & of {§} and apply (38) to the lattice vector x = (f,, ..., &,) of which all components &, vanish,

except €, = ¢, &, = e. One finds

tr Er0€ + 705 + Y¥526) 20

(Y52 = Yes)

or

tr{(B — é) y.(B-&)} = tr (B 2A). Determine & in {§} so that

(56)

8 — £ = , is reduced mod {%}. We make use of

the upper bound furnished by Lemma 3.5, but replace the largest eigenvalue v* by the sum tr y of all eigenvalues:

tr (Bo v2 Ao) Str ya: tr (Bo Bo) So? trys Hence (56) implies

WG

ne Sein, (ha a),

or more generally, for uw < 1,

Ee

es 0 tle

(57)

For a fixed y, we extend the integration with respect to the g real coefficients of the variable y,, over the entire ellipsoid (57). Since by Lemma 5.2 d-(Nmy,)~1 is the determinant of the quadratic form

tr Ey,1 4) = Ta[x] we find for the volume

V,, of this ellipsoid

Le = v2 a-'Nm y,: 0? (tr Yy) or, because of Nm y, < (g-}- tr wo

VS

where

hay

k(try,)? =k uy

(ay?

In view of (25, and (55) it remains to integrate pr (n-V2 1) C. L. StecEL, Discontinuous

Groups,

Ge

tae

Ann. Math.

5%, 1)? 44, 688

(1943).

263 with respect to

the

mg+

coordinates

of

Vir +++) Yn Over the region described by

the

variable

Oe lactate

symmetric |

SIS

(M, =g-¢,1"9) n

|

positive

E

(58)

Let v,/g+ be the volume in the space of the gt coefficients of an arbitrary symmetric element y of the bounded portion described by y > 0, try 0,

tStrySt+d

is v, ¢7—1 dt, and we obtain as an upper bound for V the integral of Br -1/2 m-v: un ue a Deeg ee 4)

= go

extended with respect to the real variables ¢,, ..., t, over (58). It is only at this last quite elementary step that we are dealing with a non-bounded domain.

(58) implies

A

conti

SEL 2

testo

tia

tee

test, = Me

ie

S My,

te

0

mioG|...|

mr, oC.

(Here some of the multiplicities m;* may be zero; this will happen if the representation is not faithful and hence the representing matric algebra €* is of lower order than ©.)

350

Any antisymmetric n-dimensional tensor 7 of rank / is completely characterized by its components (i...) with 4

as the limit

(n= 1,2,---).

0.

+ gals) ¢1(s),-*

as the

eigenkernel

Ga(t)

+, ¢n(s)

form

a unitary

basis for the eigenfunctions of H(s,t) that belong to the eigenvalue y. In this construction the invariance property (2.2) plays no réle. However

the latter is decisive for the fact that the h functions $,(s) sponding

vectors

g,—= fo, constitute

In carrying out this program

(2.4)

Deg) SABE i,

(m,n21,

0Sl I fa (4) Il? and then by integration over s and f, Tocmensty S Tomer? Ponst-

In

(2.5)

we distinguish the two cases

Set in the first case ntl=2v+1.

(i)

n —1 = 2u,

(i) n +1

even;

(ii) n +1 odd.

n+] = 2v, in the second n—1 = 2u + 1,

Tr= f Huw(s,8) = ff Hu(s, t)Ho(t, 8),

hence application of (C,) result

to integrals over the pair

(s,t)

gives the desired

Pou Tao. |? = s) ?2 ff | Holt Hu(s,t)| a| ] [Pff

372

(ii)

Apply

(Cs)

to

Pu f Huser (5,8) = f (fu(8), fo(s))

with the result

7



Taos

[Pn]? f Ufu(s) 17> fll fo(s) ll * = Pou &

Only the case 11

of (2.5)

If f40

T.— ff

also

and

||? >0 T,=|/f

then

(1=2.3,°):

bee

ee

Te

(2.9)

will be used,

[for

| (sf, tf)|? >0

st

The otherwise (sf,éf)=0 for all s and ¢, in particular (f,f)—0]. inequality (2.9) then shows that one after the other of the traces Ty,T4,° - (n=1,2,:

> +) is an increasing

(2. 10)

Qn™ FS Qnnst

precisely we find that gn =Tnu/Ta

More

are likewise positive and not zero.

while

sequence,

shows that

(2.4)

+ * Anima = Tnsm/Tn S Tm,

or that the sequence gn never grows beyond T,/”, in particular not beyond T, 1. Hence it converges to a positive number inequality qi = y or

(2.11)

Note in particular the

T, S yf.

It is even easy to make

Considering S qn"

y.

that

qn

an explicit estimate of the rapidity of convergence.

does

not

grow

one finds that gn from

beyond

T,/™

and

IPm—=Tigi?

+ dma

n = m on can not grow by more than

gn™21™(1 — gul) sf or agu= = ovp(@) + gv.

This equation

(2. 16)

8gu= Z ovn(s) «gv

shows that the g, form

an invariant set {gy}

belonging to the

unitary) representation Q: s—> Q(s) = | wpr(s)||.

(necessarily

Let us say that h vectors gy

(»=1,:--+,h) transform according to © and call (g1,- + +, gn) briefly an Q-row, if the equations (2.16) hold for every group element s. For any such row we infer from (2.16):

(2.17) (f, gv)

(sf, 9n) = (f, 8*9n) = > ovn(s*) > (f, gv). is the

conjugate

of the

Fourier

coefficient

a

= (gv,f).

Moreover

the matrix | opy(s)|| is reciprocal to | opv(s){| and the latter is unitary, hence wvp(s*) = Gpr(s), whereby (2.17) turns into

(2. 18)

(sf, 9u) — Ddp0(s)«a.

For two Q-rows g, g’ one gets

= (Gu sf) > (sf, Iu) =D end, “



and hence by integration over s:

(2.19)

p2 (9 B9u) = pe © ny.

The special set {gu} constructed above satisfies the relation

(9u, 89u) =y,

(2. 20)

and

(2.19)

yields the important

hy =D anit. #

equation

(2.15), thus

375

Let us say that the invariant set {g,} occurs in f with the (non-vanishing)

weight

When

y=h* >| an|?. # we subtract

h

e= 2 age

jel

from f, the remainder {’ = f—e is orthogonal to the vectors gu. If it is still different from zero, we repeat for f” the process carried out before on f.

One gets a new eigenvalue y’ and again a corresponding set of eigenfunctions and

(ue =1,---,h’)

bw(s)

orthogonal

g’w.

vectors

to the gyz(s), the g’w to the gy.

The

¢’w(s)

The equation

tur

out

to

be

Ty = hy" + T’a

shows that I’,/y"— 0, hence y/ < y.

It is obvious how to continue the process. It yields a string of orthogonal functions ¢n(s) and orthogonal vectors gn (n= 1,2,-- -) subdivided into sections.

section

The

p,

m1). Thus

| H')(a;+s)|* > 8)? for se Cj, and consequently

Ne ia| everywhere on o.

A

(arts) |? > 68,7

By integration over s one finds

Ne: f | H)(s)|? > %)*, * Combination

with

Mm 0. Here A* isthe conjugate of the transpose

we may

course

see to it that, whenever

of our

construction

teaches

the

following

a primitive

belongs

9;,- - -,g, transforms

basis

two

things:

(1)

irreducible

unitary

(gus gv)

set obtained

in the

~Q,

its orthogonal

further

lemma

transform

representations,

then they

(ue Ve

If they transform according to the same

Thus

g’y

gy

If the

is unitary.]

Schur’s

to © itself.

(Gu gv) = 0 (3.1)

invariant

to a representation

according

according to two inequivalent are mutually orthogonal,

(2)

A/Vp

of A, and £ denotes the unit matrix.

the

and

ey eed)

irreducible unitary ©, then

=B 8p

(wv=1-

+h)

with a factor 8 independent of » and ». Any numerical multiple Ag = (Agi,* > *,Agn) of an Q-row g is an Q-row, and so is the sum of two such

rows.

Thus

the

Q-rows

form

a linear

introduce the factor B in the equation of the two Q-rows g and g’;

(3.1)

manifold.

It

is natural

as the scalar product

to

(g, g’)

(8,8) =h*- = (9m 9'n)Indeed it has all the formal properties of a scalar product, in particular (g,g) >0 unless g—0. Call the row a= (%,---,%) of the conjugates

om = (f, gu) of the Fourier coefficients the f-component of the Q-row g and define

(a’, a) ae

aa

al]?

= (aa).

Aa and a+ a’ are the f-components of Ag and g+g’. now reads

(3. 2)

The equation

(2.19)

(8, 68) =h>- (a’,a).

An Q-row will be said to be hidden

or flat if its f-component

a is zero,

and upright if it is perpendicular to all hidden Q-rows. Clearly there can not be more than h linearly independent upright Q-rows. For let g™,--+,g(™ be any m such rows and form the linear combination g—dg +--++ Ang’. Whenever its f-component Aa +--+ Anam) vanishes, g itself must be zero, because such a g is at the same time upright and

flat, therefore

(2.14)

(g,g) =0,

g =0.

shows that the operator § changes an Q-row g into an Q-row $g;

379

according to (3.2) this Gg is necessarily upright. Any Q-row g = {gy} obtained by our construction satisfies an equation (2.15) with a positive factor y, gu = y*- Sgn, and consequently these Q-rows themselves are upright ; among all possible Q-rows the construction automatically selects the upright ones as those that actually occur in the Parseval equation (2. 23). Such a principle of selection is needed, since the linear manifold of all Q-rows may

not be of finite nor even of denumerable dimensionality. Let g,---,g(™ (m= 0.

It is this normalized form which results from the

The greatest of the m numbers y1,° * -, ym can be defined

»()

of the Hermitian form

(3.3)

for >| &|? | e(Q) | * extending over any set of inequivalent irreducible unitary Q can not exceed | f ||? (Bessel’s inequality). It is thus further clear that the

construction of 2 can not help yielding a complete basis g of the upright

Q-rows

for each

Q.

Otherwise

the

sum

(i= 1,- - -,m) (2.23)

would

short of | f | ? at least by the contributions of the missing g‘).

We

fall

sum-

marize our findings in a preliminary Statement and a Theorem. Statement.

accomplishes

Starting

from

the following.

a

given

a.p.

vector

For every irreducible

and

m>0.

representation

basis g™,-

set of inequivalent

Q’s

--,g'™)

that actually

One has m Sh, and the vectors

gu

construction

Q

of

- *, gn) which are perpendicular to all hidden Q-rows,

it picks out a complete

for which

our

unitary

of o, degree h, it determines a complete orthogonal those Q-rows g = (91,°

f,

(=e ms je 1, =

occur,

i.e.

h)

lie in 3;°. [They are even of the special form S(s) -sf where $(s) is not only f-continuous but satisfies a strong Lipschitz condition with respect to f,

(2.1).]

Q

are

The gu

mutually

for two inequivalent

orthongonal.

The

irreducible unitary representations

contribution

of

Q

to

f

is

the

sum

(2) =Daygu formed by means of the Fourier coefficients a, = (gu, f), the contribution of 2 to | f |? equals | e(Q)|?= S| au | 2%

The maximum

weight y(Q)

of h*> | Bay #

Es |? for S| & |*S1 é

is introduced as the

with which Q occurs in f.

THeoreM. The sum of the contributions > | e(Q)||? extending over the denumerable sequence of inequivalent irreducible unitary Q actually occurring in f equals | f \|.°. An explicit estimate of the convergence can be given if the terms are arranged by descending weights y(Q). It is perhaps

host of “hidden”

not justified to speak

of a completeness

relation,

since a

invariant sets are left in the dark, for the good reason

that they do not contribute to f and

| f | 2.

381

For the “interpretation by scalar functions on IL” the situation is a little simpler. Here, according to an observation made by H. Cartan [8, ef. also

11], the number

of linearly independent

any irreducible unitary representation basis

g@,---,g'™

for them

of degree h.

in order

is at most h for

We choose an orthogonal

to determine

to f, without rejecting the hidden ones. picking out the denumerable

Q-rows the

contribution

e(Q)

But even then f must be used for

sequence of ’s that actually contribute to f.

A subgroup o° of o may be treated directly by observing that f is also almost

periodic

with

respect

to

o°.

This

is

a better

procedure

than

by

breaking up the o-invariant sets obtained by our construction into primitive o°-inyariant sets. For then one would still face the task of getting rid of all but the upright

In

the

ones.

Main

Theorem

one

can

readily

pass

vector f to a finite number,

or even a denumerable

fu fa’:

is f-hidden

+.

An

Q-row

g’

when

from

a single

sequence,

(g’4,f)—0

given

a. p.

of such vectors,

(«#—1,---,h).

For any v=1,2,--- we construct an orthogonal basis g(”1),- - -, g(™) for those Q-rows g which are perpendicular to all f,-hidden Q-rows g’, (g’, g) =0, and moreover satisfy the relations (fis Ju)

Of course my +5 (fos

(finite or infinite) .

Sem

s

gen,-

sequence .

For any vector f we may

== (gu, f) and form

| eo(F 32)? =D | an

(u=1,---,h).

= 0

gu)

° : all vanish then Q “ does not contribute.”

If m1, m2, msz,°

the whole Boa)

ap)

= 05°

| 2

of Q-rows

=, gio);

determine

a:

the Fourier

coefficients

(w=1,--+,h; i=1,-- +, my).

The contribution of © to | f ||? is defined by

| e(f;2)?—=la(fsQ)l?+le(fs@leto--,

(3. 5)

and

Sl es) 1S

(3. 6)

FI

the sum extending over a complete set of inequivalent contributing irreducible unitary @. For ff» the sum (3.5) is finite, | eu(fo;92)||* equaling zero for

u=v+1,v+2,:--,

corresponding

equation.

and the Bessel inequality

As a result, an orthogonal

(3.6)

changes into the

sequence

91, g2,° °° of

vectors, subdivided into sections that form primitive invariant sets, has been found such that the Fourier series 019:+ @g2-+-~ ~~ with the coefficients

382



= (gn,f)

converges

f

to

weakly

for

of

each

the

vectors

a.p.

given

fm fas Jap

Interpretation by vector

Strong approximation.

4.

A further axiom connecting “length”

|g| with “modulus”

for the transition from weak to strong convergence.

ge= f €(s) -sg in which

é(s)

is any f-continuous

on II.

functions

| g | is needed

Let us study the integral

numerical

function and

the vector g of such nature that sg is an f-continuous vector function In the interpretation by scalar functions on II this relation reads

of s.

ge(P) = f €(s) -g(s7P), 3

and thus

| ge(P)

2

f 1é(s)1?f | g(srP)|*.

The second factor on the right is independent of P and has been denoted by f |g(P)|?=l19 |? Tf | |]? stands for ff | é(s)| *, we therefore arrive P * at the inequality

Igelee len st: lge(P)lSlNel-lglThe special case suggests the following general Axiom g=

IV.

Under

the

fi &(s) + sg satisfies the

(4.1)

conditions

axiom:

specified

above

the

integral

inequality

lg lSlél-lg-

I do not deny its arbitrary character; one would wish to reduce it to some simpler assumptions. But two things can be said for it: (1) It holds in any metric vector space & provided one identifies | g| with | g ||. Indeed for g(s) =sg the inequality (C.) gives

If és) sg? Sel?

f iso l?=éell? lol’

(2) It holds for the interpretation by scalar functions on IL. The following theorem &(s)

being

y= f Es) sf,

is an immediate

an f-continuous LMmIn

function,

consequence the

Fourier

an= (Gus ¥)>

of the new series

of the

axiom. vector

383

converges strongly towards y.

In the sum

> members

not be separated.

of the same

set may

i

Let us indicate the contribution of the first p invariant sets, ete+t+--+-+ep4, by {e+-- -}p and carry out our calculations for one of the sets, to which

h,

Jus

the old notations

On,

may refer.

We have

(4.2)

se = Dau a

therefore

> oy gu,

op (s)

o

(4.3)

(#y=1,°-°,h)

sgn —= ZX Mp op (s) gv => (Gv, Sf) * Jv, by v f &(s) -se= = (99)

and

where

=

gv = Lng

y= (Sg too det Sele) 9 e

f'”)

se shows

8

denotes the remainder

f—{e-+---},.

The

formula

(4.2)

for

that

|se—te] = Z| (gv, sf —tf)| lov]= l sf—#f | Sl gv] and thus with

peg! RON] eet [| gray | A=1+{S|g|+---:}s

Thus

sf

is

an

f-continuous

function of s, and our axiom yields for the remainder in (4.3)

| f &s) sf But one knows that THEOREM

| f'”) | ]

OF STRONG

| STE

17

f can

the estimate:

|.

= 6, tends to zero with p—

APPROXIMATION.

vector

o.

be strongly approximated

n

with arbitrary accuracy by finite sums of the form > angie The

numerical

function

| p(s) —p(t)|S|sf—tf|.

|[sf—f|—p(s)

Choose

any

«>0

kal

is

and

f-continuous

let

I(p)

be

since

a non-

negative uniformly continuous function of the variable p (= 0) which vanishes for p=e and for which the integral of the f-continuous function I(p(s)) =(s) is 1. Form the difference

J A(s) sf —f= f(s) 3f—f).

384


0 the group o may be covered by a finite number of such

f-disks D;(bi3«) (kK—=1,-+-+,n) of radius « Choose a definite point Po. For an f satisfying this condition the numerical function | f(¢*Po)|| of ¢

385

and hence the function of f by

| f(2)|| of P is bounded. [Alla

This

length

is invariant with

b

respect

We define the length | f|

F Ce).

to all transformations

|(s,t)f | =| f |. im particular to all transformations

two vector functions f(P), g(P)

w(P) = (g(P).f(P))

2. around

of oxo,

(s,s) of o.

For any

satisfying our condition the scalar product

is an almost periodic numerical function on IL, and we

may therefore form the constant f y(t"P) = f ¥(P) scalar product

(s,#)

(g,f).

Again

t

Pia

invariance prevails:

and introduce it as the

((s, t)g, (s, 4)f) = (g, f)-

Let us say that the element s lies in the f-circle C;(a;e) a if | sf(P?) —af(P)||

Se

for all points P, and now

of radius ¢

assume

can be covered not only by a finite number of f-disks D;(bg3«) (¢ =1,but

also

by

a finite

number

trarily small radius «.

of f-circles

C;(ai;«)

If s lies in C;(a;;¢)

| sf(P)

(i=1,:--,m)

and ¢ in D;(bx;)

—aif (bP) |

S| sf((P) — af (AP) + | wf)

that o

+ +, 7), of arbi-

we have

— af (OP) I

Set [FO P) f(bP) || S ee,

or

|(s,t)f— (ai, bi) f

the group

of the pairs

«xo

(s,¢)

periodic

f is almost

Hence

|S 2.

and

a fortiori with

with

respect

respect to to the

sub-

group o of the diagonal pairs (s,s). All the axioms I —III are satisfied for the space > of the almost periodic and to «.

vector functions f, both with respect to «xo

introduced

in this section holds.

Indeed consider

Also the Axiom

IV

ge(P) = ff ls, 8) -sg(MP). One finds by applying

(C2) to the (s,?)-means,

hge(PIPSSS

at

LEGO? SS I sgP) yy? a

t

The second factor on the right equals

P)le—=f lgPle—lel?

Sige t

Hence

|ge|

ip

=| £|- | gl].

value of the type

The same

argument

applies to a simple mean

g'n(P) = f als) -sg(s7P).

386

On Maak’s approach to the theory of almost periodic

5. Appendix. functions.

+, dn

{a} =ai:,°

objects

n distinct

Given

problem.

The marriage

1.

(boys) and n distinct objects {b} =0,,- + -,bn (girls) ; moreover a scheme of linkage Q, according to which an a; and a 0; are either linked (friends) JI call the number

or not linked.

A set B of

in a set its rank.

of elements

girls is said to be associate to a given set A of boys if no boy in A has friends

is the necessary and sufficient con-

Question: What

associate of {b} —B.)

is in the same sense an

{a}—A

(Then the complement

outside the set B.

dition that the boys can be paired with the girls in such a fashion that in each of the n pairs the partners are friends? The following basic condition

is obviously necessary: A set A of boys has never an associate set B of girls

of lesser rank than Proof.

[9, 4, 6].

is also sufficient

this condition

I let the

girls

asserts that

lemrna

combinatorial

fundamental

The

A.

b,,- --,b»

choose

their

partners

one

after

the

other, and therefore ask 6, first to make her choice. Suppose she chooses ay. The choice should be fair to herself, i.e. as and b, should be friends. But it

should also be fair to the other girls by not making it impossible for them to find partners among their friends, i.e. the linkage scheme Q,_, arising

from Qn by removing a; and 6, should still satisfy the basic condition. a set A B=

of boys

(1, big,*

+ +, bx,)

Then

the

set

which

does

distinguished

sets.

A

of rank

second

of

r (21)

distinguished

girls

that

is of

requires

that

postulate not

contain

dj,

or

a;

the

if it has

same

there must

rank

should be

in

an

r and

be the

no

Call

associate

set

contains

}.

distinguished

intersection

of

all

We make the basic observation that the intersection of two distinguished

sets is again distinguished and is not empty.

Indeed let A; be a distinguished

set of rank r; and B; an associate of the same rank A=A,f[] A. be of rank r and B—B,f)} Bs of rank s.

(t=—1,2). Let The rank s is at

least

r = (cxb)

389 of $(s)

and ¢(s)

= (sd)

differ by not more

of f o(sb) mean

than p, (5.3).

and f ¢(s)

values

must

INSTITUTE

oscillate by less than p.

Among

themselves they

This proves that the absolute

difference

can not exceed 3p, and as p is arbitrary, these two

coincide.

FoR ADVANCED

Srupy.

REFERENCES.

[1]

S.

Bochner

and

J. von

Neumann,

“Almost

periodic

functions

in groups,”

Transactions of the American Mathematical Society, vol. 37 (1935), pp. 21-50. {2] H. Bohr, “Zur Theorie der fastperiodischen Funktionen I,’ Acta

matica, vol. 45 (1925), pp. 29-127. [3] espace de

E. Cartan, “Sur la détermination d’un systeme orthogonal complet dans un Riemann symétrique clos,” Rendiconti del Circolo Matematico di Palermo,

vol. 53 (1929), pp. 217-252.

[4] P. Hall, “On representatives of subsets,” matical Society, vol. 10 (1935), pp. 26-29.

[5]

Anna

Hurevitsch,

“ Unitary

[6]

“ Abstrakte

W.

Maak,

“Eine

fastperiodische

neue

The Journal

representations

topological group,” Recueil Mathématique

pp. 79-86.

Mathe-

in Hilbert

space

Mathe-

of a compact

(Matematicheskii Sbornik), vol. 13 (1943),

Definition

Funktionen,”

of the London

der

fastperiodischen

Abhandlungen

aus

dem

Funktionen”

and

Mathematischen

Seminar der Hamburgischen Universitat, vol. 11 (1935), pp. 240-244, and vol. 11 (1936),

pp. 367-380. [7]

J. von

Neumann,

“ Almost

periodic

functions

in a group

I,”

Transactions

of the American Mathematical Society, vol. 36 (1934), pp. 445-492. [8] F. Peter and H. Weyl, “Die Vollstindigkeit der primitiven Darstellungen einer geschlossenen kontinuierlichen Gruppe,” Mathematische Annalen, vol. 97 (1927),

pp. 737-7 von

R. Rado, “ Bemerkungen zur Kombinatorik im Anschluss an Untersuchungen [9] Herrn D. Konig,” Sitzungsberichte der Preussischen Akademie der Wissenschaft,

vol. 32 (1933), p. 68. [10]

H.

Weyl,

[11]

H.

Weyl,

“Integralgleichungen

und

matische Annalen, vol. 97 (1927), pp. 338-356. vol. 35

“Harmonics

(1934), pp. 486-499.

on

homogeneous

fastperiodische manifolds,’

Funktionen,”

Annals

Mathe-

of Mathematics,

146. Inequalities between the two kinds of eigenvalues of a linear transformation Proceedings of the National Academy of Sciences of the United States of America 35, 408—411

(1949)

With a linear transformation A in an n-dimensional vector space (matrix

consisting of » X complex numbers a;;;) there are connected two kinds of eigenvalues: the roots = a, ..., a, of the characteristic polynomial |2E =— A| of A (E = unit matrix) and the roots = m, ..., kn of |2E _ R|

where K is the Hermitian matrix 4*4A composed of A and its Hermitian conjugate A*. The x; are non-negative, and one would naturally compare the \, =

| au| ? with the x;

If A is normal, A*A

in general, however, they do not. ee

they coincide;

Arrange the x as well as the ) in de-

scending order,

ize

= AA*,

oan Se

Nic

Ka shee

I shall prove the following TueoreM. Let y(d) be an increasing function of the positive argument d, e(A) = o(d’) for X> X’ > 0, such that o(e) is a convex function of § and ¢(0) = lim g(\) = 0. Then the eigenvalues \, and x, in descending order a7>0

satisfy the inequalities

eu) +... + Qn) S ela) +... + e(km) in particular

WWE Soo NVeSiGY ar oo ae ie

(m= 1,2,...,0), Ge = SP hat)

| (e))

for any real exponent s > 0.

According to a familiar argument! AS

Indeed the equation Ax

aa, a*A*

=

=

(3)

ax has a vector solution x =

aa*, hence a*A*da

a*Ka

mM.

=

= ),(a*a),

&0,(a*a)

eS

or

a Sa

ue

OY,

Since every vector satisfies the inequality «*Kx < k(x*x), (3) follows.

The linear vector transformation A induces certain linear trans formations AM, AP, AB, |... AM for the space elements (skew-symmet ric

391

tensors) of rank 1, 2,3,

...,.

af

3

For instance All ;

=

\|oJ9,|| is given by

itt, Vir, Giy

1 = | Getty Qixty ev Qiu, Ax, Guy

where J and J’ range over the triples (7, k, 1) and (2’, k’, l’) with the re-

strictions 7 < k < 1, i’ < k’ < I’, respectively. Application of the inequality (3) to these matrices AM, A@, ... yields the relations RESTING IE MAS SUKI Ko

Ware

AT

eA

eSe Rie

(with the equality sign prevailing in the last of them). be settled as soon as I prove the following Lema:

Let «;,\; (¢ =

ies

(4)

Everything will

1, ..., m) be non-negative numbers such that

NN

ee

Nr

(5)

and INGE

TS

OIE

ey

GNSS TA

Meee

(6)

then

Dw)

S Lek)

G=1,...,m)

(7)

for any function ¢ of the nature described in the Theorem. Of B>

two a.

real numbers a, 6 let max.(a, 8) denote a if a> B and B With a variable argument z > 0 form the functions

f(z)

= Tl pac)

Then

and

g(z) =

is max.(1, 4).

(2)

g(2) S f(z) for 22 0.

Indeed

if

set

edz)

= lforz

= 0.and.2(z)

=.

and distinguish the intervals {0}, i, Nee

Then g(z)

leo

gL ENDS,

= g:(z) for z in {i}.

sere

.:. dw

for

7=1,...,m

oun {m = 1) {m} as defined by A 1s a LA

pA ge

But, because of (6), gi(z) < fils) < f(z),

hence (8) holds in each of the m + 1 intervals. With

an increasing function ¥(z) one can form

the Stieltjes integral

So log g(z)-dv(2) = Dw), provided f-” log z.dp(z) converges. Here

e(\) = J,” log max. (1, z)-dy(z) = Sis > 1 log (Az)-db(z).

(9)

(10)

392

It is clear how

formula

the corresponding

of (9) and

(8) by means

for

a (fz) leads to (7). Set \ = ef. If (A) = G(€) is a given function satisfying the conditions of the Theorem, it can be expressed by means of a non-decreasing function G'(t) in the form

G@) = ff, Gd

(1)

= — JZ, b — dG".

(The integration per partes is justified since

-1.G'() S$ 2. f'? G'(n)-dr (10) goes over into (11) by the substitu-

converges to zero fort > —.) tion z = e-', ¥(z) = —G’(t).

Of the inequalities

m

=

(2) thus proved,

important

the most

is the last

of any arrangement of the x; and )j,

n, which is independent

WEEE coe SPINES USP ons SEE, ... yields the further relations

Its application to AW, A",

NE

nies,

ii 0,b > 0 (a < 0,6 < 0).

This follows from assumption (I’). value.

Lemma 2.

If S’ = 0 for a certain value of t, then dS'/dt

Proof. Substitute the values of a and b derived from and use the inequality (II’).

Lemma3.

< 0for the same

S’ =

0 into dS’/dt

If S’ < 0 for t = 0, then S’ < Ofort > 0.

Proof. First assume S’ < 0 fort = 0. Should S change sign as Z-travels along the ray, it would have to vanish somewhere. Suppose this occurs for the first time at t = t,. As S’ < 0 before ¢ reaches this value, S’ passes the zero level at ¢, ascending; hence

dS’/dt

>

0 for t =

t, .

But these two

relations

S’ = 0, dS’/dt > 0 for t = t, , contradict Lemma 2. If S’ = 0 for t = 0 we have dS’/dt < 0 for t = 0; consequently S’ becomes negative immediately after the point Z has started on its way from Zp , and from then on, as we have seen, S’ must remain negative.

We turn to III. Starting at a given point Z) = (po , 79) and always raising p, we follow the adiabatic through Z, : we thus obtain a continuous monotonely

407

descending function r(p), and according to II] we can make p travel over the entire infinite interval p > p).

While this happens, the directional coefficient

s = PP

(7)

of the straight line joining (pp , 79) with the point (p, 7) on the adiabatic S = So increases monotonely from a certain positive value m, to +.

m denotes

the adiabatic derivative — dp/dr and m, its value at the point Z) . Consequently, there is exactly one point on the upper branch p > pp of the adiabatic that lies on the ray from (po , 79) in a given direction s, provided s lies between My and

©.

The analytic proof for these intuitively evident statements runs as follows. Form the (non-total) differential

dr = (1) — 7) dp + (p — po) dr, We want to show that along the which vanishes along any ray through Z.. adiabatic dr is positive for a positive increment dp:

)

di

pte

d

=(m-—)+@-p)G>0

p>po-

for

By hypothesis II dR

i

(9)

Oh

dr

ew

for

p>

po,

and since R vanishes for p = po it must indeed be positive for p > po :

r= [ @- po Siar >0.

(10) Using (7) we may write

dr = (1) — 1)’ ds

(11)

and thus the monotone behavior of s is exhibited by the equation

= =R/iGs-

o>.

It is clear that s tends to m for p — po , and the combination of hypothesis HI with the trivial inequality s > (p — po)/t» leads to s > for p—@.

4, The

Fundamental

Inequality for the Direction

For the differential of

(12)

H(Z, Zo) = (e — ) — 2(p + Po)(t» — 7)

Zo Z:

408 we find by (1)

dH = de — 3(t) — 1) dp + 3(po + p) dr

= T dS — 3{(t> — 7) dp + (p — po) dr},

or

dH = TdS — 4dr,

(13) and hence along any ray

df = T ds.

(14)

Let Z) , Z, be two distinct points satisfying the Hugoniot equation H(Z, , Z) = 0. Because of the convexity of the region Z (hypothesis IV) we may join these two points by a straight segment lying in Z. Integrating (14) along it we obtain

HZ, ,2) = | Tas =0.

(15)

From this simple relation we are going to deduce

Theorem 1. For any two distinct states Zo , Z, linked by Hugoniot’s equation H = 0 the following inequalities hold

(16) (17)

W = (p: — po)(t. — 71) > 0, (Pi — Do) + Mo(t1 — 7) Before

proving

the theorem

> 0,

(i — Po) + mlm: — 1) 0, cf. (4). We now realize that we may omit the supplementary relation W > 0 because it follows, in fact in the sharper form W > 0, from the Hugoniot equation. The adiabatic derivative

Opaoages OE Re is the square of the ‘acoustic velocity” relations (17) and relation (4) give

magnitude of M.

Assume

7

>

7, .

Then

the two

— Po < m, my < M? = Pi

(18) or po Co < | M|

c.

ae L

< p, ¢,, thus confirming a statement made in Section 2 on the

In terms of the velocities wo , uw given by (2.1) or ps

To k=aeta

TO)

Ww

are

Di

ho

1

ice

409

(18) may be written as (19)

Wey ee

Co

| dal

ees

Relative to the shock front the flow in [0] is supersonic, in [1] subsonic.

Proceeding to the proof of our theorem, let us travel from ray which passes through Z, and hence set in Lemma 1,

a= Pi — fe,

b=

Z, along that

tp — %.

Were a > 0, b > 0, then S’ would be positive and hence S monotone increasing along the segment Z,Z, , which contradicts the relation (15). Consequently

this combination

(and in the same way the other, a < 0, b < 0) is ruled out,

and therefore ab < 0, or (16) must hold. Next consider

sie aegnn

gs

S'=a ap +6 ae

Were S’ < 0 for t = 0, then S’ would always be negative by Lemma 3 and hence S would monotonely decrease while Z travels along the straight segment from



Z, to Z, , in contradiction to equation (15). o(25)

+

0f%8)

Therefore > 0,

and this is equivalent to the first of the inequalities (17). The second follows from the first by interchanging Z, and Z, . But our argument shows much more. While Z travels along the ray from

Z, passing through Z, , S’ starts with a positive value; because of (15) it must

change sign before Z reaches Z, . But S’ remains negative from the moment it vanishes for the first time. Any rise and fall in H = H(Z, Z,) along the ray is coupled with the same in S by the relation dW = T dS. Hence H first rises monotonely to a positive maximum and then decreases, passing on the descent through 0 for Z = Z, . This description shows that H is positive from Z, to Z, and negative thereafter. In particular: On the ray from Z, through Z, the point Z = Z, is the only one aside from Z, itself which satisfies the Hugoniot equation H(Z, Z,) = 0. According to (16) either the inequalities p: > po , t > 71 OF Po > Pi, 7, > 7» hold for two distinct states Z) , Z, , satisfying the Hugoniot equation H(Z, , Z.) = 0. We shall indicate these alternatives by Z; > Z , Z, < Zo respectively. By the Hugoniot contour 3 (for a given Z) we understand the locus of all points Z; = (p, , 71) for which H(Z,, Zo)

= 0

and

Lie

Loe

Our It is quite essential for our argument to pick this upper branch Z, > Z. above result may then be stated thus: On the Hugoniot contour s = (p.: — po)/

410

my and

11) lies between

(to —

©; s is a uniformizing parameter

inasmuch as

in mind the value of s specifies uniquely the point Z,. But it must be borne of s > mo that so far we have not yet proved that to every preassigned value that there actually corresponds a point on the Hugoniot contour; we only know there cannot be more than one.

of the Hugoniot

and Parametrization

5. Entropy

Line

For a moment let us return to the upper branch @ of the adiabatic through

(p — Po)/(t — 7)

Z, , on which p increases monotonely from pp to + © and s = -Moving along @ we have by (13)

from m, to ».

dH

(20)

=

—3R

=

—tdr

dp

and hence (10) implies = H(Z, Z,) is negative along @.

H

Lemma 4.

Asa matter of fact, we even know that H is falling,

(9) falling with increasing rapidity, dR/dp

H(Z4,

>

0.

R > 0, and by condition

The explicit formula giving

Z) for any point Z, on the adiabatic @ is by (20), (8) and (9), DA

1

HZ, , %) = —4 | Rdp=—5R@-p)| Jo

PA

1

fr

aR

me) —3] ode @a-” Go

ar

ayes

=> 3 Po)(Pa — P)DP) Hi5 dp So or Sy < So according to whether Z, > Zo or Z, < Zo.

Theorem 3.

The Hugoniot contour is a simple line starting from Z) on which

s and S are monotone increasing

unknown destination). Proof.

Assume

(s traveling from m, to +,

the arrangement Z,

>

S from

S, to an

Z, for the two points Z) , Z,; linked

by the Hugoniot equation. Draw the ray ® from Z, passing through Z, ; its directional coefficient s = (p; — po)/(t. — 71) lies between m) and +, and hence ® meets the upper branch of the adiabatic line @ in a point Z, = (pa , Ta)

where, according to Lemma 4,

Hence Z on its road along the ray must have passed the point Z’ where it reaches its maximum,

and

running

downhill

have

passed

a point

Z, for which

H =

411

Zz. T dS = 0 before coming to Z,.

Its value S, at the point Z, of the Hugoniot

contour is therefore higher than its value Sp at Z4 (Theorem 2). Again, let s be any value between m and ». The ray going out from Z, in the direction s meets @ in a point Z, where H is negative. Hence H must

vanish at a certain point Z, between Z’ and Z, (where S is already on the down grade), and thus a point on the Hugoniot contour is obtained corresponding to this preassigned value of s = (p, — p)/(ro — 71). We know it is the only one. The question raised at the end of Section 4 is answered in the affirmative, and

we may now speak of the Hugoniot contour as a simple line Z, the uniformizing parameter s. Because the point Z,

=

=

&(s) with

®(s) is thought of as varying along the Hugoniot

line, let us drop the index 1. Along 3¢ we may use the differential relation dH = 0 or, according to (13) and (11), T dS — 3dr = T dS —

3 (1%) — 1)?ds = 0.

Because

OS. (Gua) Gh;~ Sie thus turns out to be positive, S increases all the way along the Hugoniot line with s. This gives a new proof for the inequality S, > S, by integration along that line,

8-5

=f

re mo

mens PSP as

(> m)

Thus Theorem 3 is proved, and Theorem 2 by even two different methods.

Part II. The Problem 6. Formulation

of the Shock

Layer

of the Problem

For the sake of brevity we now denote partial derivatives with respect to

(p, t) by subscripts as in

as

Beas _Eb

The rest of our argument will be based on the thermodynamical assumptions I-IV, Section 3, but in its course it will become clear that one more condition of highly plausible nature is to be added to our list: Ia. Heating temperature.

a quantity

of fluid at constant volume

raises its pressure

and

412

It must be remembered that the analytic formulation (I’), (II’) of the hyFor the previous inpotheses I and II depended on the fixation of a sign.

vestigation this did not matter greatly, but it is essential for the behavior of

the shock layer. We therefore require explicitly by the statement concerning pressure in Ia, that (dp/TdS8),-cone. > 0 or S, > 0. The other part of Ia

then adds the inequality 7,/S, > 0 or T, > 0. In a stationary field the state (Z, i) of a fluid of given thermodynamical

The nature depends on the spatial coordinates x, y, 2 only, and not on time. five equations which, in the absence of an external force, express conservation

of mass,

momentum

and

energy,

account by the flow of heat j =

Sa=pie- Sh,

take

—)grad

conductivity

heat

and

viscosity

into

7 and a stress tensor S of the form

Sh= al divas + a(S + 2)

(in the writing down of which we have for the moment adopted the subscript notation for the coordinates x; and velocity components u,). 4 = A[Z], uw, u’ are given functions of the thermodynamical state Z; \} > 0, nh > 0, uw’ + 3u> 0. As they are small we intend to let them pass to zero. In two ways such a passage to the limit \, u, u’ — 0 can give rise to a discontinuity in the plane x = 0: (A) Write ex for x and «A, eu, eu’ for A, uw, uw’, leaving all other quantities untouched, and then let the positive constant factor ¢ tend to zero (shock layer).

(B) Let ¢ tend to zero after writing ex, ew for z, u and €d, €y, eu’ for

d, uw, uw’ (slip layer). The different effect is clearly exemplified by the continuity equation

(21)

3 0) + 5p (o) + 2 (pw) = 0.

Substitution (A) changes it into

i (01) + € 5 (00) + € 5 (ous) = 0 which for « — 0 reduces to tc)

3q (Ph) = 9; whereas process

typical.

None

(B)

leaves the three-term

equation

(21)

unchanged.

but the derivatives 0/dx survive the limiting process

This

is

(A), the

others 0/dy, 0/dz disappear from the equations. The shock layer is thus seen to be a one-dimensional problem, no interaction taking place between the onedimensional fibers of space y = constant, 2 = constant. The slip layer on the

other hand constitutes a far more complicated three-dimensional problem, identical with that of Prandtl’s boundary layer. We are concerned here only with the shock layer.

413

With the abbreviation u* = yu’ + 2u, u* > 0, the limiting process (A) leads to the following equations: (22.1)

pu =

(22.2)

constant = M,

Mu + p — p* ae = constant,

(22.3)

Mv

(22.4)

- 4 2

9

= constant,

(v ant du + vet) = of

Mw-

yp aus dx

o + wt)

= ZF

where e* denotes the total energy e + 3(u’ + 0 + w*).

constant,

= constant,

Here the quantities

, », w’ are no longer small. We assume that (Z, u) approaches definite constant values (Zo , io), (Z, , ti) for + —© and x > + respectively, whereas the derivatives d/dzx tend to zero. The energy equation is equivalent to the entropy equation

cs)

ge (ms — ie) = (Ge) + (ae) + Gey} + A),

from which one learns that the flow of entropy

CESS

a ar pe

rar re

is a monotone increasing function of x and therefore MS, > MS,. The equality sign could prevail only if u, v, w, T and hence by (22.2) also p were constant. Excluding this trivial case of a constant state® (Z, %) we thus obtain the strict inequality

(24)

MS, > MS,

implying M # 0. The latter remark shows how misleading it is to link the alternative M # 0, M = 0 to the discrimination of shock and slip layer; rather

do these two possibilities correspond to the two ways (A) and (B) of passing to

the limit of vanishing viscosity and heat conductivity.

M

and

Being sure now that

does not vanish, we fall back upon our old relations connecting (Z, , u,), including the Hugoniot

Indeed, in general,

relation.

p = constant, 7’ = constant, imply

Let us assign the labels 0, 1 7 = constant.

There is, however,

one extremely exceptional case in which the conclusion does not stand, namely if M

if for some constant value of p the temperature

z-interval.

