C. Cattaneo ( E d.

Relativistic Fluid Dynamics

Lectures given at a Summer School of the
Centro Internazionale Matematico Estivo (C.I.M.E.),
held in Bressanone (Bolzano), Italy,
June 7-16, 1970
C.I.M.E. Foundation
c/o Dipartimento di Matematica “U. Dini”
Viale Morgagni n. 67/a
50134 Firenze
Italy
cime@math.unifi.it

ISBN 978-3-642-11097-9 e-ISBN: 978-3-642-11099-3

DOI:10.1007/978-3-642-11099-3
Springer Heidelberg Dordrecht London New York

©Springer-Verlag Berlin Heidelberg 2011

Reprint of the 1st ed. C.I.M.E., Ed. Cremonese, Roma 1971
With kind permission of C.I.M.E.

Printed on acid-free paper

Springer.com
 CENTRO INTERNAZIONALE MATEMATICO ESTlVO
(C. I. M. E )

1 Ciclo - Bressanone dal 7 al 16 Giugno 1970

"RELATIVISTIC FLUID DYNAMIC'S"

Coordinatore: Pro, C CATTANEO

PHAM MAU QUAN Problems mathematiques en hydro-

dynamique relativiste. Pag.
A. LICHNEROWICZ Ondes des choc, ondes infinitesima-
les et rayons en hydrodynamique et
magnetohydrodynamique relati vi stes. II 87
A. H. TAUB Variational principles in general re-
lativity. " 205
J. EHLERS General relativistic kinetic theory
of gases " 30 I
K B. MARATI-IE Abstract minkowski spaces as fibre
bundles. " 389
G. BOILLAT Sur la propagation de la chaleur en
I'elativite II 405
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(C. I ME.)

PROBLEMS MATHEMATIQUES
EN
HYDRODYNAMIQUE RELATIVISTE

PHAM MAU QUAN

Corso tenuto a Bressanone dal 7 al 16 Giugno 1970

 Chapitre 1
LES SCHEMAS FLUIDES EN HYDRODYNAMIQUE RELATIVISTE

;1. GENERALITES SUR LA

DYNAMIQUE RELATIVISTE DES FLUIDES

1. Le cadre geometrique.
La mecanique relativiste des fluides a pour cadre geometrique l'espace-
temps qui est une variete differentiable V de dimension 4, de classe C"',
sur larquelle est donnee une structure pseudo-riemannienne g de signature
+~--. La geometrie de l'espace-temps (V, g) est celle de la connexion rie-
mannienne canoniquement associee a g.
La metrique definie par g est dite de type hyperbolique normal. Elle in-
duit sur l'espace vectoriel tangent Tx (V) en chaque point x de V une
structure d'espace-temps plat de Minkowski. En coordon~es locales (xd.)
on a

(1. 1) («, ~ = 0, 1, 2, 3).

Le tenseur g,,~ dit tenseur fondamental de gravitation est assujetti a veri

fier un systeme d'equations aux derivees partielles du second ordre qui gen~

ralise Ies equations de Laplace.-Poisson et qui donne naissance aux condi -

tions de conservation. Ces equations sont Ies dix equations d'Einstein

(1.2)

ou St(~ ne depend que de la structure riemannienne g de I'espace-temps,

TC(, est de signification purement mecanique et X un facteur constant.
 -4-
Pham

Le tenseur TIf dit tenseur d'impulsion-energie du fluide doit decrire

au mieux la distribution energetique dans l'espace-temps. Le tenseur S.,
est astreint aux deux conditions suivantes:
1. S., ne dependent que de g.~, de leurs derivees des deux pre -
miers ordres, sont lineaires par rapport aux cterivees du second ordre
2. SClit est conservatif, c 'est-a.-dire tel que

(1.3)

On demontre (~) qu 'on a necessairement

Oll Rca, est la courbure de Ricci, R la courbure scalaire de (V, gl, h et k

deux constantes arbitraires. k est la constante cosmologique, ne joue pas de
rOle dans la description des fluides. On peut supposer k = O. En supprimant
d'autre part Ie facteur surabondant h, on prendra pour premier membre des
equations d 'Einstein

(1. 4)

S., sera dit tenseur d 'Einstein.

Le tenseur d'Einstein Sd, etant conservatif, il en est de m~me du tenseur
d'impulsion-energie Til, . Les equations

(4) E. CARTAN. - J. Math. pures et appliquees, 1. p. 141-203. (1922)

 - 5-

Pharn

expriment alors la conservation de l'impulsion-energie et definiront l'evo-

lution du fluide.

2. Le tenseur d'impulsion-energie.
Dans toute theorie relativiste des fluides, Ie premier pas consiste a
choisir l'expression du tenseur d'impulsion-energie T.~ . Chaque expres-
sion de T.,~ definit un schema de fluide. Til, doit @tre symetrique si lIon
veut satisfaire aux equations d'Einstein. Mais pour que T.c, puisse decri-
re un fluide physique, il faut qu'il existe un champ de vecteurs unitaires u·
orhmtes dans Ie temps

(2.1)

pour lequelle scalaire T., if'u' soit positif. u· est dit vecteur vitesse un.!.
taire du fluide et ses trajectoires definissent les lignes de courant.
En fait les fluides reels sont doues de proprietes diverses. Les forces
de liaisons internes qui jouent un rOle fondamental dans I 'etude dynamique
se traduisent par Ie tenseur des pressions propres. Les phenomenes calo-
rifiques introduisent un scalaire 9 dit champ de temperature propre. Les
proprietes electromagnetiques sont susceptihles d'@tre representees par
deux champs de tenseurs antisymetriques H., ,G., comme on Ie verra.
n convient d'autre part d'etudier l'evolution thermodynamique du fluide.
Ces diverses proprietes peuvent @tre envisagees dans une decomposition
geometrique du tenseur d'impulsion-energie.
On est aussi conduit it mettre T.c, sous la forme

(2.2)
 - 6-

Pham

ou f est un scalaire positif representant la densite propre de matiere-

energie ponderable, 1(.~ les pressions propres, Q.~ les echllIlges the~

miques par conduction et f ll, Ie tenseur d'energie electromagnetique. Si

lIon neglige certaines proprietes, les termes correspondants ne figurent
pas dans la decomposition. De m(!me on peut introduire de nouveaux ter-
mes pour etudier de nouvelles proprietes.
A chaque expression de ToI~' correspond alors un schema fluide.
Dans chaque cas, l'evolution du fluide sera defini par les equations de
conservation (1. 5) qui, tenant compte du caractere unitaire de uGl, con
duisent aux equations sliivantes

(2. 3)

(2.4)

(2.3) est dite l'equationaie continuite et (2.4) constitue Ie systeme diffe-

rentiel aux lignes de courant.
A ces equations on adjoindra eventuellement d'autres equations telles que
les equations thermodynamiques, les equations du champ electromagnetique.
On obtiendra de cette maniere Ie systeme fondamental des equations du sche-
ma considere. Ainsi Ie schema fluide pur a fait l'objet de nombreuses etudes
divenues c1assiques, en particulier celles de L. P. Eisenhart et de A. Lichne-
rowicz. Le schema fluide thermodynamique a He Hudiee par C. Eckart et
par l'auteur dans sa these en 1954. Le schema fluide champ electromagneti-
que a fait l'objet des travaux de A. Lichnerowicz, de ceux de l'auteur datant
de 1955 qui ont suscite depuis de nombreux travaux, notamment ceux de G.
Pichon.. n a conduit dans un cas particulier a la magnetohydrodynamique r~
 - 7-

Pham

lativiste dont l'etude a fait l'objet de tres beaux travaux de Y. Choquet-

Bruhat, A. Lichnerowicz.
C 'est 1'etude mathematique de quelques uns de ces schemas qui consti
tue Ie sujet de ces conferences.

3. Repere propre.
On appelle repere propre en un point x de I 'espace-temps (V 4' g) un r~

pere orthonorme (V~ ) dont Ie premier vecteur V.' coincide avec Ie vecteur
vitesse unitaire u et dont les trois autres vecteurs Vi! definissent I 'espa-
ce associe a la direction de temps u.
On peut rapporter I 'espace-temps dans Ie voisinage de tout point a un
champ de reperes propres qu 'on supposera differentiable (mais non neces-
sairement integrable). La metrique d 'univers prend alors la forme canonique

r !'
(3.1) _(I) sea)

~I
Oll les u>sont les I-formes duales des champs de vecteurs VJ: i. e. tel-
~',: ~
les que W , V"., >= $ /,-', \,.', etant Ie symbole de Kronecker egal a
<
'It'
.t
I ,
si r
=p.' et 0 si 4 p.'. Les lA.I constituent donc quatre formes de
Pfaff lineairement independantes.
La consideration du repere propre est fort utile. En effet l'espace vecto-
riel tangent T (V) a une structure d'espace-temps de Minkowski Ie repere
x
propre V~ doit Hre identifie a un repere galileenlocal ou Ie fluide a une vi-
tesse nulle. Si on connait les composantes d 'un tenseur t relativement au
repere propre, ses composantes dans un repere quelconque (eod se dedui-
sent des premieres par des formules de transformation connues.
En effet si (A~.) est la matrice de passage du repere (e,) au repere
 - 8 -

Pham

propre (Vl.') et (It'i) la matrice inverse, on a

CI II 01
(3. 2) A" " u A·,~ = V(~r

0' - (.')
(3.2') "
Ad. = utI A~
" - - Vel

Si t est un tenseur d'ordre 2, ses composantes t., dans le repere (eat)

se deduisent de ses composantes t,:"., dans le repere propre par les form~

les

(3. 3)

On a en particulier pour le tenseur metrique

Ainsi pour determiner l'expression du tenseur d'impulsion-energie d'un

fluide pur, on le rapporte d'abord au rep ere propre. Le fluide y est caract~

rise par sa densite propre de matiere-energie " son tenseur des pressions
partielles ff.'j' . Son tenseur d'impulsion-energie a pour composantes dans
le repere propre

T,',' =, T·",
,J 11'""
~J

Rapportons maintenant 1'espace-temps a des coordonnees locales xGl, on a

~
Q/~" ArJ. dx lil et l'application des formules (3.3) donne
I
 - 9-

Pham

0.5)

satisfait aux identites

(3. 6)

On voit que dans Ie cas d'un fluide pur, Ie tenseur d'impulsion-energie se

decompose relativement a uti. une composante tempGrelle , u. u. et en
une compos ante spatiale n.,.
Definition - On dit que Ie fluide est parfait si la quadrique des pressions
dans Ie repere propre est une sphere i. e. si 1l't'jl = p ~~, j' , pest dite
pression scalaire du fluide.
'1 ..
Pour un fluide parfait, on a 1I'aI. = p .; A~ A', , soit en tenant compte
de (3.2) et (3.4) 1l'G1,= P (gil, - uCil u, ). Ainsi Ie tenseur d'impuision-ener-
gie d 'un fluide parfait est donne par

(3.7)

Appelons repere principal en x Ie repere orthonorme (Wa,) dont chaque

vecteurs W" est vectenr propre de la matrice (R•• ) par rapport a la ma-
trice (gil,). Les directions definies par Wl ne sont autres que Ies direc -
tions principales de Ricci. Or en vertu des equations d'Einstein, W~ sont
aussi vecteurs propres de la matrice (T",) relativement ala matrice (gil,).
On peut exprimer les composantes de Tel, a partir des valeurs propres
et vecteur propres comme

S
(3.8) TU' = s, W(o)tJ, W(o..,- f.i s~ W(~ ,.W(~)~
 - 10 -

Pham

On voit qu' en general Ie repere propre d 'un fluide thermodynamique cha~

ge est different du repere principal, sauf dans Ie cas d 'un fluide pur OU

pour lequel f = s/l et on peut faire une rotation du 3-plan espace de

maniere a amener v.,1 sur W1..• s.1 sont alorslesvaleurs propres
de la matrice ( fl"."~
~J"I.

§I. LE FLUIDE THERMODYNAMIQUE

4. Le fluide parfait et les variables thermodynamiques.

lJ..e tenseur d'impulsion-energie d 'un fluide parfait non conducteur de
chaleur est

(3.1)

II est clair que u' est vecteur propre oriente· dans Ie temps et f la
valeur propre correspondante de (T,,_). Tout repere propre de ce fluide
coincide avec un repere principal qui est indetermine a cause de la va-
leur propre triple ~p.

En vue de I 'etude energetique, on decompose la densite propre " en

la somme d 'une densite de matiere r et d 'une densite d 'energie vitesse
se r E OU t est I 'energie interne specifique

(4.2) t= r(1+f)
 - 11 -

Pham

On est amene a introduire l'indic e f du fluide defini par

(4. 3) f 1 +£ + E.r

Dans ces formules et dans la suite, les unites physiques ont ete choi-
sies de maniE~re que la vitesse limite c soit egale a 1. Autrement,
-2 -2
il faudra remplacer E , p par Ec , pc
Le tenseur d'impulsion-energie d 'un fluide parfait non conducteur
de la chaleur prend alors la forme

(4.4)

Du point de vue thermodynamique, la temperature propre e et

I' entropie specifique propre S peuvent ~tre definis comme en
hydrodinamique classique par la relation

(4. 5) adS d£ + pdt'

ou t t ~ est Ie volume specifique. En tenant compte de (4.3), et en

r
prenant f, s, p comme variables thermodynamiques (non independantes)
on peut ecrire (4.5) sous la forme equivalente

(4.6) r9dS rdf dp

Ces relations expriment qu'il existe Iparmi les variables r, 9, f, S, p,

seulement deux variables independantes pour lesquelles on choisira
souvent f et S ou S et p.
 - 12 -

Pham

Si on prend f et S comme variables independantes et si l'on se

donne p en fonction de f et S, la relation (4.6) entraine

I' = ~p r9
" 1f

La premiere relation definit l'equation d'etat du fluide sous la forme

r r (f, S)

et la seconde relation definit la temperature.

Des conditions de conservation appliquees au tenseur d'impulsion-e-
nergie (4.4) on tire l'equation de continuite et Ie systeme differentiel
aux lignes de courant

(4.7) V. (rfu') - u"1,p 0

(4.8)

Tenant compte de (4. 6) on peut ecrire l'equation de continuite sous la

forme

(4. 7))

d' ou l'on deduit

Theoreme - Pour un fluide parfait, il est equivalent de dire qu'il y a
conservation de la matiere ou que I' entropie est constante Ie long des
lignes de courant.
 - 13 -

Pham

un fluide tel que ud,S = 0 est dit adiabatique.

De m~me tenant compte de I 'equation thermodynamique, Ie systeme
differentiel aux lignes de courant s 'ecrit avec les variables f, S

(4.8')

On en deduit
TMoreme - Si Ie mouvement du fluide est isentropique (S const. )
Ie systeme differentiel aux lignes de courant se reduit a

(4. 9)

NOliS montrerons que pour un tel fluide isentropique, il existe un

principe extremal pour les lignes de courant

5. Le fluide visqueux.
Pour caracteriser la deformation locale du fluide, nous introduisons
Ia derivee de Lie du tenseur metrique g suivant Ie vecteur vitesse u
nitaire

et nous posons

(5. 1)
 ~ 14 -

Pham

ou rli~ est Ie projecteu[' d 'espace.

Les lois contraintes-deformations sont supposees lineaires, si Ie
milieu est isotrope, les phenomenes de viscosite sont decrits par Ie
tenseur

(5.2)

011

(5.3)

(5.3')

Le tenseur d'impulsion-energie du fluide visqueux homogene est

alors donne par

(5.4)

Cette expression est utilisee par C. Eckart, G. Pichon. A. Lichnerowicz

proposirlt une autre expression de ~III' 011 intervient dans Ia definition
de Ia viscosite Ie vecteur C" = fu" . Le tenseur d'impulsion-energie
devient alors
 - 1S -

Pham

avec

V designant la derivee covariante dans la metrique g= f2 g,

6, Le fluide conducteur de chaleur,

On tient compte des echanges thermiques par conduction, Celle-ci
est definie par un vecteur qO( orthogonal au vecteur uOl, C'est l'ex-
pression de qG(' sa presence dans Ie tenseur d'impulsion-energie
qui caracterise les points de vue,
Eckart a choisi Ie tenseur:

(9,1)

Les equations qui regissent I'evolution du fluide sont donnees par les
conditions de conservation du tenseur d'impulsion-energie, la conserva
tion du courant de matiere, l'equation de definition de qC(:

'if« (ruG) 0

(9,2)

et une equation thermodynamique,

 - 16 -

Pham

L'auteur a propose en 1954 Ie tenseur d'impulsion-energie

(9. 3)

ou I 'on a neglige la viscosite ~ q« est dMini par

(9.4)

Les equations du mouvement sont constituees par les conditions de c0E.

servation Vc( Till' "0 et I 'equation thermodynamique est remplacee par
I 'equation de conduction qui generalise celle de Fourier

(9. 5)

C est la chaleur specifique a volum constant et ( la chaleur de dila-

tation du fluide. P.ichon a repris ce modele en ajoutant Ie terme de vi-
scosite 811/, .
Pour Landau et Lifchitz, Ie tenseur d'impulsion-energie d'un flui-
de conducteur de chaleur est identique a celui d 'un fluide parfait, Ie
vecteur conrant de chaleur qlll apporte sa contribution a travers 1'equ~
tion de conservation d 'un certain vecteur POI . 11 pose

(9. 6)

(9.7)
 - 17 -

Pham

Les equations du mouvement sont donnees par

VIII pOi 0

auxquelles ont ajoute l'equation de definition de qlll:

2 • • l+G
(9. 8) q. = -ore (g" - u,.u )I,( -9- )

01' G est la tiDnction de Gibbs definie par

(9. 10) G =E+ .£. - 9 s.

r

Ces modeles se justitlent par des considerations physiques et cine-

tiques et ont tops Ie merite de se reduire a la limite a la description
classique non relativiste. L'etude du probleme de Cauchy montre que les
les systemes d 'equations auxquels ils donnent lieu, sont mixtes et com
portent une partie parabolique provenant soit de la viscosite, soit de la
definition du vecteur courant de chaleur qO(. Ce qui conduit a une vi-
tesse de propagation infinie.
Pour lever la difficulte provenant de qllL' Cattaneo et Vernotte
ont suggere de modifier I 'hypothese de Fourier par un terme de Relax~

tion. Kranys a fait la traduction de cette hypothese en relativi8i~

(9. 11)
 - 18 -

Pham

C e vecteur qat n I est alors plus orthogonal a u at. Adoptant Ie point

de vue de Landau- Llfchitz et celui de Cattaneo- Vernotte-Kranys,
Mahjoreb a propose une nouvelle theorie cette annee.
 - HJ -

1'ham

~). LE CHAMP ELECTROl\TAGNETIOUE

10. Representation du champ electromagnetique.

S'il existe un champ electromagnetique, Ie fluide est soumis a des
inductions electromagnetiques qu 'on peut decrire a I' aide de deux
formes: la 2-forme champ electrique-induction magnetique H et la
2-forme induction electrique-champ magnetique G. On note In, IG
leurs formes duales au sens de l'element de volume riemannien ~ de
l'espace- champ. On a en composantes

(10. 1)

(10.2)

On appelle vecteurs champs et inductions eleclriques et magnetiques

les vecteurs definis par les 1-formes

(10.3) e = i H diG h i (. G) b i (lH)

u u u u

ou i
u
est Ie produit interieur par Ie vecteur vitesse unitaire u. En
composantes, on a

(10.4) e

Ces vecteurs sont orthogonaux a u"'.

Inversement, H, G, *II, *G s'expriment en fonction de e,d,h,b
 - 20 -

Pham

par les formules

(10. 5) H uAe-.(uAb) G (uAd) - '(UAh)

(10.6) •H u" b + ~u Ae)

Dans les deux dernieres relations Ie signe + provient du fait que sur
une vari~t~ riemannienne I 'op~rateur * satisfait a larelation .. 2
= Eg (-l)P (n-p) oil n = dim V ,p = degre de la forme et £g Ie
signe du det. g. On en cteduit les relations suivantes qui donnent lascorn
composantes

En theorie electromagn~tique de Maxwell, les inductions d~pendent

lineairement des cilamps. Dans Ie cas isotrope oil Ie fluide a une per-
mitivit~ dielectrique 1 et une permeabilite magn~tique ", on a

(1 0.7) d =1 e b =}'-h

Les deux relations (10.5) donnent alors

1 ~,,-1
(10.8) G
,,-
H + - - uAi H

'" u
 - 21 -

Pham

soit en composantes

(10.8')

qu 'on peut mettre sous Ia forme

(10.9)

ou

L'induction electromagnetique H, G satisfait aux equations de

Maxwell

(10.10) dH a

(10. 11) ~G J

ou ~ est Ia codifferentielle, June I-forme dont Ie vecteur asso-

cie dHinit Ie courant electrique. L'equation (10.10) signifie que Ia
2-forme H est Iocalement exacte i. e. qu'il existe Iocalement une
I-forme + telle que H = d +. + s 'appelle Ie potentiel vecteur eIec-
tromagnetique, (10.11) donne en remarquant que ~2 = a

(10.12) ~J a
 - 22 ~

Pham

equation qui exprime la conservation du courant electrique.

En composantes, les equations (10.10), (10.11), (10.12) s'ecri-
vent

On decompose Ie courant electrique J en un courant de convec-

tion colineaire a u et un courant de conduction r orthogonal a u.
r peut Nre ctefini par l'hypotMse d 'Ohm r =, e ou ~ est la conduc
tivite electrique du fluide. On a alors

(10.13)

t s'appelle la densite de charge.

11. Le tenseur d 'energie electromagnetique.

A partir de HGI~' Gill, on construit Ie tenseur d'impulsion-energie
electromagnetique 'ru~ dont la divergence donne la densite de force e-
lectromagnetique agissant sur Ie fluide. En generalisant un resultat con
nu dans Ie cas non inductif ), =". = 1, on obtient Ie tenseur donne par
Minkowski
 - 23 -

Pham

(11. 1)

Pour interpreter ce tenseur, on va 1'exprimer a 1'aide des vec-

teurs e, d, h, b:
On obtient

ou

(11. 3)

Pet. est Ie vecteur de Poynting et Q" = ~~PCI . Sur (11. 2) on voit Ia

signification de chaque groupe de termes.
'r~~ n 'est pas symetrique. On peut prendre 1'expression proposee
par Abraham

(11. 4) 't'cl~

On peut penser a Ie symetriser [ 1, mais Ies raisons physiques

sont obscures.
 - 24 -

Pham

Nous conservons l'expression (11.1). En prenant la divergence

de ce tenseur, nous avons

(11. 5)

Or les equations de Maxwell du {r groupe s 'ecrivent encore

Par multiplication contractee avec Clll' , il vient

En portant dans (11. 4), on a alors en tenant compte de la definition

de J

(11. 5')

On peut transformer la parenthese en utilisant les equations de liai-

son ce qui donne finalement

(11. 6)

011
 - 25 -

Pham

La signification du groupe J'H" est claire. Le tenseur supple'men-

taire (~f-1)V~u' Pf est nul si ).~= 1 oft P_ = 0, c'est-a.-dire
si 1".~ ess symltrique. II est encore nul si uCl est un champ a.
d'riv'e covariante nulle. Le terme suppltmentaire }- (e~ et'),l+
+ h.h' ~~ .... ) correspond aux phenomenes de magnetostriction et d 'e-
lectrostriction. En fait k,rdependent des variables d'etat.

I
12. Cas ou 'rill, defini par (11.1) est symetrique.
Le tenseur -r., n'est pas symetrique en general. Son antisymetri-
se est (1", -1) (UGI p~ + u, POI ). Comme Uti et Pel sont orthogo -
naux, I 'ant)'symetrise est nul i. e. que 'r_, est symetrique si:

1. - ~~= 1, cas non inductif

2. - p. = 0,
, I
ce qui est verifi' soit que e. = 0 soit que hot = O.

Dans Ie cas non inductif, on prend 1. =JA-= 1 , alors HC(~ = GcI.~ =

F«~ . Le tenseur d'energie electromagnetique s'ecrit

(12.1)

et

(12.2)

Le cas e..=O correspond a. celui de la magnetohydrodynamique ou

 - 26 -

Pham

fluides de conductivite rt = ()(). Comme Ie courant electrique doit Nre

borne ~ e< IJO, on a necessairement e = 0 . Le tenseur d'impulsion-

energie electromagnetique s'ecrit alors

Le cas h. = 0 conduit a

11 convient pour la description d 'un electron considere comme une

boule continue.
Dans chacun de ces cas, Ie tenseur d'impulsion-energie du fluide
s'obtient a ajoutant 't'0(~ a l'expression deja connue. On obtient un
tenseur d'impulsion-energie du fluide qui est symetrique, par suite on
peut ecrir les equations d 'Einstein.

13. Le fluide parfait charge sans inductions.

On a dans ce cas Ie tenseur d'impulsion-energie total

(13.1)

ou

Les equations du mouvement sont donnees par les conditions de

 - 27-

Pham

conservation

(13. 2)

I 'equation thermodynamique

(13. 3) r 9d S rdf - dp

et les equations de Maxwell

(13.4) o

(13. 5)

Ou supposons la conductivite i = 0, de sorte que J' = 1u ~ et

(13. 6)

En prenant f, S comme variables thermodynamiques, les condi-

tions de conservation (13. 2) donnent I 'equation de continuite et Ie
systeme differentiel aux lignes de courant

(13.7)

(13.8)
 - 28 -

Pham

On en deduit que si Ie mouvement est adiabatique (uCl 0« S = 0), il y

a conservation de matiere

(13.9)

De (13.6) et (13.9) on tire

diOli par difference ultOct log 1.r = O. Le rapport .lr est dOIlC cons
tant Ie long des lignes de courant, on pose

(13.10) K = 1.
r

Le systeme differentiel· aux lignes de courant se met alors sous

la forme

(13.11)
 - 29 -

Chapitre 2

LE PROBLEME DE CAUCHY

Pham

Pour chaque modele de fluide, on obtient un systeme fondamental d 'equ~

tions aux derivees partielles pour etudier l'evolution du fluide. Un pro-
bleme essentiel est de voir dans quelles mesures ces equations deter~

nent les fonctions qui representent les grandeurs physiques envisagees.

Comme c 'est un probleme d 'evolution, Ie probleme mathematique pose
es. Ie probleme de Cauchy. Les donnees initiales portees par une hy-
persurface E de 1 'espace-temps determinent - elles les grandeurs
dans Ie voisinage de z..
Le seul theoreme general c1assiquement connu en reponse a cette
question est Ie theoreme d'existence et d'unicite de Cauchy-Kowaleski
valable dans Ie cas analytique pour un systeme de N equations aux
derivees partielles a A
N fonctions inconnus dont Ie polynome caracter2,
•

stique n 'est pas identiquement nul.

L'hypothese d'analy1idte restreint considerablement la portee de ce
theoreme en Physique. On peut maintenant se passer de I 'hypothese
d'analyti.dte pour des systemes quasi-lineaires hyperboliques stricts.
Pour de tels systemes, Leray a demontre un theoreme d'existence et
d'unicite pour Ie probleme de Cauchy non analytique. Toute solution de
ce probleme possede un domaine d'influence, c'est-a-dire que la valeur
en un point ne depend que d'une partie des donnees initiales, celles se
trouvant a l'interieur d'un certain cOnoide de sommet ce point. C'est
cette notion d'hyperbolicite stricte et son critere que nous allons expo-
ser en vue de I 'appliquer aux differents systemes d'equations trouvees
dans Ie chapitre 1.
 - 30 -

Pham

I
t LTHEOREME D'EXISTENCE ET D'UNICITE
POUR LES SYSTEMES STRICTEMENT HYPERBOLIQUES

1. Systemes strictement hyperboliques.

Soit V une variete differentiable de classe Ck (k suffisamment
n
grand) et de dimension n.
Soit a(x, D) un operateur differentiel d'ordre m agissant sur les
fonctions. Localement a depend des coordonnees locales x" et
des derivees partielles de( . Ponr JeT' (V ) , a (x, ~) est un
x n
polynome reel en ~ de degre m. On designera par h (x, ~) la paE.
tie principale de a (x, ~ ) c'est-a-dire la partie homogene de degre m
de a (x, ~). Soit V (h) Ie ctme projectif defini dans T*(V) par
x x n
I 'equation h (x, ~ ) = o.
Definition 1. - L'operateur differentiel a (x, D) est dit strictement
hyperbolique au point si I 'hypothese suivante est verifiee:
xEV
n
(H) Il existe dans T* (V ) des points ~ tels que toute droite is-
x n
sue de ~ et ne passant pas par Ie sommet du cOne V (h) Ie coupe
x
en m points reels distincts.
S'il en est ainsi, l'ensemble des points ~ forme l'interieur de deux
demi-cOnes convexes ~
appos~s . r+ (a)
non vldes
x
et r- (a)
x
dont 1es
bords appartiennent a V (h).
x
Considerons maintenant un operateur differentiel matriciel dia-
gonal A(x, D) suffisamment differentiable en x
 - 31 -

Pham

A(x, D)

ou les a.(x, D) sont des operateurs differentiels d'ordre m(i).

1
Definition 2. - On dit que l'operateur differentiel diagonal A (x, D)
est strictement hyperbolique en un point x si
1) les a.(x, D) sont strictement hyperboliques en x
1
2) les deux demi-cOnes convexes apposes

r +(A)
x
= n
1
r+(a.)
x 1
r-(A)
X
n 1
r-(a.)
X 1

ont un interieur non vide.

Pour definir l'hyperbolicite stricte dans un domaine n. (ouvert con-
nexe) de V, introduisons Ie cOne
n x
dual du cOne C+(A) C+(A)
x
r +x (A).
est ferme de l'ensemble des vecteurs XET (V) tels que <: f ' X > " 0
x n
pour tout 4E r+(A).
x
Le cOne C- (A)
x
dual de r x-(A) est defini. de ma-
niere analogue. SoH

+
(1. 3) Cx (A) C (A)
x
U Cx- (A)

Un chemin r: [0,1] .... Vn differentiable est dit oriente dans Ie

temps relativement a A si la demi-tangente positive en chaque points
de r est dans C:(A). Une hypersurface differentiable r; est dite orien
 - 32 -

Pham

tee dans I 'espace relativement a A si Ie sous expace vectoriel tan

gent T (E)
x
en chaq~e point x de 1: est exterieur a CIA).
Definition 3. - On dit que I 'operateur A(x, D) est strictement
hyperbolique dans tin domaine n C Vn si les deux conditions suivan
tes sont satisfaites:
1) A(x, D) est strictement hyperbolique en tout point x 611
2) I 'ensemble des chemins temporels joignant deux points x o ' xi
quelconques de .n est compact ou vide pour la topologie de la
convergence uniforme de l'ensemble {~: [0, 1] ~V n ' r(O) = Xo ,~(~)=
= xl} •
Si A(x, D) est differentiable en x et si A(x, D) est strictement
hyperbolique en un point x.' on peut montrer qu'il existe un voisin~ge

ge ouvert connexe ..n de Xo hOm€l)Illorphe a une boule de Rn dans

lequel A(x, D) est strictement hyperbolique. Un tel ouvert sera dit
simple.

2. Systemes de Leray.
Considerons un systeme d'equations aux derivees partielles de N
equations a N inconnues (u i ) et n variables (xIII) qqe l'on ecrit
symboliquement

(2.1) A(x, u, D) u + B(x, u) 0

ou A(x, u, D) est une matrice diagonale d'elements a.(x, u, D), i = 1, ... , N

1
et B(x, u) une matrice colonne d'elements b. (x, u). Les a.(x, u, D) sont
1 1
des operateurs differentiels d 'ordre m(i).
Associons a chaque inconnu u i un entier s(i} )., 1 et a chaque equ~
 - 33 -

Pham

tion de rang un entier t(j) ~ 1 tels que:

(2. 2) m(i) s(i} - t(i) + 1

les entiers s(i) , t(j) ne sont definis qu 'a une constante additive
pres.
Definition. - On dit que Ie systeme diagonal (2.1) est quasi-li-
neaire au sens de Leray si pour tout i, l'operateur differentiel
a. (x, u, D) est lineaire par rapport aux derivees d 'ordre m{i}, si
1
les relations (2.2) sont verifiees et si les a., b. sont des fonctions
. 1 1 .
suffisamment regulieres de x·, de uJ et des derivees des uJ
d'ordre Ss(i) - t(j) , si s(i} - t(j) < 0 a. et b. sont independa~
1 1
ts de uj .
Ceci etant, Ie probleme de Cauchy pour Ie systeme (2.1) se p~

se de la maniere suivante. Soit n un domaine simple de V.

n
Les
donnees de Cauchy sur une hypersurface E plongee dans 12 consiste
tent en les valeurs des fonctions u ~, de leurs derivees d 'ordre
< m(i) . II existe (toujours) des fonctions wi admettant des deri -
vees d 'ordre ~ s(i} + 1 de casses localement integrables dont les
traces sur E sont les donnees de Cauchy envisagers.
Une solution du probleme de Cauchy pose est alors une solution
(u~) de (2.1) dont les derivees d'ordre 's(i) sont de carres locale
ment integrables et coincident su E avec celles de
i
w. Pour ce pr~

bleme, J. Leray a demontre un theoreme d'existence et d'unicite que

nous allons enoncer sans demonstration.
Theoreme. - Si les donnees de Cauchy sur 1: sont definies par
i
des fonctions w verifiant les hypotheses:
 - 34-

Pham

1) l'operateur differentiel A(x, w, D) est hyperbolique strict

dans n et l'hypersurface L orientee dans l'espace relativement a
A(x, w, D)
2) les a.(x, w, D)w + b.(x, w)
1 1
s'annulent sur 1: ainsi que leurs
derivees jusqu 'a l'ordre t(O - 1.
Alors pour tout x E1:, Ie probleme de Cauchy pour (2.1) admet
au moins une solution dans un voisinage de x. Si (n i ) et (u i )
i
sont deux solutions et si iii et u ont des derivees d'ordre ~ s(i)+l

de carres localement integrables, alors elles coincident.

Definition. - Un systeme quasi-lineaire qui verifie les hypotheses
precedentes sera dit un systeme quasi-lineaire strictement hyperboli-
que, ou systeme de Leray.
Les systemes quasi-lineaires qu 'on rencontre en Physique ne sont
pas toujours des systemes diagonaux. Pour appliquer Ie theoreme de
Leray, il faudra les ramener a la forme diagonale. On sait qu 'on Ie
peut toujours. Mais Ie caractere strictement hyperbolique doit ~tre de
montre. Nous allons exposer la methode dans un cas .

• ~. APPLICATION AUX EQUATIONS DE

L'HYDRODYNAMIQUE DES FLUIDES PARFAITS

3. Les coordonnees harmoniques.

Dans l'etude du probleme de Cauchy relatif au systeme des equa-
tions d 'Einstein, les coordonnees harmoniques sont un outil precieux.
Definition. - Un systeme de coordonnees locales (x') est dit
harmonique si chaque fonction coordonnee x fest une solution de l'e
 • 35 •

Pham

quaiion de Laplace

(3. 1)

ij~ etant les coefficients de la connexion riemannienne.

On remarque que les varietes caracteristiques de (3. 1) sont
tangentes en chaque point au cOne elementaire C de l'espace.temps.
x
Si (x') est un systeme de coordonnees locales, on pose

(3.2)

F' dependent de g.~ et de leurs derivees premieres. Si F'= 0,

Ie systeme de coordonnees locales est harmonique.
On associe a F' les quantites LCII~ definies par

(3.3)

Lemme. - Dans un systeme de coordonnees locales arbitraires,

les composantes du tenseur de Ricci peuvent se mettre sous la forme

(3.4)

ou

les F.~ etant des fonctions regulieres.

 - 36 -

Pham

II suffit pour demontrer Ie Iemme de chercher a mettre en e.vi-

dence Ies derivees du second ordre de gl}!l dans LG!~ et Rcr~'

On a modulo Ies termes en g1,. et '6r g1....

soit

D'autre part

soit

On en deduit
 - 37 -

Pham

puis Ie lemme.
Corollaire. - En coordonnees locales quelconques, on a

(3. 6)

ou

S (h) R (h) 1 R (h)

(3. 7) OI~ ~~ '2 gct~

(3. 8)

(h) ~.. R(h)

avec R g ~I" et L = g1J" L1-,..,. Si les coordonnees sont
harmoniques K.~ = O.

4. Application a I 'etude des solutions des equations d 'Einstein.

Theoreme. - Toute solution du systeme des equations d'Einstein

(4.1)

est une solution en coordonnees harmoniques du systeme

(4.2)

(4. 3)
 - 38 -

Pham

Inversement toute solution du systeme (4.2), (4.3) satisfaisant sur

une hypersurface 1: orientee dans l'espace aux conditions

(4.4)

(4.5)

est une solution du probleme de Cauchy correspondant pour Ie syst~

me (4.1) des equations d'Einstein.

En effet toute solution de (4.2) ecrite en coordonnees harmo-
niques verifie (4. 2) car L«~ = 0 et comme solution de (4. 1), el-
Ie doit verifier les conditions de conservation ,,,S II~ _-
n'
0 i. e. ()
4. 3 .
Reciproquement considerons une solution de (4.2), (4.3) corres-
pondant a des donnees de Cauchy satisfaisant sur E aux conditions
't' 0 (h). •
(4.4) et (4.5). Sur"" on a SCI! = S cI + K« et en vertu de (4.2)
et (4.5) on a

"-
K' - 0 sur I

En explicitant l'expression de K~ , on a

soit en tenant compte de la condition (4.4) qui entraine sur I , '~F'

=0
 - 39 -

Pham

K' = 2..2 g"t gOO '1 F~ = 0

G( II,
Co:nme l: est oriente dans 1'espace , g" fa, on a necessairement
'()o F' = O.

Ainsi la solution consideree de (4.2), (4.3) satisfait sur r. a

(4.6)

Pour cette solution on a encore

6)n en deduit en vertu de (4. 3)

(4. 7) V~K~,lt' = 0

Or

donne par derivation

 - 40 -

Pham

II en resulte que l'equation (4.7) peut s'ecrire

(4.8)

ou A1,- , sont des fonctions regulieres.

Ainsi F' satisfait a un systeme lineaire hyperbolique qui ad -
met un tMoreme d'existence et d'unicite. Si l: est oriente dans l'es
pace, la seule solution de (4.8) qui satisfait sur t a F' = 0 et
'ao F' = 0 est la solution nulle. 11 en resulte que la solution consi-
cteree est une solution ecrite en coordonnees harmoniques des equa -
tions d'Einstein (4. D) •

5. Analyse formelle du systeme fondamental de I 'hydrodynamique.

Le systeme fondamental des ~quations de I 'hydrodynamique des
fluides parfaits adiabatiques est forme des equations

(5.1)

(5.2)

(5.3)

(5.4) dp rdf - redS

 Pham

On prendra f et S comme variables thermodynamiques r =

r(f, S) et p=p(f, S) sont alors des fonctions connues de f et S.
C 'est un systeme de 16 equations aux derivees partielles aux 16
fonctions inconnues g«" f, S, uct. On va d 'abord faire une analyse for-
melle du probleme de Cauchy. Pour cela on se donne sur une hyper-
surface L d'equation locale x lS = 0, les valeurs des g•• , ~,gOl' ,S
et on cherche a determiner la solution au voisinage de 1:. Nous suE.
posons que I. n'est pas tangente aux cOnes elementaires i. e.

(5. 5) gH f °
et que sur Z on a

(5. 6)

Vne etude classique montre que si gO' f 0, les quantites

S'III sont connues en fonction des donnees de Cauchy gO!, , 1. gill,
Les donnees de Cauchy (g." 'a o gill, ' S) doivent alors verifier les
conditions de compatibilite

S'CII = X (rfu'11a! - pg'jI()

Supposons provisoirement connus les valeurs sur 1: de f. Les equa-

tions precedentes s 'ecrivenil

(5.7)
 - 42 -

Pham

En tenant compte du caractere unitaire (5.3) de uel, on tire

(5.7) donne alors la valeur de u'

(5. 8)

pui celle de

(5. 9)

L'equation (5.9) definit implicitement la ou les valeurs possibles

de f. On va 1'ecrire

En derivant par rapport a f, il vient en tenant compte de (5.4)

soit d'apres (5.7)

On voit que fest connue sur 1'. si F'f f 0 i. e.

 - 43 -

Pham

g
(JD
-
(1 _ fr ,.; ) u 'u·.,tr 0

Dne fois f connue sur l:, on en deduit ul a I'aide de (5.7)

si

Ceci etant en vertu des raisonnements faits au paragraphe 4 pr~

cedent, on peut substituer au systeme (5. I), (5.2), (5.3), (5.4) Ie s~

steme

(5.10)

(5.11)

(5.12)

(5.13)

oli (5.12), (5.13) proviennent des conditions de conservation Vo(TIII~= 0

et (5.3). On remarque que (5.13) entraine
 - 44 -

Pham

ce qui montre que si u~ est imitaire sur Z, i1 Ie reste au voisina

ge de Z.
8upposons que les donnees de Cauchy sont donnees en termes de
series formelles par rapport aux coordonnees locales et cher
chons les solutions formelles du systeme (5.10), (5.11),
(5.12), (5.13).
En mettant en evidence les derivees O"glll,' 1,8, ~.f, l,u'
dans ces equations, il vient

(5.14) (d. c. )

(5.15) u'1,8 (d. c.)

(5.16)

(5.17)

oil. les seconds membres sont connus en fonction des donnees de Cau-
chy. (5.14) donne ~ gil' si gel;. 0; (5.15) donne 00 8 si
uO f 0, (5.16) et (5.17) pour ~= 0 determinentalorsO,u',?,f
si Ie determinant du systeme g" - (1 - 1...Q)
r
u' u' f 0; enfin (5. 17)
pour ~ = i donnent 1. u' si u' f o.
 - 45 -

Pham

Ainsi sous la condition que:

uOf 0

on peut calculer 100 glll~' ~o S, 'a, f et 'aou· . Les m~mes conclusions

s 'etendent aux derivees d 'ordre superieure qu 'on obtient en derivant
par rapport a x O les differentes equations, de sorte que les series
formelles cherchees sont uniquement determinees.
Si Ie probleme de Cauchy est analytique, il resulte du tMoreme de
Cauchy-Kowaleski Ie resultat suivant.
TMoreme. - Dans Ie cas analytique; si les donnees de Cauchy gCIIP'
'ao g"P' S satisfont aux conditions
1) la forme get~ XCIX' est hyperbolique normale
2) I 'hypersurface 1: portant les donnees de Cauchy et dMinie loca-
lement par XO 0, est orientee dans I 'espace
3) sur Z on a F' = 0 et So. = t T'"
alors Ie probleme de Cauchy pour Ie systeme des equations de I 'hydr~
dynamique admet une solution analytique et une seule au voisinage de
tout point x E Z.
Varietes caracteristiques. - L'etude precedente montre que les vari~

tes caracteristiques du probleme de Cauchy sont dMinies par les equ~

tions suivantes
 - 46 -

Pham

Les premieres definissent les ondes gravitationnelles , les secondes

les ondes materielles ou entropiques et les troisiemes les ondes hy-
drodynamiques. Un calcul classique montre que ces trois sortes d'o~

des se propagent avec les vitesses 1, 0, v respectivement avec

v = ~ ;fr
t r~
Sous I 'exigence relativiste on doit avoir.L:.f ~ 1 pour que v 4- 1
r
ce qui entraine que les varietes caracteristiques definissant les on -
des hydrodynamiques doivent Hre orientees dans Ie temps.

6. TMoreme d'existence et d'unicite.

Reprenons Ie systeme etudie au paragraphe precedent

(6. 1)

(6.2)

(6.3)

(6.4)
 - 47 -

Pham

Il ne se presente pas sous forme diagonale sauf pour (6.1) et (6.2).

Nous Ie transformons de maniere a obtenir un systeme diagonal.
Gardons pour Ie moment (6. 1) et (6.2) tels quels.
Pour avoir 1'equation en f, prenons Ia derivee contractee V,
de (6.4) il vient

en f, 1 en uol

ou Ia notation au 2e memerre signifie que Ies termes non explicites

contiennent des derivees d'ordre maximum indique pour chaque
inconnue . Pour avoir uC(V~ V. u~ , considerons I'equa.tion (6.3) qui
se developpe comme

en Ia derivant suivant Ies lignes de courant -et utilisant 1'identite de

Ricci (VatV~ -V.~) u' = - RoI , u', on a

En portant dans 1'equation en f, on obtient

(6. 5)

1 en S, 1 en f, 1 en ull )
 - 48 -

Pham

Appliquons l'operateur urV r a cette equation

2 en S, 2 en f, 2 en ud }

Or l'identite de Ricci donne

(6.6)

si on tient compte de ud. ~el S = 0, on a

et l'equation erl f s 'ecrit finalement

(6.7)
t r! }uclu.] u1 ~OlPr f
[ g'" - (1 - r;t = F (3 en gCII~' 2 en S,

2 en f, 2 en u GI )

Pour les inconnues u~ considerons Ie systeme (6.4) auquel nous

appliquons l'operateur [g't~ - (1 - lrfi ) uCll u'1 V. V.

f [g"': (1 - :i )u" u'l u V~ V,), - [

f rvef, gd.' - (1 - lrl"t }u· u'] (l~ u' ul ).
 - 49 -

Pham

= q (3 en gOl~' 2 en S, 2 en f, 2 en uCl)

en tenant compte de cette relation l'equation en u 1. devient

= r(3 en gee" 2 en S, 2 en f, 2 en ucc )

Or ur~,s = 0 entraine (6.6) de sorte que on obtient finalement

Nous avons ainsi trans forme Ie systeme (6.1), (6. 2~, (6.3), (6.4) en
Ie systeme diagonal suivant.
 - 50 -

Pham

(6.11)

2 en f, 2 en u at )

(6.11)

2 en f, 2 en u«

ou (!6.9) s 'obtient a partir de (6. 1) gr~ce a l'expression de R (h~~.

Ordonnons l'ensemble {g",~, S, f, ulll } en Ie numerotant de 1 a 16.
Nous avons avec les notations evidentes

m(~~) 2 0,1,2,3

m(11) = 1

m(12) 3

AS80cions aux inconnus et aux ~quations les indices suivants

a(N) = ~IIlP -(1 - !;f )ullll] l?pr m(N) = 3 N = 13, 14, 15, 16

Associons aux inconnus et aux equations les indices suivants

 · 51 -

Pham

s(d~) 4 s(11) 3 s(12) 3 s(N) 3

t( ~ ~) 3 t(11) 3 t(12 ) t(N)

et dressons Ie tableau de l'ordre maximum des derivations et celui

des differences s(i) - t(j):

~ 0 0 0 l 0 0 0

0 0 4 0 0 0

~ t 1 $ ! t
,
l )

~ l t 3 ~ t t

ordre maximum des derivations matrice ( s(i) - t(j) )

Nous voyons que l'ordre maximum des derivations ainsi que l'ordre des
operateurs differentiels a(i) est compatible avec Ie choix des indices.
Nous avons ainsi montre que Ie systeme (6.9), (6.10), (6.11), (6.12)
est un systeme quasi-lineaire au sens de Leray.
Montrons que c 'est un systeme strictement hyperbolique.
Les donnees de Cauchy sur 1: sont par example les series formelles
calculees au • 5 (il suffit de prendre la somme des n premiers ter-
mes, pour n convenablel. Leurs derivees d'ordre 's(i)+J sont mani
festement de carres localement integrables. Nous supposons de plus que
les donnees d~ Cauchy (gtA~, 0, g •• ' S) sur 1. satisfont aux conditions:
1) la forme quadratique gel~ x· X. est hyperbolique nor-male
 - 52 -

Pham

2) sur Z on a F' = 0 , S'" = XT·. ; la seconde relation defi

nit une valeur admissible de f telle que 11 ~ 1. -
. r
Avec ces donnees on peut rechercher les deml-cOnes
f
r +
x
(ai).
1
A l'operateur a(~~) correspond Ie cOne l' uftolt.tr f 0 et
+
Ie demi-cOne r x ( ~)
I( dMini par

A I 'operateur a( 11) correspond Ie cOne 5« o et Ie demi-cO

r
uti.

ne + (11) defini par

x

Aux operateurs a(12) et a(N) correspond Ie cOne

dMini par

(6.12)

r
Sous I 'hypothese11"1 , nous voyons que l'intersections de ces

x
r
trois demi-cOnes r + (cl~), r+ (11), + (12) =
x x x
r
+ (N) n'est autre

que Ie demi-cOne (6.12) qui a un interieur non vide.

Nous avons donc prouve que l'operateur differentiel A = (a.) est
1
hyperbolique strict en chaque point x. Comme il est differentiable
en x , il est strictement hyperbolique dans un voisinage ouvert con-
 - 53 -

Ph am

nexe U de x. L 'hypersurface initiale ! a ete choisie orientee

dans I' espace relativement a A, il reste a preciser les valeurs
sur t des derivees d'indices 0 de ~" S, f, uGC d'ordre ~ s(i) -1,
soit:

S ,1,s, ?.,S

f, 'a,f, ?'of

ot ~.I\
U ,110 U, '00 Ud

g",., ~gll" S ont ete donnes dans Ie probleme de Cauchy formel du

t5 "g.. ,S , \S, f, ~f, u"', ~ou· sont obtenus par la resolution
de ce probleme, pour avoir ~,g." 100 S, 'blOf, ~"l', il suffit
de deriver les equations (6.1), (6.2), (6.3) et (6.4).
Les hypotheses du theoreme de Leray sont donc satisfaites. On en
deduit que Ie probleme de Cauchy pose pour Ie systeme non analytique
(6.9), (6.10), (6.11), (6.12) admet une solution unique (gGl,' S, f, uel)
au voisillage du point x ,1: .
II reste a montrer que cette solution est solution du systeme initial
(6.1), (6.2), (6.3), (G.4) . Or si les donnees sont analytiques, la solu
tion (&i~, f, S, uGl) est analytique et est necessairement la solution
 - 54 -

Ph am

analytique du premier probleme.

Si les donnees ne sont pas analytiques, on approche Ie systeme
par des systemes analytiques et dans chaque cas, la solution du se
cond probleme de Cauchy est solution du premier probleme de Cau-
chy. Par passage a la limite, la solution du probleme de Cauchy de
Leray est encore solution du systeme initial. Et nous avons demon-
tre Ie theor!me d'existence et d'unicite pour Ie probleme de Cauchy
non analytique relatif au sisteme fondamental des equations de 1'hydr~
dynamique relativiste des fluides parfaits.
 - 55 -

Chapitre 3 Pham

L'HYDRODYNAMIQUE RELATIVISTE DES

FLUlDES PARFAITS ISENTROPIQUES

tL LA FORME DIFFERENTIELLE INVARIANTE

1. Systeme diffhentiel aux lignes de courant.

Le fluide parfait est dit isentropique si I' entropie S est constante.
Dans ce cas (cf. Ch. 1, .4) Ie systeme differentiel aux lignes de cou
rant s 'ecrit

(1. 1) ulA

(1. 2) uotv.cl u~ - (gIA"

~ -u u, ) ~f - 0
-f- -

ou s est l'absciae cuririligne sur les lignes de courant; fest l'indi-

ce du fluide. On a vu que

Nous nous proposons de mettre en evidence les proprietes geometri

ques du mouvement du fluide.

