C. Ferrari ( E d.)
Dinamica dei gas rarefatti Lectures given at the Centro Internazionale Matematico Estivo (C.I.M.E.), held in Varenna (Como), Italy, August 21-29, 1964
C.I.M.E. Foundation c/o Dipartimento di Matematica “U. Dini” Viale Morgagni n. 67/a 50134 Firenze Italy
[email protected]
ISBN 978-3-642-11023-8 e-ISBN: 978-3-642-11024-5 DOI:10.1007/978-3-642-11024-5 Springer Heidelberg Dordrecht London New York
©Springer-Verlag Berlin Heidelberg 2011 Reprint of the 1st ed. C.I.M.E., Ed. Cremonese, Roma, 1965 With kind permission of C.I.M.E.
Printed on acid-free paper
Springer.com
CENTRO INTERNATIONALE MATEMATICO ESTIVO (C.I.M.E)
Reprint of the 1st ed.- Varenna, Italy, August 21-29, 1964
DINAMICA DEI GAS RAREFATTI
C. Ferrari:
Premessa al ciclo sulla “dinamica dei gas rarefatti” .................................................................
1
M. Z. v. Krzywoblocki: Some Mathematical aspects of rarefied gasdynamics as applied to hypersonics, reentry and magneto-gas-dynamics..............................
5
J. Kampé De Fériet:
La théorie de l’information et la mécanique statistique classique des systèmes en équilibre ............ 155
M. -Lunc:
Equations de transport .................................................. 223
I. Estermann:
Applications of molecular beams to problems in rarefied gas dynamics ................................................... 269
I. Estermann:
Experimental methods in rarefied gas dynamics ........... 305
S. Nocilla:
Sull’interazione tra flussi di molecole libere e superficie rigide ......................................................... 331
F. Sernagiotto:
“Solution of Rayleigh’s problem for the whole range of knudsen numbers” ......................................... 367
G. Tironi:
Linearized Rayleigh’s problem in magnetogasdynamics ................................................... 381
D. Graffi:
Alcuni richiami sulla ionosfera .................................... 407
C. Agostinelli:
Le equazioni delle onde d’urto in un gas rarefatto elettricamente conduttore soggetto a un campo magnetico .................................. 425
Premessa al CicIo sulla "Dinamica dei Gas Rarefatti" di CARLO FERRARI
Desidero porgere innanzi tutto il pili cordiale benvenuto ad ognuno dei pre senti, ed esprimere un particolare ringraziamento ai cari ed illustri Colleghi che nella Sessione C. I. M. E. che oggi si inizia, si sobbarcano alIa maggiore fatica: quella di dare in un cicIo di conferenze relativamente breve la pili larga, e nella stesso tempo la pili completa pos sibile trattazione dei problemi che si riferiscono alIa dinamica dei gas rarefatti; e senza dubbio, grazie alIa lore ben nota capacita e altissima competenza sull'argomento, allo studio del quale tanti contributi originali essi hanno dato, essi riusciranno nell'intento, che pure
e difficile
anche se l'assemblea degli ascoltatori
e per
molti effetti
eccezionale. I problemi della dinamica dei gas rarefatti hanno da tempo attirato l' attenzione di Matematici, di Fisici, di Chimici: basti ricordare i n.o-
mi di Knudsen, Smoluchowsky, Enkscog, Clausing; i tentativi fatti per ottenere soluzioni approssimate dell'equazione di Boltzmann; gli studi teorici e sperimentali sull' adsorbimento fisico e chimico di gas da parte di superfici solide ; Ie ricerche sui processi di diffusione attraverso a materiali porosi per la separazione degli isotopi. Ma Ie applicazioni recenti della meccanica dei fluidi hanno proposto nuovi problemi
0
hanno riproposto problemi vecchi sotto veste
nuova ; l'uso di satelliti artificiali e di razzi spaziali per investigare la struttura e Ie proprieta della ionosfera e del mezzo interplanetario hanno intensificato l'interesse nella studio degli effetti, che si producono in vicinanza di un corpo che si muove in un mezzo molto rarefatto. D'altra parte questa mezzo rarefatto non atomi neutri, rna
e un
e un aggregato di mole cole
e di
plasma, ossia un aggregato di mole cole, atomi,
1
- II -
C, Ferrari
ioni ed elettroni, In effetto nella parte inferiore della ionosfera, ad un' altezza di 200 Km, dalla superficie terrestre, il numero di densita delle particelle neutre
e da
2 a 5 x 10 10particelle per
cm~, con
un per.