Maybe

(Z) , tio)

= 0 and

7’ as function of 7 stays constant in an entire

this is a warning against the danger of a singularity in the shock layer

problem which one steers clear of with certainty only by assuming 7, # 0 or, what is the same,

if the hypotheses I-IV, Ia are preceded by another to the effect that p, 7 may serve as local

parameters on

everywhere.

414

to the two states Z, , Z, , and accordingly orient the x-axis, so that Z, < Z, .

Then Theorem 2 combined with the law of increasing entropy (24) gives M

> 0.

The meaning of this inequality is that the shock front moves in the direction [1] —

[0], or that Z, and not Z, is the initial state of the fluid not yet reached

by the propagating shock.

7. Character

of the Two Singular Points Z., Z:: Node and Saddle*

It is easy to dispose of the equations

(22.3) for v and w.

one for v,

(Cae

Mv

— b),

b =

Single out the

constant,

and introduce the variable

We find v(t)

— b = Ce’,

u(x) tends to positive values wo = respectively; therefore

ge =

Considering that M

a«—-+.o layer.

unless C

Man

Ho

he

1

+

C = constant.

u[Zo], m

=

u[Z,] for

we

forz>—

0)

for

rx

zr >

—©,

2 >

+o

0,

+o.

> 0 one realizes that v cannot tend to a finite limit v, for

= 0.

Thus »v and w must be constant throughout

the shock

(22.1) is satisfied by uw = Mr. Denoting by Ma the constant at the right of (22.2) and substituting from (22.2) the value of p — u* du/dx into (22.4) we are left, as is readily seen, with the following two ordinary differential equa-

tions of first order for the state

WM (25)

T=

d dT Md

Z = Z(z):

Mr — 0) +p ao

1

ec.

‘For ideal gases of constant specific heat with certain special values of the adiabatic exponent y, cf. R. Becker, Zeitschrift fiir Physik 8 (1922), 321-362; in particular, pp. 339-347.

415

The constants M’, a, c are determined by

M*r, + po = M?r, +p. = Ma, € — $M*(r. — a)? = e, — $M7(r, — a)? = 0. As the right members

of (25) vanish for

Z = Z) and Z = Z, , the two

points Z, , Z, turn out to be singular points for the differential system (25).

If

we care only for the trajectory in the (p,7)-diagram along which Z(x) moves while x runs from —© to +© we are faced with one ordinary differential equation of first order in the (p,r)-plane,

nal

id

e—4M(r— ay —c

w* dr

(+ — a) + (p/M’)

ae

But contrary to the customary viewpoint taken in the classical investigations on singular points, which were started by Briot and Bouquet almost one hundred years ago and are reproduced in all treatises on differential equations, our interest is in the parametrized trajectory Z = Z(x). Let us briefly recount the well-known facts about singular points relevant for our purposes.” Placing the singular point at the origin 0 we study two simultaneous differential equations for the “vector” ¢ = (¢,(2), g:(«)) ina w , ¢2-plane,

d.

a

dg.

= Fig, , $2),

oe = Fie , ¢2)

where F, , F, are given functions of , , g, vanishing at the origin. From the outset it is evident that any translation x — x + h carries a solution g =

(¢:(x),

¢2(x))

into a solution

solutions”).

(‘equivalent

expansions of F,; , F, begin with the terms

Suppose

the Taylor

Fale, 2) ~ Rarer + Rarer -

Fy(gi 5 G2) ~ Rugi + Rigs, The equations

define a linear mapping ¢ — and

corresponding

Ra = ka, explicitly

¢y’ = Ry.

eigen-vectors

92 = Rag

,

+ Rie

gt = Rug



+ Rog.

We determine its two eigen-values k =

a

(or

rather

eigen-directions)

by

(k — Rijoy — Ryo. = 0, Roa, + (hk — Rex)o2 = 0, 8See for direct constructive proofs: H. Weyl, Concerning a classical problem in the theory of singular points of ordinary differential equations. Actas de la Academia Nacional de Ciencias de Lima, Volume 7, 1944, pp. 21-60.

416 and assume that the two roots k, , k, of the secular equation

km Ru,

Ry

—Rn,

k— Roo

In a suitable affine coordinate system in the are real, distinct and ~ 0. ¢: , ¢o-plane our mapping is then described by gf = hig: , ¢3 = koge , so that F, , F, are of the form

Pig: , G2) = hig

+ +++,

Fig. , 2) = kag, + +++ We distinguish three cases according to the signs of the eigen-values

k, , kp .

In the case (i) of two negative eigen-values, k, < k, < 0, every solution g which

comes sufficiently near to the origin plunges into it with x tending to + @; its asymptotic behavior is described by

g(a) ~ ae,

g(x) ~ 0-e8*

(a >)

in the sense that

e*“g(c) =a + O0€*), e*o,(2) = O@ *) where ¢ is a positive constant, the same for all solutions, while the constant a is specific (node, see Figure 2). In the case (ii) of one negative, one positive b,

Fie. 2.

Node.

Fia. 3.

Saddle.

eigen-value, k, < 0, k, > 0, there exist exactly two solutions (1, , 1_ in Figure 3) approaching 0 with z > + ©; their asymptotic behavior is described by

ez) ~~ +e", (saddle).

g(a) ~ 0-8"

= (@ 40)

Of course this statement is to be interpreted so that equivalent solu-

tions are considered as one and the same.

The term saddle is better understood

by observing that in this case (ii) two other solutions, 2, and 2_ , plunge into 0 in the direction of the g,-axis, while x goes to — ©. In case (iii) of two positive eigen-values there is no solution whatsoever which approaches the origin as x tends to +, except the trivial one ¢, = y, = 0 (whereas, of course, every solution that comes sufficiently near 0 takes the plunge when x moves toward

— ©),

With these facts in mind we now advance our decisive Theorem

6.

For the differential equations of the shock layer determining the

transition from a state Z, to a state Z, > Z,, the initial state Z, is a node, the end state Z, a saddle.

Clearly this behavior in the neighborhood of the two singular points Z) , Z, is such as to favor the existence of a unique solution of the shock layer problem. All other combinations would make one expect either no solution or an infinity of solutions

Again,

for M’?

=

the proof depends,

(pi —

po)/(t.



7).

above all, on the fundamental

Since S,/S, is the adiabatic derivative m

— dp/dr we may write them as

(26)

MSS>—

S>

inequalities (18)

0;

M’SS?

— 8S? 0;

(27)

First investigate the neighborhood of Z and therefore write

+ p = M*(r — 10) + (p — Po)-

M?(r — a)

For infinitesimal 5p = p — po, br = tT — 7

6e = Ty 6S — py 57,

— a)"}

d{e — 4M(r

hence

64(7 — a)” = (10 — @) 57,

= Ty 6S —

{po + M%(r, — a)} br = Ty 48,

therefore in neglecting terms of higher order

e — 34M*(r — a)”

—c ~ T(S — &).

The following approximate linear system for 6r, 6p results:

d 2 ay (87) = GM" br + bp),

£ (an) = 9h 68 where

g = 1/Mp* > 0,

g = TM/rA>0

and 6S, 67 stand for

5S = 8

Sp + 8

br,

dT = T;” sp + T. dr.

In order to avoid unnecessary encumbrances,

equations determining (67, 6p) at Z) become

the eigen-values

k and

k-ér = g(M6r + 6p),

we drop the index 0.

corresponding

giving rise to the following secular equation for k,

(28)

=)

kT, — g'S,,

kT, — gS,

=0

or

T,-K — {o(M°T, — T.) + g’S,}-k + gg'(M’S, — S,) = 0.

Compute the discriminant

A = {o(M’T, — T,) + g’S,}? — 4g9’T,(M’?S, — S,)

=H g0t

Ty Tyg Sy)

tag iS

the

eigen-directions

-k-8T = g'88

k— gM’,

Then

eS.)

419

and thus verify by (6) that A is positive,

A = {g(M’T, — T.) — g'S,}* + 499" > 499’ = a mal Consequently the eigen-values are real and distinct, at the singular point Z) as

well as at Z,.

Their product equals

go S) —S.)/T2, and

because

of (27)

and

the

fundamental

inequalities

(26)

that product

is

positive at Z) , negative at Z, . Hence the two eigen values are of opposite sign at Z, , of equal sign at Z,. Whether the latter sign is positive or negative

can be decided by the sum of the two k’s for which (28) gives the value

{gMT, — T,) + g6S8p}/Ty . All constituents of this expression are positive, including ge

inns oy

>

COTS?

\r2

a.

Gey

=

1/8

>

0.

Thus we arrive at the result that the two roots k at Z, are positive. Our theorem has now been proved: the fundamental inequalities (17) not only show that by a shock the flow changes from supersonic to subsonic but also that one of the two points Z, , Z; is a node, the other a saddle.

Fic. 4.

The topological picture of the trajectories in the (p, 7)-plane.

420

In order actually to prove the existence of a unique solution of the shock

layer problem, one has to bridge the gap between the neighborhoods of the In that two singular points Z, , Z; by some sort of topological argument. respect one fact is of decisive importance: The differential equations of the shock layer have no singular Theorem 7. point besides Zy , Z, . Indeed if there existed three distinct singular points Z) , Z; , Z, they would satisfy the equations

Mr) + po = Mr, + p: = Mr, + p, = Ma,

€ — $M7(ro — a)” = e, — $M*(7, — a)’ = & — 4M (72 — 0)’. The first set states that Z, , Z, , Z. lie on a straight line of negative inclination —M”’, and we may assume the arrangement Z, < Z; < Z,. Then the second set requires that both Z = Z, and Z = Z, satisfy the Hugoniot equation H(Z, Z.) = 0, in contradiction to a statement previously proved by which there cannot be more than one point Z on any ray from Z, satisfying that equation. The observation is urged upon us that whereas the Hugoniot relation is not

transitive,

equations

the Mr)

relation + po

=

between M*r,

Z,

+ pi,

, Z,

defined

by

the

simultaneous

H(Z, , Zo) = 0

is (M’ being a given constant).

The topological picture of the trajectories which one expects is indicated by Figure 4. If this is correct, the region covered by all trajectories running into Z) for x +

— ~ would be bounded by the saddle line through Z, consisting

of the two branches 2, and 2_ (Figure 3).

The outlook is encouraging.

149. 50 Jahre Relativitatstheorie

Die Naturwissenschaften 38, 73—83 (1950) Einleitung. Die Historiker haben sich zuweilen der Einteilung des Geschichtsablaufs in Jahrhunderte so bedient, als wire die Jahrhundertwende mehr als ein rein auBerlicher, durch unsere Zeitrechnung bedingter Einschnitt. So sprach und spricht man etwa vom Geist des 18. Jahrhunderts. Es ist, als wolle die Geschichte der neueren Physik diesem an sich so unwissenschaftlichen Brauch recht geben. Denn wie die Quantentheorie ziemlich genau mit dem Beginn des gegenwirtigen Jahrhunderts auf den Plan tritt, so wird um die Jahrhundertwende auch das Fundament der Relativitatstheorie in der Form gelegt, wie sie heute gilt, und die dadurch gekennzeichnet ist, daB die endliche Lichtgeschwindigkeit als die obere Grenze der Ausbreitungsgeschwindigkeit aller Wirkungen erscheint. Die historische Stufenfolge der speziellen und der allgemeinen Relativitatstheorie (SR und AR) ist auch sachlich wohl begriindet. Freilich geht die SR in der AR auf, so wie ein naherungsweise in ein exakt giiltiges Gesetz aufgeht. Aber jene so viel leichter zu handhabende Annaherung ist maBgebend fir alle physikalischen Phanomene, in denen die Gravitation vernachlassigt werden kann, und spielt darum im aktuellen Betrieb der physikalischen Forschung, insbesondere der Atomforschung, die weitaus gréBere Rolle. Es ist ja eine der merkwiirdigsten Tatsachen der Natur, daB die Gravitationsanzichung zweier Elektronen 10"%mal so Klein ist wie ibre elektrostatische Abstofung. Heute wird es wohl kaum einen Physiker geben, der daran zweifelt, da die SR einen der wichtigsten und empitisch

am

sickersten

gestiitzten

Grundziige

der

Natur

wiedergibt. Anders steht es mit der auf einem viel schmaleren Erfahrungsfundament beruhenden AR, und wenn auch die meisten Physiker ihre Grundgedanken akzeptieren, so werden wenige anzunehmen geneigt sein, daB wir die allgemein invarianten Feldgesetze bereits in ihrer endgilltigen Form besitzen. Ich brauche Ihnen nicht zu erzihlen, daB in viel hdherem MaBe, als das in der Quantentheorie der Fall ist, Grundlagen und Ausbau der Relativitatstheorie das Werk eines Mannes sind: Atpert E1nsTetn. Wie er, ungleich den meisten andern Physikern, sich seinerzeit mit der von ihm errichteten spezicllen Relativitatstheorie nicht zufrieden gab und in einer an GALILET und NEwToN gemahnenden Kombination von Empirie und Spekulation zur AR vorstieB, so ist er auch jetzt noch unablassig mit dem Problem einer einheitlichen, alle Naturkrafte umfassenden Feldtheorie beschaftigt (7). A. Spezielle Relativitatstheorie. 1. Vorgeschichte und Begriindung. ‘Als Max Prancx 1900 die Quanten zur Erklérung der Formel fiir die Energieverteilung im Spektrum der schwarzen Hohlraumstrahlung einfiihrte, trat, so darf man sagen, mit der Quantentheorie auch das *) Vortrag, gehalten auf der Tagung der Gesellschaft Deutscher Naturforscher und Arzte in Miinchen am 24. Oktober 1950.

Quantenproblem zum erstenmal auf den Plan. Das Relativitatsproblem aber ist viel alter als die im 20. Jahrhundert entstandene Relativitatstheorie, in welcher wir seine Lésung sehen. Schon ARISTOTELES definiert Ort (réx0c) relativ, namlich als Beziehung eines Kérpers zu den Kérpern seiner Umgebung. Locke gab in seinem Hauptwerk eine eingehende erkenntnistheoretische Analyse [2], GALILEr illustriert die Relativitat der Bewegung hiibsch durch das Beispiel des Schreibers, der an Bord eines von Venedig nach Alexandrette segelnden Schiffes seine Notizen macht und dessen Feder ,,in Wahrheit", das ist relativ zur Erde, eine lange, glatte, nur leicht gewellte Linie beschreibt [3]. Offenkundig durch theologische Uberzeugungen mitbestimmt, verkiindet Newron am Beginn seiner Principia mit ehernen Worten die Lehre vom absoluten Raum und der absoluten Zeit. Er gibt das Beispiel zweier durch einen Faden verbundenen Kugeln, die um eine zum Faden senkrechte Mittelachse rotieren: die Spannung des Fadens zeigt Existenz und Geschwindigkeit der Rotation an. Allgemein stellt er sich die Aufgabe, aus den relativen Bewegungen der Krper und den bei der Bewegung auftretenden Kraften ihre absolute Bewegung zu bestimmen, und erklirt geradezu, daB sein treatise verfaBt wurde um der Lésung dieses Problems willen. ,,To this end it was that I composed it", heiBt es am Schluf der Vorrede [4]. Gelingt ihm sein Vorhaben? Nicht véllig. Die von ihm aufgestellten Gesetze der Mechanik (und der Gravitation) gestatten wohl, die Bewegung eines Massenpunktes in gerader Linie mit konstanter Geschwindigkeit (gleichférmige Translation) von allen andern Bewegungen 2u unterscheiden; hingegen gestatten sie nicht, unter den gleichférmigen Translationen die Ruhe auszuzeichnen. Um dies doch zu erreichen, nimmt NewTon zu einer kosmologischen Hypothese und einer Begriffsunterschiebung seine Zuflucht, die sich gar fremd in dem sonst so wohlfundierten, herrlichen Aufbau der Principia ausnehmen. Seine Hypothese ist, daB das Weltall ein Zentrum habe und dieses sich in Ruhe befinde. Vom Schwerpunkt des Sonnensystems stellt er fest, daB es eine gleichférmige Translation ausfihrt, und fahrt dann fort: ,,But if that

center

moved,

the

center

of

the

world

would

move also against the hypothesis‘ [5]. Beides, die Hypothese vom ruhenden Zentrum des Weltalls und diese durch nichts begriindete Identifizierung desselben mit dem Schwerpunkt des Sonnensystems, sind erstaunlich. Newtons Bild vom Kosmos ist offenbar noch wesentlich gebundener als das, welches schon 100 Jahre fraher GrorpaNo Bronos leidenschaftliche Seele erfiillte [6]. Raum und Zeit bilden ein vierdimensionales Kontinuum, das wir mit Minkowskt die Well nennen. Ein raum-zeitlich eng begrenztes Ereignis geschieht an einer bestimmten Raum-Zeit-Stelle, in einem bestimmten Weltpunkt, ,,hier-jetzt“. Nur raumzeitliche Koinzidenz oder unmittelbare raum-zeitliche Nachbar_ schaft ist etwas in der Anschauung unzweifelhafy

422

Gegebenes. Fin Massenpunkt beschreibt eine eindimensionale Linie in der Welt. Wenn zwei Personen sich treffen und sich die Hand reichen, so geschieht das an einer bestimmten Raum-Zeit-Stelle, in welcher sich ihre Weltlinien schneiden. Weil kein Raumpunkt und kein Zeitpunkt von dem andern an sich physikalisch verschieden ist, entsteht das Problem NEwToNs, wie U

Tr

t= const

==

fe

Abb. 1, Raum-Zeit-Koordinatensystem.

man von zwei Ereignissen entscheiden soll, ob sie am gleichen Ort (wenn auch zu verschiedenen Zeiten) oder zur gleichen Zeit (wenn auch an verschiedenen Orten) geschehen, Um ein graphisches Bild entwerfen zu

(nichthorizontale) gerade Linien die méglichen gleichférmigen Translationsbewegungen dar, Das spezielle Relativitatsprinzip, das sich aus der Newronschen Mechanik ergab und das Newrow selber nur aus theologisch-spekulativen Griinden verleugnete, besagt, da in einem in gleichférmiger Translation befindlichen Eisenbahnwagen alle Vorginge genau so ablaufen, alle Tatigkeiten, wie Briefeschreiben oder Ballspielen, genau so ausgefiihrt werden kénnen, wie wenn der Wagen ruhte; Stérungen treten nur bei Beschleunigungen auf. Schon GaLitet hatte dieses Prinzip klar erkannt und entwickelt es in seinem Dialogo sopra i dui massimi sistemi del mondo [7]. Ex fabt die Bewegung auf als einen Kampf zwischen Tragheit und Kraft; solange ein Kérper durch keine SuBeren Krafte abgelenkt wird, beschreibt er eine gleichférmige Translation. In unserem graphischen Bild kénnen wir diese Erkenntnis dahin aussprechen, da8 wohl die geraden Linien unter allen andern ausgezeichnet sind, aber nicht die vertikalen Geraden. ‘An der objektiven Bedeutung der Gleichzeitigkeit, der ,,horizontalen Schichtung, hatte vor EINSTEIN niemand ernstlich gezweifelt. Indem es dem Menschen natiirlich ist, anzunehmen, daB ein Ereignis dann geschieht, wenn er es beobachtet, dehnt er seine eigene Dieser naive Glaube Zeit auf die ganze Welt aus. war freilich durch OLar ROmeRs Entdeckung der endlichen Lichtgeschwindigkeit erschiittert worden. Reprasentiert man in unserem graphischen Bild 1 sec auf der Zeitachse durch eine Strecke von derselben Lange wie die Raumstrecke in der Ebene E, die das Licht in 14sec durchmiBt, so wird die Ausbreitung eines in O gegebenen Lichtsignals durch einen geraden vertikalen Kreiskegel vom Offnungswinkel 90° mit Spitze inO dargestellt (Abb. 2) ; ergibt den, ,Lichtkegel"

wieder, auf welchem diejenigen Weltpunkte liegen, in

Abb. 2, Lichtkegel.

kénnen, nehmen wir dem Raum eine Dimension, studieren also nur die Vorginge auf einer horizontalen EbeneE und tragen im Bild die Zeit ¢ senkrecht zu dieser Ebene £ auf(Abb. 1), In unserem Bilde erscheinen gleichzeitige Weltpunkte als in einer Horizontalebene

gelegen, gleichortige auf einer vertikalen Geraden. Newtons

Glaube

an

einen

absoluten

Raum

und

an

eine absolute Zeit bedeutet also, da er der Welt eine Struktur zuschreibt, die sich in einer (horizontalen) Schichtung und einer quer dazu verlaufenden (vertikalen) Faserung ausdriickt. Durch jeden Weltpunkt geht eine Schicht und eine Faser. Raum und Zeit kommt ferner eine metrische Struktur zu, auf Grund deren wir von der Gleichheit von Raumstrecken und von Zeitintervallen sprechen, In unserem graphischen Bilde kénnen wir das am anschaulichsten durch eine Schar konzentrischer Kreise in der Ebene E und eine Reihe Aquidistanter Punkte auf der vertikalen Zeitachse veranschaulichen. Gibt das Bild in dieser Weise die metrischen Verhiltnisse getreu wieder, so stellen

denen das Signal eintrifft. Die Begriffe von Vergangenheit und Zukunft haben, wie Lerentz nicht mide wurde zu betonen, kausale Bedeutung: durch ,,hierjetzt’, im Weltpunkt 0, abgeschossene Kugeln kann ich Carus Jutius Cazsar nicht mehr treffen, da seine Weltlinie ganz der vergangenen Welthalfte in bezug auf O angehért. Und man nahm an, daB es die Ebene ¢ =const durch 0 ist, welche diese Trennung in eine vergangene und zukiinftige Welthalfte bewirkt. Die neue Erkenntnis, welche das 20. Jahrhundert brachte, war die, daB die Kausalstruktur der Welt nicht durch die ebenen Schichten ¢ const, sondern durch die von jedem Weltpunkt ausgehenden Lichtkegel beschrieben wird. Physikalisch heiBt das, daB keine Wirkung sich mit gréBerer als Lichtgeschwindigkeit ausbreitet. Im Innern des ,,vorderen’ yon O ausgehenden Lichtkegels liegen alle Weltpunkte, auf die das, was in O geschieht, noch von Finfluf sein kann; im Innern des durch riickwartige Verlingerung entstehenden hinteren Lichtkegels aber liegen alle Ereignisse, die auf das, was in O geschieht, méglicherweise von Einflu8 waren, insbesondere solche, von denen ich jetzt hier, in O, eine auf direkte Wahrnehmung gegriindete Kunde haben kann. Die beiden Teile des Kegels grenzen nicht zwischenraumlos aneinander; mit dem Zwischengebiet ist 0 kausal tiberhaupt nicht verbunden. Wenn hier zur Beschreibung der Weltstruktur (unter Streichung einer Dimension) ein Bildraum Verwendung fand, von dem in den vertrauten geometri-

423

schen termini der euklidischen Geometrie gesprochen wurde, so geschah das nur im Interesse der leichteren Verstandlichkeit. In Wahrheit brauchen wir zur begtifflichen Darstellung der Naturvorginge eine Abbildung der Welt oder des in Frage kommenden Weltstiicks auf ein Stiick des vierdimensionalen Zahlenraums; dessen Punkte sind nichts anderes als die méglichen Zahlenquadrupel (x, Xa, 9» %4). Relativ zu einem solchen Koordinaten- oder Bezugssystem wird die Zeit eines Weltpunktes durch x4, der Ort durch (x, %», %) angegeben: es legt in dieser Weise eine rein konventionelle Gleichzeitigkeit und Gleichortigkeit fest. Die Abbildung der Welt auf den Zahlenraum ist an sich ganz willkiirlich; sie dient dazu, den Weltpunkten Namen zu geben. Das allgemeine Relativitatsprinzip ist so im Grunde nichts anderes als die Ablehnung des Namenzaubers, indem es behauptet, da Ereignisse nicht davon berithrt werden, in welcher Nomenklatur man sie beschreibt. Wenn eine Klasse von Koordinatensystemen objektiv herausgehoben werden soll, so kann dies nur dadurch geschehen, daB man yon bestimmten physikalischen Vorg’ngen ausgeht und angibt, wie sie sich in einem Koordinatensystem dieser Klasse arithmetisch ausdriicken. So wird 2.B. in E1nsre1ns allgemein-relativistischer Grayitationstheorie das Gravitationsfeld eines Zentralkorpers gegebener Masse nach K. SCHWARZSCHILD durch bestimmte Formeln in einem Koordinatensystem dargestellt, das eben durch diese auf dasselbe bezogene Formeldarstellung weitgehend normiert ist. Bisher

haben

wir uns

von

der Newtonschen

An-

sicht tiber die Struktur der Welt 2u einer solchen den Weg zu bahnen gesucht, die nicht in apriorischen Prinzipien, sondern auf Erfahrungen basiert ist. Nun miissen wir die so gewonnene Ansicht ohne den Umweg tiber Newron kurz systematisch darstellen. Wir benutzen zwei Grundvorgange: die Ausbreitung eines Lichtsignals und die kriftefreie Bewegung eines Massenpunktes. Fiir diese gelten die folgenden beiden Grundgesetze, die, wie ich gleich bemerke, auch im Rahmen der allgemeinen Relativitatstheorie bestehen bleiben:

4. Der Lichtkegel in O, das ist der geometrische Ort der Weltpunkte, in denen ein in O gegebenes Lichtsignal eintrifft,

ist durch O eindeutig bestimmt,

unabhingig von dem Zustand, insbesondere dem Bewegungszustand der das Licht in O aussendenden Lichtquelle. (Dieses Gesetz wird haufig durch den

etwas irrefiihrenden Namen

,,Gesetz von der Konstanz

der Lichtgeschwindigkeit“ bezeichnet).

2. Die geodatische Weltlinie, die ein kraftefrei sich bewegender Massenpunkt beschreibt, ist durch Anfangspunkt und Anfangsrichtung in der Welt eindeutig bestimmt. Die spezielle Relativitatstheorie ist auf die Annahme basiert, daB Bezugssysteme existieren, Abbildungen der Welt, in welcher sich diese Vorginge in einer vollig determinierten Weise darstellen: namlich die geodatischen Linien als gerade Linien und die Lichtkegel als Nullkegel, das ist als vertikale Kreiskegel vom Offnungswinkel 90%. Ein solches normales Koordinatensystem ist nicht eindeutig bestimmt, sondern nur bis auf eine LoreNtz2-Transformation; die Gruppe dieser Transformationen besteht aus allen solchen Abbildungen des Bildraums, welche Gerade in

Gerade und Nullkegel in Nullkegel tiberfiihren.

Es ist

eine rein mathematische Aufgabe, diese Gruppe zu

bestimmen, Die Theorie wird abgeschlossen durch die Behauptung, da unter diesen gleichberechtigten Koordinatensystemen

auch

unter

Beriicksichtigung

aller weiteren Naturvorginge objektiv keine engere Auswahl getroffen werden kann: die Naturgesetze sind invariant gegeniiber Lorentz-Transformationen. Unter diesen Transformationen gibt es solche, den gegebenen Weltpunkt0 festlassende, welche die verti-

kale Gerade durch 0 in eine beliebige zeitartige Gerade durch diesen Punkt iiberfiihren (d.h. in eine, die von O

ins Innere des Nullkegels in O hineinfiihrt): darin driickt sich das spezielle Relativitatsprinzip aus. MiNKowsKI erkannte, daB die Gruppe der LorentzTransformationen mit der Gruppe der euklidischen Ahnlichkeiten zusammenfallt, wenn man sich nicht scheut, der einen Koordinate x, statt reeller rein imaginare Werte zu erteilen. Die von der speziellen Relativitatstheorie der Welt zugeschriebene Tragheitsund

Kausalstruktur

ist darum

von

einer metrischen

Struktur abzuleiten, Und zwar laBt sich die der Welt

zukommende Geometrie, so wie die euklidische Raum-

geometrie, auf den affinen Begriff des Vektors auf-

bauen, und wie in der euklidischen Geometrie bestimmen nach Wahl einer Mafeinheit je zwei Vekto-

ren z,9 ein skalares Produkt (r- y), das eine nichtausgeartete Bilinearform der beiden Argumente ist. Nur ist die zugehérige quadratische Form (r- z), ,,Quadrat der Lange’ des Vektors x, nicht posi definit, (¢-z)>0, sondern mit bezug auf eine gecig nete Normalbasis ¢, ¢s, ej, ¢y fiir die Vektoren, die dem

Cartesischen

Achsenkreuz

in

der euklidischen

Geometrie entspricht, wird das Quadrat der Lange des Vektors r= >) x, ¢, durch die Formel (damit at gt

mit einem negativen Vorzeichen gegeben (quadratische Form der Signatur1). Den Zusammenhang dieser Weltgeometrie mit der physikalischen Messung von Raumstrecken durch starre MaBstabe und yon Zeitintervallen durch Uhren wollen wir erst im Zusammenhang der allgemeinen Relativitatstheorie

kurz

erlautern.

An

dieser

Stelle

nur zwei Bemerkungen dariiber. StoBt man einen Eisenklotz gleichzeitig an verschiedenen Stellen an, so werden sich die Umgebungen dieser Stellen in Bewegung setzen; aber erst, wenn

die mit Lichtgeschwin-

digkeit sich um die Stellen ausbreitenden Wirkungskugeln sich zu tiberlappen beginnen, werden diese Bewegungen sich gegenseitig beeinflussen. Diese einfache Bemerkung von Herrn von Lave zeigt, daB schon in der speziellen Relativitatstheorie der Begriff des starren Kérpers im Prinzip hinfallig wird. — Die klassische

Feldphysik la8t Kérper beliebiger Ladung und Masse mi. Sie ignoriert damit, méchte man sagen, die Grundtatsache der Atomistik, daB die Materie aus Elementarteilchen fester Ladung und Masse besteht. In der Feldphysik erscheint darum auch die Wahl der MaBeinheit, in der Raum- und Zeitstrecken gemessen werden, als willkirlich; die Gruppe der Lorentz-Transformationen umfaBt die Dilatationen. Die Unterscheidung von Ahnlichkeit und Kongruenz, auf die diese Bemerkung hinweist, ist haufig von Darstellern der Relativitatstheorie verwischt worden,

424

2. Anfiinge und Fortbildung der Theorie.

a) Friihe Geschichte.

Es ist meine Aufgabe, Ihnen

ein Referat tiber die Entwicklung der Relativitatstheorie in der ersten Halfte dieses Jahrhunderts zu erstatten.

Es schien mir aber zweckmaBig,

zunachst

einen knappen systematischen Abrif der Theorie vor-

anzustellen, in dem ich zugleich bemiiht war, verkehrte Auffassungen abzuwehren.

Holen wir nun in raschen

Schritten die Historie nach [8]: Die optischen waren von Maxwett den elektromagnetischen Erscheinun-

gen eingegliedert worden; dementsprechend spielt die

Lichtgeschwindigkeit c in den Maxwettschen Gleichungen des elektromagnetischen Feldes eine wichtige Rolle.

Solange man an einen substantiellen Lichtather

glaubte, wurde die aus diesen Gleichungen folgende Ausbreitung einer elektromagnetischen Stérung mit Lichtgeschwindigkeit in konzentrischen Kugeln als

eine Beschreibung des Vorganges relativ zu dem im Ganzen ruhenden Ather aufgefaBt. Dann muBte man aber erwarten, daB die Bewegung ponderabler Materie

relativ zu diesem Ather sich durch bestimmte Effekte kundgeben wiirde.

Die Erfahrung hingegen zeigte,

daB das spezielle Relativitatsprinzip auch bei der Ausdehnung auf optische und elektromagnetische Vor-

giinge seine Richtigkeit behalt. Aus der von ihm mitbegriindeten Elektronentheorie heraus erschlo8

H. A.

Lorentz zuerst die Tatsache, daB alle Effekte 1. Ordnung in dem Verhiltnis v/e zwischen Materie- und

Lichtgeschwindigkeit fiir einen mitbewegten Beobachter herausfallen. Als weitere Versuche, wie der beriihmte MicHELsonsche

Interferenzversuch,

auch

das

Fehlen von Effekten 2. Ordnung feststellten, nahm

Lorentz zu der Hypothese seine Zuflucht, daf ein Kérper infolge seiner Bewegung gegen den Ather eine Langskontraktion im Verhaltnis 1: //i— (o/c)? erfahrt.

1900 stellte er die heute allgemein als LoRENTZ-Trans-

formation

bekannten

Formeln

auf.

Die

véllige

Ab-

klarung brachten dann drei Arbeiten von LoRENTZ, H. Pomncaré und Einstein 1904/05. Einstery nahm die grundsitzliche Wendung, daB er das spezielle Relativititsprinzip als exakt giiltig postulierte und nun

zusah, welche Modifikationen fir unsere Raum-Zeit-

Vorstellungen resultieren, wenn man damit die ele-

mentaren Gesetze der Ausbreitung des Lichtes und der elektromagnetischen Wellen zusammenhilt, und welche Gesetze fiir die Elektrodynamik bewegter Kérper daraus folgen, wenn man die fiir ruhende Kérper geltenden aus der (phinomenologischen) MaxweELt-

schen Theorie heriibernimmt. Erst hierdurch kam es zu der radikalen Kritik des Begriffs der Gleichzeitigkeit. Minkowskt endlich gab die weltgeometrische Einkleidung, die insbesondere fiir die Weiterentwicklung zur allgemeinen Relativitatstheorie bedeutungsvoll wurde (Vortrag auf der Kélner Naturforscher-

versammlung 1908).

b) Elektrodynamik bewegter Kérper. Leicht waren die optischen Konsequenzen des Relativitatsprinzips zu ziehen. Aus dem einfachen Umstand, da8 die Phasendifferenz einer ebenen Lichtwelle in einem homogenen Medium an zwei Weltstellen A, B linear von dem Weltvektor AB abhangt, ergibt sich durch Ver-

gleich zweier gleichberechtigter Spaltungen der Welt in Raum und Zeit die Aberration als Geschwindigkeitsperspektive, der Dorrrer-Effekt und der Fizeavsche Milfithrungskoeffizient. Betreffs Ableitung der

elektromagnetischen Gleichungen fiir bewegte Kérper sei die folgende methodische Bemerkung gemacht. Im Grunde gestattet das Prinzip, von ruhenden Kérpern nur auf den Fall zu schlieBen, wo alle beteiligten Kérper dieselbe gleichformige Translation erfahren. In Wahrheit interessieren aber Aussagen tiber Situationen, in denen mehrere Kérper mit verschiedenen Geschwindigkeiten auftreten. Nehmen wir den Fall zweier durch einen leeren Zwischenraum getrennter Kérper, die sich (in unserem Bezugssystem) mit verschiedenen Geschwindigkeiten bewegen. Fir den einzelnen Kérper und den ihn umgebenden leeren Raum erhalt man bestimmte Gleichungen, indem man die fiir den ruhenden Kérper geltenden durch Lorentz-Transformation in solche umwandelt, die relativ 2u dem zugrunde liegenden Bezugssystem gelten. Indem man so fir beide Kérper verfahrt, erhalt man Gleichungen, die im leeren Zwischenraum miteinander tibereinstimmen, da die Maxwettschen Gleichungen im leeren Raum Lorentz-invariant sind. Diese Gesetze kann man somit ohne Widerspruch als giiltig annehmen, und man verwendet sie de facto als Naherungsgesetze, solange wenigstens die Distanz der beiden Kérper nicht von molekularer GréSenordnung ist. Die auf diese Weise von Lorentz, EINsTEIN, MINKOwsKI, Born u. a. gezogenen Konsequenzen haben sich durchweg in der Erfahrung bewahrt. ¢) Mechanik. Energie und Tragheit. In der Mechanik haben wir es mit Kérpern 2u tun, bei denen wir den von einem mitbewegten Beobachter zu messenden inneren Zustand von dem durch seine vektorielle Geschwindigkeit » bestimmten Bewegungszustand unterscheiden kénnen. Wir betrachten die Korper nur im Zustand gleichformiger Translation. Das Impulsgesetz besagt, daB einem solchen Kérper ein vektorieller Impuls § 2ukommt, welcher der Geschwindigkeit parallel ist, $=, und daB vor und nach einer Reaktion die Summe der Impulse der beteiligten Kérper die gleiche ist. Der skalare Faktor m ist die trage Masse; so war dieser Begriff im Grunde schon von GatiLer und HuyGens eingefithrt worden. Durch das Relativitatsprinzip erkennt man, daB das Impulsgesetz das weitere zur Folge hat, nach welchem auch die Massensumme 5° m durch eine Reaktion sich nicht Gndert. Indem man den Vorgang von zwei villig gleich beschaffenen Kérpern, die mit entgegengesetzt gleichen Geschwindigkeiten +» gegeneinander gejagt werden und sich dabei 2u einem einzigen, notwendig ruhenden Kérper vereinigen, von einem beliebigen normalen Bezugssystem aus studiert, erkennt man bei Zugrundelegung der Gatieischen Kinematik, die aus der Enysretnschen durch den Grenziibergang 2u c-> oo entspringt, daB die Masse von dem Bewegungszustand und infolgedessen auch yon dem inneren Zustand eines Kérpers unabhangig ist. Hingegen ergibt sich auf Grund der Emsrerschen Kinematik die Formel wo m die nur vom inneren Zustand abhiingige Ruhmasse, v der Betrag der Geschwindigkeit ist und ¢ gesetzt wurde. Messungen an Kathodenstrahlen bestiitigten die Formel um so genauer, je mehr sich die Messungsmethoden verbesserten. Man versteht aus dieser Formel, daB der Tragheitswiderstand eines Kérpers mit seiner Geschwindigkeit so anwichst, daB

425 diese niemals die Lichtgeschwindigkeit

Verainderung

erreicht.

Die

der Masse mit dem Bewegungszustand

des Kérpers hat zur Folge, daB sie auch von inneren Zustandsanderungen beeinfluBt wird, z.B. bei Erwar-

mung eines Kérpers einen Zuwachs erfahrt. Natiirlich ist die Massenanderung des Kérpers unabhiingig davon, wie diese Zustandsinderung zustande kommt. Zusammen mit dem Faktum, da8 fiir ein abgeschlos-

senes Kérpersystem die Summe der Masseninderungen verschwindet, zeigt dies an, erstens, daB trige

Masse und Energie dasselbe sind, und zweitens, daB das Energiema8 einer Zustandsanderung als die Differenz von den einzelnen Zustanden zukommenden Energieniveaus autzufassen ist. Dieses Gesetz von der

Tragheit der Energie, das sich in den iiblichen Einheiten in der heute so popular gewordenen Formel, Energie

E =mc*,

ausdriickt, ist zweifellos die wich-

tigste Folgerung aus der Relativitatstheorie; sie wurde

von EINSTEIN schon in seiner ersten Arbeit 1905 gezogen, und schon damals faBte er die Anwendung derselben auf Kernreaktionen ins Auge, bei denen der

der freiwerdenden Energie entsprechende Massendefekt

meBbare GréBenordnung erreichen mag. Das ist ein Gebiet, das inzwischen ins Zentrum der Forschung geriickt ist, von dem damals aber nur die Erscheinungen des spontanen radioaktiven Zerfalls bekannt

waren. Um die GréBenordnung zu kennzeichnen, gebe

ich als Beispiel die Bildung eines Lithium-Atoms aus drei Protonen und drei Neutronen; die Masse (das

Atomgewicht) des Lithiums ist 6,01692, der Massendefekt 0,03432, entsprechend einem Energieverlust von 5,11 + 10-5 erg je Atom.

Die Maxwettsche Theo-

rie schreibt dem elektromagnetischen Felde Dichte und Stromdichte der Energie und des Impulses zu. Ein

Stiick des Feldes ist nicht ein ,,Kérper‘, an dem man Bewegungs- und inneren Zustand unterscheiden kann. driickt

sich

das

Gesetz

Impulsdichte

ist.

An

dem

Hier

der Tragheit

von

der

Energie darin aus, daB der Energiestrom gleich der ruhenden

eines

Beispiel

Korpers, der eine kugelférmige Lichtwelle vom Impuls 0 aussendet, entdeckte E1nsTEIN zuerst das Gesetz von der Tragheit der Energie, indem er feststellte, daB die ausgesandte Energie durch eine entsprechende Abnahme

der

trigen

Masse

des

Kérpers kompensiert werden mu8.

lichtaussendenden

Er erkannte so-

gleich, daB dies Gesetz dieselbe universelle Giiltigkeit

besitzt wie das Relativitatsprinzip, aus dem es zwin-

gend hervorgeht.

3. Verbindung mit der Quantentheorie. In eine neue Epoche tritt die SR ein durch ihre Verbindung mit der das atomare Geschehen beherr-

schenden Quantentheorie, wie sie in mathematisch priziser Form um 1925 von HEISENBERG und SCHRODIN-

GER aufgestellt wurde [10]. Die Verbindung ist nicht ohne

Schwierigkeiten

zu vollziehen.