2. Variation d 'une integrale.

Soit V une variete differentiable de dimension n ,'/'(: T(V ) ..... V
n n n
Ie fibre des vecteurs tangents en tous les points de V et D(V )-+V
n n n
 - 56 -

Pham

Ie fibre des directions tangentes. T(V) est de dimension 2n et

n
D(V) de dimension 2n - 1. Comme T(V) et D(V) sont loca-
n n n
lement triviaux et qu 'on peut choisir comme cartes locales des car-
tes locales induites par celles de V, en coordonnees locales un
n
point de T(V) est defini par I 'ensemble (x~, X"') ou (xlt) est un
n
point de I' ouvert U de coordonnees locales de V et (XCi() les
n
composantes d'un vecteur tangent en (xU) E U relativement aces
coordonnees (xCII) ; un point de D(V) est defini par I 'ensemble
n
(x", u·) ou uC( sonLa, parametres directeurs de la direction.
Soit C: [to, ta - Vn une courbe differentiable dans V n d'ori-

gine xo= X(tD) et d'extremite x~ = x(t.). En coordonnees locales,

elle est definie par la representation patarnetrique

(2.1)

On posera

(2.2)

Cette courbe se releve dans T(V) en la courbe L: t_(xOl(t),x"(t)

r dans
n
et si x-(t) n'est pas nulle pour tout t, en la courbe D(V).
n
Soit F: W(V ) -+ Rune fonction a valeurs scalaires donnee sur
n
T(V ) positivement homogene de degre
n
1 par rapport a X i. e. pour
tout x fixe, F(x, lX) = 1 F(x, X). A toute courbe C: [to, t.] - Vn'
F(x·, x') est une fonction de t et on peut calculer 1'integrale

(2.3)
 - 57 -

Pham

Cette integrale est en fait intrinsequement attachee a F et C et ne

depend pas de la representation parametrique.
Calculons la variation de cette integrale pour une variation quelcon-
que a extremites non fixes de C. En supposant que C est dans un
ouvert de coordonnees locales, on a

dt

soit d'apres un raisonnement classique du calcul des variations

(2.4) Sx > dt

ou fA) est la 1 furme definie sur T(V) par

n

(2.5)

et P un covecteur defini en composantes par

d ~F ?F
(2.6)
ill ~ xii - 'fX"

les Pc( ne sont autres que les premiers membres des equations d'Eu
ler du calcul des variations. ix est un vecteur.
Si C n'est pas dans un ouvert de coordonnees locales, on la re -
couvre par un nombre fini de cartes locales et on peut etudier la varia
tion de l'integrale qui est la m~me.
 - 58 -

Pham

3. Principe d'extremum pour les lignes de courant.

Appliquons les resultats precedents au cas OU Vest I 'espace-
temps et

(3. 1)

On a a calculer les variations de l'integrale

(3.2)

supposee evaluee Ie long d 'un courbe C orientee dans Ie temps.

On a

~F fg", x. ~
~F ~fg~r x·~·r 1
x + Zf 0.ilg~r x~ x"'
, --
to ~U ~g ').,. x~~,.,. 'hOi Vg,.~ x~ x'"

d fgd~ x~ ~dfg~~ ~p ~r + -} f ~g~r ~, ~r

PCI(
dt
Vg)." ~x'" Vg,.,.. x" x,..
AUf lieu d'un parametre t
quelconque, prenons l'arc S de courbe com
.J dx~ -
me parametre. Le vecteur x'" = - est alors Ie vecteur vitesse unitai
ds
re, de sorte que l'on ait par un calcul facile

111." II.
(3.3) Pat = f [ u 'd. u~ - (g, - u u~
(I
'atl/ ]

(3.4)
 - 59 -

Pham

On obtient ainsi Ia formule

(3. 5) ~S =(W, ~x>~

4

qui donne Ia variation de I'integrale

(3. 6)

pour des extremites non fixes.

Si Ies variations sont a extremites fixes, bxo = bx~ 0, il vient

(3 7\

Pour que S est extremum, il faut et il suffit que P 0, c 'est-a-di

re

equat ions formellement identiques a (1. 1). D 'ou

Theoreme. - Dans tout mouvement d 'un fluide parfait isentropique,
Ies lignes de courant sont Iocalement des Iignes orientees dans Ie
temps extremales de I'integrale (3.6) pour des variations a extremi-
tes fixes.
Introduisons la met rique conforme g f2 g . On a

(3.8)
 ~ 60 -

Pham

Vis a vis de cette metrique, l'arc de courbe est defini par ds = fds,
de sorte que les lignes de courant sont definies comme extl'e nales de

(3.9)

Ces extre::nales sont des geodesiques de (V 4' g ). En posant

(3.10)

on voit que ell = fUll et

-II
C -1 fJl
= f u , de sorte que ~,
-11-' = 1
C C
et ces geodesiques ont pour equations

-II -
(3.11) C VA C~ = 0

Corollaire. - Lesllignes de courant du fluide parfait isentropique

sont geodesiques orientees dans Ie temps de (V, g).

4. L'invariant integral de l'hydrodynamique.

Considerons un mouvement du fluide, defini par example par un pr~

bleme de Cauchy. Soit " un tube de courant s'appuyant sur un cycle

ro de dimension trace sur l'hypersurface initiale 1: (non tangente aux
lignes de courant) et soit G un cycle trace sur ~, homotope a roo
Chaque ligne de courant de ~ est limite en Xo ~ ro et x4 ~ r.
Nous pouvons appliquer la formule (3.5) a chacune de ces lignes de
courant , P = 0 , la variation totale de S est nulle quand Xo de-
crit Ie cycle r. ' il vient
(4.2)
 - 61 -

Pham

La 1-forme (/J a pour expression:

(4.2)

la propriete (4. I) se traduit par l'enonce suivant qui generalise un

theoreme classique sur la conservation de la circulation.
Theoreme. - Etant donne un cycle r a une dimension non tangent
aux lignes de courant, la circulation du vecteur courant Ccc Ie long
de r reste invariante quand r se deforme sur Ie tube de courant
defini par r
Si D est une variete differentiable a 2 dimensions, a bord DD,
la formule de Stokes donne

(4.3)

L'integrale de la forme

n dW

sur la sous-variete D se conserve quand elle se deforme de manie-

re que chaque point reste sur la m~me ligne de courant.
Dans Ie langage de H. Poincare, 12 definit un invariant integral
pour Ie systeme differentiel aux lignes de courant

dx
dS
Gl
- u·
et fU definit un invariant integral relatif. La 2-forme .n joue un rOle
fondamental dans la description du mouvement. Elle admet l'expression
 - 62 -

Pham

locale

(4.4)

TMoreme. - La 2-forme nest une forme invariante pour Ie syste-

me differentiel aux lignes de courant i. e.

(4. 5) ~cll = 0

£c est la derivee de Lie suivant C.

En effet on a en utilisant l'identite du calcul des variations

comme n dW, dn 0, il reste die.n or

C"n.,
En introduisant la connexion riemannienne associee a la metrique con-
forme g, on obtient

-111- -
Comme C(I. est unitaire, C V~ CC( = 0 et d' autre part est un
champ geodesiique d'apres (3. 11) , on a bien i/l. = o.
La 2-forme 12 est une forme invariante par Ie systeme differentiel
aux lignes de courant. Nous allons rechercher tous les systemes diffe
 - r3 -

Pham

rentiels qui la laissent invariante. Il nous faut determiner tms les

champs de vecteur X tels que

L' existence de X depend du rang du systeme precedent, comme

!l.~ est antisymetrique, il vient :
1. - si n est de rang 2 , en chaqu'? point x les vecteurs carac
teristiques forment un 2-plan 'It'll:' Le champ de 2-plan 1T
admet des varietes integrales de dimension 2 engendries
par les lignes de courant.
2. - si 11 est de rang O,.n = O. Comme.n = da>, (J) est une
I-forme ferme, il existe une fonction f telle que W = d +
Par suite Car = ~, : les lignes de courant sont trajectoires

orthogonales a la famille d 'hypersurfaces ~ = cont.

Ces resultats sont importants pour I 'etude des mouvements rotation
nels et irrotationnels du fluide.

tt,l\IOUVEMENTS ROTATIONNELSET IRROTATIONNEI,s

5. Tenseur tourbillon et equations de Helmholtz.

Definition. - On appelle tenseur tourbillon Ie tenseur a llisymetrique
d'ordre 2 defini par la 2-forme invariante [l
Il constitue la veritable extension relativiste du rotationnel des vi-
tesses introduit en mecanique classique. Si I 'on se rappelle I 'expres-
-2
ou e est la vitesse de la lu -
-1-2
sion de f = 1 + te + pr ,e
 - 64 -

Pham

miere, on voit que CCII = fUg( differe de uCl( par des termes en e-2
o
.0•• = 0« C~ -,,~ Cal differe de .a«~ = ~OI u. -U, UO( par des termes
-2
en C
TMoreme. - Le tenseur tombillon satisfait aux equations de Helm-
holtz

(5. 1)

E;n effet un caIcul simple montre que ces equations sont une conse-
quence de I 'equation 1/2 = 0 qui exprime que n. est une forme
invariante. On fait Ie caIcul en metrique initiale (V, g).
Definition. - On dit qu 'un mouvement du fluide est rotationnel si
.n f 0 et irrJtationnel si !l = o.
Theoreme. - Pour qu 'un mouvement du fluide parfait isentropique
soit irrotation:aH, il faut et il suffit que les lignes de courant soient
orthogonales a une m~me hypersurface (locale).
En effet soit Z une hypersurface orientee dans I 'espace telle que
ntC~ = 0 sur I . On peut choisir des coordonnees locales telles que
!. soit representee par x O= 0 et que Ies lignes de courant par
x~ = const.. (coordonnees de Gauss). Les equations de Helmholtz mon
trent alors que 10.0 11 , = O. 11 en resulte que l1 Gl , = 0 au voisinage
de 1. .

6. Vecteur tourbillon.
On suppose que Ie mouvement est irrotationnel. Au point x 6 'Ill e-
tudions Ie 2-plan ltlt ferme des vecteurs x oC
caracteristiques i. e.
tels que

(6. 1)
 -65-

Pham

On aurait deja dans TI'jt Ie vecteur ull tangent a la ligne de courant

passant par x . Pour achever de determiner liz , il nous suffit
de rechercher un second vecteur non colineaire a uc(. Nous choi -

sissons un tel vecteur 9 orthogonal au premier. Ce vecteur est

defini par les equations

(6. 2) o

Le vecteur 9' n'est defini qu'a un facteur pres, on a par un cal-

cuI algebrique

(6. 3)

ou ~Il'r" est la forme element de volume riemannien de (V, g).

On remarque que ea= 0 entraine nIl' = o.
Definition. - Au vecteur Gel defini par (6.3) on donne Ie nom de
vecteur tourbillon, a ses trajectoires Ie nom de lignes de tourbillon.
D 'apres la definition les lignes de tourbillon sont orthogonales aux
lignes de courant. D I autre part Ie systeme differentiel aux lignes de
tourbillon

dx Cl =
dt
e'
admet la 2-forme.n. comme forme invariante. On en deduit imme-
diatement les proprietes suivantes:
Theoreme. - Etant donne un cycle r a une dimension non tangent
aux lignes de tourbillon, la circulation du vecteur tourbillon Ie long
de r reste invariante quand on deforme r sur Ie tube de tourbillons
 -66-

Pham

defini par r, ,
Soit " un tube de lignes de courant, r et r deux cycles hom~
topes sur " ' Chacun de ces cycles definit un tube de tourbillons,
soit e et 6)', Soit r4 un cycle sur e homoiope a r , Alors,
Ie tube de courant passant par ~ coupe Ie tube de tourbillons ®
suivant un cycle rj' homol 'Ope a r' , Comme OJ est un invariant
integral relatif pour les lignes de courant et aussi pour les lignes de
tourbillons, on 3.

Cette propriete constitue la generalisation relativiste d 'un theoreme

de Helmholtz en dynamique classique,
Enfin Ie champ de 2-plans x-+ 1l':Jt defini par Ie systeme ca-
racteristique de la forme fl

est un champ completement integrable, Aux varietes integrales a deux

dimensions WI on donne Ie nom de varietes caracteristiques de.£l ,
Ces varietes peuvent Nre engendrees par dE'S lignes de courant et par
des lignes de tourbillon qui sont orthogonales sur Wt ' Si donc on
mene les lignes de courant passant par les points d 'un ligne de tour-
billon, les trajectoires orthogonales de ces lignes de courant sur W,
sont lignes de tOllrbillon, Cela veut dire que si une ligne fluide est de
I) 'rbillon a un instant, elle reste de tourbillon a tout instant,
 - 67 -

Pham

• ~ • MOUVEMENTS PERMANENTS

7, :Espaee-temps stationnaire,
On dit qu 'un espace-temps (V 4' g) est stationnaire s'il existe un
groupe connexe d'isometri~s globales a un parametre ne laissant in-
variant aucun point de V4' a trajectoires orientees dans Ie temps
et tel que
1} chaque trajectoire ~ est homeJmorphe a R
2} il existe une variete differentiable V3 a trois dimensions et
un diffeomorphisme V4 .... V3 It R appliquant les trajectoires % sur
la droite facteur R
V4 apparait comme une variete fibree triviale de base V3 de fi
bre type R' Les fibres sont des trajectoires d'isometries, On les
appelle les lignes de temps, On appelle espace la variete de base V3'
Celle-ci est diffeomorphe a la variete quotient de V4 par la relation
d 'equivalence definie par Ie groupe d'isometries,
Si t est Ie vecteur generateur infinitesimal du groupe d'isome -
tries il satisfal.t aux equations de Killing

(7. 1)

II resulte de la definition qu'il existe des systemes de coordonnees

locales (x', )} tel que les x~ soient un systeme de coordonnees 10
cales sur V3 et qqe xO definisse les points sur les trajectoires de
I' de sorte que les sections d'espace xO = const sont globalement
definies et Eiiffeomorphes a V3' On dira que ces coordonnees locales
 - f8 -

Pham

(x', )) sont localement adaptees au groupe d'isometries si Ie gene-

rateur infinitesimal ~ y admet les composantes contra rariantes

.
(7. 2) ~o ,~ " 0

Si g;.~ sont les composantes du tenseur metrique dans ces systemes

de coordonnees, les composantes covariantes de t sont

Les equations de Killing (7. 1) se traduisent comme

spit

Ainsi dans les coordonnees adaptees les gr, sont independantes de

xO
En ctecomposant la forme metrique suivant la variabl€ directrice
x' on a

(7.3) g

oli

(7.4)
 - 69 -

Pham

definit une metrique definie negative sur les sections d'espace. EI-
Ie est invariante par tout changement de systeme de coordonnees a -
daptee de la forme

A
On munira V3 de cette metrique g

8. Mouvement permanent.
On dit que Ie mouvement du fluide parfait isentropique est perma -
nent si I' espace -temps stationnaire et si Ie groupe d'isometries laisse
invariants l'indice f et Ie vecteur vitesse unitaire i. e.

(8. 1) o

Si les coordonnees sont adaptees ces conditions (8. 1) se traduisent

par

(8. 2) a, f 0 ~D Ull( 0

Theoreme. - Pour que Ie mouvement du fluide parfait isentropique

soit permanent, il faut et il suffit que 1'espace-temps soit stationnai-
reo
Choisissons des coordonnees locales adaptees (xo, x") et soit L
I 'hyper surface d 'equation x 0 = 0 . Zest orientee dans Ie temps. 11
resulte du probleme de Cauchy que sur 1. et sur les hypersurfaces
voisines xO = const , on a (chp II, 5, (5.9) )
 - 70 -

Pham

En coordonn~es adaptees, ~. S~ = 0 , ~o gGl~ = 0 . Il en resulte qu 'en

derivant par rapport a x 0 on obtient compte tenu de I 'equation ther-
modynamique dp = rdf - 9d S (d S = 0)

soit en tenant compte du Ch. II, 5, (5.7)

Ou en deduit "a.f = 0 sur 1.. , puis ~o u~ O. Le mouvement

est donc permanent.
Theoreme. - Dans tout mouvement permanent du fluide, la fonction
scalaire

(8. 3)

conserve une valeur constante Ie long de chaque ligne de courant.

II nous suffit de montrer que i c ( Lt (c) ) = 0 ou W = Cel dx el
Or on a en utilisant I 'identit~ du ca1cul des variations
 - 71 -

Pham

Or il est manifeste que J,'" = 0 ce qui veut dir que Ie systeme di.!
ferentiel aux lignes de courant admet la transformee infinitesimale J'
on en cteduit egalement Jt II) = 0 . II vient

Remarque.- Le systeme differentiel aux lignes de courant admet

la forme invariante n. et la transformee infinitesimale I Comme
t,6 = 0 , on voit que Ie systeme differentiel aux lignes de tourbil-

Ion possede la m~me propriete. On en cteduit que H = Col til( est e-

galement constant Ie long des lignes de tourbillons. H est donc cons
tant sur chaque variete caracteristique W, de II .
On a

(8.4)

formule qui rend les resultats precedents evidents .

•
9. Le theoreme de Bernouil1i;..
Introduisons la grandeur d 'espace du vecteur uel relativement a la
direction de temps t . SoH

2
- v
 - 72 -

Pham

En vertu du caract ere unitaire de u, on a

d'ou

2 2
(9. 1) (uo) g,o(1+v)

L'integrale premiere H a pour valeur en coordonnees adaptees

Co = fuo . On en deduit

En posant U = goo ' i1 vient

Theoreme. - Le mouvement permanent d'un flidde parfait isentropi-
que satisfaits Ie long de chaque ligne de courant a

2 2
(9.2) f U (1 + v ) = const .

ou U est Ie potentiel principal de gravitation.

Ce theoreme generalise Ie tMoreme de Bernouilli,. En effet de l'e
quation thermodynamique, i1 vient

.1
On en deduit aux termes en C pres

+ v2 + fP dp
r
const.
Po
 - 13 -

Pham

H. PROJECTIONS DANS L'ESPACE

la, Un probleme du calcul des variations,

On se propose d 'etudier Ie mouvement permanent dans l'espace V 3'

Pour cela il nous faut etudier les projections des geodesiques de
(\' "
sur I' espace quotient (V 3 g).
4
Un tel probleme a ete resolu dans Ie cas plus general d 'une varie-
te fuislerienne (V"+ l ,t) Mfinie par une variete differentiable VM~
munie d 'une fonction ~ (x, X) positivement homogene de degre
sur Ie fibre des direetions D(V""+4). On supposera que (V"~4 ' :

admet un groupe connexe d'isometries globales definies par un champ

de vecteurs ~ tel que

On rapportera (VIM I ,£) a des coordonnees locales (x ~ '. xo)

adaptees a son groupe d'isometries et on designera par V
n
la varie-
te quotient,
L"e systeme differentiel aux ext) emales de .t admet l'invariant inte-
gral relatif

(10,1) w

, £ ne depend pas de xO , 'ot = a , on a l'integrale

premiere provenant de l'equation d 'Euler en x·

(10,2) ,.[
o h
 - 74 -

Pham

de sorte que 60£dxo = Ie dx O constitue un invariant integral pour la

famille (E k ) des extremales correspondant a la valeur h. Il en re
sulte que

(10.3)

est un invariant integral relatif pour la famille (E h ).

Si 'oo! fa, on peu t resoU.dre (10.2) par rapport a xII, soit

(10.4) xo= ~ (x',. xJ , h)

ou , est une fonction homogEme de degre 1 en

. D'autre part en
xJ .
vertu de I 'homogeneite de £ on a

Par suite la forme * = x~ 'a~£ peut s 'exprimer par ime fonction L

des variables x~, xj , k , soit

(10.5)

et lion a

On a demontre Ie theoreme.
TMoreme. - Les projections sur Vn des extremales (E h ) pour
une valeur h donne sont les extl'emales de la function L. Elles
 - 75 -

Pham

sont dHinies pour un systeme differentiel qui admet l'invariant inte-

gral relatif

11. Cas d 'une metrique riemannienne.

Considerons Ie cas OU la fonction ! est definie par

t4,~ = 0,1, ... , n. On suppose que go. r o.

Le procede de descente conduit a former I 'equation

(11. 1)

et a eliminer ~, entre cette equation et

(11. 2) L l - h~'

L 'elimination donne

g . x~
(11. 3) L + h ....!L-
gIl

o, on a

(11. 4) + g ..
~J

On supposera go~
.r0 .
x~ Le procede de descente conduit a eli-
 - 76 -

Pham

miner x' entre (11. 2) et les relations

L = £- h x G

L 'elimination donne

(11. 5) L

Application aux mouvements permanents. - II suffit de remplacer

2
dans les formules precedents gClf, par f glt, et on obtient la fonc-
tion L dont les extremales donnent Ie mouvement dans l'espace. II
y a un seul cas car goo f O. On obtient ainsi:

(11. 6) L f 2g.. x' ~ x';

~J

II serait interessant de developper les calculs.

12. Projection des geodesiques de longueur nulle.

On les considere comme limites des geodesiques orientees dans Ie
temps. Dans notre probleme

en d"nvan
.t.· t par rappor t d>. x' • , on a h ...r go. x ." , ce qui montre
 - 77 -

Pham

que h .... oIJ lorsque .r - 0 , h garde Ie signe de gOl! xII. . Les

extremales de L coincident avec celles de L/k..
Par suite les extremales cherchees qui definissent les projections
des geodesiques isotropes de (V 4' g) sont les extremales de la fonc
tion

(12.1) lim
h+c»

1er cas goo f 0 Le passage a la limite donne

(12.2)
A
II =H - -
I 1"· ~ .j
g ij x x
goo

ou ~I est Ie signe de gOIl x· et t Ie signe de goo' puis

(12. 3)

2e cas o . Le passage a la limite donne

(12.4) L

(12.5) x'= -

Nous appliquons ces resultats a I 'etude du principe de Fermat.

13. Le principe de Fermat.

On sait que les rayons lumineux [16] sont geodesiques isotropes de
 - 78 -

Pham

la variete riemannienne (V 4' g) definie par l'espace-temps V4 mu

ni de la metrique.

(13.1)

Supposons que Ie mouvement est permanent. Si 1 et}k sont constants,

Ie groupe d'isometries de (V 4' g) induit un groupe d'isometries sur
(V4 ,g)

On choisira des coordonnees adaptees.

Mais alors que les trajectoires d'isometries de (V 4' g) sont orien-
tees dans Ie temps, les trajectoires d'isometries induites sur (V4' g)
peuvent ~tre orientees dans Ie temps, dans l'.espace or isotropes . En
effet si t est Ie generateur infinitesimal du groupe d'isometries de
(V4' g) on a pour les composantes contravariantes

1 o

Ie carre de ce vecteur a pour valeur

2
(13.2) gllO - (1 - w ) u, u,

2 1
ou w est Ie carre de la vitesse de propagation de la lumiere
-;;;:
dans Ie fluide.
Si nous introduisons la grandeur d'espace du vecteur vitesse unitaire
ulll relativement a la direction de temps ~, .,1 = - g~j u~ u j , on a
vu que

2 2
(u.) t: g •• (1 + v )
 - 79 -

Pham

En portant cette valeur dans (13. 2 b, il vient

2 2 2 2
(13. 3) goo = goo (v w + W - v )

-
goo peut changer de signe.
En appliquant les formules du paragraphe precedent, on obtient Ie
theoreme suivant qui donne la loi de propagation de la lumiere dans
I 'espace.
Theoreme. - Si Ie mouvement du fluide est permanent et tel que
g., to, les rayons lumineux dans I' espace sont les extremales de
l'integrale

go~ x·
f'[E£1
}o
jXI
(13.4)
" du
x,
goo
glj x~ ~j - -..---
g,. ] du
Xo
dx~
ou x~ pour des variations a extremites fixes dans V 3 . Le
du '
temps mis par un rayon pour aller du point Xo au point x 4 est don
ne par

(13.5)

II est extremum .
Dans Ie cas g.o = 0 , on a

(13. 6) JXIA
Xo
du = f4Xo
gO
2g.~
x x
. ~ •j

*' du

-.
f' f' xj
~

(13.7) dt
=
- gil x
2g,:, xl du
x. Xo
 - 80 -

Pham

II est clair que les resultats ne dependent pas de la variable auxi

liaire u. D'autre part si l'espace-temps est statique orthogonal et
si les lignes de courant coincident avec les lignes de temps, on a la
metrique d 'univers

et la metrique associee

ou n2 ~f. On peut alors mettre (13.5) sous la forme

dr

est I 'element lineaire de (V 3' g) . Dans Ie

cas d 'un espac e temps plat U" 1 , Ie tMoreme precedent se tra -
duit par

C 'est I 'enonce du principe de Fermat en optique classique. Ce tM£

reme que nous avons demontre constitue I 'enonce du principe de Fer-
mat en relativite generale , dans Ie cas OU Ie fluide est en mouve -
ment. Par ce theoreme se trouve egalement demontree I 'equivalence
entre Ie principe d'action et Ie principe du moindre temps.
 - 81 -

Pham

14. Application: loi relativiste de la composition des vitesses.

Pla<;ons nous dans I' espace temps de Minkowski rapporte a des
coordonnees orthonormales. uel est Ie vecteur vitesse unitaire d'u-
rivers clont les composantes sont determinees classiquement a par-
•
tir de la vitesse d'espace ~ si , est prise comme unite. Un
calcul facile donne la metriql'le associee que nous ecrivons sous la
forme

(14. 1)

1 - V2 2
+ --2 (~;dx~)
1 - ~

Cette metrique est du type hyperbolique normal. II est a noter un

2 112
changement de l'ordre dans la signature au passage de V = I' . On
Ie met en evidence en choisissant l'axe } parallele a .., (vitesse
du fluide), On a ainsi

(14.2)

que l'on peut mettre sous la forme canonique par une decomposition en
carres . Si V2 f ~2 , on obtient

1-~ 2 [ V-~
2 2 D
- 22 --2- dx +
2
(1-V)~
2 dx
I J' -
2 2
(1-~)V
V -f 1-' 1_~ V 2 _ ~2
 - 82 -

Pham

2
et I 'on voit que pour V ~, rR2 on a la signature + - - - et pour
2 2
V < ~ la signature - + - - . Pour V2= rII. 2., on obtient

qui reste de signature + - - -

A partir de la metrique associee (14. 1) cherchons a exprimer
Ie theoreme en prenant Harc ~ du rayon comme parametre. On a a
remplacer dans (13.5) x· par ~~
AI = dx
\
~I d~ ov d.1 = (dx)
4 2
+ (dx 2):2 +
2
+ (dx') . On en tire ainsi:

dt
Vi - H
I O_V2) ( ,d)
d v2.. ~a
S1· V2 _,,2./.
r T 0 ,cette re IatlOn
' d onne

/1. 2 W 2 2 ~ 2
1-"12 -(1-r) (1- V ) (1- W fi A) 0

Si on interprete
..
V comme vitesse absolue et
~
W comme vitesse reJafuB
de propagation de la lumiere , on a manifestement

(14. 3)
-2 ....P + 2W·. .;. ...P + (W.... . -~ )- - W11'1]
W +
2. ....
P

On verifie par un calcul direct que cette relation reste valable dans Ie
cas V 2= ~ 2 . C 'est la formule relativiste de la composition des vites
ses. Il est aise de verifier qu 'on peut la mettre sous la forme
 - 83 -

Pham

Nous obtenons ainsi a partir du principe de Fermat une demonstra -

tion de la loi relativiste de composition des vitesses.
 - 84 -

Pham

BIBLlOG RAP HIE

M. ABRAHAM - R.C. Circ. Mat. Palermo 22 (1909), 27-35

N. L. BALAZS - The propagation of light rays in moving media.
Jour. Optical Soc. Amer. 45 (1955)
[3] C. CATTANRO - C. R. Ac. Sc. Paris, 247 (1958), 431-433

[4] Y. CHOQUET-BRUHAT - Fluides relativistes de conductivite in

finie, Astronautica Acta, 6 (1960), 354-365
Y. CHOQUET-BRUHAT - Etude des equations des fluides charges
relativistes inductifs et conducteurs, Comm. Math.
Phys. 3 (1966), 334-357
[6] .. MM. KRANYS - Il Nuovo Cimento, 50B, (1967), 48-63
[7] LANDAU - LIFCHITZ - Fluid mechanics, Pergamon, London (1958)
[8J J. LERA Y - Hyperbolic differential equations - lnst. for Advanced
Studies, Princeton (1953) , notes mimeographines
A. LICHNEROWlCZ - Theories relativistes de la gravitation et de
l'electromagnetisme, Masson, Paris (1955)
A. LlCHNEROWICZ - Relativistk hydrodynamiqBeand magnetohrdr)
dynamics, Benjamin (1967) -
[lil B. MAHJOUB - CR. Ac. Sc. , Paris, 247, (1968), 668-671 et
268A (1969), 1440-14l:2.
C. MARLE - Sur l'etablissement des equations de l'hydrodynamique
des fluides relativistes dissipatifs. Ann. lnst. Henri
Poincare vol. X,n.2, (1969), 127-194
C. ECKART - The thermodynamics of irreversible process, Phys.
Rev. 58, (1940)
PHAM MAU QUAN - Sur une theorie relativiste des fluides thermo-
dynamiques, Ann. Mat.. Pura e appl. (4), 38 (1955)
PHAM MAU QUAN - Etude electromagnetique et thermodynamique
d'un fluide relativiste charge, J. Mech. Anal. 5
(1956), 473-538
 - 85 -

Pham

[i~ . PRAM MAU QUAN - Inductions electromagnetiques en relativite

generale et principe de Fermat, Arch. Rat. Mech.
Anal., 1, (1957)
PRAM MAU QUAN - Thermodynamique d'un :t1w.ide relativiste,
Boll. U.M.I. (1960) (3) vol 15, 105-118.
PRAM MAU QUAN - C. R. Ac. Sc. , Paris, 261 (1965) 3049-3052
PRAM MAU QUAN - Magnetohydrodynamique relativiste, Ann. Inst.
Renri Poincare 2 (1965), 21-85
PRAM MAU QUAN - Sur les equations des fluides charges induc-
tifs en Relativite generale - Rendiconti di Matema
tica, 1. 2; vol. 2 Serie VI (1969) -
G. PICRON - F,tude relativiste des fluides visqueux et charges,
.
Ann .. lnst. Henri Poincare, 2 (1965) 21-85
AH. TAUB - Relativistic hydrodynamics, Arch. Rat. Mech. Anal.
3 (1959) et in Lectures in Applied Mathematics,
8 (1967)
 CENTRO INTERW\ZIONALE MATEMATICO ESTIVO

(C. 1. M. E. )

ONDES DES CHOC, ONDES INFINITESIMALES ET RAYONS.

EN HYDRODYNAMIQUE ET MAGNETOHYDRODYNAMIQUE RELATIVISTES

A. LICHNEROWICZ

Corso tenuto a Bressanone dal 7 al 16 Giugno 1970

 ONDES DES CHOC, ONDES INFINITESIMALES ET RAYONS
EN HYDRODYNAMIQUE ET MAGNETOHYDRODYNAMIQUE RELATIVISTES

Introduction.
On sait 1'importance mathematique et physique prise recemment par la
magnetohydrodynamique relativiste. Dans ce cadre, on s'est propose d'e-
tudier en ctetailles ondes infinitesimales et les ondes de choc. L 'extension
relativiste des conditions de compressibilite de Hermann Weil joue dans la
theorie developpee un rOle important.
On a introduit systematiquement un instrument mathematique commode
(les tenseurs-distributions) par 1'analyse des differentes ondes. Cet instru-
ment est d'abord applique en hydrodynamique, a l'etude des ondes soniques
et des rayons correspondants.
En passant au cadre de la magnetohydrodynamique, on etudie ensuite
successivement Ie systeme differentiel fondamental, la structure des ondes
magnetosoniques et des ondes d 'Alfven, les rayons correspondants, enfin les
ondes de choc qui font 1'objet d 'une analyse detaillee. On etablit en particulier
grs.ce a 1'introduction d 'une fonction d 'Hugoniilt convenable, un important
theoreme d 'existence et d 'unicite par les solutions non triviales des equa-
tions de choc.
En traitant les ondes magnetosoniques par Ie formalisme des ondes de
choc, on met en evidence Ia non-invariance de la direction des rayons par
I 'operateur de disci.ntinuite infinitesimale.
Ces le90ns qui se suffisent a elles-m~mes represent une synthese de
certains de mes travaux durant la periode 1966-69
- Tenseurs-distributions
II - Hydrodynamique relativiste
III - Les equations de la magnetohydrodynamique relativiste
 - 90 -

Lichnerowicz

IV - Ondes de choc en magnetohydrodynamique

V - Fonction d 'Hugoniot et orientation des ondes de choc
VI - Ondes de choc et ondes d 'Alfven
VII - Vitesses des ondes de choc et theoremes fondamentaux
VIII - Retour aux rayons magnetosoniques.
 - 91 -

Lichnerowicz

1. TENSEURS-DISTRIBUTIONS ET DISCONTINUITES

I, Tenseurs-distributions sur une variete riemannienne,

a) Soit V une variete differentiable orientee, de dimension n et
classe ch+1n(h~0);
nous disposons sur V d'une metrique riemannienne
2 n h
ds de signature arbitraire et de classe cC (O~ k~h). Localement:

2 r::I. ()
ds =g dx dx P (01, (3 =0, 1"", n-l),
Cl(j

Si T estU sont deux p-tenseurs, nous notons (T, U) Ie produit scalaire

x
de T et U au point x de V, En coordonnees locales:
n

0/1. , , CXfi'
(T, U) = T (x)U (x)
x ()( I' , 'O(p

h
SoH D(p, V ) l'espace des p-tenseurs
n
a support compact de classe C sur
V, Si T est un p-tenseur localement sommable arbitraire, nous pouvons
n
pos~r pour U E D(p, V ):
n

(1. 1)

ou y est l'element de volume riemannien de la variete, Localement:

(1. 2) 1J,='VAr.
jgl
0
dx ;\ .. , ;\ dx
n-1
,

Un p-tenseur-distribution T de Vest une forme lineaire continue,

n
a va-
leur scalaires, sur l'espace D(p, V l. Continu est ici entendu au sens
n-
usuel en tMorie des distributions, Si U E D(p, V ), nous designons par 1f[UJ
n
 - 92 -

Lichnerowicz

ou (T, U> la valeur pour U du tenseur-distribution T,

Un p-tenseur ordinaire localement sommable T de V peut Nre
n
identifie avec un tenseur-distribution au moyen de la formule (1. 1). Le
tenseur-distribution est note T D , ou parfois T par abus de notation, quand
aucune confusion n 'est possible,
b) Si T est un scalaire-distribution et V un p-tenseyr ordinaire, TV
est Ie p-tenseur distribution defini naturellement par:

TV[UJ = T [(V, U)]

Cela pose, soit.f2. Ie domaine d'un systeme de coordonnees locales (xo()

et donnons-nous dans (l, nP scalaires distributions T L'expression
0(1' , 'O<p
0<'1 ~
(1. 3) T=T dx ®," ® dx P
"'1"
'" '0(
P
0<
definit un p-tenseur-distribution: si (eo(.) est Ie repere dual du corepere dx :

0( l' , ,0<
U=U P eo(1 ®, , .8> eo<p

et T[U] est donne par la somme:

Inversement tout p-tenseur distribution T dans.n. peut Nre rapresente

par une telle expression: designons par T les scalaires-distributions
0(1' , ,O(p
definis par:
 Lichnerowicz

To( (f) = T [f eO( ® .. , ® ~ ] f€D(O,fi)

I' "o(p 1 P

On a:

oll' , ,0(
=T U P
o(l'''~

et T admet bien l'expression (1. 3), Ainsi, dans Ie domaine d'un systeme
de coordonnees locales, tout p-tenseur-distribution peut ~tre, comme un p-
tenseur ordinaire, rapporte aces coordonnees, les composantes etant des
scalmrres-distributions, Nous supposons maintenant h, k ~2,
c) Soit T un p-tenseur ordinaire, \7 T sa derivee covariante dans la
connexion riemannienne, Si U e D (p+l, V )
n

<VT ,U) = !vV

n
V0 T
J 0(1" ,0(
p
PO<"I' , ,c(,
U p "l

soit, par integration par parties:

<'VT,U)

Le premier terme due sec-ond membre est nul. On introduit ainsi I 'operateur
~ de coderivation sur les (p+1)-tenseurs, definispar:
 - 94 -

Lichnerowicz

Ainsi la formule precedente s 'ecrit:

(1.4) .( V"T , U> = <T, QU >

Cela pose, on definit naturellement la derivee covariente d'un p-tenseur

distribution T comme Ie (p+1)-tenseur distribution "T determine par
la relation (1. 4) ou U E D(p, V ). Il est aise de voir que toutes les pro-
n
prietes c1assiques de la derivee covariante dans une connexion riemannienne,
ainsi que les formules correspondantes, demeurent valables pour les tenseurs·
distributions.

2. Les distributions -/, Y- et D relatives a une shypersurface.

a) Nous considerons maintenant exc1usivement un domaine.o. de V
n
correspondant a des coordonnees locales. Soit L une hypersurface reguliere
definie par l'equation locale f =0 ('P de classe C 2) qui partage D- en
deux domaines no et n 1 correspondant respectivement a cp.( 0 et
c:P> O. Nous notons par 1/0 Ie gradient de <f'1/.
Soit Yf, ou plus brievement yO (resp Y~ ou (;) la fonction surD.
qui vaut 1 (resp 0) dans Q0 et 0 (resp 1) dans n;: ces fonctions de-
finissent dans n des distributions designees par les m~mes notations,
par l'intermediaire des formules:

(2. 1)

b) Considerons la c1asse des (n-l)-formes coverifiant la relation:

(2.2)
 - 95 -

Liehnerowiez

Si (J et Wi- sont deux formes de eel te, classe, il existe une (n- 2)-forme

).I. telle que W'= c..J +dfAf. Soit 'dno et d.o lIes bords orientes sur L
de .n 0 et 11 1 (d U o =- d .n1). D'apres la remarque preeedente, l'in-
tegrale:

a une valeur bien cteterminee, independente db ehoix de W flans la elas-

se envisagee. Nous pouvons ainsi definir un sealaire-distribution
ou plus bir.evement l , par la relation:

(2.3) <d, f) f € D(O,n.)

d<p est la mesure de Dirac relative a q> ; son support est porte par L.
c) Proposons-nous d 'evaluer les tenseurs-distributions derives des sea-
laires-distributions yO et yl. A eet effet, introduisons dans Il un sy-
0
steme de eoordonnees
0(
(y ) tel que y = f; dans ee systeme 1 a pour eom-
posantes 1'=1,1.=0 (i,j =1,2, ... ,n-l). Del'expression de ty) on cteduit
O I L
que la (n-l)-forme

n-l
w {,;j dy~ ... 1\ dy

verifie (2.2). Si UE D(1,n)' on a:

soit
 - 96 -

Lichnerowicz

soit encore, d I apres 1a formule de Stockes:

<r7Vy1, U > = - fo
~nl
uo,W
'V Igi dy 1A ... I\.dy n-l =..i anI 10< Uc;I.W =(10,
"(
U )-

o
En raisonnant sur Y , on obtient ainsi:

(2.4)

On verifie de m~me qu 'il existe un sca1aire-distribution d I de support

L tel que:

(2. 5)

3. Tenseurs-discontinuites a la traversee d'une hypersurface.

a) Considerons un p-tenseur T sur n. satisfaisant les hypotheses
suivantes:
AI) Sur chacun des domaines n 0 et n l' Ie tenseur T est un
tenseur ordinaire de classe C 1.
A2) Quand q> tend vers zero par valeurs negatives (resp. positives),
T et VT convergent uniformement vers des fonctions a valeurs tensorielles
dMinies sur L etnotees To' (VT)o (resp. T 1,(V'T)l L
Nous introduisons les tenseurs-discontinuites sur L :
 - 97 -

Lichnerowicz

[TJ =T - T
1 0

Dans n , se trouvent definis de maniere naturelle Ies tenseurs distribu-

tions yOT, Y°V'T, ylT, y\7 T. Si TD est Ie tenseur-distribu.tion defini
par Ie tenseur T defini presque partout dans n , on a en termes de
distributions:

Le tenseur-distribution V T D, cterivee au sens des distributions de T D,

s 'ecrit:

avec:

o - 0
V(Y T)=- Ih®T +y V T
o

On en deduit, compte tenu de (3. 1):

(3.2)

ou (V T)D est Ie tenseur distribution defini par Ie tenseer ordinaire defini

presque partout " T, derivee covariante usuelle du tenseur T.
b) Etudions Ia cterivee du tenseur-distribution J[T] de n . En
coordonnees (y~), on a:
 - 98 -

Lichnerowicz

OU V.1 ~ =L
• 1
JI =0. D 'apres les_ hypotheses de convergence uniforme
V.1 [T] =['V.TJ
1
. Ainsi l'on a b ["Y.TJ
1
= \7.(
1
~ [TJ). Il en resulte qIT'il
existe un p-tenseur distribution ~ T, a support sur 1.. , tel que l'on
ait la formule:

(3. 3)

c) Nous considerons maintenant des tenseurs satisfaisant toujours aux

hypotheses AI' A2 mais qui sont supposes continus dans n. Ces ten-
seurs definissent d 'une maniere naturelle une algebre tA de tenseurs.
La formule (3.3) devient alors:

(3.4)

Considerons I' application:

J : TE~ --+ ~T

OU ~T est un tenseur-distribution a support sur ~ . On deduit de (3.4)

que l'application 6 est une derivation: si a et b sont deux reels et si

T, ue ~ sont deux p-tenseurs, il resulte de (3.4):

b(aT+bU) = a bT+b ~U

Si T, U ~CA sont respectivement un p-tenseur et un q-tenseur, on a:

 - 99 -

Lichnerowicz

b(T ® U) '" bT0 U + T @ bU

J Test appele la discontinuite infinitesimale de T et ~ l'operateur de

discontinuite infinitesimale.

4. Formule concernant les derivees secondes.

Pla<;ons-nous maintenant dans les hypotheses suivantes:
B l ) Le tenseur Test continu sur n . Sur chacun des domaines n.o
et
--1
n , T est un tenseur de classe C 2
B 2) (,luand c:p tend vers zero par valeur negatives (resp. positives),
VT et 'i1 'V T convergent uniformement vers des fonctions a valeurs ten-
sorielles definies sur.L et notees ('1 T)o ('V''V'T)o (resp. (\l T)l' ('V''V T ) /
Ainsi Ie tenseur vrT satisfait lui-m~me aux hypotheses AI' A2.
De (3.3) applique a vr T, il resulte:

On en deduit d'apres (3.4):

soit:

(4. 1)

La metrique etant C 2, Ie tenseur de courbure est continue a la traversee

de r. , II en resulte d 'apres l'identite de Bianchi
 - 100 -

Lichnerowicz

Comme ~ 1~ = \7~ 10( = Vol \7~'P' on cteduit de (4.1)

soit:

On cteduit de (4.2) qu'il existe un p-tenseur-distribution T a support sur

L tel que:

En substituant dans (4.1), on obtient une formule utile:

II. HYDRODYNAMIQUE RELATIVISTE ET HYPOTHESES DE

COMPRESSIBILITE.

5. Fluide parfait thepmodynamique.

a) Soit V4 un espace- temps muni d 'une metrique hyperbolique ds 2, de
signature +- - -, satisfaisant aux hypotheses usuelles de differentiabilite. En
coordonn6es locales ds 2 =go( p., dx()( dx f3 (0(, f.> =0, 1, 2, 3). Dans un domaine de
 - 101 -

Lichnerowicz

V4' un fluide parfait est decrit par Ie tenseur d'energie:

(5. 1)

ou ~ est la den site d'energie proprl!, p la pression et UO( Ie vecteur-vi-

tesse unitaire du fluide, oriente vers Ie futur; ~ contient une densite de
matiere et une densite d'energie interne. NOlls posons:

2 E.
f=cr(1+2)
c

ou rest la den site de matiere du fluide et E. son energie interne specifiElu

que. Considerons Ie scalaire

2 £ p
o+p = c r (1+-+--)
) 2 2
c c r

Nous posons

= eo +.L = e.. +pV (ou V=l/r)

r

Vest Ie volume specifique et i I' enthalpie specifique. A cette variable

nous substituons la variable thermodynamique equivalente, appelee l'indi-
ce du fluide, defini par:

i
f = 1 +-
c2

Avec ces notations, Ie tenseur d 'energie s 'ecrit:

(f) 2
(5.2) To{ ~ = c rfud. ul?> - pg()( ~
 - 102 -

Lichnerowicz

b) La temperature propre ® du fluide et son eutropie specifique S

peuvent Nre definies, comme en hydrodynamique classique, par la relation
differentielle

~ dS = dE. +~dV = di-Vdp = c 2df-Vdp ( ®;>O)

Ainsi

(5.3)

En relativite la variable thermodynamique Z =fV (volume dynamique) joue

un rOle important et se substitue Ie plus souvent au volume specifique clas-
sique. II est commode d'adopter p et S comme variables thermodynamiques
de bases Nous considerons 't: = t; (p, S) comme une fonction donnee defi-

nissante, par Ie fluide envisage une equation d'etat

c) Le systmne differentiel de l'hydrodynamique relativiste est fourni
par les considerations suivantes: nous supposons d 'abord que la densite de
matiere (qui correspond au nombre specifique de particules) est conserva-
tive. Si \l est l'operateur de derivation covariante:

(5.4)

D'autre part les equations de la dynamique relativiste sont fournies par la

conservation du tenseur d 'energie:

(5. 5)

Nous allons transformer Ie systeme (5 4), (5.5) en un systeme equivalent;

(5.5) s'ecrit explicitement:

(5. 6)
 - 10:3 -

Lichnerowicz

Par produit par u f.>, il vient, compte-tenu due caractere unitaire du

vecteur-vitesse:

(5.7)

soit:

D'apres (5.3) cette relation peut s'ecrire:

Ainsi (5.4) entraine l'equation dite de t10t adiabatique:

(5.8)

En reportant (5.7) dans (5.6) on obtient Ie systeme differentiel aux lignes

de courant:

(5.9)

Le systeme (5.4), (5.5) est equivalent au systeme forme par (5.4), (5.8)
et Ie systeme differentiel (5.9) aux lignes de courant.

6. Vitesse d 'une hypersurface par rapport au t1uide et ondes soniques.

a) Soit L.. une hypersurface reguliere dans un domaine de V4' d'equa-
tion 'f =0 (avec I = d <p). La vitesse v r. de l'hypersurface L.. par rap-
 - 104 -

Lichnerowicz

port au fluide, c'est-a-dire par rapport a la direction temporelle u, est

donnee c1assiquement par la formule:

(v E )2 l.
(6.1) - 2 - =y avec
c

I..
On voit que, quel que soit l~ y est positif et que l'on a:

b) Dans un domaine n ou les variables thermodynamiques p, S et Ie

vecteur-vitesse u sont continues, supposons les derivees premieres discon-
tinues a la traversee de L. , de fa<;on que p, S, uo( verifient les hypothe-
ses du § 3, c. D'appes ~3. 4), nous posons dans la suite:

Etudions a quelle condition l'une au moins des distributions bp, ~S, bu ~

est non nulle.
La relation (5.4) peut s'ecrire, compte-tenu de (5.8),

(6.2) r \l. uel. +r' ucla p =0 (ou r=r(p, S))

01. p b(

De 't' =fV, an deduit en derivant en paS constant:

't" I = f' V + f V'

P P P
 - 105 -

Lichnerowicz

De (5. 3) on deduit:

Nous introduisons dans la suite la quantite y definie, a. partir de l'e-

quation d'etat, par la relation:

(6.3)

2
de telle sorte que l' =c fr~. La relation (6.2) peut s'ecrire:

(6.2')

En ecrivant cette relation de part et d 'autre de L et retranchant, il vient,

apres produit par b:

ce qui s'ecrit:

(6.4)

On deduit de m~me de la relation (5.8):

(6.5)

Enfin Ie systeme (5.9) donne:

(6.6)
 - 106 -

Lichnerowicz

Pour uc( 10(=0, 68 peut ~tre non-nulle. On obtient ainsi les hypersurfaces
engendrees par des lignes de courant, dites ondes d'eutropie (ou de ma-
tiere). Leur vitesse par rapport au fluide est nulle.
8upposons uo(lOl./o; on a b 8=0 de (6.6) il r~sulte que si ~p=O, on
a bu~ =0. Nous sommes ainsi conduits d'apres (6.4) a multiplier' (6.6)

par 1~. II vient

soit d'apres (6.4):

Ainsi si J prO, l'hypersurface L. verifie l'equation:

(6.7)

qui est l'equation aux ondes soniques du fluide; (6.7) exprime que des

ondes, qui sont naturellement des caract~ristiques du systeme differen-

tiel (5.4), (5.5) sont les hypersurfaces tangentes aux cOnes du second
degre d~finis par dualite a partir du tenseur:

8i vest la vitesse des ondes soniques par rapport au fluide, on deduit

de (6. 1) et (6.7)
 - 107 -

Lichnerowicz

Nous postulons dans la suite que v<:c (ou 1>1), ce qui revient a postuler
que les ondes soniques sont orientees dans Ie temps. Pour que v soit
<c, il faut et il suffit d'apres (6.3) que 'r;/ soit..( 0
c) Pour l' > 1, Ie tenseur hol (3 defini~e forme quadratique de type
hyperbolique normal; ho(r.> designe Ie tenseur covariant inverse de hol~.
La generatrice de contact de L., onde sonique, avec Ie cOne est definie
par Ie vecteur:

(6.8)

Soit v ~ la composante tangente aL de la vitesse u (3 du fluide:

Notons que, NI3 etant tangent a :z:., la direction de N~ est celle de v 13 :

(6. 9)

7. Propriete fondamentale des rayons en hydrodynamique.

a) Les bicaracteristiques ou rayons associes aux ondes soniques (et qui
les engendrent) sont les trajectoires du champ sur L des vecteurs N ~ =hClf.;1
0(:
Ce sont des geodesiques isotropes relatives a la metrique hyperbolique nor-

male definie par hol~ .