corso libero medio di 8 x 10 cm,; mentre il numero di densita degli elettroni
0
degli iOl1i
e da
3 a 50 x 104/cm 3 , con un percorso libero
medio di 9 x 103 cm. A 300 Km. di altezza i numeri di densita delle particelle neutre e di quelle cariche diventano rispettivamente 3 x 10 9 e 1+2 x 106.ei lora percorsi liberi medi 105 cm.; 7 x 10 3 cm. Infine a 3000 Km. di altezza si puo presumere che Ie particelle neutre siano , numero d'1 uno per cm 3, mentre Ie particelle elettrizzate present!, m 1 en' medi sono rispettivamente sono 7 x 103/ cm3; ,1 percorsl'l'b
2 x 10
14
6
e 3 x 10 cm.
Con i gradi di rarefazione del mezzo quali risultano dai numeri
,.
ora detti, gli 'usuali metodi dell'aerodinamica non possono essere applicati per descrivere i fenomeni che sono prodotti nel mezzo stesso dal moto di un corpo;
e necessario
Ie particelle neutre, molecole
0
applicare la teoria
cinetica. Ora,
atomi, interagiscono solo colla super-
ficie dell'ostacolo ; Ie particelle cariche invece, non soltanto presentano questo tipo di interazione, rna sono anche influenzate nel loro moto dai campi elettrico e magnetico, ed il campo elettrico a sua volta
e
prodotto e dalle cariche spaziali, che si producono nel plasma per effetto della differenza di concentrazione degli ioni e degli elettroni causata dalla presenza del corpo, e dalla carica che questo assume. I problemi che sono connessi allo studio di queste interazioni possono variare notevol mente : nella parte inferiore della ionosfera, la velocita
del corpo
.Yo e generalmente molto maggiore della velocita termica degli ioni molto
P'u piccola cl
quella v
-e
~i'
e
degli elettroni. COSl, all'altezza di 200 Km,
2
- III -
C. Ferrari dette velocita sono rispettivamente dell'ordine: v v. = 1
DC
10 4 cm/sec.
i v = oc 10 7 cm/sec.
e
6 0( 10 cm/ sec; o e pertanto il mota del
corpo e supersonico rispetto agli ioni, subsonico r"ispetto agli elettronL Ma passando dagli strati inferiori a quelli superiori della ionosfera i parametri caratteristici del plasma nel quale il corpo si muove cambiano sostanzialmente: la velocita termica v. degli ioni aumenta , di-1
ventando dell'ordine di grandezza mentre la velocita del corpo
v
-0
10 5 cm/ sec.
(all'altezza di 700 Km. ),
diminuisce, di guisa che
v
-0
e v.
-1
diventano uguali, od anche pub verificarsi la condizione inversa a quella prima indicata,
ed il moto
e subsonico
rispetto ad entrambe Ie
particelle cariche. Dlaltra parte, la lunghezza di Debye, che nella ionosfera
e alquanto
pili piccola della lunghezza che caratterizza la dimen-
sione trasversale dell'ostacolo (precisamente,
1
-+ 4 cm.