Die

als Funk-

tion von Ort (x, %g, 9) und Impuls (f,, Pe, P.) ausgedriickte Energie p,, welche nach den Hamittonschen Gleichungen der klassischen Mechanik die Bewegung

eines freien Elektrons bestimmt, lautet

Py = mt + E+ BE

2),

wo m die konstante Ruhmasse ist. Bei der quantentheoretischen Ubersetzung dieses Ausdrucks, bei wel-

cher fy durch den Operator * 2 an ersetzen ist " (i = 1, h die durch einen Faktor 22 modifizierte Prancxsche Wirkungskonstante), macht die Quadratwurzel Schwierigkeiten. Schreibt man aber die Gleichung in ihrer rationalen Form, Pi— (bE + P3 + P3) = m2,

(1)

welche die relativistische Invarianz unmittelbar in Evidenz setzt, so ergibt die Ubertragung nicht die von der Quantentheorie allgemein vorgeschriebene Form. Dirac itberwand dieses Hindernis durch die den Mathematikern wohlbekannte, aber in ihren Handen unfruchtbar gebliebene Bemerkung, daB die quadratische Form der Variablen p, auf der linken Seite von (1) sich mit Hilfe gewisser hyperkomplexer Zahlen 7, deren Multiplikation nicht kommutativ ist, als das Quadrat einer Linearform 5, yy , schreiben laBt, und er setzte darum die Gleichung an Dnte Pe =

die der quantenmechanischen Ubersetzung keine Schwierigkeiten in den Weg legt [11]. Die Physiker miissen mir verzeihen, wenn ich hier um der Kiirze willen diese reichlich abstrakte Fassung von Diracs Grundidee wahle. Gema8 seinem Ansatz wird das Wellenfeld des Elektrons nicht durch eine skalare Funktion p der vier Weltkoordinaten x, beschrieben,

sondern durch eine Gréfe mit vier Komponenten y,, die sich unter dem EinfluB einer Lorentz-Transformation der Koordinaten in einer ungewhnlichen, in der

iiblichen Vektor-

und

Tensorrechnung

nicht

vor-

geschenen Weise transformieren. Ein gemeinsamer

konstanter Phasenfaktor e~'* vom absoluten Betrag 4 bleibt in den y, unbestimmt.

Aus dem Hami-ronschen

Prinzip entnimmt man ferner die Regel, da man das auf das Elektron einwirkende elektromagnetische

Feld

einfach dadurch beriicksichtigen kann, daB man die Operatoren 2— durch m +igy ersetzt, wo g die mit ¢/h multiplizierten Komponenten des elektromagnetischen Potentials sind (e = elektrische Elementarladung). Diese Dinge werden hernach im Rahmen der AR unter dem Titel Eichinvarianz von besonderer Wichtigkeit werden. Die geschilderte Lorentz-invariante Diracsche Theorie des Elektrons gibt in wunderbarer Weise Rechenschaft iiber den aus der Analyse der Atomspektren von S. Goupsmir, G. E. UHLENpeck und W.Pautt erschlossenen Elekironenspin, iiber den anomalen ZeEMANN-Effekt, die Feinstruktur des Wasserstoffspektrums und viele andere Dinge. Man kann wohl sagen, daB die Tatsachen der Spektroskopie heute die zuverlissigste Stiitze fiir das Relativitatsprinzip abgeben. Der Drracsche Erfolg ist um so bemerkenswerter, als bei dem Ubergang von einem Teilchen, Photon oder Elektron, 2u einer unbestimmten Anzahl von in Wechselwirkung miteinander stehenden Photonen und Elektronen die Verschmelzung von Quanten- und Relativitatstheorie auf Hindernisse stoBt, die, trotz vielversprechender Ansitze, noch nicht aus dem Wege geriumt werden konnten, Ich glaube aber, niemand denkt daran, deswegen der Relativitatstheorie den LaufpaB zu geben; sie ist dafiir 2u fest in dem ganzen Gebiude unserer theoretischen Physik verankert.

426 B, Allgemeine Relativitatstheorie.

1. Der Grundgedanke. Es lieB sich nicht vermeiden, daB ich hier fiir einen Augenblick das Gebiet der gestrigen Vortrige, die Quantentheorie!), berahrte, Aber nun ist es héchste Zeit fiir mich, von der speziellen zur allgemeinen Relativitatstheorie iiberzugehen, Es wurde schon oben gesagt, da® wir zur begrifflichen Beschreibung der Naturvorginge die Weltpunkte auf Koordinaten bezichen miissen, daB aber die Erscheinungen selber durch diese an sich willktirliche Namengebung natiirlich nicht beeinfluBt werden, Der Ubergang von einem zu einem andern Koordinatensystem geschieht durch eine (stetige) Koordinatentransformation. Von was fiir Gesetzen auch immer die Natur beherrscht sein mag, ich kann deren Formulierung ein willktirliches Koordinatensystem zugrunde legen, und sie werden alsdann invariant sein gegeniiber beliebigen Koordinatentransformationen, Freilich muB ich dabei das Tragheits- und Kausalfeld oder das metrische Feld, aus dem beide abgeleitet werden, mit unter die physikalischen ZustandsgréBen aufnehmen. Das Prinzip der Invarianz gegentiber beliebigen Koordinatentrans-

formationen ist also an sich nichtssagend, und der physikalisch entscheidende Gedanke, der von der spe-

ziellen zur allgemeinen

Relativitatstheorie

fihrt, liegt

denn auch woanders als in diesem Prinzip. Die Ei sicht in die Relativitat des Ortes scheint uns 2u zwingen, alle Bewegungszustdnde und nicht nur die gleichformigen Translationen als gleichwertig zu erachten So haben denn schon zu NEwrons Zeit Denker wie Lepniz und Huycens, spiter Euter, sich um das Riitsel bemiiht, was der offenkundigen dynamischen Ungleichwertigkeit der kinematisch gleichwertigen

Bewegungszustinde zugrunde liegt. In neuerer Zeit

war es Ernst Macu, der mit allem Nachdruck allgemeine Prinzip der Relativitat der Bewegung focht, und das Studium von Macs (neben dem Home) war auf Envstem nach seinem eigenen

stindnis von maBgebendem Einflu8 [22].

das vervon Ge-

Wir haben als Tragheitsfeld diejenige Struktur der Welt bezeichnet, die einem Kérper eine durch Anfangsort und -richtung in der Welt eindeutig bestimmte Bewegung aufnétigt, in der er zu beharren bestrebt ist, solange er nicht durch duBere Krafte abgelenkt

wird. Die wirkliche Bewegung resultiert aus dem Kampf zwischen Trigheit (Beharrungstendenz) und Kraft. Lassen Sie mich an ein Beispiel ankniipfen, das, wenn ich mich recht besinne, Putuipr Lenarp

auf

der

Naturforscherversammlung

in

Bad

Nauheim

1920 in die Diskussion iber Relativititstheorie hineinwarf: Ein Zug stoBt mit einem entgegenfahrenden Zuge zusammen, wabrend er an dem Kirchturm eines Dorfes vortiberfhrt; warum, fragte LEwarp, geht der Zug in Trimmer, und nicht der Kitchturm, wo doch der Kirchturm relativ zum Zug einen ebenso starken Bewegungsruck erfihrt wie der Zug relatiy zum Kirchturm? Die unvoreingenommene, von keiner Relativitatstheorie angekrankelte Antwort ist wohl Klar: der Zug wird zerrissen durch den Konflikt seiner eigenen Tragheit mit den auf ihn von dem zusammenstoBenden Zug ausgeiibten Molekularkriiften ; wahrend der Kirchturm ruhig der ihm durch das Tragheitsfeld

4) Vgl. die Vortrage von W. Hersenperc, M. von Lave und P, Hanreck, Naturwiss. 38, 494f. (1951).

vorgeschriebenen Bahn folgt. In dem von der SR aufgestellten, bis auf eine Lorent2-Transformation normierten Koordinatensystem erscheint das Tragheitsfeld als eine starre, der Welt ein fiir allemal innewohnende geometrische Beschaffenheit: wahrend es enorme Wirkungen auf die Materie ausiibt, ist es selbst iiber alle Einwirkungen der Materie erhaben. Dagegen straubt sich unser Gerechtigkeitsgefiihl: was Wirkun-

gen auf die Materie ausiibt, muB auch Wirkungen von ihr erleiden. Wo aber sind in der Natur Anzeichen dafiir vorhanden, daB das Tragheitsfeld nicht vorgegeben, sondern den Einwirkungen der Materie gegeniiber nachgiebig ist? Hier setzt EmvsTemns fundamentaler Gedanke ein: die Gravitation ist dieses Anzeichen. Wenn dies stimmt, wenn in dem Dualismus von Kraft und Tragheit die Gravitation auf die Seite der Tragheit gehért, so wiirde auf einmal die seltsame Tatsache der Ubereinstimmung von schwerer und trager Masse verstandlich, die durch die feinsten Messungen immer wieder besttigt wurde. Die Kraft, mit welcher ein elektrisches Feld an einem geladenen Kérper angreift, ist seiner Ladung proportional; so ist die Kraft, mit welcher das Gravitationsfeld an einem Kérper ai greift, seinem Gewicht = schwerer Masse proporti nal. Aber wihrend die elektrische Ladung ponderab-

ler Kérper in keiner Weise mit ihrer tragen Masse

musammenhangt, stimmt merkwiirdigerweise ihre schwere Masse, ihre Gravitationsladung, stets mit der tragen Masse iiberein.

Ist EinsTerns

Erklarung rich-

tig, welche die Schwerkraft anf eine Linie mit der Zentrifugalkraft stellt, so miissen wir die an sich unbefriedigende Vorstellung eines fest vorgegebenen Tragheitsfeldes fallen lassen und miissen statt dessen nach Differentialgesetzen suchen, welche das Tragheits- = Schwere-Feld so mit den vorhandenen Massen verkniipfen wie die Maxwettschen Gleichungen das elektromagnetische Feld mit den dasselbe erzeugenden Ladungen. Dies war das Programm, welches EINSTEIN konzipierte. Zu seiner Durchfihrung war er gezwungen, mit beliebigen Koordinaten und allgemeinen invarianten Differentialgleichungen zu operieren. Inmathematisch zwingender Weise ergaben sich dabei die Feldgleichun-

gen der Gravitation, die an Stelle des NewTonschen Gravitationsgesetzes treten. So sehr ihre mathematische Form von der Newronschen abweicht, fihren sie doch in grofer Annaherung zu den gleichen Resultaten. Nur drei kleine Abweichungen erreichen ein der Beobachtung zugingliches Ma: eine Stérung des Merkur-Perihel, die sich tiber die von den andern Planeten der Newronschen Theorie gemaf verursachten Stdrungen iiberlagert [73], die Ablenkung eines nahe an der Sonne voriibergchenden Lichtstrahles und die Rotverschiebung der Spektrallinien im Gravitationsfeld. In allen Fallen ergaben die Messungen Ubereinstimmung mit der E1stemschen Voraussage innerhalb der Fehlergrenzen. Die eklatanteste Bestatigung wird vielleicht von den Spektren jener Zwergsterne von enormer Dichte geliefert, von denen der lichtschwache Begleiter des Sirius ein Beispiel ist. Nach meiner Meinung hat der Etnsreinsche Grundgedanke, der mit einem Schlage das alte Ratsel der Bewegung lst, in Kombination mit diesen empirischen Resultaten, eine solche Durchschlagskraft, daB ich nicht glauben kann, daB man je zur speziellen Relativitats-

theorie mit ihrer festen metrischen Tragheits- und

427

Kausalstruktur zuriickkehren wird; wozu neuerdings E. A. Mite und G. D, BrrKHoFF uns verleiten wollten [14),

Die Envsternsche Lésung des Bewegungsproblems lehrt, daB es tiberhaupt nicht um den Gegensatz von absoluter und relativer Bewegung geht. LaBt man beliebige Koordinaten zu, so kann man nicht nur einen,

sondern alle in der Welt vorhandenen Massenpunkte

simultan auf Ruhe transformieren ; der Begriff der rela-

tiven hat so gut wie der der absoluten Bewegung seinen Sinn verloren. Dagegen bleibt die dynamische Auszeichnung der geodatischen Weltlinien als der reinen Tragheitsbewegung bestehen, aber das sie bestimmende metrische Feld steht in Wechselwirkung mit der Materie: Gatttets dynamische Auffassung der Bewegung erfahrt dadurch eine konkretere Deutung. Solche Spekulationen wie die von Macu, wonach die Sterne des Weltalls die Ebene des Foucaurtschen Pendels fiihren, die der Ausbildung der Theorie vielleicht Vorschub

geleistet haben,

sollte man

nicht lan-

ger mit dem niichternen physikalischen Gehalt der Theorie vermengen. Freilich ist es eine Tatsache, daB in einem geeigneten Koordinatensystem das die Gravitation mitumfassende metrische Feld wenig von dem homogenen Zustand abweicht, der durch die MvKowskische Geometrie beschrieben wurde. Gebrauchen wir mit ExsTErn das alte Wort Ather fir das metrische Feld, so wiirde dies darauf hinweisen, daB in der Wechselwirkung von Ather und Materie der Ather awar kein die Materie bewegender und von ihr unbewegter Riese,

Gott ist, aber doch ein iibermachtiger

und daB hierauf das nahe Zusammengehen

TragheitskompaB und Sternenkompa® beruht.

von

2. Mathematischer und physikalischer Ausbau.

RIEMANN hatte, nach dem Muster der Gaussschen Behandlung krummer Flichen, in der Mitte des 19. Jahrhunderts eine Infinitesimalgeometrie _n-dimensionaler Mannigfaltigkeiten ausgebildet. Dieses Werkzeugs konnte sich E1Nstetn bei der Durchfiihrung seiner Theorie bedienen. Die Linienelemente, welche einen Punkt P der Mannigfaltigkeit mit den unendlich benachbarten Punkten P’ verbinden, bilden die unendlich kleinen Vektoren des dem Punkte P zugehérigen Vektorkompasses. In der Tat erleiden bei beliebiger Transformation der Koordinaten x die Differentiale dx,, ‘die relativen Koordinaten des variablen Nachbarpunktes P’ mit Bezug auf den festen Punkt P, lediglich eine lineare Transformation, Indem RIEMANN im Unendlichkleinen die Giltigkeit der euklidischen Geometrie, das ist im wesentlichen des pythagoreischen Lehrsatzes annimmt, kann er in allgemein invarianter Weise dem Linienelement mit den Komponenten dx, eine Lange ds zuschreiben, deren Quadrat eine positive quadratische Form der dx, ist, ds? =D), 58:42; dx;;

die Koeffizienten g;; hangen vom Punkte P ab, E1nSTEIN konnte diesen Ansatz ohne weiteres fiir die vierdimensionale Welt tibernehmen, mit dem Unterschied natirlich, daB hier die quadratische Form nicht definit ist, sondern die Signatur 4 besitzt. Der symmetrische Tensor mit den 10 Komponenten g;;=g); beschreibt das metrische Feld und figuriert zugleich als Gravitationspotential, Das metrische Feld bestimmt

eindeutig die infinitesimale Parallelverschiebung eines beliebigen Vektors in P nach den unendlich benachbarten Punkten P’ und damit den a/finen Zusammenhang der Welt, welcher durch die 40 Cartstorretschen Drei-Indizes-Symbole I's, =’, (die Komponenten des Gravitationsfeldes) beschrieben wird. Aus ihnen entspringt durch abermalige Differentiation der RreMANNsche Kriimmungstensor vo Range 4 (das ist mit vier Indizes). Das Wort Kriimmung hat hier oft zu MiBdeutungen AnlaB gegeben, und man sollte in der Tat diesen Tensor lieber Vektorwirbel nennen, Fuhrt man néimlich die Vektoren des Kompasses in P durch fortgesetzte infinitesimale Parallelverschiebung lings einer nach P zuriickfiihrenden Kurve herum, so kehtt der KompaB nicht in seine Anfangslage, sondern in einer dieser gegentiber verdrehten Lage zuriick; die Vektoriibertragung ist, wie man sagt, nicht integrabel. Eben diese Drehung gibt der Vektorwirbel an. Wenn er verschwindet, hat der R1eMaNNsche Raum oder die Exvsteinsche Welt die besondere homogene, ihr durch die euklidische bzw. MinKowskische Geometrie zugeschriebene Struktur. Nach dem Ervstemschen Gravitationsgesetz ist aber dieser Tensor oder vielmehr ein daraus durch die mathematische Operation der Kontraktion hervorgehender Tensor R;, vom Range 2 nicht Null, sondern gleich dem die Materieverteilung kennzeichnenden Energie-Impuls-Tensor, multipliziert mit einer universellen Konstanten, der Gravitationskonstanten x. Ich hoffe, ich habe hier vom Aufbau der Relativitdtstheorie ein die wesentlichen Zusammenhange leidlich getreu wiedergebendes Bild entworfen. Natiirlich mufte ich simplifizieren. So einfach, wie es hier erscheinen mag, ist die Bezichung zwischen Erfahrung und Theorie nicht; so leicht hat es die Natur dem Forscher nicht gemacht, von den beobachtbaren GréBen zu den FundamentalgréBen vorzudringen, auf welchen die Theorie aufgebaut werden muB! Lassen Sie es mit dieser allgemeinen Verwahrung sein Bewenden haben, und lassen Sie mich nun zum Ausbau und dann 2u dem im Laufe der letzten Jahrzehnte vorgenommenen Erweiterungen der AR tibergehen. Beim Ausbau macht die Einfihrung von Dichte und Stromdichte von Energie und Impuls der Gravila~ tion eine gewisse Schwierigkeit. Es liegt ja geradezu in dem von Erxsrein erkannten Wesen der Gravitation, daB sich das Gravitationsfeld, die KomponentenI” des affinen Zusammenhangs, lokal ,,wegtransformi ren‘ lassen; damit miissen dann auch jene Energi Impuls-GréBen zum Verschwinden kommen. Dennoch ergibt sich durch Integration ein nichtverschwindender Totalbetrag von invarianter Bedeutung [15]. Wie die aktive elektrische Ladung eines Teilchens durch den FluB definiert werden kann, den das elektrische Feld durch eine das Teilchen umschlieBende gedachte Hiille sendet, so kann auch die akfive, die gravitationsfeld-erzeugende Masse als FluB des Gravitationsfeldes durch eine solche Hille gewonnen werden, Die Berechnung des Impulsstromes ergibt im elektromagnetischen Feld, daB die aktive Ladung 2ugleich als passive Ladung auftritt, an der die elektrischen Krifte angreifen; dasselbe Verfahren liefert im Gravitationsfeld die Gleichheit von aktiver und passiver oder schwerer Masse. Hiermit hingt eine andere wesentliche Leistung der allgemeinen Relativitatstheorie zusammen: die

428

Herleitung der Bewegungsgleichungen eines mit Ladung und Masse begabten Teilchens aus den Feldgleichungen. Gestatten Sie mir, mich einer Ausdrucksweise zu bedienen, welche die vierdimensionale Welt mit ihrem metrischen Feld durch eine ziemlich, aber doch nicht villig ebene zweidimensionale Fliche ersetzt. Darin beschreibt ein Teilchen wie ein Elektron eine feine, aber tiefe Furche. Wir wissen nicht, was diese Furche birgt, doch ihre Béschung ist uns zugiinglich. Ohne uns also Gedanken iber die innere Konstitution des Teilchens 2u machen, kennzeichnen wir das Teilchen durch das dasselbe umgebende lokale Feld. Indem wir ausdriicken, daB dieses Feld sich in den Gesamtverlauf des den Feldgleichungen unterworfenen Feldes einbettet, erhalten wir die Bewegungsgleichungen [16]. Es ist nicht richtig, daB das Wirkungsprinzip, aus welchem die Emxstetnschen Gravitationsgesetze entspringen, durch die Forderung der Invarianz (zusammen mit der Forderung einer méglichst niedrigen Differentiationsordnung) vollig eindeutig bestimmt ist. Zu der von EINsTEIN urspriinglich angenommenen WirkungsgréBe kann ein 2weites, besonders einfach gebautes Glied, mit einer willktirlichen Konstanten A multipliziert, hinzugefiigt werden. E1NsTe1n fihrte dieses , kosmologische Glied' zuerst ein, um den schon von der Newronschen Theorie her bekannten Schwierigkeiten zu entgehen, die sich aus der Annahme eines im groBen ganzen gleichfrmig mit Sternen erfiillten unendlichen Weltraums ergeben. Spiter hat er dieses sein Kind wieder verleugnet; aber man wird es wohl in der Diskussion der kosmologischen Fragen zulassen miissen, solange kein zwingender formaler oder empirischer Grund fir seine AusschlieBung ersichtlich ist. Die Ohnmacht der Gravitation im Haushalt der Atome wurde am Beginn durch eine reine Zahl 10! ausgedriickt. Die ungewohnliche GréBenordnung dieser Zahl hat zu Spekulationen AnlaB gegeben, die sie mit dem MiBverhiltnis zwischen Ausdehnung oder Masse der Elementarteilchen einerseits, des Universums andererseits, und damit letzten Endes mit der zufalligen Anzahl der in der Welt vorhandenen Teilchen zusammenbringen, oder die in der Gravitationskonstanten x eine von dem Alter des Universums abhangige und mit ihm vernderliche GréBe sehen, Aber dies sind Fragen, deren Diskussion ich gerne meinem Nachfolger an diesem Pult tberlasse. 3. Versuche einer einheitlichen Feldtheorie.

Die Maxwettschen Gleichungen fiir das elektromagnetische Feld im leeren Raum flieBen aus einem sehr einfachen Wirkungsprinzip, das sich sofort von der speziellen in die allgemeine Relativitatstheorie iibertragen laBt. Aber beide Felder, das metrische oder Gravitationsfeld und das elektromagnetische, stehen unverbunden nebeneinander. Es entstand natiirlicherweise das Desideratum einer einheitlichen Feldtheorie, welche alle Erscheinungen umspannt. Von vornherein verband sich damit die Hoffnung, durch eine solche Theorie auch die atomare Konstitution der Materie erklaren zu konnen. Noch vor der Entstehung der AR und mit Beschrankung auf die elektromagnetischen Erscheinungen hatte Gustav Mir 1912 das Programm einer reinen Feldtheorie der Materie entworfen. Das Ziel, das ihm vorschwebte, war, die Maxwettschen Gleichungen so zu modifizieren, daB sie

eine oder wenige singularitatenfreie statische kugelsymmetrische Lésungen besitzen; diese wiirden dann dem Elektron und den Atomkernen der in der Natur

vorkommenden Elemente entsprechen.

Davin Hit-

BERT hatte zur selben Zeit, als ErvsTEIN seine Grundgleichungen des Gravitationsfeldes aufstellte, dieses Miesche Programm auf die allgemeine Relativitats-

theorie iibertragen [17]. E1nsTEmN selber war weise genug, in seiner Fassung der Gravitationsgleichungen dem Beispiel der Newronschen Theorie zu folgen: wie hier in der Gleichung A® —ko fiir das Gravitations-

potential ® auf der rechten Seite die (mit der Gravitationskonstanten & multiplizierte) Massendichte 9 auftritt, so stellte er auf die rechte Seite seiner Glei-

chungen (deren linke der kontrahierte Kriimmungstensor ist) einen Energie-Impuls-Tensor, der, wie er sagt, , eine formale Zusammenfassung aller Dinge war, deren Erfassung im Sinne einer Feldtheorie noch pro-

blematisch war. Natiirlich, fiigt er hinzu, ,,war ich keinen Augenblick im Zweifel, daB diese Fassung nur

ein Notbehelf war‘ [18]. Viele Versuche sind seither unternommen

worden,

zu

einer

einheitlichen

Feld-

theorie zu gelangen, insbesondere auch von EINSTEIN

selbst. Ich glaube nicht, daB das Ziel erreicht ist, oder auch nur, da® wir dem Ziel in den letzten drei Dezen-

nien wesentlich naher gekommen sind. Jede die Gra-

vitation mitumfassende Feldtheorie, welche die Atome nicht als Femdkérper einfithrt, steht dem Ratsel der

reinen Zahl 10" gegeniiber, des Verhiiltnisses von elektrischem und Gravitations Radius des Elektrons. Den-

noch méchte ich mit einer kurzen Ubersicht tiber diese Versuche mein Referat beschlieBen.

Ich strebe keine

Vollstindigkeit an. Insbesondere soll die von E1Nstern eine Zeitlang verfolgte, aber dann aufgegebene Idee des Fernparallelismus unberiicksichtigt gelassen werden,

kommt.

durch

weil sie fast einer Riickkehr zur SR gleich-

die

Im iibrigen unterscheide ich drei Gruppen Stichworte:

Preisgabe der Symmetrie.

Eichinvarianz,

Affintheorie,

a) Eichinvarianz. Die mathematische Aufgabe, als welche Mrz und Hirgert das Problem angriffen, war die Bestimmung aller Invarianten, die von den

vier elektromagnetischen Potentialen g und ihren

ersten Ableitungen sowie von den 10 Gravitations-

potentialen g;; und deren ersten und zweiten Ablei-

tungen abhingen.. Unter ihnen, nabmen sie an, miisse sich die WirkungsgréBe befinden. Aber die Auswahl war grof; es galt, ein Prinzip zu finden, das darunter eine engere, womidglich eine eindeutige Wahl traf. Sprecher glaubte 1918 dies im Prinzip der Eichinvarianz gefunden zu haben [9]. Beim Herumfahren eines Vektors lings einer geschlossenen Kurve durch fortgesetzte infinitesimale Parallelverschiebung kehrt

dieser im allgemeinen in einer andern Lage zuriick;

seine Richtung hat sich gedndert. Warum nicht auch

seine Lange? Dies war mein Einfall. Ich nahm also, an die Relativitat der Lange glaubend, an, da8 ein

willkiirliches Eichma® zur Messung der Langen von

Linienelementen lokal festgelegt werden muB, und daB wohl eine infinitesimale Cbertraeun desselben von

Weltpunkt zu Weltpunkt statt hat, daB aber diese so wenig integrabel zu sein braucht wie die Parallelliibertragung der Richtungen von Vektoren. Es zeigte

sich dann, daf zur Beschreibung des metrischen Feldes

neben dem Tensorfeld g,; noch ein Vektorfeld ¢, notig

ist,

daB

aber

Invarianz

statt hat bei gleichzeitiger

429 Ersetzung der g;; durch e~*-

g;; und

der g,

durch

ate, wo A eine willkiirliche Ortsfunktion in der Welt ist (,,Eichinvarianz“). Da man wei8, daB eine solche Willkiir wie die durch die Substitution

aa

>

M+ om

(2)

ausgedriickte in den elektromagnetischen Potentialen steckt — eine Erfahrung, welche Miz und H1pert beim Aufbau ihrer Theorie ausdriicklich verleugnet hatten, — schien es plausibel, diese y, mit den (in einer unbekannten kosmischen Einheit gemessenen) elektromagnetischen Potentialen zu identifizieren. In der Tat ergab das Wirkungsprinzip, das durch die Forderung der Eichinvarianz wenigstens nahezu eindeutig festgelegt ist, daB die y, diese Rolle spielen. Die resultierenden Gleichungen sind den E1NsTEINMaxwettschen Gleichungen geniigend hnlich, um das erkennen zu lassen, weichen aber doch geniigend davon ab, um der Hoffnung Raum 2u geben, dab sie singularitatenfreie statische kugelsymmetrische Lésungen gestatten.

Die entgegenstehenden mathe-

matischen Schwierigkeiten haben es freilich yerunméglicht, dies zu entscheiden; aber in keiner der noch zu erwahnenden konkurrierenden Theorien steht es damit besser, und darum sind sie alle physikalisch ohne Frucht geblieben. E1NsTerN machte sogleich den Einwand, daB mein Prinzip von der Nichtintegrabilitat der Langeniibertragung mit der absoluten Stabilitat der Frequenzen von Spektrallinien in Widerspruch stehe. Die Definition des MaBfeldes im Ather mit Hilfe von wirklichen Mafstaben und Uhren kann natiirlich nur als eine vorlaufige Ankniipfung an die Erfahrung gelten, Erst wenn die physikalischen Wirkungsgesetze aufgestellt sind, muf man aus ihnen ableiten, in welcher Beziehung die an jenen Kérpern abgelesenen MeBresultate zu den FundamentalgréBen der Theorie stehen. Die Erfahrungen, auf die sich ErNSTEIN mir gegentiber berief, zeigen gewiB, daB die physikalisch gemessenen Langen nicht der kongruenten Verpflanzung von Strecken folgen, die zum Fundament meiner Theorie gehért. Ich habe keine Lust, diese Theorie, an die ich langst nicht mehr glaube, zu verteidigen. Aber ich konnte damals doch mit Recht auf das Faktum hinweisen, daB sie im Kriimmungsradius der Welt, sozusagen nachtraglich, ein absolutes lokales EichmaB liefert, auf das sich spektrale Frequenzen und andere LAngengréBen einstellen kénnen und vielleicht gemaB dem geltenden Wirkungsprinzip wirklich einstellen. Heute, nach Einfiihrung der ScrépINGER-Dirac-

schen y, durch die Quantentheorie, glaube ich, kénnen wir mit

groBer

Bestimmtheit

den

Finger

auf den

Punkt legen, in welchem meine Theorie irrte: die Eichinvarianz verbindet die elektromagnetischen Poten-

fiale nicht mit den g,, der Gravitation, sondern mit den y, des Materiefeldes.

Das konnte ich freilich 1918

nicht ‘wissen! Damals waren diese

noch véllig un-

bekannt. Im Rahmen der AR wird der willkiirliche Phasenfaktor e~‘4, der den yp anhaftet, von einer Konstanten zu einer willkiirlichen Ortsfunktion in der

Welt. Es muB dann notwendig dem Differentialoperator 0/@x,, um ihm eine invariante Bedeutung zu

sichern, die allgemeinere Form 7 + ig, gegeben

werden, wobei die g, ein Vektorfeld bilden: verwandelt

man y, in e~**- y,, so geht p, ing + ae iiber. Genau

Giese Vorstiaifeist co aber, nach welcher tie Dinacsche Theorie die Einwirkung des elektromagnetischen Feldes auf das Elektron wiedergibt, wenn g, als das mit e/h multiplizierte elektromagnetische Potential gedeutet wird. Hier sind wir nicht im Gebiet der Spekulation, sondern der Erfahrung, und die Einheit, in welcher die g gemessen werden, ist nicht eine unbekannte kosmische, sondern eine bekannte atomare GréBe. Man sollte freilich jetzt lieber von Phasen- statt von Eichinvarianz sprechen. Die durch (2) zum Ausdruck kommende Unbestimmtheit in den elektromagnetischen Potentialen 9 ist jedenfalls, auch wenn man diep mit keinen anderen GrdBen verkniipft, eine gesicherte Tatsache, und die Invarianz gegeniiber der Substitution (2) mit der willkiirlichen Ortsfunktion A ist auf die gleiche Weise mit dem Gesetz von der Erhaltung der Ladung verkniipft wie die Invarianz gegeniiber Koordinatentransformation mit dem der Erhaltung von Energieimpuls.

Dem Umstand, daB nicht die Potentialeg,, sondern nur die daraus abgeleiteten FeldgréBen

eine physikalische Bedeutung haben, kann man innerhalb des Miz-Hnperrschen Schemas Rechnung tragen und dadurch wenigstens eine gewisse Einschrinkung in der Auswahl der zur Verftigung stehenden invarianten WirkungsgréBen erzielen. So verfuhr Born [20]. Statt der Maxwetrschen WirkungsgroBe L schligt er insbesondere eine vor, welche unter Vernachlssigung der Gravitation so lautet: yi+2

()

(B ist eine kleine Konstante). Damit errang er wenigstens einen partiellen Erfolg, insofern die statischen kugelsymmetrischen Lésungen seiner Gleichungen zwar nicht singularitatenfrei

endlichen Energie fihren. Katuza

hatte

1921

den

sind, aber doch zu einer

Gedanken,

ob sich nicht

die Invarianz gegentiber der Substitution (2) als Er-

weiterung der Invarianz gegeniiber Koordinatentrans-

formation auf eine fiktive 5. Weltkoordinate x) deuten

lieBe [21}. Er machte die spezielle Annahme, daB die Koordinaten %, %», %3, %4 Sich wie bei EINSTEIN nur

untereinander transformieren, wahrend fiir x» ein be-

liebiges Transformationsgesetz von Form

Ky

der besonderen

Kot A (ys Mar Xa, a)

(4)

mugelassen wird. Setzt man dann eine quadratische Differentialform der fiinf Variablen fiir die Beschreibung des metrischen Feldes an, ds* = Sra, p8ap4%qd%,

(a,8 =0,1,2,3,4),

so stellt sich heraus, daB gy9 eine Invariante ist, die Katuza durch go) —1 normiert (dies scheint zulassig, wenn man annimmt, da8 nicht die Form ds? selber, sondern nur die Gleichung ds =0 eine physikalische

430 Bedeutung hat), wahrend die vier GréBen gy =£,o sich gegentiber Transformationen von %,, %», %j, %4 $0 wie die Komponenten eines Vektors verhalten, bei der Transformation (4) aber die Substitution (2) erleiden. (Der Index z lauft hier immer nur von 1 bis 4.) Man macht die zusitzliche Annahme, daB alle g, Von % unabhangig sind. Man kommt so in der Tat auf natiir-

liche Weise

auf

die Maxwett-Ernsternschen

Feld-

gleichungen, und die Bewegung nicht nur von ungeladenen, sondern auch von geladenen Teilchen verlauft

lings geodatischer Weltlinien. Dennoch liegt hier kaum mehr vor als eine formale Zusammenfassung der beiden

Felder,

die zu

keinem

Erkenntnisfortschritt

fiihren kann, da sie dem von Born korrigierten MizHirsertschen

Schema

keinerlei

Einschrinkungen

auferlegt. Eine ansprechende geometrische Einkleidung des Formalismus liefert die projektive Geome-

trie. Anstatt der vier Weltkoordinaten x, benutzt der

projektive Geometer die durch x, —X,/X, eingefiihrten fiinf homogenen

Koordinaten X,.

Ein Punkt be-

stimmt nur die Verhiltnisse dieser Koordinaten; indem man diese selbst festlegt, erteilt man einem Punkt ein Gewicht, Setzt man X,—=e%, so bedeutet der Umstand, daB x,, %, %, % sich nur untereinander transformieren, dies, daB das Zusammenfallen von Punkten

verschiedenen Gewichts eine invariante Bedeutung hat, wahrend die Transformation

Gewichts

ausdriickt.

Neben

(4) die Willkiir des

Karuza

hat

Oskar

Kern diese Ansitze verfolgt; er ist spater, im Zusammenhang mit quantentheoretischen Erwigungen, dazu tibergegangen, die Annahme, daB ZustandsgréBen

und Transformationsfunktionen einschlieBlich 4 von x»

nicht abhingen, dahin zu verallgemeinern, daB sie periodische Funktionen von x, sind, mit einer durch die universellen Naturkonstanten vorgeschriebenen Periode [22]. Die angedeutete projektive Form wurde, von 1930 ab, ausgebaut von vAN DANTzIG und ScHouTEN, ferner von Paut. Sie war in einer etwas anderen Gestalt schon vorher von O. VEBLEN und B. HorrMANN entwickelt worden. Pavitt hat auch die Aus-

dehnung der Theorie auf die y-GréBen verfolgt. E1NSTEIN zusammen mit W. MAyeER hat in den gleichen Jahren eine nahverwandte Theorie konstruiert, deren

Formalismus gleichfalls fiinf unabhangige Variable benutzt [23]. Aus jiingster Zeit waren Arbeiten von PascuaL JORDAN zu erwihnen.

b) Reine und gemischte Affintheorien. EDDINGTON dehnte meine suggestion, da8 der Kriimmungsradius der Welt das Eichmaf liefert, von der skalaren Kriimmung auf den Kriimmungstensor R,; aus und wurde so dazu gefiihrt, der Welt von Hause aus keine Metrik, sondern einen durch 40 GréBen I, =", ausgedriickten affinen Zusammenhang zuzuschreiben, In der Tat entstehen die KriimmungsgréBen R,, aus ihnen allein, Ervstem; Gravitationsgleichungen im leeren Raum, R,,=0, hatten sich durch Hinzufiigung des kosmologischen Gliedes in Ry =A Bi;

verwandelt; diese werden nun fiir EDpINGTON aus einem Naturgesetz 2u einer Definition des metrischen Tensors g,;. Diesen Gedanken aufgreifend, wies EINstein alsbald darauf hin, da8 dann die Gleichungen, welche die KomponentenI’ des affinen Zusammenhangs durch die-g,; ausdriicken, nicht langer als Defi-

nitionen aufgefaBt werden kénnen, sondern aus einem Wirkungsprinzip abzuleitende Naturgesetze sein miissen. Under fand in der Tat, daB sie sich ergeben, wenn man die einfachste Invariante, die fie Rahmen der Eppincronschen Affintheorie méglich ist, als WirkungsgréBe wahlt; das ist die Quadratwurzel aus der Determinante der R,; [24]. Es treten dabei auch Terme auf, die sich als elektromagnetisches Potential deuten lieBen, und die resultierenden Gleichungen sind einschlieBlich der kleinen kosmologischen Glieder mit ihren numerischen Koeffizienten genau mit den Feldgleichungen meiner metrischen Theorie identisch. Es ist mir schleierhaft, worauf diese merkwiirdige Ubereinstimmung beruht [25]. In dem Dilemma, ob man der Welt urspriinglich

eine metrische oder eine affine Struktur zuschreiben soll, ist vielleicht der beste Standpunkt der neutrale, der sowohl die g wie die I’ als unabhingige ZustandsgréBen behandelt.

Dann

werden

die beiden

Satze von

Gleichungen, welche sie verbinden, zu Naturgesetzen, ohne daB die eine oder andere Halfte als Definitionen eine bevorzugte Stellung bekommen. In der Tat

zeigte Ernstety, daB ein Wirkungsprinzip von beson-

ders simpler Bauart hier dieselben Gesetze liefert wie seine urspriingliche rein-metrische Theorie [26]. Freilich fiihrt dieser neutrale Standpunkt auch nicht tiber die rein-metrische Theorie hinaus, selbst nicht bei Einbezichung des y-Feldes der Elektronen. ¢) Preisgabe der Symmetric. Ein beliebiger Tensor fy vom Range 2 spaltet in invarianter Weise in einen symmetrischen und einen schiefsymmetrischen Bestandteil:

Wa=ti ths Diese kann

man

gta

te=—hi

bzw. als Gravitationspotential

(6) und

elektromagnetisches Feld deuten, die so zu einem einzigen Tensor h,; zusammengefaBt erscheinen. Bei der Aufstellung seiner besonderen WirkungsgréBe des elektromagnetischen Feldes (3) war schon BORN von diesem Gedanken ausgegangen [27]. Systematisch haben dann Ernste1n und ScuR6DINGER untersucht,

was geschieht, wenn man fiir die Ji, (wie auch fiir die g,;) die Symmetrieannahme Tj, =Tj, fallen 1aBt. SCHRODINGER stellt sich dabei auf den rein-affinen, E1nstern auf den neutralen metrisch-affinen Stand-

punkt [28]. ScHROpINGER glaubt durch seine Theorie zum mindesten eine Art von Mesonen mitzuumfassen,

erwartet aber wohl, daB das ganze System von Feld~

gleichungen erst dem QuantisierungsprozeB unterworfen werden muB, ehe es die atomaren Erscheinun-

gen zu erklaren fahig ist. EINSTEIN nahrt in seinem

Busen noch immer die kithne Hoffnung, da8 die Feldgleichungen selber ohne quantentheoretische Umdeutung dies leisten. Ich gestehe, daB ich als Mathematiker mir von

einer so formalen Verallgemeinerung wie dem Fallen‘ lassen der Symmetriebedingungen nichts versprechen kann, Die Symmetrie der g,; und der Ij, hat eine tiber das Formale weit hinausgehende Bedeutung, namlich die, daB die Natur der Metrik und des affinen

Zusammenhangs eine und allerorten die gleiche ist. Statt an der Symmetrie zu riitteln, sollte man nach

einer andersartigen reicheren Struktur fahnden, deren Natur aber wiederum iiberall die gleiche sein miiBte.