II resulte de (6.6) que, par une onde sonique 2:

(7. 1)
 - 108 -

Lichnerowicz

s 'exprime simplement en fonction de ~ po Nous nous proposons de

montrer que ~ p (et par suite les JJ ufJ ) se propage Ie long des
rayons, c 'est-a.-dire que bp verifie un sisteme differentiel de la
forme:

Nous postulons dans la suite les hypotheses Bl et B2 du § 4 pour

p, S, u ~o De la formule (403) il r~sulte qu'il existe des distributions
_ - -'A
p, S, u telles que:

(702) ~['Yc(~~p] = \7oll~ ~p+1O(V~ ~P+l~'7ol ~P+lo(l~p

(7 3)
0 ~[\7o(~sJ =lo(l~S

(70 4) ~[~O/ ~ u] = \10( II!- ~ u" +lol 'V~ ~ul). +1~ ~c( bu~ +10( l~ ul).

b) Partons de la relation (602')' divisons la partc 2rf et d~ri­

vons-lao II vient

En ~crivant cette relation de part et d'autre de L et retranchant, on

obtient:
 ~ 109 -

Lichne rowic z

est une combinaison lineaire de termes en !J p et bu'>- et est donc

proportionnel a ~ p. Nous ecrivons (7.5) sous la forme

(7. 6)

ou I'J signifie modulo des termes proportionnels a J p.

En procedant de m~me sur la relation (5.8) relative a 1 'entropie,
on a:

soit d 'apres (7. 3):

Ainsi S est ~ 0 et il vient:

(7.7)

Enfin en se livrant au m~me raisonnement sur (5.9), il vient.

En tenant compte de (7. 6) dans (7. 8) on obtient:

soit d'apres (7.2):

 - 110 -

Lichnerowicz

Or L. ~tant sonique verifie P(I)=O et Ie terme en p disparait. II reste:

soit:

(7.9)

On a Ie theoreme qui traduit la propriete fondamentale des rayons:

Theoreme. Sous les hypotheses du § 4, les discontinuites infinitesi-
males Jp, ~ u').. relatives a une onde sonique L. s1 propagent Ie
long des rayons associes selon les systemes differentiels:

c) Nous avons vu que I 'operateur ~ definit une derivation; la me-

trique et L etant supposees de classe C~ dans iO... , on a:

~ gO/~ = 0

On peut mettre les equ ations fondamentales relatives aux ondes soniques
sous une forme commode.
En ecrivant (5.4) de part et d'autre de f- et retranchant, il vient im-
mediatement
 - 111 -

Lichnerowicz

Ainsi Ie scalaire:

(7. 10) a= rue( 10(

est invariant par Ia derivation ~

De m~me, en ecrivant (5.5) de part et d'autre de L et retranchant,
on obtient

et Ie vecteur,:

(7.11)

est invariant par ~ Cevecteur peut Nre decompose en Ia somme:

(7.12) W
~ =
2 ~ 2 ueXIo(
c afv +(c af - - - p) 1
~
10\ 10(,

et ses composantes tangentielles et normale par rapport a L. sont

invariantes. Ainsi:

~ (fv ~)=o

En particulier la direction du rayon, qui est d'apres (6.9) celIe de

v f.,. est invariante par 1'operateur de discontinuite infinitesimale ~ .
 - 112 -

Lichnerowicz

8. Hypotheses de compressibilite.
a) J'ai ete conduit a adopter par les iluides parfaits relativistes
les hypotheses de compressibilite suivantes portant sur la fonction
"t (p, S), Dans Ie domaine envisage des variables p et S, on suppose
que l'on a:

'(;'p (0

et la condition de convexite:

~ ~2 >0

L'inegalite rr'p <0 exprime que vest <c ou que les ondes soni-
ques sont orientees dans Ie temps. Les hypotheses (HI) et (H 2) se re-
duisent a l'approximation classique aux hypotheses usuelles dites de
Hermann-Weil et nous montrerons qu 'elles jouent Ie m~me rMe qu 'elles,
pour la thermodynamique des ondes de chac en hydrodynamique et ma-
gnetohydrodynamique (relativistes). On verifie facilement que 7:: I <.0
p
implique les deux autres conditions par 1es gaz polytropiques relativi-
stes. Dans des travaux encore partiellement inedits Israel et Lucquiaud
en ont donne des justifications du point de vue de la mecanique stati-
stique
b) En inversant la fonction 1:' = (;' (p, S), on obtient line fonction
p=p ("t', S) exprimant la pression en fonction des variables -"t; et S. On
a identiquement en 't et S

(8. 1) p = p {'t(p, s), s}

 - 113 -

Lichenerowicz

Nous nous proposons d 'evaluer les derivees partielles de p(~ S), Par
derivation de (8. 1) par rapport a p, a S constant, il vient:

/
(8. 2) p' 'U =1
1; P

On en deduit:

(8.3) p' =
'C 't~

De m~me, par derivation de (8. 1) par rapport a S, a p constant, on a:

p' 't;' + p' =0

't S S

II en resulte:

'C"S
(8.4) p' = - - -
S t;'p

Les hypotheses de compressibilite (HI) se traduisent denc par les inega-

lites Pz,<o, Ps ~o .
c) En derivant (8.2) par rapport a p, a S constant, il vient:

On en deduit:

(8.5) "
P'\;'2

et (H 2) se traduit, modulo (HI)' par rf2> 0

 - 114 -

Lichnerowicz

Nous n'etudierons pas, pour elles-m~mes, les ondes de choc de

l'hydrodynamique relati viste, mais nous considererons une telle etude
comme un cas particulier de l'etude complete des ondes de choc de la
magnetohydrodynamique, a Ihaquelle nous procederons.

III. LES EQUATIONS DE LA lVIAGl\TETOHYDRODYNAMIQUE

RELATIVISTE.

9. Le tenseur d'energie de Ia magnetohydrodynamique.

a) Supposons Ie fluide envisage soumis a un champ electromagnetique
Mcrit par l'ensemble de deux 2-tenseurs antisymetriques H et G; H est
ici Ie tenseur champ electrique-induction magnetique et verifie Ie pre-
mier groupe des equations de Maxwell dH=Q (ou d designe la differentia-
tion exterieurel. Si *' est I'operateur d'adjonction sur les tenseurs
antisymetriques, les vecteurs orthogonaux a u, donc spatiaux

01..
e~ =l!l Ho{~

sont respectivement Ie vecteur champ electrique et Ie vecteur induction

magnetique relatifs a la direction temporelle u. Soit 1" constante don-
nee, la permeabilite magnetique du fluide. Le vecteur champ magneti-
que h est suppose relie a l'induction magnetique b par la relation:

Le courant electrique Jest sensiblement la somme de deux termes:

 - 115 -
Lichnerowicz

J~ = -VUP-:> + ()e~

ou vest la densite propre de charge electrique du fluide et () sa

conductivite.
b) La magnetohydrodynamique est ici I 'etude des proprietes d'un
fluide ideal relativiste de conductivite infinie () = 00; J etant essentiel-
lement fini, il en est de m~me pour rr ,e, et I 'on a necessairement
e=O.
Par rapport a la direction temporelle definie par Ie vecterr-vitesse u
du fluide, Ie champ electromagnerjque est reduit au champ magnetique h.
D'apres des resultats c1assiques, ce champ admet Ie tenseur d'energie:

ou IhI2=_h~ h f est strictement positif pour h

f fO. Le tenseur d'ener-
gie total fluide-champs s'en deduit:

(9. 1)

ou l'on a pose:

c) Le systeme differentiel fondamental de la magnetohydrodynamique

relativiste est constitue par les equations suivantes: I 'equation de conser-
vation de la densite de matiere:

(9.2)

les equations de Maxwell (dH=O) qui peuvent s'ecrire ici:

 - 11 G -

Lichnerowicz

(9.3) """~
Vo( (u n -no{(O
u )=0

et les equations de la dynamique relativiste:

(9.4) \J0( To(~ =0

ou T.ol~ est donne par (q-l).

10. Consequences du systeme fondamental.

Nous utiliserons dans la suite un certain nombre de relations, con-
Seql!enCeS du systeme (9. 2), (9. 3), (9.4).
a) Partons des equations de Maxwell explicitees sous la forme:

(10. 1)

et projetons-les successivement sur les directions definies par les

deux vecteurs orthogonaux u et h. Par produit scalaire par u il vient,
compte-tenu du caractere unitaire de u:

(10. 2)

Par produit scalaire par h, on obtient:

soit, compte-tenu de l'orthogonalite de u et h:

(10. 3)
 - 117 -

Lichnerowicz

b) En explici tant (9. 4), on a:

Par produit par u ~, on en deduit:

c 'est-a.-dire:

qui, compye-tenu de (10.3), peut s'ecrire

\7 2 eX. 0(
Yo<. (c rfu )-u 00( p =0

relation indentique a. (5.7). En faisant intervenir ® et S, il vient encore:

Ainsi (9.2) entraine toujours I'equation de flot adiabatique:

(10.6)

En utilisant (10.6) paur transformer .(10.4), on obtient Ie systeme dif-

ferentiel aux lignes de courant:
 - 1 1g -

Lichnerowicz

En developpant, il vient:

(c 2rf+ }it hl2)uO<Vo( u0 -(t ~ _uo{ u~ ) ~ p - :~ JL g o{~'Vo( Ih /8 +

+ t )-l-uo( u~~lhI2 + }lh'A u,u,V" h}l- u~ - ~ v;"h« h~ -fA.h~Y'o( ! =0

En tenant compte de (10.3), on obtient:

(10.7) (c 2 rf+ }-LlhI2)J<\7~u~ -(g~~-uotu~)aol.p-tfg~~~lhI2+

+ fU"'- }V'oc: Ihl 2+}L Ihl 2 ~ uo( U~ - J1:~ho( h~ - f ho("V<I,! =0

Par produit par h~, on deduit de (10.7):

2 JL Ihi 201.~\7
(c rf+
C<" 1 0<\71121120(
)u h Vo( u~ -h CJo( P - '2 Jih Yo{ h + p.- h \/0( h +

soit, apres simplification,

2 2 0( ~r1 C(
(c rf+ )-l~hl )u u ~ h[:J Hi ()().p- fA' Ih.12 \7vol... h0(- =0

Compte-tenu de (10.2), on obtient la relation simple:

 - 119 -

Lichnerowicz

2\7 O<DI.
(10.8) c rf Vr;I. h +h ~p =0

11. Ondes magnetosoniques et ondes d'Alfven.

a) Dans un domaine[L ou les variables thermodynamiques p, S
et les vecteurs ue>(" h cI. sont continus, soit L une hypersurface d 'e-
quation cr =0 (Cf de classe C 2; l=d f), a la traversee de laquelle
les derivees premieres sont discontinues, de fac;on que p, S, uC<, hO<
verifient les hypotheses du § 3, c. Comme precedemment nous posons:

~ [~SJ =k ~ S
rV h~J
- rJ..
=1
0(
~ h~
Etudions a quelle continion l'une au moins des distributions ~ p, Js,
~ u ~, J h~ est non nulle.
Le systeme differentiel (9.2), .(9.3), (9.4) est equivalent au systeme
differentiel forme par (9.2), (9.3), (10.6), (10.7), Les relations (9.2)
et (10. 6) donnent d 'abord, exactement comme dans Ie cas de l'hydrody-
namique:

(11.1)

et

(11. 2) (uc( 1 )
0,
JS =0

Des equations de Maxwell (9.3) ou (10.1), on deduit par un raisonnement

 - 120 -

Lichnerowicz

identique a celui du § 6:
(11.3)

Enfin Ie systeme differentiel aux lignes de courant donne de maniere

analogue:

(11. 4) (c 2rf+}-'-1 hl 2}(uo( lei.) ~ u~ -(l~ _uc( i

u~ )10£ ~p - }U~ ~I hl 2 +

+ ]A- fttc{ 10( )u~ ~ Ih 12 + Ji Ihl 2 u~ Ie{ ~ uo(_ f" h~ 10( ~hO< -f(ho( lo()J Jl =0

Pour ucJ.lo\ =0, JS peut t!tre nulle et on obtient de nouveau les ondes
de matiere. Nous supposons dans la suite uo/ lCl{FO et par suite dS=O.
b) Des relations (10.2) et (10.3), consequences des equations de
Maxwell, on deduit les relations suivantes consequences de (11. 3):

10{ ~ho( -(JX lol.)U~ h h~ =0

~(tFlo()~lhI2+ IhI21c<~uo( _(hoi. ~)u~ cS h~ =0

Par elimination de u p., ~ h~ entre ces deux relations, il vient:

Considerons maintenant la relation (10.8), consequence de (10.7).

On en deduit:

2 (0/ 0( (
(11.6) crflolu h +olo(c)p=O
 - 121 -

Lichnerowicz

Notons que comme bs=o, il resulte de (5.3) ()p=c 2r ~f. Par suite
(11.6) peut s'ecrire:

Ainsi Ie scalaire b=fhO( 1 eX est invariant par la derivation ~

Enfin multiplions (11. 4) par 1 ~. II vient:

(c2rf+J-LlhI2)(UO<lo1.)l~~U~-(gO<~.U<XU~)l0l1~ ~p-~ 1~1~~lhI2+

+JJ-(u ol lo()2 ~ Ih/ 2+ f' IhI2(uo('lo()l~ clu~-2 f(hcX\:()l~~ h~ =0

c'est-'a-dire:

2 rio. ,PI 0I.P.J 0( B 1 ~ (I 2

c rf(u lo1.)l~ 0 ur -(g ,- -u ur )lo(l~Jp -'2 fl l~cl hi +

+y.~ Ih l2(uoIlo( )1~ ~u~ +(uO/ 10(.)2 ~ ~h \2 _2(ho( 101.)l~ ~ h ~ } :0 0

D'apres (11.5), on obtient:

Eliminons Jlhl2 entre (11.5) et (11.7). Par produit de (11.5) par

IN 1Y 1 f' de (11. 7) par (uol 1~)2 et addition, on obtient:

c2rf(uollol)331~ ~u~ _(ud.l~/(g~~ _ueJ. u~ )10{ 1~ Jp+ n 1

+)A-19 ly{lhI2(tP101)1 ~ cl u~ _tho( 10/)1 ~ ~ h~l =0

 - 122 -

Lichnerowicz

c 'est-a.-dire:

~
{c 2rf(uO< 10/ + Jl'lh121 f 1 } (u"\.( )1~ bu~ -(uollo()2(1 ~ 1~ -f) 1,e )2) ~ p-

- IJ, h0{ 1 1P1 1 S h~ =0

I 0{ f'~

A cette relation, nous pouvons adjoindre (11. 1), soit:

et (11. 6) soit:

ot { 2 (~
(h 10{) c> p+c rn~ U h =0

Le determinant de ces trois equations l~eaires aux inconnues 1~ du~,

bP,l~~h~ s'ecrit:

tc2rf(UcX.\,/+ J.l'lhI21Y1~}(~lo() -(U(\/l~l~ _(U~1~)2) -)J.hd.. 1o( 19 1

H: 2
c rf O(u~ 10{) 0

hOi 1 2
0 C/ c rf

On obtient par deve1oppement:

 - 123 -

Lichnerowicz

Pour que Ie systeme considere admette des solutions autres quella

solution nulle, il faut et il suffit qhe H=O. S\~l en est ainsi, les rela-
tions (11. 1}, (11. 5) et (11. 6) fournissent l~ ~ u~ , l~ ~ h ~ , ~ , h \2 en
fonction de ~ p.
c) L'etude precedente concernait ~ p et les composantes normales
a L de u ~ et h~ . Decomposons ces vecteurs selon leurs compo-

santes tangentielles et normales a L . 11 vient:

ou ) 1~)1~=0. Compte-tenu du b, les formules (11.3) et (11.4)

fournissent relations de la forme:

(11. 9) (hO\>:) ~ v~ -(u'\,()~ t~ ~O

(11.10) (c 2rf+p: IhJ2)(uo{ ~ ~ v~ .:.}l(ht><lc{)S t~~ 0

ou Ie symbole (V signifie encore modulo des termes proportionnels

a ~ p. Le determinant des equations (11. 9), (11. 10) aux inconnues
~ v~, ~t~ s'ecrit:

D(l) =(c 2rf+ f I h 12 )(u0( l~) 2-f.l,(h~ lei) 2

d) Nous avons ainsi mi-8 en evidence trois types d'ondes ou d'hyper-

surfaces caracteristiques du systeme fondamentale de la magnetohydro-
dynamique
1) Les ondes d'entropie ou hypersurfaces engendrees par des lignes
 - 124 -

Lichnerowicz

de courant dlequation

Ii la traversee desqueUes ¢ S peut ~tre #0.

2) Les ondes magnetosoniques, hypersurfaces telles que l=d<p ve-

rifie:

A la traversee d lune onde magnetosonique Jp et par suite 1~ ~ uf

Ip ~ J , ~ Ihl 2 peuvent ~tre fO
3) Les ondes dlAlfven, hypersurfaces verifiant:

A la traversee diune onde dlAlfven, il peut y avoir discontinuite effec-

tive des derivees des compos antes tangentielles de la vitesse et du
champ magnetique: dv~ et ~ t~ peuvente ~tre fO pour ~ p=O.
Posons pour abreger ~ ='V(c 2rf+ JA-I hI 2)/fI' . Llequation D(l)=O
aux ondes dlAlfven peut slecrire:

et lIon voit que les ondes dlAlfven sont engendrees par les trajectoires
des champt de vecteurs:
 - 125 -

Lichnerowicz

i- = ~ ual +hol
ce qui d~finit deux types d'ondes dites ondes A ou ondes B. On note que:

Les vecteurs A et B sont temporels et orient~s vers Ie futur.

12. Vitesses des ondes magn~tosoniques et des ondes d'Alfven.

a) Une hypersurface L (d'~quation <p =0, avec l=dcp) ~tant donn~e,

nous avons associ~ A u et 1 (voir (6.1)) Ie parametre:

r
y ~o

La dMinition pr~cMente est ~quivalente A la relation:

(12.1) (l-y
1.. 0( 2 r.o(
)(u 10() = -y (1 10()

De maniere analogue, associons au vecteur h et A 1 , une composante

h Mfinie par:
n

h 2 v~rifie Ie lemme suivant.

n
 - 126 -

Lichnerowicz

Lemme 1°) On a toujours h 2n 61hl 2

--
2°) Pour que h 2 =lhI 2 , il faut et il suffit que I appartienne au 2-plan
n
n defini par u et h.
En effet considerons au point x de L un repere orthonorme
(e ,e.) (i, j=l, 2, 3) tel que e =u. On a hO=O et il vient d'apres l'inegalite
° 1
de Schwarz:
°

c'est-a-dire

On obtient

ce qui etablit Ie 10. Pour que I 'egalite ait lieu il faut et il suffit que
si nou~ decomposons I selon un vecteur colineaire a u et un vectaur
k orthogonal a u, k soit colineaire a h, ce qui ctemontre Ie 20.
b) En introduisant h 2, Ie polynome P(l) peut s'ecrire:
n

D'apres (12.1), P(l) peut s'exprimer en termes de yL. par:

 - 127 -

Lichnerowicz

(12. 2)

ou n (y) designe le trinome du second degre:

c 'est-a-dire en developpant et ordannant en y:

(12. 3)

Sou3-.Jil. seUl~potl:..~~e.. _ryJp <~ (soit r> n

1), (y) ales proprietes

suivantes

D'autre part:

2
-rr v
1\ (-:d
c

Ce qui peut sreerire:

.. .
f\lO~l 1\ Y a d eux zeros
1\ () • en t re O
et lque " "eSlgnons par y :VlL et
dnous

y "vIR. II eXlste
. d one pour.1es rmdes magnetosomques
.. . et par rapport au
 Lichnerowicz

f1 Ul'd e d
eux '
vltesses v IVlL et v MR , d'f"
e lmes par (v:VfL)2/,c 2 c y NfL et
(V 1\11''1)2/Ie 2=y MR ,satlsialsant
.. . 1
es' 'mega
< 1"ltes:

ML :YIn
(12.4) v ~ v~ v <" c

lViL . i\iIn
v e s t (hte vitesse des ~lDdes magnetosoniques 1entes et v
vitesse des ondes magnetoeoniques rapides.
?
c) En introduisant h~ dans D(l), on a:
n

2
(12. 5) )-I.hn _)
2 2
c rf+yl h]

Ainsi les ondes d 'Alfven admettent par rapport au fluide une vitesse
A
v donne par:

(V A)2 -'
(12.6) --2-
c

Evaluons:

soit, apres semplifications,

 Lichne rowicz

1\ A)2 A)2 ')

I\ (J;-- ) ~ = f- (11 ~ - IL I ") ('¥ . l) ~ 0
c c

On etablit ainsi que:

(12. 7)
i\IL ~ A/ .Im
v ~ v ~ v

Soit L une onde d'alfven de vitesse vAiO (h 2 fa); elle verifie

D(1)=O. i\'otons que paur que P(l)=O i1 faul et il suffit que
') ')
n .
n(~)=O,A)2

c~

c'est-a-dire que h~=

n
/111-. D'aprcs Ie lemme doit appartenir au
2 -plan n defini par u et h( (puur urX. 10{ f 0).

13. Representation des cOnes d 'ondes dans H 3.

a) En un point x du fluide, soit T~
x
l' espace des covecteurs tangents
a 1'espace-temps. Le ctlne fondamental de 1'espace-temps peut-Nre de-
fini par Ie cOne r des covecteurs 1 verifiant:

r go{ ~ 10( l~ =0

D'apres l'etude du S 11, les caracteristiqnes ou ondes de la magneto-

hydrodynamique sont definies par les hypersurfaces tangentes a 1'un des
trois cOnes d 'equations:

Introduisons en XI un repere orthonorme (e ,e.l (i' I, 2, 3) tel que l'on ail:

o 1
 - 130 -

Lichnerowicz

e ~ u
o

Nous posons provisoirement dans ce paragraphe:

1 =t 1 = "K 1 = y 1 = z
o 1 2 " 3

Avec ces notations, nous obtenons pour les differents cbnes les equations
explicites:

222 2
t -x -y -z = 0
t = 0
2 42 222222
c rf(1' -l)t +(c rf+ flh\ (t -x -y -z )- i')t

avec "( > 1. En introduisant ~ , on obtient pour equations de r,

fH' r A (apres avoir pose ~2= )=>2/1 hJ2)

r 222 2
t -x -y ··z = 0
P, 2 1 4 -2 2 2 2 2 2 2 2 2 2 2
(l~ -l)('¥-d t +(~ +'i'-l)t (t -x -y -z )-z (t -x -y -z )=0

-22 2
~ t -z = 0

Nous appelons indicatrices dans R3 les sections de ces cbnes par

l'hyperplan too I _ Nous obtenons ainsi les trois indicatrices suivantes:
 - 1:11 -

Lichnerowicz

2 2 2
1-x -y -z =0
-2
~ l' -( P
-2
+ r -l){x +y +z }-z {l-x -y -z }=o
2 2 2 2 2 2 2

z =t.~

11 est aise de discuter la forme de ces indicatrices qui admettent Oz

comme axe de rotation. II suffit de couper les indicatrices par Ie plan
x= 0, ce qui conduit au cercle

2 2
Y +z =1

aux deux droites

z = ±~

et par SH a la quartique:

CH : ~
-?
Wi -( -2 2 2 2
~ +~ -l){y +z }-z {l-y -z )=0
2 2

admettant Oy et Oz pour axes de symNrie. L'equation de C H peut

s 'ecrire:

2
Y = -
(z
2
-1 2 -?
){z - \? ~}

z2( ~ \1' -l}

2 ?
Pour z = l' et on a y'"=0 et elI admet les asymptotes
1:'
.) -'J
z-= f.> ~+ -1.
 - 132 -

Lichnerowicz
- 2
Nous appelons cas general Ie cas ou l'
f ~ . Par exemple pour
-2
l' t... ~ , on a pour CH Ia forme de Ia fig. 1; H est decomposee r
topologiquement en deux parties
et rH r
Hila partie exterieure
1 . . 2'... l~
~H correspond aux ondes Ientes et 1a partle lllt.::rleUre H aux
2
ondes rapides.
Nous appelons cas singulier Ie cas ou 'Y = ~-2 , c'est-a-dire ou:

2
v
(13.1)
2 2 2
c c rf+ J-"Ihl

C admet des points singuliers sur Oz et a Ia forme della fig. 2.

H
z

------.~--"-=- -- - -

/
/
/
-, "-
\
/
\
0 I
\ I J
'- ...... /

fig. 1 fig. 2
 - 133 -

Lichnerowicz
r 2 2
Le cCne ~H admet deux generatdcessingulieres x=O, y=O, z =1' t
contenues dans Ie 2-plan 11 defini par ru et h,
Dans tous les cas, rA est l'ensemble des deux hyperplans z=± ~t;
leur intersection est Ie 2-plan n I defini par z=O, k=O, c'est-a-dire
l'orthocomplement du 2-plan n, Dans Ie cas general les hyperplans
r A sont tangents a r H le long des generatrices contenues dans 1T
et dMinieS par z2= ~2 t 2 , Dans Ie cas singulier, ils passent par ces
generatrices qui sont les generatrices singulieres,
L'intersection de l'hyperplan r (t =0) soit avec r H, soit avec
chacun des hyperplans de, rAn lest autre que Ie 2-plan TT
I'H n rA,
I,

b) Etudions l'intersection Si nous posons z2=~2t2

dans I'equation de r A il vient
-2 4 -2 2{ 2 -2 2 2}
(~ -1)(~-l)t +(~ +t-l)t t (1-~ )-x -y -

- 2 2 J 2 - 2 2 2)
- ~ t L t (1- ~ )-x -y = J °
soit apres simplifications:

(i' -l)t 2(x 2+y 2) = °

L'intersection se compose donc du 2-plan 1t I (d'equations z=O, t=O)

et des deux generatrices de r H contenues
dans Ie 2-plan IT (x=O,
-2
y=O) et dMinies par z=± ~t, Dans Ie cas general f ~ ), les ('1
hyperplans de r A
sont tangents a r:H
Ie long de ces generatrices;
dans Ie cas singulier (1t = ~2), nous avons vu que ces generatrices
sont singulieres,
c) Designons par a(x, () ) (x f n.) un operateur differentiel homogene
 - 134 -

Lichnerowicz

d 'ordre m. Si 1 E: T~ a(x,l) est un polynome homogEme en 1 de degre

x
m. Soit V (a) Ie ctlne de T* defini par l'equation a(x, 1)= 0, OU x est
x x
fixe. BNous considerons V (a) comme un ctlne projectif. L'operateur
x
a est dit strictement hyperbolique en x, si V (a) verifie l'hypothese
x
suivante:
II existe dans T* des points 1 telle que toute droite issue de 1
x
et ne passant pas P<:ir Ie sommet du ctlne coupe la surface du ctlne en
m points reels et distincts.
Etudions les operateurs P(l) et D(l) qui apparaissent dans Ie syste-
me differentiel de la magnetohydrodynamique. Pla90ns -nous dans Ie cas
general: l'operateur P(l) est strictement hyperbolique, mais il n 'est est
manifestement pas de m~me pour l'operateur D(l) qui n 'est que Ie produit
de deux operateurs strictement hyperboliques du 1e ordre correspondant
aux deux hyperplans. Le systeme differentiel etudie n'est donc pas
strictement hyperbolique.
II en est a fortiori de m~me dans Ie cas singulier OU l'operateur
P(l) lui-m~me n 'est pas strictement hyperbolique.
En me plarant dans Ie cas general, j 'ai etabli cependant ailleurs
(Relativistic hydrodynamics and Magnetohydrodynamics, Benjamin, New
York (1967))un tMoreme d'existence et d'unicite pour Ie probleme de
Cauchy relatif au systeme de la magnetohydrodynamique, sur une classe
convenable de fonctions Coo.

14. Propriete fondamentale des rayons associes aux ondes magnetoso-

niques.
a) Soit 'f une solution de l'equation aux ondes magnetosoniques
P(l)=O. Nous supposons que pour les ondes L.. envisagees D(l) est
 Lichnerowicz

#0. Il resulte du § 11 (en particulier (11.9) et (11.10) que les ''tli-

sconiinuites infinitesimales 11 Ju ~, ~rf peuvent s 'exprimcr de ma-
niere unique en termes de ~ p. Nous aurons besoin en particulier de
l' expression de ~ /hl 22. D 'apres (11. 7):

On en deduit en exprimant l~ ~ u~ a partir de (11. 1):

c'est-a-dire

(14.1)

b) Les ondes magnetosoniques sont les shypersurfaces tangentes

aux ctmes P(l)=O. La generatrice de contact avec Ie ctme est dCfinie

par Ie cvecteur:

I() P(l)
2
'C)1~

soit:

(14.2) n:'{~ =2c2rf(O-1)(U<><'lo()3u\'J+(c2rf+flhi1 )uo(lll((lflfu~+(U~lo()l~)

0( 0 ~ 0( ~
- JA- h 100J.l) 1 ~~l +(h 10()1 )
 - 136 -

Lichnerowicz

Ce vecteur etant tangent a l'onde magnetosonique L, (14.2) peut

encore s 'ecrire en prenant les composantes tangentielles des differen)s
termes:

(14.3) N~ =2c2rf(l-l)(Ull/lo()3v~+(c2rf+flhI1 )(UOC:lo()(lfl~)VP­

-f (hoi. lo()(l~ If)t f

Les bicaracteristiques ou rayons associes aces ondes L sont les

trajectoires sur L.. du champ des vecteurs Nf... Nous nous proposons
de montrer que cl p (et par suite les ~ u~ , ~ h~ ) se prop agent Ie
long des rayons, c'est-a-dire que ~p verifie un systeme differentiel
de la forme:

ou Ie symbole ~ signifie modulo des termes proportionnels a ~ p.

Nous postulons dans la suite les hypotheses 58 1 et B2 du § 4 pour
p,S,uiX,hiX . D'apres (4.3), il existe des distributionsp,S,u~,h:t
telles que:

(14.4) ~[Vd.V~ pJ = 'V'" l~ b p+lD( 'V~ ~ p+l~ VoZ~ p+lo( l~ P

(14.5) X[VD\V~ SJ = 10{ 1~ S

1['Vo\v~uaJ = Vd., 1~ J u'l. +10( ~ 6U~+1~ \70{ ~U?.+~l~

-A-
(14.6) u

~ Vc ~ ~ 'A -~
(14.7) ~['Vo( V~h 'A] = 'Yo( l~ ~ h +lc;( /?> ¢h +1~ ~ h +1",-1(0 h
 - 137 -

Lichnerowicz

c) Des relations (11. 1) et (11. 2), on deduit comme precedemment:

(14.9)

Des relations (10. 2), (10. 3) consequences des equations de iVraxwell,

on deduit par derivation, en raisonnant comme au ~ 7:

et

En multipliant la premiere de ces relations par h~ , la seconde par

u~, et retranchant, il vient:

De (10.8) il vient de m~me:

(14.11)

l{>e (14.10) on tire, compte-tenu de (14.11) et (14.8):

 - 13B -

Lichnerowicz

d) Prenons enfin la derivee covariante contractee de (10.7). On ob-

tient, compte-tenu de (14 ..5):

(c2rf+)llhI2)l~[V'~ Vo\~ _(go/~_u«u~ )JlV'... ~p] - trb[~ V~lhI5+

+}J-uo/ u~~[Y'c{ V~ Ihl 2J+t l h\2l ~[V~~o(uo/J -2Jh~ ~ [V~ Vo( ho(] ~ 0

Ce qui peut s 'ecrire:

~- [ 1 - ~
2
c rf u b ~Vo( u
01.1
_(go(l~P. j?l
_ur:i.. u )
-
b [Vo( ~pJ - 2"JL 6 LV~'i~Ihl 2 J+
+t{2 h\2 I u~ ~[~ ~ u1 +uol.u~b[V~~ Ih\2] -2h~ ~ [~~h1]~0
ou la seconde ligne disparait en vertu de (14.10). En utilisant (14.8) il
vient:

(14.13)

Introduisons l'expression:

On a:

et (14.12) peut s'ecrire:

 - 139 -

Lichnerowicz

u u -h h

Apres produit par c2rf(u~ l~ )2, la formule (14.13) prend la forme:

soit en ordonnant:

(14.15) tc2rf((O-l)uol. u~ +i~ )(u~lf)2+}A-lhl\ Ifl?Uol.U~_

-y-191?holh~}X[Y't¥\7~pJ +f"ll~rf(u~l?) {(uVl~)l~ -l~ u~l V~ ~lhI2~ 0

Substituons a 1['Yo( V~ pJ sa valeur tiree de (l4.4). Le coefficient de

pest:

soit:

Ainsi Ie coefficient de pest nu1. n vient ainsi a partir de (14.15):

 - 140 -

Lichnerowicz

(14.16) 2 { c 2rf ("t _l)(uol. 10( )3u~ +c 2rf(u« 10( )2l +~I hl 21 l~ If (uo\ le()} -

- }'-l~ If (hoi ho( )he}v~ b p + r:c 2rfu ~ l~ {(U9 l~ )l ~ -l~ l~ u~l V~ ~ Ih \2 ~ 0

D'apres Ie calcul precedent concernant Ie coefficient de p, Ie vecteur

coefficient de ~6P dans (14.16) est tangent a L. II en est de
m~me manifestement pour Ie coefficient de 'V~ J I h ~2. La relation
precedente peut donc s 'ecrire apres division par 2:

t c 2rf (O _l)(uo( 1ell/V ~ +,fI-1 hl 21 11> If (UC)(lo(, )v~ -f If If (ha! 1

lo()t~ 'l~ p

- trvc2rfl91~(uolltJ()V~~-cS lhl2~ 0

soit d' apres la relation (14. 1):

{2C 2r f(r _l)(uO< 10()3 vf.> +(c 2rf+}-Lj h/21 )1 f l~ (ue( lo()v 13_

-}A-l~l~(hQ(lol)t~l'7~~p~ 0

c'est-a-dire d'apres (14.3):

Nous obtenons ainsi :

Theoreme. Sous les hypotheses du ~ 4, les discontinuites infinitesi-
males ~p, ~u').., Jh" relatives a une onde magnetosonique L se
propagent Ie lang des rayons associes selon les systemes differentiels:
 - 141 -

Lichnerowicz

N~ 'V, J h'A = 0
~

15. Propri~e des rayons associes aux ondes d' Alfven.

a) Soit 'f' une soluzion de I 'equation aux ondes d 'Alfen par example
d'espece A. On a:

(15. 1) AoI. "d <l> =A01.. I =( ~ uo( +h O()l =0

0(\ 0( r eX

Nous SuppoBons que pour les ondes envisagees p(l)fo. Par suite:

~p =0 I ~ uol. = 0
01..

De plus (14.15) se reduitt a P(I)p=O; par suite 1>=0 et l'on a:

(15.2)

De (14.8);, (14.11)(14. 12) il resulte:

b) Pour les ondes d'Alfven envisagees ici, seules les discontinuites

~vA, ~t'). peuven1 ~tre non nulles. Des relations (11. 9), (11.10) ou
uollo(fO et ou les seconds membres sont nuls, on deduit que ~t'A.
est proportionnel a ~ v OX. Le symbole ~ signifie dans ce § mo-
dulo des termes lineaires par rapport aux bv~ (ou aux ~ t'l- ).
 - 142 -

Lichnerowicz

Compte-tenu de (15.2), (15.3), les equations de Maywell est Ie systeme

aux lignes de courant donnent par derivation:

et:

Multipliant la 1ere relation par h~ , la seconde par u ~ et retranshant,

il vient

soit en explicitant:

Le coefficient de U?. est nul et il reste:

soit d'apres (15.1):

Nous obtenons:
 - 14:3 -

Lichnerowicz

Theoreme. Les distributions.{)(vA, V(t A a supports sur une on de

d 'Alfven L d 'espece A se propagent Ie long des rayons associes
selon les systemes differentiels:

ou ~ signifie modulo des termes lineaires par rapport aux ~ vJA-

(resp. 6t}l.l ).

Des resultats symetriques sont valables pour une onde d'Alfven

d'espece B.
c) Reprenons une onde d'espece A et etudions l'action de la deriva-
tion b sur Ie vecteur

D 'apres l'etude du a, on a:

1 6uo( =0 10( J ho( = 0

c\.

II en resulte:

soit d'apres (11. 9) ou Ie second membre est nul:

 - 144 -

Lichnerowicz

II vient <5 } =0 et Ie vecteur A(3 lui-m~me, et sa direction en par-

ticulier, sont invariants par la derivation ~ .

IV. ONDES DE CHOC EN MAGNETOHYDRODYNAMIQUE

16. Le systeme fQndamental des ondes de choc.

a) Dans un domaine ..0.. de V4' soit encore L une hypersurface re-
guliere d'equation cp =0 (Cf de c1asse c 4, avec l=dtp). L'hypersurfa-
ce L. est une onde de choc magnetohydrodynamique si uO{, ho( ou l'une
au moins des variables thermodynamiques sont discontinus a. la traver-
see de L.. , conformement aux hypotheses Al et A2 du ~ 3. Nous e-
tablirons que sous les conditions de compressibilite HI' r.. est neces-
sairement orientee dans Ie temps, done de vitesse admissible au point
de vue relativiste.
Soit Y un etat du systeme fluide-champ ,defini par les valeurs de
p,S,u~,h~ en un point x de L. Un tel etat est defini par la donnee
de 8 quantites scalaires, c 'est-a.-dire depend de 8 parametres. Nous
notons Yo 1'etat anterieur au choc, Y1 I 'etat posterieur au. choc et
[Q]la discontinuite, Q1- Qo d'une quantite a. la traversee de L. .
Le systeme fondamental (9.2), (9.3), (9.4) est suppose satisfait au
sens des distributions:
 - 145 -

Lichnerowicz

Ces equations impliquent:

II resulte de l'equation (3.2) que l'on a:

(16.1)

Po sons

a(Y)=ruc( 10(

et:

IhI 2 ~ f3 is
w,., (Y)=(c 21:
jl
+ f -2-) a(Y)ru -ql - j4(H 1~)h
0(

II est clair que Ie vecteur V ~ est tangent a I... Le systeme (16. 1)

exprime que:

Vi' (Y )=V~ (Y )
1 0

Le scalaire a et les vecteurs V i3 et wfJ definissent les invariants

du choc.
b) Un choc est dit tangentiel si a=O. S'il en est ainsi:

est de vitesse nulle par rapport au fluide dans les deux etats et
 - 146 -

Lichnerowicz
orientee dans Ie temps. De 1'invariance de V (? et w (3 il resuIte:

(16.2)

(16.3)

~~ !.od~ u~ et u~ etant unitaires et colineaires d'apres (16.2),

orientes vers Ie futur, on a u~ =u t et par suite h~ 10( =h~ 1«. De
(16. 3) il re suIt e :

et par suite [q] =0. De (16.3) resulte ainsi [if] =0, [PJ =0, la di-
scontinuite de r
0( _
restant indeterminee si he< lex' =
-0- -
° il resulte de (16. 2)
hI 10< -0 et les equations de choc donnent seulement ici:

les autres discontinuites restant indeterminees.

c) Nous supposons desormais alO (choc non tangentiel) et introduisons
Ie scalaire H, invariant par Ie choc defini par:

01 2 2
(16.4) H(Y) = ..!. V~ V(.l, = (h 10/) Ih I
2 \~ 2 2
a a r

En substituant dans wi3 a h(3 son expression en termes de vt3 et ui3 :

 - 147 -

Lichnerowicz

il vient (nous supprimons la reference a Y lorsqu'elle est inutile):

W
{3 2 Ihl 2 i3 f3
=(c"C +}'--2 ) aru . -ql - ~
(hO< 10d 2 {3 rhO/leX Vi3
ru + f - -
r a: a

soit:

ou l'on a introduit la variable importante:

2 2
0( = c t -fH = D(l)/a

Si nous decomposons wf3> ell sa partie tangentielle et sa partie normale

a L, on obtient

ou

(16.5)

est tangent aL . L'invariance de W f-> est equivalente a l'ensemble de

celIe de xP-> et de celIe de la composante normale. Ainsi Ie scalaire:

2 2
c a
(16.6) e = q- - - 't
10( 1
01.

est invariant.
 - 148 -

Liehnerowiez

Considerons en partieulier Ie produit sealaire invariant au eours

duo ehoe:

D'apres la definition de: qt , on obtient Xf> Vf-> =e 2ab, OU b est Ie

sealaire invariant:

(16.7)

d) Consi~rons enfin Ie sealaire invariant:

K
2 2
e a

done nous allons donner deux expressions importantes:

Lemme 1. L'invariant K admet l'expression:

(16.8)

OU l'on a pose:

(16.9)

Il admet aussi l'expression:

(16.10)
 - 149 -

Lichnerowicz

En effet de:

on deduit

Il vient aussi:

2 a2 0(2}LeX 2 0/ )2
(16.11) K =(r _ - - ) - +2 - - r2(ht:X 1 )2+u? r (11 ld II
22 0( r 2?
l ()( 1 ()( _.. - -
?
c (' a c a

En substituant a sa valeur, on a:

soit.:

Or d I apres 1a definition de H:
 - 150 -

Lichnerowicz

n vient aussi:

ce qui etablit (16.8). De cette relation on deduit:

Ce qui peut s 'ecrire:

22 2 II 2
K = c f +)1! hI (2't- ##-)- -a - (c
II
rc:
2 2 2 2
-2c 't:,l.lH+}J- H )
c2 c 2l I c\ /
rX

En reintroduisant 0<., on obtient (16.10), ce qui demontre Ie lemme.

Transformons enfin l'expression de HK, ou K est donne par (16. 11).
De

il resulte en substituant a ,..H sa valeur 2

crt: - 0(:
 - 151 -

Lichnerowicz

2 2 2 2(ho( 1 ) 2 2 2
HK = H(r - _a_)~ + r c{ (c'i:' +c< Hc 1:" -01)
I e( 1 0( c 2 2 2
c a

soit:

? (2 2 2 0< )2
HK= H(r~ _a_) ~ + r (h Lx (c 41: 2_ OIh
1.'Y. 1eX c 2 22
c a

Nous obtenons ainsi:

2
HK ={iI(r2 __
a_)_
10( 10(

soit:

?
2 b~ 2
(16.2) HK = c - )CO<
2 2
a c

Ainsi L= X0( 2 est un invariant qui q, H, K etant fixes peut ~tre sub-
stitue a b en convenant que Ie signe de (hOI 10() demeure inchange au
cours du choc.
e) En ce qui concerne 'X. ' nous allons etablir Ie lemme suivant
qui nous sera utile a differentes reprises.
Lemme 2. 1) On a (1 Id,) X{,
c(
a.
2) Si lo{levest ta, pour que :x =a, il faut et il suffit que I appar-
Henne au 2-plan IT Mfini par (u, h)
3) Si 10( lex est ';y a, ion a ll~a et par suite 0< > a.
En effet d 'apres les definitions de 'X et de H
 - 152 -

Lichnerowicz

. en mtro
SOlt . Ulsant h 2 :
d'
n

ce qui, compte-tenu du lemme du § 12, demontre les 1) et 2).

L'inegalite h2~
n
Ihl 2 peut s'ecrire:

\ hj2
10( 10(
2
a

~ 2
Pour 1 lct)O, on a donc H~O et 0( =c't: -fH)O.

17. Analyse des equations de choc et chocs d 'Alfven.

a) De l'etude du ~ 16, il resulte que les deux arariables thermo-
dynamiques et les trois scalaires Ih1 2, ue< 10(, ho( 1 cl( verifient les cinq
relations scalaires fondamentales:

(17.1) r1u~lo(=rou~ 10( =a

IhC\l )2 Ih \2 (hO< lo'i Ih ,2

. 1 c< 1 0 0
(17.2) - --= = II
2 2 2 2
a r1 a r
0
 - 153 -

Lichnerowicz

2 2 2 2
c a c a
(17. 3) q - - - f'( =q - - - 'L = e
1 {l( 10( 1 0 10( 10( 0

2 2 2 2 2
2 2 c a 2 ).tH 2 2 c a 2
(17.4) c f - - ' 1 : ' +2 ~'l:' X - -'X =c f - - ' 1 : +21LX T-
1 10\ Ie( 1 I 1 1 c2 1 0 leX 10( 0 roo

2
-~ tV = K
2 IVO
C

2 2
(17.5) 1)1 0{ ='V 0< =L
Iv 1 1 .IIi 0 0

ainsi que 1a relation onon independante des precedentes:

(17. 6)

Nous etablions que, sous des conditions convenables, les systeme

precedent definit d 'une maniere et d 'une seule les valeurs apres Ie
choc des deux variables theormodynamiques et des trois scalaires
2
Ihll ,u~ 10(' h~ 10(' Il est clair que si (17.3), (17.4), (17.5) definissent
les valeurs de Iv (ou de Ih1 2) et des variables thermodynamiques,
(17. 1) fournit alors la composante normale de la vitesse et (17. 2)
ou 07.6) celle du champ magnetique.
D'autre part, d'apres l'invariance de V~ et W J5 , les composantes
tangentielles de la vitesse et du champ magnetique verifient:

(17.7)
 - 154 -

Lichnerowicz

Le determinant des premiers membres de (17.7), (17.8) aux inconnues

vf,tf s'ecrit:

Si eX1fo, (17.7), (17.8) determine v t, t~ ep fonction des quantites de-

terminees par Ie systeme des 5 equations scalaires.
b) Supposons d... 1 =0 en un point x de L... ; l'hypersurface est alors on-
de d'Alfven en x pour l'etat posterieur au choc et elle est orientee dans
Ie temps (lo(lo(<:!l). La relation (17.5) donne ry <x2 =0 et ou bien
/\"0 0

0(0=0 ou bien Xo=O.

Dans me cas = Oi =0 L definit un choc d'Alfven. Les cas
01
o 1 '
0( fO,
o
A0 =0,
0< =0 et 0< =0
1 0 '
sont dits des chocs
singuliers. Nous etablirons qu'ils sont incompatibles avec les ondes
d'Alfven infinitesimales et que, par suite, il n'ont pas de realite phy-
sique. On a
TMoreme: 1) Si dans un choc 0(1= 0< rO, Ie choc est nul.
-0'--'--------'---'-
2) Dans un choc d'Alfven (eX 1= d =0) les variables thermodynamiques
-0
lhl 2 , (uO(lcX), (hO( lo() sont continus si l'hypotMse '1;' <0 est satisfaite.
p
1) Si [0(] =0, on a

['1:J=O

Sous notre hypothese Xl = 'Xo et par suite:

 - 155 -

Lichnerowicz
D'apres (17.3), on a [qJ=O et par suite

[pJ =0

Dans K donne par (16. 10), les trois derniers termes sont invariants
au cours du choc. On a done

[rJ =0 [r J =0

et d' apres (17. 1) et (17. 6)

L. n'etant pas onde d'Alfven, il r~sulte de (17.7) et (17.8) que Ie choc

est nul.
2) Pour un choc d' Alfven, on a toujours [t:] =0 et il resulte des
equations de choc, K etant donne par (16. 10):

[ c 2f 2+flh\ 2:]
'tj =0 ~
r: 2rf+}Llhl 2
LC =0

(17.9)

Examinons l'independanca des deux variables theormodynamiques:

tf 2
=c rf-2p
 - 156 -

Lichnerowlcz

Par derivation de cr en p il vient; d'apres Ie § 6:

De m~me en derivant en S:

2
Cl)'=c fr' +r9 =r(c f - + @)
2 rS
ISS r

Or:

2 ,f";\ 2.
c 1:' = 101 V+c fV'
S S

et par suite:

2
c r't:' = ®
2
-c f -
rS
S r

II en resulte:

Le jacobien de c.p et "l;' par rapport ap et S a pour valeur:

d(CP '1:) 2 2 2
--,--'--'-'~=c r '1:" 't' -r(2®-c r-c')"C'
dip, S) p SSp

c'est-a-dire
 - 157 -

Lichnerowicz

d( <p , 't' )
=2-2r8 X'
d (p, S) P

Le jacobien etant different de zero dans I 'hypothese '1:' 0, les va-

p
riables 'f et 'l: sont independantes. Ces variables etant continues

a la traversee de r.., il en est de m~me pour toutes les variables

theormodynamiques. En particulier [r] = 0, [r] =0, [p] =0. D'apres
(17.1) et (17.6), on a:

et [I h1 2] =0, ce qui demontre Ie tMoreme.

c) Dans un choc d 'Alfven, la vitesse tangentielle du fluide apres
Ie choc demeure indeterminee, la direction du champ magnetique
tangentiel etant alors determinee par example a I 'aide du resultat
qui suit.
Considerons un choc d 'Alfven de type A; L... est engendree par les
B
trajectoires du champ A et 1'on a

ainsi que:

De (17.8) on deduit:
 - 158 -

Lichnerowicz

c 'est-a.-dire:

Comme Ie choc est non tangentiel [R] =0. II vient:

Theoreme. Dans un choc d'Alfven d'espece A (resp. B), Ie vecteur PI>
(resp. B~) reste invariant dans Ie choc.

V. FONCTION D'HUGONIOT ET ORIENTATIO"!\; DES O]';1)ES

DE CHOC.

18. Relation d'ITugoniot relativiste.

La relation (17.4) peut s'ecrire:

.)

c 2 [r2] -(T o + 'r 1) [qJ +2f[X'L]- ~~II [X] =0

c

.)

(18.1) c 2[r2] -(t'o+'r1) [p] - t~('Co+T1)[XJ +2f[~rc]- }t~l [XJO

c
 - 159 -

Lichnerowicz

Nous nous proposons de substituer a (18.1) une relation plus maniable.

a) Examinons Ie terme [X~J qui peut s 'ecrire:

[X.'t1J = (X 1 ) 'r1+ X
- X' 0 (t' - 't )
010

ou

[ X. rr], = "V
~1
('t' -
1
-r0 )+(X 1-X0 )t' 0

Par addition membre a membre il vient:

(18.2)

En reportant dans (18.1), on obtient aprus simplifications:

c2 [r 2J '-( 't'o+ 't'l)[P] + tf('t O+'t1- ¥)[X] + f (Xo+X 1l['t} 0

c

soit:

b) Proposons-nous de transformer la somme des deux derniers ter-

mes. On a d'abord, comme dans (18.2):

1.2 (0( +0<

0
)['\11 + 1.2 (X 0 +}:: 1)[o<J = Lrx,o<J
1 I'v

Or d'apres (17.5), on peut ecrire:

 - 160 -

Lichnerowicz

Xo o(~ Xo <Xo
[x«] = X1 ot 1-X 0 c<0 =--
0(1
-X 0 0( 0 =- -0(-
1
(~-OJ. )
1 0

La relation (18. 3) devient ainsi:

(18.4)
2 2 2 1 0(0
c (f -f H't;'+7,;')(p -p )+(1:' -7:) -u.(X +X -2--)=0
Xo
1 0 0 1 1 0 1 0 2I 0 1 ot1

Nous appelons relation d 'Hugoniot cette relation que nous substi-

tuerons desormais a ((7.4) dans Ie systeme des cinqu equations sca-
laires.
c) Un etat initial Y du fluide etant donne, nous considerons
o
jusqu'a nouvel ordre l'ensemble des etats Y verifiant les conditions:

(18. 5) H(Y)=H(Y )=H L(Y)=L(Y )=L

o o

de telle sorte que, pour 'l:' T./. ltH on a:

2'
c

L
X= - - -
2 2 2 2
(c 1:: - fH) (c't' -jUI)

II est commode de substituer a q la variable:

1 1 a2
q = p +-2fX =q+-)A--- H
2 loll
d.