contro 1 me-
tro), diventa ne 110 spazio interplanetario dello stesso ordine di detta dimensione. Al variare delle caratteristiche del plasma variano, ovviamente, gli effetti dovuti al moto del corpo; in particolare, variano il carattere del "riempimento" e Ie dimensioni della regione di "rarefazione" (rispetto all'ambiente indisturbato), dietro all'ostacolo, che superano notevolmente queUe dell'ostacolo stesso, e che presentano tanta importanza nella studio delle perturbazioni della propagazione delle onde elettromagnetiche. Non ho voluto, e non ho potuto, che fare un cenno a problemi di grande interesse attuale, che la dinamica dei gas rarefatti presenta, e che gia da soli giustificherebbero 11 esistenza di una speciale Sessione del C.1. M. E. dedicata a questa importante branca della Fisica-Matematica. I fondamenti matematici e fisici per 10 studio di questi e di altri problemi di uguale interesse, insieme colla trattazione di alcune tipiche
3
- IV C. Ferrari applicazioni dei metodi e delle equazioni fondamentali della dinamica dei gas rarefatti verrano esposti nelle conferenze dei professori Estermann,
Kamp~
de
F~riet,
Krywoblocki, Lune, ai quali come gia ho
detto in principio, molto deve il progresso delle ricerche in questo campo. Conferenze di Seminario sopra problemi particolari, relativi alIa interazione tra superfici solide e flusso di molecole libero, e al problema di Rayleygh in magnetogasdinamica, saranno tenute dal professor Nocilla e dai dottori Tironi e Sergianotto, mentre altre conferenze sull'onda d 'urto in gas rerefatto in magnetogasdinamica , e sulla struttura dell'alta atmosfera saranno fatte dai professori Agostinelli e GraffL Dai
nomi che ora ho detto appare l'internazionalita dei Confe-
renzieri; d'altra parte anche gli altri Studiosi qui presenti sono venuti da ogni parte d'Italia e da diverse Nazioni,
COS1
che questa
e certo
una assemblea internazionale altamente qualificata, in cui 10 spirito di collaborazione, come sempre avviene nelle riunioni di persone accomunate dall'amore
alla ricerca scientifica,
e pieno
e sincero. Ma qui
a Varenna c 'e una particolare atmosfera, l'atmosfera di "Villa Monastero" che rende i contatti personali particolarmente cordiali e fecondi, come hanno dimostrato i vari Corsi che qui si sono succeduti. Auguro che questa atmosfera sia ugualmente salutare per 10 sviluppo delle ricerche, di cui ora qui ci stiamo occupando, e delle relazioni e collaborazioni personali con amici vecchi e nuovi, e con questo augurio apro la presente Sessione del C. I. M. E.
4
CENTRO INTERNAZIONALE IvIATEMATICO ESTIVO (C.!, M. E. )
M..Z. v. KRZYWOBLOCKI
SOME MATHEMATICAL ASPECTS OF RAREFIED GASDYNAMICS AS APPLIED TO HYPERSONICS, REENTRY AND MAGNETO-GASDYNAMICS.
5
SOME MATHEMATICAL ASPECTS OF RAREFIED GASDYNAMICS AS APPLIED TO HYPERSONICS, REENTRY AND MAGNETO-GASDYNAMICS by M. Z. v. Krzywoblocki (Michigan State University)
T ABLE OF CONTENTS
Summary Introduction
1. Free Molecule Flow Technique and the Fundamental Systems of Equations 1. 1. Free Tvlolecule Flow 1. 2. Kinetic Theory of Gases Approach 1. 3. Equations Based on the Theory of Continuous Media 1. 4. Macroscopic Equations Based upon the Kinetic Theory of Gases 1. 5. Asymptotic Expansion 1. 6. MHD and Plasma 2. Hodograph Transformation and Integral Operators 2. 1. Preliminary Remarks 2. 2. Hodograph Transformation in Two Dimensional Flow 2. 3. Bergman's Operator Technique in General 2. 4. Generalization of the Hodograph Technique to Diabatic Flow 2.5. Generalization of the Hodograph Technique to Magneto-Gas-Dynamics 2. 6. Transonic Flow 2.7. Gilbert's Technique
7
- II -
M. Z, v.. Krzywoblocki
3. Reduction of Independent Variables 3. 1. Preliminary Remark 3. 2. Early Approaches 3. 3. Michal's Approach 3. 4. Examples of the Reduction to Parametric Functions 3. 5. Application of Theory to the Hypersonic Boundary Layer 3. 6. Particular Cases of the Direct Reduction of Independent Variables (with no Parameters) 4. Topological Technique 4. 1. Fundamental Concepts 4. 2. Application 5. Sub-Light Relativistic Hypersonics Based on the Relativistic Energodpamics 5. 1. Preliminary Remarks 5. 2. Fundamental Equations 5, 3. Transformation Equations 5.4. Transformation Based Upon the Principle of the Invariance of the Total Energy Space-Time Metric 5. 5. Generalized Transformation Equations for Velocity 5. 6. Generalized Transformation Equations for Acceleration 5, 7. Generalized Special Relativity and Mechanics, The Dynamics of A Particle 5.8. Three Kinds of Fundamental Laws 5.9. Optimum Energy Principle 6. Concluding Remarks Acknowledgement Appendix List of References
8
- 1-
M. Z. v. Krzywoblocki
SUMMARY
The author presents briefly the fundamental equations of the free molecule flow regime and of the remaining regions of the hypersonic flow. He discusses the recent developments in three techniques, possibly applicable to the hypersonic flow: integral operators, reduction of independent variables and topological technique. A brief presentation of the fundamentals of the relativistic energodynamics and some concluding remarks close the paper.