431 concerning Human Understanding, Book II, Chap. 13, Sections — [3] ,,Dialogo sopra i dui massimi sistemi del mondo", physikalischer Theorien wichtige Lehre entnehmen in7-10. Bd. VII, S. 198, der Opere, Edizione nazionale. Florenz 1890 kann, so ist es die, daB nur GroBen, die unter ihrem bis 1909. Neudruck 1929—. — [4] Siehe z.B.S. 10—12 der englischen spezifischen Transformationsgesetz unzerlegbar sind, Ausgabe der Philosophiae naturalis principia mathematica von F. Cajori, Berkeley, Calif., 1934, zweiter Druck 1946; auch das den eine einheitliche physikalische Entitat darstellen ; eine Definitionen I, II und IV folgende Scholium auf S. 6—7; betrefis, solche GréBe ist der symmetrische und der schiefder Beziehung zu Newtons Theologie: Cores’ Vorrede zur 2. Aull symmetrische Tensor, g und , aber nicht ihre Zusam- der Principia, S. XXXII, (Cajori) und Principia, S.46, (Cajori); ferner menfassung (5). PAULI formuliert dieses Prinzip so: ‘Newrons Optics, S. 370, (Ausgabe von E.T.Whittaker, London 1931) und der Briefwechsel zwischen Leibniz und S, Clarke in G. W. LersWas Gott getrennt hat, soll der Mensch nicht zusamPhilosophische Schriften, Bd. VII, S. 352—440, (Ausgabe von Germenfiigen. Durch die Preisgabe der Symmetrie ist N12, hardt).— (6] Principia, S. 419, Ausgabe von Cajori.— (6] Brunos Del die Mannigfaltigkeit der als WirkungsgréBe zur Ver- infinito universo e monci erschien 1584, die erste Aufl. von Newtons fiigung stehenden Invarianten gewaltig gewachsen, Principia 1687. — [7] Opere, ed. naz., Bd. VII, S, 212214. — {[s] Uber die Entwicklung bis 1920 orientiert am besten der Artiwahrend doch das Bestreben sein sollte, die Méglichkel V 19: Relativitatstheorie von W. Pautt, in der Enzyklopadie keiten einzuschrinken. Offenbar sind wir doch nicht der mathematischen Wissenschaften, Bd. V, Teil 2, S. 539—775, kdug genug, um durch reines Denken — ,,aus dem hoh- 1904—1922 (zitiert als ,,Pauli, Enc."’). — [9] Ich entnehme diese len Bauch“, glaube ich, war frither der Ausdruck der Zablangaben dem schénen Artikel von Max von Lave Uber Inertia and Energy, AE, S. 503—533. Andere frihe Hinweise auf Miinchener Physiker dafiir — die universelle Struktur kernphysikalische Anwendungen siehe bei Pavus, Enz., S. 681. — der Welt und die sie beherrschenden Feldgesetze zu [10] Die grundlegenden Arbeiten von HetseNBeRc und von ihm, finden. Und ich glaube auch nicht, da8 unser gegen- Born und JorpAN erschienen in der Z. Physik 33—35 (1925/26); wartiges Wissen tiber die Wellenfelder der Elementardie von E. Scuréprncer sind zusammengefabt in den Abhandiungen zur Wellenmechanik. Leipzig 1927. — [11] Proc. Roy. Soc. teilchen dafiir irgend zureichend ist. Hier wie andersA 117, 610; 118, 351 (1928). — [12] Siehe AE, S. 52. — wo ist dafiir gesorgt, daB unsere Baume nicht in den Lond. [13] Siehe die genaue Diskussion in dem Vortrag tiber Relativity Himmel (oder in die Hille) wachsen. Effects in Planetary Motion, den G. M. Crewence im Rahmen des Princetoner Symposium aus Anla8 von Einsteins siebzigstem GeUm aber nicht mit bloBer Kritik zu enden, will burtstag, Proc. Amer. Philos. Soc. 93, 532—534 (1949). — ich zum Schlu8 noch mein Scherflein zur Spekulation [14] Siehe hielt: fiir Mune: AE, S.415—435; fiir Brrxnorr: Proc. Nat. beitragen [29]. Die Maxwetrschen Gleichungen, in Acad. Sci. USA 29, 231 (1943) und die Bemerkungen von H. Weve mégliche lineare Theorien der Gravitation in Amer. J. Math. denen die Potentiale p als die unabhangigen Zustands- liber 66, 591—604 (1944). — [15] Etnsretw: Sitzgsber. preuss. Akad. groGen figurieren, sind linear, und es besteht Invari- Wiss., Math.-naturwiss. Kl. 1918 448. — (16) Wevi, H.: Raum, anz gegeniiber der Substitution (2). EINSTEINS Gra- Zeit, Materie, 5. Aufl. Berlin 1923, § 38. Mit viel groBerer Sorgfalt, vitationstheorie ergibt fiir die 10 Potentiale y,,=y;; auch 2ur Bestimmung der Wechselwirkung mehrerer Teilchen, wurde Weg dann in mehreren Arbeiten von Exstem und INFELD eines unendlich schwachen Gravitationsfeldes eben- dieser beschritten; vgl. die letzte abschlieBende Arbeit, die im Canadian J. falls lineare Gleichungen, und diese sind invariant Math. 1, 209—241 (1949) erschien. — [17] Mre, G.: Ann. Physik gegeniiber der zu (2) analogen Substitution 37, 39, 40 (1912/13). — Hivperr, D.: Die Grundlagen der Physik. Nachr. Ges. Wiss. Gottingen, Math.-physik. KI. 1915 u. 1917. — [18] AE, S. 74. — 19] Vgl. die Darstellung in Raum, Zeit, Materie, non + (BE ae ae) 5. Aufl, S. 298—308. — [20] Borw,M.: Proc. Roy. Soc., Lond. A Ox; 143, 410 (1934).— ScurOvincER, E.: Proc. Roy. Soc., Lond. A 150, 465 (1933). — [21] Katuza: Sitzgsber. preuss. Akad. Wiss., Physik. mit vier willktirlichen Funktionen &,; eine Invarianz- math. Kl. 1921, 966. — (22] Kuetw, O.: Z. Physik 37, 895 (1926); 46, 188 (1927). — Vel. ferner: O. Kiem, Ark, Mat., Astronom. Fysi eigenschaft, die die Koordinateninvarianz der strengen Ser. A 34, Nr 1 (1946). Weitere Publikationen stehen in Aussicht. — Gleichungen widerspiegelt. Vielleicht sollte man zu- [23] ScHouren Proc. Amsterdam 34, 1398 (1931).— nachst einmal auf diesem linearen Niveau nach einer Z. Physik 78, 639u. vax(1932).Dantz1G: VAN: Math. Ann. 106, 400 Vereinigung von y,; und g, fahnden. Die Quanten- (1932). — Pavrs, W.: Ann.— Danrzic, Physik (5) 18, 305—372 (1933). — O., u. B. HOFFMANN: Projective relativity. Physic. Rev. theorie 14Bt die elektromagnetischen Potentiale y als Vepten, 36, 810-822 — Veten, O.: Projektive Relativitatstheorie, ein vom Prinzip der Phaseninvarianz gefordertes An- Ergebnisse der(1930). Mathematik, Bd. 11/1. Berlin 1933, — E1nsrer, A., hangsel an die das Elektron darstellenden FeldgroBenp u. W. Maver: Sitzgsber. preuss. Akad. Wiss., Math.-naturwiss. Kl. erscheinen, Frage: Sind die ,, in analogem Sine ein 1931, 541-557; 1932, 130-137. — [24] Einstein: Sitzgsber, preuss. Akad. Wiss., Math.-naturwiss. Kl. 1923, 32, 76, 137. — Appendix an das Wellenfeld X eines unbekannten Eppinorow, Theory of Relativity, 2. Aufl., Elementarteilchens ,,Graviton? Erst nachdem diese Note 14. — [25]A.S.:Vgl. Mathematical zu allen diesen H. Wey, GeoFrage beantwortet ist, sollte man jenen Ubergang zur metrie und Physik. Naturwiss, 19, 49~58Ausfihrungen (1931).— [26] EINSTEIN, A. nichtlinearen Theorie versuchen, bei welchem die ,; Sitzgsber. preuss. Akad. Wiss., Math.-naturwiss, Kl. 1928, 414. — H.: Physic, Rev. 77, 699-701 (1950). — [27] Born, M.: sich in die wirklichen ZustandsgrdBen g,, des metri- Wer, Proc. Roy. Soc., Lond. A 144, 425—454 (1934). — [28] Von Scurd: der ProzeB, schen Feldes (zuriick-)verwandeln; ein pINGER zitiere Arbeiten betitelt ,,The final affine Field dann notwendig auch die g (samt den y und dem Laws‘ in Proc. Roy.ich dieIrishdrei Acad. A 51, 163171, 205—216; 82, 1-9 unbekannten X) mitergreifen wiirde und so in organi- (1947/48); voraufgehende Arbeiten sind dort angefihrt. Besonders ist die Ubersicht der verschiedenen Theorien in der 2. Abh. scher Weise eine nichtlineare Theorie des MaxwELt- niitzlich Einsteins letzte Version seiner Theorie findet man im Appendix I schen Feldes ergibe. Ich bin weit davon entiernt, von The Meaning of Relativity, 3. Aufl., S.109—147. Princeton, dieses Programm durchfiihren 2u kénnen. N. J. 1949. Vorbereitet war sie durch die Arbeiten von ErstEI und Straus in NS, Math 46, 578 (1945) und 47, 731 (1946). Nach Abschlu8 dieses Artikels erschien E, Scardpincers Buch Literatur. Space-Time Structure. Cambridge 1950.— [29] Wevt, H.: Amer. [1) Uber die Entwicklung Yon Exxsrets Iden vgl. seine ,,Auto- J. Math, 66, 602 (1944). biographical Notes‘ in Albert Einstein Philosopher-Scientist, Bd. VII der Library of Living Philosophers, herausgeg. von Paut A. Scuitrr. Evanston, Ill., 1949. (In der Folge zitiert als AE.) — (2] Enquiry

‘Wenn man der Mathematik

eine fiir die Aufstellung

150. Ramifications, old and new, of the eigenvalue problem

Bulletin of the American Mathematical Society 56, 115—139 (1950)

Since this is a lecture dedicated to the memory of Josiah Willard Gibbs let me start with that purely mathematical discovery which Gibbs contributed to the theory of Fourier series. Fourier series have to do with the eigenvalues and eigenfunctions of the oldest, simplest, and most important of all spectrum problems, that of the vibrating string. In preparing this lecture, the speaker has assumed that he is expected to talk on a subject in which he had some first-hand experience through his own work. And glancing back over the years he found that the one topic to which he has returned again and again is the problem of eigenvalues and eigenfunctions in its various ramifications. It so happens that right at the beginning of my mathematical career I wrote two papers on what we now call the Gibbs phenomenon. 1. Gibbs phenomenon. Take a simple periodic function with a discontinuity, for example, the function 1°(x) of period 27 which equals letter to the editor of Nature 0 for —r0 and q(s) are given real continuous functions in this interval. Let the abbreviation ¢/ be used for p(s)dp/ds. A real linear boundary condition for s=/ is of the form

435,

(2)

¢'(l) — h-4(l) = 0

with a real constant A (not excluding h= ©). For a solution ¢ of Iy(¢) =0 and a solution $x of Zy,(d+) =0 we have the simple Green’s formula

@)

[oon — b40} = O— 4) f dbads. t

0

It shows that eigenfunctions ¢, ¢« belonging to two distinct eigen-

values

\, Ax

are

necessarily

orthogonal,

/}¢dx-ds=0.

On

taking

dx =X, dx =¢ one finds that for a non-real \ the function ¢(s) cannot

satisfy a real linear boundary condition at both ends without vanish-

ing identically;

for under

these

circumstances

our equation

would

must

that

give J¢¢¢ds=0. The positive-definite character of the integrand $6 is decisive here. The spectrum of the eigenvalues is discrete provided the differential equation is regular at both ends, that is, provided p(s), q(s) are continuous and (s) actually positive throughout the closed interval 0 $s Sl. If we make this assumption only for the right-open interval 0 1/2; it seems difficult to go beyond the vertical ¢ =1/2. Standard devices familiar from the theory of Riemann’s {-function permit one to deduce from this behavior of the {-function of the membrane Carleman’s asymptotic formulas

D @(P))?~d/4n,

AnZd

NO) = DY 1~ VA/4e AnSd

and also the “incoherence relation”

bn(Q) = 0(d) $n(P) DX And

for P #0.

I feel that these informations about the proper oscillations of a membrane, valuable as they are, are still very incomplete. I have certain conjectures on what a complete analysis of their asymptotic behavior should aim at; but since for more than 35 years I have made no serious attempt to prove them, I think I had better keep them to myself.

In general, it can not be expected

that our {-function satisfies a

functional equation of the Riemann type; one may guess that this

449

feature depends on the homogeneity of the domain of integration. Such a domain is the circumference of the unit circle. Functions on it are functions f(x) of period 27. The periodic eigenfunctions ¢ and corresponding eigenvalues d of d’p/dx?+d=0 are

o(%) =e",

dd, = n?

This leads straight to the Riemann

(r= ¢-function

OPE £2)

)°?_.-%

=)

usually de-

noted not by ¢(s) but by ¢(2s). It was therefore natural that Minakshisundaram should investigate the spherical harmonics on a k-dimensional sphere (in k+1-dimensional space). Here he found indeed a sort of Riemann functional equation, the structure of which is, however, essentially more complicated than in the classical Riemann R=1. case The two-sphere is homogeneous because it permits a compact transitive Lie-group o of transformations s into itself, namely the group of rotations. The spherical harmonics of order / form a (2/-++1)dimensional linear manifold that is invariant with respect to the group of rotations and has the property of irreducibility in this regard. Consider arbitrary (complex-valued) continuous functions on the sphere and define the scalar product of two such functions f and g by the integral [Z(P)f(P)-dwp formed by means of the invariant area element dw . It is obvious how to generalize this situation to any homogeneous manifold S of points P, that is, any manifold that permits a compact transitive Lie-group o of transformations s, PsP. The existence of an invariant volume element on such a manifold (which itself is of necessity compact) follows easily from the fact that acompact Lie-group has an invariant volume element ds. We normalize the unit for measuring volumes on the group so that the total volume fds of the group becomes 1. The integrals with respect to s are then in truth mean values. The transform sf of a function f=f(P) on S is defined by sf(sP) =f(P) or sf(P) =f(s-P). A set $i(P), ---, ¢n(P) of functions on S, or the manifold of their linear combinations O(P) =x1-gi(P)+ - > + +xn-ba(P), is invariant if each s¢,(P) is a linear combination )\,w;(s)-¢.(P) of the ¢; themselves. Then s||o.(s)]| is a representation of degree h of the group a. Inequivalent irreducible invariant sets are orthogonal to each other. Besides orthogonality there is the completeness relation, to which we shall presently return. Thus we are in possession of the “eigenfunctions” of the homogeneous manifold, the sequence of which is subdivided into irreducible invariant sets of finite length. Theorems about summability of expansions in terms of these eigenfunctions have been proved by S. Bochner [20]. But so far they are eigenfunctions with-

450

out eigenvalues. It was the young Dutch physicist H. B. G. Casimir who, prompted by the applications of group theory to quantum mechanics,

found

the

eigenvalues.

Indeed

he

constructed

an

in-

variant self-adjoint differential operator A working on arbitrary functions f(P) which is the analogue of the Laplace operator on a sphere, and he was able to show that the functions of a given irreducible invariant set satisfy an equation A¢+A¢=0 with a constant » characteristic for the entire set [21]. Having the eigenfunctions, one can, following a suggestion by Bochner, form the {-function

8, 0:2) = E

n(P) bn ROBO

and might expect that this function, in addition to having the properties quite generally established by Minakshisundaram, will satisfy a functional equation of Riemann’s type. But this is a question that remains to be investigated. 6. Integral equations and the group-theoretic completeness relation. The proof of the completeness relation for invariant sets on a homogeneous manifold S is one of the most surprising applications of the eigenvalue theory of integral equations. If the manifold S is the compact Lie group itself under the influence of its left translations, then this theorem states the completeness of the totality of all irreducible representations of the group. But the method for its proof, developed in 1927 by F. Peter and the speaker [22], not only carries over to the homogeneous manifolds, but applies to a far more general situation, that is best described in axiomatic terms [23]. We replace the functions on the homogeneous manifold by vectorsf in a vector space 2 and suppose that 2 bears a Hermitian metric defined by a scalar product (g, f) with the usual properties including the positive character of (f,f) = llell2. An abstract compact Lie groupc is given and a representation of its elements s by linear transformations f—sf in our vector space. The invariance of the metric is assumed, (sg, sf) =(g,f). Let f be a given vector. All vectors that will occur in our con-

struction are prepared from f by forming linear combinations of its transforms sf. Besides 2 we envisage the “vector space” & of all continuous functions = &(s) on the group manifold o and define a linear mapping £—g of & onto 2 by

g=fe = f &9)-sfas, and its Hermitian conjugate,

a mapping g—é of Y onto X, by

451

&= f*g = (sf, g). The mapping f*f=§ of & into itself has the positive-definite Hermitian kernel H(s, t) =(sf, tf). By Erhard Schmidt’s method we construct its largest reciprocal eigenvalue y and an orthonormal set of eigenfunctions ¢(s), - - - , @x(s) for it. Repetition of the construction gives the reciprocal eigenvalues in descending order, y>y'> ---, and the sought-for completeness relation results from the well known fact that the trace of H equals the sum of the reciprocal eigenvalues,

Il? = GD = byt hy tee,

Indeed fH(s, t)-$:(t)-dt=y-¢; may be written in the form

for =7'?-g:,

Pgs = y!?-b;

where the first equation is to be taken as the definition of the vector

gi. The g; then form an orthonormal invariant set, and if the Fourier coefficients (g;, f) =a; are introduced one finds that

TE

neous aegis

Thus the completeness relation follows, stating that the orthonormal sequence £1, £2, ° ++, (Zm, Zn) =Smn, resulting from our construction and consisting of sections of finite length, each of which is an invariant set, makes the absolute square sum Jaa| a4 | as| 24... of the Fourier coefficients a;=(gi, f) not only s|lyil2, as is trivial (Bessel’s inequality), but actually =|[sl|2. The construction picks out those invariant sets that contribute to f and ll#ll2. This feature is quite essential when, with Harald Bohr and J. von Neumann [24], all restrictions concerning the group ¢ are abandoned. The construction still works, provided one supposes f to be “almost periodic.” But in general there are under these circumstances more than denumerably many inequivalent irreducible invariant sets of vectors; but f itself picks out those among them that matter for f— The assumption of almost periodicity is highly restrictive. One may instead impose some slight restriction on the group, e.g., local compactness,

and

at

the

same

time

admit

a far wider

class

of vectors

f. In that case nothing resembling completeness is to be expected unless one includes also representations of infinite degree. This step has recently been taken by D. Rykov and I. Gelfand in Russia, by V.

Bargmann and I. Segal in this country [25].

I think it is time for me to stop here. I have not even touched on the extension of Hilbert’s theory of bounded to non-bounded linear operators, which came about under the pressure of quantum me-

452

chanics, nor to the connection between spectral decomposition and ergodic theory. Other mathematicians at other times have spoken or will speak on these subjects with more competence than I could. I hope you have taken this lecture for what it was meant to be: a Plauderei, the chat of a man who has reached the age where it is more pleasant to remember the past than to look forward into the future. Even so, it gives him a little satisfaction to see that the issues to which the efforts of his youth were dedicated have kept alive over the years and are still in the process of unfolding their implications. Notes

(A) There are two classes of eigenfunctions. But since one of them does not contribute to the expansion of the discontinuous function

1°(x), only the eigenvalues \? belonging to the other class have to be

taken into account; they are determined by the transcendental equation

adr DAE

+ a-tan

BAr ETT, = 0.

It is easily seen that for every integer 7 this equation has exactly one root X, of the form \,=22+0n, —10, involves only (1, the second f; and 2, and so on. If these inequalities are fulfilled then there are two possibilities, which are distinguishable by a convergence criterion. In the first, the limit point case, the problem has a unique solution; in the second, the limit circle case, the manifold of all solutions w(A) is obtained from that of all positive functions 2(A) by a certain Mobius transformation

w(a)

_ 4Q):20) + BA)

C(A)-2(A) + D(A)

with coefficients A, B, C, D that are regular analytic functions of \ in the upper half \-plane.

(C)

Indeed replacing Xo, X by X, X in (6) one finds 5¢(s) = on: f

i

G(s, t)-(4)-de,

453

and therefore 39(s) has a uniformly convergent expansion in terms of the eigenfunctions 4(s; \,). For the integral Sob(s)n(s; Xa) ds one obtains from (3) the value 1/(A,—A). (D) The eigenvalues X=, of Au+du=0 corresponding to the boundary condition du/8x=0 (normal derivative of u equal to zero; “acoustic eigenvalues”) are the reciprocals of the successive maxima of J(u) under the restriction D(u)=1, or the successive minima of D(z) under the restriction I(w)=1. The boundary condition du/dn=0 gets lost, as it were, in the process of closure under the metric defined by D(u). Hence unSXna, where dq, as before, are the membrane eigenvalues corresponding to the boundary condition u=0.

The equation for the oscillations of a plate, AAu



?-4

= 0

in S,

arises from minimizing

(20)

i (Au)?-dP s

under the auxiliary condition I(u)=1. the clamped plate, boundary conditions

Let uy? be the eigenvalues of

xu = 0,

Ou — = 0, on

and }%, those for the “half-free” plate,

(21)

boundary conditionsu=0,

Au = 0.

Again the boundary condition Au =0 gets lost in the process of closure under the metric defined by (20). Hence \%Su?; one can further expect that \, and yp, follow the same asymptotic law. Weinstein observed that the eigenvalues of the half-free plate (as their notation

indicates) are simply the squares of the membrane eigenvalues. Indeed if AAu=)?2u (A>0) set Au=—)dv, so that Av+du=0. The

boundary conditions (21) give w=0, v=0 along S’, hence (u+v)/2 is a membrane eigenfunction with the eigenvalue \ and (u—v)/2 for —). But the membrane has no negative eigenvalues; consequently w—v=0 and (u+v)/2=u. It was by a somewhat similar remark that I had previously reduced the elastic oscillations of a three-dimensional body asymptotically

to the three-dimensional membrane problem [26]. The vector field v(P) describing a proper oscillation of the elastic body satisfies an

454

equation (E)

a-grad

div » — b-rot rot»

+ A»

with two elastic constants a, 6. Impose (E’)

bv normal,

div

= 0

in S

the boundary » = 0

conditions

on the surface S’ of S.

For ¢ =div » one finds a-A¢-+-Af =0 in S and the boundary condition

%=0;

hence if ¢ is an eigenfunction of the membrane

the eigenvalue \/a then p=grad ¢ div »=0 throughout S we have

6-Av + Av = 0,

satisfies (E),

(E’).

div v = 0 in S;

problem

with

If, however,

» norma onlS’

(since Ap =grad div »—rot rot »), or \/d is an eigenvalue of the problem of radiation in a Hohlraum S whose wall S’ is a perfect mirror. Denoting the numbers of eigenvalues S$) of the membrane, the radiation, and the elastic problem (E) & (E’) by Nn(A), N-(A) and N,(a,b;))

respectively,

we thus find the relation

N (a,b; ) = Nn(d/a) + N,(0/2). For a=b=1 the left side of (E) turns into Av-+-Av. Since asymptotically the boundary conditions are of no influence we must have the asymptotic relation

N.(1,1;2) ~3N

(A),

that is,

Nmm(A) + N,(X) ~ 3Nn(A)

or N,(A)~2N (A) and thus

N (a, 6; X) ~ Nm(A/a) + 2Nn(X/8). BIBLIOGRAPHY 1. H. Weyl, Die Gibbs’sche Erscheinung in der Theorie der Kugel funktionen, Rend. Circ. Mat. Palermo vol. 29 (1910) pp. 308-328 ; Ueber die Gibbs ’sche Erscheinung und verwandte Konvergenzphinomene, Rend. Circ. Mat. Palermo vol. 30 (1910) pp. 1-31. 25 1 Ueber die Gleichverteilung von Zahlen mod. Eins, Math. Ann. vol. 77

(1916) pp. 313-352.

3. , Ueber gewshnliche Differentialgleichungen mit Singularitit en und die sugehdrigen Entwicklungen willkirlicher Funktionen, Math. Ann. vol. 68 (1910) pp. 220-269; Ueber gewihnliche Differentialgleichungen mit singuliren Stellen und thre Eigenfunktionen, Nachr. Ges. Wiss. Géttingen (1910) pp. 442-467. 4. E. Hellinger, Zur Stieltjes'schen Kettenbruchtheorie, Math . Ann. vol. 86 (1922)

pp. 18-29.

5. G. Pick, Ueber die Beschrinkungen analytischer Funk tionen, welche durch vorge-

gebene Funktionswerte bewirkt werden, Math. Ann. vol. 77 (1916) pp. 7-23. R. Nevanlinna, Ueber beschrankte Funktionen, die in gegebenen Punkten vorgeschriebene Werte

455

annehmen,

Annales Academiae

Scientiarium

Fennicae vol. 13, No.

1, 1919;

Ueber

beschrinkte analytische Funktionen, ibid. vol. 32, No. 7, 1929. 6. H. Weyl, Ueber das Pick-Nevanlinna’sche Interpolationsproblem und sein in-

Jinitesimales Analogon, Ann. of Math. vol. 36 (1935) pp. 230-254. 7. , Chapter 3 of the first, and §1 of the second, paper quoted under [3]. 8. M. H. Stone, Linear transformations in Hilbert space, Amer. Math. Soc. Colloquium Publications, vol. 15, 1932. 9. E. Hellinger, Newe Begriindung der Theorie quadratischer Formen von unend-

lichvielen Verdnderlichen, J. Reine Angew. Math. vol. 136 (1909) pp. 265-326.

10. E. Hilb, Ueber gewohnliche Differentialgleichungen mit Singularititen und die dazugehdrigen Entwicklungen willkiérlicher Funktionen, Math. Ann. vol. 76 (1915) pp.

333-339.

11. E. C, Titchmarsh, Eigenfunction expansions associated with second-order dif-

ferential equations, Oxford, Clarendon Press, 1946.

12. A. Wintner, Stability and spectrum in the wave mechanics of lattices, Physical

Review vol. 72 (1947) pp. 81-82; On the normalization of characteristic differentials in continuous spectra, ibid. vol. 72 (1947) pp. 516-517; On the location of continuous

spectra, Amer. J. Math. vol. 70 (1948) pp. 22-30; Asymptotic integrations of the adiabatic oscillator, Amer. J. Math. vol. 69 (1947) pp. 251-272, and Duke Math. J. vol. 15 (1948) pp. 53-67. P. Hartmann and A. Wintner, An oscillation theorem for

continuous spectra, Proc. Nat. Acad. Sci. U.S.A. vol. 33 (1947) pp. 376-379.

13. H. Weyl, Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differentialgleichungen, Math. Ann. vol. 71 (1911) pp. 441-469; Ueber die Abhingigkeit der Eigenschwingungen einer Membran von deren Begrenzung, J. Reine Angew. Math. vol. 141 (1912) pp. 1-11. 14, See his account in: R. Courant and D. Hilbert, Methoden der mathematischen Physik, vol. 1, 2d ed., Berlin, 1931, in particular Chap. 6, §4.

15. A. Weinstein, Etude des spectres des équations aux dérivées partielles de la théorie des plaques élastiques, Mémorial des Sciences Mathématiques, vol. 88, 1937.

16. N. Aronszajn, Rayleigh-Ritz and A. Weinstein methods for approximation of eigenvalues, 1: Operators in a Hilbert space, Proc. Nat. Acad. Sci. U.S.A. vol. 34 (1948) pp. 474-480; II: Differential operators, ibid. pp. 594-601.

17. T. Carleman, Propriétés asymptotiques des fonctions fondamentales des mem-

StockMatematikerkongressen Skandinaviska Forhandlingar branes vibrantes, holm, 1934, pp. 34-44; Ueber die asymptotische Verteilung der Eigenwerte partieller Differentialgleichungen, Berichte iiber die Verhandlungen der Sachsischen Akademie

der Wissenschaften zu Leipzig vol. 88 (1936) pp. 119-132. 18. A. Pleijel, Propriétés asymptotiques des fonctions et valeurs propres de certains

problémes de vibration, Arkiv fr Matematik, Astronomi och Fysik vol. 27 A, No. 13,

1940; Sur la distribution des valeurs propres de problémes régis par l’équation Au +dR(x, y)u=0, ibid. vol. 29 B, No. 7, 1943; On Hilbert-Schmidt’s theorem in the

theory of partial differential equations, Fysiografiska Salskapets i Lund Férhandlingar vol. 17, No. 2, 1946; Asymptotic relations for the eigenfunctions of certain boundary

problems of polar type, Amer. J. Math. vol. 70 (1948) pp. 892-907. 19. S. Minakshisundaram, A generalization of Epstein zeta functions, Canadian Journal of Mathematics vol. 1 (1949) pp. 320-327. S. Minakshisundaram and A Pleijel, Some properties of the eigenfunctions of the Laplace-operator on Riemannian manifolds, Canadian Journal of Mathematics vol. 1 (1949) pp. 242-256. 20. S. Bochner, Summation of derived Fourier series (An application to Fourier

expansions on compact Lie groups), Ann. of Math. vol. 37 (1936) pp. 345-356.

456 21. H. B.G. Casimir, Rotation of a rigid body in quantum mechanics, Leiden thesis,

1931.

22. F. Peter and H. Weyl, Die Vollsténdigheit der primitiven Darstellungen einer

geschlossenen kontinuierlichen Gruppe, Math. Ann. vol. 97 (1927) pp. 737-755. 23.

H. Weyl,

Almost periodic invariant vector sets in a metric vector space,

J. Math. vol. 71 (1949) pp. 178-205.

Amer.

24. I quote but the two most important papers: H. Bohr, Zur Theorie der fastpertodischen Funktionen I, Acta Math. vol. 45 (1925) pp. 29-127. J. von Neumann, Almost periodic functions in a group 1, Trans. Amer. Math. Soc. vol. 36 (1934) pp. 445—

492. For further references see [23].

25. I. Gelfand and D. Rykov, Irreducible unitary representations of locally compact groups, Rec. Math. (Mat. Sbornik) N. S. vol. 3 (1943) pp. 301-316; also: Academy of Sciences of the USRR. Journal of Physics vol. 10 (1946) pp. 93-94. V. Bargmann, Irreducible unitary representations of the Lorentz group, Ann. of Math. (2) vol. 48 (1947) pp. 568-640.

I. E. Segal, Irreducible representations of operator algebras,

Amer. Math. Soc. vol. 53 (1947) pp. 73-88.

Bull.

26. H. Weyl, Das asymptotische Verteilungsgesetz der Eigenschwingungen eines beliebig gestalteten elastischen K orpers, Rend. Circ. Mat. Palermo vol. 39 (1915) pp. 149

11. Elementary proof of a minimax theorem due to von Neumann

Contributions to the theory of games I, Annals of Mathematics Studies, Princeton University Press 24, 19—25 (1950) J.

von

Neumann's

minimax

to the theory of linear

inequalities

elementary

way

I

pyramids.

As

field

K

or

which

elementary

of numbers

multiplication

>0

in

=0

and

or

proved

are

the

division,

in

the

theory

fundamental

such

facts

operations

nothing but addition,

and

the

Decisions

of

and can be approached

considered

as require

anes

of

Archimedes,

a matrix

of

numbers

any

for

and

point

€ =

suitably

satisfying

the

a point

in

are

= Hg

m-space

a

set

(EyasiM)

»

in

are

field

K

no

assumed.

2)

m-space:

(2584;,§;)

A,

satisfying

f=

a point

in

n= n°

n-space

direct

the

corresponding

conditions

=1, ..., m),Z,89=1, following

my) as

n-space

the

Doe = 1, ey Btn =I

going to establish the Fundamental facts:

* accepted

ees

for

conditions

£2204 we

= aie

constructing a number

no and

in

++.) 8)

(£,>

M(€)

After

Deeg Me = 1s 1)

mn)

As

axiom

ay,(1=1, For

max Hye

contribution

to

and

= 4°,

being obtained for

\,

the upper bound

=I

and which I am now

of the rest)

going to repeat as Lemma

1 is concerned with a configuration

which

span

the

Given

a linear

x

n-space,

and

form

(¥x)

lies in the plane

({x) > 0

set

2

support

respectively.

in

=



is

extreme

if

the

plane

support,

A form 1

In

1.

or

corollary:

If

We

and

the

n

Cy

Nes some

may

are

unit

0

further

of

YX

¥

¥

let

Either

the

point

== Perse

ery

(¥x)

ea

in

(Y¥x) =0

bd;

the

or

of the set, for

point

of the

and this

instance

Blea)

is proportional

configuration is

to

Pd

has

representable

in

an the

extreme form

en ye)

ayia

ye

adjective

the

that

Ln x! p

the

say

by,

Mines

extreme

has a support it has an extreme

interested

if

independent

(ly

Nei Uy,

configuration.

us

is called a support

case

here

this

of points

2

such that all points

¥ +0

linearly

this

every

cancel

2

eset

sees a

Be One

+

supports

or ‘in the half-space

n-

¥.

LEMMA

of

¥,x,

lie in the half-space

Die lie

extreme

I proved

which

pyramids

convex

about

case

where

the

and

then

support.

add

configuration

the 2

consists

points

Sac

(ne

points

OS

Oe

Wh

Beas

Ny

aes

O91

(05

00502-55.

1)

459 Set

e=-e,

the

+

...+e,=(1,

...,

hence

we

may

por OC

normalize

support



(+0)

satisfies

the

second

alternative

ye

of

Lemma

tiles One

thus

finds

a

Sain

oanA Al

by

Ce If

Every

inequalities

Ont and

1).

Heh)

representation

o=Zp,e,

* of

ty

1

takes

@,

FO

place,

represent

By eter

tt

0

-e

Sa

and

add

ye

zero,

+ Fa,

with

p,>1,£, 20,

hence

21818, = - Py

SO

(k= 1,

..., 0)

Thus we may state the following two facts: LEMMA 2. e,(k then

If the system

= UD Gare 2) eke it has an extreme LEMMA

3.

If

it

2.

consisting of the points

a,(i =1, support.

has

no

...,

support,

m)

has

i.e.

if

a support

the

inequali-

ties

Neo 0

(ola,

have none but. inequalities

the

nl, 2 egne trivial

solution

PESTA are

We

rather k,

than

n,“.

solvable

find

to

Then

it

by

||a,,||

and

Lemma

3 takes

LEMMA

4.

PL

apply

to change

the

(nearly

Either

the

system

m,

the

n

matrix

-

G59 2)

Lemma

the

...5

then

$4:

on

(L=1,

or the system

to

ibm ty Gate) = 0,

Vey

non-negative

convenient

OBC

H

3 to

the

notation

1,

identical)

20

(k=

m,g

Hapgll

into

form:

1,

...,

n)

2 fi $0 (eh = 1, «2, m), £520 (1 = 1, «0 mM, 2ZE= 1 is

solvable.

460 disjunctive

either-or.

¥, 2% and

holds

as

ey

Ae)

a consequence

of

Take any point N=

hence

this

»

satisfying

= %, (¥,

v,

nue

¥ of

> 0

+ (dA - At) (¥e)

28

20

\'

DO.

(224%)

= am)

Me,)

2 Ok = 1,

.-,

- de)

= Ce

(1).

finds

for

good.

..., m),

and

d-

For

On the other hand,

given

\

n)

anda

5 ¥ +0

relations

he)

Od

every

= 1,

point

...,.m),

of

the

Yi,= (Ye)

2 0c = 1,

-.., n)

form

(hha

eases

n=)

inequality

(8, p-de) Hence at least one of the

A M(€)

therefore

of

ae

Tee

bj)

= bs

indices,

0,

max p= max

unified

the

of Lemma the

p, - >

an upper bound for those

yO In view

20 or Fiyip, - dr) > 0

corresponding to a point

Use

out

- Ae)

satisfying

(ayn)

p =2,€,8,

Any

a; a;

2 O(L = 1,

makes

the

(Yjpay

the

We)

first A' Cd,

has

(M, ay > Ae)

one

In the every

thus

(%,

one

«25%, 2%

points

n

of the

Indeed a support

and the same support ¥ will do for all of them. LZ, satisfies the inequalities

a

try to find

and

time

has or has not a support. If X» makes good so does

a; - de m points A makes good. that

the say

and e, case we

& y consisting

the system

»

of

out for which values

called

A

a parameter

introduce

Let us now

is

this

that

see

to

order

in

bi F1cM

Zs

form

but

needs

One

2 pick j=

Oy ¥y04,.) = Mig) »



fulfilling

the relations

(2) is

that make good.

notation

any i)

or

Moen - Abie (n 222,

Sayan (qi=

4,

Fee

eee,

- 1)-combination rr.

If

Ey

1

NOSES

hd

ang re

rls tigean) J = cei,

LS

Soe

Yn-1

(A)

are

ad

1)

Tinessly

461 independent and entire set ax alive

J=

at

U1,

¥+0

the

2,

the plane going through them of the by(A) then we shall

moment

...,

>.

Let

n- 1)

such that

is

us

express

alive

at

(8 b5(s)) 20 and

This

implies

Therefore

DA'), The

-e+5

¥, > 0, e

is

the

of

or

Consequently

expression

of

(¥x)

one

has

ly, (ig

eas]

\'

simultaneous

(and

to

form

the

banarne.v)

vice

vanish

linear

c

> ©

of

independent

=1.

points

functions

soos BL ADs BGI If

J

is

alive

at

the

moment

DO

Notice

that

conn Se

Cas

ENON

SAY

Write

therefore

gO)

met

indices

ye)

= tne

A

Fj)

A.

between

an

index

he OL cys‘ j

of

the

j

of

the

first

first

and

second

class

by

oe

0,

the

the

moment

for all

(¥e)

independent

BL_y(A'),

that

= Oe fore

hence

is a supporting plane for say that the combination

combination should

serves

ignore

the

merely

special

as

an

value

For

example 0

which

AD» for

$,

an

arbitrary

takes

on

for

J.

k=

Hence

1,

...,

the

n.

462 states

©

>

F509

inequality

the

class

second

the

of

j

index

an

For that

1

an

Ics The

times

second

class

cannot

A>",

impossible.

be

and every

Determine

correspond

to

vacuous;

the

A

the

for

J

would make good,

minimum

indices

then

j

of

ds

the

II)

of

the

second

would

stay

alive

at

which we know to be quotients

class.

«5/4;

all

that

Then

RS

and 180) and

every

j

of

5X3)

203

second

class.

the

J

is alive at the moment

the whole period

A'

J

terminates

d=

at

to

life.

determinant

the

are

J=

out

least

of

the

quotients

(j's

of

the

second

class)

and

which

passes

the

away unless class.

we

may

A

(3)

J

state

(2)

our

sible

Az 2+ Indeed support

20 as

the

now

for

combinations

Ap

°

NG

the J

fais

form

1}

the

call

The

Ay:

for all three

j,

death

n-1

Jo =

2

»

has

linearly

The combination

let

(j,,

ruled

“,

if

life of

that

of

there

the

J = J,

+++»

5 be

Jp_,]

are

a

the

which

out

never

(1)

25)

if

j

for

- aj

take

the

Gy

ae

which

combination

J

come

the

= “s

are positive

those

consecutive

whose

and

show:

is

thrown

even for those of the first

J

system

how

others

functions

to

combination

largest

is

corresponding it

more

the

But if some

good

moment A. Lemma 5 shows

Ne

n-

makes

Be OO me aes Bee at

...,

A

by

III)

keeps alive during

tests

is

admissible.

there

is

an

follows:

5.

the

J

once

from

If

if

determined

(1,

«Sy

result LEMMA

separated

all 45 0 (i=1,

= min

gs

¢=¢°

(¥a,)

that

we

(3) But

A

M(g),

Xs

makes

good whenever m(n)

< X

+

good,

the

The

upper

be

We,)

no

¥

D0

4

to

that

(k=1,

would make good, Lemma

such

...,

against

the

matrix

n),

(Ye)

=1

our better knowledge. Bay

7 ro

we

find

00 ee, ety

have

every point

is actually

Thus

facts.

..., m),

Mie MKC NONE for

*£X,5)-

i = 1, .., wy, dy < mi)

can

Zireyp Ne) so

x!

and therefore indeed

is,

applying

of

(1), n°;

there

>,



)> Xo does. is a A that makes

Fundamental hand

(Ao)?

Bye)?

for 2 »2

according to Lemma 5 no We have seen that m(w)

On

(8,

eeaninDj

r\ satisfies the conditions is actually reached for w=

ti, This

(and hence

=

TNE Neb. Ibs (A)

¢

in particular reached

for

0 Gee

satisfying

>, < M(g).

€ =¢°.

mel eee) ees se (2)

This

and every

2

According to (3) completes

that makes

good,

the lower bound

the proof.

\,

a

152. A half-century of mathematics

The American Mathematical Monthly 58, 523—553 (1951) 1. Introduction. Axiomatics. Mathematics, beside astronomy, is the oldest of all sciences. Without the concepts, methods and results found and developed by previous generations right down to Greek antiquity, one cannot understand either the aims or the achievements of mathematics in the last fifty years.

Mathematics

has been called the science of the infinite; indeed,

the mathe-

matician invents finite constructions by which questions are decided that by their very nature refer to the infinite. That is his glory. Kierkegaard once said religion deals with what concerns man unconditionally. In contrast (but with equal exaggeration) one may say that mathematics talks about the things which are of no concern at all to man. Mathematics has the inhuman quality of starlight, brilliant and sharp, but cold. But it seems an irony of creation that man’s mind knows how to handle things the better the farther removed they are from the center of his existence. Thus we are cleverest where knowledge matters least: in mathematics, especially in number theory. There is nothing in any other science that, in subtlety and complexity, could compare even remotely with such mathematical theories as for instance that of algebraic class fields. Whereas physics in its development since the turn of the century resembles a mighty stream rushing on in one direction, mathematics is more

like the Nile delta, its waters fanning out in all directions. In view of all this: dependence on a long past, other-worldliness, intricacy, and diversity, it seems

an almost hopeless task to give a non-esoteric account of what mathematicians have done during the last fifty years. What I shall try to do here is, first to describe in somewhat vague terms general trends of development, and then in more precise language explain the most outstanding mathematical notions devised, and list some of the more important problems solved, in this period. One very conspicuous aspect of twentieth century mathematics is the enormously increased role which the axiomatic approach plays. Whereas the axiomatic method was formerly used merely for the purpose of elucidating the

foundations on which we build, it has now

become a

tool for concrete mathe-

matical research. It is perhaps in algebra that it has scored its greatest successes. Take for instance the system of real numbers. It is like a Janus head facing in two directions: on the one side it is the field of the algebraic operations of addition and multiplication; on the other hand

it is a continuous manifold,

like such

and

the parts of which are so connected as to defy exact isolation from each other. The one is the algebraic, the other the topological face of numbers. Modern axiomatics, simple-minded as it is (in contrast to modern politics), does not ambiguous

mixtures

of peace

and

war,

therefore

cleanly

sepa-

rated both aspects from each other. In order to understand a complex mathematical situation it is often convenient to separate in a natural manner the various sides of the subject in question, make each side accessible by a relatively narrow and easily surveyable

465,

group of notions and of facts formulated in terms of these notions, and finally

return to the whole by uniting the partial results in their proper specialization.

The last synthetic act is purely mechanical. The art lies in the first, the analytic

act of suitable

separation

and

generalization.