La relation (17. 3) peut ainsi s 'ecrire sous la forme:

 - 161 -

Lichnerowicz

2 2
_ _ c a (_ )
(18.6) q -q = - - .... - 't:
1 0 10( 101. 1 0

Sous les conditions (18.5) un etat thermodynamique ('C', p) du fluide

definit (pour 1:: t ~H ) un point Z du plan ('L, Cj') et inversement.
Entre les variables c 't', S et q, nous avons la relation fonction-
nelle:

1 L
(18.7) q= 2' (c2t:- P.H)2 + P ( 't;, S)

OU p( 't;' ,S) est definie par inversion de 1'equation d'etat.

Nous sommes conduits a introduire la fonction d'Hugoniot ~(Z ,Z)
o
de la magnetohydrodynamique, consideree comme une fonction de Z
pour un point initial donne Z :
o

<f,p 2 2 2 1 'XotXo
(18.8) d'~(Z ,z)=c (f -f }-('t'+'t' )(p-p )+('t'-'t')- u..(x.,+X - 2 - - )
o 0 0 0 0 2/ 0 0(

On a manifestement 'jl{.(Zo'Zo)=O et (18.4) peut s'ecrire ~(Zo,Zl)=O.

C 'est 1'etude detaillee du comportement de la fonction % qui permet
d'etablir la plupart des tMoremes chercMs.

19. Differentielle de la fop.ction

En differentiant (18.8) sous les conditions (18.5), il vient:
 - 162 -

Lichnerowicz

Or:

2
c fdf = f e dS+ 'Cdp 2
doc. =c d'Z:'

On en deduit:

M' 1 Xoo(o
d-dl'O =2f (9dS+('t"-'t }dp-(p-p }dt:+-~(X+X -2--}d't'+
o 0 21 0 ct

1 2 'XoCX 0
+('1:- 't" ) - u. (dX +2c - - d't')
2I d,2

En introduisant la variable q a la place de la pression p il vient:

Soit apres simplification:

2
~J Xo~o
cd-J'!\o=2f@)dS+('t-"C }dq-(iHi }d'C+ IJ,.(%- 2 }dt'
o 0 I 0(

Ce qui s'ecrit, compte-tenu de (17.5):

(19. 1)
 Lichnerowicz

Dans Ie plan ("t,q) introduisons commme parametre la pente m de

la aroite joignant Z a z, On a:
o

(19,2) q-q =m ('r-"C)

o 0

En differentiant il vient

dq=md?: +( -C - 't: )dm

o

et en multipliant par ('!' - 1: ):

o

La relation (19,1) peut done s'ecrire:

(19,3)

20, Differentielle de S Ie long de la droite (Z ,Z 1).

0-
Considerons en x E l. un choc Yo -+ Y1 , L 'etat Y 1 posterieur au
choc verifie manifestement les conditions (18,5). Nous nous proposons
d 'evaluer la differentielle de S pour une famille d 'etats telle que Ie
point Z correspondant decrive la droite du plan (re, <j) joignant Z a
o
Zl' D'apres (17,3), cette droite a pour pente:

2 2
c a
m
 - 164 -

Lichnerowicz

a) De:

on deduit:

/'f -1
1:' dS= ~ dp+d 'C
S 2 2
c r

Or, Ie long de la droite envisagee:

et en differentiant X 0( 2=const.:

On aen deduit:

Ce qui peut s 'ecrire, compte-tenu des valeurs de 0( et X

(20. 1)

b) Le second membre de (20. 1) peut s 'evaluer aisement en fonction

de P(l). La relation (17.1) definissant P(l) peut se mettre sous la
forme:
 - 165

Lichnerowicz

ce qui peut s 'ecrire:

II en resulte:

P~)
a
- if (c2a2~+r-lh,2
r
10/ loi)+~r 10(

On deduit ainsi de (20.1) la relation importante: Ie long de la droite

(2 0 ,2 1):

(20. 2) , ....J dS d"" = ~ dq-

P(l)
't"S...... ~ = 2 .. 2 4
a flo( c a

21. Orientation dans Ie tempos des ondes de choc.

a) Considerons au point x ~!. un choc qui n'est ni nul, ni d'Alfven.
Nous nous proposons de montrer que sous les hypotheses (H 1) ~
compressibilite ('t:' -< 0, 'C s' :> 0) que nous postulons, l'onde de choc
p--
magnetohydrodynamique Lest orientee dans Ie temps.
Considerons la famille des etats Y verifiant (18.5) et telle que
Ie point 2 correspondant decrive Ia droite (2 0 , 2 / II resulte de
 - 166 -

Lichnerowicz

(19.3) et (20.2) que lIon a dans ces conditions:

(21.1) d% =2f e dS

et

(21.1-) ..,s
n-I N
1,1\
dS =~
2 4
dq-
c a

b) Si {i. 10( est> 0, l- est> 1 et on deduit de (12.2), soit:

que P(l) est > O.

Si 10(10(=0, on a P(1)=c 2rf(1-1)(Uo(la)4> 0
Aussi si le(lQ(~O, on a P(l)O et d'apres Ie Iemme 2 du § 16,
0< est > O. 11 resulte de (21. 2) que dS/ dq est> 0 Ie long de la

droite Z0' Zl'

D'autre part, Ie long de cette droite, cJlb (Z o,Z 0 )=0, 1&z 0 ,Zl)=O
et la fonction ~ est stationnaire en un point au moins du segment
(Zo' Zl)' D'apres (21. 1), il en est de m~me pour S, ce qui est en
contradiction avec dS/ Uq )0.
C lest pourquoi 10( 10« 0 et E est oriente dans Ie temps. Nous
enon<;ons:
Theoreme. Sous les hypotheses (H1)-.tt~<.0, 't's).O), toute onde de
choc magnetohydrodynamique est necessairement orientee dans Ie temps.
Si v; ~v~ sont les vitesses de L par rapport au fluide avent et
 - 167 -

Lichnerowicz
r !
apres mle choc, on a v
'0
< c, v, <: c.
1
2
D'apres Ie lemme 2 du § 16, % est;? a et nous posons X =k ;
nous pouvons, pour un choid convenable des signes substituer a la
seconde condition (18.5) la relation k=04=k eX • On a alors:
o 0

2 2 2
k +k -2kk =(k-k )
000

La fonction d 'Hugoniot (18. 8) peut s 'ecrire:

A,P 2 2 2 1 2
(21. 3) 'J1I,:J(Z ,Z)=c (f -f }-('t'+"t )(p-p )+('t"- 't:) -u(k-k)
o 0 0 0 0 21 0

et la relation d 'Hugoniot:

2 2 2 1 2
(21. 4) c (f -f H't" +'t')(p -p )+(1:' -"C )-LI,..(k -k) =0.
1 0 1 0 1 0 1 0 21 1 0

22. Approximation classique de la magnetohydrodynamique relativiste.

Nous nous proposons de deduire les equations de choc de la magne-
tohydrodynamique classique par approximation a partir de celles cor-
respondant au cadre relativiste. Nous nous limitons aux chocs non
tangentiels (arO). Nous posons dans la suite:

uti 1 = w j=rw,
ol c

de telle sorte que a=j/c.

-2
Cherchons les parties principales, relativemente a c , des equa-
tions de choc (17.1), (17.2), (17.3), (17.5), (17.6). On a d'abord:
 - 168 -

Lichnerowicz

(22.1) Dl = [rw] =0
L'invariance de b=fhcx loe donne l'~quation:

-4
On en d~duit qu'a des termes en c pres:

(22.2)

On retrouve qu 'a l'approximation classique:

(22.3)

De (17. 3) il r~sulte:

ce qui peut .s'~crire a l'approximation classique:

2 1 2
(22/4) rw +p+ 2ft Ihl =0

Compte-tenu de (22.2), l'invariance de

 - 169 -

Lichnerowicz

donne alors a des termes d 'ordre superieur pres:

2 jh \2
_c_ (hO( I }2 _0_
.2 0 ~ 2
J r
o

n en resulte:

t 2i
(ho(I0(}2
_0_ _
.2
J
+ Ihrl: ] .Q

soit, d'apres les relations precedentes:

(22.5)

La relation (22.5) peut encore s'ecrire:

soit

(22.6)

Considerons enfin l'invariance de L. Tout d'abord:

c( 2
Ii (HIed 2
-2 u. ( . _ - -~)
c 0( = - (1+-)-
r 2 J .2 2 2
c J c r
 - 170 -

Lichnerowicz
1 (ho( 1N)2
est equivalent a -r - ru. .2
a l'approximation classique. D'autre
J
part:

est equivalent a I q 2. Ainsi:

(22.7)

Nous avons retrouve en (22.1), (22.3) (22.4), (22.6), (22.7) les equa-
tions de choc de la magnetohydrodynamique classique, ecrites dans
un repere lie au choc.

23. Thermodynamique des chocs.

a) p et 8 etant prises comme variables theormodynamiques de base,
f(p, 8) verifie d'apres (5.3):

2 2
(23. 1) c f' = V:>O c f' = €J > 0
p 8

On en deduit par derivation:

(23.2)

Les etats Zo et Zl sont relies par la relation d'Hugoniot (21. 4) qui

est symetrique en 0 et 1. Au cours d 'un choc, on a bien entendu
8 0 ~ 8 1 ell chaque point de L . Nous allons etablir les resultats
 - 171 -

Lichnerowicz

suivants, valables en chaque point de L ,

Theoreme 1. Pour un choc qui cn'est ni nul, ni d'Alfven, on a sous
les hypotheses de compressibilite (HI) ~2t

En effet supposons qu 'au point x de L. , on ait Sg=S 1 et p jP I' En

modifiant au besoin Ie mnumerotage des etats Z et Z , on peut
o 1
supposer Po < PI' On a alors 't' 0> "t'1 puisque 't~ « 0, De (23,2)
on deduit

J P1
Po
~(p, S )dp
0

II en resulte d' apres la condition de c onvevite (H 2):

2 2 2
c (f 1 -f ) «P1- P )("C(p ,S )+ 't'(P1' S ))
o 0 0 0 0

soit

2 2 2
c (f -f}-('1; +'l:;)(p -p ),0
1 0 1 0 1 0

On deduit Ide la relation d'Hugoniot que "t" 1 > '00 ce qui implique
contradiction, On a donc PI =p 0 et Ie choc envisage ne peut Hre que
nul ou d'Alfven,
Theoreme 2, Pour un choc qui n 'est ni nul, ni d 'Alfven, on a sous
 - 172 -

Lichnerowicz

p >i'=p
1 0

En particulier toute onde de choc est une onde de compression et

:Y 1 <V o '
Supposons en effet ~ P0 au point x de
PI -..;: L. De (23.2), on de-
duit:

c 2[ f 2 (p ,s }-f 2 (p l' S) } = 2 fPo 1;' (p, S )dp

o 0 0 PI 0

D'apres la condition de convexite (H 2), il en resulterait:

Comme So< SI' on aurait a fortiori puisque fS>O, "t'S»O

soit:

2 2 2
c (f -f)-( 't' + 't )(p -p ) > 0
1 0 1 0 1 0

La relation d 'Hugoniot donne alors 't'1 < ~ o' D' apres (HI) cela est
contradictoire avec Pl~Po' Sl>So' On a donc PI"? Po et, d'apres
(23.1), f 1>fo '
Pour etablir '<='1 <'to' on part de:
 - 173 -

Lichnerowicz

n en resulte puisque ~'<O:

p

ou a fortiori":

2 2 2
c (f 1 -f }-2~1(P -p )
o 1 0
>0

De la relation d'Rugoniot il resulte alors:

('t' - '?:
1 0
}{p 1-p 0 + 1..2/M.(k 1-k 0 }2l<0
j

soit 't' 1 < 't'o' ce qui demontre Ie tMoreme. On a par suite 0(1 < 0(0'

VI. ONDES DE CROC ET ONDES D'ALFVEN.

24. Ondes de choc et ondes d'Alfven.

Considerons Ii la traversee de L. un choc non tangentiel qui ne
soit pas choc d' Alfven
a} Nous nous proposons d'~tablir Ie lemme suivant
Lemme. n existe toujours au moins une dicection n orthogonale a 1,
u ,u1,h
-0 -
,h , .
'--()-l

Examinons les differents cas possibles:

 - 174 -

Lichnerowicz

1) Si 01. 0 011 f 0, il existe une direction n au moins orthogonale

aux vecteurs 1;, V et X avec:

(24. 1)

De (24. 1) il resulte que n est orthogonal a v et

o
v
l'
donc a u0
et u l . Comme:

(24.2)

nest aussi orthogonal a h 0 et hI

2) Si ex ~ =0, on a '\I =0 et il resulte du lemme 2 du
--0---.- /Yo
So. 16 que 1 est dans Ie 2-plan (u ,h ). Le vecteur Vest orthogonal
:5 0 0
a 1 dans ce 2-plan. II existe une direction n au moins orthogonale
a 1, V et u l . Cette direction est or~ogonale au 2-plan (u ,h ) et
o 0
etant orthogonale a V es u l est orthogonale a hI.
3) Si 0( o~c(lfO, il suffit d'echanger Ie rt'lle des indices ·0 et 1.
b) En xE L, introduisons une perturbation infinitesimale de l'etat
anterieur au choc. II en resulte une perturbation infinitesimale de
I'etat posterieur au choc reliee a la precedente par les relations ob-
tenues en differentiant les equatinns fondamentales de choc. II vient
ainsi:
 - 175 -

Lichnerowicz

[b rue/. +rbucAJ l()( =0

(24.3) [~(ho/ lo<)ue> +(hO( lotl~u,B -cS(~)h~ - ~ q)~J =0

[6(p2 ;)uf.> + ~2 ~bl' _y- ~ql> _6(ho/.lol)Jf _ho( 10/ £J3] =0

Adoptons en x un repere orthonormee {e(o()} tel que eO) soit coli-

neaire a. I et e(3) a. Ia direction n. Dans ce repere il vient:

3 3
u =0 u =0
o 1

Le systeme differentiel (24. 3) se partage en deux systemes dont Ie

premier contient exclusi vement Ies perturbations du 3, dh 3, soit:

(24.4)

(24.5)

Nous supposons que seuis

---~o--o
{3
ou,
(3
oh sont to avant Ie choc. Les
variables thermodynamiques n' ayant pas ete perturbees, il en resul-
te que, dans les etats respectivement ante rieur ou posterieur au choc
[ , de telles perturbations correspondent a. des chocs d'Alfven in-
finitesimaux, c'est-a.-dire a. des ondes d'Alfven.
C'onsiderons, dans l'etat anterieur a. L , une onde d 'Alfven de
type A. Le vecteur A0( etant invariant a. la traversee de cette onde
. 0 .13
infinitesimale, une telle onde porte en x une perturbation (JuoA' Jh oA )
 - 176 -

Lichnerowicz

telle que:

(24. 6)

De m~me une onde d'Alfven de type B porte en x une perturbation

3 3
( bu oB )' bhoB) telle que:

3· 3
(24.7) ~ ~u - Jh =0
;-0 oB oB

La superposition en x d Tune onde de type A et d Tune on de de type B

fournit une perturbation ( hu 3, ~h 03) arbitraire avec:
o

c) b Les vecteurs A(J. et Bot' verifient en x E L :

o 0

01. a ()/.
(24.8) !\ 1 =A
- +l:\ I P, ~
FfXo 1eX. = Fo -hoallo(
ocXTor 001 r
o o

On en deduit:

2
01. 2 a
-(h 10() =- 0(
o fA. 0

Convenons d 'orienter 1 de 1 'etat anterieur vers 1'etat posterieur au

choc L . On a alors a <0. Supposons, pour fixer les idees, b")' 0
0( 0(
et par suite jUolo(>O (resp.hl~ >0). D'apres (24.8), Ie vecteur B
o
(resp. B 1) est oriente par rapport a L du m~me ctlte que I. Quant
 - 177 -

Lichnerowicz

au vecteur Ao (resp. AI), son orientation par rapport a L. est celle

de I ou l'orientation opposee selon que Of 0 (resp. eX 1) est positif
au negatif. Pour 0( nul, A est tangent a L Si b etait < 0, les
rOles des vecteurs A et B seraient simplement inverses.

25. Compatibilite d'une onde de choc avec les ondes d'Alfven.

a) Nous examinons d'abord les cas OU Ie choc L envisage, non
choc d'Alfven, est tel que 01. 0 0(1=0. Supposons en premier lieu

D(
1
=0 0<
0
>0

Dans l'etat Y;, les ondes d'Alfven de type A et B qui abortissent

o
en X6 peuvent creer en ce point une perturbation ( C\U 3 , ~h 3)
z.. o 0
arbitraire et, dans (24.4), (24.5), on a puisque eX 1=0

2
a 0( 2
-(h 10\.) =0
2 1
r1

Pour que ces relations (24.4), (24.5) admettent une solution quels
que soient ~u o3, bh 03, il faut et il suffit que lIon ait identiquement
par rapport a ces variables:

soit:

(25. 1)
 - 178 -

Lichnerowicz

(25.2)

Si nous mettons (25.1) sous la forme:

2 -a 0(

ftl rl hI IQ{
I.
a
~! -
r
0
hoe I
o CI.

il vient d'apres (25.2):

~o< =0
o

soit ~ =0 ce qui est impossible. n y a incompatibilite de I 'onde de

choc L. avec les ondes d'Alfven
b} Supposons maintenant:

0( =0
o

2
ainsi f-H=c 1: 0 est> O. On a:

(25.3)
;>0

Avant Ie choc, une onde de type B et une onde de type A tangents a

L peuvent abortir en x E L creant une perturbation arbitraire
( bu 3, ah 3). Mais peuvent s'eloigner de x d'une part l'onde de type A
o 0
pour I 'etat Y port ant une perturbation (
A
0 0 0 A J u3 , J' h 3 ) verifiant:
 - 179 -

Lichnerowicz

(25.4)

d'autre part une onde d'Alfven de type B pour l'etat Y1 portant une
perturbation ( 6u~B' dh~B) verifiant

(25. 5)

Les relations (24.4), (24.5) relient cette derniere perturbation a une

perturbation ( Ju 3, Jh 3) anterieure au choc 'l.vec
o 0

Ainsi pour qu 'une perturbation ( l uo3, J"h 3) puisse

0
t!tre transmise a
travers Ie choc L et s 'eloigner, il faut et il suffit que:

(25.6)

Nous etablirons dans un instant que:

f31 a
ro
+ holol..J
""
f 0

L'onde de choc L. sera compatible avec la perturbation arbitraire

( ~ u 3, 6 h 3) s'il existe toujpurs une decomposition
o 0
 - 180 -

Lichnerowicz

avec:

r 3 3
oh = -13 C\U
oA fo oA

ou l'on a pose:

TI
B ~ +h~ I
r1 ro 0 '"

On est ainsi amene a etudier Ie systeme lineaire:

Le determinant de (25.7) est donne par:

131hQl0 IeX +V(.2, 02 ~ +

r
(l,
J-' 0
((.J,
J~ 1
a +ho( I )
roo( 0
4tH

1T+~= 0 0
o Po ~ + hoi I
n ro O d

( A ~ +hoi I )=0
ror 0 0<
o
 - 181 -

Lichnerowicz

Ainsi (25.7) n 'admet pas de solution pour toutes valeurs des seconds
membres et L n'est pas compatible avec les ondes d'Alfven.
II nous res t e a' mont rer que rf'.. 1 r_a +licx.0 l ctJ40 . S'1 non, on aural't
o

ce qui peut s 'eerire:

220
!h 12
r 0 (e 1:0+f -2- )
r
o

ou

II en resulte:

1 0( ( _1_
ttl 1 1;2
o
't;2
1

soit en divisant par 0<

1
Ie 2= "Z: - 7:
1 0

e2 "C! (h~ 10/

('{; 0+ 1: 1)= - - =H't' 09 2
a
 - 182 -

Lichnerowicz

Apres simplification il vient:

(heX I )2
1 d. 'C' =0
2 1
a

Cette relation implique h1 =0 donc H=O, ce qui est absurde.

Nous pouvons enoncer pour un choc non tangentiel.
Theoreme 1. Si z: est une onde de choc telle que 0( 0(
o
1=0, elle
est incompatible avec les ondes d 'Alfven a moins qu 'elle ne corre-
sponde a un choc d'Alfven (01. = 0( =0) .
.-....-----------'- 0 - 1-
Les chocs singuliers n 'ont pas de realite physique.
c) Supposons maintenant que pour Ie choc L envisage, on ait

Les ondes d 'Alfven de type A et B dans l'etat Y creent en x

o
une perturbation ( ~ u , Jh ) arbitraires;
3 3 <0(1 etant fO, Ie
o 0
systeme (24.4), (24.5) admet une solution unique definissant une
perturbation de l'etat Y1 pouvant s 'eloigner de x selon les ondes
d'Alfven A1 et B1 correspondant a cet etat. II y a compatibilite
de l'onde de choc avec les ondes d'Alfven
d) Supposons que lion ait

eX 1 < 0 < C<o

Les ondes d' Alfven de type A et B dans l'etat Y creent toujours

o
en x une perturbation arbitraire (~u 3 6h 3)
o 0
 - 183 -

Lichne rowic z

Mais seule s'eloigne de x, dans l'etat Y l' une onde d'Alfven de

tipe B port ant une perturbation verifiant

De (24.4) et (24.5) resulte:

11 est necessaire que (J u 3,

o
5h03) verifie la rEllation:

(25.3) A.2 ~ + A hOI 1 ) c\u 3 _( A ~ +holl )c5h 3 =0

r
J'" 0 rIo 0( 0 ''''I r o d < 0
o 0

Cette relation ne peut Nre une identite en (du 3 , Jh 3 ), sinon l'on

o 0
aurait

~ +ho( 1 =0
ro
(l..
roO(
o
 - 184 -

Lichnerowicz

et par suite 0< =0. Il yaa ainsi incompatibilite de l'onde de

o
choc L avec les ondes d'Alfven
e) Supposons enfin que l'on ait:

Avant le choc, seule une onde d 'Alfven de type B aboutit en x

creant une perturbation ( 0 u!B' hh!B) verifiant

3 ,3
t;
ro
Ou oB - oh oB
=0

Mais peuvent s'eloigner de x une onde d'Alfven de type A pour

l'etat Y et une onde d 'Alfven de type B pour l'etat Y 1 qui avant
o - 3 - 3
le choc correspond a une perturbation ( ~ u o , cS h 0 ). Nous posons:

3
cS u
oB
=&- u 03 + ~ u oA
3

avec, d'apres (25.3):

ou:
 - 185 -

Lichnerowiczz

NOBS sommes amenes comme au b a resoudre les equations:

- 3 3 3
~u o +~u 0 A'" =~u oB

'IT -~ u o3 - ro 3
t2. ~ uo
u A= ""o 0u 0B
J-
3

ou Ie determinant

Si 1'1: +h~lo( =0 on obtient aussi trivialement une solution.

II y g compatibilite de l'onde de choc 2: avec les ondes
d'Alfven. Nous enon90ns
Theoreme 2. Pour qu 'un choc non d' Alfven soit compatible avec
les ondes d' Alfven, il faut et il suffit que 0( 0~ 1 > 0
f) Les chocs envisages peuvent done ~tre decomposes en

chocs lents tels que

et chocs rapides tels que:

0<0( <0(
1 0

Pourtant de la relation fondamentale:

 - 186 -

Lichnerowicz

Pour un choc lent 0< ~ >.o<!, donc k~ < k! et par suite

lh 1 2< Iho\2, P our un ch oc rapl'd e, I es m",ga
'''' l't'"
1 "'s sont 'mvers",es,
'"
'
On a donc:
Proposition, Dans un choc lent, la grandeur de champ magnetique
diminue; dans un choc rapide elle augmente,

VII, VITESSES DES ONDES DE CHOC ET THEOREMES

FONDAMENTAUX,

26, Entropie et courbes isentropiques,

a) Un etat initial Y etant donne, nous considerons la suite des
o
etats Y verifiant

(26, 1) H(Y)=H(Y )=H kg( =k lX (avec ~O( '> 0)

o o 0 0

et etudions les points representatifs correspondants Z dans ~e plan

('t:', q); si H est > 0 la droite 't' = f HI c 2 est interdite, Dans Ie
plan ('t', eil, la courbe isentropique j correspondant a la valeur
S de l'eutropie est definie par:

2 ~
LLk 0(
1 roo
(26,2) p('t',S)+'2 -----
2 2
(c't: -}'lH)

En derivant (26,2) Ie long de ~ , il vient:

 - 187 -

Lichnerowicz

2 2 2
c LLk 0(
roo
0<.3

et en derivant line seconde fois:

On en deduit d'apres (8.5)

2_
(26.3) (~) = 1 M
d~j' -~
P

ou l'on a pose:

(26.4)

qui est strictement positif en vertu des hypotheses (Hi), (H 2).

Ainsi les courbes isentropiques -;J du plan ('t, cD sont convexes

Notons que la fonction 91 ('l:,S)definie par (26.2) verifie

1:'
(26. 5) ~ s= -
S
1:'p
 - 188 -

Lichnerowicz

b) Soit A une droite de pente m du plan ('t', q) issue du point

Zoo On a, Ie long de A

q-q
o
= m(t - or0 )

Lemme. En tout point Z de A ou S est stationnaire, qn a:

'!

I
-( e,2 , M)p
p 1:S ~

d 2S
A'lssi (-2 ~ est -< 0, sous les hypotheses (HI )~H2l.-
Par1:fns de la relation fonctionnelle:

q= 9' (t, S)

et derivons la Ie long de A par rapport au parametre 'l:. Il

vient:

dS
CD' + CD' ( - ) =m
11; IS d'C~

et en derivant une seconde fois:

d2S
dS dS2,
'f~2 +2q>~'t: ( d"ClA + <P~2 (d"C + <fs h _
( d'C.1 ~ -0
En un point Z de /1 ou (dS/d't:1.., s'annule, on a:
~ -., 2
(<P~2~ +(CP's)1 (~~ =0

On en deduit d'apres (26.5)

 - 189 -

Lichnerowicz

ce qui demontre Ie lemme.

c) Considerons dans Ie plan (t', q), la courbe d 'Hugoniot cjG
definie par l'equation ~{z, Z)=O. On deduit de (19. 1) que Ie
o
long de ~ :

dS d-
(26. 6) 2f ® {d,,.)',1
.. '11'0
+("t - ~0 )(~)
d'C''lIC.
-(q-q0 )=0

Aussi au point Zo' on a (dS/d"t:)~ =0. En derivant (26.6), on

obtient:

2
(26.7) 2f 9) (d ] +2{ d{f 9 )) (dS) +('t' -"t" )( d2q) =0
d't 1f(. d t: 11' dt; 'II' 0 d t:; 2 'll'

et en Zo' on a aussi (d 2S/d'L2)1' =0. Aussi la courbe d'Hugoniot

et l'isentropique S=S ont un contact du second ordre en Z . En
o 0
ce point, nous avons done:

(26.8)

En derivant (26.7), on obtient en Z

o
3 2-
d S +(~)
2f @) (g't:>3~ d -z;2 ~ =0

II resulte ainsi de (26.8):

 - 190 -

Lichnerowicz

( 11 M)
(26. 9)
2f@ ~
1; p
z0

D 'apres Ie contact entre ~ et l'isentropique -S (S=S ), on a en Z :

o 0

( d-r;) = ( d't') (0
dP1b dp'-j'='tpZ
o

et il vient en ce point:

On obtient ainsi en Z :
o

(26. 10)

et dans Ie voisinage de Z , nous avons Ie long de 1t

o

(26. 11)

ou Ie coefficient de (p-p )3 est positif. Nous enon<;ons:

o
Theoreme. La courbe d 'Hugoniot et l'isentropique S=S du plan (t:, ti)
o
ont au point initial Z un contact du second ordre. Pour un choc faible,
o
I 'accroiseement d 'entropie est du troisreme ordre par rapport a la
puissance du choc mesuree par I 'accroissement de pression
 - 191 -

Lichnerowicz

27. Conditions necessaires sur les vitesses des ondes de choc.

a) Considerons un choc qui, au point x de L , fait passer de Z o
a Zl' Nous avons vu que Ie long de la droite A joignant Zo a Zl'
la fonction 11<, (Zo' Z) est stationnaire en un point au moins Z~ situe
entre Z 0 et Z 1 et que d 'apres (21. 1), il en est de m~me paur S. Du
lemme du § 26, il resulte que Ie point Z-5 de b ou S est sta-
tionnaire est unique et correspond a un maximum strict de S sur b
La variable 't decroissant Ie long de /J.. orientee de Z 0 vers Z l'
on a:

dS
d't: '> 0 pour 't .:::.. "r:-t

et aussi:

(dS) < 0
d't: 0

On en deduit d'apres (20.2):

(27. l) 0< P(l) ') 0

o 0

b) La relation (12.5) peut s'ecrire:

2 r. 2 p.)
a oI..=(v) (vr
(27.2)
2 2 ~ 2 ot 2 2
(c rf+JPhl )((u 1«) -1 Ie*;) c c
 - 192 -

Lichnerowicz

Considerons un choc rapide (0 < 0( 1 < do). D 'apres (27.2), on a:

L. A L A
v
o
>v0 v
1
> v
1

De (27.1) il resulte que Pill est2)0 et Pill <0; d'apres (12.2), il

o z: 1
en est de m~me pour TT (L.:--i). On en deduit
c

Considerons maintenant un choc lent ( IXI (.0(0 <0). D'apres (27.2):

De (27.1) il resulte que P(l)o est <0 et P(I)l/'O .. On en deduit de

m~me

l:
v
o

Nous obtenons:
Theoreme. Sous les hypotheses de compressibilite (Hl)~), les
vitesses v 0 et
I. r.
----- - - -v1 -'-_
par _
rapport
.!....L_ _au_fluide
___ d_
'une
__onde
__ de_choc,
_

respectivement avant ou apres Ie choc, verifient vnecessairement

les inegalites suivantes:
1) pour un choc rapide

ML A MR!. ML A L MR
v n < V0 < v0 < vVo VI <.v 1 <;: v 1 < vI
 - 193 -

Lichnerowicz

2) pour un choc lent:

ML
v o <v 0
!.
-< v A <v MR
0 0

ou interviennent les vitesses magn~tosoniques lentes et rapides et

les vitesses d'Alfven avant et apr~s Ie choc.
(27. 1) r~sume la situation.

28. Theoreme d'existence et d'unicit~.

a) En inversant la. fonction 't' ="2:' (p, S), on obtient une fonctioR
S=S( p, 'Z;) exprimant l'entropie en fonction des variables p et 't: .
En raisonnant comme au § 8, on obtient:

~ 1
't;' 't;' =
p - S' S S!c
't:

Les conditions de compressibilit~ (H 1) se traduisent par les in~galit~s

S~>O, S~~o.
Par une nouvelle d~rivation, on obtient:

Nous supposons ici que la fonction S( p ,'t:') satisfait les hypotheses

(H 1) et (H 2) pour 't: '" ~ et paur des valeurs arbitrairement grandes
o . MR ~
de p. Nous nous proposons de montrer que Sl v < V <
A
V
- - ML 1: A 0 0 0
(ou v < v < v.A. ) c 'est-a.-dire 0< P(l) >0 les ~quations g~n~rales
o 0 0 0 0'
de c)1oc admettent une solution unique non triviale avec eXo 0(1 ,. O.
 - 194 -

Lichnerowicz

b) Considerons dans Ie plan ('t;, q) la branche de l'istmtropique J

correspondant a S=S qui est telle que
o
"C'~ "'C avec CXO( > O. Le long
-5 , nous avons d'aprus
0 0
de (19.1)

En derivant une seconde fois, il vient:

soit d'apres (26.3):

Aussi sur la branche consideree, (diG /d't"~ est» 0 et 1fe, (Zo~

est <.0.
c) Soit 11 une droite de pente m issue de Z . Il resulte de (19.3)
o
que Ie long de A:

( d~ ) 0.( dS )
(28. 1) d't: !::. =2f ~ d't' t.

Soit m
o
la pente de l'isentropique -&
0
en Z . De l'hypothese globale
faite, il resulte que Ie long de la branche envisagee de ~ , q peut
prendre des valeurs arbitrairement grandes. Or la courbe --! est
convexe. On en deduit que si m <m ; la droite
---0
A rencontre la bran-
 - 195 -

Lichnerowicz

che consideree de ~ en un point unique Z AfZ 0 et pour ce point:

Mais Ie long de A
, S(Z ):S(ZA)=S et S(~) est necessairement
o 0
stationnaire entre Z 0 et Z A en un point Z~ necessairement unique
d'apres Ie lemme et qui est un maximum strict pour S sur /l
Nous avons donc (dS/d't)b. < 0 en Zo et d'apres (28.1), (dlie/d~)A
< 0 en Z . Quand Z decrit
0 0 0
t::. de Z vers zA,1tG (Z ,Z) commence
par ~tre positif et il existe donc au moins un point Z
1
sur A entre
Z et Z pour lequel ~ (Z ,Z1)=0.
o A 0
Nous savons que sur Il, S croit de Z 0 a Z-&~ passe par un
maximum en Z~ et decroit constamment ensuite. D'apres (28.1),1-'
a Ie m~me comportement et Ie pojnt Z 1 est necessairement unique.
a tout nombre m <m , correspond un point unique Z 1 -de
Aussi
"--"--------'-0
la courbe d'Hugoniot de Zo avec 't'1< 'to ~o( 1 ~o ';> 0) tel que

m=q1'~)rt' 1-~'
Nous voyons que m fournit une parametrisation simple de la bran-
che consideree de la courbe d 'Hugoniot. Nous notons qu'il reesuHe de-
(19.3) que Ie long de .-x:

et que par suite S croit quand m decroit.

c 2a Z ~,P
Pour m=-(J{--(O, Z1 satisfait (17.3) et 'l'I'<>(Zo,Z1)=0; Z1 etant
connu, t 1 letlOllPl Ie soit et f1 est donne par exemple'par la
relation d 'Hugoniot. De (17.5), on deduit la valeur de I hl, de (17.1)
 - 196 -

Lichnerowicz

celle de u ~I d et de (17. 2) la valeur de ~ I 01. (qui a Ie signe de

h~lo<)' Les equations (17.1), (17.2), (17.3), (17.4), (17.5) ont aussi
une solution unique telle que DC 10(0 >0
d) Composons la tcondition

et la condition necessaire trouvee au § 27:

D'apres les resultats du ~ ~O, on a:

(28.2)

Or

2
c't' -- ~2
p
r

En divisant (28.2) par "t" 0(, il vient:

p

Or d'apres Ie ~ 26 a):
 - 197 -

Lichnerowicz

1 1 c 2}Lk 2
---
'7:;' ~- 0(
p P

On obtient aussi:

1
(28.3)

Au point Z il vient:
o

2 2
c a
= m -
o

2 2 0(
On voit aussi que c a /1 10( < m 0 est )equivalent a ()( 0P(l) 0> 0,. N ous
enon9 0ns
Theoreme. Si la fonction S( p, 't') satisfait les hypotheses (H 1).JB 2) pour
'to" '- ~ et pour des valeurs arbitrairement grandes de p , a tout etat
y
--0
verifiant P(l)
·0-0
0( >0
correspond une solution unique non triviale
des equations de choc telle que Oi. 1 ~ > O.
Les inegalites figurant dans cet enonce s'interpretent immediatement,
comme au 927, en termes de vitesses de l'onde de choc et des vitesses
magnetosoniques et d' Alfven.
 - 198 -

Lichnerowicz

VIII. RETOUR AUX RAYONS MAGNETOSONIQUES.

29. Etude de la direction N ~.

a) A partir des equations fondamental de la magnetohydrodynamique,
on deduit que pour une onde infinitesimale L non engendree par des
lignes de courant (uO{ 10( fo), on a d'abord:

Aussi on a l'invariance par ~ du scalaire et des deux vecteurs:

0(
a=ru 10( fO V~

Les invariants scalaires de choc mis en evidence aux ~ ~ 16 et 17

sont donc aussi des invariants pour l'operateur b de discontinuite
infini te simal e.
Nous supposons l'onde envisage non pas d'Alfven (0. fO), mais
magnetosbnique de telle sorte que

P(l)=O

D'apres les ca1culs du ~ 20, ceci peut s'ecrire:

(29. 1) r -21 (c2a21::+f\hI210"\)I)+oIloilcx. =0 (c 2a 2-c +)-,-lh,2 lo(lo(f O)

r

b) Transformons de m~me l'expression du vecteur NP-> qui donne

la direction du rayon (voir (14. 3))
 - 199 -

Lichnerowicz

(3 2 a2 \3> 2 Ih 12 ~ $
N =2c't (0'-1) 2" arv +(c 't+fl 2 r)(1 l~)ar v-
r r

Compte-tenu de:

il vient:

Or d'apres (29.1) (c'est-a.-dire P(l)=O)

2 2 d 2
(h l~) l~l --0
2 a ( 1) (2_
c 't 2 D - + c ...
r
+r Ihl
-2- ~
r)
)It?l
p - fA
a
2
f

et l'on a:

N~ =c2~ :: (0' -1)ar v ~+r; (Ii' loi)l fIr v f3

2
c 'est-a.-dire en tirant 1-1/r de (29.1):
 - 200 -

Lichnerowicz

;? 2 2 ~ ~
Nj"-" = - _c_a_'C'_l_l-ll..."_ _ ad.r v~+u ~a (he{ la>l )l\' lIZ) VI'"
2 2
c a 't +r- Ih \2 1~ If I ,

On deduit de (16/5):

La direction de N ~ est donc celle du vecteur proportionnel:

En utilisant b=f If lei et en divisant par 1:';, on obtient Ie vecteur

colineaire a N F:

ou lIon a pose:

(29.3)

30. Action de b sur la direction du rayon.

a) Pour que la direction de N~ soit invariante par ~ , il faut
et il suffit que:
 - 201 -

Lichnerowicz

II en est donc en particulier ainsi pour b=O, c'est-a.-dire si Ie champ

magnetique est tangentiel
b) Cherchons a. evaluer ~ Q. II vient:

(30.1) 'C2~Q=-c2a2~~+tI~Iff-lh\2_1~1't-'-lh\2 ~

D'autre part, d'apres I'invariance de l par ~ , on a:

c 'est-a.-dire:

En reportant dans (30.1), on en deduit, compte-tenu de &8=0

soit, comme:

2 I
c"C p =-
.1....:..!..
2
r

on a:

(30.2)
 - 202 -

Lichnerowicz

Pour que ~ Q soit nul, il faut et il suffit que

c 'est-a.-dire: que:

(39.3)

Nous avons ainsi etabli que, contrairement aux resultats concernant

les rayons aessocies aux ondes soniltues et waux d'&des d'Alfven
(en magnetohydrodynamique), la direction des rayons associes aux
ondes magnetosoniques n 'est pas en general invariante par I 'operateur
b de discontinuite infinitesimale.
II y a invariance seulement si Ie champ p magnetique est tangentiel
ou si la relation (30. 3) est satisfaite. Mais celle-ci n 'est autre que
(13.1). Cette relation correspond donc au cas singulier ou Ie cOne P(I)=O
admet deux generatrices doubles.
A I 'approximation classique la relation (30.3) s 'ecrit:

I \2 2
(30.4) ~h =rv.
 - 203 -

Lichnerowicz

BIBLIOGRAPHIE

[d F. Hoffman et E. Teller, Phys. Rev. t.80, p.692, (1950).

[2] A.H.Taub, Arch. Rat. Mech. Anal. t.3, p.312, (1959).
[31 Y. Choquet-Bruhat, Astron. Acta t. 6, p.354, (1960)
[ 4] W. Israel, Proc. Roy. Soc. A. 259, p.129, (1960).
[ 5] Pham-Mau-quan, Ann. Inst. If. Poincare t. 2, p.151, (1965).
[61 A. Lichnerowicz, Relativistic HydrodyIIllmics and Magnetohydro-
dynamics, W. A. Benjamin New York (1967).
[ 7] A. Lichnerowicz, Ann. Inst. Poincare t. 5, p. 37, (1966).
[8J A.Lichnerowicz, Ann. Inst. Poincare t. 7, p.271, (191i7).
[9J A. Lichnerowicz, Comm. Math. Phys. t. 12, p.145, (1969).
UOJ A. Lichnerowicz, Comptes rendus Acad. Sc. Paris t. 268, p.256,
(1969).
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(C.I.M.E. )

VARIATIONAL PRINCIPLES IN GENERAL RELATIVITY

A. H. TAUB

Corso tenuto a Bressanone dal 7 al 16 Giugno 1970

 LECTURE I

VARIATIONAL PRINCIPLES IN GENERAL RELATIVITY

1. Introduction

In these lectures we shall derive the Einstein field

and the equations of motion for uncharged and charged self-

gravitating fluids from variational principles. We shall

also see how singular hyper-surfaces (shock waves) and the
equations governing their behavior may be treated by means
of these principles. In addition we shall show how the

"second variation" problem is related to the discussion of

the stability of the solutions of the Einstein field equations.
Before taking up these problems we shall discuss some
general properties of variational principles and show how a
form of the principle of equivalence may be used to formulate
in general relativity a field theory described in special
relativity by a variational principle involving a Lagrangian

fUnction which is a scalar function in Minkowski space-time

and which in turn depends on tensor fields and the first
derivatives.
Given a four dimensional space-time with a metric tensor

and a tensor field Vu (written as ¢A) defined

vm
over the space-time. Let be a scalar function formed from
the metric tensor of the tensor field ¢A and the derivatives
of these tensor fields.

= Jf ;:g d 4 x
V
where V is a fixed but arbitrary four volume in the space-time,
 - 208 - Taub

is a functional of the metric tensor and the tensor field

For given g~v' I may be evaluated in any coordinate

system in the space-time by using the assumed transformation
properties of the function I . Thus under the

I (x) = I (x(x*» = I *(x*)

and we may write

I(g , ~A) = I(x) r-g d4 x

J
V

= I(x(x*» r-g (x(x*» J d4 x*

J
V*

(1.1)

= I * (x*) ;:gw d 4 x*
J
V*

where J is the Jacobian of the transformation of coordinates,

that is,

J =

we of course assume that J # 0,

and V* is the same four volume as V but now expressed in the
x*~ coordinate system.
If the tensor fields g~v (x) and ~A (x) are embedded
A
in families of tensor fields g~v(x;e), ~ (x;e) such that
g~v(x) = g~v(x;o) and ~A(x) = ~A(x;o) the functional
 - 209 -
Taub
I(g ¢A) becomes a function of the parameter e. Thus we have

ICe) = I (g~v(xie) i ¢A (x e)).

An example of the embedding referred to above is given by

= g~v (x) + e h ~v (x), ¢A (xi e ) = ¢A (x) + e ~A (x)

Where h is an arbitrary symmetric second order tensor field

~v

and ~A is a tensor field with the same transformation properties

as ¢A .

A variational principle is said to apply if

(dI) = I' (0) = 0

de e=O

for arbitrary
dg
(~) = g'~v(O)
de e=O

or

(d¢A) = ¢' A( 0)
de e=O

or both.

2. Embeddings Induced by Coordinate Transformations.

Let g'-' (x,~) and ¢*A(~*) be the components of fixed
~v

tensor fields and ¢A in the x*~ coordinate system and

let the equations of transformation to another coordinate

system x~ depend on the parameter e, that is, let

(2.1 )
 - 210 -
Taub

Then since

and

~A(X) 'V ~lll lln (x)

\)1 \)m

1
foOl a n d/ l dxlln dX *1, .1. dX* m
= ~ 1 -----
1 1m dx*Ol ox i : On dx\)l dX \)m

A
'V ~ (xje)

the and ~A also depend on the parameter e. Such

dependence will be said to be induced by the coordinate trans-

formation (2.1).

It may be readily verified that if

= + ~ll (x)

then

= - ~ll;\) - ~\);ll (2.2)

where the covariant derivative is taken with respect to the

metric

= gll \) (x *; 0 ) ,
 - 211 -
Taub

and

(2.3)

These expressions for g~v(O) and ¢,A(O) when the e

dependence is induced by a coordinate transformation will be

used below.
When the volume over which I(e) is defined is independent
of e, that is the limits of integration in the right hand
side of the first of equations (1.1) are independent of e,

then the limits of integration in the other integrals occurring

in that equation depend on e. Thus if f(x) is a scalar

function depending on the g~v(x;e) whose dependence on e is

due only to the fact that this dependence has been induced by

a coordinate transformation then the expression for I,

4
He) = J f ;g d x
V
depends on e only because the integrand depends on e.
Whereas the equivalent formula

He) = J f'~!gil d4 x'~

V*
depends on e only because the limits of integration which

determine the volume over which the integral is to be carried

out depends on e. The integrand is by assumption independent
 - 212 -
Taub

of e. We may write the last equation for I(e) as

I(e) = f fi: (/'(x)) Vg 1:(x":(x» J- l d 4 x

V

and now the integrand depends on e through the dependence

of the functions x*(x) on e. It follows then that

1'(0) = f (fE;°);or-g d 4 x (2.4)

V

3. The Principle of Equivalence

If in the special theory of relativity a field theory can

be described by a variational principle then there exists a

scalar Lagrangian function £ which depends on the dependent

variables of the theory and their derivatives. We shall assume

that the dependent variables are given by fields which behave

as tensors under Lorentz transformations. The case of spinor

fields can be treated in a fashion similar to that described

below. Thus we are considering the case where the function £

is a scalar function of a tensor field and its derivatives

under Lorentz transformations. We postulate that the dependent

variables of the theory are tensors under general coordinate

transformations in Minkowski space-time and write £ in such

a coordinate system as a function of the tensor field ~A, its

covariant derivatives and the metcic tensor evaluated in the

general coordinate system.

We now form the integral

1= f(R-k£)r-gd 4 X. (3.1)
V
 - 213 -
Taub
In this integral, the g~v are no longer assumed to be the
metric of a flat space-time and hence one can compute a
combination of it, its first, and second derivatives which is
a scalar curvature of a space-time with metric tensor g~v'

The volume of integration entering on the right-hand side of

equation (3.6) is an arbitrary one, and K is a constant which
may be related to the Einstein graviataional constant.
Next we form the function I(e) by embedding the tensor
fields and into the families g (x;e) and
~v

$A(x;e) as discussed earlier and study the conditions for

1'(0) = O.

We shall denote

ag I = I' (0)

when

$,A(O) = 0

and

15$ I = I' (0)

when

g~v(O) = O.

In general we shall have

I' (0) = ag I + 0$ I
 - 214 - Taub
The general relativistic formulation of the field equations

determining the field tensor ¢A will be given by

6¢ I = o (3.2)

and the Einstein field equations for ir,e gravitationa.l field

created by the sources dependent on the tensor field ¢A will

be taken to be

= o (3.3)

It is evident that the equations obtained from equations

(3.2) are those that would hold in a general coordinate system

in Minkowski space-time. Hence in view of the principle of

equivalence which states that some non-galilean coordinate

systems in Minkowski space-time are locally equivalent to the

equivalent to the presence of gravitational fields, equations

(3.2) should represent the equation determining the tensor

field ¢A in general relativity.

In the subsequent discussion we shall see that equations

(3.3) lead to the Einstein field equations in the form

= -KTIJ\! (3.4)

and TIJ\! will be a symmetric tensor determined from the

tensor field ¢A.

4. Notation
In order to avoid an excessive use of indices, we employ

the following notation

¢A (4.1 )
T
n
 - 215 -
Taub
The symbol '\, is to be read as "stands for. n

A '\, ¢ol ... a

n ,mj].J
¢ ;lJ
'1

= ¢ol ... a
,m,lJ Oi_lpoi+l .•• On
Ttl + ~¢ol rOo
l '1'''' m P].Jl

_ r¢ol ... a
n,
1 ... 'j-1P'j+l ... P
' mr T.].J (4.2)
]

where the comma denotes the ordinary derivative and

1
P
r a, = g PA 2' (gaA ,, + gAT ,a - go, , A)' (4.3)

We also write IjJA' and IjJlJ where

A

'1
,m
IjJA '\, IjJ
01 a
n
(4.4)
'1
,m lJ
1jJ].J IjJ
A '\, 01 a
n
and

'1
,m al ... a
n
IjIA¢A '\, IjI
al a
n
¢
'1
,m
(4.5)

(4.6)

That is, the quantity given in (4.5) is a scalar and that in

(4.6) is a vector.
We consider £ as a function of three sets of variables:

and define
 - 216 -
Taub
1 )1V
e~V = ~+ "2 g J:. (4.7)
ag)1V

aJ:.
qA =
a¢A (4.8)

p)1 = aJ:. (4.9)

A -::7\
a¢ ;)1

In each case the remaining two sets of variables are kept

constant in the partial differentiations.

We also define

(4.10)

It follows from equations (4.3) that

= = g PA -21 (g' + g' _ g' )

OA;1 A1;0 01;A

(4.11)

and from (4.2) that

A
[(~) ] = ("A
;)1
),
de e=o;)1 'I'

~
0 " .0, IPO'+l"'O
¢ 1 1- 1 n r' a,1
1 11, .. 1 m P)1
 - 217 -

Taub
where the variations are produced by varying the ¢A and the

gjJ\)'
Hence

PjJ (,!,A )1 =
A 'I';jJ
(4.12)
=

where

(4.13)

It follows from this equation that

It is a consequence of the definition of the Ricci tensor.

RjJ\) = rOjJo,\) rOjJ\),O + r jJo

P rO
P\)
rjJ\)
P rO
po (4.15)

and equation (4.1l) that

dR
( jJ\) rIa rIa
"""'"de e=o = =
RI
jJ\) jJo;\) jJ\);O

and that

jJ \) po jJ P \)0 1
= [ (g g
]
- g g )gjJ\);P;O (4.16)
 - 218 -
Taub

5. The Euler Equations

The relation between the energy-momentum tensor for the

~-field which appears in the Einstein field equations and the

variation of £ with respect to the g~v was first pointed

out by Hilbert [1] as is noted in Pauli's classical discussion

of the theory of relativity[2] where additional references may

be found. In this section we shall derive the Einstein field

equations and the equations that must be satisfied by the

~-field.