9
- 2-
M. Z. v.. Krz.ywoblocki
INTRODUC TION
There exists no precise definition of the boundaries of the regime known under the name of hypersonics. On one side it includes the reentry
phenome~
na which are inseparably connected with all the domains of the classical aeroand fluid-dynamics, involving all of the classical regions of the sub-, trans-, and super-sonic nature. On the other side it involves the relativistic phenomena with the velocity of light being another barrier, the light-barrier, having a some sort of analogy to the old sonic barrier. Practically, the hypersonics is interested in the sub-light regime, leaving the super-light region still in the sphere of speculations of the theoretical physics. Hypersonics applies all the possible tools, ever invented, developed and worked out by the mechanics of continuous media, kinetic theory of gases, Newtonian free molecule technique and finally the classical special theory of relativity. This refers to all kinds of gaseous media, 1. e., ideal, perfect, real, ionized, etc. There is no limit in that respect from any point of view. Concerning the problem of solving the differential and integral equations occurring in the fields in question, there are used all of the possible te"chniques and methods known in the theory of partial and ordinary differential as well as of integro-differential equations. One can mention here the classical techniques,
special
functions, algebraic, integral operator, topological techniques, reduction of the number of independent variables, etc. Each of these methods possesses certain advantages and disadvantages. The present work is concerned with certain aspects of the enormously large field of hypersonics, namely with the discussion of some recent developments in a few techniques of remodelling the equations, governing the flow of a gas in that region. Obviously, not only the state of a neutral gas but as well that of an ionized gas (electro-magneto-hydrodynamics or plasma-dyna11
- 3M. Z. v. Krzywoblocki
mics) should be considered. In general, the classical techniques are not strong enough to attack the nonlinear systems. Integral operator technique discussed in the present work, involves a hodograph transformation with all its enormously complicated formalism of returning back to the physical plane, difficulties with boundary conditions (u.nknown) in the hodograph plane, etc. The technique of the reduction of independent variables is actually based upon the elementary fundamentals of the theory of invariant groups. As such one, it does not take account of the boundary conditions. The topological techniques furnish usually in a very simple manner the existence proof. But concerning the formal solution at a point they seem to require a lot of formalism involving necessarily numerical procedures usually with the use of high-speed computing machines. Actually, all the techniques, discussed in the present work require necessarily the application of numerical solutions and the use of the high speed computing machines. There is briefly discussed the most recent tendency in calculating the reentry phenomena of blunt bodies (Apollo's shape) by simply programming directly the equations of motion for the high speed computing machines without any whatsoever remodelling them from their original forms. Finally, tha last part of the work is concerned with the approach to the hypersonic flow regime from the light-barrier point of view. This is accomplished by discussing the fundamentals of the relativistic energo-dynamics. It involves the invariance of the total energy of the system in question filled
out by a continuous (matter-full) mp.diuin under the transformation group of coordinates. The assumption is that the matter (i. e., a certain level of energy) can be transformed into the energy of the light (electro-magnetic-matterless energy). This approach could be of some value in motion of matter-full particles in the range of velocities approaching the velocity of light. Some concluding remarks close the work. 12
- 4-
M. Z. v. Krzywoblocki
1. Free Molecule Flow Technique And The Fundamental Systems Of Equations
1. 1. Free Molecule Flow In this chapter we shall briefly discuss the fundamental aspects of the free molecule flow {no collision between particles of the gas only the impact of the gas-molecules on the surface of the moving body}, reflecting them from more than one point of view. Next we shall collect some fundamental systems of equations used to describe hypersonic flow in its various aspects. If
\',
"'\' ,
~' are the components of velocity of a molecule in the
directions x', y', z' of a coordinate system in which the macroscopic velocity of the gas is zero, then since the velocity distribution is not modified by the impact with the body because of absence of collision between molecules in the free stream and re-emitted molecules from the surface of the body, the distribution is Maxwellian, or
N("i~'
(1. 1. 1) NEI"l'~'
ties
=N{H/'tt)
3/2
{ 2 2 2 eXPl-h{~' +'1\' +~')
l J'
is the number of molecules per unit volume in the range of veloci-
~'to
;' + d ~ "
divided by d ; 'd fY\ 'd
~
"1\'
~'+ d 'Y\' ,and
to
~'to l; '+ d ~ ,
'. N is total of molecules per unit volume. h is
related to the most probable velocity c. of the molecules in the free stream 1
by the expression (1. 1. 2)
2 where c i = 2RT = 2{p/ P
h = l/c.