Our

mathematics

of the last

decades has wallowed in generalizations and formalizations. But one misunderstands this tendency if one thinks that generality was sought merely for generality’s sake. The real aim is simplicity: every natural generalization simplifies since it reduces the assumptions that have to be taken into account. It is not easy to say what constitutes a natural separation and generalization. For this there is ultimately no other criterion but fruitfulness: the success decides. In following this procedure the individual investigator is guided by more or less obvious analogies and by an instinctive discernment of the essential acquired through accumulated previous research experience. When systematized the procedure leads straight to axiomatics. Then the basic notions and facts of which we spoke are changed into undefined terms and into axioms involving them. The body of statements deduced from these hypothetical axioms is at our disposal now, not only for the instance from which the notions and axioms were abstracted, but wherever we come across an interpretation of the basic terms

which turns the axioms into true statements. It is a common occurrence that there are several such interpretations with widely different subject matter. The axiomatic approach has often revealed inner relations between, and has made for unification of methods within, domains that apparently lie far apart. This tendency of several branches of mathematics to coalesce is another conspicuous feature in the modern development of our science, and one that goes side by side with the apparently opposite tendency of axiomatization. It is as if you took a man out of a milieu in which he had lived not because it fitted him but from ingrained habits and prejudices, and then allowed him, after thus

setting him

nature.

free, to form associations in better accordance

with his true inner

In stressing the importance of the axiomatic method I do not wish to exaggerate. Without inventing new constructive processes no mathematician will get very far. It is perhaps proper to say that the strength of modern mathematics lies in the interaction between axiomatics and construction. Take algebra as a representative example. It is only in this century that algebra has come into its own by breaking away from the one universal system of numbers which used to form the basis of all mathematical operations as well as all physical measurements. In its newly-acquired freedom algebra envisages an infinite variety of “number fields” each of which may serve as an operational basis; no attempt is made to embed them into the one system &. Axioms limit the possibilities for the number concept; constructive processes yield number fields that satisfy the axioms.

In this way algebra has made itself independent of its former master analysis

and in some branches has even assumed the dominant role. This development in

mathematics is paralleled in physics to a certain degree by the transition from

466 classical to quantum physics, inasmuch as the latter ascribes to each physical structure its own system of observables or quantities. These quantities are subject to the algebraic operations of addition and multiplication; but as their multiplication is non-commutative, they are certainly not reducible to ordinary

esti

numbers.

At the International Mathematical Congress in Paris in 1900 David Hilbert,

convinced

that problems are the life-blood of science, formulated

twenty-three

unsolved problems which he expected to play an important role in the development of mathematics during the next era. How much better he predicted the future of mathematics than any politician foresaw the gifts of war and terror that the new century was about to lavish upon mankind! We mathematicians have often measured our progress by checking which of Hilbert’s questions had been settled in the meantime. It would be tempting to use his list as a guide for a survey like the one attempted here. I have not done so because it would necessitate explanation of too many details. I shall have to tax the reader’s patience enough anyhow. Part I. ALGEBRA. NUMBER THEORY. GROUPS. 2. Rings, Fields, Ideals. Indeed, at this point it seems impossible for me to go on without illustrating the axiomatic approach by some of the simplest algebraic notions. Some of them are as old as Methuselah. For what is older than

the sequence

of natural numbers

1, 2, 3, ---,

by which

we

count? Two

such numbers a, 6 may be added and multiplied (@ + b and a-b). The next step in the genesis of numbers adds to these positive integers the negative ones and zero; in the wider system thus created the operation of addition permits of a unique inversion, subtraction. One does not stop here: the integers in their turn get absorbed into the still wider range of rational numbers (fractions). Thereby division, the operation inverse to multiplication, also becomes possible, with one notable exception however: division by zero. (Since b:0 = 0

for every rational

number

4, there is no inverse } of 0 such

that 6-0

=

1.) I

now formulate the fundamental facts about the operations “plus” and “times”

in the form of a table of axioms:

Table T (1) The commutative and associative laws for addition,

a+b=b+4.

a+(b+c =(c+ db +6.

(2) The corresponding laws for multiplication. (3) The distributive law connecting addition with multiplication

c-(a@ + (4)

b)

=

(6-a)

+

(¢-5).

The axioms of subtraction: (41) There is an element o (0, “zero”) such that

@ +0 = 0 + a4 = a for every a. (42) To every a there is a number such thata + (—a) = (-a) + a = 0.

—a

467

(5)

The axioms of division: (51) There is an element e (1, “unity”) such that a-e = e-a = a for every a. (52) Toeverya # o there is an a~! such that aa! = a-a = e

By means of (42) and (52) one may introduce the difference b — a and the quotient b/a as b + (—a) and b-a-—, respectively. When the Greeks discovered that the ratio (./2) between diagonal and side of a square is not measurable by a rational number, a further extension of the

number

concept

was

called

for.

However,

all

measurements

of

continuous

quantities are possible only approximately, and always have a certain range of inaccuracy. Hence rational numbers, or even finite decimal fractions, can and do serve the ends of mensuration provided they are interpreted as approximations, and a calculus with approximate numbers seems the adequate numerical instrument for all measuring sciences. But mathematics ought to be prepared for any subsequent refinement of measurements. Hence dealing, say, with electric phenomena, one would be glad if one could consider the approximate values of the charge e¢ of the electron which the experimentalist determines with ever greater accuracy as approximations of one definite exact value e. And thus, during more than two millenniums from Plato’s time until the end of the nineteenth century, the mathematicians worked out an exact number concept, that of real numbers, that underlies all our theories in natural science.

Not even to this day are the logical issues involved in that concept completely clarified and settled. The rational numbers are but a small part of the real numbers. The latter satisfy our axioms no less than the rational ones, but their system possesses a certain completeness not enjoyed by the rational numbers, and it is this, their “topological” feature, on which the operations with infinite sums and the like, as well as all continuity arguments, rest. We shall come back

to this later.

Finally, during the Renaissance complex numbers were introduced. They are essentially pairs z = (x, y) of real numbers x, y, pairs for which addition and multiplication are defined in such a way that all axioms hold. On the ground of these definitions e = (1, 0) turns out to be the unity, whilez = (0, 1) satisfies the equation 7-i = — e. The two members x, y of the pair ¢ are called its real and imaginary parts, and g is usually written in the form xe + yi, or simply x + yi. The usefulness of the complex numbers rests on the fact that every algebraic equation (with real or even complex coefficients) is solvable in the field of complex numbers. The analytic functions of a complex variable are the subject of a particularly rich and harmonious theory, which is the show-piece of classical nineteenth century analysis. A set of elements for which the operations a + 6 and a:b are so defined as to satisfy the axioms (1)-(4) is called a ring; it is called a field if also the axioms (5) hold. Thus the common integers form a ring J, the rational numbers form a field w; so do the real numbers (field &) and the complex numbers (field *). But these are by no means the only rings or fields. The polynomials of all

468 possible degrees h, (1)

f

= f(x)

=

a9

+

ae

+

ax?

+ +++

+

ane,

with coefficients a; taken from a given ring R (e.g. the ring J of integers, or the

field w), called “polynomials over R,” form a ring R[x]. Here the variable or indeterminate x is to be looked

really nothing but the sequence

upon as an empty

symbol;

of its coefficients do, a1, 2,

the polynomial

is

+++. But writing

it in the customary form (1) suggests the rules for the addition and multiplication of polynomials which I will not repeat here. By substituting for the variable x a definite element (“number”) ¥ of R, or of a ring P containing R as a sub-

ring, one projects the elementsf of R[x] into elements a of P, f — a: the poly-

nomial f = f(x) goes over into the number a = f(y). This mapping f > a is homomorphic, i.e., it preserves addition and multiplication. Indeed, if the substitution of y for x carries the polynomial f into a and the polynomial g into 6 then it carries f + g,f-gintoa + , a-B, respectively. If the product of two elements of a ring is never zero unless one of the factors is, one says that the ring is without null-divisor. This is the case for the rings discussed so far. A field is always a ring without null-divisor. The construction by which one rises from the integers to the fractions can be used to show that any ring R with unity and without null-divisor may be imbedded in a field k, the quotient field, such that every element of & is the quotient a/b of two elements a and 6 of R, the second of which

(the denominator)

is not zero.

Writing 1a, 2a, 3a,---,fora,a + a,a + a + a, etc., we use the natural numbers n=1, 2, 3, - - - , as multipliers for the elements a of a ring or a field. Suppose the ring contains the unity e. It may happen that a certain multiple ne of e equals zero; then one readily sees that na = 0 for every element a of

the ring. If the ring is without null divisors, in particular if it is a field and p

is the least natural number for which pe = 0, then p is necessarily a prime number like 2 or 3 or 5 or 7 or 11 - - - . One thus distinguishes fields of prime characteristic p from those of characteristic 0 in which no multiple of e is zero. Plot the integers ---, —2, —1, 0, 1, 2, - - - as equidistant marks on a line. Let ” be a natural number 22 and roll this line upon a wheel of circumference n. Then any two marks a, a’ coincide, the difference a—a’ of which is divisible by 2. (The mathematicians write a = a’ (mod m); they say: a congruent to a’ modulo .) By this identification the ring of integers I goes over into a ring I, consisting of m elements only (the marks on the wheel), as which one may take the “residues” 0, 1,- +--+, — 1. Indeed, congruent numbers give congruent

results under both addition and multiplication:a = a’,b a+b =a’ + b’,a-b = a’-b’ (mod m). For instance, 7

+

8

=

3, 5-8

=

4 because

15 leaves the residue

= 6’ (mod n) imply modulo 12 we have

3 and

40 the residue

4

if divided by 12. The ring Jy is not without null divisors since 3-4 is divisible

by 12, but neither 3 nor 4 is. However, if p is a natural prime number, then T, has no null divisor and is even a field; for as the ancient Greeks proved by

an ingenious procedure (Euclid’s algorism), for every integer a not divisible by

469 p there is one, a’, such that a-a’

=

1 (mod p). This Euclidean theorem is at

the basis of the whole of number theory. The example shows that there are fields of any given prime characteristic p. In any ring R one may introduce the notions of unit and prime element as follows. The ring element a is a unit if it has a reciprocal a’ in the ring, such that a’-a@ = e. The element a is composite if it may be decomposed into two

factors a@;:d2, neither

of which

is a unit. A

prime

number

is one

that is

linear

factors

neither a unit nor composite. The units of J are the numbers +1 and —1. he units of the ring k [x] of polynomials over a field # are the non-vanishing elements of k (polynomials of degree 0). According to the Greek discovery of the irrationality of 1/2 the polynomial x? — 2 is prime in the ring w[x]; but, of

course, (x



not

»/2)(x

in

Q[x],

+

for

there

it

splits

into

the

two

+/2). Euclid’s algorism is also applicable to polynomials f(x)

of one variable x over any field k. Hence they satisfy Euclid’s theorem: Given a

P = P(x) in this ring k[x] and an element f(x) of k[x] not

prime element

divisible by P(x), there exists another polynomial f’(x) over k such that { F(x) -f'(x)} — 1 is divisible by P(x). Identification of any elements f and g of k[x], the difference of which is divisible by P, therefore changes the ring k[x] into a field, the “residue field « of k[x] modulo P.” Example: w[x] mod x? ~ 2. (Incidentally the complex numbers may be described as the elements of the residue field of Q[x] mod x? + 1.) Strangely enough, the fundamental Euclidean theorem does not hold for polynomials of two variables x, y. For

= x — y is a prime element of w[x, y], and f(x, y)

instance, P(x, y)

an element not divisible by P(x, y). But a congruence

is impossible.

— 9)

(mod

a-f'(*,y) = 1 Indeed, it would imply



4: /'@, 2) = 1+



0, contrary to the

=

x-f’(x, x)

1 +

= x

fact that the polynomial of one indeterminate x,

=

1

Flav

bas

+ ---,

is not zero. Thus the ring w[x, y] does not obey the simple laws prevailing in Tand in w[x]. Consider x, the residue field of w[x] mod x? — 2. Since for any two polynomials f(x), f’(x)

which

are

congruent

mod

x?



2 the

numbers

f(V2),

f'(\/2) coincide, the transition f(x) — f(/2) maps x into a sub-field w[/2] of © consisting of the numbers

tion would be f(x)

@ +

64/2 with rational a, 6. Another such projec-

— f(—-+/2). In former times one looked upon x as the part

w[./2] of the continuum @ or 0* of all real or all complex numbers; one wished to embed everything into this universe 2 or Q* in which analysis and physics operate. But as we have introduced it here, « is an algebraic entity the elements of which are not numbers in the ordinary sense. It requires for its construction no other numbers but the rational ones. It has nothing to do with ®, and ought not to be confused with the one or the other of its two projections into Q More generally, if P=P(x) is any prime element in w[x] we can form the

470

residue field xp of w[x] modulo P. To be sure, if 6 is any of the real or complex roots of the equation P(x) =0 in Q* then f(x) — f(8) defines projection of xp into Q*. But the projection is not xp itself. Let us return

to the ordinary

integers--+,

—2,

—1,

a homomorphic

0,1,

2,---,

which

form the ring I. The multiples of 5, é.e., the integers divisible by 5, clearly form a ring. It is a ring without unity, but it has another important peculiarity: not only does the product of any two of its elements lie in it, but all the integral multiples of an element do. The queer term ideal has been introduced for such a set: Given a ring R, an R-ideal (a) is a set of elements of R such that (1) sum and difference of any two elements of (a) are in (a), (2) the product of an element in (a) by any element of R is in (a). We may try to describe a divisor a by the set of all elements divisible by a. One would certainly expect this set to be an ideal (a) in the sense just defined. Given an ideal (a), there may not exist an actual element a of R such that (a) consists of all multiples 7 = m-a of a (m any element in R). But then we would say that (a) stands for an “ideal divisor” a: the words “the elementj of R is divisible by a” would simply mean: “j belongs to (a).” In the ring I of common integers all divisors are actual. But this is not so in every ring. An algebraic surface in the three-dimensional Euclidean space with the Cartesian coordinates x, y, 2 is defined by an equation F(x, y,z) = 0 where Fis an element of 32 = Q|[x, y, 2], i.e., a polynomial of the variables x, y, 2 with real coefficients. F is zero in all the points of

F of F (L being any elethe surface; but the same is true for every multiple Lment of *Q), in other words, for every polynomial of the ideal (F) in 3Q. Two simultaneous polynomial equations

Fy(x, 9,2) = 0,

-Fa(x, y, 2) = 0

will in general define a curve, the intersection of the surface 7,

=

0 and the

surface F, = 0. The polynomials (Zi: F\) + (L2:F2) formed by arbitrary elements Li, Lz of *2 form an ideal (Fi, F2), and all these polynomials vanish on the curve. This ideal will in general not correspond to an actual divisor F, for a curve is not a surface. Examples like this should convince the reader that the study of algebraic manifolds (curves, surfaces, efc., in 2, 3, or any number of dimensions) amounts essentially to a study of polynomial ideals. The field of

coefficients

nature.

3. Some

is not

necessarily

achievements

2 or *,

of

algebra

but may

and

be a

number

field of a more

theory.

general

I have

finally

reached a point where I can hint, I hope, with something less than complete

obscurity, at some of the accomplishments

of algebra and number

theory in

our century. The most important is probably the freedom with which we have learned to manage these abstract axiomatic concepts, like field, ring, ideal, etc. The atmosphere in a book like van der Waerden’s Moderne Algebra, published about 1930, is completely different from that prevailing, e.g., in the articles on

algebra written for the Mathematical Encyclopaedia around 1900. More specif-

471 ically, a general

theory of ideals, and

in particular of polynomial

ideals, was

developed. (However, it should be said that the great pioneer of abstract alge-

bra, Richard Dedekind, who first introduced the ideals into number theory, still belonged to the nineteenth century.) Algebraic geometry, before and around 1900 flourishing chiefly in Italy, was at that time a discipline of a type uncommon in the sisterhood of mathematical disciplines: it had powerful methods, plenty of general results, but they were of somewhat doubtful validity. By the abstract algebraic methods of the twentieth century all this was put on a safe basis, and the whole subject received a new impetus. Admission of fields other than 0%, as the field of coefficients, opened up a new horizon. A new technique, the “primadic numbers,” was introduced into algebra and number theory by K. Hensel shortly after the turn of the century, and since then has become of ever increasing importance. Hensel shaped this instrument in analogy to the power series which played such an important part in Riemann’s and Weierstrass’s theory of algebraic functions of one variable and their integrals (Abelian integrals). In this theory, one of the most impressive accomplishments of the previous century, the coefficients were supposed to vary over the field Q* of ali complex numbers. Without pursuing the analogy, I may illustrate the idea of p-adic numbers by one typical example, that of quadratic norms. Let p be a prime number, and let us first agree that a congruence power f* of p for rational numbers a, b has this meaning that a = b moduloa (a — b)/p* equals a fraction whose denominator is not divisible by p; 39

£—-

12

-—=

pitas

0 (mod.

ie"

'

72) b

Let now a, } be rational numbers, a #

2

39

12

cause

4.

5

a

5)

2

20

0, and b not the square of a rational

number. In the quadratic field w[+/b] the number a is a norm if there are

rational numbers x, y such that

a = (x + yVb)(@ — yb),

oro

= 2? — byt,

and Necessary for the solvability of this equation is (1) that for every prime p every power p'of p the congruencea = x? — by? (mod p*) has a solution. This

js what we mean by saying the equation has a p-adic solution. Moreover there must exist rational numbers x and y such that x? — by? differs as little as one

wants from a. This is what we mean by saying that the equation has an ~ -adic

solution. The latter condition is clearly satisfied for every a provided 0 is posicase tive; however, if } is negative it is satisfied only for positive a. In the first namely, every a is ~-adic norm, in the second case only half of the a’s are, c norms. the positive ones. A similar situation prevails with respect to p-adi

@ is a norm One proves that these necessary conditions are also sufficient: if

on for every locally everywhere, i.e., if @ = x? — by? has a p-adic soluti then it has a “finite prime spot p” and also for the “infinite prime spot ©,” rs x, y. “global” solution, namely an exact solution in rational numbe

472 This example,

the simplest

I could

think

of, is closely connected

with

the

theory of genera of quadratic forms, a subject that goes back to Gauss’ Disquisi-

tiones arithmeticae, but in which the twentieth century has made some decisive

progress by means of the p-adic technique, and it is also typical for that most fascinating branch of mathematics mentioned in the introduction: class field theory. Around 1900 David Hilbert had formulated a number of interlaced theorems concerning class fields, proved some of them at least in special cases, and left the rest to his twentieth century successors, among whom I name Takagi, Artin and Chevalley. His norm residue symbol paved the way for Artin’s general reciprocity law. Hilbert had used the analogy with the Riemann-Weierstrass theory of algebraic functions over 0 for his orientation, but the ingenious, partly transcendental methods which he applied had nothing to do with the much simpler ones that had proved effective for the functions. By the primadic technique a rapprochement of methods has occurred, although there is still a considerable gap separating the theory of algebraic functions and the much subtler algebraic numbers. Hensel and his successors have expressed the p-adic technique in terms of the non-algebraic “topological” notion of (“valuation” or) convergence. An infinite sequence of rational numbers a, dz, - - - is convergent if the difference a; — a; tends to zero, a; — a; — 0, provided i and j independently of each other tend to infinity; more explicitly, if for every positive rational number e there exists a positive integral N such that — e < a; — a; < € forall iand j > N. The completeness of the real number system is expressed by Cauchy’s convergence theorem: To every convergent sequence di, dz, - - - of rational numbers there exists a real number a@ to which it converges: a; — a — 0 for

4 — , With the ~-adic concept of convergence we have now confronted the p-adic one induced by a prime number p. Here the sequence is considered

convergent

if for every exponent

h

=

1, 2, 3,---,

there is a positive integer

Nsuch that a; — a; is divisible by p*as soonasiandj > N. By introduction of p-adic numbers one can make the system of rational numbers complete in the p-adic sense as the introduction of real numbers makes them complete in

the »-adic sense. The rational numbers are embedded

in the continuum of all

real numbers, but they may be embedded as well in that of all p-adic numbers. Each of these embedments corresponding to a finite or the infinite prime spot p is equally interesting from the arithmetical viewpoint. Now it is more evident than ever how wrong it was to identify an algebraic number field with one of its homomorphic projections into the field @ of real numbers; along with the (real) infinite prime spots one must pay attention to the finite prime spots which correspond to the various prime ideals of the field. This is a golden rule abstracted from earlier, and then made fruitful for later, arithmetical research; and here is one bridge (others will be pointed out later) joining the two most fascinating branches of modern mathematics: abstract algebra and topology. Besides the introduction of the primadic treatment and the progress made

in the theory of class fields, the most important advances of number theory

473

during the last fifty years seem to lie in those regions where the powerful tool

of analytic functions can be brought to bear upon its problems. I mention two such fields of investigation: I. distribution of primes and the zeta function, II. additive number theory. I. The notion of prime number is of course as old and as primitive as that of the multiplication of natural numbers. Hence it is most surprising to find the distribution of primes among all natural numbers is of such a highly irregular and almost mysterious character. While on the whole the prime numbers thin out the further one gets in the sequence of numbers, wide gaps are always followed again by clusters. An old conjecture of Goldbach’s maintains that there even come along again and again pairs of primes of the smallest

possible difference 2, like 57 and 59. However, the distribution of primes obeys at least a fairly simple asymptotic law: the number w(x) of primes among all numbers from 1 to

is asymptotically equal to n/log n. [Here log n is not the

Briggs logarithm which our logarithmic tables give, but the natural logarithm as defined by the integral {%dx/x.] By asymptotic is meant that the quotient tends between m() and the approximating function ”/log infinity. In antiquity Eratosthenes had devised a method to numbers. By this sieve method the Russian mathematician obtained, during the nineteenth century, the first non-trivial distribution of primes. Riemann used a different approach: called zeta-function defined by the infinite series

HG) =

@)

to 1 as ” tends to sift out the prime Tchebycheff had results about the his tool is the so-

hp Op Sp

Here s is a complex variable, and the series converges for all values of s, the real part of which is greater than 1,Rs > 1. Already in the eighteenth century the fact that every positive integer can be uniquely factorized into primes had been translated by Euler into the equation

VEO)

Css

Bath

= SGC

oe II) ace

where the (infinite) product extends over all primes 2, 3, 5, ---. Riemann showed that the zeta-function has a unique “analytic continuation” to all values of s and that it satisfies a certain functional equation connecting its

values for sand 1 —

s. Decisive for the prime number problem are the zeros of

the zeta-function, i.e., the values s for which {(s)=0. Riemann’s equation all showed that, except for the “trivial” zeros at s = —2, —4, —6,---, zeros have real parts between 0 and 1. Riemann conjectured that their real parts actually equal 4. His conjecture has remained a challenge to mathematics now for almost a hundred years. However, enough had been learned about these zeros at the close of the nineteenth century to enable mathematicians, by means of some profound and newly-discovered theorems concerning analytic functions, to prove the above-mentioned asymptotic law. This was generally considered a great triumph of mathematics. Since the turn of the century Rie-

474 mann’s functional equation with the attending consequences has been carried over from the “classical” zeta-function (ii) of the field of rational numbers to that of an arbitrary algebraic number field (E. Hecke). For certain fields of prime characteristic one succeeded in confirming Riemann’s conjecture, but this provides hardly a clue for the classical case. About the classical zeta-function we know now that it has infinitely many zeros on the critical line Rs = 4, and even that at least a fixed percentage, say 10 per cent, of them

lie on it. (What

this means is the following: Some percentage of those zeros whose imaginary part lies between arbitrary fixed limits — T and +T will have a real part equal to 3, and this percentage will not sink below a certain positive limit, like 10 per cent, when 7 tends to infinity.) Finally about two years ago Atle Selberg succeeded, to the astonishment of the mathematical world, in giving an “elementary” proof of the prime number law by an ingenious refinement of old Eratosthenes’ sieve method. II. It has been known for a long time that every natural number n may be written as the sum of at most four square numbers, e.g.,

THRE

PEP

+1,

87 = 94 274 124 12 = 724 524 32 +4 22,

The same question arises for cubes, and generally for any k® powers (k=2, 3, 4, 5,+--). In the eighteenth century Waring had conjectured that every non-negative integer may be expressed as the sum of a limited number M of k* powers,

(3)

pa

ees

where the ; are also non-negative integers and M is independent of 7. The first decade of the twentieth century brought two events: first one found that every 7 is expressible as the sum of at most 9 cubes (and that, excepting a few comparatively small », even 8 cubes will do); and shortly afterwards Hilbert proved Waring’s general theorem. His method was soon replaced by a different approach, the Hardy-Littlewood circle method, which rests on the use of a certain analytic function of a complex variable and yields asymp totic formulas for the number of different representations of n in the form (3). With some precautions demanded by the nature of the problem, and by overcoming some quite serious obstacles, the result was later carried over to arbitrary algebraic number fields; and by a further refinement of the circle metho d in a different direction Vinogradoff proved that every sufficiently large n is the sum of at most 3 primes. Is it even true that every even m is the sum of 2 primes?

To show this seems to transcend our present mathematical power s as much as Goldbach’s conjecture. The prime numbers remain very elusiv e fellows.

III. Finally, a word ought to be said about investigations concerning the arithmetical nature of numbers originating in analysis. One of the most elementary such constants is 7, the area of the circle of radius 1. By proving that m is a transcendental number (not satisfying an algebr aic equation with

475

rational coefficients) the age-old problem of “squaring the circle” was settled in

1882 in the negative sense; that is, one cannot square the circle by constructions with ruler and compass. In general it is much harder to establish the transcendency of numbers than of functions. Whereas it is easy to see that the exponential function

Oe is not algebraic, it is quite number. C. L. Siegel was a sort of general method results in this field remain

aa

ib ae =a aesa

oe

ee 1-2-3

difficult to prove that its basis e is a transcendental the first who succeeded, around 1930, in developing for testing the transcendency of numbers. But the sporadic.

4. Groups, vector spaces and algebras. This ends our report on number theory, but not on algebra. For now we have to introduce the group concept, which, since the young genius Evariste Galois blazed the trail in 1830, has penetrated the entire body of mathematics. Without it an understanding of modern mathematics is impossible. Groups first occurred as groups of transformations. Transformations may operate in any set of elements, whether it is finite like the integers from 1 to 10, or infinite like the points in space. Set is a premathematical

concept: whenever

we deal with a realm

of objects, a set is de-

fined by giving a criterion which decides for any object of the realm whether it belongs to the set or not. Thus we speak of the set of prime numbers, or of

the set of all points on a

circle, or of all points with rational

coordinates

in a

given coordinate system, or of all people living at this moment in the State of New Jersey. Two sets are considered equal if every element of the one belongs to the other and vice versa. A mapping S of a set o into a set a’ is defined if with every element a of o there is associated an element a’ of a’, a > a’. Here a rule is required which allows one to find the “image” a’ for any given element a of o. This general notion of mapping we may also call of a premathematical nature. Examples: a real-valued function of a real variable is a mapping of the continuum Q into itself. Perpendicular projection of the space points upon a given plane is a mapping of the space into the plane. Representing every space-point by its three coordinates x, y, z with respect to a given coordinate system is a mapping of space into the continuum of real number triples (x, y, 2). If a mapping S, a — a’ of a into o’, is followed by a mapping S’, a’ — a"’ of

a’ into a third set o’’, the result is a mapping SS’:

a —

a” of o into a’. A

one-to-one mapping between two sets a, a’ is a pair of mappings, S: a — a’ of o into o’, and S’: a’ — a of oa’ into a, which are inverse to each other. This means that the mapping SS’ of o into a is the identical mapping Z of which sends every element a of o into itself, and that S’S is the identical mapping of o’. In particular, one is interested in one-to-one mappings of a set @ into itself. For them we shall use the word transformation. Permutations are nothing but transformations of a finite set.

476 The inverse S’ of a transformation S,a — a’ ofa given set, is again a transformation and is usually denoted by S-!. The result ST of any two transformations S and T of o is again a transformation, and its inverse is T—4S—! (according to the rule of dressing and undressing: if in dressing one begins with the shirt and ends with the jacket, one must in undressing begin with the jacket and end with the shirt. The order of the two “factors” S, T is essential.) A group of transformations is a set of transformations of a given manifold which (1) contains the identity Z, (2) contains with every transformation S its inverse S~!, and (3) with any two transformations S, T their “product” ST. Example: One could define congruent configurations in space as point sets of which one goes into the other by a congruent transformation of space. The congruent transformations, or “motions,” of space form a group; a statement which, according to the above definition of group, is equivalent to the threefold statement that (1) every figure is congruent to itself, (2) if a figure F is congruent to F’, then F’ is congruent to F, and (3) if F is congruent to F’ and F’ congruent to F’’, then F is congruent to F’’, This example at once illuminates the inner significance of the group concept. Symmetry of a configuration F in space is described by the group of motions that carry F into itself.

Often manifolds have a structure. For instance,

the elements of a field are

connected by the two operations of plus and times; or in Euclidean space we have the relationship of congruence between figures. Hence we have the idea of structure-preserving mappings; they are called homomorphisms. Thus a homomorphic mapping of a field k intoa field k’isa mappinga —> a’ of the “numbers”

a of k into the numbers a’ of k’ such that (@ +

6)!

=

a’ +

b’ and

(a-b)!

= a'-b’. A homomorphic mapping of space into itself would be one that carries any two congruent figures into two mutually congruent figures. The following terminology (suggestive to him who knows a little Greek) has been agreed upon: homomorphisms which are one-to-one mappings are called isomorphisms; when a homomorphism maps a manifold ¢ into itself, it is called an endomorphism, and an automorphism when it is both: a one-to-one mapping of into itself. Isomorphic systems, i.e., any two systems mapped isomorphically upon each other, have the same structure; indeed nothing can be said about the

structure of the one system that is not equally true for the other, The automorphisms of a manifold with a well-defined structure form a group.

Two sub-sets of the manifold that go over into each other by an automorphism

deserve the name of when he says that sidered in itself”; he geometric notion of

equivalent. This is the precise idea at which Leibniz hints two such sub-sets are “indiscernible when each is conrecognized this general idea as lying behind the specific similitude. The general problem of relativity consists in

nothing else but to find the group of automorphisms. Here then is an important

lesson the mathematicians learned in the twentieth century: whenever you are concerned with a structured manifold, study its group of automorphisms. Also

the inverse problem, which Felix Klein stressed in his famous Erlangen program

(1872), deserves attention: Given a group of transformations of a manifold og,

477

determine such relations or operations as are invariant with respect to the group. If in studying a group of transformations we ignore the fact that it consists of transformations and look merely at the way in which any two of its transformations S, T give rise to a composite ST, we obtain the abstract composition schema of the group. Hence an abstract group is a set of elements (of

unknown or irrelevant nature) for which an operation of composition is defined

generating an element st from any two elements s, ¢ such that the following axioms hold: (1) There is a unit element e such that es=se=s for every s. (2) Every element s has an inverse s~! such that ss-! = s—!s = e. (3) The associative law (st)u = s(tu) holds. It is perhaps the most astonishing experience of modern mathematics how rich in consequences these three simple axioms are. A realization of an abstract group by transformations of a given manifold ¢ is obtained by associating with every element s of the group a transformation S of ¢, s — S, such that s —

S,t



T imply st —

ST.

In general, the commutative

law st

=

¢s will

not hold. If it does, the group is called commutative or Abelian (after the Norwegian mathematician Niels Henrik Abel). Because composition of group elements in general does not satisfy the commutative law, it has proved convenient to use the term “ring” in the wider sense in which it does not imply the commutative law for multiplication. (However, in speaking of a field one usually assumes this law.) The simplest mappings are the linear ones. They operate in a vector space. The vectors in our ordinary three-dimensional space are directed segments AB point B. The vector AB is considered equal to A’B’ leading from a point A toa if a parallel displacement (translation) carries AB into A’B’. In consequence of this convention one can add vectors and one can also multiply a vector by a number (integral, rational or even real). Addition satisfies the same axioms as enumerated

for numbers

in the table T, and

it is also easy

to formulate

the

axioms for the second operation. These axioms constitute the general axiomatic notion of vector space, which is therefore an algebraic and not a geometric

concept.

The

numbers

which

serve as multipliers

of the vectors

may

be the

elements of any ring; this generality is actually required in the application of the axiomatic vector concept to topology. However, here we shall assume that they form a field. Then one sees at once that one can ascribe to the vector space a natural number u as its dimensionality in this sense: there exist m vectors €1, * + +, é, such that every vector may be expressed in one and only one way as a linear combination xe;

+

+++

-

xen, where the “coordinates” x; are defi-

nite numbers of the field. In our three-dimensional space 7 equals 3, but mechan-

ics and physics give ample occasion to use the general algebraic n-dimensional vector space for higher 2. The endomorphisms of a vector space are called its linear such they allow composition ST (perform first the mapping S, they also allow addition and multiplication by numbers : if

notion of an

mappings; as then 7), but S sends the

478

arbitrary vector « into «S, T into xT, then S + T, yS are those linear mappings which send x into (xS) + («T) and y-xS, respectively. We must forego to describe how in terms of a vector basis e:, - - + , é, a linear mapping is repre-

sented by a square matrix of numbers. Often rings occur—they are then called algebras—which are at the same time vector spaces, i.e., for which three operations, addition of two elements, multiplication of two elements and multiplication of an element by a number, are defined in such manner as to satisfy the characteristic axioms. The linear mappings of an m-dimensional vector space themselves form such an algebra, called the complete matric algebra (in 2 dimensions). According to quantum mechanics the observables of a physical system form an algebra of special type with a non-commutative multiplication. In the hands of the physicists abstract algebra has thus become a key that unlocked to them the secrets of the atom. A realization of an abstract group by linear transformations of a vector space is called representation. One may also speak of representations of a ring or an algebra: in each case the representation can be described as a homomorphic mapping of the group or ring or algebra into the complete matric algebra (which indeed is a group and a ring and an algebra, all in one).

5. Finale. After spending so much time on the explanation of the notions I can be brief in my enumeration of some of the essential achievements for which they provided the tools. If g is a subgroup of the group G, one may

identify elements s, ¢ of G that are congruent

mod

g, i.e., for which sé~ is in g;

g is a “self-conjugate” subgroup if this process of identification carries G again into a group, the “factor group” G/g. The group-theoretic core of Galois’ theory isa theorem due to C. Jordan and O. Hélder which deals with the several ways in which one may break down a given finite groupGinstepsG = Gp, G1,G:,---, each G; being a self-conjugate subgroup of the preceding group G;_;. Under the assumption that this is done in as small steps as possible, the theorem states, the steps (factor groups) Gia/G; (i = 1, 2,- +--+) in one such “compo sition series” are isomorphic to the steps, suitably rearranged, in a second such series. The theorem is very remarkable in itself, but perhaps the more so as its proof rests on the same argument by which one proves what I conside r the most fundamental proposition in all mathematics, namely the fact that if you counta finite set of elements in two ways, you end up with the same number » both times. The Jordan-Hélder theorem in recent times received a much more natural and general formulation by (1) abandoning the assumption that the breaking down is done in the smallest possible steps, and (2) by admitting only such subgroups as are invariant with respect to a given set of endomorphic mappings of G. It thus has become applicable to infinite as well as finite groups, and provided a common denominator for quite a number of important algebraic facts. The theory of representations of finite groups, the most systematic and substantial part of group theory developed shortly before the turn of the century

479 by G. Frobenius, taught us that there are only a few irreducible representations, of which all others are composed. This theory was greatly simplified after 1900 and later carried over, first to continuous groups that have the topological property of compactness, but then also to all infinite groups, with a restrictive imposition (called almost-periodicity) on the representations. With these generalizations one trespasses the limits of algebra, and a few more words will have to be said about it under the title analysis. New phenomena occur if representations of finite groups in fields of prime characteristic are taken into account, and from their investigation profound number-theoretic consequences have been derived. It is easy to embed a finite group into an algebra, and hence facts about representations of a group are best deduced from those of the embedding algebra. At the beginning of the century algebras seemed to be ferocious beasts of unpredictable behavior, but after fifty years of investigation they, or at least the variety called semi-simple, have become remarkably tame; indeed the wild things do not happen in these superstructures, but in the underlying commutative “number” fields. In the nineteenth century geometry seemed to have been reduced to a study of invariants of groups; Felix Klein formulated this standpoint explicitly in his Erlangen program. But the full linear group was practically the only group whose invariants were studied. We have now outgrown this limitation and no longer ignore all the other continuous groups one encounters in algebra, analysis, geometry and physics. Above all we have come to realize that the theory of invariants has to be subsumed under that of representations.

modular

Certain

groups,

infinite discontinuous groups,

which

are of special

importance

like the unimodular and

to number

the

theory, witness

Gauss’ class theory of quadratic forms, have been studied with remarkable suc-

cess and profound results. The macroscopic and microscopic symmetries of crystals are described by discontinuous groups of motions, and it has been proved for 7 dimensions, what had long been known for 3 dimensions, that in a certain sense there is but a finite number of possibilities for these crystallographic groups. In the nineteenth century Sophus Lie had reduced a continuous group to its “germ” of infinitesimal elements. These elements form a sort of algebra in which the associative law is replaced by a different type of law. A Lie algebra isa purely algebraic structure, especially if the numbers which act as multipliers are taken from an algebraically defined field rather than from the continuum of real numbers ®. These Lie groups have provided a new playground for our algebraists. The constructions of the mathematical mind are at the same time free and necessary. The individual mathematician feels free to define his notions and to set up his axioms as he pleases. But the question is, will he get his fellowmathematicians interested in the constructs of his imagination. We can not help feeling that certain mathematical structures which have evolved through the combined efforts of the mathematical community bear the stamp of a necessity not affected by the accidents of their historical birth. Everybody who looks

480 at the spectacle of modern freedom and necessity.

algebra will be struck by this complementarity

Part II. ANatysis. TOPOLOGY. 6. Linear

operators

and

their

GEOMETRY.

spectral

of

FOUNDATIONS.

decomposition.

Hilbert

space.

A

mechanical system of m degrees of freedom in stable equilibrium is capable of oscillations deviating “infinitely little” from the state of equilibrium. It is a fact of fundamental significance not only for physics but also for music that all these oscillations are superpositions of » “harmonic” oscillations with definite frequencies. Mathematically the problem of determining the harmonic oscillations amounts to constructing the principal axes of an ellipsoid in an n-dimensional Euclidean space. Representing the vectors « in this space by their coordinates (x1, «2, - + +, %,) one has to solve an equation a —

\-Kx=

0,

where K denotes a given linear operator ( = linear mapping); ) is the square of the unknown frequency »v of the harmonic oscillation, whereas the “eigenvector” x characterizes its amplitude. Define the scalar product (x, y) of two vectors x and y by the sum xy) + +++ + on¥q. Our “affine” vector space is made into a metric one by assigning to any vector x the length ll-l| given by

||xl|?_ =

(w, «), and this metric is the Euclidean one so familiar to us from the

3-dimensional case and epitomized by the “Pythagoras.” The linear operator K is symmetric in the sense that (x, Ky) = (Kx, y). The field of numbers in which

we operate

here

is, of course,

the continuum

of all real numbers.

De-

termination of the m frequencies v or rather of the corresponding eigen-values = »? requires the solution of an algebraic equation of degree n (often known as the secular equation, because it first appeared in the theory of the secular perturbations of the planetary system). More important in physics than the oscillations of a mechanical system of a finite number of degrees of freedom are the oscillations of continuous media, as the mechanical-acoustical oscillations of a string, a membrane or a 3-dimensional elastic body, and the electromagnetic-optical oscillations of the “ether.” Here the vectors with which one has to operate are continuous functions x(s) of a point s with one or several coordinates that vary over a given domain, and consequently K is a linear integral operator. Take for instance a straight string of length 1, the points of which are distinguished by a parameter 5 varying from

0 to 1. Here

(x, x) is the integral

fix2(s)-ds, and the problem

of harmonic

oscillations (which first suggested to the early Greeks the idea of a universe ruled by harmonious mathematical laws) takes the form of the integral equation

{1] where

x(s) — rf

KG, )x)dt=0,

(0S

s S

1),

481

(2) 7

K(s, 0) =

(1']

and a is a constant determined solutions are > =

ce = 3) 36. GS o. A non-complete vector space can be made complete by the same construction by which the system of rational numbers is completed to form that of real numbers. Later authors have coined the name “Hilbert space” for a vector space satisfying these axioms. Hilbert

himself

first tackled

only integral operators

in the strict

sense as

exemplified by [1]. But soon he extended his spectral theory to a far wider

class, that of bounded (symmetric) linear operators in Hilbert space. Boundedness of the linear operator requires the existence of a constant M such that

\|Kxl]?

J lie in the nth neighborhood U, of po. Of course, the choice of the neighborhoods

U, is arbitrary

to a certain extent.

For instance,

one could also have chosen as the nth neighborhood V, of (xo, yo) the square of side 2/n around (xo, yo), to which a point (x, y) belongs, if

488



1/n

< «x —

%


O

7 is the imaginary

with

R> ©

unit, 2, the

sphere

of radius

R

The main question is: can one assign arbitrary bound-

ary values on © for such a field? The place of (1) is now taken by

Q(pp") = e-™ /2ar. The

theorem

uniqueness

Rellich in 1943.1

has

established

been

(6) by

W.

Magnus

and

F.

The formula (2) (with the new Q) gives an admissible field in W with the

boundary values

Bae Kn where K(s, s’) is the kernel (3).

a

(2)

What stands in the way of solving the ex-

ternal boundary problem in this fashion is the circumstance that the homo-

geneous equations

(E+

Ke

=0

may have non-vanishing solutions.

and

+ K) =0 n(E

They form linear manifolds ® and H of

the same number h of dimensions, and the equation (2) is solvable for a given y only if y is orthogonal to all the elements 7 of H, (ny) = 0. (The

number h has now nothing to do with the number of connected components

of V.)