It may be verified that

= f [(G~v + Ke~v) g~v + K(qA~A' + pX (~~~)')

V
(5.1)

_{(gPO ~v p~vo)g' } ].r;:Qgd 4 x

g - g g >IV;P ;0

where G~v is defined ln equations (3.4), in (4.8) and

pX in equation (4.9).
It follows from equation (4.12) that equation (5.1) may

be written as

-1'(0) = f [(G~v + Ke~v) g~v + K(qA~A' + pX (~A');~

V

+ KM~vO , ) {( pO ~v P~ VO) g' } ] ~g d 4 x

g~v;o - g g - g g ~V;P ;0
 - 219 -
Taub

On integrating by parts this in turn may be written as

-1'(0) = J {[G~v + KT~vJ g~v + K(qA - p~;~) ~A'} ;:g d4 x

V
( 5•2)

+ J [KH~va g'~v + KP~ ~A' - (gPag~v_gP~gva)g'~v;pJ;ar-g d4 x

V

where the symmetric tensor

= = (5.3)

and M~va is defined by equations (4.10) and (4.13).

By requiring 1'(0) to vanish for arbitrary variations

which vanish on the boundary of the region V we obtain the
Euler equations

G~v KT~v
+ = 0 (5.4)

and

~
FA - qA - PA;~ = 0 (5.5)

Equations (5.4) are the Einstein field equations with a matter

tensor.given by equation (5.3). Because this tensor arises

from the variation of I with respect to the gravitational

field we call the gravitational matter tensor. Note

that even in the Minkowski space-time T~v is different from

e~v in a general coordinate system.
Equations (5.5) are the equations for the field ~A. They

may be obtained from the special relativity equations ln a

galilean coordinate system by replacing every ordinary derivative
 - 220 -
Taub

by a covariant one. They obviously reduce to the equations of

special relativity in case the tensor g~v is the metric tensor

of Minkowski space-time.

6 Conservation Laws

In this section we shall use a technique similar to that of

E. Noether [3] to relate the tensor T~v occurring above with

a not-necessarily-symmetric tensor which, will be called the

inertial stress energy tensor and to derive various conservaTion

laws. The inertial stress energy tensor of the field ¢A is

defined by the equation

Q'
t P = (6.1)

and it is evident that in general

Further we see that as a consequence of the equations satisfied

by the ¢ field, equations (5.5), we have

when £ does not depend explicitly on the coordinates. In

special relativity we have In general can

at most depend linearly on and the curvature tensor in

view of the Ricci identity. We shall evaluate this dependence

in equation (6.7).
 - 221 -
Taub
The computations carried out in the sequel make use of

the fact that when g'~v and ~,A are given by equations
(2.2) and 2.3») 1'(0) must be given by equation (2.4) with
f = R - Kt. That is, we must have

1'(0) = -f«R o
Kt)~ )'o~-g
,
r-
d
4
x
V

when equations (2.2) and (2.3) are substituted into equation (5.2).

When the former equations are substituted into the latter

one we use the identity

2G~v~ _ ( po ~v _ p~ ov)(~ + ~) =
s~;v g g g g ~;v sV;~ ;po

equation (26.1) and the definition

to obtain

1'(0) = -f[(R
V

+ K ( 6 • 2)

where N~vO is defined by equation (4.14).

 - 222 -
Taub
Note that

= o

when equations ( 5 .5) are satisfied.

Since the tensor Nl1v O' is antisymmetric in v and 0' we have

(~ Nl1vO') = 0
11 ; va

That is,

(~ Nl1v O' + ~ Nl1va) = 0

I1;V 11 ;V;O'

or

In view of this equation and equation (2.4) we may write equa-

tion (6.1) as

J[(2T I1V + Ll1v) ~11;V + FA¢A;p~P];:g d4 x

V

On integrating this equation by parts we obtain

(F ~A _ 2T 0' _ L 0' ) ~P;:gg d4 X

.. A'" ;P P;O' p;a
V
 - 223 -
Taub

When the tensor field ¢A is a solution of the Euler equa-

tions (5.5) the above equations become

= o
(6.3)

and
-2fT\1\) = o
;\)
V
(6.4)

Both equations must hold for arbitrary volumes and arbitrary

vectors ~.
)J

If is a Killing vector, that is, satisfies

~)J;\) + ~\);)J = o

then equation (6.2) inplies for such vectors that

= [t o~p + NP\)O~ ] o (6.5)

P ., "p;\) ;0 =

Killing vectors need not exist in a spa~e-time satisfying the

Einstein field equations. As is well-known there are ten linearly

independent Killing vectors in Minkowski space-time--the generators

of the inhomogeneous Lorentz group. Equations (6.5) for theBe

Killing vectors are the conservation laws of energy momentum and

angular momentum discussed by Belinfante and by Rosenfeld. Since

the tensor t PO satisfies these conservation laws, in special

relativity, we may consider it as the inertial energy tensor.
Since equation (6.4) must hold for arbitrary vectors and

arbitrary volumes, we must have

 - 224 - Taub
Til \I = o ( 6 . 6)
;\1

and

= o (6.7)

Equation (6.6) is a consequence of the Einstein field equations.

We have now shown that it follows from the invariance properties

of £ and the ¢A field equations, equations (5.1). Thus

equations (6.6) and (6.&) hold in special relativity.

Equations (6.7) relate two energy tensors, the gravitational

one and the inertial one til\l.

It follows from equations (6.6) and (6.7) that

= (6.8)

This equation, which may be derived directly from the definition

of t PO and equations (5.5), reduces to

= o

ln the case of special relativity.

It should be noted that as a consequence of equation (6.6),

equation (6.3) may be written as

J[(2T pO + tpO + NpO~\I) ~p];O~ d 4x = a

v

which in turn may be written as an integral over the hypersurface

S boundary the volume V, namely
 - 225 -
Taub

f[2Tp O + tpO + Np;v

ov ] scP no dS = 0 (6.9)
S

where nodS is the element of volume ~n S, and ~p is an arbitrary

vector. Equation (6.9) may be used to relate the time rate of
change of the three-dimensional volume integrals of T~ and t~ by
choosing the hypersurface surface S to consist of the hyperplanes

t = constant and t + dt = constant.

7. Generalizations

The results obtained above may be readily generalized to

the case where there are a number of~-fields present. In such

a case for each such field there will be an associ~ted T~v and a
corresponding t~v. The right-hand side of the Einstein field
equations will contain the sum of the T~v and this tensor will be
related to the sum of the t~V by equations analogous to equa-
tions (6.6).
In case the ~-field is a spinor field a similar discussion
to that given above can be made. The special relativistic
Lagrangian must first be generalized by replacing ordinary
derivatives of the spinor field by covariant ones. The variations
in the met·ric tensor may be performed by varying the generalized
Dirac matrices which satisfy the relation

= 2g~V'
 - 226 -
Taub
It should be pointed out that if we define the scalar

x = =

then equations (5.5) become

0 ax
PA;o = - acpA (7.1)

and

A ax
,
cp '0 = 0
(7.2)
apA

These two equations are similar to the Hamiltonian equations

for particles.
Note, however, that X is not tp
p
nOr 4 Thus if in
t4 •
virtue of equations (7.1) we call X the Hamiltonian it is

related to t 0 through the equations

p

x = t P
P
+ 3£ (7.3)

Thus the connection of the Hamiltonian with the stress energy

tensor involves the Lagrangian function.

References

1. D. Hilbert, Nachr. Ges. Wiss. Gottingen, p. 395 (1915).

2. W. Pauli, Theory of Relativity, Pergamon Press, London,
p. 158, (958).
3. E. Noether, Gottingen Nachtrichten, p. 235, (1918).
4. L. Rosenfeld, Acad. Roy. Belgique, ~, p. 6, (940).
5. F. J. Belinfante, Physics 7, p. 887 (939).
 - 227 -
Taub
LECTURE II

A VARIATIONAL PRINCIPLE FOR PERFECT FLUIDS

8. Co-Moving Coordinates
The general discussion of variational principles given
above may be applied to the derivation of the equation of motion
of a self-gravitating perfect fluid and the Einstein field

equations for the case when such a fluid is the source of the
gravitational field. In order to make such an application
appropriate field variables ¢A must be chosen and a Lagrangian

function £ must be specified. We shall use as field variables

the rest density of the fluid p, a variable related to the rest
temperature S of the fluid and a set of functions which
characterize a three-parameter congruence of curves, the world

lines (particle paths) of elements of the fluid. We shall

not vary these quantities arbitrarily but shall restrict

the variations of p, the particle paths and the metric so

that the mass of the fluid is conserved under the variation.

The use of co-moving coordinates will enable us to

represent the congruence of particle paths and their variation
in terms of the metric and its variation. Thus, some of the

field variables ¢A are absorbed into the tensor g~v' The

use of this special coordinate system greatly simplifies
the calculations. The resulting simplifications are one
implication of the fact that as a field theory relativistic

hydrodynamics is special in that the equations of motion of

the fluid are a consequence of the Einstein field equations
and are not independent of them. That is, in the case of a
 - 228 - Taub

perfect fluid, we are dealing with a situation where equa-

tions (5.5) are implied by equations (5.4).

We shall be dealing with a one-parameter family of

space-times with metrics g~v(x;e). In each space-time of

the family there is a congruence of curves determined by the

solutions to the ordinary differential equations

dx *~
as = U*~ (x *;e) (8.1)

where the are the labels assigned to events in the

space-time in an arbitrary coordinate system in which the

,', '/;
metric has components g ~v(x ;e), and the are the

components of the velocity four vector of the fluid in this

coordinate system. They satisfy

(8.2)

We may write the solutions of Eqs. (8.1) as

(i=1,2,3) ( 8• 3)

where

are required to be the parametric equations of a hypersurface

i
The four variables ~ ,s which we shall denote as

form a comoving coordinate system in each of the space-times.

Eqs. (8.3) which may be written more generally as

(8.4)
 - 229 - Taub

with
xi = ~i,
(8.5)
x 4 i
4
= x (~ ,s;e)

may be regarded as the transformation between the x * co-

ordinate system and a general comoving one which uses the
x~ as labels for events. Eq. (8.1) is then to be under-
stood as

* ,~
= u ~(x (x;e);e) ( 8• 6)

where In the partial differentiation the xi are kept con-

stant for when these variables are fixed a particular world-
line is selected.
In the general comoving coordinate system we have the
components of the four velocity vector given by

= u
*0 * dX~
(x)---wo
dX

as follows from Eqs. (8.6). Hence

(8.7)

where are the compoenents of the metric tensor in the

comoving coordinate system. Eqs. (8.7) are a consequence of
Eqs. (8.5) and (8.2).
We shall be using a comoving coordinate system in each
space-time of the one parameter family of space-times with
which we shall be concerned. In a particular one of these
the metric tensor in the comoving coordinate system will be
 - 230 -
Taub

written as g~v(xje). The tensor

( 8• 8)

with x~ kept constant will measure the change in the metric

tensor evaluated in the comoving coordinate system at an event

labelled by the coordinates x~, produced by a change in the
parameter e. Similar statements will apply to other tensor

fields which depend on e. In particular we shall have

Ul~ = (8.9)

That is, the transformations given by Eqs. (8.4) for various

values of e, produce comoving coordinates in each of the

space-times associated with that value of e.
We shall use the notation

.*~(x *' je)

V = (8.10)

with kept constant, where the V*~ are the components

of a vector field in a general coordinate system using the
labels x *~ • It is of interest to determine the relation
between V'~ and V • To do this we define
.~

(8.ll)

where is given as a function of x and e by equa-

tions (8.4) and x~ is kept constant under the differentia-

tion. Since

6~
p

must hold for all values of e, it follows from the differentiatior

 - 231 -
Taub

of this equation with respect to e keeping xll fixed that

a ax v ax v a~*11
ae (--rp)
ax
= - --r,;- -,-;;-
ax 11 ax P
(8.12)

that
,\ v
a (~)
ae = _a_~*v (8.13)
ax P ax P

From the transformation law of vectors we have

*v * dxll
= V (x (x;e);e)~.
dX v

On differentiating this equation with respect to e, keeping

xll fixed we find

V' =

In virtue of Eq. (8.12) we may write this as

= (8.14)

where

= (8.15)

*v with
and is of course the Lie derivative of the vector V
1')J
respect to ~ . It may be shown by similar arguments that
for any tensor the operation of taking the prime derivative
of the tensor compenents differs from the transform of
taking the dot derivative by the appropriate Lie derivative

of the tensor.
 - 232 - Taub

In particular for a scalar we have

f'(x;e) = (8.16)

where

f*(x*;e) = f(x(x *;e);e).

For the metric tensor we have

I
g aT = (8.17)

It follows from equation (8.9) that

~s is to be expected. If we apply equation (8.14) to the

four-velocity field, we find that

U*v U*0 )Z" *

"'o;p
(8.18)
- ( g*vo - U*0U*v )U * Z"
*p
o;p'"

Hence

= o (8.19)

as a consequence of equation (8.18) and the fact that equa-

tion (8.2) holds.

9. The Variations of the Field Variables

The particle paths which are a three parameter C0nbruence
of curves, are described by equations (8.4). In these equations
4
the xi (i = 1,2,3) label a particular particle and x

measures a "co-moving time" which is allowed to differ from

 - 233 -
Taub

the proper time along the world line of a particle (cf. equa-
tion (8.5». The vector field defined in terms of the
former equations describes the variations in this congruence
of curves. That is, if x*p(x;o) is the description in the

starred coordinates of the position of and time when particle

i 1+
x is located at the comoving time x , as this particle moves
along an unperturbed path, then x*p(x;O) + e~*p(x;O) describes
the corresponding event as the particle moves along a
perturbed world-line.
The quantities measure a variation in the metric

tensor in the sense that gOt(x;O) + eg~t(x;O) describe two

almost equal space-times. In each of these space-times, the
particle paths, whether perturbed or not, are described by the
curves xi = constant. Because we are allowing general space-
1+
time hypersurfaces to be described by the equations x = con-
stant, a variation in gOt produces a variation in UV; tte
unit tangent vector to the particle paths.
Equations (8.17) and (8.18) describe the variations pro-
duced at the event labeled by of the metric tensor and
the four-velocity vector field when these tensors are varied
in the comoving coordinate system and then evaluated in the
x*p coordinate system. It should be noted that although f our
arbitrary functions, the enter into equations (8.18),

the manner in which they enter is restricted so that equation

(8.19) holds.
We now turn to a discussion of the variations which we
shall allow in the rest density p and the rest temperature ®.
 - 234 -
Taub

If ~(x;e) is the rest density of the fluid and U~ is its

four-velocity, then the conservation of matter is expressed
by the equation

= __1__ (f:g p U~) = o (9.1)

f:g ,~

In a comoving coordinate system when equation (817) holds we

may integrate this equation to obtain

pf:g = vi44 Ho(x1,x2,x

3
,,e) (9.2)

The function Mo which is independent of x 4 measures

the amount of fluid in the hypersurface x4 = constant between
the world-lines labeled by xl and xt + dx t , x 2 and x 2 + dx 2 ,
X3 and x3 + dx 3 • The requirement

MI = 0
o

is then the statement that for each of the space-times g (x;e)

. . ~v

and for every perturbation of the world-lines of the particles,

the amount of matter in the region de~cribed above, is constant.
It follows from the results of the preceeding section
and equation (9.2) that the requirement MI =0 is equivalent to

(9.3)

In the x*~ coordinate system, we have

1
'2 U
*0 *1 •
U
*
(gOl +
* *
~O;l + ~t;o) = o.
(9.4)
 - 235 -
Taub

The rest temperature ~ and the rest specific entropy

S of a fluid are defined by the equation

i&:ls = de: - ~ dp (9.5)

where p is the pressure and e:(p,p) is the rest specific

internal energy given by the caloric equation of state of the

fluid. In this equation ID(p,p) is determined as an inte-
grating factor of the right-hand side of the equation and

S is then determined up to a constant of integration.

It follows from equation (9.5) that if p and pare
functions of a parameter e, then

®S' = e:'-~P' (9.6)

p

Instead of considering the function ® as a field

variable we introduce another function a defined by the
equation

(9•7)

In a comoving coordinate system

1 <la
® = ----4
Ig 44 ax

and
®' = (9.8)
 - 236 -
Taub

10. The Lagrangean For A Perfect Fluid

We shall define this function as

SS) 00 .1)

where p, E, ® and S are defined as above. Then the

variational principle discussed in Lecture I may be written

as 1'(0) =a where

I = I g - 2Kl r 00 .2)

with
Ig = IRA d\ 00.3)
V

If = Ip(C 2 + E - «I)S)A d x
4
00 .4)

It then follows from the results of the preceeding section

that the equation

is equivalent to

2 I' I[T\.IV g , 2pSU\.la' Jr-g d 4x , 00.5)

r = r \.IV ,\.1

where
T\.IV pCc 2 + E + £)U\.lUV _ pg\.lV 00.6)
r = p
and use has been made of equations (9.3), (9.6) and (9.8). We
may integrate by parts the last term in equation (10.5) and
obtain
21' = I[T\.IV Q , + 2CpSU\.I) a'J;:;g d 4x
r r .:l\.lV ;\.1

(10.7)
 - 237 -
Taub
We now turn to the calculation of I'g • We have

]..Iv
r-gR = r-g g R]..IV

and

R]..IV = rO
]..IO,V rO P O
]..IV,O + r ]..10 r pv
P rO
r ]..IV po (10.8)

where
P 1 pI..
r ]..IV = 2"g (gA]..I,V + g AV,]..I - g]..IV,A ). (10.9)

Hence
,P 1 PAC ,
r ]..IV = 2"g g A]..I;V + g'A]..I;V - g']..IV,A ) (10.10)

where the covariant derivative is taken with respect to the

metric g]..lv' In deriving this result we use the fact that

g ']..IV = -gOT
'g0]..l gTV , (1o.n)

as follows from the equations

= 6]..1 o·
It is a consequence of the above that

rIO )
= ]..I P va), ]
]..IV;O g g g ]..IV;P ;0

(10.12)
Therefore
I'

(10.13)
where

G]..IV = R]..IV I ]..IV R

2" g •

The Euler equations are obtained by requiring 1'(0) =0

,
for arbitrary g]..lv(O) and u'(O) which vanish on the boundary
of the region of integration. Since the second integral in
 - 238 -
Taub
equation (10.13) has an integrand which is a divergence of a
vector field, it vanishes for such variations and the Euler
equations become

= o (10.14)

and

= o (10.15 )

where T ].1V is given by equations (10.6).

F
In view of the definition of and the definition of

specific entropy, equations (10.14) and (10.15) are equivalent

to (10.14) and

(10.16)

This is so, because equations (10.14), together with the

Bianchi identity

G].1V = 0
;V

imply that

T 2
].1V
F;v = [p(c + e: + £)U].1U
p
V pg].1v] . = 0 (10.17)
'v
Hence

= = o

In view of equation (9.5) we may write this equation as

Cc 2 + e: + £)(pU v) + pU v 8S = 0
p ;v ;v
or as

(c 2 + e: - ®S + ~) (PUV);V + ®( PUvS) = 0
;v
Hence equation (10.15) implies equation (10.16) and conversely.
 - 239 -
Taub

Equations (10.14) and (10.16) are the Einstein field

equations for a self-gravitating fluid when matter is conserved.
Equations (10.17) are the equations of motion of the fluid.
Equations (10.16) and (10.17) reduce to the special relativistic
equations of conservation of matter, energy and momentum when
the space-time is Minkowski space. They may therefore be
properly called the generalization of these equations to
general relativity. The special relativistic equations re-
duce to the Newtonian ones in case the fluid particle veloc-
ities are small compared to the velocity of light.
It should be noted that through the use of the comoving
coordinate system in which both the unperturbed and perturbed
i
particle paths are represented by the curves x = constant.
The variation of the particle paths does not appear in the
evaluation of I'. Thus the field variables corresponding
to the particle paths do not appear in this integral and there
are no field equations for these variables. The equations
describing the particle paths in a general coordinate system
are of course derivable from equation (10.16) and (10.17).
The latter ones may be written as equations (10.15) or as

pS U~
= 0 (10.18)
,~

and
2 _
= P ,v (gV~
')UvU~ U~Uv) (10.19)
p(c + ~ ;v
where
i = E + E (10.20)
p
 - 240 - Taub

LECTURE III

SINGULAR HYPERSURFACES

11. The Existence of Shock Waves

It has been shown [1, 2] that solutions of equations
(10.14) for g ,p, p, and u~ may be found which reduce
~v

to solutions of problems in fluid flow in the special theory

of relativity when K = O. If we further take the limit of
these solutions when c ~ • , we obtain solutions to problems
in classical hydrodynamics. Further, the approximation
method which was used is such that any special relativistic
solution of a plane-symmetric problem in hydrodynamics could
be obtained by such a reduction process.
In particular those solutions of the special relativistic:
equations describing the motion of perfect fluids which are
physically unacceptable in certain regions of Minkowski
(flat) space-time have their counterpart among the solutions
of equations (10.14). Such solutions are associated with the
formation of shock waves in special relativity and in classical
theory [3]. For this reason it was suggested that the theory
of a perfect fluid in general relativity must allow for the
existence of shock waves. That is, we must contemplate the
existence of three-dimensional hypersurfaces in space-time
across which there may be discontinuities in the stress energy
tensor, the g~v' and their derivatives.
If such hypersurfaces are to be considered, then we must
consider equations (10.14) as holding on each side of such a
hypersurface, and we must supplement these equations by
 - 241 -
Taub
conditions which relate the values of the g , the
~v

derivatives of these quantities, and the stress energy tensor

on both sides of such a hypersurface. The relations that must
hold between the components of the stress energy tensor across
a hypersurface of discontinuities must be the generalization
of the Rankine-Hugoniot equations of classical and special
relativistic hydrodynamics [3J.
It is the purpose of this lecture to derive and discuss
a set of conditions of the type described above for a space-
time in which the matter present is a perfect fluid. The
method used is based on the existence of the variational
principle discussed previously.
If the field equations and the equations of motion, the
conservation equations, can be derived from a variational
principle, then we may generalize the variational principle
by allowing the region of integration involved to include
regions of space-time which contain hypersurfaces across which
the matter distribution and the metric tensor and its de-
rivatives are discontinuous. We may even vary these singular
hypersurfaces. In the classical theory such a generalization
is known to give the classical Rankine-Hugoniot equations [5].
It will be shown below that such a generalization of the
variational principle leading to the field equations and the
equations of motion for a space-time containing a perfect
fluid leads to a general relativistic generalization of the
Rankine-Hugoniot equations which reduces to the appropriate
equations in the special relativistic and classical limits. We
shall also obtain conditions that must be satisfied by the
 - 242 -
Taub
g~v and their derivatives across hypersurfaces of discon-
tinuities. Such conditions have been discussed by S O'Brien
and J. L. Synge [6] and by A Lichnerowicz [7]. The results
given below are obtained in a general coordinate system.
When the coordinate system is chosen to be that used by

O'Brien and Synge, the equations obtained reduce to those

given by them.
A transformation of coordinates with a discontinuous

second derivative may be show [8l to reduce these conditions

to the requirement of Lichnerowicz, namely that g~v and
the derivatives of the be continuous in the new co-
ordinate system.
The equations governing the behavior of thin shells of
material discussed by W. Israel [9] and A. PapapeTrou and
A. Hamoni [10] may also be derived from a variational
principle. These equation differ from those describing shocks
because in the former situation, the integrand occurring in
the integral defining the variational principle is such that
its evaluation involves an integral over the singular hyper-
surface. In the case of shocks no such hypersurface integral
occurs directly in the definition of the variational principle.

12. The Generalization of the Integral I

The variational principle we shall now consider will
involve the integral I defined by equation (10.2). However,
we shall assume that there exists a three-dimensional hyper-
surface L which divides the region of integration V into
two four-dimensional regions VI and V2 in each of which
 - 243 -
Taub
the integrand exists and is integrable. There mayor may not
be a contribution to I involving an integral over r.
Thus

I = f (R - 2K£)r-g d 4x + §
VI + V2
02.1)

where
4
IA = f (R - 2K£) r-g d x, (A= 1, 2) 02.2)

VA

S =
I'd r , 02.3)

dr is the invariant three dimensional volume measure on the

hyper surface r and I is a function of the g).1V and the

field variables of the fluid evaluated on r. In case I = 0,

we will see that the singular hypersurface is a shock wave or
a boundary between two regions of space-time across which no

matter flows (slip-surface). In case f *- 0 the singular

hypersurface will be shown to be a thin shell.

In a general coordinate system with labels a

family of hypersurfaces ~(e) may be defined by the para-

metric equations

= x *).1 (Ct,B,y;e) 02.4)

where the parameters Ct, B, y define a point on the hyper-

surface and the parameter e labels particular hypersurface.

The vector field

=
 - 244 -
Taub

where a, Band yare kept constant in the differentiation

is then defined on the hypersurface ~(e).

We observe that in the comoving coordinate system we have

where ae-
dx ll
is evaluated for fixed a, 6, y and e by use of
equations (12.4) and the inverse of equations (8.4).

The coordinate system used to write equation (12.1) may

be an arbitrary one in space-time. In case it is the co-

moving one the variation in I due to a variation in the

singular hypersurface is given by

(12.5)
=

where

(12.6)

and we have used the notation

[h] = li m {h(x ll _ E=Il) - h(x ll + E=Il)} = (12.7)

E+O

Equation (10.13) which holds in each of the regions

Vl and V2 may be used in the evaluation of II where I

lS given by equation (12.1). Under a variation of the

variables gllv' the fluid field variables and the singular

hyper surface we then have

 - 245 -
Taub

I'

+f[{;:g(g~VgPO - gP~gVO)g~v;P + 2Kpa'SU o ;:g}AO]do

(12.8)

+f::~=~d~ +f[f:g(R - 2K£)=OAO]do + f 1~Vg~vd~

~ ~ ~

In this equation the last term arises from the variation of

S in equation (12.1) with respect to the field variables.
The third and fourth terms come from the variation in I due

to the variation of the hypersurface ~ and the first two

terms come from the application of equation (10.13). The
divergences occurring in that equation have been integrated
under the assumption that
, vanishes on all boundaries of
g~v

VI and V2 other than the hypersurface ~.

13. Shock Waves

We shall first discuss the conditions

I' (0) = 0
~v
in the case f = 1 = O. That is the case for which

4
I = II + 12 = (R - 2K£)!=g d x.
f
V +V
l 2

In equation (12.8) we may set

, =
g~v

where g~v represents the variation in g~v due to the

 - 246 - Taub

variation of the metric tensor at a fixed x*~ and the

second terms represent the variation in g~v due to a

variation of the world-lines of the fluid particles. After
an integration by parts, equation (12.8) may be written as

~v ~v •
I' = -f {(G + KTF )g~v
Vl +V 2

+ I[r-g(g~VgPO - g P~ g)g
vo •
~V;P
Ao]do - I[I=gT o~·
F ~
~
Ao]do

+ f[1=g{R(=o +
~ (13.1)

We now require that 1'(0) = 0 for g~v ' and a'

which vanish on the hypersurface ~ and are otherwise arbi-
trary. It then follows that in region VI and in region V2
equations (10.14) and (10.15) hold. It 1S a consequence of
the first set of these that

T~v
F;v = o (13.2)

1n regions VI and V2 . Thus no additional conditions are

obtained by setting to zero the coefficient of ~~ in the

first integral on the right-hand side of equation (13.1).
We restrict the variations of a' on the hypersurface

(13.3)

In view of the fact that equations (10.14) hold we may write

this equation as
 - 247 -
Taub

<13.4)

This equation, which relates the variations of the hypersurface,

the variations of the particle paths and the variations of the

temperature is the general relativistic analogue of the

relation which was previously discussed in the classical

theory [5J. It arises because in the theory of shock waves,

one does not have a conservation of entropy flow across shocks

and hence one does not have the generalisation of equations

(10.15) holding. Instead one requires a generalisation of

the equation describing the conservation of mass.

If equation (13.3) holds, it follows from equation (13.1)

and the requirement that I'(O) = 0 for arbitrary ~~ that

we must have

(l3.5)

These equations give the conditions that must be satisfied by

the discontinuities in the stress-energy tensor of the fluid
across the hypersurface ~. They are a natural generalization

of equations (13.2) and may be derived from them by rewriting

the latter equations as

for arbitrary vector fields f~ and integrating these equa-

tions over an appropriate volume in space-time containing the

hyper surface ~. These equations have been discussed in some
detail in reference [3J.
 - 248 - Taub

14. The Conditions on the Metric Tensor Across a Shock

We now turn to a discussion of the second integral in

equation (13.1). In order to evaluate this integral we

introduce coordinates adapted to the hypersurface r. We

shall treat the case for which

< o Cl4.1)

and indicate the changes that must be made if this condition

does not hold. We shall assume that 1n the region Vl (and

in the region V2 ) the hypersurface is defined by the

equation

1
x1 = x0
Cl4. 2)
i
x = xi(oj) a, i, j = 1, 2 , 3

1
where x is a constant.
o
The induced metric on the hypersurface r is then

i
=
lJ V
g llV x I1.Y IJ.da da
i j
= y .. da da j Cl4. 3)
1J

where we may choose the ai so that

Cl4.4)

are three vectors tangent to the hypersurface. The unit

normal to the hypersurface is

= Cl4. 5)
 - 249 -
Tauh

where

N2g11 = -1 (14.6)

and
11
n Il x l'1 = 0 (14.7)

If we define yij so that

ij
Y Yjk = 6ki (14.8)

we then have

gllV = _nlln V + yijxll"xV" (14.9)

1 'J

where

nil = gllVn v = Ng ll1 (14.10)

Equations (14.6) and (14.7) may be written as

v
gIlV x Il I,n
1
= 0

and
g nlln v = -1
IlV

These equations determine the components of nil to be

1 1
n = if = - Ng ll

1yij
ni = -N g1j = Ng l i

2 ij 2
N2 = g 11 (-1 + y g4i g4j)

It follows from equation (14.9) that

(14.11)
 - 250 - Taub

where y is the determinant of the y ..•

1J
We note that under the transformation of coordinates

given by the equations

1
we have on the hypersurface where x

_1
n =
-].
n =

as follows from the equations

cr3x ll
= n

Thus if nl *0 we may always choose the coordinate system

in VI so that

nl _ 1 = = o
That is, so that

= 1.

= o.

Such coordinates are of course Gaussian coordinates based on

the hypersurface r. We shall restrict ourselves to such

coordinates in the remainder of this section.

In Gaussian coordinates, we have

and hence

= o (14.12)
 - 251 -
Taub
Therefore
\!
= n n)J;\! = o (14.13)

The second fundamental form of the hypersurface ~ is

defined as

S"l, ,
lJ
= = S"l ..
Jl
04.14)

The last of the above equations holds in view of equations

04.12). It follows from the definition of S"l" and n
lJ )J
that in the Gaussian coordinate system

-n f~, 1
S"l ..
lJ
=
p lJ =. -f"
lJ
= 04.15)

Since the four vectors n)J, xli are linearly independent

we may write

)J p ij)J p
= n \! (n ; pn) + S"l x "x
1
"g
] P\!
•

In view of equation (14.13) this becomes

ij v
}I,' x''J g
p
= 04.16 )
S"l l PV

where

Hence

= S"l = Yij S"l lJ

..

We now apply the above results to the evaluation of the

integral

J = I ;:g(g)JagPa P va'
g)Jg)g
)J\!;pa
d 4x
Vl
 - 252 - Taub

where is an arbitrary tensor defined in VI and such

that it vanishes on all boundaries of VI except the hyper-
surface ~. We may evaluate J in an arbitrary coordinate

system in VI in particular in the Gaussian coordinate

system based on ~. Then we have

J =

where equation (12.6) defines and is the unit

normal to the hypersurface r and d~ is the invariant
volume element of this hypersurface, that is, we may evaluate
J in the Gaussian coordinate system based on ~ where we have

;:

(nPx lJ .x\)_ v I' P ij' • A

=
11 fJ
- n X1.xl.)y
1 .) (g llV P gA/llP)

ij' ij • ij
= Y goo 1 + Y P . . + gun - goon
1J, 11J 1J

where

P.
1
=

P. . = P.. Pk {koo}
11 J 1,-J 1J

and the {k.. } are the Christoffel symbols formed from the Y.. o
1J 1J
Hence we have

J = J (y ij·g .. 1
1J,
~
 - 253 -
Taub

Thus the second term in equation (13.1) may be written as

= f[1Y(y ij·g1J,
.. 1
~

(14.17)
The vanishing of I I given by equation (13.1) for
.
g)lV such
that

-g g g)lV ) +
( ~)lV = = o (14.18)

but otherwise arbitrary on ~ and for arbitrary .. 1

g 1J, then
implies that

(14.19)
and that
[ If( yijn - n ij )] = 0 (14.20]
Equations (14.19) imply that the induced metric on the
hypersurface ~ takes on the same values when computed from
the metric in Vl as when computed from the metric in V2 •
Since we may use Gaussian coordinates based on the hyper surface
~ in each of these regions we have that in these coordinates
the metric tensor is continuous. Equations (14.20) then
imply that the second fundamental form of the hypersurface ~

takes on the same values when this hypersurface is regarded

as one in Vl or one in V2 • This result obtains without
the restriction imposed on
.
g)lV by equation (14.18). However

that equation is the statement that the variations in volume

element, as measured by the variations in the ;:g, vanish
on the hypersurface ~.
 - 254 -
Taub
15. Thin Shells

The equations that determine the behavior of thin

shells may be derived by considering the conditions for which

I' = 0 where I' is given by equations (12.8) with f ~ o.

We shall discuss the case for which P =0 inside VI and

Vw but not one the hyper surface ~. In that case we have

£ = 0 inside VI and V2 and equation (12.8) becomes

I' =

when we set =0 = 0 and f = O.

It then follows by using the argument given above that.

in the regions VI and V2 we must have

GlJ V = 0 (15.1)

Then on setting
g +, +,
lJV '>lJ; v '>v.; lJ

we find that

I' = f[l=g(glJVgPO g PlJ g)g

va • A ]do +
lJV;P a
f lJV'glJV ;Y dl:
t

l: l:

(15.2)

The requirement that I' = 0 for arbitrary vectors

SlJ which vanish on the boundaries of VI and V2 other than

l: requires that we consider the last term in this expression.
 - 255 -
Taub

We shall evaluate it in the Gaussian coordinate system used

in the previous section. In such a coordinate system we have
as the non-vanishing Christoffel symbols

r~.
1J
= -no1J. =

r..
k
1J
= ..
{ k },
1J

Hence we have

~l;l = ~l ,1 ~l;i = ~l , i ~ nk .
k 1

k . k
~i;l = ~ v, l-~kn.1 ~i;j = ~.
1,J
~k {. . l- [1 n ..
- 1J
1J

The last of these equations may be written as

when we have used the symbol to denote the covariant

derivatjve with respect to the three-dimensional tensor y ..

1J
We may then write

+ 1
ij (~·I· - ~ln ..
» d~
1 1 1J

The vanishing of this integral for arbitrary ~~ and

then implies that

11
1 = 0
li
1 = 0
ij n .. (15.3)
1 = 0
1J
ij
1 • = 0
IJ
 - 256 - Taub

These are the equations which govern the behavior of

the thin shell described by the tensor 1~V which vanishes

off the hypersurface L. We may use the results of the

preceeding section to determine the equations which determine
1~v in terms of the geometry of the hyper surface L. If we
use equation (14.17) we may write equation (15.2) as

I' = J(r ~v·g IY + [1Y(y

ij'
g .. 1 nij )g• .. ]}dL
~v lJ, lJ
L

If we impose equations (14.18) on the g~v we find that

II = 0 for arbitrary
.
g~v and
.g~v,l subject to these condi-
tions if equations (14.19) hold and

= (15.4

Equations (15.3) and (15.4) are the equations given by Israel [9]

and Papaetrou and Hamoui [lOJ for the theory of thin shells.
 - 257 -

Taub

References

1. A. H.
pp. 454-

2. , Approximate solutions of the Einstein equations

for isentropic motions of plane-symmetric distribu-
tions _?Lp~d~ct fluids, f>hysical Rev. Vel Ie 7 II ~ t: I)
~r~~if-100
3. , Relativistic Rankine-Hugoniot equations,
Physical Rev., vol. 74 (1948), pp. 328-334.
4. , General relativistic variational principle for
perfect fluids, Physical Rev., vol. 94 (1954),
pp. 1468-1470.

6. Stephen O'Brien and John L. Synge, Jum~ conditions at

discontinuities in general relatlvity, Communications
of the Dublin Institute for Advanced Studies, Ser.
A, no. 9 (1953).

7. A. Lichnerowicz, Theories relativistes de la gravitation

et de l'electromagnetisme, Masson et cie, Paris, 1955.
8. A. H. Relativit ,

9. W. Israel, Singular Hypersurfaces and Thin Shells in

General Relativity, II Nuovo Cimento, 44 (1966)
pp. 1-14.
10. A Papapetrou and A. Hamoui, Couches Simples de Materie
en Relativite Generale, Ann. Inst. Henri Poincare
~ (1968), pp. 179-211.
 - 258 - Taub
LECTURE IV

FLUIDS OBEYING AN EQUATION OF STATE

16. Equations of State

In this lecture we shall derive a simpler variational

principle from which we may derive the Einstein field equations
for a self-gravitating fluid that satisfies an equation of

state of the form

p = p(w) (16.1)

where p is the pressure and w is the energy density of the

fluid. That is, in terms of the quantities used previously

we have

w = p(c 2 + d (16;2)

where p is the rest mass density and e is the rest specific

internal energy.
We have previously made use of the fact that function

e = dp,p),

the caloric equation of state, describes the nature of the

material with which we deal and serves to determine the
temperature e and the entropy S by means of the equations

®is = de + (16.3)

Hence for every fluid we may express the pressure as a function

of two thermodynamic variables, say wand the entropy S.
 - 259 - Taub
Thus for every fluid we may write

p = pew,S). 06.4)

The assumption made above is that all thermodynamic variables

are functions of one of them, say w. This assumption is

satisfied in case the fluid motion is isentropic, that is

S = S
o

where So is a constant. It is also satisfied in other

~ircumstances.

In the general situation when equation (16.4) holds the

velocity of sound ca is determined by the equation

= 06.5)

where c is the special relativistic velocity light and the

entropy is kept constant in the differentiation occurring on
the right-hand side of this equation. If there is a family of

flows such that the thermodynamic variables are functions of a

parameter e as well as the coordinates, then it follows from
equations (16.4) and (16.5) that

p' = a 2w' + (.~)

as w
SI 06.6)

When we restrict this family by the condition that

SI = 0,

we shall say that the family of motions is an adiabatic family

(or that the perturbations which distinguish one member of the
family from another member are adiabatic perturbations).
 - 260 - Taub
For such families we have

(16.7)

17. Integration of the Equations of Motion

In case equation (16.1) holds we may derive from the
equations of motion of the fluid,

TjJ\)
F;\I = 0, (17.1)

expressions for various components of the metric tensor in the

comoving coordinate system in terms of the thermodynamic
variables. In this coordinate system we have

=
1 .rjJ
-- u4 (17.2)
Ig 44

and

= =

(17.3)

Equations (17.1) are in general equivalent to the

equations

(w + P)U\)'\) + U\) (17.4)

, W
,\)
= 0

and
(w + p)UjJ;\)U\) = (o\)
jJ U\)U ) (17.5)
,
:f: \) jJ

When equation (17.1) holds there exists a thermodynamic

function o(w) defined up to a constant by the equation
do dw (17.6)
o = w-+p
 - 261 - Taub

Hence equation (17.4) becomes

(17.7)

In case equation (16.1) is equivalent to the statement that

the entropy is constant we have

o = p

and equation (17.7) is the conservation of mass.

In the comoving coordinate system equation (17.7)

becomes

= o

This equation may be integrated to give

(17.8)

where f(x i ) is an arbitrary function of the variables

4
xl, x 2 and x 3 but independent of x •

If equation (16.1) holds

do + d(w + p)
~ =
w + P
o w + P

Thus, in the comoving coordinate system equations (17.5)

become

= (17.9)

These equations are identically satisfied when ~ = 4. Their

integrability conditions may be integrated to give

g'4
(~_1_) ,
o ~ ,J
44
 - 262 -
Taub
where the F .• = -F .. are arbitrary functions of the three
lJ Jl
i 2 4
variables x ,x and x 3 but not of x and are such that

F .. k + F' k . + Fk · .
lJ, ] ,l l, ]
= 0

Hence we must have

_0_, c.(x j ) +
= w+p 0 ¢
l W"+P' ,i

(17.10)
i
where the c· are arbitrary functions of x and may
.l
i 4
be a function of x and x • The solutions of equations
(17.9) then become

= (17.n)

It is no restriction to take

¢ = constant

k = 1

for if these conditions are not satisfied we may make the

coordinate transformation
_4
x =
-i
x =

where

The x~ coordinate system is also a comoving one and in it

we have

U. = = (17.12)
l
 - 263 - Taut.

= =
w+ P ° (17.13)

IS. The Vorticity Vector

The three functions c.(k j ), which enter into the

1

components of the metric tensor in the comoving coordinate system

and which may be determined from the initial conditions satis-

fied by the motion of a self-gravitating fluid, determine and

are determined by the amount of rotation in the fluid. This

may be seen by examining the vorticity vector

1 ElJ\lOTU U
= (1S .1)
0,[
;:g \I

in the comoving coordinate system. We have

k (_o_)2 1 kij
v = c ..
w+ k E
1, 1
v"-g
OS.2)
1+ 2 1 kiJ' ckc . .
v = (_0_)
W+P;:g
E
1,]

It is evident from these equations that the necessary and

sufficient condition for vlJ = 0 is that

c.. c.. = 0,
I,] J,l

that is c. be the gradient of a scalar. In that case there

1

exists a comoving coordinate system In which.

The world lines of the fluid particles are then orthogonal to the

hypersurfaces xl+ = constant In the comoving coordinate system.

In case the flow is isentropic, that is S is a constant, we

 - 264 -
Taub
have

o 1
w+ P =

where

i = E: + .e.p

is the specific enthalpy of the fluid. Further we have in

the comoving coordinate system

1
=

These are the results obtained earlier [1].

19. A Variational Principle

In the comoving coordinate system used above, we may

use equation (17.13) to determine a thermodynamic variable

such as p as a function of g44' that is as a function

of the coefficients of the metric tensor. Variations of the
metric tensor in the comoving coordinate system will then
produce variations in the pressure. Thus with the notation
we have used earlier we have

p' =

as follows from differentiating equation (17.13) with respect

to e in the comoving coordinate system. This equation may

be written as

(19.1)

Thus we have p as a function of and an expression

 - 265 -

Taub
for the variation of this function.

Now consider the variational principle based on the

integral

(19.2)

It follows from the results given earlier and equation

(19.1) that

I' = I<G~V + KT~V)g~V;:g d4x

(19.3)
I< g~V g po - g ~p g va), H
+ g ~v;po -g d 4x

where
T~v
F
= (w + p)U~Uv _ g~Vp. (19.4)

This variational principle then leads to the field

equations
G~v = (19.5)

with equations (19.4) holding. In the latter equations

p = pew) and hence one of the consequences of the Bianchi
identities

=
is that

where a is a function of w (or p) defined by equation (17.7).

 - 266 -

Taub

LECTURE V

A VARIATIONAL PRINCIPLE FOR CHARGED FLUIDS

20. Introduction
It is the purpose of this lecture to discuss a variational
principle from which the equations governing the motion of a
gravitating charged electromagnetic fluid with dielectric
permitivity, magnetic permeability and conductivity are

derived. In case the conductivity is zero or infinite, we

will have a 'holonomic' variational principle. In other cases
the principle will be 'non-holonomic' in that not all terms in
it will be de~ivable from the variation of a function of the
dependent variables which enter into the problem. This is to
be expected, since for non-vanishing and finite conductivity
one is dealing with a non-conservative problem since ohmic
heat is involved and for such problems holonomic variational
principles do not exist.
We begin with a discussion of the electromagnet~fields.

We shall use the notation given by A. Lichnerowicz [1] for

the Minkowski formulation of the Maxwell equations governing

 - 267 -
Taub

these flelds. They are descrned by two antisymmetric tensors

H~V and G'lV' If u~ is the four-velocity vector of the

fluid satisfying
(20.1)

Then the electric field eu ' the electric induction du'

the magnetic field ha and the magnetic induction bu as
measured by an observer whose world line is given by a
solution of the equation
dx~ ~
(fT"" :: U ,

that is, one who is at rest w1th respect to the fluid, are
given by

(1IJ
(20.2)
ibf) = u '!lUf3

where 1:: J:I and

v 1 ~5
Huf3 = ¥u:3~5H
(20.3)
Haf3 _ lEaf3~5HIJ
- 2" ~5 '

E(((3~5 and EUf3~5 are pure imaginary tensors defined by the

equations

EUf3~5 :: -rgEUf3~5
(20.4)
EUf3'Vr 5 ::
-1. af3'V5
.{gE r

and the E'S are the Levi-Civita alternating tensor densiUes.

\'Ie have introduced the pure imaginary quantities in order to
preserve the commutativity of various methods for manipulating
 - 268 -
Taub

indices. It follows that

It follol-IS from equations(20.2) and(2o .3) and the prop-

erties of the EIS that

ibf3Ef3AIlV = _u"R llv _ u~vA _ uVHAIl

Hence
HIlV = ulle V - uVe ll - ibf3UAEf3AIlV (20.5)
Similarly we have

GIlV = u~Ldv _ uVd ll - ihf3UAEf3AIlV (20.6)

For the purpose of simplifying the ensuing discussion

we shall assume simple constitutive equations, namely

where A, the dialectric permitivity of the matter and Il,

the magnetic permeability are scalar functions which may
depend on the coordinates. The situation in which A and
Il are functions of the vectors ea and ha respectively
and of other variables characterizing the matter may be
dealt with in a manner analogous to the discussion given
below.
It follows from equations (20.5), (20.6) and(2o.• 7) that

( 2o.• 8)

where
AIlVa'T _- -.l ( all 'TV ov 'Til) + 2
21l g g -g g
1 (A - 1) (u ~Lu ag 'TV -u Vu ag'!jl
~

- u Ilu 'Tg av + u vu 'Tg all) (20· .9)

 - 269 -
Taub
Hence
(20.10)
It is a consequence of equations (20.5) and(20.6) that

H,LV H !-tV ,I v
!-tv = 2(e Ve v-b Vbv ) H H!-tv = 4ie bv
v
a~LVa
!-tv = 2(d vd v_hvh v ) a!-tvG
!-tv = 4id Vh v oo.n)
G!-tv H = 2i(b Vd +h Ve )
J
a 11VH
!-tv = 2(e vdv-h Vbv ) !-tv v v
 - 270 -
Taub

21. The l'lax~leJl Equations

These equations are
v IlV
(21.1)
H
j v = 0
and
a[J·V = JIl (21.2)
iV

where the semi-colon denotes the covariant derivative. The

vector JIl defined by equations (21.2) is the electric
current and as a consequence of equations (2L2) satisfies
the conservation equation
JIL = 0
ill
\ole may write

where E is the proper electric charge density and III

is the conduction current. Ohms law then is given by the
statement that

"lhere C5 is the conductivity of the fluid. Equation (21.5)

is the third constitutive equation for the matter. It may
be replaced by a more general (non-linear) lal'l without
affecting many of the results given below.
Equations(21 .1) are equivalent to the statement thnt

HAlliV + HIlViA + HVAjll = 0 • (21.6)

These equations imply of course that there exists a four-

potential such that
 - 271 -
Taub

(21.7)

The Minko\'lslci stress-energy tensor is defined as

1"IlV = HllpGPV + .kllvHPO"G (21.8)
lj~ pO"

and satisfies
(2L9)

It follows from equations(20.5) and(20.6) that

IlV
1" = (21. 10 )

where
Vv = le Ph O"u 1"EpcJ'rV (21.11)

w.v = id(lb CYu1"E PO"1"V (21.12)

and hence satisfy

ullv
Il
= ullvi Il = 0 (2L 13)

When the constitutive equation~(2Q.7) hold we have

and
<21.14)

It fo11ov/s from the Maxwell eoua tions, when these are

written as equations(21.2) and(21.6), that

,-IlV = HllpJP + aPO"gllV (H + ~l )

;v VpjO" 2 pO"jV

+ .kIlV(H GPO" _ G nPO")

lj~ pO" jV po' ;v
 Taub

In view of equations C21 .6) and(2D.8) we may write this equation

as
~v = H~ JP + 1 ~vAPoa(3H H
't" j V P "IF j v Pcr-u(3

On making use of equation(zo.9) we may in turn write this

equation as
(21.15 )

+ u A gP~(w -v)
j P A A

If we define
W= - 1(e 0d +bPh ) (n.16)
2 P P
and
JlV = + (g~V_ul!uV)W + e~dv _ h~v (21.17)
then equation (2.10) becomes
't"~v = Wu~u V _ u~v v _ wlll- ollV (21 .18)
and

{21.19)

Substituting from equation(21.19) into the left hand side of

equation(21.15) and multiplying the resulting equation by
Ull we obtain

(Wuv-VV).v + aIlvu = (0+ 1 1.. uD)eOe

, ~jV 2 jp 0"
(21.20)

This is in the form of a conservation equation and relates

the change in ii, the energy density of the electromagnetic
fields as measured by an observer at rest with respect to it,
the divergence of v Il and the electromagnetic stresses crltv.
The vector w~ is related to the momentum of the electro-
magnetic field.
 - 273 -
Taub

22. A Variational Principle for the ~1axwell Equations

Let us assume that the space-time with the metric tensor

g~v 1s given and is not influenced by the presence of the

electromagnetic fields described by H~v and G~v' This is

the case in special relativity where the space-time is the
Minkowski space-time and in a galilean coordinate system
the g~v. are constants. Vie shall not restrict ourselves
to galilean coordinate systems nor require that the curvature
tensor of space-time vanish.
Consider the integral over an arbitrary region of space-
time defined by

IE = J fxrLvH~vJi d4x = J *A~vaTHaTHlJ.vJg d4x (22.1)

where H~v is assumed to satisfy equation(21:6), that is,

it determines a four potential ~IJ.' We shall assume that
~IJ. in addition to depending on the space-time coordinate,
also depends on a parameter e. For the present, we assume
that the glJ. v and u~ are independent of e. Then

where
~, = ( d~~)
~ de e=O

If we require that ~~ vanish on the boundary of integration

then we have by integrating by parts
 - 274 -
Taub

The variational principle which requires that

(22.2)

where Jil is given by eruations(21.4) and(21.5) is then

equivalent to equations (21.2). In case III is absent from
the latter equation, equation(22.2) may be derived from
looking for the extrema of the integral

The second term on the rlght hand side of equation(22.3)

represents the interaction Lagrangean.
It is l'lell known that if the Lagrangenan of a variational

principle depends on the coordinates only because of its

dependence on the field variables being varied, then there
exists a second order, non-symmetric tensor t~ which
satisfies a conservation theorem. For the Lagrangean
~ = bllVHllv
we have

and

or
t~ = - 1"~ + (GIlO'cpp);1l - G~~CPp (22.4)
where 1" is the rUnkowski tensor. In case Jil = 0, and the space-
tlme is flat the divergences of the tensors t and T are equal
and equal to zero as a consequence of the field equations.
 - 275 -
Taub
23. The General Variational Principle
We now turn to the discussion of a variational principle

from which we propose to derive the equations governing the

motion of a perfect fluid which has a dielectric permitivity,

a magnetic permeability, is self-gravitating and is subject to

and electromagnetic field. We shall have to assume that the

conductivity vanishes. If it does not do so we will have to

resort to a non-holonomic variational principle. In the sub-
sequent discussion we shall introduce as field variables, the
metric tensor g~j\) of space-time, the particle paths of the

elements of the fluid, the density of the fluid, a variable

related to the temperature and the four-vector potential.
The variational principle will be described in terms of

I = (23.1)

where K is the Einstein constant of gravitation, IE and

II are given by equations (22.1) and (22.3) respectively,

Ig = f r-g Rd 4 x (23.2)

where R is the scalar curvature determined by the metric

tensor gil\! '

and

= (23.3)

with p the rest density, £ the rest specific internal

energy, ® the rest temperature and S the rest specific

entropy. IF is the integral which entered into our previous

discussions.
 - 276 -

Taub
We have seen (cf. Lecture II!) that

and if this is to vanish for arbitrary g~v and at which

vanish on the boundaries, we obtain as the Euler equations

(23.4)

and
(23.5)

Equation (23.5) may be shown to be a consequence of the fact

that the conservation of mass holds and

eJ..lV
;V = o (23.6)

as a result of equation (23.4).