2
1
).
If the observer moves with a velocity e 1U, e 2U, e 3U in the directions
x', y', z', then in the relative coordinate system x, y, z where the observer is considered at rest, the velocities
5 ' 13
IV)
,~
of the molecule in the
- 5M. Z. v. Krzywoblocki
directions x, y, z are (1. 1. 3)
Therefore, in the new coordinate system, the number of molecules having velocity component between ~ +d~
~,
III
'
~
S + d S ' I'It + d'yt
and
,
is
Suppose there is a surface dS whose normal is the x-axis. During a unit second the molecules, having velocity components between and
r +d ~
t ,'1\,
~
,"I + d 'l\ ' ~ + d ~ , and striking the surface dS, will be con-
tained at a given moment in the cylinder with dS as base and the length
V~
2 +IY) 2 +
~2
in the direction
~, "1 ' ~ .
The volume of the cylinder is equal to the area of the base dS multiplied by the heights (- ~
). The number of this kind of molecules is then Nf"'~ (- ~ )dS.
The number of molecules between f , ~ ,~ ~
-
+d, ~ Nh~
and
~ +d ~ ,
rtt + d rtt '
that will strike a unit area with x-axis as normal is then d~
d"l
d~
. The total number n of molecules striking this unit
area is
11.1.5) n- _NI:)3 j 2
I~
E, f.~dE
exp (-h[IE +eju)2+1,+e2u)2+1'+<>3U)lj
the result being
(1. 1. 6)
n
=
Nt [2( 11 h)1/2] -1 exp [-h(e 1U)2] +2 -1 e1U [1+ erf(e 1U\["h)J}
where erf(t) is the error function defined as (1. 1.7)
erf(t) = 2( 11 )-1/2
ft exp(-s 2)ds. 0
Hence, the number of molecules per second striking a unit area of the plate inclined at angle 9
to the stream with velocity U is
14
- 6M. Z. v. Krzywoblocki
(1.
1. 8)
~
n=
l('!l'h)-1/2 exP [-h(USine )2]tUsin9 [lterf(Uhl/2sin9n
The mass m. of the stream per second striking a unit area of the plate is 1
(1.
1. 9)
m. = ci 91/2 1 21T
fexp [-(Uc.-1 sin Q) 2] t
J}
{\ 'IT1/2 Uc.-1 sin (I [ l+erf(Uc.-1 sin I7)J
1
1
1
The component of velocity of the molecule in a direction, having directional cosinese l , e 2, e g with the axes is elf te 2,,\ teg~ . If this molecule is absorbed by the surface after striking it, the corresponding momentum will be transferred to the surface. For the surface with x-axis as normal consi-
or
(1. 1. 11)
The pressure p. due to impact of molecules on a plate inclined at an angle 9 1
to the stream of velocity U is calculated by using (1. 1. 12)
e 1 = sinG, e 2 = cosQ , e 3 = 0; e l = -1, e 2 = 0, e g = 0,
and the impact pressure p. is 1
Pi
'IT -1 / 2sin9 (c.U -1 )exp [ -(Uc.-1 sine) 2] t 1
1
(1. 1. 13)
2_] . [ l+erf(Uc -1 sino) C'\] . t [ 2-1 (ciU -1 ) 2 tsin9 i To calculate the shearing stress
1;. l
due to impact of molecules on the pIa-
te, the direction cosines are 15
- 7M. Z. v. Krzywoblocki
(1. 1. 14) By substituting Eq. (1. 1. 14) into Eq. (1. 1. 11), we obtain
1'i pu2
- - = cos
~
(1. 1. 15)
-1 -1 2) 9 . c.U exp[ -(Uc. sin G) + 1
1
'
2
+ sin e cos e (1 + erf(Uc~lsin 9 )). 1
If the molecules reflect specularly from the surface, then the pressure
and the shearing stress due to re -emission are (1. 1. 16) On the other hand, if molecules re-emit diffusively, I: r = 0, and the pressure Pr due to diffuse-reemission, is given by =
2 -1 1T l/2 c U-1~ c.U -1 '"IT -1/2 exp [ -(Uc.-1 sine) 2] 4r
1
1
(1.1.17)
where c
r
is the most probable velocity of the molecule in the termal equili-
brium at a temperature T (1. 1. 18)
c
of the reemitted gas, having the relation
r
2 r
=
2RT
r
The above presentation is actually a part of the work by Tsien
(72]
1. 2. Kinetic Theory of Gases Approach. The approach to the free molecule flow from the kinetic theory of gases was proposed by Heineman [37] and Keller [42] . One can find the momentum per second imparted to dS ' by the impinging only:
16
- 8M. Z. v. Krzywoblocki
o
f. .