The external boundary problem for scalar and also for electromagnetic radiation has recently been treated by W. K. Saunders.? He overcame the

498

difficulty just mentioned by an ingenious but highly artificial device which

he ascribes to a suggestion by H. Lewy. I asked myself whether the construction of capacity coefficients for radiating fields would not provide a better and more natural way out. Checking the above four facts I found that the first, second and third hold good, but since the argument by means of the definite Dirichlet integral fails there is no reason to expect the unifact.

versal validity of the fourth

Nor

is this supported

by the general

theory of integral equations—at least not unless one extends the manifold ® of the eigen-functions to that of the ‘‘principal functions’ which the theory of elementary divisors deals with.

Let K(ss’) be an arbitrary kernel regular enough for the applicability of

Fredholm’s theory, dimensional

and let A be the operator

E + K.

manifolds @9, $;, ®:, . . . of the solutions

We form the finite¢ of the successive

equations g=0,

Ag

=0,

PA? oe

Oe

The sequence @) ¢ ®, ¢ &, € ... becomes stable after a number / of steps:

@, = $4) = Oy. = ... while &_, is still a proper part of ,.

- The manifold

H,, of the solutions 7 of 7A™ = 0 is of the same dimensionality h,, as ®,. Set 4, = h, ®; = 6, Hi = Hand h, = n. According to the general theory an 7 of the n-dimensional manifold H, vanishes if it satisfies (7¢) = 0 for all ¢ge®,: the bilinear form (ng) for n eH, ¢ € , is non-singular. Use the abbreviations Ag = ¢’, 7A = 7’, and observe the simple rule

(n'e) = (ng’). The elements of the form ¢’ (¢ € ®,) constitute a manifold ,’ of dimensional-

ityn —h.

Hence let ¢:*, ..., ¢,* be h elements of ©, that are linearly inde-

pendent mod. 6’, and m, . . ., », a basis of H. matrix

dy = (ne*) is non-singular, det d,,, += 0.

I maintain

that the square

Gj=1,...,4)

(7)

Indeed every element 7 ¢ H is orthogonal to

the elements ¢’ of ©’, because (ny’) = (n’¢) = 0; hence if it is also orthog-

onal to g:*, .. ., g,* it is orthogonal to all ¢ ¢ 4, and therefore zero.

With this in mind we return to the kernel (3) derived from (6) and again

examine the four facts, now for , and H, instead of 6; and Hy.

H, into @,?

n = (E+

The answer is yes.

K)

= 7A.

in W the equation holds

Does

map

For any 7 eH, form g = Mn and again set

One infers from Green’s formula that for points p

S (0Q(ps’)/On,(5!) .) ds’ = SQ(ps'\n'(s") ds’. Letting p tend to a point s on Q from the exterior one gets

(E + K)g = My’;

(8)

499

in other words:

if M

carries 7 into ¢

it carries n’

=

7A

into

g!

=

Ag.

This proves even considerably more than the “‘first fact.” Fact 2 holds good for the mapping M of H, upon , because of Rellich’s uniqueness theorem. Fact 3 is trivial, and fact 4 is identical with the universally valid statement that (ng) is non-singular for

¢ H,, ¢ ¢ @,.

Thus

the external problem can be solved in the manner outlined for the static

case if only one uses the manifolds H, and ©, instead of H and ®.

Still there remains something to be desired. For we know that the “potential” u(p), equation [D], of a suitable double layer u(s) solves the

problem for given boundary values y = 7/(s) if y is orthogonal to the elements 7 of H:= Hj; it need not be orthogonal to all the elements of H,!

But for every element ’ of H,’ the formula (8) shows that the potential of the simple layer n’ is identical with that of the double layer ¢ = Mn.

This remark clarifies the situation completely and suggests the following pro-

cedure.

One chooses m*(s),.. ., 2,*(s) as h elements of H, that are linearly

independent modulo H,’.

They are mapped by M

¢n* of ®, that are linearly independent mod. ®,’.

into h elements ¢:*,.. .,

If m,..., , is a basis of H

the determinant of the coefficients (7) does not vanish.

choose h constants a; such that

can therefore

We

y*() = (5) — De a, @,* (5) is orthogonal to m, ..., 7. For the boundary values y*(s) we have a solution in form of the potential of a double layer; to it we add the linear combination> a,2,*(p) of the potentials v,*(p) of the simple layers 7,*(s). These

z

are exactly those potentials of simple layers chosen from the manifold H, which cannot be carried over into double layer potentials. The analysis may be pursued to finer details. It is not without interest that here we have a case in mathematical physics where proof of non-de-

generacy of a quadratic form is not based on its definite character, and where

one

can

cope

successfully

with

the

complications

of

elementary

divisors of higher degree than 1. 3. The best choice of surface layers for the treatment of electromagnetic

radiation is set forth in §3 of an old paper of mine’ on the eigen-frequencies of elastic bodies. It superseded an earlier paper‘. Since, unfortunately,

Mr. Saunders followed the bad example set by me in the latter I propose to

come back to the external boundary problem of outgoing electromagnetic waves in a more systematic paper on radiation capacity.

1 Rellich, F., Jahresber. d. Deutschen Mathematiker-Vereinigung, 53, 57-65 (1943). 2 Saunders, W. K., Reports No. 175 and 176 of series 7 (1950-1951) issued by the Univ. of California, Dept. of Engineering, Antenna Laboratory.

3 Weyl, H., Rend.

Circolo Mat. di Palermo, 39, 1-49 (1914).

4 Weyl, H., J. f. d. reine u. angew. Math., 143, 177-202 (1913).

154. Kapazitat von Strahlungsfeldern

Mathematische Zeitschrift 55, 187—198 (1952) Diese Arbeit

behandelt

nicht unterschligt,

in breiterer Ausfiihrung,

denselben

Gegenstand

welche die Beweise

wie eine jiingst

von mir in

den Proceedings der National Academy

of Sciences der USA. veréffent-

Den

bisher

nur

in

allem

mit

dem

lichte

Note

Ansto&

iiber skalare dazu

Strahlungsfelder

gaben

zwei

(vol. 37,

Dezember

vorlaiufiger

1951).

Form

vom

Antenna Laboratory der University of California bekannt gegebene Mitteilungen (Nr. 175 und 176 der Serie 7, 1950/51) von Herrn W.

K.

Saunpers,

die

Strahlungsfeld befassen. Ansatz

sich

vor

elektromagnetischen

Darauf lieBe sich meine Theorie,

der ,Belegungs-Potentiale*,

leicht

ausdehnen'),

bei richtigem

wie

denn

iiber-

haupt im n-dimensionalen Evxtipischen Raum das skalare Feld dureh eine lineare Differentialform beliebigen Ranges ersetzt werden kénnte.

Mir ist es genug,

das Prinzip am

Bezeichnungen.

Ein

dreidimensionaler

Cartesischen Koordinaten

mit

v(pp'),

Evkuipischer

2,, «,, 2, liegt zugrunde.

Punkte p’ zum Punkte zp fiihrende

werde

skalaren Fall (Rang 0) zu erlautern.

seine

Linge,

Raum

mit

den

Der vom beliebigen

Vektor mit den Komponenten 2;—2x;

die Distanz

der

beiden

Punkte,

mit

r =r(pp’) bezeichnet. Zp ist die Kugel vom Radius R um den Ursprung

O =

(0,0, 0). Runde

Vektoren

der

Verwendung.

Vektoranalysis

#8

Operat tritt

die

Klammern

oe

Es wird

das Innere“

,

finden

fiir das

Die Bezeichnungen

iibernommen.

@

4 =

skalare Produkt zweier

grad

div grad

und ist

div werden

der

aus

LaprLacrsche

+on z Da wir im Gebiet komplexer Zahlen operieren, S Binheit

angenommen,

i auf.

da8 der Raum in zwei komplementiire Teile,

V und da® , AuBere“ W, geteilt ist.

V wird als beschrinkt,

1) Erst nach dem Druck der Proceedings-Note (und nach Abschlu8 des Manuskripts dieser Arbeit) ist mir die jiingst veréffentlichte Arbeit von CLaus MULLER »Uber die Beugung elektromagnetischer Schwingungen an endlichen homogenen

Korpern*

(Math.

Annalen

123, 1951, S. 345-378)

zuganglich

geworden,

Er ver-

wendet, fiir den Fall der vollkommenen Reflexion, einen dhnlichen Kunstgriff wie Herr Saunpers: Modifikation der Flache Q. Ich glaube, da® der hier entwickelte Kapazitats-Begriff die natiirliche Lésung der Schwierigkeit liefert, welche aus der Existenz von Kigenfunktionen des Schwingungsproblems fiir den Innenraum entsteht.

501

W als zusammenhingend vorausgesetzt. Ihre gemeinsame Grenze 2 modge aus einer oder mehreren getrennt verlaufenden geschlossenen glatten Flaichen bestehen. Glatt bedeutet, da® der Einheitsvektor n=n(s) der duSeren Normalen im variablen Punkt s von & nicht nur eine stetige Funktion von s ist, sondern auch einer Hoper-Bedingung geniigt. Die normale Ableitung eines skalaren Feldes w(p) wird mit

du/dn = (grad u,n), iar Wert im Punkte s von 2 zuweilen der DeutlichKeit halber mit dw/dm, oder gar (vielleicht nicht sehr gliicklich) mit Au(s)/On. bezeichnet. Bei Integration iiber den Raum mit Bezug auf

den

variablen Raumpunkt

gration

element.

Das

iiber einen

p bedeutet

variablen

Randwertproblem

dp

Flachenpunkt

und

das Volumelement,

s ist ds

der Eindeutigkeitssatz.

gebene positive Konstante. Wir betrachten skalare Felder w(p), welche der Schwingungsgleichung

(1)

das

bei Inte-

Oberflachen-

/

sei

eine

ge-

(komplex-wertige)

4u+ku=0

geniigen. Fiir ein derartiges Feld, das auBerhalb einer geniigend groBen Kugel 2p, definiert ist, gilt das folgende Lemma

von

Re ticu.

~ verschwindet

[iupas

+o

gleichmdpig in allen Richtungen. Nur solche Liésungen von (1) gelten

fiir R > co

als zulassig,

strahlungshedingung

R(OulOn

identisch, vorausgesetzt, dap

+ ik Want re -> 0

gleichmafBig in allen Richtungen geniigen. punkt* p’ sich beziehende Funktion p ist eine

solche

Lésung

der Ausfiir R > co

Die auf einen

QPP’) = Q(r) =e von

welche

festen

ar

,Auf-

Ir = r(pp')|

(Grundlisung),

die

Existenz-

sich

einer

punktférmigen

die

Eindeutigkeits-

Quelle in p’ entspricht. Die Aufgabe ist, eine zuliissige Lésung in W zu bestimmen, die am Rande von W, d.i. auf 2, vorgegebene Randwerte

y(s)

frage.

annimmt.

Sie wurde

erwiihnten

zeigten,

daB

einer

Beweis

ganz

auf ganz

setzung

W

von

Lemmas

der

W. Macyus

im

Jahre

und

1943

stellt

F. Retuicn

bejahend

auf Grund

beantwortet,

y(s) = 0 auf 2 die Gleichung u(p) = 0 in W

2) Jahresber. sche

Vor

d. Deutschen

liefert,

genau

genommen,

V einschlieBenden

auszudehnen,

bedienen,

sammenhingend

und

hier

Math.-Vereinigung

mu8

Kugel

man

kommt

die

27g

sich

58, S.57—65.

Gleichung

um

des

OQ, nicht

w=

schon

Prozesses

die Voraussetzung

ist. Sei o eine samt ihrer Obertlache

0

nur



des oben

indem

sie

impliziert?).

Der Retitcnim

in ganz

AuBenraum

W.

Um

sie

dab

W

2u-

der analytischen

zur Geltung.

Fort-

ganz innerhalb IW liegende

502

Zuriickfiihrung auf eine Fredholmsche Integralgleichung. Man kann den Gradienten von Q(pp’) sowohl mit Bezug auf p als p’ bilden, und es

ist

dQ r(pp')

grad, O(PP) = So

ist klar,

fiir irgend

In

was

“ay sippy — — BtAdw O(pP).

mit

8 Q(ps')dnw = (grade Q(ps'), n(s')

einen Raumpunkt

Analogie

zur

p und

eigentlichen

einen Punkt

Potentialtheorie,

s’ auf 2 gemeint

die

mit

dem

ist.

Fall

k = 0 zu tun hat, bilden wir mit Hilfe einer beliebigen stetigen Punktion w(s) auf 2 das Potential‘ der Doppelbelegung u(s):

(2)

u(p) = f (0 Q(ps/dns) u(s') ds’. iy

Es geniigt

die

sowohl

in W

wie

in V der Gleichung (1). Es

Ausstrahlungsbedingung.

Seine

Randwerte

erfiillt

sind

ferner

£u(s f K(s, )+ 8’) w(s!) ds’, io

wobei das Zeichen + fiir Anniéherung an 2 vom AuBeren W, Zeichen — fiir Anniherung vom Inneren V gilt. Auf den Kern

das

K(s, s') = (grady Q(ss’), n(s’)) ist die Frepuotmsche Theorie anwendbar,

das Produkt (r(ss’), n(s’)) verschwindet. Integration auf

die

ganze

Fliche

&.

ist.

Das

man

Feld

(2) im

AuBern

y(s) annehmen,

wird

somit

die

vor-

y(s) stets

eine

falls u(s) eine

Lésung

u(s)+ f K(ss')u( ds’ s') = p(s) Diese

Gleichung

(eindeutige) Lésung

Kugel

der Glattheit von 2

fiir s = s’ von héherer als erster Ordnung mit Bezug auf s oder s’ erstrecke sich stets

gegebenen stetigen Randwerte der Integralgleichung

(3)

da wegen

vom

Radius

in bekannter

besitzt

u(s),

vorausgesetzt,

a um den

Weise

fiir beliebig vorgegebenes

Punkt

p).

die Gleichung

2u(p) = Je

6

daB die homogene

Pir einen

Punkt

Gleichung

p innerhalb

o erhalt

Q(sp)/Ons) u(s) — Q(sp')Oujdn,} ds.

Daraus ergibt sich eine Entwicklung von w(p) nach Potenzen der relativen Koordinaten «;— 2} von p in bezug auf p,, die zum mindesten innerhalb der konzentrischen Kugel o’ um p) vom Radius a’ = a(y2 — 1) Giiltigkeit besitzt. Dar-

auf fubt die kettenartige »Buchten*

des

,,Meeres“

analytische

W

hinein.

Fortsetzung

der Gleichung w(p) = 0 in alle

503

mit

dem

transponierten

(4)

Kern

n(s') + f(s) K(ss')ds=0

keine andere Liésung gestattet als die triviale 7 = 0. Indem Gleichungen (3) und (4) in der abgekiirzten Form

(E+K)u=y,

(E+

wir die

K)=0

schreiben, ahmen wir den Symbolismus des Matrixkalkiils so nach, als ob K eine quadratische Matrix wire und die Werte von w in einer Spalte, die von 7 in einer Reihe arrangiert sind. E ist der Operator »ldentitit*.

Aber die Méglichkeit, Gleichung (5) (E+K)p =0

oder

g(s) +f K(ss')p(s')as' =)

nicht-triviale

oder

@ besitzt,

geschlossen

Lésungen

werden.

dann lésbar, wenn orthogonal ist,

Die

ihre

da

y

die Gleichung

inhomogene

rechte

(Vy) =0,

Seite

di.

(4) nnd

kann

Gleichung

y(s)

zu

dann

natiirlich

(8) ist dann

auch

nicht

allen Lisungen

frls)y(s)ds

und

die

aus-

nur

y von (4)

= 0;

und die Liésung w(s) ist in dem Sinne vieldeutig, da zu ihr eine beliebige Lésung ¢(s) von (5) addiert werden kann. Nach dem Eindeutigkeitssatz hat das freilich auf das zugehérige Doppelbelegungs‘Potential (2) keinen Einflu8. Die Lésungen y von (4) bilden eine lineare Mannigfaltigkeit H von endlichvielen, sagen wir h, Dimensionen; die lineare Mannigfaltigkeit ® der Lésungen » von (5) besitzt die gleiche Dimensionszahl. Es ist nétig, in Analogie zu dem elektrostatischen Problem

der

Elektrizitatsverteilung

auf

vorgegebenen

Konduktoren

zunichst diese homogenen Gleichungen zu studieren. Man gelangt aber zum Ziel nur, wenn man dabei den Kreis der Eigenfunktionen in der aus der Elementarteilertheorie geliufigen Weise tiberschreitet. Wir stellen

kurz

zunichst

die

zusammen.

einschligigen

Tatsachen

der

allgemeinen

Theorie

Die charakteristischen Funktionen eines Kerns. Sei also K(ss') ein beliebiger ,hinreichend regulirer“ Kern. [Es geniigt z. B. anzunehmen,

da

K(ss’) gleichmiBig

gebung

Punkte

Integrale

r(s, s')

=)

gentigt,

nach

ist,

h

dem

da®

oder

a7, Null

ist,

d.h.

die Bilinearform

man

welche sein

det dij+0.

mub.

zeigt

den

direkt,

die Existenz einer Funktion

Liegt

» in @,

A’

so y

fiir 7H, eine

man

entweder

» © ®

lineare

(nig) =

0

fiir

aus

nicht-aus-

Kombination

i=

ieee

ergibt sich aus diesen Bedingungen

y, fiir die Ay=g@

von Integralgleichungen gilt. Da somit

441 y=0

Nach dem Hauptsatz iiber Integralgleichungen

= @ eine Lésung

in ©.

da®

folgert

tiber die Lésung

*) Hier ist ein rascher Beweis:

hat die Gleichung

(7)

Gleichungen

In der Tat

Fundamentalsatz

Dies

(7 = 1,..., 7)

Ex ist aber

», falls

6, = 6,,

orthogonal

somit

zu allen 7 € H,

pe®,,

ist.

Ay=p=0.

505

ist,

liegt

y

gF,---, 9;

=a,

=

in

O1,

=

®

und

linear unabhingig

0 zur

Folge.

darum

g

mod @

in

sind,

9.

Das

die

hat

Gleichungen

aber,

da

a, =---

Die Abbildung M. Wir kehren nunmehr zu unserm speziellen Kern K

zuriick.

legung

Neben

(2)

v(s),

betrachten

wir

das

Potential

einer

einfachen

Be-

v(p) = J Q(ps) »(s) ds.

(1)

Hier ist v(s) eine beliebige stetige Funktion auf 2; und wie immer erstreckt sich die Integration tiber ganz 2. Die Funktion v(p) geniigt sowohl in V als in W der Gleichung 4v+hk’v = 0, sie erfiillt die Ausstrahlungsbedingung, und sie geht stetig tiber & hiniiber. Hingegen erleidet die normale Ableitung 0v/On auf 2 den Sprung 2(s), indem

dieselbe

die

Werte

auf

der

Aufen-

und

Innen-Seite

F v(s')+ il v(s) K(ss') ds, annimmt. eine

Man

Potential

das

bilde

charakteristische

(7)

Funktion

im

Punkte

kiirzer

von

2

bzw.

dai

(s)

v(+ E+ K) fiir

insbesondere

7» des

s’

den

transponierten

Fall,

ist:

Kernes

v(p) = [ Q(ps')uls) ds’,

(8) und

bezeichne

seine

Randwerte

mit

g(s) = i; Q(ss') y(s') ds’.

(9) Ich

behaupte: 1.

Satz

Die

Gleichung

(9)

definiert

eine

lineare

>

Abbildung

9

= Mn von Hin ©. An Stelle dieses werde sogleich ein schirferer Satz bewiesen, in dem der Angelpunkt unserer ganzen Methode liegt: © Hi in itiber, Satz 2. Fiihrt die Abbildung M das Element

so geht durch dieselbe Abbildung y = (E+ K) in g = (E+K)@ iiber. muB

M

verwandelt also auch 7A™ in A"g; und wenn 7A”= 0 A"p = 0 sein. So folgt Satz 1 in der Form, daB M nicht

ist, pur

H, in ® verwandelt, sondern auch jede der Teilmannigfaltigkeiten ints (m =1,..., 1) in das entsprechende ®n.

von

Beweis

Formel

Greens

a

flu

Satz

2. Sei p’ ein Punkt

auf

Man

wende

o ul dp ° a utk v(d v)v+k —v ja) ds = f (ula Vv

2

an

im Aufern W.

Innere

das

V

und

die

von p: Q(pp’) und

Funktionen

f

beiden

(8). Man

folgenden

findet

daselbst

ns)ds = f Ql(sp')(6v/9n)(s)-ds. @Q(spV/av(s)

reguliren

506

Hier ist mit 0v/On die normale Ableitung auf der inneren Seite von 8 gemeint, welche den Wert 7(E+K) = 1A = yj’ besitzt. Andert man die Bezeichnungen p’,s in p,s' um, so erhilt man folgende fiir alle Punkte

Gleichung

giiltige

pW

fo Q(ps'\/dns) p(s’) ds! = i Q(ps') yl (s') as’.

(10)

Q

a

Lagbt man jetzt den Punkt p von auBen in riicken, so geht die linke Seite in g(s) tiber, wihrend man fiir die rechte Seite

einen Randpunkt s wo g' = (E+ K)q,

fas s')n'(s') ds’, d. i. das Bild in der Tat

das

Satz

3.

einzige

My’

von

7’,

erhilt.

Darum

ist, zugleich

mit

gp =

Mn,

gy = My. Die

Abbildung

Element

von

Hi,

M

das

ist

nicht-ausgeartet,

durch

M

in » =

Daraus folgt sogleich, da8 die Abbildung Beweis. Fiir das Potential (8) bedeutet

0

d. h.

4 = 0

tibergeht.

ist

M eineindeutig ist. die Gleichung Mn = 0,

da® seine (inneren und diuBeren) Randwerte verschwinden. Retticuschen Hindeutigkeitssatz verschwindet darum v(p)

Nach dem im ganzen

AuBenraum, und darum ist auch die normale Ableitung (dv/0n)a auf der AuGenseite von @ gleich Null, 7(— £ + K) = 0. Infolgedessen gilt

n(E+K)=2y Daraus

und

4 =

374A.

folgt aber

yWA= 4-94, durch

Wiederholung

Wir bilden Bilinearform behaupten: Satz 4. Die

ausgeartet.

nun

aus

y= }yA=—]-44',

dieses

GS

Garead:

und

oder

Verfahrens

21

irgend

schlieBlich

A,

zwei

Elementen

4,

4*

von

H,

die

C(n, 0) = (7%, Mn) Kapazitdtsform

Beweis. Die Symmetrie pliziten Ausdruck

C(y*, y)

ergibt

sich

ist ohne

symimetrisch weiteres

und aus

nicht-

dem

ex-

3) Q(ss') nls’) ds' ds. Die

zweite

Behauptung,

wonach

» =

0 das

einzige

Element

von

H;

ist, das fiir alle Elemente 7* von H, die Gleichung C(y*, 4) = 0 erfiillt, entspringt

aus

der Kombination

der beiden Tatsachen,

daB

1) fiir ein

507

festes 1 © H, das Element » = My fiir alle 7* € H, nur dann geniigt, Gleichung

My = 0 die andere

Wegen

der

Symmetrie

von @ der Gleichung (y* g) = 0 wenn » = 0 ist, und daB 2) die

7 = 0 impliziert

geniigt

es,

die

(Satz

3).

quadratische

Form

C(n7)

statt der Bilinearform C(y*, 7) zu betrachten. Mit Bezug auf eine Basis 7,,..+, %» von H; laBt sich jedes » © H, als eine lineare Kombination

(8) = @, 1,(8) +--+ + &n Mn (5)

der 4; schreiben, dratische

schen (Gap)

Form

}

und

C(77)

geht

cy, x; x; der

Koeffizienten

dann

in die

Variablen

2;

nicht-ausgeartete

tiber

mit

Gy — 10 (0.5 1)

den

qua-

symmetri-

(gp Sty

ee

Diese nennen wir die Kapazitatskoeffizienten. [Im Fall der eigentlichen Potentialtheorie, k = 0, ist die Form C(7) sogar nicht-ausgeartet, wenn

wir

gleich dem

uns

auf

iiber den

die

Elemente

ganzen Raum

7

von

H =H,

beschrinken,

da

sie

erstreckten Diricuterschen Integral

von (8) und darum positiv-definit ist. Dies hat tibrigens, wie man sofort sieht, zur Folge, daf 7 = 1 ist*). Im Falle der Strahlungstheorie

versagt dieser Integral iiber ist v komplex Sei 7,,..., .., 7 von

* == My¥

darum

SchluB, da hier an Stelle des Diricater-Integrals das den indefiniten Ausdruck (grad v)’ — k* v’ tritt; zudem und das Integral von zweifelhafter Konvergenz.] 7, eine Basis von H=H,. Wir wihlen fh Elemente H;, die linear unabhiingig sind mod Hj; ihre Bilder

(i = 1,...,h)

ist die quadratische

(12)

in

®

sind linear

Matrix

unabhingig

der

dij = (nin 9) = Elms,7)

mod @,

und

(i, j= 1,...,h)

nicht-ausgeartet. (Symmetrie kann man fiir sie natiirlich nicht erwarten.) nennen

Wir

Hie

(12)

sogleich

“ h-zeilige

die

noch

niher

analysieren,

D =

||di||

,grofen“

oder

Kapazitiitsmatrix

engeren

im

Kapazititskoeffizienten

in welcher

in der

Weise

,groBen*

Sinne.

diese

Wir

_,,klei-

n-zeiligen

Sone idishen Matrix C der cy enthalten ist. Lésung der Randwertaufgabe. Nunmehr sind wir in der Lage, eine mulissige Lésung w der Gleichung (1) im AuBenraum W fiir beliebig vorgegebene stetige Randwerte y(s) auf 2 zu konstruieren. Wir kénnen dabei

uns

entweder

der

,kleinen*

der

Matrix, der

Kapazitiitskoeffizienten bedienen. Im ersten Falle verfahrt man so: Aus einer Basis 7,,.--, %™ von H; bilde man die Basis gi = Mi von

®;. Die Funktion gi(s) besteht werten des Potentials

al

aus den (innéren

und

auBeren)

= [G(ps')n(s')ds'.

»€H, das In der Tat zeigt der erwihnte Umstand, da® ein in ® orthogonal ist, notwendig verschwindet. Fiir ein 4 in H, ist aber 4)

ches

Element,

Rand-

darum

ist 7’ = 0, d. h. y liegt

in H,,

oder

H, = Hy.

au allen 7 7! ein sol-

508

Von der vorgegebenen

lineare

Kombination

Randfunktion

der

g,(s),

y(s) subtrahiere

daB

die

Differenz

man

eine

solche

) = (9) -— I=13 apils) zu

allen

y; orthogonal

wird.

Das

ergibt

die

» Gleichungen

(my) = S cya; Diese besitzen

eine eindeutige Lésung

der Kapazitatskoeffizienten

(i,j =1,..., 7).

a;, da die symmetrische

Matrix

Cay = (ey Yi) = (Ni, My) = Cui, 11) nicht-ausgeartet ist. Fiir y*(s) hat die Integralgleichung (3), (E+ K)u* = y* eine Liésung y*(s), und diese fiihrt mittels (2) zu einem Doppelbelegungs-Potential w*(p) in W mit den Randwerten y*(s). Indem wir der Mannigfaltigkeit H; das Element 7 = 5) ajy; entnehmen und zu u* das

erhalten

der einfachen Belegung wir

in

ES

4(s) entspringende Potential

addieren,

u(p) = w¥(p) +S a7e;(p) eine

zulissige Lésung in WW mit den vorgegebenen Randwerten y(s). Dies Verfahren ist darum unbefriedigend, weil ja zur Lésbarkeit

der inhomogenen

zu

allen Elementen

H =H,

notwendig

Gleichung 4 von

H;,

(3) gar nicht die Orthogonalitat

ist. Darum

sondern

nur

ziehe ich vor,

zu

von

allen Elementen

y(s)

y von

Elemente ey 2) He Von

H, zu wihlen, die mod H; linear unabhingig sind, und auBerdem eine Basis 1,, ..., 7, von H =H,. Dann ist die Matrix (12) nicht-ausgeartet. Ich subtrahiere also von y(s) eine solche lineare Kombination DG oF 7

der 4 Funktionen = Myj, daB die Differenz 7* zu den Basiselementen 7,,..., 1%, von H orthogonal wird; das ergibt die in der Tat lésbaren

Gleichungen

h

(ny) = Di dia;

=

7=1

Fiir y* bekommen

belegungs-Potential

wir wie

w*(p) in W mit

h

2ST), dazu

addieren,

Randwerten

7(s).

vorhin

resultiert

ein w*

1,9).

und ein zugehoriges Doppel-

den Randwerten

y*(s). Wenn wir

F0) = [Glps')nfls') ds’, eine

zulissige

Lésung

u(p)

in

W

mit

den

Die Aufklirung tiber das Verhiiltnis der beiden Verfahrensweisen bringt die Gleichung (10). Sie zeigt nimlich, da das aus einer zu H} gehdrigen Belegung 1’ entspringende einfache Potential dem von der

509

Doppelbelegung

sondert

Verfahren

Belegungen

aus,

hinzufiigen

mu,

g = My

erzeugten

Potential

Schar

aus H; eine lineare

deren

Potentiale

sich

gleich

ist.

Das

zweite

>) a; 77(s) von einfachen

durchaus

nicht

in

Doppel-

belegungs-Potentiale umwandeln lassen und die darum das Minimum dessen darstellen, was nan zum allgemeinen Doppelbelegungs-Potential

um

die Losung

der Randwertaufgabe

fiir beliebiges

y(s) zu erzwingen. Genauere Analyse des Enthaltenseins der kleinen in der grofen Kapazitats-Matrix. Eine Basis (,,Normalbasis*) von H; kann in folgender Weise konstruiert werden. Man fiingt an mit einer Basis 6, = {no1. -.-, Nor}

Folge,

daB_

sind 6) = mod H;-2,

von

H; mod Hi.

2914,

...,

Die

ord’

linear

{n1,..., mor} Elemente 60 =

{m61, ..-, aor}

Unabhiingigkeit

unabhiingig

in H;1,

besteht

linear unabhingig sind mod H;-3,

mod Hj

sind,

und

hat

die linear unabhiingig

aus

Elementen

von

und so fort bis 6-).

zur

darum

H-2,

sind

die

So erhalten wir

r-linear unabhiingige Elemente, welche in die Abschnitte 6,, 6,..., 6(/-))

je von der Linge r =r, zerfallen. Darauf ergiinzen wir 71, ..., 40, durch Hinzufiigung von 7 weiteren Elementen 6; = {yu1,..., yir,} zu einer

Basis

vollen

und bilden

mod Hs

H;1

von

In dieser Weise fortfahrend lassen stehen, die aus den Abschnitten

of und

Die

der Gleichung

darum

geniigen

eine

Elemente

(f, i).

Sektionen

Ass s

von

6; gehéren

1 von

zu

der Ind

Hy

unterschieden sind, zerfallt selbst

Basis

Die

H, ent-

f=0,...,0-1)

7°~? = 0. Die Reihe

durch die unsere Basiselemente voneinander i

Normalbasis

(i=0,1,...,2-1—-f;

7; besteht.

der Linge

je von

wir

6,, 6, ..., 6.

wiederum

besteht

aus

Ketten

ip divecongs verschiedener

einer

Kette

Linge

von

der

1—/

Linge

(f =

/—/

0, 1,...,/—

gehért

zu

1);

oy.

Anfangselement

das

7

Bei Wahl dieser Normalbasis steht in der aus den Kapazitatskoeffizienten (11) gebildeten ,groBen Matrix“ C vom Grade n dort, wo sich

die Sektion index

kreuzt,

die

der Sektion

mit

(f, i) des Zeilenindex rechteckige

Matrix

Clu? ng),

wo

Cz; der

(g, j) des Kolonnen-

Elemente

m€ oy, ug © oy.

Nach der Regel (6) und Satz 2 hingt diese Matrix nur von i+j7 = 9 ab und werde darum mit C?, bezeichnet. Da 7 =0 ist, sobald

2/—g, verschwindet diese Matrix, o=2l—f, und i” = 0, sobald 9 wenn 9 =/—max(f, g). Sie verschwindet insbesondere, wenn entweder

ist. Nehmen

stellung

wir

oder

21-f

(47

fiir einen

der Zeilen

vor,

i+j7—1—1-—f,

Augenblick

welche

zustande

g>f

in der Matrix kommt,

wenn

C diejenige Um-

man die Num-

510

wiirts

durchliuft,

Form

C= Ch,

Be-

g>f,

ti +j=l—-1-f/,

oder

(f%) riick-

hinreichenden

Cj’;

von

Versehwinden

das

fiir

¢+j>l-1-f

in der

M

t=l-1-f-i,

die

so kann man dingungen

der Zeilensektionen

des Index

i= 0, 1,...,7-—1—/

mern

schreiben:

yi—44.9'>7.

moder

W >t

(13)

Bei alphabetischer Anordnung der Sektionen (fi) nach der Regel, dab (gj) auf (fi) folgt, wenn (13) besteht, tritt also fiir die Matrix C Redukein,

tion

j=i, besetzt nanten ar ve}

indem

g =f sind. der

= Cy’.

von

alle Rechtecke

C, die

auf

der

Seite

einer

durch

gekennzeichneten Hauptdiagonale stehen, mit Nullen Darum zerfillt die Determinante von C in die Determi-

in der

Hauptdiagonale

von

C stehenden

Matrizen

Cy}=

Setzen wir C7}/ = C;, so besteht demnach die Gleichung

(14)

[C] = +[C,|--- |Cial’.

Sie wo

beruht auf dem Umstand, daf& sich (f, i’) mit (f, i) kreuzt, die

einer Hiilfte teilung, da

in den L ,Hauptfeldern* von Matrizen C; stehen, wihrend

der L(L—1) Nebenfelder Nullen Reduktion eintritt. Die Regel

Chg = 9 ergibt, wie man leicht ausrechnet,

in dem

Raster

von

C den

Wert

aye

Ne

wahrend

ee

Ne ist. Also



tragen

noch

L(L—1)

z

in solcher

Ver-

fiir i+ 7 2 l—max(f, 9)

als Anzahl der leerstehenden Felder

(—1)-1-( + 1?

6

=

stehen,

C, in

(J—1)-1-@

7 +

1)-(1+ ee

2)

(—2)-(U—1)-1-0 41) 1.2.3.4

mehr

wire.

Die

Rechtecke

(fi) >< (gj) die Matrix

Endelemente

unserer

Ketten,

0 als

fiir die Reduktion

noétig

d. i.

Ge)

bilden zusammen

eine Basis 9,,..., 9, fiir

Indizes

2 in

(f is

0,

oy

Sql

1)

H=H,, wiihrend ihre Anfangs-

elemente, d. i. die 6;, zusammen eine Basis 7*, ..., n; von H; mod H; ergeben. So zerfiallt die Reihe der hier einen Augenblick benutzten 1, 2,...,

nattirlicher

Weise

in 2 durch

den

Index

a0,

511

kleine

, /—1

ergibt

Matrix

sich

unterschiedene D

ein

der

Raster,

g die Matrix Cl'~’

der

aupedieecnale

Hauptfeldern

von

triigt. die

DT man

will,

mag

man

von

Cp—

Cnty; )

im

Durchschnitt

gleichen

Somit

[ee die

Matrizen

Chore

Formeln

der

Linge

7;.

(Gil

Diese verschwindet

C stehen.

(15) Wenn

dafi

Sektionen

f> /, wiihrend lings auftreten,

die

in

den

benutzen,

um

Gal (14),

(15)

dazu

aus [D| + 0 die Ungleichung {C| + 0 zu erschliefen. Die Analyse zeigt, daB D bereits alle in C wesentlichen Bestandteile C; enthalt ;

withrend kommt,

aber die Matrix C; in den Hauptfeldern

tritt

sie in D

nur

einfach

auf.

von C (J — f)mal vor-

155. Die natiirlichen Randwertaufgaben im Aufenraum fiir Strahlungsfelder beliebiger Dimension und beliebigen Ranges

Mathematische Zeitschrift 56, 105—119 (1952)

lung

Die

von

der

mir

in

,Kapazitit“

einer

von

voraufgehenden

Note’)

Strahlungsfeldern

hat

gegebene

inzwischen

Behand-

durch

Herrn C. MUiier eine wesentliche Vereinfachung erfahren, indem es ihm gelang zu zeigen, da® hier so wenig wie in der Potentialtheorie Elementarteiler héherer Ordnung auftreten®). Im m-dimensionalen Euklidischen Raum ,(m = 2) existieren zwei zueinander duale natiirliche Randwertprobleme, das ,,elektrische“ und ,magnetische“, fiir ein

Schwingungsfeld

vom

Range

gy;

wobei

g

der

Werte

0,1,...,

7

fahig ist. Das magnetische Problem vom Range n—g ist mit dem elektrischen vom Range q identisch. Nachdem die Herren W. K. Saunpers und C. MULLER dem fiir n = 3, g = 1 sich ergebenden Randwertproblem des elektromagnetischen Feldes von neuem ihre Aufmerksamkeit geschenkt haben*), méchte ich auch die von mir fiir den allgemeinen Fall eines beliebigen n und q vorgesehene Behandlung bekannt geben, mit der Modifikation jedoch, welche die schéne MUtiersche Entdeckung méglich gemacht hat. Dafiir ist der heute

vielen Mathematikern rentialformen,

wie

ich

geliufige Carransche Kalkiil der linearen Diffeihn

kurz

in

§ 2

auseinandersetze,

das

ange-

eines

varia-

messene Werkzeug. Im Unterschied von neueren Untersuchungen harmonische Integrale ist das Operationsgebiet hier euklidisch, nicht-kompakt.

iiber aber

oule Die

blen

Im

&,, seien

Punktes

z,,..., z, die

Grundlésung.

p, k eine positive

Cartesischen

Konstante.

Koordinaten

Fiir irgend

zwei

Punkte

*) Kapazitét von Strahlungsfeldern, diese Zeitschrift 55 (1952), 187—198. *) C. Minter, Zur Methode der Strahlungskapazitét von H. WEYL, diese

Zeitschrift 56 (1952), 80—83. 5) W. K. Saunpers,

On solutions of MaxweEL1’s equations in an exterior region,

Proc. Nat. Ac. of Sciences, USA., 38 (1952), 342—348. ©. MiuEr, Randwertprobleme der Theorie elektro-magnetischer Schwingungen (erscheint demnichst in dieser

Zeitschrift).

513

Pp, p' bezeichne t(pp')—=p' p den von p' nach p fiihrenden Vektor, r seine Lange. Xe sei die Kugel um den Ursprung 0 vom Radius R. Indem wir zunachst

p' mit

o zusammenfallen

lassen,

bestimmen

wir

eine

kugel-

symmetrische, d.i. nur von r abhingige Lésung (Grundlésung) G(r) der Schwingungsgleichung 4u+k*u = 0 aus der Hanxexschen Zylinderfunktion H(r) der Ordnung > —1, die sich im Unendlichen asymptotisch wie e-#/r verhalt, indem wir G(r) = H(kr)/r? * setzen. Im Nullpunkt wird G(r) singulir wie const. 7—“—®) (man weiB, wie diese

Aussage

ktirlich

fiir n = 2 zu modifizieren

bleibender

ist).

Ein

zunichst

konstanter Faktor werde so normiert,

in G(r)

will-

daB der FluB

von G, namlich das Integral von - dG/dr iiber die Kugel vom Radius r, mit r+0 gegen 2 konvergiert. Die Wahl der besonderen Hankexschen Zylinderfunktion hat zur Folge, daB n—1



(grad

G+ ika*)

mit

r— co gegen 0 strebt (Ausstrahlungsbedingung). Ist r = r(pp') die Entfernung des variablen Punktes p von einem beliebigen festen Aufpunkt p’, so werde die Grundlésung G(r) mit

G(pp')

den

bezeichnet,

Komponenten

und

I’= I'(pp’)

0G/0z;.

Offenbar

sei der Vektor

grad,

G(pp’) mit

ist I’ = gradp G(pp') = —T.

is) 2 Kalkiil

der

schiefsymmetrischen

Tensorfelder

im

Euklidischen

Raum.

Ein schiefsymmetrischer Tensor mit den Komponenten fi, ...i, vom Range g werde kurz als g-Tensor, ein derartiges Tensorfeld als q-Feld bezeichnet. g ist die Summe

Das

innere

Produkt

Die

erstreckt

iiber

(f-g) zweier g-Tensoren

f und

Fas

die (7) Kombinationen

verschiedener

Ziffern

i,,..., ig

aus der Reihe 1,2,..., ”. Ist « ein Vektor mit den Komponenten «, so ist (fa) ein Tensor vom Range g—1, [fa] einer vom Range q+1, mit den Komponenten

(1)

Ge

eae

fa

eateats

ty Caen SS cai — [fle etee

S + zeigt eine alternierende Summe von g+1 Gliedern an, die dadurch zustande kommt, da® der letzte 4,,, in der Reihe der Indizes

schrittweise

die Regeln

(2)

von

hinten nach

(fa)e) =0

vorn

und

durchgezogen

[[fela] = 0

wird.

Man

beachte

514

(3)

insbesondere

«, B

Vektoren

fiir zwei

ergibt

triviale Rechnung

Eine

(FB) a) = 7(Be) — [(Fe) Bl,

fiir B = «:

(4)

f(aa) = ([fo] a) +[(Fa) a).