We now turn to the evaluation of

We have evaluated the contribution to IE from the variation

of the in the preceeding section. Thus after an

integration by parts

II = J[_G].lV,.1
E ;v'i'].l

where AJ..lV01
1
is to be computed from equation (20.9), with

J..l and A assumed to be independent of e.

It may be verified that

I I =
E
 - 277 - Taub

where
=

(23.7)

The symmetric tensor E~v is similar to the Minkowski tensor.

It is known as the Abraham tensor [2]. In fact we have

(23.8)

Hence on evaluating

I' =
we find that

I' = J{[(R~v _ l2g~vR) + KT~v]g'

~v

+ 2K(pSU~) a' + 2K(G~V - J~)~'}~g d4x

;~ ;v ~

where

= e~v + E~v • (23.9)

Thus the requirement that I' = 0 for arbitrary g'

~v '
a',
and ~'~ leads to the Euler equations

R~v h~vR + KT~v = 0 03.10)

(pSu~) = 0 03.ll)
;~

G~v J~ £u~ (23.12)

;\1 = =

The latter equations are of course the Maxwell equations.

Equations (23.10) are the Einstein field equations with the
source of the gravitational field given by both the matter
present and the electromagnetic field. Equations (23.11) are a
 - 278 -
Taub

consequence of equations (23.10) and (23.l1hlhen matter is

conserved. That is, the equations

hold and there are no ohmic losses. For, it is a consequence

of equations {23.lq)and the Bianchi identity that

T~V = 0 (23.14 )
jV

These four equations which are the generalization of the

Lorentz pondermotive force equations contain the definition
of the pondermotive force acting on the matter and a
statement concerning the conservation of energy. The latter
statement is equivalent to the statement about the rate of
change of rest specific entropy along a world-line of the
fluid vanishes.
 - 279 -
Taub
24. Summary

The general variational principle given above, from

which one may derive the Einstein field equations, the

Maxwell equations, and the equations of hydrodynamics may

be applied to the problem of general relativistic magneto-
hydrodynamics by setting e U = O. The restriction to a =0
may be removed by using a non-holonomic variational principle,
that is by replacing II by an appropriate expressioninvol-
ving u' and ¢'
It is worthy of notice that the pondermotive force
acting on the charged fluid with electric permitivity and
magnetic permeability is not given by the divergence of the
Minkowski tensor. Rather, it is given by the divergence
of the symmetric tensor E~v the Abraham tensor defined by

equation (23.7) related to t~V by equation (23.8). A similar

result holding for less general circumstances has been given by

Penfield and Haus [3J. It is reasonable to expect that the
pondermotive force is to be derived from the symmetric
tensor E~v even in non-conservative cases where the varia-
tional principle does not apply in the form discussed above
in detail. This would resolve the old controversy over the
appropriate stress-energy tensor to be used in describing
general electromagnetic fields in moving bodies.
From the discussion of the variational principles it
is clear as to why E~V replaces t~V in the calculation
 - 280 -
Taub

of the pondermotive force. This tensor describes the energy

of the electromagnetic field alone as follows from the
discussion at the end of section22 above. However, because
A~vaT depends on u~, as well as on A and ~, the
Lagrangean ~ contains interaction terms between the fluid
and the electromagnetic field, even in the case of an
uncharged medium. The presence of these terms then is
~v
responsible for the additional energy which changes T

into E~v
We conclude with the remark that the methods given above
will apply to mare general situations then those treated. Thus
we need not assume that the electric properties of the matter
are isotropic or that that matter is a fluid. All that is
required is that a Lagrangean exists that is a function of
the field quantities describing the various properties of
the medium. Thus if in general

and i f
:-. ~v _, ~vaT
00 ,aT~v
~=" =" ,
aT

we may define a Lagrangean for the electromagnetic fields.

Similarly if the matter has a more general stress-energy
tensor than that given by a fluid, but involving no viscous
or similar forces, it too may be described by a more general
Lagrangean.
 - 281 -

Taub

References

1. A. Lichnerowicz, Relativistic Hydrodynamics and Magneto-

hydrodynamics, W.A. Benjamin, Inc., New York (1967).
2. W. Pauli, Theory of Relativity, Pergamon Press, New York
(1958), p. 110.

3. Paul Penfield, Jr. and Hermann A. Haus, The Physics of

Fluids 9 (1966) 1195-1202.
 - 282 - Taub

LECTURE VI

STABILITY OF GENERAL RELATIVISTIC GASEOUS

MASSES AND VARIATIONAL PRINCIPLES

25. The Second Variation

In this lecture we shall derive and then apply the well-
known result that if a set of equations are the Euler equa-
tions of a variational principle based on an integral I(e)
then the perturbations of solutions to the Euler equations
satisfy equations which may be derived from another varia-
tional principle. The latter principle is given by an
integral equal to 1"(0) where the prime denotes the

derivative of I(e) with respect to e. The variational

principle based on 1"(0) is called the second variation.
We shall apply these results to the discussion of
the stability against radial perturbation of spherically
symmetric static solutions of the Einstein field equations
for a self-gravitating fluid which obeys an equation of
state. In this case we may define I to be that given
by equation (19.2). We shall show that the second variation
problem for this integral is the same as the variational
principle given by Chandrasekhar [1] for this problem.
Let £(~A;~~~) be a scalar density formed from some
scalar or tensor fields $A and the derivatives of these
fields with respect to the coordinates in space time,
1, 2, 3, 4
=
A = 1, 2, , N.
 - 283 -
Taub

Then

(25.1)

where the integral is carried out over an arbitrary four

volume in space time determines a variational principle in
the following sense. We assume that the ¢A are functions
of the x~ and a parameter e, thus
A
= ¢ (x;e).
Then I is also a function of e and we may require that

I' (0) = drl

de e=o
= o

for arbitrary
A
= (d¢ )
de e=o

We have

rICe) =

(25.2)

rICe) =

since
=

On integrating the above expression for rICe) by parts we

obtain

r' (e) =
(25.3)
 - 284 -
Taub
where
= C25.4)

The second integral in Eq.C25.3) may be written as an

integral over the hypersurface bounding the four-volume of
integration.
The requirement that I'CO) =0 for arbitrary ¢,ACO),
in particular, for those that vanish on the boundary of the
region of integration then leads to the Euler equations

= = o (25.5)

The equations satisfied by the difference between two

"almost equal" solutions of these equations, or the equa-
tions satisfied by perturbations of solutions to these
Euler equations are

= o

where FA(e) is given by Eq. (25.4). Thus

FA(O) = [
a2.c
. ¢,B + a2.c ¢,B
a 2¢Aa¢B 3¢Aa¢B ,lJ
,lJ

- ( a¢Aa2.ca¢B ¢IB + a¢A a2a¢B.c ¢,B) ]

, v , II e=o
(25.6)
,lJ ,ll v

= 0

These are a set of linear equations for the variables ¢,A(O)

A
whose coefficients depend on the ¢ (x;O) and their
derivatives. The ¢,A(O) are called the perturbations and
 - 285 -
Taub

the ¢A(x;O) the unperturbed solutions.

Now it follows from Eg. (25.2) that

I' (e) = fF A¢"Ad 4x + (F,¢,A d 4x +

. A f (~ ¢"A) d 4x
3¢A ,11
,11
(25.7)
+ f[ 32£ ¢,A¢,B + 32£ ¢,A ¢,B] d4x
3¢A3¢B 3¢A a¢B ,11 ,v
,11 ,11 ,v

or
I"(e) = fF A¢"Ad 4x + f (~¢"A) d 4x
a¢A ,11
,11
(25.8)

:0 2£
2 a 2£
+j[ ¢,A¢,B + 2a £ ¢,A¢,B + H,A¢,B d 4
,11 ,v' x.
a¢Aa¢B a¢Aa¢B a¢A a¢B
,11 ,11 ,v

From Eq. (25.8) we have

1"(0)
32£ ¢,A¢,B + 23 2£ ¢,A¢,B + 32£ ¢ ,A¢ ,B] d 4x
= f[ ,11 ,v
a¢Aa¢B a¢Aa¢B ,11 a¢A a¢B
,11 ,11 ,\I

(25.9)
when the ¢A(O) are such that FA(O) = 0, that is the
¢A(O) are unperturbed solutions of the Euler equations
associated with I, and the ¢"A(O) = 0, on the boundary of
the region of integration. If we now consider the ¢,A (not
the ¢a) functions of x and a parameter f we may define

J(f) = 1"(0)

and examine the Euler equations resulting from the condition

= o
 - 286 - Taub

This is the "second variation problem." We find

2JF'Ao~ ,Ad 4x + 2J ( a 2£ ~,Ao~,B)

oJ = d 4x
a~Aa~B ,]1
,]1

~ 2J ( a 2£ ~ ,Ao~B) d 4x
d~A a~B , ]1 ,v
,]1 ,v

Hence for variations of the ~ ,A such that ~,A vanish on

the boundary of the region of integration, 1"(0) takes on
extreme values when the ~,A satisfy the equations

the equations satisfied by the perturbations.

Thus the solutions ~A of the Euler equations FA(~) = 0,
when considered as ~A(x;O) are such that I' (0) = o, for
~ ,A which vanish on the boundary of the region of integration,
and the solutions ~ ,A of the equations FA(~;~') = 0 , where
the ~A satisfy the Eurler equations and are coefficients
in the linear differential equations, are such that 1"(0)
takes on extreme values.

26. The Spherically Symmetric Case

In a spherically symmetric space-time we may write the

line element as

= e 2~ dt2 - e 2~ d r 2 - e 2]1 dn2

" (26.1)

where
d6 2 + Sln
. 26 dX 2 (26.2)
 - 287 -
Taub

¢, W and ~ are functions of r, t and a parameter e and

these are comoving coordinates for each value of the para-

meter e. It then follows [2J that the non-vanishing

components of G~ are
v

2~ W ) + e-2~
r r

(R 2
2
~)
2 = - (R 33 _ ~)2 (26.3)

= e- 2¢[W tt +
~tt
+ ]It2 + w2 - W ¢ + ~t~1jJt - ¢t)J
t t t

- e -2¢[ ¢rr + ]lrr + ]lr2 + ¢2r - ¢r1jJr +]l(¢

r r
-W)J
r

R4 -2¢
1 = 2e [~rt - ]It¢r ~rWt + ~ t ]l r J

Rl
4 = - 2-2W[ ]l rt - ~t¢r - ]lrWt + ]lt~rJ

where the subscripts r and t denote the derivatives with

respect to these variables.

The Einstein equations become

-F =(R 4 !R) + kw
2 = 0
¢ 4

-F =(R l !R)
2
kp = 0 (26.4)
W 1

-F ]l =(R 22 !R) - kp
2 = 0

and

Rl = 0 (26.5)
4
 - 288 -
Taub
The four equations (26.4) and (26.5), are not all independent
in view of the Bianchi identities. It may be shown that the
solution of these equations is determined by the solution
of equation (26.5) and Fw= 0 for a range of values of t
and of F = 0
¢
for t = O.
The unperturbed solution we shall consider will be

assumed to be static, that is ¢, W and ~ will be assumed

to be functions of r alone. In that case it is no restrictior
to take

~ = log r.

Equation (26.5) is identically satisfied and equations (26.4)

reduce to

(26.6)

It is a consequence of these equations that

(26.7)

It is a further consequence of Eqs. (26.6) that

= (26.S)

The last equation also follows from the equation of state

assumption.
 - 289 -
Taub
The equations satisfied by the perturbations,

~' and ~' are obtained by differentiating Eqs. (26.4)

and (26.5) with respect to e and setting e = O. We

then obtain from Eq. (26.5) and the last of (26.4) the
equation

~r't - ~'~
t r - !~,
r t + !~,
r t = 0
(26.9)

where now ~r is determined by Eqs. (26.6). The solution

of Eq. (26. 9) is given by

~' - ~' 0 = e~(e-~r~') r (26.10)

where
,
~o = ~' (r,O)

and we have chosen our comoving coordinates so that

,
~o = ~'(r,O) = 0 (26.11)

This can always be achieved by a coordinate from transformation

involving r alone.
The function ~' may be evaluated by using the integral
of the field equations given by Eq.(17.8) which holds for all
values of e. That equation may be written as

oe ~+2~ = oo e\jJo+2~0 (26.12)

where now the subscript zero on fCr, t, e) is defined by

fo = fCr,O;e)

On differentiating Eq. C26.12) with respect to e, setting

e = 0 and using Eqs. (17.6), C16.5), and (26.11) we obtain
 - 290 -
Taub

= (3~' + r~'l' - ~ l'r~') (26.13)

or

= I' e't'(r 3 ~'e -"'t')

-2" (26.14)
l'

where a is the velocity of sound in the unperturbed fluid,

is given as above and ~ ,o is ~'(r,O) for e = O.
Thus ~' and ~' are determined in terms of~' This
function may be determined by solving the equation

F' = 0
~

The quantities ~' and ~' enter into this equation but may
be eliminated by means of Eqs. (26.10) and (26.14). We shall
discuss this equation in the next section.
When the field equations, Eq. (26.4) and (26.5) are

applied to a problem in which there exists a hypersurface in

space-time across which the stress-energy tensor is discon-
tinuous, the equations must be supplemented by conditions
satisfied by the metric tensor, its derivatives and the stress
energy tensor on this hypersurface. Thus for the problem
we wish to consider, namely that of a gas occupying a limited
region of space-time and bounded by a vacuum there exists
the hypersurface L defined by

r =

where is the constant comoving coordinate of the boundary

element of the material.

 - 291 -
Taub

It is well known that the conditions referred to above

become in this case

and that ¢, W, and ~ are continuous across the hypersurface

L. In addition all first derivatives of these quantities

except ,!,
'l'r must be continuous across L. These conditions

m~st hold for the perturbed as well as for the unperturbed

equations. Hence we must have

= = o (26.15)

In view of Eq. (26.14) this condition becomes a boundary

condition on the function ~'.

Another condition is the requirement that for the

perturbed and the unperturbed solutions the function

R = e~ = r = o

at the origin. This function is the analogue of the Eulerian

coordinate of an element of the fluid which has the Lagrange
coordinate r. Hence we must have

R' = e~~' = r~' = 0 (26.16)

at the origin.
Eqs. (26.15) and (26.16) provide boundary conditions for

the second order partial differential equation FW' = o.

We close this section with a discussion of the implication

of the Bianchi identities.

 - 292 - Taub

If we define

these identities are

1 ).1
= -~ (;:gK v ) ,).1 = o
v-g

They hold for all values of e. If the above equations are

differentiated with respect to e and then evaluated for
e = 0, and if it is assumed that K).1 (x' 0) = 0, it follows
v '
that

1
= o (26.17)
;:g
where now g).1V = g).1v(x;O) is the unperturbed metric, and
r ).1PV is determined from this metric and this metric satisfies
the field equations.
We now evaluate Eqs. (26.17) for the case considered
above, when the unperturbed metric is spherically symmetric
and static and the perturbed metric depends on time but is
still spherically symmetric. In that case Eqs. (26.17)
reduce to two equations corresponding to v =4 and v = 1.
These are

(26.18)

and
2rF'
).1
= 0 (26.19)

respectively.
 - 293 -
Taub
Hence when R'41 = (R,l)
4 t
=0 as is the case when
Eq. (26.10) holds, Eq. (26.18) becomes

= F~(r,O) (26.20)

and Eq. (26.19) becomes

(26.21)

The first of these equations implies that the equation

is only a restriction on the functions ~'
o and
~' It may be verified that on substituting Eqs. (26.10)
o
into the expression for F~ one obtains

2 -2'" + k
-F~ = r2 (re 'f lji 0, ) r a2
(w + p) ~ ,
0
= o (26.22)

27. The Equation F~ =0

The equation F~ = 0 is derived by differentiating the

second of Eqs. (26.4) into which Eqs. (26.3) have been
substituted setting e = 0, and making use of the values of
the unperturbed solution. One then obtains
,
-2 , , -2lji(1 + ~ ) _ u ! e-2lji~,
F' = 2[ e lltt llr e .r 'f r 2" - r 'f r
r
(27.1)

when Eqs. (26.10) and (26.14) are used to express lji' and

~' in terms of U', one finds that

 - 294 -
Taub
re2iJ!Cw + 12) 1 2
F' = e
2iJ!-2¢ Cw+p)~ + -4 p ~ - w-+p Pr~
2 iJ! tt r r

2
- e - iJ!-2 ¢[ e 3¢+ iJ!Cw + })a Ce-¢r2~) r Jr C27.2)
r

+ ke2iJ!Cw + p)p~ + ~ )_e-iJ!-¢Ce¢+iJ!¢')

iJ!~C~ + r r 0 r
r

where

~ = rll' C27.3)

The equation F' = 0

lji
where F' is given by equation
iJ!
C6.2), has a boundary condition indicated in Eqs.C26.1S) and

C26.16). It is the equation given in [2J for the case of

the radial perturbations of a self gravitating fluid when

the equation of state was such that the fluid was isentropic.

In that case cr = p, the rest mass density of the fluid.

When ¢' = lji' = 0, the equation is the same as the equation

o 0

given by Chandrasekhar [lJ as may be seen by writing

and thus defining y. This definition of y is that

given by Chandrasekhar as may be verified by writing

w = NCl + uCp,N))

where u is the internal energy. If one then computes

Cop/3w)S and remembers that

TdS =

one verifies that the definition of y given above is that

used by Chandrasekhar.
 - 295 _
Taub

28. The Evaluation of 1"(0)

In this section we shall use the results obtained above

to express 1"(0) in terms of 11 ' ,

,/,'1'0'
, and ¢'.
0 We begin by

observing that when I is defined by Eq. (19.2) and when the

perturbed and unperturbed metrics are of the form given by

Eq. (26.1), then it is sufficient for the purpose of calculating

I'(e) and I"(e) to evaluate I(e) in the coordinate

system In which Eq. (26.1) holds.

Thus we have

1 ~+,/,
ff{e'l' + e'l'~-'/'+2"
~(11
2 + 211 ¢ ) - e -~+'/'+2"~(11
2 + 211 ljJ )
811 I(e) : 'I' 'I'
r rr
'I' 'I'
t tt

(28.1)
Hence

~1II' (e) : ff{e¢+ljJ+2 11 (F ¢' + F ljJ' + 2F 11')dr dt - See)

¢ ljJ 11
(28.2)

where F¢,FljJ' and FI1 are defined by Eqs. (26.4) and (26.3) and

See) = (28.3)

with

A =
(28.4)

B = e¢-ljJ+2 11 (_ljJ'(211 r + ¢r ) + ¢'¢ r + 211'11 r + 211'r + ¢')

r
(28.5)

The integration in Eqs. (28.1) to (28.3) may be taken to

be the region bounded by the inequalities

a ~ r < 00 (28.6)
 - 296 - Taub

Across the boundary

r = (28.7)

There is a discontinuity in the stress energy tensor. The

pressure p must be continuous at p = rb but the energy

density w need not be. The requirement that

L811 I , (0) = 0

for arbitrary ~', ~I, and~' which vanish together with their
derivatives on the boundary of the region given by the in-
equalities (28.6) and such that ~" ~', I.l' .and ~'
r and ~
rI
may

take on arbitrary values on the interior boundary given by

Eq. (28.6) leads to the field Eqs (26.4) and the boundary
conditions discussed in Section 26 (cf.[4]).
We also have

1 1"(0) =
8TI
where ~, ~ and ~ are evaluated for e = 0 and these functions
satisfy the unperturbed equations. In view of Eqs. (26.10),
(26.7), and (27.3) we have

= e~+~r2~'F' +(e~+~r2~F')
o ~ ~ r

_e~+~r2(~-~(e~r2F') + i re2~k(w + p)F,:).

r ~ r 2 'I'

On using Eq. (26.21) we obtain

 - 297 -
Taub

This equation holds for all values of r, however for

r ~ rb w = P = 0 and for r ~ rb we may use Eq. (27.2).
Hence we have

1
81i 1"(0) = (28.8)

where

J =

(28.9)

+ r 2e ¢+1jJ ~ 2 (ke 21jJ (w+p)p + 4 Pr - r I

W+'P 2
Pr )] dr dt

tl'"
JI = -f fr 2e¢+1jJ[F¢¢' + F1j!1jJ~] dr dt
o 0

(28.10)
tlrb
r r)
+kf J [r2e¢+1j!~1j!~(~ + bP· _ r2~(e¢+1j!¢,) ] dr dt
o r
o 0

and

=
(28.11)

with F~ given in terms of ¢', 1j!' and~' by Eq. (27.1).

If the functions ¢', 1j!' and ~ are to be such that

1'(0) = 0, that is if they and their derivatives are to

vanish on the e_ erior boundaries, and if the boundary

conditions on ~ are to hold ar r = rb and r = 0 we

must have
L871 1"(0) = J (28.12)
 - 298 -

Taub

where J is given by Eq. (28.9).

The Euler equations of the variational principle

1 or"(o)
81T = oJ = 0 (28.13)

is the equation

re 2lj!(w + ;e) F' (28.14)

lj! =
0
2

where the explicit form of this equation is given by equation

(27.2). This equation is equivalent to F' = 0 and the

lj!
variational principle defined by Eqs. (28.13) and (28.9) was

of course to be expected in view of the general discussion

given in the introduction.

29. The Stability Criterion

The variational principle defined by Eqs. (28.13) and

(28.9) may be related to that given by Chandrasekhar in [2J

by observing that if one writes

~ = sinCat + a)~(r) (29.1)

we have
1 tl 2
871
r"(o) = J = ! sin Cat + a)} dt (29.2)
0

where
rb
j = ! - a 2e 3lj!-¢ r 2 (w + p) ~
2
dr + 11 (29.3)
0
where
rb

/1
4Pr
= J [r 2e ¢+lj! ~ 2 (ke 2lj! (w+p)p + r ~Pr
1 2)
0
(29.4)
2
+ e +3¢+lj! (w+p) a (re'¢~);J dr
2"
r
 - 299 - Taub

The variational problem

(29.5)

has as its Euler equation, Eq. (28.14) with ~ given by

Eq. (29.3). The functions ~(r) satisfying this Euler

equation, that is the extremal ~(r) = ~o(r) are such that

Chandrasekhar has pointed out (cf. [2]) that the

variational problem given by Eq. (8.4) expresses a minimum

principle for the determination of the lowest value of

0 2 and that a sufficient condition for the dynamical instability

of a mass is that )1 = 0 for some "trial function" ~ which

which satisfies the required boundary conditions.

However, if such a trial function exists we shall have

~~ 0 and in view of Eq. (29.2), for this trial function

L1,,(o) < 0 (29.6)

811

Thus the sufficient condltion for instability used by

Chandrasekhar is equivalent to the condition that there exists

a trial function such that the inequality (29.6) holds. The

latter criterion may be applied to discussion of the stability

of general solutions of the Einstein field equations. We

need not restrict ourselves to a static unperturbed solution

and consider perturbations of such solutions which depend

on the then defined time coordinate in an exponential manner.

 - 300 -

Taub

References

1. Chandrasekhar, S. The dynamical instability of gaseous

masses approaching the Schwarzschild limit in
general relativity. Astrophys. J. 140, 417-433
(1964). -

2. Taub, A.H. Small motions of a spherically symmetric

distribution of matter. Les Theories Relativistes
de la Graviation, pp. 173-191. Centre National
de la Recherches Scientific, Paris (1962).
3. • Singular hypersurfaces in general relativity,
--IITinois J. I'lath. !, 370-388 (1957.
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO

(C.I ME)

GENERAL-RELATIVISTIC KINETIC THEORY

OF GASES(1)

J. EHLERS

Corso tenuto a Bressanone dal 7 al 16 Giugno 1970

 GENERAL-RELATIVISTIC KINETIC THEORy
( i )
OF GASES

Introduction

The relativistic kinetic theory of gases, which will be presented in the

following lectures, is of interest for a number of reasons: It offers a sim
pIe, microscopic model for matter in bulk 'which is sufficiently general to
provide a oasis for hydrodynamics and thermodynamics of simple and multi-
component systems. Definite conservation laws, balance equations, equations
of state, transport and reactions can be derived from it, and if cross from
a microscopic scattering theory are fed in, kinetic theory gives transport and
reaction coefficients. As in the non-relativistic theory, the arbitrariness of
the constitutive equations and the indefiniteness of the transport coefficients
inherent in the phenomenological continuum approach are overcome by the ki-
netic theory.
Moreaver, kinetic theory provides a description of gases under conditions
where fluid dynamics does not apply, e. g., when collisions are rare and the
mean free path is large.

This work was supported in part by NSF - grant GP 20033

 - 304 -

Ehlers

Applied to macroscopic particles like stars or galaxies, kinetic theo-

ry offers a method of treating systems or the sustem of galaxies, the
"gas" of cosmology.
Another asset of relativistic kinetic theory is its uniform treatment
of gases consisting of particles with positive mass and those having zero
mass particles; its application to photons gives the cosmologically and
astrophysically important theory of the transport of radiation.
Specific applications of relati vistic ki~ tic theory to astrophysical pro-
blems which illustrate the usefulness of this theory will be mentioned later.
Although the domains of applicability of fluid dynamics and kinetic theo-
ry overlap, neither of them contains the other one. Nevertheless, kinetic
theory may be considered as the more fundamental of the two theories, sin
ce within it one can derive from simple microscopic laws and plausible sta-
tistical assumptions and approximation methods the general forms of all the
laws which are postulated in fluid dynamics; only the numerical values .of
(e. g.) transport coefficients have to be changed on leaving the domain of va-
lidity of kinetic theory.
Ideally, one would like to derive both kinetic theory and fluid dynamics
from statistical mechanics; at the relativistic level, this has not yet been
achieved. Therefore, we have to introduce the basic concepts and laws of
kinetic theory on the basis of plausibility considerations as did Boltzmann.
There are many unsolved problems in relativistic kinetic theory, questions
concerning the foundations, the mathematical structure, and specific physical
applications. We shall refer to some of them in the following lectures.
 - 305 -

Ehlers

Several systematic expositions of relativistic kinetic theory exist

which naturally have much in common with the following lectures, in
particular those by N. A. Chernikov (1963,1964), C. MarIe (1969), J. Ehlers
and R. K. Sachs (1968), and J. Ehlers (1969). The elementary aspects of
the special-relativistic theory which precede the Boltzmann equation (or
sidestep it) are contained in the well-known book by J. L. Synge (1957) who-
se geometrical spirit has strongly influended the present lectures. (More
specific references will be given at appropriate places in the lectures.)
In order to free equations of inessential factors, we shall use the
following convention regarding physival dimensions and units: We put

c = 8 Tt G =1i = 1,

where c is the speed of light in vacuo, G Newton's constant of gravitation,

h the quantum of angular momentum, and k Boltzmann's constant. All
physical quantities are then measured by pure numbers.
 - 306 -

Ehlers

1. Assumptions on spacetime. Notation

Let X denote spacetime which we assume to be a real, four-dime~

sional, connected, differentiable Hausdorff manifold. In addition, we as-

sume X to be oriented, and take always oriented local coordinate sy-
stems (x a ), a = 1, ... ,4.
The tangent space to X at p is denoted as T (x); its dual, T"'(X).
d PaP
Natural, dual bases in T T4t are (--) and (dx ), respectively.
and
pp dX a
X carries a normal hyperbolic metric whose signature we take as
+ 2. The metric tensor or gravitational potential is written gab' the Rie-
mannian connection is r~c' and the Riemann! curvature tensor is R~Cd'
The Ricci tensor is given by Rab: = RC b' and the Einstein tensor by
1 ac
Gab: = Rab - 2 Rg ab , where R: Ra a · The sign of the curvature ten-
sors is fixed by the Ricci identity.

d
= vd R abc'

We assume that X is time-oriented with respect to gab' so that it is

meaningful to distinguish between future directed and past directed timelike
( 1)
and lightlike vectors, respecti vHy
An orthonorlaal basis (e.) of T is always chosen to be oriented
J p
and such that e4 is future -directed.

(1)
An example of a pair (X, g b) which is not time-orientable is given in
appendix I of Ehlers (1969).a
 - 307 -

Ehlers

A coordinate- system (x a ) is said to be inertial at p, p ~ X, if

r : c Ip = 0 and (?xalp ) is orthonorlaal. -

The physical interpretation of general relativity theory is largely b~

sEld on the correspondence principle that physical laws in the presence

of gravitation retain their special relativistic form at p if expressed
with respect to coordinates which are inertial at p. This guiding pri!!.
ciple is not unambigious, however.
The assumption that spaceti me is oriented is not necessary for kin~

tic theory; it is made here only for convenience. Without this assumption
several quantities appearing in kinetic theory would have to be defined with
respect to oriented domains of X, and it would have to be shown that a
change of orientation preserves all relevant equations. This can be done.
The assumption that spacetime is time-oriented is also not necessary
for those pArts of kinetic theory which are independent of the Boltzmann
equation. Without it, some quantities would have to be defined relative to
time oriented domains of X, and the relevant equations would have to
be shown to be insensitive to changes of the time orientation; that can
easily be done. The Boltzmann eqllation, however, can only be formula..!
ed in a time-oriented spacetime, and its form is not preserved under a
change of that orientation. The reason is that the occupation numbers of
initial and final states enter the collision integral in a non- symmetrical
manner, as will be seen later and as is known from ordinary kinetic
theory. The arrow of time built into the Boltzmann equation shows up
particularly clearly in the Htheorem, to be derived later.
 - :W8 -

.Ehlers

2. Some facts about differential forms and integration (1 )

Kinetic theory deals with various kinds of averages which are expre~

sed as integrals. The domain of integration is sometimes a hypersurf~

ce in X, sometimes a specetime region, sometimes a hypersurface or

a region in phase space (to be defined below). The appropriate tools for
forming such integrals-volume elements, hypersurface elements etc. -
are differential forms. We assume that the elements of the theory of
differential forms on manifolds are known, and collect here a number
of facts which we need later.
On an n-dimensional manifold N, the differential form fields can be
expressed, with respect to local coordinates, as sums of homogeneous
forms like UI = J.
\II dxai ' .. a" where the components \II
I r! I at' .. a r ' I at' .. al'
are real functions and

are exterior products of the coordinate differentials. With respect to the

operations of addition, multiplication with real numbers and exterior mil.!.
tiplication the form fields from an associative algebra. The exterior

r
differentiation operator d maps this algebra into itself.
An r-form can be contracted, like any covariant tensor, with avec
tor A; the result is a (r - l}-form cP= A·U/ with components I.D =
I I I a 1 · .. a r
= Aat \II . A coordinate-independent definition of this operation
I at a, ... a~

( t ) See the "references"about mathematical tools" in the bibliography at

the end of these lecture notes.
 - 309 -

Ehlers

is contained in the following assertion:

For any system of r-1 vectors A 2,···, Ar' we have

For a fixed vector A the mapping 0/-+ A· f of the algebra of

forms into itself in an antiderivation, i. e., it is linear and satisfies the
product rule

Moreover,

0.

A trivial, but useful consequence of the definitions is the

Lemma 1. If n is a nonzero n-form at some point p of N, then the
map L-"L·.n. = : W is a vector space isomorphism of T (N) onto
p
the space of (n-l) -forms at p.
This lemma immediately leads to
Lemma 2. If.n. is an n-form at p, n f 0, and LET (N),
p
L f 0,

then the most general (n -1) -form W at p such that t\){A l' ... , An _1) f °
whenever (L, A , ... , A ) is linearly independent, is given by to\) = a vCt
1 n-1
where a f 0.
Corollary. If fA) has the property stated in Lemma 2, then L . "" = 0,
 - 310 -

Ehlers

and w (AI' .... , An_I) o whenever (L, AI' .... ,An_I) is linearly depe~

dent.
Lemma 2 and its corollary should be visualized by considering n
and w as volume-functions for n- dimensional and (n-l)- dimensional
parallelotope s, respectively.
Another useful fact needed later the proof of which is left as an ex-
ercise is
Lemma 3. If n is an n-form field on N, L a vector field and f a
function, then

df " (L • n ) L (f) n. (1)

We here recall that a vector is (identified with) a linear differential

operator acting on functions: L(f) = La f,
a
Finally we recall the fundamental theorem (of Stokes):
If M is an oriented, compact, m- dimensional subma.nifold-with - boun .
dary ~M of an n-manifold N, and 'f is a (smooth) (m-l}-form field of
N defined on M, then

(2)

The assumption that M is compact can be omitted provided if

decreases to zero sufficiently strongly at infinity of M; also, dM may
be allowed to have "corners".
 - 311 -

Ehlers

3. Volume elements in spacetime

Under the assumptions about spacetime stated in section 1 the ex-
pression

(3)

where g:= det(gab) and y-:g >0, is a nonvcinishing4-form such that

'? (e 1,e 2,e 3,e 4) = 1 for any orthonormal basis (e j ); it is the volume
element of spacetime.
Let A be a vector field on X and D a 4 -dimensional, oriented)
compact submanifold-with-boundary of X - henceforth called a region. Then,
according to (2),

Id(AO"1) = SAO "l (4)

D '() D
The integrand on the left can be rewritten as

d(Ao,)

and that on the right as

A 0 fJ) Aa 6.:' (5': = -.!.

fT) dx bcd (5)
( a' a 6 (abed '

where
 - 312 -

Ehlers

~ abcd '7 [abcd] , (1234 (6)

are the components of "I .

~ are the components of the (vectorial) hypersurface element in
a
X; the latter is a vector-valued 3-form.
With this notation, (4) goes over into

Er,
a
(7)

D
the familiar metric-dependent version of Gauss's theorem in Rieman-
nian space.
We shall heneeforth use the term hypersurface for "oriented hyper-
surface".
Since each tangent space T of X is itself a (flat, oriented)
q
pseudoriemannian space, it has its own volume element

..,... ,r:r;"d 1234

I&. V -g P . (8)

g is to be evaluated at q with respect to coordinates (x a), and the

p a from p" p a :f a.
~X
define an oriented coordinate - system on T
q
Physically important hypersurfaces of T are the mass - shells for
q
masses m? O. The mass- shell P (q) con sists of all future directed
m
(4 momentum) vectors p at q which belong to (proper) mass m;
p 2 " - m 2. An oriented coordinate- system on P (q) is defined as follows.
Take coordinates on X around q such that
dm _
--~, v- 1,2, 3, are space-
;;)X
 - 313 -

Ehlers

like and S X4 is future- directed and timelike at q. Then the restric-

tions of the natural coordinates p" to P (q) form an oriented co-
m
ordinate- system on P (q), and p 4 (
m
> 0) is determined by

cab 2
gab(x )p p = - m . (9)

Po (q) is the future light cone of q.

In order to obtain a scalar volume element on P (q), consider
m
the T -analogue of (5),
q

1 bcd
La:= 6'(abciP . (10)

Its restriction to P (q) has values proportional to the normal of P (q),

m m
hence there exists a 3-form 1T such that
m

(11)

since is a normal of P Setting a 4 in (10) and (11) gives

Pa m
explicit ely

(12)

The same volume element is formally obtained from

(2 2
It = 2 H(p) d(p + m ) 1t. (13)
m
 - 314 -

Ehlers

in which H is the Heaviside function of p 4 and J is the Dirac

distribution.
For m> 0, m 1t: is the induced Riemannian volume element of
m
P (q) as a hypersurface of T .
m q
In inertial coordinates at q, we have the familiar expression

~3
(14)
E

4
where E P is the energy. Taking polar coordinates in p-space we
obtain

Vm 2 +p 2'
1(m (15)

or also

2'
1'( m (16)
m

The consideration which led to the volume element TC o on the tan -

gent null cone P (q) can be generalized to the actual null cone of
o
q in X. We leave it as an exercise to the reader to verify
Lemma 4. Let N~ be the past null cone of q, and let u be a
q q
future-directed timelike unit vector at q. A normal k of W is ob-
a q
tained by drawing null geodesics through q, choosing tangent vectors k
to them such that, at q, k.u and parallely propagating these
= 1,
q a dx a
k's along the null geodesics. Also, put v = 0 at q, and put k = dv'
 - 315 -

Ehlers

obtaining a field of affine distances v on J.q ne>note as dfl q

the solid angle abtained by proJecting a small bundle of null rays
through q into that 3-space through q which is orthogonal to u,
q
and call D the distance from apparent size of an arbitrary point
r E Jq from q, as measured by an observer at q with 4-velocity u .
q
Then

\ 17)

so that D2d n q 1\ dv is a natural scalar volume element on N.

q

(i )
4. Basic assumptions about a relativistic gas. Geometry of phase space.
The history of a system of many (classical) particles of negligible size
is represented in relativity theory as a complex of timelike or lightlike
wordlines. The particles may be thought of as being macroscopic ( stars,
galaxies) or microscopic (molecules, atoms, ions, nuclei, photons, ... ),
and they may be interacting through long- range and/or short range forces.
Without attempting to give a detailed description of the dynamics of
such a general system, we lay down a special, simple model for some
systems which we call gases. In these systems, the particles are assumed
to move like test particles in a mean gravitational field gab and elec-

( i) The geometric treatment given in this section follows essentially

that of Bichteler (1965). See also Chernikov (1963), Lindquist (1966)
(Appendix), and Marie (1969).
 - 316 -

,Ehlers

tromagnetic field F ab' except during encounters due to short range in-
teractions which are idealised as pOint collisions. (I. e. ,the range of
these interactions must be much smaller than the mean free path. )
The mean fields may be external fields - we then speak of a test gas
- or may be collectively generated by the gas p8rticles themselves, in
which case we have a selfgr2vitating gas (or a Vlasov plasma).
We proceed to formalize this qualitative picture of a gas.
A particle of mass m (~O) and charge e has a worldline
xa(v) which obeys the Lorentz-Einstein equations of motion

a Dpa a b
(18)
dv = e F bP ,
p,

if radiation reaction is neglected. The parameter v is so chosen that

the tangent vector pa is the (future-directed) 4-momentum. If m > 0,
mv is proper time. ~v denotes, here and in the sequel, the abso-
lute derivative along the world line,

Dpa d
p aa
r c b
cis = cis +bc P P . (19)

If a particle participates in a collision at x E X, its world line may

have a corner at x, or the world line may end or begin at x, if the

particle is annihilated or created in the collision.
In the case of many particles the spacetime figure of a gas is a com-
plicated network of curves, since several trajestories with different direc
tions can pass through the same event, and the trajectories through near-
by events can have quite dilferent directions,
 - 317 -

Ehlers

A simplification of the geometrical representation is achieved, as

in nonrelativistic kinetic theory, by introducing a phase space. Since
in relativity no preferred space sections t = const. exist, the relati-
vistic phase space cannot be defined in strict analogy to the ordinary
(1, p) phase space (of one particle), but will correspond to the (;t, t, p, E) -
~space. We define the (relativistic) one particle phase space for parti-
cles of arbitrary mass m to be the manifold

M: = { (x, p): x E X, P (Tx(X), P~ 0, p future directed.} (20)

This set is indeed a 8-dimensional manifold, if we agree to take as

local coordinates (x a , pal, where (x a ) is a coordinate-system on X
and pa are the corresponding natural vector components.
M is, in fact, a manifold with boundary, the boundary dM being the
2
set of states (x, p) having mass zero, p = O.
M is a fiber bundle with base X. The fiber at x is the set
of non-spacelike, future-directed vectors at x, i. e., the 4-momentum
space at x. (If all vectors had been admitted, M would be the tangent
bundle T(X) over spacetime.)
M is obviously oriented, the (x a, p a)_ systems being oriented coor-
dinate-systems.
The equations of motion (18), (19) define on M a vector field

a~ aarabcd
L =P d X" + (e F bP -, bc P P ) d pI. (21)
 - 318 -

Ehlers

called the Liouville vector (or operator). The oriented integral curves
(xa(v), pa(v)) form a congruence in M, the phase flow generated by
L. Physically, the phase flow represents the set of all test particle mo-
tions which are possible in the combined gravitational and electromagne-
tic fields occuring in L.
The rest mass m as given by equation (9) is a scalar function on
M. It is constant on each phase orbit,

L(m) o. (22)

Hence the restriction L

of L to the hyper surface M of M
m m
defined by m = const. is tangent to M . We note that
m

M
m
= Up m
xeX
(x). (23)

M , with its Liouville vector L and its phase flow, is the phase
m m
space for particles of fixed mass m; it is seven-dimensional and
- t
corresponds to the Newtonian (1; t,p) -space. ( ) It is also a fiber

( 1) In classical mechanics, this space is sometimes called "augmen-

ted phase space". See e. g. Liboff (1969). p.16.
 - 319 -

Ehlers

bandle with base X, the fiber over x now being P (x), the mass-
m
shell at x.
M , being the boundary of t~ oriented submanifold of M given
by Pf1~ _ m 2, is also orientable. We orient it by choosing a coordina-
a a 4
te system (x ,p) on M such that p 4P 0 whenever <
pEP (x),
m
and then take (x a , p'l1 ) as an oriented coordinate-system on M . We
m
then have

L = pa ~ + (e F" pb ,.,V pbpc) 'j (24)

m dXa. b - I bc <JP"
We know from ordinary statistical mechanics the usefulness of a mea
sure on phase space which is invariant under canonical transformations
and, in particular, under the phase now.
Let us conseder, therefore, the coordinate-independent 8 form

n : = ~ "1" = _ g dx 1234 A dp 1234 (25)

on M (formed by means of (3) and (8)) and the 7 - form

...:.Ld 1234" d 123 (26)

Ip41 x p

on M . Obviously, at each Doint.

m

0, n m
I o. (27)
 - 320 -

Ehlers

nand n m
are related as follows (exercise):

!l = m dm ",{lm . (28)

To see whether n is invariant with respect to the phase flow we com-

pute £L!L , the Lie derivative of n with respect to L. Because of
the identity ( ~ )

iLn. = d(L·..n) + L·d.n.

and d.n. = 0, we get tL n = d W, if we put

"" : = L' n = p
a
era" Tt 1
+ 6" ~ abcd( eF dP -
a d ra d e bcd
de P p )dp "1- (29)

The differential of this 7- form vanishes. This is reall.y verified by using

inertial coordinates at some (arbitrary) event x.
Hence,

( .. ) See, e. g. Hicks (1965), p. 94.

 - 321 -

Ehlers

dl.l) 0; (30)

i. e., n is invariant under the phase flow (Liouville's theorem).

The t - form W which arose here rather naturally as a tool will
be seen in the next section to be important in itself; let us note some
of its properties. From its definition (29) and from iLW = d(L' W)+
+ L(dw)
we infer:

L' ~ 0, o. (31)

These properties express that '" induces a nonzero 7 form on the

quotient manifold MIL; W can be consedered as a measure on tre 7-

SII '
=.:m=.:an=i.::.;fo:..::l..:;d_o"--f,--,, p_h.=cas=..e, -,o-,-r.. .:.b-=..it:..=. ;s. Indeed, if we int roduce on M comoving local
coordinates ~A with respect to L, i. e., such that L = then
(31) means that I.U =
} r-t ~,.
f';if""WA('l> , ... ~ )d;>
. "."
'eA ,whIch IS a form
on MIL. If ~ is a tube of phase orbits and !: a cross section of
'l, JE W measures, loosely speaking, the "number" of orbits contaiR
ed in 1 it is independent of the cross section.
The preceding consliderations can be carried over straightforwardly
from M, L, to M ,L (exercise); one obtains
m m

1 IY) ). b rl b c ,.."
W
m
:=L'n. =pa
m m
6"" A1t +
a m 21 p,1 P"f"''' (F bP -, bcP P )dp 1\'9 (32)
 - 322 -

Ehlers

(.n. ) =d
m
W
m
= ~L (W m ) L'CV
m m
=0 (33)
m

5. Distribution function, collision density, Liou ville's eguation

An individual gas-history - a particular complex of world-lines -
is too complicated to be useful; we are interested only in the typi~al,

average properties of gases. Therefore, we imagine a large collection

of microscopically different, but macroscopically indistinguishable gas
histories, a Gibbs-ensemble of gases. The average properties of such
an ensemble are the subject of kinetic theory. (The averaging may have
the additional merit that it disposes of certain all-too-classical features
of our gas model like sharply defined worldlines and collision events;
the average properties may well provide an appro&.imate macroscopic
description of a gas whose particles obey quantum laws. ( , )
Consider, then, a gas consisting of particles of different species.
Concentrate on one component the particles of which have mpss m and
charge e. A definite microstate, or history, of Ue gas can be represen~

ed, as far as the specified component is concerned, as a collection of

( 1 \ Nonrelativistically the Boltzmann equation, e. g., can be "derived"

from classical as well as from quantum mechanics; see Kadanoff
and Baym (1962): Lowry (1970).
 - 323 -

Ehlers

segments of phase orbits in M , the states occupied by particles

m
between collisions. (We do not assign phase-orbits to particles dur-
ing collisions; hence there are no particle orbits in M transverse
m
to tre phase flow. )
The distribution of occupied states in M can be fully charac-
m
terized by the functional 2:--+Nm [ 2:] which assigns to any co~
pact hypersurface 1: the number of occupied orbit segments intersec-
ting it. By a hypersurface in M we mean here and henceforth an
m
oriented, 6 -dimensional submanilbld with boundary of M . The inter
m -
section of an orbit k with I: is counted positively (negatively) if,
at the event of intersection, the vector basis (L m , AI' ... ,A 6 ) has the
same (opposite) orientation as the basis of an oriented coordinate system
of Mm' Lm being the tangent to k and (AI"'" A6 ) an oriented ba
sis tangent to ~ .
If D is any region in Mm' then Nm [ d D] is the number of
collisions in D, if creations are counted positively, and annihilations
negatively.
For a macrostete, let Nm[X]be the ensemble average of
m
N .
Since Nm[X] is a kind of flux through L of a fictitious fluid steeaming
in M with velocity L , we expect it to be expressible as an integral.
m m
WEI thus need a volume element for hypersurfaces in M .
m
It is natural to ask whether there exists a 6 - form on M which
m
could serve as such a volume element. From the meaning of Nm it is

clear that this form would have to assign a nonzero volume to any hype.!:
su rface - element not tangent to L , since there could be a flux through
m
 - 324 -

Ehlers

it. Using the fact that n m

is a non-vanishing 7-form on M
m
(see eq. (27)) and remembering lemma 2, we infer that such a
6-form must coincide with W as defined in (32), except for a non
m -
vanishing factor. Because of Liouville's theorem, eq. (33), it is ad-
visable to choose this factor to be constant on phase orbits in order
that the 6 - form is L -invariant ( 0" d(f W
) " df /\ c.J "df 1\ (L •
m m m m
n )"
m
L (f) n ~ L (f)" 0 ; we have used (1)). Hence, iN
m -"'In --...,,- m m
reoommends itself as an almost unique candidate for the required me-
asure.
A hypersurface 2: M
m
whose projection into X is a space-
in
like hypersurface corresponds to a region of an "inttantaneous" ordin~
ry (it,p)-phase space. On such a L: (and, more generally, on any I:;
whose projection into X is a hypersurface), W from eq. (32) reduces
m
to its first part,

a
p t5"'a A ~ . (34)
m

l: an inertial coordinate sy-

r'
If we choose at some point x with (x, p) E
stem such that is, at x, normal to (34) gives, at x,

(35)

which is, except for the (conventional) sign, the ordinary. phase volume
element of an observer at x with 4 velocity ::x~ . Therefore, Wm
 - 325 -

Ehlers

is the appropriate 6-form we have been looking for ..