""
. !w. u(u cos 9 + v sin 9
-m [ ....du J.;v
).
(1. 2. 1)
.f(u+Vcos9 ,v+Vsine ,w,xyz)dS'. Here the symbol f denotes the distribution function of the gas. Then total momentum per second D. for the impinging molecules is 1
D. = -m 1
Ii
f f
,0
dS '
["'"
du
_00
dv
rot:>
1
_ ..
dwu(ucos 9 +vsin
-00
e ).
.f(u+Vcos9 ,v+Vsine ,w,xyz)-
- fdS " IoOOdu
(1. 2. 2)
fO<> dv f:w u(u cos
-~
....
e + v sin e ).
. f(u+Vcos9 ,v+Vsin9 ,W,XYZ)}. The total momentum per second exerted by the reflected molecules is
() rf
r. .
[00 J.O" ).~wu(ucos B +vsin9 o
Drs = -mt dS' J du .;v
).
. f(u-Vcos9 ,v+Vsin@ ,w,xyz)· (1. 2. 3)
- [dS"
[~u
f,!v r:;U(UCOS
9
+vsine ).
,
. f(u-Vcos6 ,v+VsinB ,w,xyz)] , for specular reflection, and
D~d) = -m [ fdS' fo~u f.!V [~WAfU(U cos g + v sin g) exp( -hr c 2)-
f f. ~ f.~ [; '\
(1.2.4)
_ dS"
u(uco, 9 +v ,ill
e )oxp( -hr 2) J C
I
for diffuse reflection as the distribution of reflected molecules is of the form
2
-1
A exp( -h c ); here h = m(2KT) where T is the absolute temperature r r 2 r2 2 2 r of the reflected stream, c = u + v + w. Af and Ab can be determined from the conservation of energy and number at the surface. Af and Ab are given by
17
- 9M. Z. v. Krzywoblocki
(1. 2. 5)
. uf(u+Vcos 9 , v+Vsin& ,w,xyz), Ab ("'''du J:v r:wueXP(-h c 2) = - fdU
I.
00
(1. 2. 6)
. uf(u+Vcos
r:v r:w .
o.~
r
_0<>
_ ...
e , v+Vsin8
. .,
,w,xyz).
In the first approximation we neglect collisions between the molecules. At infinity the function f is Maxwellian, i. e. , (1. 2. 7)
f = /M) = N(h "/l'-1)3/2exp
f-hUu+V cos e )2+(v+V sin e )2 +w2]} .
Here N is the number density for the monatomic molecules and at the boundary, f = f(M) + fIR) where fIR) is the distribution function of the reflected stream. Eqs(1. 2. 5) and (1. 2. 6) now take the form
(1 2 8)
Ab = N 'lr - 3/ 2h~ / 2h 2 {_I 1T 1
r
(1. 2.9)
+ 2V cos 9
h~ 1)1 / 2V cos e + h~l exp( _h.V 2cos 29
r
1
1
)+
1
Vcos9 2} exp( -hiu )du ,
0
where h. = m(2KT ,)-1, T. being the temperature of the impinging stream. 1
1
1
Heineman furnishes the drag coefficients for a plate, sphere, right circular cone and prolate ellipsoid. We present here the drag coefficient of a plate: Specular reflection : CD = 8(2'1T()
-1/2 -21 1/2 2 2 M l(2/~) (1+ ~M cos
(1, 2. 10)
+ Mcos
e )erf(t/2) 1/2 Mcos9)+
e exp(-(
2
2
~ /2)M cos ~
18
1.
)j
- 10 M, Z, v. Krzywoblocki
Diffuse reflection :
CD=4(21(~) (1. 2. 11)
-1/2
2
(Micos9)
-If
112
1(2 / K)
2 2,' + M.exp[-(~ 12)M.cos!1H 1
G
1"'
2
(1+(Mi)cosger'f(((/2)
I
1/2] Micose
11 2
+(2'11' V-)
1 M- cosF
,
r
,)
I
where M. and M are respectively equal to VIv(i) and VIv(r), v(i) and 1 r s s s v(r) being the sound velocities in the impinging and reflected streams.
s
Keller employs the Jaffe's method I 41.' with more general type of boundary condition. Let s denotes arc length along a trajectory, then the Boltzmann equation may be written:
3
(1. 2. 12)
). = a IN!f
where a is a typical macrt1scopic dimension;
2
,
\ is (when divided by 'iT' )
the mean free path in a gas of spherical molecules of radius cr
; N' = num-
ber of particles fa 3 ; (1. 2. 13)
J(f, f) =
fJ
J Lf(c'lf(ci)
- f(C)f( c 1
iJ; c-c 1: bdbd:: iC 1
.