Wird der Vektor a vom Betrage 1, (w«) = 1, als ,Normale* angesprochen, so liefert diese Gleichung die Zerlegung von f in einen »tangentiellen* Tensor f: = ([f@]a) und einen ,normalen* /, = [(f«) a]; denn

wir

werden

dizes

7 gleich

f

normal

oder

tangentiell

nennen,

je

nachdem

[fa] = 0 oder (fa) = 0 ist. Wird @ als (0,..., 0,1) normiert, so faft fc die (‘7’) Komponenten fi, ... i, zusammen, fiir die keiner der g Inn ist,

Pythagoreische Die

folgende

/, die

Gesetz

Regel,

findet vielfache

Cy

Komponenten

fi, ... ig4n-

Daher

das

q+ 1

ist,

beiden

auf

(fg) = (he g) + (fr > G)-

in der

Verwendung:

(5)

/ vom

Range

g,

w vom

Range

statt (1)

die

7 (wa)) = ([Fa]- w). Nimmt

ein

g-Feld

man

« —0/02;,

/ wirkenden

:

Clie

Formeln

erhilt

man

Differentialoperatoren

(iv/i-, — > Den

so

gas |

(2), (4) entsprechen

div

und

rot,

Ohneat (ro ea t = SEE oe 2. tot1

die Gleichungen

(6)

div divf —0,

(7)

4f = div rot f+ rot div f.

rot rot? = 0;

Aus einem g-Tensor f entsteht in einer gegentiber eigentlichen orthogonalen Koordinaten-Transformationen invarianten Weise der

duale (n —q)-Tensor /* nach

Fiea4q'— Tereoy

in der 1,...,

i,,...,%, i4,,--., i irgend eine gerade Permutation von ist. Ersetzt man konsequent alle Tensoren durch ihre dualen,

so vertauschen rotf.

der Gleichung

.

sich

die Operationen

(fe)

und

[fa]

sowie

div/

und

§ 3: Greensche

Seien u,v zwei Felder vom beiden Vektorfelder F und F:

Fr= Fo

>) " : eee, thy tga SDR: diesen

Uy.

Formel.

Range

q.

Man

bilde

as (div), 0.0 tq—q-1 y= (rotten. ays,

>

die

folgenden

515

(die

q—1

Summen bzw.

erstrecken

g). Es

mehreren

sich

tiber alle

sei V ein beschrinktes

glatten Oberflichen

2

Kombinationen

Gebiet,

begrenzt wird.

das

von

Auf 2

(i,,...)

zu

bezeichne

»

einer oder

den Vektor der 4uBeren Einheitsnormalen. Ein variabler Punkt auf Q werde mit s (oder s') bezeichnet; bei Integration tiber den Raumpunkt p oder den Flichenpunkt s bedeute dp das Volumen-, ds das

Flachen-Element. und F

Anwendung

ergibt die

der

Gleichungen

Gavssschen

f {div w- div v) + (u- rot div v)}-dp =f

Integralformel

(ue)-div 0)

auf F

ds,

5

f {rot w- rot v) + (u- div rot v)}-dp = f (u-(rot x, »))-ds, 1

Q

aus denen durch Forme! folgt:

Addition

nach

(7)

die

fundamentale

Grernsche

ff {div w- div v) + (rot u-rot v)+(u-4v)}-dp

= i {((wv) «div v) + (w- (rot v, v))} ds. Q

Auf der rechten gentiell

sind,

Seite

div v und

kann

man,

wu auch

da die Felder (wy), (rot v,v)

durch ihre tangentiellen Teile

u ersetzen. Vertauscht man in dieser Formel u mit hiert, so ergibt sich die zweite Greensche Gleichung

[Re

uds = f {ul4v + kv) —v(4ut+k

Q

Hier

TON

subtra-

wh} dp.

((uv)- (div ve) + (ue: (rot v, v))

(esi (Gtavua este Abe

NT

gilt

(8) wenn wu und in V sind.

und

(div v),

Vv

ist

Insbesondere

v

tan-

[ &u, eds =0, v Lésungen

der

Schwingungsgleichung

4u+k*’u=0

§ 4. Eindeutigkeit. Der Raum &, sei durch eine oder mehrére glatte Flachen Q in ein beschrinktes Inneres V und ein zusammenhingendes AuBeres W geteilt. Zur Untersuchung stehen (komplex-wertige) g-Felder wu im Aufenraum W, die daselbst der Schwingungsgleichung 4u + ku = 0 geniigen. Eine derartige Lisung gilt als zulassig, wenn sie der Aus-

516

n—1

R?

besagt,

Diese

strahlungsbedingung geniigt. 2p gebildeten Ausdriicke

-{(divust+tik(ur)}

und

R

i

?

der Kugel

auf

die

daB

-{(rotu,v)+ikwu}

gleichmiBig in allen Richtungen gegen 0 streben, wenn der Radius R iiber alle Grenzen wiachst. Wir suchen eine zuldssige Lésung u, fiir welche (divu), und um auf 2 mit einem vorgegebenen tangentiellen (g—1)-Feld y bzw. g-Feld g zusammenfallen (elektrisches Randwertproblem).

Lésung

Der

Eindeutigkeitssatz

uw in ganz

W

verschwindet,

behauptet,

da

falls (div uw): und

m

eine

zulassige

auf 2

ver-

schwinden. Der Beweis folgt dem Retiicuschen Schema. Neben uw benutzt man das konjugierte Feld % Gema&& der Ausstrahlungsbedingung konvergiert das tiber 2p erstreckte Integral von

(div u) + ik(ur), (div a —ik(ar))

(9)

f ((rot u,v)

+iku,

(rot a, v) —ikt%)

mit R— co gegen Null. Sobald Zz ganz V einschlieBt, kann man die Greensche Formel (8) auf die beiden Felder uw, # in dem von Zp begrenzten Teil von W anwenden. Wegen der auf 2 geltenden Randbedingungen folgt alsdann, daB das iiber 2p erstreckte Integral von L(u, a) exakt gleich 0 ist, und aus unserer Aussage ergibt sich demnach, daB das Integral von

mit Roo

(I(div ue + |(rot a, v)) + gegen

0 strebt

und

(lee? + eee?)

demnach a

fortiori

das

Integral

von

[uel + [al = (ue) = Bl, ah Anwendung des Retiicuschen Lemmas auf Ui,...i, fiihrt zu dem gewiinschten Resultat.

Integralgleichung Die

der

Komponenten

805.

des elektrischen

Randwertproblems.

GroBen

(10)

G-}-

kénnen

jede

als

Gii5

++) Siity

Oni

onue “eg,

bg

die Komponenten

= Gir -.- tgs y+

einer

Matrix

ty

betrachtet

werden,

fiir

welche (i,, ..., iq) als Zeilen-, (k,, ..., kg) als Kolonnen-Index figuriert.

Fiir

variables

p

und

festes

p’

ist

die

einer

(i,,..., ig) entsprechende Zeile G:, +++ig(pp') ein Komponenten wi,,..., % = Gi, soe tay Bie ee ky

festen

g-Feld

Kombination

w

mit

den

517

Die

Greensche

gesuchten

zweckmiBig

Formel

Lésung

w(p)

gewihlter

(8)

weist

des

den

Weg

elektrischen

zur Konstruktion

Randwertproblems

Oberflichenbelegungen.

der

mittels

Sei » = g(s) ein ste-

tiges (q¢— 1)-Feld und f = f(s) ein ebensolches g-Feld auf 2; beide seien tangentiell, so daB y = ([py]v) und f = ({f»]»), sonst aber beliebig. Man bilde aus dem q-Feld w(p) = Gi, ... ig(pp') und dem Paar

vy = {p(s), f(s)} dieser beiden

(1)

Felder

den Ausdruck4)

—((wr)-p) + ((rot w, v)- f),

(wobei in w und rotw der variable Flichenpunkt s fiir p eintritt) und integriere alsdann nach s tiber ganz &: die entstehenden Zahlen Ui, .-+ ig(p') sind die Komponenten eines g-Feldes, das sowohl im Innen-

wie im AuBenraum als Funktion von p’ der Gleichung Zu+k?u = 0 geniigt und im AuBenraum der Ausstrahlungsbedingung geniigt. Es gebrauchen die Abkiirzungen

zu

s’,p

s, p’ gegen

Bezeichnungen

die

ist bequem,

g, @’ fiir y(s) und

vertauschen.

(s'),

und analog

Wir

fiir

alle Funktionen auf 2. I’ = I''(s' p) bedeutet nunmehr den Vektor grad, G(p'p) fiir p’ = s’. Integration iiber 2 nach s oder s’ werde durch il bzw. i angedeutet. Das beschriebene Verfahren ergibt den folgenden Ansatz:

(12)

u(p) = ulp; ») = — fle Es

gilt,

die

auBeren

div uw zu bestimmen.

Randwerte

Da

I’ = —TF

"1G + fv"). der Tangentialteile

und

somit

von

wu

und

errichte

man

[@.0 = - div ) fir, 1G ist, kommt Um

die

von

(6):

divu = — f'(g',= vI1 f'(ly' L) v1’).

die gesuchten

Normale

s den

vergieren.

delt

wegen

es sich

Randwerte

in s,

Abstand

wahle

+e

an

der Stelle s zu finden,

fiir p den

oder

—e

Punkt

hat, und

Indem wir fiir einen Augenblick also

zunichst

darum,

den

auf

dieser

Normalen,

lasse dann ¢ gegen

der

0 kon-

yp fiir [pv] schreiben, han-

Grenzwert

von

[div u,v] = f(T)» zu ermitteln.

(13)

Nach

(8) ist

[(y'D')y] = w'(L'y) - ((y'y], DP).

Es ist wohl bekannt, was aus dem Integral des ersten Teiles, f v' (Cv), wird, wenn « gegen 0 strebt, naémlich

+ o(s) + fo v'(I'). *) So verfuhr ich schon in einer friihen Arbeit iiber die Eigenschwingungen elastischer Kérper, Rend. Cire. Mat. di Palermo 39 (1915), 1—49, insbesondere S. 16.

518

Vorzeichen

obere

gilt das

Hier

aufen,

von

bei Anndéherung

das

un-

kann

im

tere bei Anniherung von innen, und in J” ist p= s zu setzen. Der Kern (I"(s's), v(s)) ist regular genug, da wegen der Glattheit der Oberfliche (= (s’s), v(s)) von positiver Ordnung fiir s’' = s verschwindet. Da die Komponenten von [v’»] = [[p’»’]»] die GréBen »j,» nur in vz(s')¥i(s) 1; (s’)— (s)

,schiefen“ Kombination

der

Teil der Grenziibergang von Wenn man im Endresultat

zweiten werden.

riickgingig

macht,

so erhalt man

enthalten,

p zu s unmittelbar ausgeftihrt die Verwandlung (13) wieder

[div u, ¥] (s) = + ¥(s) + f'[(w(s"), 1 (8's), v(9)] den

Wert

Tangentialteil

den

fiir

damit

und

+ ot f(y I") >] »).

Auf genau die gleiche Weise us, von u selber den Wert

bekommt

man

fiir

den

tft fv) +) »)— f G(ly'»')y]»).

Sei y = {y, g} ein gegebenes Paar

bestehend

aus

tiellen (¢g—1)-Feld y und einem tangentiellen g-Feld elektrische Randwertproblem, ein zulaéssiges AuBenfeld

fiir welches

baw.

die

System

auf

dem

vorgegebenen

Rande

Werte

von Frepxotmschen

& die Tangentialteile y und

g

annehmen,

Integralgleichungen

Tangentialteil

einem

tangen-

g auf 2. Das w(p) zu finden,

von

divw

ergibt

und

dann

u

das

y= 9+ f(y’) I)2)»), g=F+ fC IL) 1») = [' Gilg») } 9),

(14) die

s

an der Stelle

divw

(div wv); von

wir

symbolisch

zu

y= (E+K)p

zusammenfassen mit dem Einheitsoperator E. Wenn y und g tangentiell sind, sorgt diese Integralgleichung von selbst dafiir, daB die Lésung {y, a

aus

zwei

tangentiellen

Feldern

besteht.

Die Méglichkeit der Existenz von nicht-trivialen Loésungen 9 der homogenen transponierten Gleichung »(E+K) = 0 erfordert das

Eingehen

auf

§ 6. Das

duale

Problem.

Dazu verfolgen wir den zu (12) dualen »Magnetischen* Ansatz. Wiederum wahlen wir eine stetige Belegung » = {n, e}, die aus zwei tangentiellen Feldern 7 und e auf & vom Range q—1 bzw. q besteht,

519

und

integrieren

nach

verstehend.

des

folgenden

(15)

von

sp’

an

Stelle

von

(11) den

Ausdruck

(div w- 7) +(w-e) w wiederum eine der Zeilen

s tiber 2, unter

(10)

nung

diesmal

Wir

unterlassen

in s'p. Wir

q-Feldes

Gi... i, (Sp') von

diesmal die Abinderung

erhalten

so

die Komponenten

der Bezeich-

Vi, +++ ig (P')

v(r') = f nt] + fea,

wo I und G fiir I'(sp’) und G(sp’') stehen. Es gentigt als Funktion von p’ im Innen- und AufSenraum der Schwingungsgleichung, und

der Ausstrahlungsbedingung im Auf enraum. Jetzt lassen wir den Punkt p' von auBen oder innen gegen den Punkt s’ von & auf der

in s’ errichteten Normalen v' = v(s’) riicken. Es ist zu bestimmen, was dabei aus ~ (v(p’), v') und (rot v(p'),v') wird. Was —(v(p’), »’) anlangt,

v(p') zum

so ist klar,

Grenzwert Im

G(ss') steht. Umwandlung

da

den

Beitrag

—[(er)G

ersten Teil ([yI°]v’) nehme

(15)

von

wieder

die

des Ausdrucks

der zweite Summand

leistet,

man

wo

zunachst

G jetzt

fiir

— (aT)y’) = —a(Pv') + [nT]

vor und nutze dann die tangentielle durch ([yv]v) ersetzt; dann hat man

Natur von 7 aus, indem man mit dem Integral if von

4

= n(L0') + (yr) 9) ¥) PJ

zu tun. Aus dem gleichen Grunde wie friiher ergibt sich fiir dieses Integral im Limes + y(s') plus dem Integral iiber den gleichen Aus-

druck,

in dem

nun

p’ durch

den Punkt

s’ auf 2 ersetzt ist, und aus

dem gleichen Grunde ist auch hier der Kern hinreichend regular. Indem man den Ausdruck in seine urspriingliche Form zuriickverwandelt, erhalt man darum

= (vr) = F a + f nr) P= f (ero) »)G.

Dabei ist noch I durch — I” ersetzt worden. Fiir ein Feld v(p’) wird man unter rot’ natiirlicherweise den Operator rotp

verstehen.

Wegen

flat) =—f nr]

ist

und

rot!v = fel”) =-fler, durch

wurde,

von

=—rot' fnG

dieselbe

kommt

Ist darum

—(vv)

kleine Rechnung,

(rot' v, v') = 4 = (4,1)

und

das

die

oben

fiir [yJI’] ausgefiihrt

= e +f (((ev] v) I] v'). Paar,

(rot v, v) besteht,

das

aus

den

inneren

Randwerten

so resultiert die Gleichung

A=n(E+K).

520

In der Tat demjenigen,

Denn

gemi®

ist der Operator, der 9 in 4 iiberfiihrt, transponiert | zu der nach den Gleichungen (14) y aus y hervorgehen laBt.

(5) ist ja z. B.

(f o(((fer] v) I") »')) = (fF v'] £) »] ») ee), da nach jener Gleichung die Vektor Faktoren von e in umgekehrter Reihenfolge an /’ angehiingt werden kénnen, indem man jede runde in eine eckige, jede eckige in eine runde Klammer verwandelt. Die Gleichung

(16)

9 (E+K)=0

bedeutet

—(vyv)

also,

und

(rot

da®

v,v)

fiir v(p)

verschwinden.

GréBen den Sprung — 27

&

von

innen

nach

Der

= v(p3 9)

aufen.

Operator

bzw.

M

—2e

die

Auf

inneren

jeden

Fall

beim Durchgang

Randwerte erleiden

von

diese

durch die Fliche

§ 7. und

die Kapazitatsmatrix.

Die Lésungen , 7 von (E+ K)y = 0 bzw. »(E+K) = 0 bilden lineare Mannigfaltigkeiten ® und H von derselben Dimension h. Es sei p’ ein Punkt im AuBern W. Dann ist G(pp') als Funktion von p eine regulare Losung der Schwingungsgleichung fiir p €V. Wir wenden die Gleichung (8) an auf das q-Feld w(p) = Gi, ...i,(pp'), das an Stelle von

wu tritt,

und

das

q-Feld

v(p) = v(p; 7),

nommenen Belegung 9 durch (15) definiert Paragraphen zeigen, daB die Randwerte von v und div v existieren und auf der Q die gleichen Werte haben. M bedeutet

das

mittels

der H

ent-

ist. Wir werden im nachsten f und » der -Tangentialteile Innen- und AuBenseite von den linearen Operator, durch

den 9 in » = {g, f} iibergeht, p = Mn. Beriicksichtigt man, daB die inneren Randwerte von —(vyv) und (rot », v) versthwinden, so ergibt unsere Gleichung +

[& (w- r) 9) + (rot w, »)-7}}

oder

LaBt

man

so kommt

u(p;~)=0

fiir

p= My,

jetzt p von

aufen

in einen

=0

7 €H und pew. Flachenpunkt

s hineinriicken,

(E+ K)p = li),

Verfahrt man in gleicher Weise mit (p), so zeig t sich, wie Herr MULLER zuerst bemerkte, da8 nicht nur y = My, sondern auch das konjugierte » derselben Gleichung geniigt. Sind » = {y, e}, » = {g, ft irgend zwei Belegu ngen von der hier durchgehend benutzten Art, die erste als Zeil e, die zweite als Spalte

521

aufgefaBt,

so kann

man

das innere

Produkt

(n, ep) = fa-9)+ fen bilden. Aus Bilinearform

zwei

Belegungen

7, 7*

von

der Art

wie 7 entsteht

die

C(y, *) = (y, Mn*).

Ihre Symmetrie beweist man am besten, indem man die Greensche Formel (8) auf v = v(p; 9) und v* = v(p; 7*) anwendet, nicht jedoch fiir

V,

so groB

da&

sondern

ist, dai

—(vv)

iiber erleiden,

man

das

Innere

der

Kugel

ganz V umschlieBt,

2p.

wahrend

durch

gy = (div v); und

welche

Integral

2 {(7,

von

hier &(v, v*) in der Form

Sobald

der

erhilt man dann,

(rot v, v) den Spring —2n

erstreckende

baw. —2e

Radius

R

bedenkend,

tiber & hin-

f = uv stetig hindurchgehen,

My*) —(y*, My)}

&(v, v*)

verwandelt

in das iiber Zp wird.

Schreibt

(vv), (div v¥) + ik (v* v)) + (v1, (rot v*v) + ik vé) — ((v* v), (div vp + ik (vv) — (oF, (rot v, v) +iku),

(17) so

2

und

eine Gleichung, sich

fiir

sich

ergibt

mit Roo

aus

gegen

den Ausstrahlungsbedingungen, 0 konvergiert,

(18)

daf

folglich

f&wrryds

2R

(9, My*) — (*, My) = 0. Das

entsprechende

Verfahren

fiir v und #

liefert zunachst

29, Mm) — 2, Mn) = f &(, ds. Parallel zu (17) bildet man®)

(19)

{

a

((vv), (div 0),—ik (ov) + (v%, (rot %, v) —ik%) a i ‘ 2 ; —((vv), (div vk +ik(vv))—(%, (rot v, v) +ik~).

Gemi8 der Ausstrahlungsbedingung konvergiert das Integral Ausdrucks tiber 2p mit R— co gegen Null, somit

dieses

k- [ |vPds+2-3(q, Mn). 2R

Daraus

(20)

folgt die Ungleichung

3, M7) 2 0,

und nach Retuicus Lemma kann hier das Gleichheitszeichen nur gelten, wenn vu(p) in ganz W verschwindet; diese Bedingung hat das Ver-

5) Dieser Gedanke riihrt von Herrn den Rexitcuschen Ausdruck (9) fiir v.

Mixver

her,

nur benutzt

er

statt

(19)

522

(21) zur

(rot v, v) oder die

und

—(vv)

von

der duferen Randwerte

schwinden Gleichung

n(-E+ K)= 0

Folge.

Die Beziehungen (18), (20) samt diesem Zusatz gelten unabhingig davon, ob 9 zu H gehért. Ist aber 9(E+K) = 0, so fiihrt (21) auf nm = 0. Fiir die Elemente 9 von H gilt somit (20*)

3a, Mn)> 0

aufer

fiir

fiir

= 0.

da8

My

von H auf ©® ein-eindeutig.

Aus

einer Basis 4; (i =

wenn

kann,

»

ist,

= 0

und

darum

ist

die

Null

dann

nur

7 €H

lehrt,

Ungleichung

Diese



Abbildung

sein

= My

1,...,h)

von H

einzige

Linear-

entspringt darum eine Basis y: = Mm, fiir ®. Es gilt der fundamentale Satz: Die Kapazitétsmatrix der ci = (ni g;) ist nicht-ausgeartet.

Zum

Beweise

kombination

mu8

der 9; ist,

man

zeigen,

die zu allen

daB

7 = 0

gj; orthogonal

die

ist.

Da My

zu

®

gehort, folgt aus (ng) =0 (j = 1,..., A) in der Tat (4, Mn) = 0, was der Ungleichung (20*) widerspricht, es sei denn 7 = 0. Danach

fiihrt das

Randwertaufgabe:

Element

2a«;y; von

Von

klassische Vorgehen zur Lésung der elektrischen dem

®, da®

gegebenen

y subtrahiert

man

ein solches

die Differenz y* zu allen 9; orthogonal

(iy)

= 2 cy ay

ist.

@,7=1,..., 4).

Fiir y* hat die inhomogene Integralgleichung (E + K)g* = y* eine Loésung y*. Zum ,elektrischen Schwingungspotential* w*(p) = u(p; p*) der so gefundenen Belegung g* addiert man das magnetische Potential v(p; 9) der H entnommenen Belegung 9 = 2; «;;. Die Summe u(p) ergibt die gewiinschten duBeren Randwerte {y, g} fiir (div uw), und w. § 8. Die

Randwerte

von

2, und

(div v),.

Ein heikler Punkt, der im Falle g=0 noch nicht auftaucht, ist der Nachweis der Existenz und des stetigen Durchgangs der tangentiellen Teile vy, und (div v); des im Innen- und AuSenraum durch (15)

bestimmten

Potentials

v. In der

Tat,

fiir

g

0 hat

man

einfach

v(p') = fels)G(sp'), und

es

ist

sofort

klar,

daB

die Randwerte

von

v auf

der

Innen-

und

AuBenseite durch v(s') = foes) Gss’) gegeben sind. Fir g>1 erfordert die Bestimmung des Oberflichenwertes des tangentialen Teils von far) eine zusitzliche Uberlegung, die aber aus der klassischen Potentialtheorie wohlbekannt ist. Wenn p lings der Normalen in s

523

in den Oberflachenpunkt aufen der gleiche Limes, Integrals

s hineinriickt, ergibt sich von innen und n&émlich der Hauptwert des uneigentlichen

fn),

LG's], »(s)] »(s).

Dabei wird es dann freilich nétig, von y(s) vorauszusetzen, nicht nur stetig ist, sondern auch einer Ho.per-Bedingung

Der

Hauptwert

des

Integrals

tiber

2

wird

bestimmt,

indem

daB es geniigt.

man

aus

um den Punkt s zunadchst ein kleines kreisférmiges Loch ausstanzt, dessen Radius ¢ man dann gegen 0 gehen l&éBt. Genauer gesagt: in der Tangentenebene im Punkte s von & beschreibt man um s einen (n—1)-dimensionalen Kreis vom Radius ¢ und errichtet iiber ihm in Richtung der Normalen v einen geraden Kreiszylinder; dieser wird zum Ausstanzen des Loches benutzt. Die Rechnung gestaltet sich am einfachsten, wenn man s als den Ursprung des Koordinatensystems annimmt und der z,-Achse die Richtung der Normalen vy gibt, v = (0,..., 0, 1). Dann besteht [y'I}: aus denjenigen Komponenten ,

d

{y ENS

von

[n’/'], in denen

dG,

,

alle Indizes i,,..., i++

sofort das gewiinschte

,

cote Dies pe s E My... ty, Lig

Resultat.

Ohne

sind.

Normierung

Daraus

ergibt sich

des Koordinaten-

systems kann man das Argument auch darauf stiitzen, da® jede Komponente i,,...,%—1,%, von {[n/I]v] die Vektoren /’ und » nur in einer der schiefen Kombinationen Iy7— I, enthiilt. Hier

Fiir ¢ = 1 ziehen

ist ferner der Oberflichenwert von (div v), zu bestimmen.

wir zunichst

vor,

die Bezeichnung

sp’

beizubehalten

und

z. B. divp v(p’) als div’ v. Zu dieser Divergenz liefert der Teil jj eG

von

nicht mit s’p zu vertauschen. Wir operieren also mit v(p') und schreiben v(p') den Beitrag [ (e Le

— fer).

Fiihrt man

[ev] ='c

(el) = ((cr)F'), und dies ist wiederum ein Ausdruck, in der schiefen Kombination »; 1), —

daB

e und

darum

auch

sich als Grenzwert

c

einer

ein, so ist

der v und I’ nur

J; enthilt. Setzt man also voraus,

HoLper-Bedingung

von | (eI”) der Hauptwert

geniigt,

so ergibt

des Integrals

(cls) »(9) P'(ss')), wenn ob

p’ auf der Normalen

v' in s’ dem Oberflichenpunkt

s’ zustrebt,

dies nun von der Innen- oder Aufenseite her geschieht. Bleibt noch der Beitrag zu diskutieren, den der Teil [ (nT)

v(p') au div’v liefert. zu tun,

auf

die

man

Hier hat man

in der

von

es mit der gleichen Schwierigkeit

klassischen

dreidimensionalen

Potential-

theorie bei der normalen Ableitung des Potentials einer Doppelbelegung

524

st6Bt. Ich verfahre so, wie ich es seit langem in Vorlesungen tiber Potentialtheorie zu tun pflegte. Es ist

furl =—f ti), P'(sp)] = — rot! f (als) G(s7'). Darum ist div’ f [ DP) S—div rot! f (m G); da aber div rot = 4 —rotdiv und 4'G = —#°G ist, so kommt

(22)

div’ f [nP'] = k- f

Was

mit

dem

ersten

Oberflichenpunkt

Summanden

s’ hineinriickt,

(7G) + rot! f (I). rechts

ist klar:

geschieht,

er geht

in

wenn

p’

in

den

k- [ (8) G(ss!) ds liber. Wir

in 22), liber

untersuchen

eine

fal)

jetzt das

nur

=-fa@l)=—f

geschlossene

Oberflache

fiir

g = 2 auftretende

zweite Glied

sich

Integral

(TP),

8

wo ®= [yx].

erstreckende

Um

das

zu

berechnen, bedeckt man die Flache ,dachziegel-artig“ mit sich tiberlappenden Gebieten, die je auf ein Koordinatensystem £,,..., f:-1 be-

zogen sind. Man kann dann# additiv in Bestandteile zerlegen (DiEUDONNS), deren

jedes

nur

in

einem

der

Stiicke,

genauer

in

einem

kompakten

» Kern“ eines solchen Stiickes nicht verschwindet. Sei 7; = «;(&,,..., &,4)

die Parameterdarstellung der Flaiche in dem von dem Parametersystem & bedeckten Stiick. Wir haben angenommen, da8 diese Darstellungs-

funktionen 2;(&) stetig differentiierbar sind, da die Ableitungen einer Hotper-Bedingung geniigen und daB die Jacosische Matrix



die aus der hingeschriebenen Zeile entsteht,

indem man

n—

1 Zeilen

untereinander schreibt, in welchen € durch £,, +++, Eri ersetzt ist —, vom Maximalrang n—1 sei. Das ist, was wir eine glatte Fliiche nannten.

Wir

fiigen jetzt die Voraussetzung

hinzu,

da® auch

die zweiten

Ab-

leitungen von 2;(&) existieren und einer Hovper-Bedingung gentigen. Wir sprechen dann von einer glatt-gekriimmten Fliche. Es gilt z. B.

(richtige Orientierung des Systems

So ergibt sich fiir (», T,—»,I,)-ds ee 0&

vorausgesetzt)

Ox. =|, bees 0aeay SHS cco eee,

yds

0G Ox,

§,,.--, 1

OG Ove

Ox, Oe

Aus TOE

der Ausdruck OXy, aoe

“Oe,

see

dba.

525

Indem

man

0G

zu

0G

FER

der

ersten

Spalte

dieser

Determinante

die

bzw.

mit

EE ., multiplizierten folgenden Spalten hinzuaddiert, verwandelt sie sich in OG

Ox,

Oz,

Of 2 oHe

Ich

nehme

jetzt

an,

da&

4

und

We

darum

auch

®

als

Funktion

der £

differentiierbar ist und die Ableitungen einer Hotper-Bedingung geniigen. Dann kann man auf das Integral

(ieee |ac es da,aedOty

c

(23)

-d&,... dena

partielle Integration so anwenden, da& G von der Differentiation befreit wird. Auf diese bees verwandelt sich (23) in

—{GSchreibt

(24) wo

man

Ziffern In

= S+|—

sich alternierend

1, 2 mit 3,...,

(25)

ey

Oe

Wa... 2(s)ds die Summe

a

eae

Ax,

foamy

ae

tiber die

erstreckt,

|

Oh

G)

dees 02n

aoe 3 ae "5, 1 Linn, n(n —1)/2

so erhalt

man

Mischungen

der

Sal) = f als) G(sp)-ds = fue. Wahrheit

bezieht

sich

die

Zwischenrechnung

auf

ein

durch

die

Koordinaten & bedecktes Stiick von ® und den zugehérigen Bestandteil von #. Aber wegen der Invarianz des durch (24) eingefiihrten w

gegeniiber Transformationen der Koordinaten

& ist im Endresultat (25)

die Zerstiickelung von 2 und # wieder tiberwunden. In diesem Stadium finden wir es zweckmiBig, die Bezeichnung sp’ wieder gegen s’p zu vertauschen. Es handelt sich dann darum, wahrend p auf der Normalen v im Flachenpunkt s gegen s riickt, den Grenz-

wert des tangentiellen Teils von rot f(y! G) = ii u' I] zu ermitteln [G = G(s'p), [= I'(s'p)]. Das geschieht genau so, wie vorhin fiir Var

verfahren

wurde.

grals f'[u(s’), I'(s’s)k.

Das

Resultat

ist der Hauptwert

Geniigt 7 der Gleichung »(#+ K)=0, so ist 7 =—yn

des Inte-

K=7 K’—...,

und daraus schlieBt man leicht, da » unsern Forderungen geniigt, unter der oben gemachten Voraussetzung, da8 & glatt-gekriimmt ist.

§ 9. SchluBbemerkungen. Wird in der elektrischen Randwertaufgabe y(s) = 0 angenommen, so erfiillt die Divergenz w= divw der im AuSenraum konstruierten

526

zulissigen

Lisung

u(p)

die Randbedingung

w=

0 auf 2.

ist divw = 0 in ganz W. Da man aus dem Ausdruck von daB

auch

w der Ausstrahlungsbedingung n—1

R

?

{(rotw, v) +ik we} aut

> 0 mit

AuBerdem

wu entnimmt,

R-> oo

geniigt, so ergibt sich nach dem auf w anzuwendenden EindeutigkeitsArgument, daB w im AuBenraum identisch verschwindet, divu = 0.

Damit reduziert sich die Schwingungsgleichung auf div rot w+ k’u = 0. Ohne diese Bemerkung ware die Behandlung des elektromagnetischen Feldes

unvollstandig.

156.

Uber den Symbolismus der Mathematik und mathematischen Physik Studium generale 6, 219—228 (1953)

Die vertrauteste und wohl auch grundlegendste Form, in der in unserem geistigen Leben die symbolische Funktion, die Reprasentation durch Zeichen auftrite, ist die Sprache. Darum meint H. Noack (11, p.97)'z »Das mit der Ausbildung der Sprache einhergehende Symbolverstindnis kann als der entscheidende Schritt des Menschen iiber das animalische Leben hinaus bezeichnet werden. Wie in einem Brennpunkt treffen sich hier die groRen philosophischen Probleme: das des Verhaltnisses von Sachverhalt — Gedanke — Aussage, an dem noch jeder Versuch einer Beruhigung im Realismus blo&en Seins gescheitert ist; aber auch des Verhiltnisses vom Ich des immanenten, im Strom des nur-eigenen Erlebens stehenden Bewuftseins zu dem individuellen, der Welt und dem Tode verhafteten, mit ‘Wesen seinesgleichen kommunizierenden Menschen. »Viele andere sinnliche Gebilde, die zum Zwecke der Bezeichnung, Vergegenwartigung, Mitteilung usw. verwendet werden, verdanken ihre ,semantische® oder ,symbolische’ Bedeutung erst der Sprache, insofern sie nach deren Analogic geformt oder mit deren Hilfe verabredet werden.* (11, p. 19). Haufig wird die Ansicht vertreten, da das begriffliche Denken an die Sprache gebunden sei; das Tier (das ja sehr wohl in seiner Welt sich zu orientieren vermag) entbehre darum mit der Sprache auch des Begriffs. So sagt Wilhelm v. Humboldt in der Einleitung zu seinem Kawi-Werk (Bd. I, Berlin 1836, pp. 68/69): »Indem in ihr (der Sprache) das geistige Streben 1 Die

Nummern

in

Schrigschrife

Ende stehende Literaturverzeichnis.

verweisen

auf

das

am

sich Bahn durch die Lippen bricht, kehrt das Erzeugnis

desselben

Vorstellung

wird

zum eignen Ohr also

in

wirkliche

zuriick.

Die

Objektivitat

hintiberversetzt, ohne darum der Subjektivitat ent-

zogen zu werden. Dies vermag nur die Sprache; und

ohne

diese, wo Sprache

mitwirkt

auch

still-

schweigend immer vorgehende, Versetzung in zum

Subjekt zuriickkehrende Objektivitat ist die Bildung des Begriffs, mithin alles wahre Denken, un-

méglich.“ Aus einer kilteren Zone klingt das Gleiche wieder in Ludwig

Wittgensteins Tractatus

logico-

philosophicus (15, 5.6 und 5.62): ,Die Grenzen

meiner Sprache bedeuten die Grenzen meiner Welt ... da die Welt meine ist, das zeigt sich darin, da

die Grenzen der Sprache (der Sprache, die ich allein

verstehe) die Grenzen meiner Welt bedeuten ...“. Merkwiirdig, wie solipsistisch hier die Sprache ge-

sehen wird; deren existentieller Ursprung und Auf-

gabe doch wohl in erster Linie in der Kommunikation liegt*. Hierin stimmen Denker so weiter

geistiger Distanz wie Jaspers und Hilbert iiberein. (Ich denke an die Rolle der Sprache zur Mitteilung tiber die Weise, wie mit den selber bedeutungslosen

2 Ich wei®, da® dies auf das Ganze der Humboldtschen Sprachphilosophie nicht zutrifft. An der angefiihrten Stelle fahre er, nachdem noch einmal eingeschirft ist, da® das Sprechen eine notwendige Bedingung des Denkens des Einzelnen in abgeschlossener Einsamkeit“ ist, so fort: yIn der Erscheinung entwickelt sich jedoch die Sprache nur gesellschaftlich, und der Mensch versteht sich selbst nur, indem er die Verstehbarkeit seiner Worte an Andren versuchend gepriift har.“ Man lese den ganzen Absatz bis zu dem michtigen Finale: Alles Sprechen, von dem einfachsten an, ist ein An-

kniipfen des einzeln Empfundenen an die gemeinsame Natur der Menschheit.*

528

Zeichen und Formeln in Hilberts formalisierter Mathematik zu verfahren ist.) Uber die wissenschaftliche Kunstsprache im Gegensatz zur Sprache des taglichen Lebens, in der sich der Mensch in seinem Umgang mit Welt und Mitmenschen ausspricht, heifft es bei Karl Vofler (12, p.225): »Vor den mathematischen und naturwissenschaftlichen Begriffen gelten alle Sprachen als etwas Auferes gleich; diese Begriffe sind fahig, sich in jeder Sprache anzusiedeln, da sie nur in der auferen Sprachform Wohnung nehmen, die innere aber aufzehren und entleeren. Nicht alle Symbole sind sprachlicher Natur. E. Cassirer handelt in seiner Theorie der symbolischen Formen der Reihe nach yon Sprache, Mythos und den Konstruktionen der wissenschaftlichen Erkenntnis, Zahlen wir einige besondere Formen auf, wie Wort, Bilder- und Buchstabenschrift, Rangabzeichen, Fahnen, allegorische Attribute, Traumsymbole, Kunstwerke, magische und religiés-kirchliche Symbole sowie die Zahlen und andere Begriffssymbole der exakten Wissenschaften, so wird uns klar, wie mannigfaltig der Sinn ist, in dem ein Zeichen auf das Bezeichnete hinweisen kann. Immer bleibt die Deutung cin Problem; diese mag sogar mehrschichtig sein. Eine steinzeitliche Hohlenzeichnung etwa, auf der Biiffel und Boot und Baum dargestellt sind, wird vielleicht durch diese gegenstindlichen Dinge hindurch auf Geister und Damonen zielen. Leibniz erklart die Zeichen fiir ,gewisse Dinge, durch welche die gegenseitigen Bezichungen anderer Dinge ausgedriickt werden und deren Behandlung leichter ist als die der letzteren“. Das macht sie zugleich zu cinem Werkzeug fiir die Entdeckung neuer, in den dargestellten Bezichungen griindender Zusammenhinge. Das Operieren mit Zahlzeichen ist das eindruckvollste Beispiel. In der reichen Skala von modi der Bedeutung, an die wir eben erinnerten, sind die extremen Falle: auf der einen Seite das Zeichen, das eine getreue Reproduktion des Bezeichneten ist (oder anstrebt), auf der andern das rein konventionelle oder gar das ,leere“ Zeichen, von dem Hilbert versichert, da es tiberhaupt nicht tiber sich hinausweist. Helmholtz sagt von der Qualitat unserer Empfindung, ,insofern sie uns von der Eigentiimlichkeit der auSeren Einwirkung, durch welche sie erregt ist, eine Nachricht gibt, kine sie als ein Zeichen derselben gelten, aber nicht als ein Abbild', da sie ja zugleich ganz wesentlich von der Natur des Sinnesapparats abhingt, auf den gewirkt wird. E. Cassirer (6, pp. 34—41) sieht in der Weise, wie sich fiir unsere Anschauung der Raum aus systematisch verkniipften und aufeinander bezogenen Perzeptionen aufbaut, die ,Urfunktion der

Reprasentation® tatig; nicht ganz verstandlich ist es mir, wenn er dann fortfahrt, daf§ man auf diese

snatiirliche Symbolik* zuriickgehen miisse, wenn man

odie kiinstliche Symbolik begreifen wolle, die sich das

Bewuftsein in Sprache, Kunst und Mythos schafft*.

In der Deutung der uns gegebenen Empfindungsdata und Wahrnehmungen, die (nach Helmholtz) »Zeichen® fiir die Wirklichkeit sind, mute die Wissenschaft bestrebt sein, die innewohnende Subjektivitat zu tberwinden. Die von ihr dazu entwickelten Grundbegriffe enthiillen sich letzten Endes als vom Geist in freier Schépfung dem Gegebenen gegeniibergestellte Symbole; und es ist gerade diese zur symbolischen Konstruktion gewordene thcoretische Erkenntnis, nicht eine reine Phanomenologic der

Natur, die es uns erlaubt, Ereignisse vorauszu-

sagen. Um ein Beispiel zu nennen: ,Die abstrakte chemische Formel, die als Bezeichnung eines bestimmten Stoffes gebraucht wird, enthale nichts mehr von dem, was die direkte Beobachtung und die sinnliche Wahrnehmung uns an diesem Stoff kennen lehrt; — aber statt dessen stellt sie den besonderen Kérper in einen auferordentlich reichen und fein gegliederten Bezichungskomplex ein, von dem die Wahrnehmung als solche tiberhaupt noch nichts wei®.* ,,Was hier die eigentliche Kraft des Zeichens ausmacht, ist eben dies: daff in dem Mafe, als die unmittelbaren Inhaltsbestimmungen zuriicktreten,

die

allgemeinen

Form-

und

Relationsmo-

mente zu um so schirferer und reinerer Auspriigung gelangen.“ (6, pp. 44—45.)

Die Symbole, deren sich der Mathematiker am haufigsten bedient, sind Scbriftzeichen; z.B. die aus hintereinander gestellten Strichen bestehenden pnatiirlichen* Zeichen fiir die natiirlichen Zahlen, wie [II fiir drei und [III] fir fiinf. So trite fiir ihn die fixierende Schrift neben die der Kom-

munikation dienende Sprache. Sie ist wichtig, wenn ich recht sehe, vor allem vermoge der durch

sie méglichen Dokumentierung.