We return to our study of M . According to its physical mea!!
m
ing, we make the following smoothness assumptions about N , i. e.
m
about a macrostate of a gas:
D 1) On any fixed hypersurface l: C Mm there exists a continuous,
nonnegative density function f~ such that for all compact parts l:'c:E

Jr' fE W
m
(36)

D2 ) Every point (x, p) , Mm has a neighbourhood U such that for

every region D cue M
m

INm [Cl oJ ~ A n J m (37)

D
for seme constant A dppending on U.
Dl asserts that on any fixed hypersurface ~ the measure defined by
the expectation value of the number of occupied states contained in parts
~I of l: has a continuous derivative, or density function fE
with respect to the geometrical measure c.J.
m
Equation (35) shows that fE (x, p) equals, for any observer who -
se worldline intersects the projection of !; into X orthogonally at x,
the ordinary density of states in his infinitesimal, ordinary (1.-P) phase
space.
D2 asserts that the expectation value of the number of collisions in
 - 326 -

Ehlers

D is at most of the order of the nm volume of D; this assum

ption excludes, e. g., the possibility of having all (or particularly
many) collisions occuring on one hypersurface of X.
The two assumptions D1 and D2 imply theexistence of an in-
variant, i. e., hypersurface or observer-independent (one particle)
(1)
distribution function f on M such that for any hyper surface
m m

N
m J
E
fm (A)m; (38)

To prove (38), we have to show that if a point ~EM is

contained in two hypersurfaces 2:1 and ~2' then fI:l($~=ft2. (~)
For that, consider a tube ~ of phase-orbits having 'S on its boun
dary. Then l.: f'\
t
17
J , L f1 J are
'1.
two cross sections of ~ . With~
out loss of generality we assume that these two cross sections to -
gether with the part I\.. of the cylindrical buundary ~ j which lies
between 2:1 ('\~]: and ~'Ll"\dj form the boundary 'dD of an orien-
table, compact region D of M . Since no phase orbits can inter -
m
sect A we have Nm [dD 1 Nm[~1"'1] Nm[l:1.,,1]. Because of

\ 1) The preceding introduction of f is a generalised and "rigori-

sed" version of that given in S!J(ge (1957), p.12 14.
 - 327 -

Ehlers

D1 and with the mean value theorem for integrals this can be re-

written as N
m
I wm - f~~1
[dD] = fop~1 ($)11:;.":1 (~)
t
J
t.n)m
W , where ~.
~
€ L:'. ""g,
'I.
But we know from Liouville's theorem that the two integrals on the
right-hand side are equal. Hence, using also D 2, I fE1 (~1) - f1;1. (~2.) \
~ A Jnm (1:",'.}IIlJm )
J)
-1. If one now lets
~
~ shrink towards the or-
bit !Jassing through S , the right-hand side tends to zero since the
numerator is "one order smaller" than the denominator. Also, ~i.--' g,
Consequently, f~ (~) = f~ (s). We call the common value f (S),
~f ~~ m
in order to emphasize that f
is defmed on 1\1 •
m m
It is easy to verify that our orientation and sign conventions im-

ply

f
m
) o. (39)

It is techilically desirable and physically not harmful to require

also
D3 ) fm is continuously differentiable on Mm.
Having obtained a phase space density which measures the f
m
average density of occupied states, we obtain straightforwardly a
collision density in M . The average number of collisions in the
m
region D eM, i. e., the difference between creations and annihi-
m
lations of particles of the specified kind in D, is given by N [dD]
= "\
C1D
f fm W '
mOm m
J
d(f w ) =
D
Jdfm A LV =
m
DI dfm ,,7Lm ,Wm )
 - 328 -

Ehlers

=I D L
m m
(f
m
) n. .
We have used equations (38), (2), (33), (32), and (1). Hence
a j} f ). b r). b c d f"", (40)
Lm (fm) = p ;} x":. + (e F bP - , bcP P ) e)P>"
is the collision density in M with respect to.n (in the sense
m m
defined above).
a a
Note that if (x (v),p (v)) is the phase orbit passing through
(x, p) for v = 0, then the expression (40), evaluated at (x, p, ),
d
equals (Tv fm (x(v), p(v)))v=O' a fact that is often useful.
The preceding considerations prove the following theorem. The
distribution function f
of a component of a (possibly heterogeneous)
m
gas satisfies Liouville's equation

L (f) 0 (41)
m m

in a region D C M if and only if there is detailed balancing every

m
where in D, i. e., if the average number of creations of particles of
that component equals everywhere in D the average number of annihi-
( i )
lations .

( 1.) Note that, in our terminology, even an elastic collision involves

two annihilations and two creations.
 - 329 -

Ehlers

( i )
Corollary 1.. If the particles of a particular species do not pa!:,
ticipate in any collisions in D, then the corresponding distribution
function satisfies, in D, equation (41).
Corollary 2. If the assumptions of the theorem hold, then f is, in
m
D, an integral of the motion defined by (18).
As an application of the invariance (observer-independence) of
the distribution function, let us consider a radiation field as a pho-
ton gas with distribution function f r . Relative to an observer with
4-velocity u a , it is customary to define a specific intensity I~ of
the radiation field, as the limit of the radio "(energy of photons with
frequency in d oJ and direction in solid angle dn passing in time dt
normally through an area dA/ (doJ dndtdA)." It is related (exercise)

to fr by

(42)

Since V 2 rr )-1 J u a Pa I ' the observer-independence of f r im

( 1) For geodesic motion (e = 0), this assertion has first been stated
by Walker (1936).
 - 330 -

Ehlers

plies that of 1-.,) / '\)~, a fact that is important, e. g., in cosmology;

its direct, kinematical proof is somewhat cumbersome.

If the photons are emitted by a source S (galaxy, e. g.) and do
not interact with matter on their journey to the observer 0, Liou-
ville's equation (41) for fr and (42) give the important relation

Iv
s (43)
(I+Z )~

between Iv ' "measured" near the source by a fictitious comoving

5
observer, and I" ' the intensity actually measured by O. z is the
o
usual redshift of S relative to O. (43) is basic for the derivation of
observable relations in cosmology. Notice that the derivation just
sketched holds in any spacetime, not only in the standard Robertson-
Walker universes.
If one assumes that the famous 3° K "fireball" radiation was
emitted thermally from the recombination hyper surface (T~' 3500 0 )

in the early universe, one obtains from (43) the predicted intensity
distribution in each direction in an arbitrary model universe, provided
( t )
one can compute z from the null geodesics.
This idea was used by R. K. Sachs and A. M. Wolfe (1967) to estimate the
influence of material "lumps" on the radiation, and similar applications
have been made more recently. The same method has been employed by
W. L. Ames and K. S. Thorne (1968) to determine the optical appearance

( 1)
It is also assumed that no scattering occurred between emission and
absorption.
 - 331 -

Ehlers

of a collapsing star to a distant observer. Several other applica-

tions of (41) have been made, particularly in cosmology and stel-
lar dynamics.

6. Macroscopic fluid variables, balance equations, conservation laws.

Let us rewrite (38) for a hypersurfa(;e E whose projection in-
to X is a hypersurface G. We obtain, using (34),

Nm [L] J f rJTt
G
f). {
Kx
m }
(44)

K is that part of the mass shell P (x) which is contained in L

x In particular, the integral 15"a m{ Jfmpa 1tm } gives the
average total number of particles 1f the species considered whose world
lines intersect G. Here we have used the convention, to be maintained
throughout the remainder, that J~. .. denotes an integral over the
whole mass-shell P (x). Therefore, the spacetime vector field
m

N
m
a (x): = J f pa
m
rc m (45)

is the particle 4-current density of the respective species. It is always

timelike and future-directed under our assumptions. (If we would permit
to be a distribution, N a could be lightlike in one particular case:
m m
m = D, and theOre is no 4-momentum dispersion at any event).
 - :1:l2 -

Ehlers

Similarly,

a a
J : e N (46)
m

is the electric 4 - current density of the species considered.

In analogy with (45) we define

T ab(x)
m
f a bf
PPm T(m (47)

as the kinetic stress energy momentum or matter tensor of the

species. (If is possible to define a 4-momentum flux through a
hypersurface G C X and to show that (47) is the corresponding
4-momentum flux density, but this has no further use and is there-
fore not treated in detail here. )
We have assumed here, and will do so throughout these lectures,
that f vanishes at infinity on P (x) so that integrals like (45), (47)
m m
exist. (Sufficient for this is exponential boundedness on P (x), as de-
m
fined at the end of section 8.)
Excluding the trivial case where vanishes on P (x) (and the
m m
singular distribution mentioned below (45)) we infer from (47)
 - 333 -

Ehlers

Lemma 5. If va is not spacelike and va f 0, then

T
ma
bVa v b> O. (48)

(i )
This lemma and a theorem due to J. L. Synge imply

(t)
Lemma 6. Any kinetic stress energy momentum tensor is normal ,
i. e., admits a decomposition

ab a b ab
T
m jU u +P (49)

with

a
u u - 1, O. (50)
a

a
u can and will be chosen future-directed, and then (49) is unique.
The physical meaning of
N a, T ab for a local observer in
m m
terms of 13-dimensional" quantities is obtained by evaluating (45) and

( 1) See Synge (1956), p.292.

( 2.) See Lichnerowicz (1955).
 - 334 -

Ehlers

(47) in an inertial coordinate system at x. We obtain, for an

arbitrary observer at x:
N 4 is the number density
m
-Nm '. = N).m ~ 4 <_v~m)
d xl = N m /X is the particle flux density,

T 44 = N 4
m m
<E >1Tf\,)
X
is the energy density, (51)
).4 ~
<p I.
-
4 (m)
Tm T 30: =N
X
is the momentum density,

T
m
m OIX
). ~ ';\
T rr! ~
4
aI
m
';)
~xl4 = N m
4
p ® v
<-. .. 'X
{m\
is the kinetic

pressure tensor.
Here V. E (= p \ and p are the 3- velocity, energy, and
3-momentum, and ( ),"") denotes the conditional expectation value
X
at x, evaluated by means of the probability distribution defined by
f with respect to the chosen inertial system.
m
We also define mean kinetic pressure p by (tr : = trace)

p : = 31 tr -Tm ~3 4
Nm <"Po"v '\
/ x(m) (52)

and recognize the classical Bernoulli formula.

The rest mass density is p: = m N~. Writing for the
energy density in (51)3' we formulate

Lemma 7. For any observer and any distribution f , the inequa-

m
 - 335 -

Ehlers

lities

hold.
Most of these inequalities are obvious from (51), (52), and E =
m
(1 _ Vi ) Vl ' P = E ~; the only nontrivial inequality is the
third one, due to A. H. Taub (1948). It follows by considering
2 t 2 !
(1 - v fi and (1 - v ). as elements of the Hilbert space
"p'I.(P , f dpl2\ and applying Schwartz's inequality to them.
(J.. m m
Equations (51) and (52) imply the well-known relations:
p« p , then fA' ~ f+ ~
r
If p (nonrel. monatomic gas),
If p» f ' then ~ 3p (ultrarel. gas), (54)
If m = 0, then r= 3 P (photon or neutrino gas).
In order to obtain balance equations for various macroscopic
fluid variables we observe that these latter quantities are moments
of the distribution function in 4-momentum space, given by

J at a\
p p ... p
a.
fm 11m' The Oth moment is, at least for m > 0,
essentially the trace of the matter tensor. Indeed. (47), (51), and
(52) give

r- 3p. (55)
 - 330 -

Ehlers

The 1st moment is the particle current density N a, and the

m
2nd moment is the matter tensor. We would like to evaluate the
divergence of the r~th order moment. We first establish

Lemma 8. If g is a C 1 -function on M , then

m

J L
m
(g) 1lm . (56)

To prove this, take an arbitrary region D C X, and let

'S : = { (x, p) : x E. D, P E Pm (x)} be the cylindrical region of
M lying over D. Then, as in the derivation of the collision densi-
ty r:bove eq. (40), rg w =
;}/' m (~m
L (g) r n m . We transform both
these integrals into integrals over D:

J"g Wm =f G'a { f pa g 7t m } =[~ (J pa g nm ) ; a J

dD dD
(Use (34) and (7))

f Lm (g)!l m = Ji~JLm(g) 'Tt m } , from (26).

"'D" D
Since D is arbitrary, (56) follows.
 - 337 -

Ehlers

Next, we generalize this to obtain an "equation of transfer".

(1 )
Lemma 9. If f is an arbitrary distribution function on
m
M , then (for r ~ 2)
m

L (f)'TC +
m m m

J
to

,
+ ~ e Fa). b p a ...
1 ·· b
p ... p a" fm TC m · (57)
}.: t

( The integrand in the sum is. to be understood such that b replaces

a). in the sequence a 2... a r .)
Proof: Take an arbitrary tensor field v ... which satisfies
at ... a ..
.
= 0 at somearbltrary event xO'
a~ ~
v Put p ... p Y .••
a 2: ... a .. ; b at
f = g, and apply lemma 8, to obtain at xO:
:: a .. m

Evaluate L m {. .. ) at Xo by taking the phase-orbit through Xo and

( i) Tauber - Weinberg ( 1961)

 - 338 -

Ehlers

differentiate C.. ) with respect to the parameter at xo' getting

L C.. ) ~()_
d ... - v ~(a'1.
d P ... p a,.)
f
m v a1... a.. v m

= va..... at' (p at ... Pa" L m (f)

m + e F a \ bPb ... par f m + ... ),

where we have used (18). Insertion into (58) and "dividing" by

v... gives (57).
(Generalizations of Lemma 9 have been given by Ph. M. Quan
(1966) and C. Marle (1969), but they do not seem to have found
applications yet.)
Applying (56) (with g = f ) and (57) (with r 2) and u-
rn
sing the definitions (45), (46), and (47) we obtain

N a·a =
m'
J L (f ) 11
m m m
(59)

and

(60)

Equation (59) is the balance equation for particles of the

species considered; since L (f ) has been shown to be the col-
m m
lision density in phase space, tile right-hand side of (59) is the
 - 339 -

Ehlers

(spacetime) production density of these particles.

Equation (60) is the 4-momentum balance equation for the gi-
ven species; the vectors on the right-hand side represent the elec-
tromagnetic and the collisional 4-force densities acting on the col"!!.
ponent of the gas with distribution function f . An example for the
m
latter is the force exerted on an electron gas by photons due to Comp-
ton scattering.
So far, we always concentrated on one component of a gas which
may contain other kinds of particles as well; all our equations are
valid for any component of a mixture.
Let us now first specialize to the case of a monocomponent, or
simple gas consisting of particles all having (proper) mass m and
charge e. Then, assuming conservation of particles in collisions (59)
gives the conservation law

N
m ;a
a
S L(f)'t(
m m m
0

a
which, of course, implies also charge conservation, J ;a o.
Assuming also 4-momentum conservation during collisions, (60) re -
sults in

T ab o. (62)
m ;b
 - 340 -

Ehlers

These equations give on the one hand the macroscopic conserva

tion laws basic to fluid mechanics, and they impose restrictions on
the evolution of f ,required by the microscopic conservation laws.
m
For a simple gas, there are two sensible ways to define the
mean 4-velocity. One can either use the fact that the particle cur-
rent vector N a is- timelike and put
m

N a a a a
n uk ' uk u ka = -1, uk future-directed, (63)
m

or one can use the normality of the matter tensor T ab and use
m
the ua of Lemma 6, i. e., require

a a
0, uDu Da = -1, u D future-directed (64)

a (1 )
uk is called tre kinematic mean velocity. An observer travell-
ing with uk a is characterized by the property that in his local iner-
tial frames there is no particle flux density (see (51) 2); n from
(63) is the proper particle number density.
a (( )
u D is called the dynamic mean 4-velocity. An observer travellr

( i) The distinction and terminology is due to J. L. Synge'; see Synge (1960)

 - 341 -

Ehlers

ing with it will measure no momentum density, and this characte-

rizes u Da (see (51)/ The energy density f in (49) is the
minimum of the energy densities measured by all possible observers;
this property also characterize s u Da (exercise).
The two mean velocities uk a and aD a are in general distinct,
their equality characterizes (by definition) adiabatic processes. They
are physically characterized by the existence of an observer ua who
finds neither a Jll. rticle flux nor an energy flux in his local inertial
systems. The necessary and sl~t'ficient condition for that is that
N [aT b] N c = 0, a very complicated restriction on the distri-
m m c m
bution function.
a
If one chooses ...!!!!Lmean 4-velocity u , one can decompose the
matter tensor uniquely according to the scheme (Eckart 1940):

T
m
ab
r- (a b)
u au b + 2 u q + P h ab + 'T(..ab, (65)

where

ha
$a.b + uaub (66)
b

a
projects T (X) onto the 3 - space orthogonal to u , and where
x
 - 342 -

Ehlers

u
a
o. (67)

U. is the mean energy density, qa the mean energy flux densi-

J ab
ty, p the mean kinetic pressure and rc; the shear pressure
a
tensor relative to u a , These quantities change with u a . If u =
a
then q 0, Adiabatic processes are characterized by the
property that q
a
o for u a = uk a . If, in addition, Tt~ = 0, the
gas behaves, in the process considered, as an ideal gas. We shall
extend these mechanical considerations later on to the regime of ther
modynamics.
Consider next a multicomponent gas. We distinguish the particle
species by indices A, B, ' , ,; particles of species A have mass m A'
charge e A' and (if we have microscopic particles) further characte-
ristics like baryon number bA etc, Each species has its phase space
which we denote by MA (instead of M ), and its Liouville opera-
mA a a
tor LA; its distribution function fA' current densities NA' J A'
and its matter tensor TA ab, and the quantities which we have defined
in terms of these, like u~ , A.
Requiring again 4-momentum conservation, we have instead of (62)

T ab = Fa Jb (68)
k;b b'
 - 343 -

Ehlers

where

T ab
k
LT~b (69)
A

is the total kinetic stress energy momentum tensor of the mix-

ture, and

Ja: = LJAa (70)

A

is the total electric 4-current density. Moreover,

(71)

The indlVidual particles will in general not be conserved during

collisions, but certain combinations of the NA a will have vanish-
ing divergence. For example, if we define the baryon current density

(72)

and assume conservation of baryon number during collisions, we obtain

;a
o (73)
 - :344 -

Ehlers

and

(74)

Similarly, we will have

(75)

and

(76)

Thus, we obtain macroscopic conservation laws for a mixture and

corresponding integral conditions for the distribution functions.
Resonable mean 4-velocities for a mixture are the dynamic
mean 4-velocity ua defined as in (64), with Tab replaced
ab D m
by Tk '
the barycentric mean velocity, defined by

(77)

and the baryonic mean 4-velocity, defined by

(78)
 - 345 -

Ehlers

provided Ba is timelike, as it is for "ordinary"matter.

With any choice of mean 4-velocity, one can decompose Tk ab
according to (65), obtaining r' p etc. for a mixture. Which 4-
velocity is the most useful one depends on the circumstances; a
careful investigation is not known to the author.

7. The selfconsistent Einstein-Maxwell-Liouville equations (t )

Consider a collisionless mixture of particles, so that (41)
holds for each component, and consequently (71), (74), (76) are
trivially satisfied. Then, we have the macroscopic conservation
laws (73), (75) and the (generalized) Poynting equation (68). It is,
therefore, permissible to assume that gab' F ab are the mean fields
produced by the gas, i. e., to require that they satisfy the Einstein-
Maxwell field equations:

ab
G + /\. gab TK ab + T ab
M '
(79)

F 0,
[ab, c]

( 1) Compare with Tauber-Weinberg (1961) who apparently first advocated

these equations.
 - :146 -

Ehlers

where

T ab (80)
M

is the Maxwell stress energy momentum tensor. Indeed, (80) and

. ab _ a ,b
(79)2 Imply Tl\T 'b - - F b" and if this is combined with
(68), there results' (T ab + Tab) = 0, as required by (79).
k M;b
Hence, the equations (79) together with the Liouville equations

(81)

seem to provide a closed, consistent system of dynamical equations

for a gravitating plasma (in the Vlasov approximation).
For neutral particles, (79)1 (with T Mab = 0) and (81) give
a relativistic version of the equations of stellar dynamics (for col
lisionless systems).
It is natural to pose the Cauchy initial value problem for the
system (79), (81). Formally, there seems to be no obstacle to sol~

ing it in the usual way by separating initial constraints from evoluti-

ona equations, the former being propagated off the initial hypersurf~

ce in consequence of the evolution equations, which in turn can be

 - 347 -

Ehlers

solved for the highest derivatives off the initial hypersurface, provid-
ed that is not characteristic. A careful elaboration for the present
system (79), (81) does not seem to have been performed, however. (1)
Examples of sulutions to the equations (79)1 (with T Mab = 0 ),
(81) are known, see E. D. Fackerell (1966), 1968), J. Ehlers, P. Geren

and R. K. Sachs (1868), R. Hakim (1968), R. Berezdivin and R. K. Sachs

( t )
(1970); see also Misner (1968), Stewart (1969), Matzner (1969).
Solutions with electromagnetic fields do not seem to be known at pre-
sent (Problem).

8. The Boltzmann eguation

Consider again a multi component gas with particle species A, B, ...
If collisions occur, then the phase space density of all collisions in
which particles of type A participate, LA (fA)' will be a sum (or in-
tegral) of various contributions due to different kinds of collisions, e. g. ,
elastic and inelastic binary collisions, absorptions and emissions.

( 1) For a series of papers on the stability theory of static, spheri-

cally symmetric solutions of (79), (81) (for Fab = 0), see J. R. Ipser
and K. S. Thorne, Ap. J. 154, 251 (1968), and subsequent papers by
Ipse r in the same Journal.
( 1) (Note added in proof) Meanwhile, the problem has been solved by
Y. Choquet-Bruhat; see Journ. Math. Phys., 1970, and another
forthcoming paper.
 - 348 -

Ehlers

Let the symbol

( x;PP A' PB' ... (82)

stand for a collisions in which particles of types A, B, . .. with

respective 4-momenta p A' PB' . . . collide at x E. X and produce
particles C,... with PC""; the numbers of incoming and out-
going particles may be arbitrary. (If, e. g., A = B, one has to
write pA' PA instead of pA' PA; this is tacitly assumed here and
in the sequel.)
The set of all collisions (82) of a particular type, with x € X,
PA € PA(x), . .. is again a bundle over X, the collision bundle. It
carries a measure, viz., '? 1\ 1tA A 1TB 1\ ... A 1t C A ...

Augmenting our former smoothness assumption D2 concerning the

probability distribution of collisions we make the hypothesis:
C 1) In any macrostate of a gas, the average number of collisions
(82) in a compact region U of the collision bundle is

J
U
V (x; PA' PB" ..-PC' ... ;) S(Ap) ,?AnAA 1'CB· .. A 'fCC' .. (83)
i
where V is a nonnegative (ordinary, measurable) function. ( ) (In order
to avoid ambiguities in the definition of V, U must be such that 4-momen
tum ranges KA (x), KB(x), ... of indistinguishable incoming or outgOing parti-
cles

( i) Because of the S-factor in (83) the physically important domain of

definition of V is (X; p A' PB'" . -PC'" .) : Ap = 0
 - 349 -

Ehlers

( A = B =... ) do not overlap.

In (83), Jis the Dirac distribution (on R\ and

(84)

is the 4-momentum difference between "in" and "out" states. The

J -factor in (83) expresses that collisions (82) occur only if they
conserve 4-momentum.
It then follows that the distribution functions fA' fB' . . . of
a gas satisfy equations of the form

In this "collision balance" the sum is to be taken over all kinds of

collisions in which A-particles participate, either as incoming or
as outgoing collision partners. The integral goes over the mass -
shells of all colliding particles except the one whose state occurs
on the left-hand side of (85). r~ is a numerical factor depending
on the type of collision and on whether the state pA on the left-
hand side of (85) is an "in" or an "out" state; it is defined thus:
r~ >0 «0) if A is an "out" ("in") state, and
 - 350 -

Ehlers

Ir~ I
-1
n A ( n A ! n B ! ... nC !) ,where nA' n B, . .. are the
=

numbers of (indistinguishable) particles of types A, B, . .. ente£

ing or leaving the V-collision, and nA refer to the number of
particles to which the left-hand state in (85) belongs.
"
(If we h ave a co11lSlon (p A' PA
I ~ ..
"-7 PA' PB' PB
I )

f
with A
hand side of
B
(85), then
and
r~ = -2 ( 2! 1! 2! )-1=
factor is necessary in order that the various collisions involving
+ .)
PAis the state occuring on the 1eft-
This

identical in (or out) particles are not counted several times in

the balance (85).
The equations (85) are useless as long as the dependences of
the functions V on the state of the gas are not specified. ( ') It

is clear from nonrelativistic statistical mechanics that in a rigo-

rous many-particle theory V will depend, not only on the one-par-
ticle distribution functions fA' fB' ... , but also ( at least) on pair

(1 )
See, e. g., Liboff (1969).
 - 351 -

Ehlers

correlations gAB (x, p A;x! pj/ No attempt will be made here to

cope with these difficulties which pose important and interesting
problems. Rather, I shall write down a IIreasonable Ansatzll (as
people say); then I shall make some remarks about the IIphilo-
sophy II which is used to motivate that Ansatz; then modify it so
as to account for the non-classical symmetry character of Bosons
and Fermions; and then simply proceed on the basis of the result-
ing (generalized) Boltzmann equation.
Consider the hypothesis

C2 ) V(x;p A' PB' .. ~ PC' ... ) " fA (x, pA) fB(x, PB)" . R(p A' PB' .-:t' PC' .. )

(85)

.
In wh'IC h th e fac t ors fA' . .. re fer t 0 th e II·
In II st t on
a es l y.
(C 2) is suggested by the assumptions that
(a) particles which are about to collide have uncorrelated momenta,
(b) the ranges of the coll.isional interactions are small in compari-
son with the scale on which the fA'S change appreciably with x,
(c) collisions take place in spacetime regions so small that the mean
differential gravitational field Ra bcd (geodesic deviation etc. !) and
the mean electromagnetic field F ab do not affect their frequency
 - 352 -

Ehlers

appreciably.
(d) the presence of particles not participating directly in a col-
lision does not affect the probability of occurence of that colli-
sioll.
These assumptions, which essentially express that the gas
is dilute ((d) and, for a gravitating gas, (c)), not too inhom~

geneous in spacetime (b) ), and in a state of high randomness

( (a) ), indicate the range of validity of the "Boltzmann collision
hypothesis" C 2; each of them poses a problem of justification and
indicates desirable generalizations. If, e. g. ,
(t )
(c) were not true, then R might be expected to depend on
the principal directions and eigenvalues of R abcd and Fab.
In order to support the assumption C 2 further and to relate
it to scattering theory, let us consider a collision (p A' p~ PC' ... )
with two incident and q emerging particles, and let us consider

Compare Marle (1969), p. 88.

 - 353 -

Ehlers

those collisions for which the momenta are contained in small ranges
KA CPA (x) etc.
According to (83) and (86), the number of those collisions
per unit spacetime volume is

(87)

Regarding the KA -particles as a beam which hits the KB -parti-

cles forming the target, we recognize that the number densities of
projectiles and target particles, relative to any inertial frame with
4-velocity u, are given by (see (44) )

whereas the relative velocity of these particles is

I
(u.p A) PB - (u'PB) PAl
(88)
(u'p A) (u'PB)
 - 354 -

Ehlers

Hence, we can rewrite (87) as

(89)

where

(90)

Equation (89) is recognized as the standard definition of the dif-

ferential scattering cross section dQu for scattering of pA' PB-
particles into the ranges K, .. , relative to the u-frame, and e-
c
quation (90) is indeed the correct expression far that cross sect-
ion which can be derived
0(.) in the non relativistic limit either from classical or from quantum
mechanics, and
!) in the relativistic domain from quantum scattering theory.
( £)

( t)
See, e. g., Brenig and Haag (1959)
 - 355 -

Ehlers

(1 )
In this case R is simply related to the S operator .)
( In the relativistic case, a classical derivation is not availa-
ble, since there is no well developed theory of interacting particles. )
In a certain sense, we have now justified (85), since under the
assumtions stated above the A- and B-particles in the gas should
behave as if they were members of beams in a collision experiment.
One correction, or generalization, of (85) shall now be made. If
the particles are atomic or sUb-atomic, then assumption (d) is defi-
nitely wrong. In the case of Fermions, the presence of particles in
the final states decreases, because of the Pauli principle, the collision
probability, whereas for Bosons it enhances that probability (stimu-
lated emission and scattering). This is incorporated by writing, in-
stead of (86),

(91)

( 1 )
See,e. g., Brenig and Haag (1959)
 - 356 -

Ehlers

Here and in the sequel the upper sign refers to Bosons, the lower
one to Fermions.

is the volume of a phase-cell which corresponds asymptotically to

( 1 )
a non-degenerate p-eigenstate of a free (quantum) particle of spin
degeneracy rC' Hence, sCfC(x, PC) equals approximately the average
occupation number of simple one-particle pc-eigenstates localised near
x. (In the "classical limit" fC«s~l, (91) reduces, of course, to
(86).
A "pseudoproof" of (91) can be given within the Fock-space for-
( t )
malism, but that will not be reproduced here .
One simplification is possible and useful in (90).
If u = ApA+ fPB - and these frames include the center - of - mass
frame of the collision as well as the rest frames of the incoming particles-

( 1 ) Weyl (1911), Peierls (1936)

( \)
See Bichteler (1965), Ehlers and Sachs (1968), Ehlers (1969).
 - :357 -

Ehlers

of u, whence the corresponding expression

dQ (92)

is often called "the" (relativistic) cross section.

We have now suppressed the arguments KC"'" considering dQ
henceforth as a(3q - 4) - form, where q is the number of final sta-
tes and we imagine that ~ ' .. ) has been "absorbed" into four of the
differentials T(C 1\ ... : (If q = 1, dQ is a ~- "function":
dQ = ~(-PA'PB)
J(l2 2 2 2
[mA+mB-mC ] - (Pi PB)) G"a is the ab-
sorption cross section.)
Inserting (91) into (85) we obtain the generalized Boltzmann
equation

where we have simplified the notation in an obvious way. In particu-

lar, in (93) and henceforth,

(94)
 - :l5H -

E:hlers

the factors Sc have been absorbed into E. .. (r R

l\
equals the

former r~.)
The equation (93) has first been formulated in special relati-
vity for a classical (i. e., Boltzmannian) gas with elastic binary in-
teractions by Lichnerowicz and Marrot (1940); for other treatments
and generalisations see Tauber-Weinberg (19CiO, Israel (1963),
Bichteler: (965) and the papers mentioned in the introduction.
Henceforth we shall require the Boltzmann equation (93) to
hold for the distribution functions of any gas. (Other "reasonable"
alternati ves for V which lead to different kinetic equations are po~.

sible, but will not be discussed in these lectures.)

One important symmetry needs to be mentioned. If the microsco-
pic collision law (S-matrix) is invariant with respect to the total re-
flection, PT, then the collision "matrix" R:: is invariant with res-
pect to an interchange of incoming and outgoing states
( , ):

(" )
See Brenig and Haag (1959).
 - 359 -

Ehlers

R C .. . RAB .. . (95)
AB .. . C .. .

We also add that, for Fermions, it is necessary that

(96)

due to the exclusion principle.

The conservation law (71) is satisfied by (93), because of thEl

~ (~p) - factor. Other conservation laws like (74) can and have
to be incorporated by similar restrictions on the R-functions; this
will be assumed in the sequel.
It is now clear that we can generalize the selfconsistent field

equations of section 7 so as to take into account collisions; we just

have to replace equation (81) by (93). The remarks about the Cauchy
problem made in section 7 still hold; a rigorous analysis for the system
(79), (93) has not been performed, however.
In a given spacetime X (with a metric of class C 2 ), i. e., for
a test gas, Bichteler (1967) has solved the (local) Cauchy problem
(t )
for (93). Besides existence and uniqueness of exponentially bounded ,

~(x)pa.
( t )
on the mass-shell, i. e., IfA (x, p) I~b(x) e J& ,with band Ja
depending continuously on x.
 - 360 -

Ehlers

nonnegative, continuous and a. e. differentiable solutions for given

initial distributions of the same type, Bichteler has established
the continuous dependence of the solution on the initial distribu-
tions, the metric, and the cross section (i, e., R::). He assumes
throughout that the total cross sections
(all final states)
f dQ are

bounded. (This last aSBumption, though perhaps valid for strong in-
teractions. ( , ), does not seem to hold, e. g., in the case of weak
interactions. ( t) Bichteler obtained his results by applying Banach's
fixed point theorem to an operator given naturally by means of (93).
defined on a suitably chosen complete metric space of exponentially
bounded distribution functions. As Bichteler pointed out, his results
lend some credibility to the (formal) Chapman-Enskog approxima-
tion which will briefly be discussed later.

( i) See Eden (1966)

( t.) See Bahcall (1964)
 - 361 -

Ehlers

9. The second law of thermodynamics (H-theorem).

We define the entropy current density of a gas to be the

4-vector field

(97)

with sA clefined as before (below (91) ).

The expression (97) can, in a sense, be derived from an in-
formation-theoretic point of view as indicated in Ehlers (1969). In
the classical limit sA fA.-.. 0 it reduces to

sa = L {f pafAlog (s AfA) TCA - N~ }, (98)

and one recognizes in the first term of S4 the Boltzmann entropy

density. Generally, _Sau is to be interpreted as the entropy densi-
a
ty relative to an observer with 4 - velocity ua.
Using Lemma 8 one obtains

sa; a =~ f LA (fA) log( (sAfA)-b) rcA' (99)

i\
inserting LA (fA) from the Boltzmann equation (93) and assuming the
 - 362 -

Ehlers

PT - symmetry (95) for all collisions involvee one gets a

sum of terms; one from each kind of collision and its inverse,
of the form

where we have again used the notation (94). Each such integral
is nonnegative, since its integrahd has the form
a
( log b) (a - b). Hence,

~o. (101)
;a

This is the relativistic form of Boltzmann's H-theorem ( Tauber-

Weinberg (1961), Ehlers (1961), Chernikov (1963), which expresses
locally the content of the second law of thermodynamics in the
framework of kinetic theory.
(101) implies that the flux of Sa through any closed hyper-
surface in X is nonnegative. Hence, for an adiabatically enclosed
 - :l(j:l -

Ehlers

or isolated gaseous body the total entropy

S[~] : (102)

evaluated on a spacelike cross section of the world tube of the bo-

dy, never decreases towards the future. (Notice that in the classical
limit (93), srI] consists of the total number of particles and the
Boltzmann contribution. If the total particle number is not constant,
the Boltzmann S-term alone does not necessarily increase. )
Notice that (101) does not follow from (93) if the collisions are

due to PT - violating interactions.

If (95) does hold, and if collisions occur frequently in a gas, then
the competition between collisions of a certain kind and their inverses
suggests the tendency of the gas to evolve in such a way that the diffe-
rence in the integrand of ( 100) te'lds towards zero, so that ultimately
the entropy production density Sa vanishes and the Liouville equa-
;a
tions (41) holds, orovided there are no disturbing external influences.
Unfortunately, precise theorems supporting this physical expectation are
so far missing in relativistic kinetic theory; even at the nonrelati vistic
level little is known. (For a brief discussion see, e. g., Uhlenbeek and
Ford (1963), p. 31.) Any result in this direction would be of interest. It
would also be of some interest to know whether in situations of gravita-
tional collapse S may increase toward>: infinity,
 - 364 -

Ehlers

10) Stationary states, equilibrium, and thermostatics

A gas given by gab' Eab , fA is said to be in a stationary

state in a region D c: X if there exists, in D, a one-dimen
sional local group G of fixed-point free local isometries with
timelike orbits which leaves F ab and the fA invariant. In terms
of the generating vector field ~a of G the last two conditions
can be expressed as

(103)

(104)

moreover , we then have Killing's equation

r?( a;b) =0 (105)

The last two equations imply

(106)

and similar statements for Na Tab etc, Because of (105) it fol

A' A
lows further that
 - 365 -

Ehlers

0, (107)

i. e .• the entropy production is constant on the G-orbits.

Let us assume now that an adiabatically isolated gas is 111

a stationary state in D, and that the boundary of the world tube

'} of the gas is . G-invariant; ~C D. Let ~ be a space like
cross section of ~ and a E G. Then a( E) is again such a
cross section, and because of the assumed stationarity S [a(E )] =
= S rr]. Applying Gauss's theorem to the part of!J between E

and a( L), using the adiabatic condition along the wall'dj- , and
taking account of (101) we obtain in '}

Sa O. (108)
;a

This conclusion, combined with the expectation described at the end

of the previous section, leads us to define:
A gas. is in local equilibrium at x E X if, at x, Sa ~a = O.
a
The formula (100) for a summand of S;a shows the validity
of the first part of the
theoreme. If the collision functions R'" of a gas are all strictly
positive almost everywhere (w. r. t. the measure a(.t:)p) 1tA 1\ . .. )
and continuous, then the gas is in local equilibrium at x if and
 - :1 (j G -

Ehlers

only if at x

AB
f f ... fC... (109)

whenever ~ p = 0, for all types of collisIons which occur; or,

equivalently, if and only if for each particle species. on P A (x)
there holds

o. (110)

The second part of this theorem follows from the first part by
means of equations (93) and (99).
The restriction R:: >0 is not unsatisfactory from the physical
point of view, since the R - functions are usually analytic functions of
the momentum variables on the "collision fiber" Ap = 0, and hence
they vanish only on sets of measure zero.
The problem of finding the general continuous solutions ( fA' ... )
of (109) has been solved for binary elastic collisions between Boltz-
mann particles, where (l09) reduces to

fl fl whenever PA+ PB = pIA + pIB· (111)

A 13
 - 367 -

Ehlers

In this case, the general solution is given by

(112)

and a similar formula for fB and with Ja the same for both spe-
cies. (Chernikov (1964), MarIe (1969) and, in the case where the
1
fA'S are assumed C, Bichteler (1965), Boyer (1965). The nicest
proof is that of MarIe, the shortest that of Bichteler.)
If we consider elastic binary collisions between Bosons or Fermions
( or a mixture ) and assume that all factors in

(113)

are positive on their mass-shells, we may divide by fAfBf~f~ and

obtain for __ f - :±: 1 etc. the same relation as for Boltzmann par-
I_
sA A
ticles, so that we obtain

a
_()( (x) - B (x)p
e A Ja -+ (114)

Whereas it is easy to deduce from (111) that fAfB f 0 everywhere

provided that holds for some pair pA f PB' this does not seem so
 - 368 -

Ehlers

obvious in the case (113). Nevertheless I shall accept (114) as

the general form of an equilibrium distribution at an event x
for parjicles participating in some kind of .binary elastic collision.
If particles in a gas undergo not only b-inary elastic collisions,
but in addition other kinds of reactions, then (114) and (109) show
that the OC A must obey

Oi A + O(B + ... (115)

for all permissible collisions A + B + ... ~ C +

With (114) and (115) we have obtained the general local equili-
brium distributions (fA' fB, ... ).
Since the fA'S have to vanish at infinity on the mass-shells, ,a(x)
must be a future-directed timelike vector. We put

(116)

It is a straightforward matter to obtain from (114) the quantities

a ab a
NA , T A , SA' nA, rA' a a
PA' uK,u D defined in eqs. (45), (47),(97),
(63), (64), (49), (52), respectively. Working in the rest frame of u a
one gets

(115)
 - 369 -

Ehlers

(116)

T ab = a b ab
A (fA + p A) u u + p A g , ( 117)

with the scalars (we omit temporarily the index A) n, f' p, s

by

( 118)

( 119)

( 120)

r
s ---
- 2 rc'l.
m
These functions and further thermostatic relations obtained from
them have been studied extensively; see, e;g., Landsberg and Du!:.
ning-Davies (1965) and the references given there.
The thermostatic meaning of the two parameters oc..! is reco-
gnized thus: observe that

s = - Otn + !r+2~1 f
00

log( 1+ ecJ.-/E) f E2_ m 2 ' E dE.

m
 - 370 -

Ehlers

Transform the last term by partial integration and get, with (120),

s =- (122)

Use (120) and compute, again integrating by parts,

dp = } dO( (123)

(122) and (123) give

dr = ! -1 ds + c:J., -1 dn ( 124)

Now,
where
r
T
(s, n) is a thermostatic pot~tial, and
is the temperature
df = T ds + fdn,
and fA'iS the chemical potential
(per particle). Hence we conclude

-1
T , ( 125)

(125) can now be rewritten in terms of the fA's and reveals

itself as the law of mass action.
For the thermodynamics of mixtures see Ehlers (1969), and
for applications of the preceding theory to cosmology see Ehlers
 - 371 -

Ehlers

and Sachs (1968).

Let us now investigate which restrictions are imposed on
the parameters 04., J and on the mean velocity u a by the requi-
rement that there is global equilibrium, i. e., that there is local
equilibrium at each event of a region D C X. According to the
theorem above, the functions (114) must then obey Lionville IS
equation; i. e. LA ( O(A + }apa ) = 0 in D. This equation is ea-
sily evaluated (see, e. g., Ehlers (1969) and leads to the
theorem. Global equilibrium requires that
(a) J a is a conformal Killing vector and, if at least one comp~
nent of the gas consists of particles with positive rest masses, a
Killing vector, and
b
(b) the electric field strength E: = F u is related to T and
a ab
0( by

T dot = e E. (126)

For a gas containing (also) ordinary particles (m> 0), equilibrium

requires a stationary spacetime. Defining in such a spacetime a
scalar gravitation;]] potential IT in terms of the Killingvector
ra.
~ = T oj a by e 2U = 't:'
~ -
2 we obtain Tolman IS law

T
U o
e (127)
T
 - 372 -

Ehlers

and if E = 0, then Ol = const., so that r.... depends on the p~

tential like the temperature. (For the general evaluation of (126)

see Ehlers (1959).)
It is possible to characterize the global equilibrium solutions

in a given, stationary spacetime by means of a variational princi-

Ele in which S is maximised under certain constraints, see MarIe
(1969) pp.107. For examples of equilibrium solutions, see Cherni-
kov (1964).
By means of (42) and (114) it can be verified that Planck's
distribution law results for rr = 2, OC r= 0, as it should be; OJ.t °
results from the relations (115), since there are always some pre-
cesses which change the photon number but not the numbers of the
other particles involved (ex.: e-e collisions).
A gas is said to be nondegenerate if the + I-term can be ne-
glected without serious error, so that (112) holds. Otherwise, it
is called degenerate.
One consequence of the last theorem is that a gas with m> °
cannot maintain an equilibrium distribution if it expands isotropical-
~ in contrast to an (m = OJ-gas (photons, nel1ltrinos). A physical re~

son for this deviation from the nonrelativistic behaviour of a (m 0)

gas will be given in the last section.
 - 3n -

Ehlers

Since the thermostatic functions of a relativistic gas are ex-

plicitely known (cf. eqs. (118)-(121) ) one can compute, e. g., the
velocity of sound in such a gas, and one can check the validi-
ty of Weyl's condition for shock waves. For a Boltzmann gas
with m 0 this has been done in detail by Synge (1957), with the
result that the sound velocity increases monotonically with the
c
temperature and approaches the limit as T.--. 00
{3'
(the value for a photon gas); shock speeds are always less than
c. Shock waves in a gas of Fermions or Bosons have been in-
vestigate by israel (1960).

11. Irreversible processes in small deviations from equilibrium;

hydrodynamics.

Whereas the equilibrium solutions of the Boltzmann equation

can be written down exactly, there is not much hope to find rigo-
rous solutions describing irreversible (Sa
;a
> 0) processes-in fact
no such (relativistic) solution is known at present. In physics, however,
one is mostly interest in non-equilibrium situations. Therefore,
in order to proceed one has to resort to approximations. We shall
briefly describe such approximation methods in this sectivil, and
refer to research papers for details. Our main goal here will be
to indicate how one may obtain from kinetic theory a complete sy-
stem of equations for thermo-hydrodynamics which is sufficiently
 - :17 4 -

Ehlers

general to include heterogeneous systems in which transport pro-

cesses and reactions take place, by applying suitable approxima -
tions to the Boltzmann equation. Partly our exposition will be a
program rather than the exposition of a completed theory. For
simplicity I shall consider here only neutral fluids, thus in the se
quel "e A = Ja = F a b = 0 " . Also, we shall only consider proces -
ses close to equilibrium, whicl: will (for most of the sequel) mean
states which are infinitesimal perturbations (first order variations)
away from local equilibrium.
Two distributions fA' ~ will describe nearly the same ma-
crostate of a gas if their moments in p-space are everywhere near
ly equal. This will be the case if fA = fA (1 + f ~"') provided ~A
is a. e. bounded on !'vIA and the numerical "perturbation" para-
meter Eo is small. With this motivation, we shall now consider a
one-parameter family fA ( £) of states which is, for € = 0, in

local equilibrium, i. e., is such that for € 0 the f 's have the

,a'
=
A
form (114), with unspecified spacetime fields OiA' and we

s ha 11 deno t e by l th e varla
fA ' t'lOns --~
dfA I~ =0. Notl'ce that the "10

cal equilibrium functions" O(A' J a

=T
lJa
are independent of £.
For "small" E , the moments computed by ,means of the "pertur~
ed distribution functions" fA(O) + E. fl will be considered to be the
A
macroscopic variables describing a "state close to equilibrium".
 - 375 -

Ehlers

It is clear that the perturbed macroscip variables will satisfy

the conservation laws

0, (128)

and similar ones, if we impose additional "scalar" conservation

laws like b-conservation. Also, we shall have the "Clausius ine-
quality"

~: = ( 129)

Again we can write the decomposition (65) for the total, perturbed
ab . a a I ab
tensor T , wIth f = rIO) +
I
£f I
,p = p(O) + Eo p, q = € (q ) • 1C =
= E ( rc ab ) I, because of (117) for £ = o.

Similarly,

( 130)

and

a a a ( 131)
S =~u +'6.
 - :376 -

Ehlers

with n A = n A (0) + f n A, i/ f. (i }\a)', ~ = ~O) + £~', ~/ =

E(~a)" from (115) and (116) for £=0.

It is a straightforward matter to derive from (128) the e-

nergy balance eguation

where the kinematical quantities {7, O"ab' Ua' Wab are defined
by

u =W +<S" 1
+- -& h -uu.
a;b ab ab 3 ab a b'
(133)
a
cr}=w
ab (ab)
=3"[ab] 6"
a
= 0

and are interpreted as the rate of rotation (Wab), rate of shear

( crab)' rate of expansion (-&), and 4 -acceleration (u a) of the
flow given by u a (see, e. g., Ehlers (1961), Synge (UJ60) ).
Here and henceforth we write

a
)' : ) ; u . ( 134)
;a

Also, one obtains the momentum balance equation

 - 377 -

Ehlers

(r+p) ua + hba (qb + p • b +T(cb;c ) + (Co\) + t3' ) qb+ -.! ~q = 0 (135)

ab ab 3 a
(h ab has been defined in (6"6).)
Let us now assume that there are Q conserved scalar quan-
tities. like b. which we call cq A' where 1 ~ q ~ Q ~ Nand
where N is the number of species A of particles; the c qA
are given. constant "gharges". Then the reactions in the system
are restricted by
(136)

We assume the Q "vectors" (c •...• c N) to be linearly inde-

q qt
pendent. and denote by (I' ••.•• 1' N)' l~p~R: = N - Q a
Pt (") p
basis in the orthogonal space. The vectors (1', •... I' N) of the
latter can be interpreted as (chemical or nuclear. e. g.) reaction
coefficients. as is seen from the equations

Na = ~v I' (137)
A;a L p pA
P
which express the general solution of (136) in terms of the con-
stants r pA and the reaction rates vp' giving the spacetime den-
sities of reactions of type p.
From (130). (137) we obtain the particle balance equations

(t) I;e .• LCqA r pA 0 for l"q~Q. l~p~R.

A
 - 378 -

Ehlers

(138)

Similarly, we rewrite (129), using (131), as

~ + :S{}' + -'ja 'a =

, ~~o. (139)

To proceed further we vary the expression (97) for the entro-

py current density Sa; because of

hog(.l. ± 1)-1
sf

and (114) we obtain

T \\' (r- IF-AnA) ( 140)

A

and

at a' ~ "" .a'

T~ = (q - £..fA lA ), (141)
A

where the fA are the chemical potentials defined in (125).