We now assume that f can be represented by a convergent power series in : (1. 2. 14)
-1 -1 2 f = fo + a ,'I f1 + (a.l ) f2 + ....
If this solution is inserted into Eq. (1. 2. 12) and coefficients of like powers of a \-1 f\,
are equated, the following infinite set of equations is obtained : dfl/ds = J(f ,f ), o 0
df/ds=O, o (1. 2. 15)
df Ids = J(f ,f 1) + J(f 1,f ) = 2J(f ,f 1),··. 2 0 0 0 When ther external forces F l' F 2' F 3' are constants, the trajectories may be calculated explicitly, and one finds that the general solution for f
o
is:
(1. 2.16) fo=h(X-Ut+Flt2/2, y-vt+F2t2/2, z-Wt+F 3t 2/2, u-F1t, v-F 2t, w-F 3t ),
19
- 11 -
M. Z. v. Krzywoblocki
where h is an arbitrary function. Keller assumes that for every gas molecule which strikes the surface element
dS
of a solid or liquid surface during the time
with velocity ( relative to that of
dS) between
c' and
dt
c r + dc'
there is a probability p(c, c') dc dc' that a molecule with velocity between c and c + dc will leave the element dS during the interval dt. Further, he assumes that there is a probability g(c) dc dS dt that a particle with velocity between c and c + dc will spontaneously leave the element dS dudng the time interval dt , and finds the relation which should be satisfied by ftc) at a solid or liquid surface : (1. 2.17)
J
-f(c)c.n=
p(c,c')f(c')c'.ndc'+g(c)
ct. n)
for
c.n< 0
0
where n is a unit normal to dS pointing out of the gas, arid p( c, c') is given by: (i) Specular reflection : (1. 2. 18)
p (c,c') = s
roO
l
if c = c' - 2(c'. n)n
0 if c
rc
l -
f , J'
ps(c, c')dc = 1 ;
c. n(
2(c ' .n)n
0
(ii) Diffuse reflection :
Jrc. n < po(c,c')dc=l;
(1. 2. 19)
0
P (c, c l ) = (2 1r) o
-1 2
m (KT)
-2
;2 -1' (-c. n)exPI-mc (2KT) j ~
where temperature T is given by T = T + a. (T - T ), T = temperature g s g g of incident molecules, T = temperature of the surface, a. = coefficient of
s
accomodation, 0
<. 0..
< 1.
(iii) Part specular and part diffuse : (1. 2. 20)
p(c, c I) = frPn + (l-fr)P S ;
20
J
c. n < 0
p(c, c ')dc = 1 .
- 12 M. Z. v. Kr.zywoblocki
The theoretical determination of p.(c, c ') and g(c) can be based on the quantum mechanical investigation of the interaction of molecules with a solid surface (see [63J ). The drag force and the torque (with respect to a given origin) exerted by a gas on a body in the gas are given by : (1. 2. 21)
D=m
J Jcfc. dSdc
,
S c
(1. 2. 22)
J
J
T = m S c c X rfc. dSdc
1. 3. Equations Based On The Theory Of Continuous Media.
The equations of this kind used in hypersonics in any region, mentioned above (sub-, transon, super-sonic), are the well-known equations of Euler and Navier-Stokes. In the field of hypersonics there are usually introduced the following improvements : ~ must be properly p v adjusted to the physical nature of the gas (not necessarily the air in the zero-·
(i) the specific heats c ,c ,and their ratio,
level conSistency); since in the mechanics of continuous media these parameters cannot be evaluated from the fundamental concepts, they must be furnished by the experimental physics; (ii) similarly, the both coefficient of viscosity, their relation and the
coefficient of heat conduction resulting from the transport phenomena must be furnished by the experimental physics. Since the above equations are so well-known, we do not need to present them in the present work.
1. 4_ Macroscopic Equations Based Upon The Kinetic Theory Of Gases.