Die Sprachlaute

verhallen, die Schriftzeichen beharren. Ich zahle

etwa, wihrend ich einer die Stunde schlagenden

Turmuhr lausche, die Schlige, indem ich sie durch Bleistiftstriche auf einem Blatt Papier vermerke. Die Objekte, wie hier die Schlage der Turmuhr, mégen

sich

einem

solchen

aufldsen,

melt,

thaw

and

resolve

themselves into a dew“, aber ihre Anzahl kann in Zahlzeichen

niedergelegt

und

auf-

bewahrt werden. Tue ich dies fiir zwei Reihen von Ténen, so kann ich mich der Zeichen zu der nach-

traglichen Feststellung bedienen, daft es ,,das zweite

Mal mehr Téne waren als das erste Mal“ — was mir beim direkten Anhéren vielleicht zweifelhaft geblieben ware. Als Zeichen dienen sichtbare Ge-

529 bilde von einer gewissen Bestindigkeit (nicht etwa Schalle und Rauchwolken; zum mindesten so lange standhaltend, als zur Ausfiihrung der an ihnen vor-

zunehmenden Operationen benétigt wird). Sie miis-

sen leicht und immer wieder herstellbar sein; ihre

Gestalt mu sich, wie Hilbert sagt, ,unabhingig

von Ort und Zeit und von den besonderen Bedin-

gungen der Herstellung des Zeichens

geringfiigigen

Unterschieden

sowie von

in der Ausfiihrung,

von uns allgemein und sicher wiedererkennen lassen“

(9, p. 163). Eine solche Beschreibung ist nicht gleichgiiltig, wenn einem (wie Hilbert) ,die Zeichen selbst die Gegenstinde der Zahlentheorie* sind. Der idealistischen Einstellung, fiir welche die Zahlen ideale Objekte sind oder aus einem Akt des reinen BewuBtseins entspringende Méglichkeiten, tritt hier eine ,anthropistische* Einstellung gegeniiber, die das konkrete Tun des Menschen ins Auge faft. Da

geht der Mathematiker nicht viel anders mit seinen aus Zeichen gebauten Formeln um wie der Tischler

in seiner Werkstatt mit Holz und Hobel, Sige und Leim. Mit einem Unterton von Ironie sagt Brouwer,

der Intuitionist, dariiber: ,Op de vraag, waar de

wiskundige exactheid dan wel bestaat, antwoorden

beide partijen verschillend; de intuitionist zegt: in het

menschelijk

papier.“ (4, p. 7.)

intellect,

de

formalist:

op

het

Wenn Newton die wahrnehmungsmafig erlebte Welt erklaren will durch die Bewegung fester Partikeln im Raum, so benutzt er den Raum,

der ihm

zugleich anschaulich gegeben und objektiv ist, zur Konstruktion der hinter den Erscheinungen verborgenen wirklichen Welt. Er verwirft, wie schon

Demokrit, die Sinnesqualitaten ob ihrer Subjekti tat als ungeignet zu ihrem Aufbau, aber behilt den Raum bei. Als Leibniz die Phanomenalitat von

Raum und Zeit erkannte, wurde man gezwungen,

auch diese zu climinieren. Gliicklicherweise stand das Mittel dazu in Descartes’ analytischer Geome-

trie bereit; denn diese lehrt, wie man (mit Bezug auf ein gegebenes Koordinatensystem) einen jeden

mehr, nicht einmal Raum

Geist

fiir seine

und Zeit, entnimmt der

symbolische

Reprasentation

der

Welt dem Gegebenen. Es ist wesentlich, da das Symbol als Symbol und nicht als Bestandstiidk der zu reprasentierenden Wirklichkeit verstanden wird. Huygens konnte noch mit gutem Gewissen sagen, da ein monchromatischer Lichtstrahl in Wirklicheit aus einer Oszillation des aus besonders feinen Partikeln bestehenden Lichtathers besteht. Wir reprasentieren den Strahl durch eine Formel, in der

ein gewisses Symbol F, elektromagnetische Feld-

starke genannt, als eine rein arithmetisch konstruierte Funktion von vier anderen Symbolen x, y, z, ¢, genannt Raum-Zeit-Koordinaten, ausgedriickt wird.

Das symbolische construct, das uns so in Handen

bleibt, kann niemand mehr im Ernst als eine den Erscheinungen

zugrunde

Anspruch nehmen.

liegende

Wirklichkeit

Natiirlich braucht darum

in

das

Band zwischen Symbol und wahrnehmungsmigig Gegebenem

nicht

durchschnitten

zu

werden;

der

Physiker versteht, wie der Symbolismus ,gemeint“

ist, wenn er die in ihm niedergelegten physikali-

schen Gesetze an der Erfahrung priift. (Vergl. hier-

zu 13.) Die hier kurz angedeutete Entwidclung der

Physik zu einer rein symbolischen Konstruktion gipfelt in unserem Jahrhundert in der Relativitatsund Quantentheorie. Die Weise, wie die Quantenphysik die beobachtbaren Groen durch Hermitesche Formen in einem unendlich-dimensionalen Hilbertschen Raum darstellt, ist ein besonders mar-

kantes Beispiel symbolischer Reprisentation.

Nach dieser allgemeinen Orientierung ist es nun an der Zeit, etwas konkreter auf die Konstruktionen der Mathematik und ihre Symbolik einzugehen (vergl. 1). Da finden wir zunachst Zeichen fiir einzelne Zablen, wie 2 oder , und Zeichen fiir einzelne bestimmte Operationen oder Relationen, wie + (plus), = (gleich), < (kleiner als). Damit lassen sich bereits bestimmte Aussagen in Formeln fassen,

wie 2 A, sondern auch die inverse A > Z stetig. Man beachte, dass

ein Dreieck auf der Flache nicht als Menge der zu ihm gehérigen Punkte definiert ist, sondern durch die Z auf A stetig abbildende Funktion A(&,, &, &s). Zur Triangulation einer Flache gehort, dass auf ihr gewisse Punkte e als Ecken ausgezeichnet sind, die in gewisser Weise zu Paaren (= Kanten) (ef) und zu Tripeln (== Dreiecken) (e/g) zusammengefasst sind. Zu einem Paar (ef) gibt es genau zwei Ecken g, so dass e/ mit g ein Tripel bilden. Zu einer Ecke e gibt

es nur endlichviele von ¢ verschiedene Ecken f = /,, ..., /, derart, dass (e/) als Kante auftritt. Diese bilden einen Zykel, indem e/; die gemeinsame Kante der beiden Dreiecke

Aia= (hah)

und

Ay= effin)

(@=1,..475 fo=tn fru =f)

ist. Zu jedem vorkommenden Tripel (efg) gehért eine stetige Funktion

P= AesalSe, §4 &4) » die jedem zulassigen, das heisst den Bedingungen

20,

B20,

60,

B28 e eget

gentigenden Wertetripel (&,, ¢,, €,) einen Punkt p auf der Fliche so zuordnet, dass verschiedenen Tripeln

verschiedene

Punkte

entsprechen.

Die Funktion

definiert das Dreieck A,,, der Triangulation. Die Bezeichnung ist so zu ver-

stehen, dass, wenn efg im Index des Funktionszeichens A und zugleich als In-

dizes der Variablen &,, &;, &, der gleichen Permutation unterworfen werden, der

Wert

der Funktion sich nicht andert. Auf den beiden Dreiecken AgolAi:

573 mit der gemeinsamen

Kante ef gelte Aesa(Ee gp 0) = Aes a (Fer &, 0) -

Damit ist also die die Kante (ef) definierende Funktion p = 6,,(¢,, &,) eindeutig erklart. Fiir jedes vorkommende Tripel (e/g) mit der Ecke e gelte A,,, (1, 0, 0) =e. Jeder Punkt der Flache erscheint als Wert mindestens einer dieser den Tripeln (efg) zugehérigen Funktionen A,;,. Die Topologie kommt in den folgenden Forderungen zum Ausdruck: Zu einem inneren Punkt eines Dreiecks

Abra

Whtge2 hes, Ws

Sy

gibt es eine Umgebung, deren Punkte keinem andern als diesem Dreieck ange-

h6ren; zu einem innern Punkt einer Kante 6,,: §, > 0, &, > 0, gibt es eine Um-

gebung, in der keine andern Punkte liegen als solche, die den beiden Dreiecken Acyq, Mey Mit der gemeinsamen Kante ef angehéren (Dreieckspaar) ; zu einem Eckpunkt e gibt es eine Umgebung, zu der keine andern Punkte als solche gehdren,

die in dem zyklischen

Stern der Dreiecke

Age Aer

a

= ere,

um ¢ liegen. Fiir die Folge nehmen wir an, dass die Flache § in einer bestimmten Triangulation ¢ vorliegt. Der Umstand, dass eine Flache zusammenhdngend ist, gibt sich jetzt in der kombinatorischen Eigenschaft kund, dass die Ecken sich in keiner Weise so in zwei Klassen teilen lassen, dass nur Paare und Tripel solcher Ecken vorkommen, die der gleichen Klasse angehéren. Eine Flache ist dann und nur dann geschlossen,

wenn

das

Schema

der Triangulation

nur aus endlichvielen

Ecken

(und darum auch aus nur endlichvielen Kanten und Dreiecken) besteht. Ein einheitlicher, auf der Fliche festgelegter Drehsinn gibt sich in jedem Dreieck A.7, der Triangulation dadurch kund, dass ihm eine bestimmte der beiden

Indikatrizen (= Ecken-Reihenfolgen) e/g = /ge = gef oder gfe = feg=egf zugewiesen ist; in solcher Weise, dass die Indikatrizen zweier in einer Kante zu-

sammenstossender Dreiecke (e/g) und (e/g’) koharent sind. Die Indikatrix efg eines Dreiecks induziert auf den begrenzenden Kanten je einen Durchlaufungssinn ef, fe, ge. Kohdarente Indikatrizen der in einer Kante ef zusammenstossenden

beiden Dreiecke induzieren auf der gemeinsamen Kante entgegengesetzte Durchlaufungssinne. Jedesmal ist hier eine Kontinuumseigenschaft (zusammenhangend, geschlossen, ergibt sich daraus,

orientierbar) in eine kombinatorische umgesetzt; und es dass die betreffende kombinatorische Eigenschaft von der

Triangulation unabhingig ist.

Aus der Beschreibung geht ferner hervor, dass das Innere des um eine Ecke e

sich gruppierenden Dreieckssterns Ay + 4; + --- + 4,4 sich topologisch auf das

574

Innere eines ebenen regularen Polygons so abbilden lasst, dass die von e ausgehenden, in den Dreiecken verlaufenden geradlinigen Strahlen in geradlinige Strahlen in der Ebene tibergehen («Strahlabbildung»). Da das Innere eines solchen Polygons einfach zusammenhingend, namlich dem Inneren eines Kreises topologisch aquivalent ist, gilt F(y) = 0 fiir eine jede Integralfunktion F und eine jede geschlossene Kurve y, die ganz im Innern des Dreieckssterns verlauft. Ein Kantenzug ey €, e) ... &, (der Lange 7) ist eine Folge von Kanten unserer Triangulation, in welcher der Endpunkt e, der (¢-ten) Kante e,_,e, zugleich der

Anfangspunkt der nachsten Kante e, e,,, ist. Der Kantenzug ist einfach, wenn alle seine Ecken éq, é;, ..., &, voneinander verschieden sind. Er ist geschlossen, wenn é,, mit é) zusammenfallt.

Statt der Kanten, welche die Ecken miteinander

verbinden, kann man auch die Cokanten (sit venia verbo!) x* betrachten, deren

jede die Mittelpunkte 7, 7’ der beiden Dreiecke A, A’ eines Dreieckpaars miteinander verbindet; x* besteht aus der Strecke 7c in A, die von z zu der Mitte c

der den beiden Dreiecken gemeinsamen Kante x lauft, und aus der Strecke ci’ in A’. Da wir annehmen, dass unsere Flache zweiseitig ist, konnen wir jedem

Dreieck der Triangulation eine positive Indikatrix so erteilen, dass die beiden Dreiecke eines Dreieckpaars stets kohdrente Indikatrizen bekommen. Eine Kante

x = ee’ sei mit dem

durch

die

Schreibweise

angedeuteten

Durchlau-

fungssinn versehen; von den beiden Dreiecken A, A’ mit der gemeinsamen Kante ce’ sei A dasjenige, dessen positive Indikatrix auf x den Durchlaufungs-

sinn ce’ induziert. Wir sagen dann, dass die oben beschriebene Cokante x*, von

i nach 2’ durchlaufen, Sinne) tiberkreuzt.

die Kante x von links nach rechts (oder im positiven Ansatz des Problems

Jetzt sind wir in der Lage, anzudeuten, wie wir das Problem der Charakteristik in Angriff nehmen wollen. Wenn f ein geschlossener Cokantenzug und « ein geschlossener Kantenzug ist, so ist der Begriff der algebraischen Summe der Uberkreuzungen von f iiber « vollkommen klar. Wir hoffen, dass folgende Tatsachen wahr sind: Jede geschlossene Kurve ist sowohl einem geschlossenen Kantenzug wie einem geschlossenen Cokantenzug homolog. Wenn immer a, a!

zwei zueinander homologe geschlossene Kantenziige und £, B’ zwei homologe

geschlossene Cokantenziige sind, besteht die Gleichung ch(B, «) = ch(B’, «’) .

Damit ist es dann méglich, ch(8, «) fiir irgend zwei geschlossene Wege «a, B

dadurch

zu definieren,

dass man

« durch

einen beliebigen zu % homologen

geschlossenen Kantenzug, 6 durch einen beliebigen zu B homologen geschlos-

senen Cokantenzug

ersetzt.

Wir hoffen

endlich,

dass die so allgemein

erklarte

575

Charakteristik das Gesetz der Antisymmetrie (1) erfiillt. Freilich haben wir uns den Nachweis dieses Faktums dadurch erschwert, dass wir die beiden Argu-

mente «, 8 ungleichartig behandeln, namlich fiir « Kantenziige, fiir 8 Cokantenziige einsetzen.

Der Poincarésche Formalismus. Man erteile willkiirlich jeder Kante x einen bestimmten «positiven» Durchlaufungssinn und der zugehdrigen Cokante x* dass sie x von links nach rechts tiberkreuzt.

Ein geschlossener

identifiziert werden, wenn sie gleiche Durchlaufungszahl, x,

— x, , fiir eine jede

einen solchen,

Kantenzug « durchlauft jede Kante x in summa eine bestimmte Anzahl von Malen x,, wobei eine positive Durchlaufung mit +1, eine negative mit —1 in Ansatz gebracht wird. Wir schreiben symbolisch « = 2 x, x und sprechen von a als einem Zykel in dem Sinne, dass zwei geschlossene Kantenziige « und «’ Kante x haben. Von den ganzen Zahlen x, verschwinden alle bis auf endlichviele. Sei x = ee’. Wir setzen fiir eine beliebige Ecke a: e(a x) = +1, wenn a=e',e(a x) = —1, wenn a =e, e(a x) = 0, wenn a verschieden von ¢ und e’ ist. Die Bedingung

der Geschlossenheit

dass unser Kantenzug

bedeutet,

« in

irgendeine Ecke a ebensooft ein- wie auslauft, und das driickt sich in der Gleichung aus: ela #)

>

(2)

= Ole

Ist A irgendein Dreieck, so setzen wir e(x4) = 0, ausser wenn A eines der

beiden Dreiecke A,, A, mit der gemeinsamen Kante ~ ist. Hingegen setzen wir

e(%A,) = +1, e(~Ay) = —1, wenn die positive Indikatrix von A, auf

den posi-

tiven, die von A, den negativen Durchlaufungssinn induziert. Es besteht fiir jede Ecke a und jedes Dreieck A der Triangulation die Gleichung

OG Ein

werden,

indem

positiven Sinne Gleichungen

kann

Cokantenzug

geschlossener man

angibt,

als ein

wie oft er, y,-mal,

durchlauft. Die

(3)

2) e(x A) = 0. Cozykel

Geschlossenheit

X y, x*

in summa

angesprochen

die Cokante

des Cozykels

x* im

findet in den

(4)

Dy ¥8(% A) = 0

jhren Ausdruck. Dies ist die Weise, wie H. Porncaré uns gelehrt hat, das kom-

binatorische Schema der Triangulation den Methoden der linearen Algebra zu unterwerfen. Schreibt man die Zahlen e(a x) als eine Matrix E, in welcher a der

Zeilen-, x der Spaltenindex ist, die Zahlen x, als eine Spalte x, die Zahlen y,

als eine Zeile y, die Gréssen (x4) endlich als eine Matrix E* mit x als Zeilen-

und A als Spaltenindex, so lauten die Gleichungen (2), (3) und (4) im Matrizen-

kalkiil beziehungsweise

Ex=0,

EE*=0,

yE*=0.

576

Der Cozykel B = ¥ y, x* tiberschneidet den Zykel « = 2 x, % so oft, wie die folgende Charakteristik angibt :

ch(B, a) =2'y, %,=y%Algebraisierung

der Integralfunktionen

Eine Integralfunktion F ordnet jeder Cokante %* einen bestimmten Wert

fy, = F(x*) zu. Da der geschlossene Cokantenzug, der die um eine Ecke a herumliegenden Elementardreiecke

Aba iiby cera tlleaty lll

edb)

von Mittelpunkt zu Mittelpunkt miteinander verbindet, im Innern des einfach zusammenhangenden

Zug,

Dreieckssterns verlauft, muss der Wert

das ist

von F fiir diesen

Sela x) f,=0

(5)

sein. Der Wert F(8) der Integralfunktion F fiir den Cozykel B = 2 y, x* ist 2 y, f,- Ich behaupte zweierlei: 1. Durch die Bestimmungszahlen f, ist die Integralfunktion im Sinne der Kohomologie eindeutig festgelegt. 2. Unter

der Voraussetzung,

dass

sie den

samtlichen,

den

Ecken

a entspre-

chenden Gleichungen (5) gentigen, kénnen diese Bestimmungszahlen beliebig vorgegeben werden. Daraus folgt, dass es zu jedem Zykel « = 2'x, x eine beschrankte Integralfunktion F, gibt (mit den Bestimmungszahlen /,, = x,), so dass ftir beliebige

geschlossene Cozykel

8 = Ly, x* die Gleichung

besteht. Da der Wert F,(8) sich nicht andert, wenn man § durch irgendeinen ihm schwach homologen geschlossenen Weg ersetzt, tibertragt sie die Definition der Charakteristik ch sofort von Cozykeln auf beliebige geschlossene Wege B und liefert das Gesetz

ch(B, «) = ch(p’, «)

fiir irgend zwei geschlossene Wege 8, A’, die einander schwach homolog sind;

insbesondere ch(f, «) = 0, wenn immer B~ 0. Jedoch bleibt das Argument «

einstweilen auf Zykeln beschrankt. Da die Gleichungen (5) ganzzahlige Koeffi-

zienten besitzen, ist jede beschrinkte Integralfunktion einer linearen Kombination endlichvieler beschrankter Integralfunktionen kohomolog, deren Bestimmungszahlen 7, ganze Zahlen sind. Darum gilt auch die umgekehrte Tatsache:

517

A. Hat ein geschlossener Weg B die Eigenschaft, dass ch(B, «) = 0 ist fiir jeden Zyhkel x, so ist B schwach homolog Null. Um die beiden oben erwahnten Tatsachen betreffend die Bestimmungszahlen einer Integralfunktion zu erweisen, haben wir an erster Stelle zu zeigen, dass jeder geschlossene Weg y einem Cozykel homolog ist. Da jeder Punkt der Flache im Innern mindestens eines Dreieckssterns liegt, folgt aus den Grund-

lagen der Analysis, dass sich y in endlichviele Teilbégen Yo Yeon

nar =)

zerlegen lasst, deren jeder ganz im Innern eines Dreieckssterns verlauft. Aufeinanderfolgende Bégen, die im gleichen Stern liegen, vereinigen wir zu einem einzigen Teilbogen. Dann hat der Dreiecksstern 2}, innerhalb dessen y, liegt, ein Dreieckspaar mit demjenigen Stern gemein, in welchem y, liegt. Der Endpunkt a, von 7, welcher zugleich der Anfangspunkt von 7, ist, liegt in einem, A, der beiden Dreiecke dieses Paars

(liegt er in beiden, so nehmen

wir fiir A

eines davon). Wir hangen dann an y, die in A verlaufende Strecke an, die von a, nach dem Mittelpunkt 7 von A fiihrt, und bringen dieselbe im umgekehrten Sinne durchlaufene

Strecke vor y, an. Sei F eine Integralfunktion. Durch das

geschilderte Verfahren verwandeln sich die Bégen 7, 72, ... durch Hinzufiigung

je einer Strecke am Anfang und am Ende in Bégen yy, 7, ..., die je innerhalb eines Sterns vom Mittelpunkt eines Dreiecks zum Mittelpunkt eines andern fiihren. Dabei bleibt die Summe

YN ES TO ungedndert und damit auch der Wert von F(y), da ja die zwischengeschalteten

Strecken, zweimal in entgegengesetztem Sinne durchlaufen, sich aufheben. Der

im Stern Y, vom Mittelpunkt des Dreiecks A zum Mittelpunkt des Dreiecks A’ fiihrende Weg y; kann durch einen dieselben beiden Punkte verbindenden, im Dreiecksstern im einen oder andern Sinne herumlaufenden Cokantenzug yj ersetzt werden, und wegen des einfachen Zusammenhangs des Sterns ist dabei F (yi) = F(y{). So erhalten wir schliesslich einen geschlossenen Cokantenzug y!=yl + yh + ++, der homolog y ist; denn er geniigt fiir jede Integralfunktion F der Gleichung F(y) = F(y”). Ebenso ergibt sich, dass jeder geschlossene

Weg homolog einem geschlossenen Kantenzug ist. Die zweite Tatsache, dass die Zahlen /, unter Einhaltung der samtlichen Be-

dingungen (5) beliebig vorgegeben werden kénnen, ergibt sich etwa so. Fir die

y in a miindenden Kanten »,..., x, setzen wir f; = e(a x;) f,,. Dann ist die Summe f, + ++ + f, = 0, und es lassen sich also Zahlen g5, ---, 8-1 (8+ = 80) SO finden, dass f; = g; — g;. _Im positiv umlaufenen Zykel der Dreiecke mit der Ecke a trennt x, das Dreieck A, von A,. Wir ordnen dann den inneren Punkten p von A; den Funktionswert g(f) = g; zu, den inneren Punkten p der

578 Kante

x; den Wert

ep)

= 2

(erate 82)

schliesslich dem Zentrum a den Wert

era)

(Sos

Fiir eine Kurve y, die innerhalb dieses Sterns vom Punkte p zum Punkte lauft, setze man



Fy) = g(b') — g() -

In die Definition von g(p) geht eine willkiirliche additive Konstante ein, die aber bei der Bestimmung von F(y) wieder herausfallt. Liegt der Kurvenbogen y in zwei Dreieckssternen,

die dann ein Dreieckspaar gemein

haben,

so ist der

resultierende Wert F(y) auch unabhangig davon, welchen dieser beiden Sterne man der Berechnung zugrunde legt. Ist schliesslich y irgendeine Kurve, ge-

schlossen oder nicht, so teile man sie in endlichviele Bogen y; (i = 1, ..., ), deren jeder ganz in einem Dreiecksstern verlauft, und bilde dann

Ey Ge

Vado

Dieser Wert ist von der benutzten Teilung unabhangig, wie man sofort sieht, wenn man zwei Einteilungen in Bégen iiberlagert. Auf diese Weise erhalt man eine Integralfunktion F mit den Bestimmungszahlen /,.

Berechnung

des

Zusammenhangsgrades

einer

geschlossenen

Flache

F ist dann und nur dann homolog Null, wenn sich jedem Elementardreieck

A eine Zahl g, so zuordnen lasst, dass allgemein fiir die von A nach J’ fiihrende Cokante x* der Wert F(x*) = f, gleich gy, — g4 wird; oder

f=

—Xy e(% A) 84.

(6)

Im Falle einer geschlossenen Flache seien die Anzahlen der Ecken, Kanten und Dreiecke e, f und d. Alsdann hat man f Unbekannte f,,, denen die e Gleichungen

(5) auferlegt sind. Da zwischen ihren linken Seiten nur eine Identitat besteht,

namlich diejenige mit den Koeffizienten A, = 1, so ist die Anzahl der linear unabhangigen Lésungen f — e + 1. Auf der andern Seite ist die Anzahl der linear

unabhangigen Lésungen, die aus (6) mittels beliebiger Zahlen g, hervorgehen,

gleich ) — 1. Der Uberschuss (=e

l=

)=W=t eas

2

579 ist darum die Maximalzahl h der linear unabhangigen Integralfunktionen, und daraus folgt, dass diese «Eulersche Charakteristik» — — e — d + 2 von der Triangulation der Flache unabhangig ist. Ein Satz iiber primitive geschlossene Wege

Fiir geschlossene Flachen kann der Satz A in bemerkenswerter Weise verscharft werden. Wir nennen einen geschlossenen Weg f primitiv, wenn er nicht dem Vielfachen » y (n = 2 oder 3 oder 4 oder ...) eines geschlossenen Weges y homolog ist. Ich behaupte: A’. Zu einem primitiven Weg B gibt es einen Zykel a, fiir den ch(B, «) = 1 ist (der also von f in summa genau einmal iiberkreuzt wird). Beweis. Zykeln « und Cozykeln B werden durch ihre f Durchlaufungszahlen x, bzw. y, gekennzeichnet; die x, werden als Spalte x, die y, als Zeile y geschrieben und x und y als Vektoren in zueinander dualen Vektorraéumen angesehen, fiir welche das innere Produkt

2g Xe = YH eine invariante Bedeutung hat. Die Vektoren mit ganzzahligen Koordinaten

sollen Gittervektoren heissen. Werden die Koordinaten x, einer beliebigen uni-

modularen Transformation unterworfen (das ist einer linearen Transformation mit ganzzahligen Koeffizienten, deren Determinante = + 1 ist), so bleibt die Kennzeichnung der Gittervektoren als der Vektoren mit ganzzahligen Koordinaten erhalten. Werden die Koordinaten y, des Vektors y der kontragredienten linearen Transformation unterworfen, so bleibt auch der Ausdruck des inneren

Produkts der gleiche:

Via cea eer Ea

Man kann nach dem Verfahren, das von M1nKowsktals Adaptation eines Gitters

an ein

enthaltenes

bezeichnet

wurde,

eine

unimodulare

Transformation

im

x-Raum so ausfiihren, dass die den Gleichungen (2) geniigenden «geschlossenen » Vektoren (4,,..., x1) durch das Verschwinden der f —/ letzten Koordinaten gekennzeichnet sind. Mit andern

Worten:

Wir bekommen

/ Zykel a, ..., «, So,

dass jeder Zykel sich in einer und nur einer Weise als eine Summe ganzzahligen

Koeffizienten

x; schreiben lasst. Man

u

Dy %_ %,; mit

t=1

fiihre die kontragrediente

Transformation im dualen y-Raum aus. Wegen der Gleichung E E* = 0 fiir die Matrizen E und E* verschwinden nach dieser Transformation in jeder

Spalte von E* die letzten £ — 1 Glieder, das heisst die letzten k — 1 Zeilen von

E* sind mit Nullen besetzt. Die die Geschlossenheit eines y-Vektors zum Ausdruck bringenden Gleichungen (4) im dualen Raum betreffen demnach nur die Koordinaten ¥,,..., y,; und nach einer geeigneten unimodularen Transformation dieser 1 Koordinaten besagen jene Gleichungen, dass die letzten / — h der

580

Koordinaten 4, ..., ¥, verschwinden. Nachdem %,, ..., ¥; der kontragredienten Transformation unterworfen sind, nehmen die geschlossenen Gittervektoren im x-Raum die Form (Phypeoas dis Sivep conn tips can Ul]

an, die im y-Raum

die Form

(Ya 209 Yar Oy 007 0 Yarns +09 Mt) + und die Charakteristik wird

Mt

ch(B, «) = y X= Der Cozykel

£ wird dann und nur dann

druck fiir beliebige ganzzahlige Werte wenn 4, =--:= y, = 0 ist. Wir haben dass jeder geschlossene Weg f homolog linearen Kombination y, 8, + --- + y,

homolog

(7)

+ Yn tne Null sein, wenn

dieser Aus-

von xj, ..., x, verschwindet, das heisst, also h Cozykel ,,..., 8, gefunden, so einer eindeutig bestimmten ganzzahligen By derselben ist, und damit erweist sich

h als der Zusammenhangsgrad. f ist primitiv, falls y,,..., y, den gréssten gemeinsamen Teiler 1 besitzen. Dann kann man aber ganze Zahlen 4, ..., x, und

h

damit einen Zykel 3’ x, «, finden, fiir welche (7) gleich 1 wird, und damit ist t=1

unser Satz bewiesen.

Das

Gesetz der Antisymmetrie

Wir haben die Regel angegeben, nach welcher der Wert a= 3) x, x gehérigen Integralfunktion

der zum

Zykel

E(B) = chi(B, «) fiir einen beliebigen Weg f zu berechnen ist. Hier spezialisieren wir nun zu-

nachst auch £ als einen Zykel 2’ y, x. Eine Kante x mit den beiden Endpunkten e, e’ zerlegen wir durch ihre Mitte c in die zwei Halbkanten

(x e) = ce und

(x e’) = ce’, so dass

x= Die + in ¢ einmiindenden Kanten,

J

@

ee x) (xe). wie die Halbkanten

¢,é, numerieren

wir in

zyklischer Reihenfolge, e im positiven Sinne umlaufend, mit i =1,...,r. Es ist zweckmiassig, den Index 7 als einen solchen zu verstehen, der alle ganzen Zahlen

durchlauft,

aber so, dass Werte 7, die einander kongruent

modulo ¢ sind, als

gleich gelten. Die den in e einlaufenden Kanten x, zugehérigen Zahlen x, genauer also die Zahlen e(e x,) %,, wollen wir, wie schon friiher, mit *,,..., x,

bezeichnen. Ihre Summe ist 0, und wir kénnen darum Zahlen g; so bestimmen,

581 dass x; = g, — g,_;. Der Beitrag zu F,(B) = ch( «), der von den in e miindenden Halbkanten (der «Spinne in e») herriihrt, ist dann

w(B a) = DS) ¥: F,(c;2) Hier ist fs

A

Fle, é) = — z i+

ga) +8,

wo g der Mittelwert der v Zahlen gp, ..., g,_, ist. Da auch die Summe ist, kénnen wir fiir diesen Beitrag schreiben: 7

al

oy Ye (e+

2 y, = 0

81-2) -

Eben wegen der Gleichung 2’ y, = 0 kann die in die g; eingehende willkiirliche additive Konstante irgendwie gewahlt werden, etwa gemiss gy = 0. Dann ist

1

elds

=

q>t

ar

1

ah

= Py

1

i DV

m=



aS

Gare

Pare

(8)

Der von der Spinne in ¢ herriihrende Beitrag zur Charakteristik ist also eine halbganze Zahl. Wenn wir in der zweiten Formel die Indizes i und 7 vertauschen, so folgt ohne weiteres das fiir die Verkehrsregelung so wichtige Gesetz der Antisymmetrie

Setze einen Augenblick

w,(B a) = —w(a B) -

Lr=2yin

(mt+lsjcism+n.

Aus unserer Berechnung geht hervor, dass Ly unabhingig davon ist, wo wir im

Zykel der Kanten,

die in ¢ hineinlaufen, mit der Numerierung

durch die Zif-

fern 1 bis 7 beginnen. Dies ist sofort zu verifizieren. Denn L,, entsteht aus Ly,_4

dadurch, dass man

fortlasst und dafiir

(Yr + +11 + Ymtr—a) Xm Ver Fmia

+ °° + Xm4r)

hinzufiigt, diese Terme sind aber beide gleich —y,, ¥,. Darumist Ly = LS.

582

Man mache sich klar, dass unsere Formel fiir die Schnittpunktmultiplizitat

w, der beiden Zykeln f und « in e dem entspricht, was man anschaulich erwartet. Ein der Fahrbahn « folgendes Auto fahre langs der Strasse x, in die Kreuzung e hinein und verlasse sie langs x, ; ein zweites Auto 8 komme zur gleichen Zeit langs der Strasse x; auf die Kreuzung zu und verlasse sie langs xj. Dann wird man erwarten, dass die Schnittmultiplizitat 0 ist, falls in der zyklischen Anordnung der in e einmtindenden Strassen das Paar (x;, xy) das Paar (x;, 3) nicht trennt; man wird die Multiplizitat -- 1 erwarten, falls die beiden Paare

sich trennen, und zwar sollte 1 herauskommen, wenn f den Weg « in e von links nach rechts iiberkreuzt, —1 im entgegengesetzten Falle. Es kann aber auch geschehen, dass die beiden Autos die Kreuzung langs derselben Strasse

verlassen. Dann ergibt unsere Formel einen Wert + 1/2. In der Tat bleibt in diesem

die andere kreuzt, in e noch

ob die eine Bahn

Falle die Entscheidung,

suspendiert. Erst wenn die Wege sich spater in einer andern Kreuzung trennen,

wird durch deren Beitrag ++ 1/2 entschieden werden, ob Kreuzung stattgefun-

den hat (falls namlich die Summe der beiden Beitrage nicht 0 ist), und wenn ja, in welchem Sinne. Unsere Formel umfasst in praziser algebraischer Form diese und alle andern Méglichkeiten.

Natiirlich muss die tiber die Ecken e erstreckte Summe der Beitrage w,(B «)

eine ganze Zahl sein. Auch dies ist sofort zu verifizieren.

halbganze Anteil von w,, (1/2)

X ¥; x;, ist ja gleich

1

D D»

Der méglicherweise

ee 2) y, (ex) %,,

und darum ist die iiber alle Ecken e erstreckte Summe dieser Anteile

% dy Hex) y,,%, ex

wegen

Dd eex)—2

gleich



Das fiir Zykeln «, 8 nunmehr erwiesene

Sy, x, .

Gesetz der Antisymmetrie

(1) er-

moglicht es uns, in ch(8, «) die Beschrankung von «, und nicht bloss von £, auf

geschlossene Kantenziige aufzuheben. Seien namlich «, a, zwei geschlossene Kantenztige, welche dem geschlossenen Weg « schwach homolog sind, und f, ein Zykel, der schwach homolog dem geschlossenen Weg ist. Dann gilt a, ~ a und

darum

ch(B,, #4) = —ch(a,, B,) = —ch(a,

B,) = ch(A,, oi)

folglich ch(8, 0) = ch(B, a1). Man kann demnach

ch(f, «) eindeutig dadurch

definieren, dass man den geschlossenen Weg « durch irgendeinen ihm schwach

homologen Zykel (a oder o) ersetzt. Sind a, 8, zwei Zykel, die beziehungsweise

den

geschlossenen

ch(B, «) = ch(By, 0). Das

Wegen

«, 8 schwach

homolog

Gesetz der Antisymmetrie

sind,

so gilt

dann

bleibt fiir beliebige «, B

583 erhalten. Das Ziel einer universellen Definition der Charakteristik ist damit erreicht.

Immerhin

Triangulation

ist zu beachten,

dass diese

Definition

€ der Fliche § stiitzt. Soll dies in Evidenz

schreibe man ch; statt ch.

sich auf eine feste

gesetzt werden,

so

Ist die Flache geschlossen und h ihr Zusammenhangsgrad, so kénnen irgend h voneinander linear unabhingige geschlossene Wege 7, ..., 7, als eine

Wegebasis benutzt werden, indem jeder geschlossene Weg « homolog einer linearen Kombination x,y, + --- + %, 7, ist. Fiir dieses « und B~ y,y,+--+ Va Vn ergibt sich I

ch(B, 0) = D7 sis Ve Xs,

Ses = chlyes V5) -

jal

Der Satz A besagt, dass diese schiefsymmetrische bilineare «Charakteristikenform» nicht ausgeartet ist und dass somit ihre Determinante

verschwindet. Dies kann aber fiir eine schiefsymmetrische stattfinden, wenn die Dimension h gerade ist. Darum ist der grad h einer geschlossenen zweiseitigen Fliche stets eine gerade Zahl h/2 heisst Geschlecht. Die Zahlen x, sind nicht notwendig

aus den Gleichungen

d = dets;; nicht

Form nur dann ZusammenhangsZahl; die ganze ganz, doch folgt

h

ch(ys, 0) = DF sis %5, i=

1

dass sie ganze Zahlen dividiert durch d sind. Durch den scharferen Satz A’ und seinen Beweis hat sich gezeigt, 1. dass die geschlossenen Wege eine Integritdtsbasis y,, ..., Yn besitzen, so dass in der Homologie

am

Hy Yi

+ Man

fiir jeden geschlossenen Weg « die Koeffizienten x, ganze Zahlen werden, und 2. dass bei Zugrundelegung einer solchen Basis d = 1 wird. Aus dem letzten Umstand

folgt, dass man die Integritatsbasis so waihlen kann, dass die Charak-

teristikenform die kanonische Gestalt

(9 Xe — Yo 41) +++ + (Yaa Xn — Yn ¥n-1)

annimmt (RIEMANNS «kanonische Riickkehrschnittpaare»). Wir haben jetzt alle wesentlichen Eigenschaften der Charakteristik beisammen und méglichst einfach kombinatorisch abgeleitet. Um aber nun festzu-

stellen, dass die Charakteristik in Wahrheit von der Triangulation, auf die sich

ihre Definition stiitzte, unabhingig ist, benutze ich eine ganz andere Definition derselben, die fiir die funktionentheoretischen Anwendungen viel geeigneter ist und mir durch diese Anwendungen nahegelegt wurde. Zur Vorbereitung stellen

584

wir zunichst fest, dass die Charakteristik sich nicht andert, wenn die Triangulation ¢ in gewisser elementarer Weise einer Unterteilung unterworfen wird. Unterteilungen

Sind a; = (%;1, &jg, %3) drei (nicht in einer Geraden liegende) Punkte (i =1,

2, 3) im Zahlendreieck Z, mit den Ecken

b= Sam q

m 29, ein Zahlendreieck

Ae 20,

Hs 20,

als das Teildreieck

Z,

1, 2, 3, so wird durch die Substitution

Gi=12,3) 1 +

Hot

Hge=1,

von Z; mit den Ecken

aj, dg, a3 fest-

gelegt. Eine Triangulation ¢’ ist eine Unterteilung der Triangulation ¢, wenn die Elementardreiecke von ¢’ Teile der Elementardreiecke von € sind. Wir wollen

hier aber nur zwei ganz spezielle Typen von Unterteilungen betrachten. Ist a

ein Punkt von Zz, so zerlegen wir Z; in die drei Dreiecke 23a, 31a, 12a.

nicht im Innern von Z,, sondern etwa auf der Kante 23 gelegen,

das erste Dreieck,

23a, fort. Ist a ein innerer Punkt

Ist a

so fallt hier

des Elementardreiecks

4

der Triangulation ¢, so fiihren wir die angegebene Teilung nur in diesem Drei-

eck aus. Ist aber a ein Kantenpunkt, so muss sie in den beiden an diese Kante

anstossenden Dreiecken bewerkstelligt werden.

Art» wird dazu benutzt,

um

einen Punkt

Diese «Elementarteilung erster

a, der noch nicht

Eckpunkt

ist, zu

einem Eckpunkt der Triangulation zu machen. Die «Elementarteilung zweiter Art» ist die Normalteilung, durch welche Z; in die vier Teildreiecke (23c,), (31 cy), (12c3), (Cy, Cg C3) zerlegt wird, wo

i

ol

1

am (pg). a= (F.%7). a= (FZ 9-9)Auf der Flache wird diese Normalteilung in allen Elementardreiecken A der gegebenen Triangulation ¢ gleichzeitig ausgefithrt. Durch Iteration dieses Prozesses der Normalteilung erhalt man Triangulationen, die schliesslich «beliebig

fein» werden. Geht die Triangulation ¢’ aus ¢ durch einen der beiden Elementarprozesse hervor, so ist ein ¢-Kantenzug zugleich ein ¢’-Kantenzug, und aus der Formel (8) fiir den Beitrag w, geht ohne weiteres hervor, dass, wenn «, B irgend zwei

geschlossene ¢-Kantenziige sind, die Gleichung

chy, (B, «) = ch,(B, «)

(9)

besteht. In der Tat, wird das Strassennetz einer Stadt durch neu angelegte

Strassen erweitert, so kann man diese sowie die etwa dadurch neu entstehenden

Kreuzungen

ignorieren,

solunge niemand

die neuen

Strassen

benutzt.

Die

Glei-

585 chung (9) tibertragt sich dann aber durch Homologie auf beliebige geschlossene

Wege «, 8; und in derselben Allgemeinheit gilt sogar die Regel

(10)

chyx(B, a) = ch-(B, a)

fiir eine Triangulation ¢*, die durch wiederholte elementare Unterteilungen aus & entsteht.

Die « Kontinuumsdefinition» der Charakteristik

zu welcher wir jetzt tibergehen, benutzt keine Triangulation, sondern die Bedeckung einer Flache mit Umgebungen und die topologische Abbildung der Umgebungen

auf das Innere von Kreisen. Es sei « eine gegebene geschlossene

Kurve auf §. Durch endlichviele Teilpunkte OI

ee Te

PSO)

werde sie so in Bogen Ce

Ulsm

Golo)

hes,

geteilt, dass der Bogen «, in der Umgebung

nett

0

U, eines Flachenpunktes liegt, und

diese Umgebung sei durch die topologische Abbildung A; auf das Innere EF des Einheitskreises der (x, y)-Ebene bezogen. Wir nehmen zunachst 7 = 1. Es sei

K das Innere eines zu E konzentrischen Kreises von einem Radius g, der kleiner als 1, aber

doch

so gross ist, dass die beiden

Punkte

0, 7, oder vielmehr

ihre

durch die Abbildung 4 = A, erzeugten Bilder noch in K liegen. Wir sprechen von K als der «geschrumpften Umgebung U = Up. Fiir einen von 0 und 7 verschiedenen Punkt p in E sei 2 (pf) der Winkel, um den man den Strahl £0 im positiven Sinne drehen muss, damit er in die Lage pl kommt. (p ist gleich — Mp, wenn 22 Gp, 27 g, die Winkel bedeuten, welche die Strahlen p0, pl mit der positiven x-Achse bilden.) y(p) ist nur mod 1 bestimmt. Doch auf der Peripherie k von K ist y() eine eindeutige stetige Funktion (deren Werte absolut kleiner als 1/2 sind). Diese kann man leicht als eine eindeutige stetige Funktion «(f) auf die ganze gelochte Flache § — K so ausdehnen, dass sie ausserhalb einer kompakten Teilmenge von § verschwindet (zum Beispiel ausserhalb einer konzentrischen Kreisscheibe K,, deren Radius > g, aber