If we combine (140) with the thermostatic Gibbs relation

( 142)
 - 379 -

Ehlers

which results from (124) by summing over the species A,

and which holds for the unperturbed equilibrium functions (on

the manifold f s, n
j , .•• ,n A)} of equilibrium states), we get
the rather remarkable
Lemma 10. The perturbed thermodynamic variables f' s, n A
satisfy

( 143)

where F is the thermostatic potential of the system (as detel

mined from the exact equilibrium relations of section 10),
It is, therefore, "reasonable" to use, for near-equilibrium
processes, the ordinary Gibbs equation of state for the perturb-
ed variables, neglecting the error term in (143), as we shall

henceforth.
Also, we rewrite (141) for the perturbed variables:

( 144)

We also recall that, from (122), the thermostatic pressure

Po associated with the perturbed state is

Po Ts -f+ 2:fA A; n (145)

A
 - :380 -

Ehlers

there is no reason why p should equal the total kinetic pressu-

o
re p in (65).
We are now ready to derive an explicit expression for the
entropy production density, ~, in terms of appropriate thermody-
namic and hydrodynamic quantities. Compute ~ from (142) for
the perturbed state, which is permissible because of Lemma 10;
insert f from (132), nA from (138), and rearrange terms,
using (445), (144) and the definition

( 146)

for the volume viscosity 1C , to obtain the entropy inequality,

- T~ = 1'Cab e-ab +'Jt~-+~a_ 4fA i~] [ (log T), a + Ua ] +

( 147)

+LiA
a
( rA,a + fAUa) +~Ivp2:fArAp} ~ O.
A P A

This expression has the usual form known from ordinary irre-
versible thermodynamics (see, e. g., de Groot and Mazur (1962) );
in relativity, it has also been worked Qut on the basis of pheno-
menological assumptions by several authors (see, e. g. Stiickel-
berg and Wanders (1953), Kluitenberg, de Groot and Mazur (1953),
Kluitenbelfg and de Groot (1954), (1955).
 - 381 -

Ehlers

We wanted to show that (147) and the previous formulae

follow, in the sense we have specified, from kinetic theory,
just as in the non-relativistic case; this does not seem to have
been pointed out before with the generality we have retained he-
re, The crucial fact is tliat equations (143) and (144) follow
from the kinetic expression (97) for the entropy current,
The expression - T~ as given by (147) is bilinear in
"fluxes" ')tab,1("" and "forces" <;rab, -e-"" ,
We have shown earlier that the "fluxes" vanish at an event x if
there is local equilibrium at x, and that the IIforces" vanish in
a region if there is (global) equilibrium in that region, Hence,
one is driven to conjecture that, in a near-equilibrium process, the
fluxes (which are "caused" by the forces) depend linearly and ho-
mogeneously on the forces, with coefficients depending on the ther-
mostatic variables s, n A' This assertion is indeed used as an as-
sumption in phenomenological approaches, and leads to (more or
less) well-known relativistic linear transport and reaction eguations
for the shear viscosity 1'Ca b' the volume viscosity 1t, the heat
flow wa : = qa ~ fA ~
i , the diffusion currents i Aa, and the
reaction rates v, The corresponding matrix which transforms
p ----
( 0ab'.{t",,) into (1'tab , TC",,) must be positive-semi-defini-
te because of (147), If one requires, as is natural for a fluid,
that this matrix is invariant with respect to rotations ( in the 3-
 - 382 -

Ehlers

tangent-space orthogonal to u a ), the matrix reduces; and one

obtains a further simplification by assuming Onsager-Casimir
symmetry. All this follows strictly the standard theory.
However, we should not make these assumptions, but de-
ri ve them from kinetic theory. This has not yet been done in
the generality maintained here, but it will undoubtedly be done
soon ( 1). Such a derivation will supposedly) give not only the
form of the transport and reaction equations, but will also pro-
vide formulae for the transport and reaction coefficients in terms
of thermostatic variables and cross sections.
Two classical methods for doing this offer themselves; the
Chapman Enskog method, and the Grad method of moments.
Both these methods have, in fact, been adapted to relativity; the
former by Israel (1963) and, in a mathematically more com-
plete form, by MarIe (1969). (Israel, however, gives more de-
tailed results, particularly for a special type of "Maxwellian"
gas.) The method of moments has been taken over into relativi-
ty by Chernikov (1964) and in a more geometrical (and also
analytically more powerful) manner by MarIe (see Marle (1966),
(1969) ) and, independently, by Anderson and Stewart (see Stewart
(1969), Anderson (1970). Mathematically, MarIe's tre.atment is the
most complete one as regards the discussion of the "relativistic
Hermite-Grad polynomials", whereas Anderson and Stewart have

(') (Note added in proof) See a forthcoming paper by J. M. Stewart,

to appear in Lecture Notes in Physics, Springer- Verlag.
 - 383 -

Ehlers

gone further towards physical applications (transport coefficients

from cross sections.) In all of this work, the gas is a simple
Boltzmannian one; in that case, both methods give the transport
equations for T(b' 1C and q expected on the basis of (147). In
a a
particular. Israel (1963) and Anderson and Stewart (1969,1970)
both emphasized that a relativistic gas has (in general) a posi-
tive bulk viscosity, in contrast to a non-relativistic gas. The bulk
viscosity vanishes both in the nonrelati vistic ( T - 0 ) and in the

ultrarelativistic (T-OOI limit. This result "explains" the differe~

ce between m = D)-gases and (m> D)-gases with respect to proper-
ty a) ot the theorem in section 10: A gas of the latter type behaves
irreversibly if expanding isotropic ally, because of the term 1'(-&
in (147); a photon gas, however, behaves reversibly, since Ta =0
a
implies that 1t = 0 always.
For more details concerning the transition from kinetic theory
to thermo-hydrodynamics within the framework of relativity we refer
to Chernikov (1964), and to the papers cited above.
a
The roles of temperature T, entropy S (or s) and of the main
theorems of thermodynamics are completely clear within the frame-
work of relativistic kinetic theory; there is no room for assumptions.
(Of course, this changes if one wants to leave the domain of appli -
cability which we have delineated above.) In particular, integration
of (129) over a section of a world tube of streamlines, bounded by
two space like cross sections Land,"" , gives with the help of (131)
i L....l
 - 334 -

Ehlers

a
w
and ~a = (~(l44) ):
T

S[ 2:t] - S [L:J ~ J a
T- 1 w <3"a ( 143)

A
where./\. is that part of the boundary of the world tube which lies
between ~ , and
~1
L: t ,and where'" is assumed to be later
~2
than
L . This inequality is a precise version of the somewhat vague
ass;rtion ~S ~ J~ which has first been postulated in general
relativistic thermodynamics by Tolman (1934). In a similar fashion
one can derive other "global" thermodynamical laws for moving, fi-
nite systems enclosed in containers (timelike cylinders in X) from
the basic differential relations discussed here; again, there is no
ambiguity. (For another example of such a derivation, see Starusz
kiewicz (1966).)
Last - but not least - I would like to mention that the long-discu~

sed paradox concerning the acausal nature of temperature propagatlOn

(mathematically: the parabolic character of the corresponding system
of equations) has been resolved by the observation that the general,
"anormal" solutions resulting from the method of moments obey hy -
perbolic equations with non- spacelike characteristics (Stewart 1969),
and only the special, so-called "normal" solutions give rise to the p~
radox, which is, therefore, due to an inadequate approximation.
 - 385 -

Ehlers
REFERENCES

A. References about mathematical tools

R. ABRAHAM and J. E. MARSDEN, Foundations of Mechanics,

Benjamin 1965
Y. CHOQUET-BRUHAT, Geometrie differentielle et systeme
exterieurs, Dunod 1968
H. FLANDERS, Differential forms with applications to the phX
sical sciences, Acad. Press 1962
N. J. HICKS, Notes on differential geometry, van Norstrand
1965
A. LICHNEROWICZ, Theories relativistes de la gravitation
et de 1'electromagnetism, Masson 1955
A. LICHNEROWICZ, Linear algebra and analysis, Holden-Day
1967
M. SPIVAK, Calculus on manifolds, Benjamin 1965

B. References on or related to relativistic kinetic theory

( The following list is not meant to be complete, but includes

only papers which are closely related to the preceding lectures.)
J. L. ANDERSON, in: Relativity; M. Carmeli, S. J. Fickler,
L. Witten (Ed.), Plenum Pro 1970, p. 109
W. L. AMES and K. S. THORNE, Ap. J. 151,659 (1968)
J. N. BAHCALL, Phys. Rev. 136, B 1164
R. BEREZDIVIN and R. K. SACHS, in: Relativity; M. Carmeli,
S. J. FICKLER and L. WITTEN (Ed.), Plenum Press 1970,
p. 125
 - 386 -

Ehlers

K. BICHTELER, Beitrage zur relativistischen kinetischen

Gastheorie,_ Dissertation, Hamburg 1965
K. BICHTELER, ·Z. Physik 182 , 521 (1965)
" , Commun. math. phys. !, 352 (1967)
R. H. BOYER, Amer. Journ. of Physics ~, 910 (1965)
W. BRENIG and R. HAAG, General Quantum Theory of Col-
lision Processes, in: Quantum Scattering Theory,
M. Ross (Ed.), Bloomington 1963
N. A. CHERNIKOV, Acta phys. Polon. 23, .629 (1963)
"" " " " ~, 1069 (1964)
"" " " " ~, 465 (1964)
"" Physics Letters ~, nO.,2, 115 (1963)
C. ECKART, Phys. Rev. 58 919 (1940)
R. J. EDEN, Lectures on High Energy Physics, University of
Maryland Technical Report, 1966
.J. EHLERS, Akad. Mainz Abh., Math. Natur. Kl., Jahrg 1961,
791
" " , General Relativity and kinetic theory, in Rend. d.
Scuola Internaz. di Fisica "Enrico Fermi", Corso 47
(1969)
J. EHLERS, P. GEREN and R. K. SACHS, J. Math. Phys . .!!.' 1344
(1968)
J. EHLERS and R. K. SACHS, Kinetic theory and Cosmology,
Brandeis Summer Institute in Theoretical Physics, 1968
E. D. FACKERELL, Ph. D. thesis, Univ. of Sidney 1966,
 - 387 -

Ehlers

E. D. FACKERELL, Ap. J. ~, 643 (1968)

" " , Proc. Astron. Soc. Australia 1., 86 (1968)

s. R. de GROOT and P. MAZUR, Non-equilibrium thermodyna-
mics, North-Holland, 1962
R. HAKIM, Phys. Rev., 173, 1235 (1968)
J. IPSER and K. S. THORNE, Ap . .J. ~, 251 (1968)
W. ISRAEL, Proc. Roy. Soc. 259, 129 (1960)

" " , Journ. Math. Phys. i, 1163 (1963)

L. P. KADANOFF and S. BA YM, Quantum Statistical Mechanics,
Benjamin 1962
G. A. KLUITENBERG, S. R. de GROOT and P. MAZUR, Physica~,

689 (1953)
" , Physica ~,
" " "
1079 (1953)
G. A. KLUITENBERG and S. R. de GROOT, Physica 20, 199 (1954)
"" " " 21, 148 (1955)
" " " " ~1, 169 (1955)
P. T. LANDSBERG and J. DUNNING-DAVIES, in: Statistical Mecha-
nics of Equilibrium and Non-Equilibrium,
J. MEIXNER (Ed.), North-Holland 1965, p. 36
R. L. LIBOFF, Introduction to the theory of Kinetic Equations,
Wiley 1969
A. LICHNEROWICZ and R. MARROT, C. R. Acd. Sc. Paris 210,
759 (1940)
R. W. LINDQUIST, Ann. of Physics ~,487 (1966)
J. LOWRY, Ph. D. Thesis, Univ. of Texas at Austin, 1970
 - 38t! -

Ehlers

C. MARLE, Ann. Inst. Henri Poincare A .!Q, 67 (1969)

" "" " " A.!Q, 127 (1969)
11 ,C. R. Acad. Se. Paris 263, A, 485 (1966)

c. W. MISNER, Ap. J. ~ 431 (1968)

R. PEIERLS, M. N. 96, 780 (1936)
PHAM MAU QUAN, C. R. Acad. Paris 263, A, 106 (1966)
R. K. SACHS and A. M. WOLFE, Ap. J. 147,73 (1967)
A. STARUSZKIEWICZ, Nuovo Cimento ~,ser. 10, p. 684 (1966)
J. M. STEWART, Ph. -D. thesis, Cambridge (Engl.), 1969
E. C. G. STUECKELBERG and G. WANDERS, Helv. Phys. Acta 26,
307 (1953)
J. L. SYNGE, Relativity: the special theory, North.- Holland 1956
" " , The relativistic gas, "" 1957
" " , Relativity: the general theory," "1960
A. H. TAUB, Phys. Rev. 74, 328 (1948)
G. E. TAUBER and J. W. WEINBERG, Phys. Rev. ~,1342 (1961)
G. E. UHLENBECK and G. V. FORD, Lectures in Statistical Mecha-
nics, Ann. Math. Soc., Providence 1963
A. G. WALKER, Proc. Edinb. Math. Soc. i, 238 (1936)
H. WEYL, Math. Ann. !!.' 441 (1911)
R. A. MATZNER, Ap. J. 157, 1085 (1969)
 CENTRO INTERNAZIONALE MATEMATICO ESTIVO

(C. 1. M. E. )

ABSTRACT

MINKOWSKI SPACES AS FIBRE BUNDLES

K. B MARATHE

Corso tenuto a Bressanone dal 7 al 16 Giugno 1970

 ABSTRACT

MINKOWSKI SPACES AS FIBRE BUNDLES

K. B. Marathe
University of Rochester

A new definition is proposed for representation spaces

of Special Relativistic physical events. These spaces - call-
ed here Minkowski Spaces - are given a structure of a fi-
bre bundle over a four dimensional manifold. The Lorentz
group appears as the structure group of this fibre bundle.
It is shown that there is a unique torsion- free connection

in this bundle which induces the usual pseudo-Riemannian

structure on the base manifold.
 - 392 -

Marathe

MINKOWSKI SPACES AS FIBRE BUNDLES

K. B. Marathe
University of Rochester

1. Introduction

The study of Minkowski spaces arose early in this century

in connection with the special theory of Relativity. The concept
of four- dimensional space as a setting for physical phenomena
was introduced by Hermann Minkowski. However, we do not find
many investigations related to the discussion of the structure of
these spaces from the mathematicians viewpoint. The questions
of topology and differential structure if regarded asaa four-dimen
sional manifold have not received due consideration.

The Lorentz group plays fundamental role in all considerations

related to the special Theory of Relativity. It is then natural to ex
pect that this group should arise in a characteristic manner in the
consideration of the appropriate mathematical structure taken as a
setting for the physical phenomena in the domaine of the special
Theory of Relativity. However, it is known that regarding a physi-
cal event as a point in IR 4 or in a suitable four- dimensional ma-
 - 393 -

Marathe

nifold does not lead to the Lorentz group (Ref. 1).

To overcome this difficulty alternative topologies have

been proposed for Minkowski Space by Zeeman (Ref. 1) and
Celnik (Ref. 2). The topology MF proposed by Zeeman for
fR 4 has among other properties the property that the
Lorentz group is the group of homeomorphisms of MF.
Though this space has some interesting properties it is to-
pologically quite complicated. In particular it is not locally
compact and therefore standard methods of calculus cannot be
employed in it.

In this paper we propose for the space of physical events

(Special Relativistic) the structure of a fibre bundle over a four
dimensional manifold M. The new definition is equivalent to sal
ing that a physical event is characterized not just by a point P
in the manifold but rather by the point P and a fixed coordina·
te chart at P. This coordinate chart corresponds to a unique ~

near frame in the bundle of linear frames Lover M. The po-

int P together with this frame give the point in the bundle of
linear frames over M, which we shall call a representation of
the physical event under consideration. Which points of the bun-
dle L are to be regarded as representing physical events? The
answer is provided by the following theorem which is the main -
 - 394 -

lVIarathe

stay of the proposed definition.

Theorem 1: Let lVI be an arbitrary four-dimensional mani-

fold, L the bundle of linear frames over lVI, G the four-
dimensional general linear group, H the Lorentz group and
(5'" a cross- section of the associated bundle
E (lVI, G/H, G, L) then

1. There exists a unique (depending on tr) reduced subbundle

P (lVI, H) of L w::'th structure group H,
2. There exists a unique torsion-free linear connection in P
which makes lVI into a pseudo-Riemannian space of signatu-
re ---t,
3. The holonomy group of this connection is a subgroup of H,
and there exists a connection whose holonomy group is H.

We call P (lVI, H) a representation space for physical events or

in the standard terminology a lVIinkowski Space. With this defini
tion two representatives correspond to the same physical event
if and only if they belong to the same fibre of P. These two
points of P are then related by a Lorentz transformation.

2. Geometrical Preliminaries

By a differentiable four- dimensional manifold lVI of class

 - 395 -

l\1arathe

c" we mean a Hausdorff, connected, locally Euclidean topo-

logical space with a fixed four-dimensional C r atlas. r need
not be infinite, but in what follows we asSl me it to be large
enough to ensure the smoothness of the operations involved.
We denote the tangent space at x E M by T (M). A linear
x
frame I at xEM
x
tangent vectors in the tangent space T (M). We define the bun
x
dIe of linear frames L(M, G, 1'(), written as L, as follows. As
a set it contains all the linear frames I at all pOints of M.
The group G acts freely on L by the following action. If

g E G and I E L then gl is the frame mEL given by

x x x

m
Y = gl' X
i m where

m
mx = (Y I , Y2, Y3, Y4 ), and g = (gi ), the summation conven-
tion being used.

The action of G defines an equivalence relation on L in

a natural manner. Denoting the relation by 'V we have I....,m if
there exists g(G much that m = gl. Clearly 1- m if and
x y
only if x = y. Under this equi valence relation the factor space
L/G can be identified with M.
Let rc: L---" M be the natural projection given by
 - 396 -

Marathe

n(l)=x
x

A differentiable structure on L is introduced by making

,(1 (U) diffeomorphic with U XG where x E U eM and
k
( U, d ) is a coordinate chart at x. Let x k = 1, 2, 3, 4 be
coordmate functions. Then -~ k = 1, 2, 3,4 form a basis
~x
for T (M). If 1 = (X.) then with respect to this basis
x x 1

k k
In L then (x , x.) can be taken as local coordinates.
1

We regard H, the Lorentz group, as a subgroup of G.

The bundle E with fibre G/H (considered as a right coset
space) associated with L is constructed as follows: Consider
the action of G on LX G/H as a given by

a.oJ) = (a.l,f.a-1)( LXG/H for aEG and (l,S) € LXG/H.

The quotient space 01 LX G/H under this action of G is

denoted by E (M, G/H, G, L), written as E. The map
 - 397 -

Marathe

and a differential structure is introduced in E in a natural

manner by using 1't E . (Ref. 3)
The surjective map (l,~ )-+~'l of LXG/H ---+ L/H fac-
tors through E and allows us to identify E with L/H. Co!!.
sequently L can be regarded as a fibre bundle over E with
structure group H. Let 1:': L---+ E = L/H be the natural
pro jection.

3. Proof of Theorem 1
The proof is divided into two lemmas.

Lemma 1: There exists a unique reduced subblIDdle P of L

with the Lorentz group H as its structure group.
Proof: Define Pc: L by

p = 't- 1 (crM) ---(0)

For x E M, G'(x) E E = L/H and Since -r is onto by defini-

tion, 1:'-1 [G' (x)] is non-empty. Therefore, for every x ~ M
there exists 1 E peL such that TO) = e"(x). Now if
'to) = "((m), then 1, mare H-related i-e. There exists
hE H such that m = hI and conversely. Thus the set of all H-re
lated frames in a fibre of L form a fibre of P. The verifi -
cation that P is a principle fibre subbundle of L is now strai-
 - 398 -

lVIarathe

ght forward.

Lemma 2: There exists a unique torsion-free connection in

the bundle P which makes lVI into a pseudo-Riemannian
space with fundamental tensor of signature - - -+ ..

Proof: We first define a pseudo-inner product in 1R. 4

equipped with the fixed basis e 1,e 2,e 3,e 4 where
e 1 = (1, 0, 0, 0), e 2 = (0,1, 0, 0), e 3 = (0, 0,1, 0), e 4 = (0, 0, 0,1)

by (x,y) =-xIYl-x2Y2-x3Y3+x4Y4 (1)

The product defined by (1) is invariant under H.

i. e. h EH implies <hx, hy) = (x, y) ---(2)

Now regarding each 1 E Pc L as a linear isomorphism of

1R4 onto T (lVI)
x
we define a bilinear form on T (lVI)
x x
by
g (X, Y) ---(3)

where X, YET (M).

x

Definition of P and (2) show that definition (3) is indepen-

dent of choice of 1 E P. (1) shows that the form g is sym-
x
metric. It then can be identified with a symmetric covariant
 - 399 -

Marathe

tensor of order 2. If g.. are the components of g with

IJ k
respect to the coordinate system (x ) then the unique torsion-
free connection r in P is defined by the usual relations

J gjk
') xh ) - (4)

ij
where are the components of the dual tensor g'" to g.
r
g
is the unique torsion-free metric connection for the pseudo-
Riemannian structure on M defined by the tensor g. The si-
gnature -.-.-.+ of g follows from its definition and (1).

The theorem of Hano and Ozeki (Ref. 4) states that the struc
ture group G of L can be reduced to a subgroup H if and
only if there exists a connection in L whose holonomy group
is H.
This last result and lemma 1 and lemma 2 constitute the
proof of theorem 1.

4. Conclusion

In view of the results of theorem 1, it seems reasonable to

define P as a representation space for Special Relativistic events.
Each event is assigned an element of P. The assignment is not
unique as is to be expected.
 - 400 -

Marathe

Two points (x, 1) and (y, m) in P correspond to the

x y
same event if and only if x =y and there exists h E H the
Lorentz group such that M = hI, Thus the Lorentz in va
x x
riance of Special Relativistic Laws is built into the definition
of p,

The result proved here applies to manifolds which are not

necessarily flat. It may thus serve as a spring-board for a
corresponding definition in the case of General Relativistic e-
vents, However, no definite results in this direction have yet
been obtained,

Taking M = IR. 4 we can precisely characterize the distinct

representation spaces; they are in 1-1 correspondence with the
elements of G/H, This follows from the fact that for a E G/H
the map -va defined by

G": x - - - . (x, a) of M ---. E(M, G/H, G, L) is

a
a cross- section and 6" f G' if and only if a f b a, b E G/lL
a b
The bundle obtained by using
a
rs
in definition (0) is denoted by
P 6" a' This may be interpreted as saying that within the same P
the representatives of the same physical event are connected by
transformations belonging to the Lorentz group, but an arbitrary
non- singular transformation a E G will take representatives in P e"e
 - 401 -

Marathe

into P 5 a' e being the identity of G. If we define inertial

frame as a cross-section of L then we have a unique bundle
P. whose fibres are inertial frames. For example a frame
1
e = (e 1,e 2,e 3,e 4 ) satisfying

and g(e.,e.)=0
1 J

may be defined as an inertial frame at each pOint of

For this last case a different approach has also been pro-
posed by Eberlein (Ref. 5)
 - 402 -

.i\larathe

ACKNOWLEDGEMENT

My thanks are due to Professors Eberlein (Rochester),

Ste.n (Rochester), Ehlers (Texas) and Lichnerowicz (Col -
lege de France) for useful discussions.
 - 403 -

Marathe

REFERENCES

1. E. C. Zeeman, Topology, Vol. 6, pp. 161-170.

2. F. A. Celnik, Soviet Math. Dokl. Vol. 9 (1968)

No.5, pp.1151-52.

3. S. Kobayashi, K. Nomizu, Foundations of Differential

Geometry, Vol. 1, 1963, Interscience.

4. ,J. Hans, H.Ozeki, Nogoya. Math. J. 10 (1956), pp.97-100.

5. W. F. Eberlein, Bull. Am. Math. Soc. Vol. 71;No. 5, pp.731-736.

 CENTRO INTERNAZIONALE MATEMATICO ESTIVO
(C. 1. M. E. )

SUR LA PROPAGATION DE LA CHALEUR EN

I
RELATIVITE

G. BOILLAT

Corso tenuto a Bressanone dal 7 al 16 Giugno 1970

 /
SUR LA PROPAGATION DE LA CHALEUR EN RELATIVITE

G. Boillat

1.- Preliminaires.
Supposons qu'un champ ~(x~) veri fie un systeme d'equations aux derivees
partielles quasi lineaire du premier ordre (ou ramene a cet ordre),

(1) A"((~)~oI. "f(~,xf», ~,,0,1,2,3,

ou les A'" sont des matrices carrees NX N, xo une variable de temps,
x
i
(i = 1,2,3) des variables d'espace. A la traversee de la surface d'on-
de ~(x~) = ° Ie gradient du champ est discontinu et on a,
(2) [~1= 0, [~J.l'" ~ctb~, ~.t'do(l(, A<J,<f ... ~~ = 0,

d'ou, en introduisant la vitesse normale A de propagation dans la

direction Ii,
4
~ = - Cfo/\VI(\ n VI{ /\V~\
i
0, An = A n.1
Si AO est reguliere (det AO J 0) on peut supposer que AO = I. On dira
alors que Ie systeme (1) est hyperboligue (au sellS large) si les va-
leurs propres de la matrice A sont reelles (sans etre necessairement
n
distinctes) et s'iI existe un systeme complet de vectellrs propres,
c'est-a-dire N vecteurs propres lineairement independants (ce qui est
toujours vrai pour N valeurs propres distinctes). Dans ces conditions
une perturbation initiale arbitraire peut toujollrs se decoQPoser en
une somme de perturbations (~~) sur la base des vecteurs propres, per-
turbations qui se propagent ensuite de fagons differentes.
Lorsque Ie systeme n'est pas hyperbolique, distinguons trois cas.
a) D'apres (4) il est clair que la somme Sk des produits
k a k des racines du polynome caracteristique est proportionnelle a
l'inverse du determinant de AO. En particulier,
 - 408 -
Boillat
, Sl' IL A 1= °
1coer • II det A I -1

, SN 1 m ~ 1= Idet A 1 Idet AOI- l

n

11 en resul te que si, dans certaines conditions, det AO ~ 0, l'une

au moins des sommes Sk tendra vers l'infini et il en sera par conse-
quent de m~me de l'une au moins des vitesses de propagation, ce qui
n'est pas acceptable du point de vue physique. Ainsi on parle d'action
instantanee a distance quand on considere I' equation de Laplace 1 u= 0
comme limite de celIe de d'Alembert c-2Utt - A
~ u .. 0 lorsque c ~ ~ •
b) Si une valeur propre est complexe il appara!t un manque
de stabilite par rapport aux donnees initiales (1).
c) Si Ie nombre de vecteurs propres est inferieur a N, il est
possible de produire une perturbation ~B. telle qu'un choc (disconti-
nui te du champ lui-m!!me : [B. J ~ 0) se produise immediatement (2).
Nous verrons diverses circonstances se manifester suivant Ie choix
de l'equation de la chaleur. Bien entendu, nous chercherons a obtenir
un systeme hyperbolique.

2.- Tenseur d'impulsion-energie.

Pour un fluide parfait quand on neglige la conduction de la chaleur
ce tenseur s'ecrit (3),
(5) 1;o(r = (~+ p)uo(u~ - pg"-f'> = rfuol.uf'> _ pgo«('> ,

ou rest la densite pro pre de matiere, f l'index du fluide,

(6) f .. 1 + i,

i l'enthalpie specifique. La temperature 0 et l'entropie S sont intro-

duites par l'equation differentielle,

df .. Vdp + OdS, v .. l/r.

Nous ecrivons l'equation de conservation de la matiere,

et celIe du tenseur d'impulsion-energie,

(9)
ou q~ est Ie vecteur (transverse) courant de chaleur (4),
 - 409 -
Boillat

(10) quo(=O, (uuo(=l).

0( 0(
Compte tenu de (7),(8), les equations (9) s'ecrivent,
(ll) (rfu.t.. - qo<..)V""uf> - 't~\~rJ..p + ru'fV,,(q./r)\ = 0,

(12) rOu.... fd.J s = V q«" + ul\uo(V of'> = V qO< - q i·v ur' ,

'" 01.. ,,"" • J.. r at.
-tf' = l(f' - u"'ul" •

On ajoutera eventuellement les equations d'Einstein,

et on supposera que lion se donne les equations d'etat, par exemple,

p = p(g,r), s .. S(g,r).
II reste encore a definir Ie courant de chaleur et nous etudierons
plusieurs modeles dans les paragraphes qui suivent mais des a present
nous pouvons obtenir des equations aux perturbations en faisant dans
(8),(11),(12), Ie remplacement,

~cA. ' VrJ.. ~ «0< ~ .

Pour abreger nous poserons,

U .. u""!(.t.' Q .. q,(~a( , C .. t"t.8r·

Alors,
r C{a(~ua( + U br .. 0,

(14) (rfU - Q)~uf' - ,(a<fr", h + ul h/r

Nous introduirons encore en un point Ie rep ere profTe oartesien du

fluide,
(16) g .. diag(l, -1, -1, -1), un .. 1, ui = o.
a<r
Dans ce repere,
V°d.
, = 0, .. _ ~ij, qO = °
et d'apres (}) nouE aurons la correspondance,
\ ~-+
(18 ) U ~ - c\ , Q --')0 q. n ~ qn ' C ~ - 1.
 - 410 -
Boillat

,.- Equation de Fourier.

On peut transposer directement l'equation classique de Fourier en
relativite,
c(. vo(~'\
(19) q =- K0 vrQ.

Cependant on voit immediatement (Eq. 17) que dans Ie repere pro pre les
equations (19) ne contiennent auuune derivee par rapport au temps. On
se trouve donc dans Ie cas (a) du §1. Eerivons alors avec Eckart (5),
(20) qo(, = _ K'loif'(UrQ _ QU"\,u(J = - Ktf>\(Q'I~).
Nous evitons ainsi l'inconvenient pree~dent et nous avons,

(21) QU ~uo(, = 'f'f'~f'd Q.

Contra.ctant eette equation avec ~ on obtient en utilisant (13),

(22) U2(Er/r) + C(dQ/Q) a O.

II nous suffit ici de calculer les vitesses de prlpagation dans une

direct ion orthogonale au vecteur q" . Hous ferons done Q. = 0 e t en con.
traetar,t (14) avec ~r' en supposant U I 0, en se servant de (13),(15)
et en remarquant que (10),(21) entrainent,
u ~qo(. = _ q iuc<. = 0,
rA J..
nous aurons.

Avec (22), on en deduira,

2 4 2 2
9 SQU + U C(QVPQ - rQSr - f) - C Pr = O.
Sous les hypotheses usuelles (6),

p ') O.
r
Ie produit des racines de cette equation bicarree est negatif; il exis-
tera des racines imaginaires (§l,b).

Cependant il est interessant de calculer ces rarinec. Avec les equa~

tions deG gaz polytropiques(avec Po I 0 et G = 0; voir § 13) or. trouve

~ 4 _ ('(_ 2) ~2 _ ((_ 1) = O.

En dehorc de Ie solution imaginBire ~2 - 1, on a,

~2 y_ 1,
 - 411 -
Boilln
une valeur que nous retrouverons plus loin (§ll).

4.- Equation de la chaleur.

En 1948, C. Cattaneo (7) proposa un nouveau modele d'equation de la
}haleur qui fut redecouvert dix ans plus tard par P. Vernotte (8) et
mis recemment sous forme covariante par M. Krany~ (9). C'est ce modele
que nous etudierons apros y avoir fait quelques changements.
Nous chercherons une modification simple des equations (19),(20) sans
choisir a priori l'une ou l'autre,

(23) qd... =_'{rJ.f'(j)~,.,g+ rl\,u,..).

(C'est l'expression classique dans Ie repere propre s'il est repere
d'inertie.)
Ajou tons un terme T'"
au premier membre. Comme il n' y a aucune raison
pour que ce terme soit orthogonal a u( nous devonn recrire (23),

(24) Q.rJ.. + T"" = second membre,

avec
(25) Q~ = qu~ + q~ •
- Pour que Ie
Nous avons introduit une nouvelle variable de champ q.
champ soit hyperbolique il est necessaire que sa derivee par rapport
au temps figure dans les equations du champ, c'est-a-dire dans T~ puis-
qu'elle n'est pas ail1eurs. On volt tout de suite que l'on n'obtient
rien d'interessant en prenant T'" proportionnel a g<t d q ou
U1Vy(qU~). On est ainsi conduit a faire apparaitre la deCivee temporelle
de Q.~ c'est-a-dire a prendre,

T" =~lVrQ.( '" lv./ 1 Qri..)

(s1 i- = cte). Nous ecrirons (10)(11)
(26) Q~ + lVf"(hQ~) = - '(rA("(V 'd,..g + rlvyu t )·
On trouvera cette equation dans (9) a cela pres que nous ne supposons
pas, dorenavant, h et V constants.
Nous definirons,

(27) h = log Flh .,' ~/h v ,

h,
'j1 rf + t"" + q =e. + p + r + q
c A 'j1 U - Q,

et nous adjoindrons a (26) l'equation d'etat,

 - 412 -
Boillat

(28) F(Q,r,q,h) = O.
5.- Perturbations.
De (26) on deduit,
(29) - UOqoC. .. Ui'(qbh +~q) + Uqo(Dh +v'toll'f,Jg + U(q +r»)ucl.,
que lion porte dans (14) pour obtenir,

(30) Aou~ - -r~\fo(.(~p -ll'SQ) + Uqt'(fh + ~r/r) .. O.

Nous y ajoutons (13),(15) ainsi que les equations qui derivent des
equations d'etat,
(31) rll ~ug( + UOr
1,(
= O..
(32) rQU ~S = (~rJ. + UUol.)~qo( ,
(33) u ~uo(
rA
= 0' I A
q DUo( + U 'bqo{
rA
.. 0,

(34) op = Po dO + Pr ~ r, 'b S = Sg bQ + Sr ar,

(5) F a + Fr ~ r + Fq ~ q + lo'h ~ h = O.
Q Q

6.- La surface d'onde U .. O.

Cette surface se deplace a. la vitesse du fluide. 11 vient immediate-
ment,

et (si Q. ~ 0),
d ut' " 0, r, ~qGC. = u b'q.( = 0, F- dq + Fh ~ h " O.
iGC. g( q
II en resulte qu'il correspond trois vecteurs propres a cette vitesse;
Oq.( a deux degres de liberte et Oq (ou Oh si Fh etait nul) peut @tre
choisi arbitrairement.
1.- Determinant ce.racteristigue.
Nous supposons U I O. Contractons (30) avec ~r et utilisons (31),
(36) U(A - Q.)~r/r + C(dp -I>~O) - UQ. ~h " O.

(32) et (29) donnent ensuite,

(37) rQU2~S + 2U 2(q ~h + dq) + UQ. ~h + )..Icbo - U2(r+ q)(jr/r) .. O.
Enfin, ou1tipliant (30) par qr et tenant compte - d'apres (33) et
(29) - de,
(38) (U ;. 0),
 - 413 -
Boillat

on obtient,

q2 = _ q qO<
0(

Nous avons maintenant avec (35) un systeme lineaire et homo gene de

quatre equations pour les quatre inconnues 1, • (dO, 'dr/r, ~q, ~h).
Nous ecrivons que Ie determinant est nul,

FO rF F-q Fh
r

(PO -)))c rp C + U(A - Q) 0 - UQ

r
(40) " O.
(PO -,.»Q rp Q + q2U - A
2
q U - qA
r

JJa + rOSOU 2 ( r 2OSr -r- -)q U2 2U 2 (2qU + Q)U

8.- HYperbolicite. Existence de la surface d'onde A a O.

II est necessaire d'examiner d'un peu p~es cette question afin de
corriger une erreur que nous avons faite (10) et qui est d'ailleurs
sans oonsequence pour la Buite sinon que les conditions qui seront
imposees a F (28) ne Ie Beront plus par un motif d'hyperbolicite.
Le systeme des equations du champ comporte 9 variables independantesl
r, Q, u~(3 composantes seulement Bont independantes car u u~ " I). Q<.
,(
Nous devons done trouver N " 9 vecteurs propres pour que Ie systeme
80it hyperbolique. On pressent tout de suite que (40) est un pOlyn6me
de degre 5 (au plus) en ~ : les deuxieme et quatrieme lignes du deter-
minant sont de degre 2, la deuxieme de degre 1. On verifie que le de-
gre est bien egal a 5 (nous ferons le calcul plus loin).
1") 6i A • 0 ne verifie pas (40), a toute solution ~W ;. 0 corres-
pondra un vecteur 0 uo( donne par (0) et un vecteur ~ q0( donne par (29).
Nous aurons ainsi (certainement si toutes les racines sont distinctes
et non nulles - differentes de U • 0) 5 vecteurs propres.
D'autre part si ~ n'est pas une racine de (40) }, = 0 et ouo( •
• q~ • 0 a moins que A • O. Dans ce cas (cf. 29,31,33.38).
(41) - .l .. (q + ....I )~uG(, u ~uo(
0(
= (811\. }uo( • q ~ uO(
ri..
.. O.
 - 414 -
Boillat

Une des composantes de auP<. est arbitraire. A la solution A = 0 cor-

respond un vecteur propre. Le nombre total des vecteurs propres est
donc
3(U = 0) + 5(aol. de 40) + l(A = 0) = 9.
Le systeme est hyperbolique. (En supposant, bien entendu, que les va-
leurs proprea sont reellea, ce qui.reste a etablir.)
2°) Supposons maintenant que (40) soit Ie produit par A d'un poly-
nome P4 du quatrieme degre. Aux racines de ce polynome correspondront
4 vecteurs propres de la fagon qui vient d'@tre dite. Pour A = 0 on pour-
ra choisir arbitrairement une des composantes de O'W, les autres seront
determinees de fagon a satisfaire les equations (30),(35),(37) (nous
verrons que cela est possible). Le vecteur ~ uo( sera seulement soumis
aux trois conditions (31),(33),(38) I

(42) u~ ~u,( = 0, qQ(~/' = - U~r/r, qo( ~uo(= qh + ~q ;

une de ses composantes pourra encore 3tre prise arbitrairement. Ce qui
revient a dire qu'a la solution A = 0 correspondront 2 vecteurs pro-
pres dont Ie nombre total sera I

3(U = 0) + 4(P4 = 0) + 2(A = 0) = 9.

(On a suppose, cela va sans dire, que P4 ne s'annule ni avec U ni avec A.)
Le systeme sera encore hyperbolique.

9.- Conditions d'ex~ence de la solution double A = O.

Nous allons montrer que par un choix convenable de F il est pasible

d'obtenir une solution double A c 0 ce qui permettra comme on l'a vu
oi-dessus de simplifier Ie determinant caracteristique (40).
On tire de (30) avec A = 0,

(43) 1p = ).lbo, "h + Ur/r = 0,

soit, avec (35),

(44) (PO -].I )~o + rPr '$r/r = 0,

(45) FO ~O + (rFr - Fh )(3r/r) + Fq bq = o.

Supposons F~ f o. On peut alors (comme Pr f 0) a l'aide des equations

ci-dessus exp~imer ) h, ~r, dq proportionnellement a. ~ O. Le premier
membre de (37) devra 3tre identiquement nul si l'on y porte ces expres-
 - 415 _
Boillat

sions, compte tenu de 11ega1ite Q = ~U, ce qui est impossible en rai-

son de 1a presence du terme en iJ C. On do it donc avo ir
F-q = 0.
A10rs 07) donne 'Oq et (44),(45) 11equation aux derivees partielles
que doit satisfaire F(Q,r,h) a 0,

(47) a(F) = 0,
II est commode de prendre F sous 1a forme suivante reso1ue par rap-
port a. h,

(48) F • - re h + ~(Q,r), h = eh = p/r.

(47) devient,

(49) (v- PQ)~r + P~Q a 0,

clest-a.-dire, si lIon ecarte 1a solution p D cte (qui n'est d'ai11eurs
pas l'hy,othese qui conduit aux resu1tats les plus simples; voir §11),

(50) ~)p = V •
10.- Llouue excentionnelle A .. 0.
A= ° clest ~ U- Q = ° soit,dans Ie repere propre (cf. 18),
(51) ~ = - qn/l1
En ecrivant que ~ 2 ~ 1, ou ce qui revient au m~me ici que 1a vitesse
radiale ~ u~ - q~ est du genre temps (ou isotrope) on a 1 l inegalite,
(52) q2 ~ 'rI2 •
Cherchons maintenant sli! est possible que llonde soit exceptionne11e
(12)(13) crest-A-dire ne produise pas de chocs sur Ie front dlonde.
Pour qu 101
1 en soit ainsi on deVl~. avoir (14),

soit,
U~" + 11 q.(6u O< _ ~.( ~qo(. • O.
De (32),(42) on tire,

~.(Oq'" = U(rQ dS - Uv( ~qO<) = U(rQ bS + qh + ~q),

de sorte que 11equation ci-dessus slecrit avec (31),
 - 416 - Boillat

6(rf + t'"+ q) - 'i1'br/r - (r01S + q1h + ~q) = O.

Mais, d'apr~s (7),

r~f-rOh= ~p,
si bien que,

- r)r/r = 0,
~p + Sf'"
ou, puisque r
= I- e -h et compte tenu de ( 43,
)

hh +br = 0, jjbO +br ,,0,

c'est-a-dire, en vertu de,

(53) ~p=O
- qui derive de (48) et
~-
~)p " - h,

Ces equations (qui se deduisent l'une de l'autre avec (50)) ne sont

pas satisfaites par ~ = O. II en resulte que pour un tel choix
l'onde A = 0 produira des chocs. (Les discontinuites des derivees pre~
mieres de variables de champ, discontinuites qui se propagent a la
vitesse radiale ~ u~ - q~ , deviendront infinies au bout d'un temps
fini si l'onde est initialement acceleree (13).)
En revanche, supposons

(55) V = K(~),
une certaine fonction de p, alors (54) donne,

r = -K(p)O + funct(p).

Ainsi l'onde A = 0 ne produira pas de chocs (semblable en cela aux on-

des d'Alfven de la magnetohydrodynamique). On remarque que si l'on
prend simplement,

(56) P. = - K(p)O,

Ie second membre de l'equation de la chaleur est celui d'Eckart. Nous

recrirons done (26),

(57) Q..i. + iVr(pl/r) = - K(p)fl.rV~ (O'tt),

 - 417 -
Boillat

avec la condition qui decouIe de (50), (55),

11.- Expression expljcite de I'eguation caracteristigue.

Revenons au determinant (40) et cherchons l'expression de l'equation
du quatrieme degre apres mise en facteur de la solution A = O.
Avant de considerer Ie cas general notons une approximation particu-
lierement simple (11). Elle consiste a prendre h ( .. ¢/r) = ~ = cte.
On obtient alors facilement, pour un gaz polytropique (pY = p/r = RO),

(59) Q< + rlVfQr{ = - ~Rryo(rV~(or(.),

et dans Ie repere pro pre (outre les vi tesses A.. 0, ~ = - q/ II ),
(60) +

J.
(61) =+(Y_l)2,

ou )' est la constante des gaz polytropiques, egale au rapport des

chaleurs specifiques :)' • Cp/C y • On remarque que (61) atteint la
valeur limi te pour t . 2 (vo ir §14).
Une combinaison lineaire de oolonnes (indiquee par les t@tes de colon-
nes) puis I'addition de la deuxieme ligne a la derniere mettent (40)
sous la forme,

FO a(F) + q(V- PO)Fq F-q Fh - qF-q

(PO - V)C ( I> - PO)UA 0 - UQ

(62) .. 0,
(PO - t> )Q 0 -A q2u
2 2U 2
rOSOU + POC (~ - PO)B 0

avec,

B .. r~)p u2 - 2UQ + r~)p C.

En realite, on n'obtient pas directement cette expression mais plu-

tOt (a la place du terme (V - PO)B),

ru 2 { rOPrSO + (V - PO)(rOSr + f»)- 2(V - PO)UQ + rprv r..

 - 418 -
Boillat

On met ~ - Pg (suppose maintenant non nUl) en facteur et on tient

compte de (49),

J> - Pg = - p)g/~r •

Le coefficient de r( 1> - Pg )U 2 devient,

rQ(Sr~g - SgPr)/Pg + f = rQf)s~r)p + f .. ()f(dr)p ,

en raison de la relation,

(64) dt = fdr + rOdS,

qui decoule immediatement de (7) (at rf = e+ pl.
Pour ce qui est du coefficient de r( V- Po)c, il slecrit,

- V p!Po .. )} ~)p .. ~)p~~)p .. ~)p ,

en utilisant (50). On a ainsi (63).
En vertu de (46),(47), Ie determinant se developpe facilemant at il
reste apres mise en facteur de A et un calcul de coefficients analogue
a celui que lIon vient de faire,

(65) P(X,Y) = a 2X2 + 2b 2XY + a

o
• 0,

X = - U2/C, y. QU/C,

a 2 = 2q2/p + if ~)r' 2b 2 = [#)rp - 2~)0' c2 .. - 2/p,

al = r(,*)p~)g - 'if lij)r' bl = 2(11)0 - ~)rp ,

Dans Ie repere pro pre (cf. 16),

X = ~2, y.. ~q •
n
On doit avoi;,
(66 ) o <. X { 1,

Non seulement la surface dlonde doit @tre du genre espace mais la vi-
tesse radiale doit encore @tre du genre temps. Ced implique,
 - 419 -
Boillat

soi t,
(67) 4(X - l)(XPX + YPy)P X + (q2X - y 2jX)Pi ~ 0,
pour toute solution de (65).

12.- Fluide incompressible.

Lorsque la conduction de la chaleur est negligee on sait que la vitesse
du son dans Ie repere propre est egale a. la racine carree de (dp!de)s
On est done amene (15) a. definir Ie fluide incompressible comme cor-
respondant au cas limite,

(68) ~)s = 1 ==.a,. ~ - p .. Cf(S).

lei nous procederons de maniere analogue. Nous chercherons une equa-

tion d'etat reliant la pression p a. deux variables thermodynamiques.
Toutefois la vitesse normale ~(qn) depend maintenant de la direction
de propagation ~ par l'intermedi~e de la quantite q . Nous supposerons
n
done que la vitesse limite est atteinte a. la frontiere de l'intervalle
de variation de qn: ~2(~ q) a 1. Nous ecrirons, par consequent,
2 2 t
p( 1, £,q) = 0, ~ = 1, q,. (q) ,
soit,

a'est-a.-dire, en posant,

(69) e- =«',p

~ ~)r + r~)Q~)~
L'equation d'etat ne doit pas dependre, bien entendu, de q ni de -q qui
figure dans ~ • II faut done,

(71)

Afin de satisfaire la derniere equation on peut faire deux hypotheses:

a) ou r = cte,
b) ou bien supposer que la vitesse de la lumiere est atteinte quand
l'onde A = 0 attaint aussi cette vitesse ce qui a lieu quand q2 ,.~2
(cf. 52). Dans ces conditions 'if +£q = 0 est verifiee pour l'une des
 - 420 -
Boillat

deux valeurs de £ et il suffi t de supposer £f.. 'f{~).

Venons-en a. une autre possibilite : A2(qn) = 1, qn€ 1- q, + q [
Puisque cet te valeur de ~ doit 3tre extremale : d ~ I dqn = O. En
eliminant Y entre P(l,Y) = 0 et Py(l,y) = 0, on trouve,
2
4c2 (a 2 + a l + ao ) = (2b 2 + bl ) ,

qui ne fournit aucune equation d'etat et renforce notre premiere fa-

gon de voir.

13.- Fluides polytropigues.

La comparaison de (68) et (71) nous invite a etudier Ie cas ou ~ ne
dependrait que de l'entropie,

(72) ~ = ~(S).
En utilisant l'identite thermodynamique,

nous pouvons recrire (58),

(l/v) ()vlOs) p = ~(s)/K(s),
soit,
(74) V = pi (p)/(S) , F(S) .. f~ dS.
Avec (73) et (7), on obtient,

(75) o = ~ p/ + G' ( s) , f .. 1 + i .. Pe F + G,

ou P et G sont des fonctions de p et S respectivement et ou la deriva-

tion est notee par un accent. D'apres (27),(48),(56),

(76) ~ .. r(f - 7 0) + q = r(G - 1 GI) + q.

En raison de l'arbitraire sur P on en deduit que
2 2
';1 ,. q ~ G = O. q ~)f

Alors,
(78) e+p = f/v • p/p' ~ ~ = e(p); f .. 70 .
Si nous suposons G' • 0 (G .. ete) nous avons, pour les chaleurs spe-
cifiques a pression et volume constants,
 - 421 -
Boillat

1 1. ~Q)
Cp Q 'OS p P (!)
= K
K
I + !
K'

(80)

Par integration de (79),

f~; EO Log f + F(S)

et Ie coefficient de la derivee au fil du courant du vecteur Q~ dans

llequation de la chaleur (57) slecrit encore,

(81) ~ = K(S)P' (p)exp J~;

Si nous supposons de plus que Cv nlest fonction que de S,

(82) - ~;~ .. cte .. t-1 ~ P '" a(p + po) (Y-l)/Y ,

1. _ 1. .. (¥_ I)! .
Cv Cp K

En particulier, on obtient les equations des fluides polytropiques en

prenant,

KTSY
iC.§l
= cte.. C
1
p

Pour les gaz, 1(,«2 et si lIon prend,

p
o
= 0, G .. 1,
on aura,
i=CQ,
p
cp .. rR/(l-l), pV=RQ.

14.- Le fluide incompressible comme fluide polytropigue.

rf '" e+ p .. f/V.
Dans llhypothese la moins restrictive (b,§12),

t -p P G -F
= pI - 2p + pIe .. lY()
1 S ,

ce qui entraine,
 - 422 -
Boillat

(86) P = a(p +p
o
)t, G = 0.

Ainsi Ie fluide incompressible correspond a '( = 2. Comme (16),

(87) ~ - p = 2po = cte,

~ - ;p '> 0....,
o p <p •

On se retrouve done dans Ie cas (a) du §12; la vitesse de la lumiere

est atteinte dans une direction parallele au vecteur courant de chaleur.

15.- Vitesses de propagation.

Avee (84), (65) donne, dans une direction orthogonale au veeteur <t,

Plagons-nous dans Ie cas des gaz (p = 0, G = 1) et supposons que lion

o
puiase negliger Ie terme en ~I (en admettant en particulier que ~ varie
peu). Alors,

(89) ~4 _ (y _ 1 + rf) ~ 2 + 'U .. 0,

If 'i1

Si Ie dernier terme est(generalement)petit par rapport au carre du

coefficient de ~ 2,

(90) ~2 ~'(-l+!!.
1 /1
IV
= ~ j (r - 1 + r;).
On voit que

(91) \(\12 --'-'

- , v_I,
, ~ 0,
\ 2 ----r
("2

lorsque q~ ()O (cr. (61). D'autre part quand q ~ 0, rf/11 =f

qui est equivalent aI dans 1 I approximation non relativiste; par con-
sequent,

La deuxieme valeur coincide sensiblement avec (60) (qn = 0, e ~ r).

C'est, toujours a la meme approximation, v 2jt
s
ou vest la vi-
s
tesse duson quand on neglige la conduction de In chaleur,
 - 423 -
Boillat

Passons au fluide incompressible. En negligeant encore ~, on cherche

les solutions de p*(X,Y) = 0 obtenu en remplaQant les coefficients de
(65) par,
(9}) lim ~'a2 ' etc •••
~, ~O

On s'apergoit alors que la vitesse limite (~= i 1) est atteinte dans

toutes les directions: P*(l,y) = O. En consequence, on obtient aisement
les deux autres vitesses (en exprimant par exemple la somme et Ie produit
des racines en fonctions des coefficients),

soit,
q ~2 +}q ~ _ C rO
n p
= 0, CprO = 2(p + P0 ).

On notera que la condition ~2 <1 dans toutes les directions entraine,

(95) q > }q + CprOo

On sait que l'eau peut @tre consideree comme un flu ide polytropique
avec Y = 7, p > O. Si jamais (88) peut @tre utilisee on se rend compte
o
en tout cas qu'on ne peut plus negliger Ie terme en~' (pour que
0<~2 (1).

Remerciements
Nous exprimons notre gratitude a MM. C. Cattaneo et A. Lichnerowicz.
Nos remerciements vont aussi a Mme Y. Choquet-Bruhat et M. J. Ehlers.

References

(1) G. BOILLAT, C. R. Acad. Sc. Paris, ~ A (1970), 217.

(2) G. BOILLAT, Ibid., 11}4.
()) A. LICHNEROWICZ, dans ce livre.
(4) PllAM MAU QUAl!, Ibid.
(5) C. ECKART, Phys. Rev., ~ (1940), 919.
(6) D. ter HAAR & H. WERGELAND, Elements of Thermodynamics, Addison-
Wesley (Reading, 1966).
 - 424 -
Boillat

(7) C. CAT'l'ANEO, Atti del Seminario matematico e fisico della Uni-

versita di Modena, ~ (1948), 1; c. R. Acad. Sc. Paris, £iI (1958), 431.
(8) P. VERNOTTE, C. R. Acad. Sc. Paris, ~ (1958), 3154.
(9) M. KRANY~, Nuovo Cimento, ~ B (1966), 51; ~B (1967), 48.
(lO)G. BOILLAT, Lett. Nuovo Cimento, l (1970), 521.
(ll)G. BOILLAT, Velocities of Heat propagation in Relativistic poly-
tropic Fluids, Ibid. (a paraitre en 1970).
(12)P. D. WL~, Ann. Math. Studies (Princeton), 2i (1954), 211; Comm.
Pure & Appl. Math., 1Q (1957), 537.
(1 3)G. BOILLAT, La propagation des ondes, Gauthier-Villars (Paris, 1965).
(14)G. BOILLAT, Journ. Math. Phys., 10 (1969), 452. Cf. Y. CHOQUET-
BRUI~T, Journ. Math. Pures & Appl., ~ (1969), 117.
(1 5)A. LICmmRO"IIICZ, Theories re1ativistes de la gravitation et de
l'electromagnetisme, Masson (Paris, 1955).
(16)c. CATTANEO, Lincei-Rend. Sc. fis. mat. e nat., ~ (1969), 698,
a propose de remplacer t? par ~ + p de fa90n a obtenir I' equa tion
d'etat ~ = cte au lieu de (8 7 ).
Editoriale Grafica - Roma