These equations were derived in the past and actually present some ave-
21
- 13 M. Z. v. Krzywoblocki
rage description of the medium in question, which supposedly consists of discrete particles subject to the collision phenomena. Since these are macroscopic equations we can deal with them in a way similar to that applied to equations based on the mechanics of continuous media.
[70J :
We briefly quote the following system of equations from
1L + o(pui ) = 0
(1. 4. 1)
~ t
"\ ui ~ " t
(1. 4. 2)
*a
(1. 4. 3)
"0
+U
x.
1
'0 u i 1 d P .. + p- __ lJ '0 u. d x.
-
j
J
J
"3 (pu·) 0 qi + ~ + (2/3) 'Ox. + (2/3)P ij 1
(1. 4. 4)
(1.4.5.)
dt
aUi
k
= 0,
J
1
t)q.
()(;'ij +
=0
d ( "t'.. )-(2/5)~ -x- uk 1J ax
j
k
_1 0P 'k
oT
'I>
U,
()
u.
+1:'k~ - p~ = - pl1't .. 1 (IX ClX. 1J, k
j
aT
-1
- (7/2) "t"kR~ +'t' .. p ~+ (5/2)p R ~ = -(2/3)pf1
UXk
1J
vX k
where
(1.4.6.)
22
uXi
qi'
- 14 M. Z. v. Krzywohlocki
=f
(1.4.7.)
f
(1. 4. 8)
f0
(1. 4. 9)
U.=
1
(1. 4.10)
o
(21TRT) -3/2 n exp [_ V2(2RT) -1],
::
n
-IJ v . fdv. . ml
= mJ(V
p .. 1J
(1.4.11)~.
[1 - (2 pRT) -11' .. V.V. - (pRT)-l q .V .(l_ V2 (5RTfl)] , IJIJ 11
(n)
IJ
m
q." 2-1 m
,
1
.
- v ) (v -v ) m 1· m. m. m.
J
J
1
JV.V 2 f d ...vm, 1
fd~m;: mfV.v.fd~ =pL-~., 1 J m 1J 1J
-1 [D't ij (n-l) (n-l) C>uk i p =r--r +"r. ~x. Dt IJ '()xk i)U
J
oqi(n-l) - (2/5)~ +
ux.
J
Cl u· ]
+~. (n-l) _J , 1k () xk r:-.
q1.(n)
= _ k .!! - 3r(2p)-1 thox.
1
au.
Dq.(n-l) iJ u. ..(n-l) (_1 +(7/5) q.(n-l) _J _ RT ()t'1J Dt lOX. '0 x. J J
ou. ";Iu.)
+ q.(n-l)[ _1 + (2/5) ( 1 +.--l) J h. dX, OX. J
and
A .. :: A .. 1J IJ
r=
+ A ..
J1
J
1
- (2/3), .. Akk ' IJ
0.243 (2mA -1) 1/2 mRT ,
kth = 15RI"/4. 23
- 15 M. Z. v. Krzywoblocki
Ikenberry and Truesdell [37) present the n-th iterate:
(1,4,13)
I
-(P/~)
,-
p(n+l) 2 "'" + p' (n)E 2/3 p (n)E O I .. ~ p.l!i .. .. -( )kl kl ' IJ IJ IJ 1J (n) +r, (n) / (n) f + 2P k (illj)k ~ijk-(2 3)h k °ij
1
jk'
(n+1) (n+l) _ .(n) + (n) E + (n) - (p/21") (3p"k 2h(.l ) ) -. P"k P"k p.IJ, 'kl 1 ik IJ 1\J IJ IJ
(1.4.14)
-1 ( - 3D pOI •. 1) J l1J'I~
+
)-3
11 I
(n)
+3 P l (" ) l' IJ uk ' where:
Pi.il"; .,.
1 ....,
(1.4.16)
(1.4.17)
(1,4,18)
=Oc. c .... c. - ) 11 12 In
(')
=mJ
g\) ft
00
c. c .... c. fd.}, h.=(1/ 2)p ... ,(1.4.15) 11 12 In 1 IJJ
+( )'iU,il
P .. ='P .. -p6 .. , E:uk'k' E .. =. {l/2)(u . . +u . . )-(1/3)El .. , IJ IJ IJ IJ 1, J J, 1 IJ
r = (1/3)
-1 1/2 -1 (2m G) mp (pA 2 ) ,
G being the constant of proportionality between the intermolecular
24
- 16 M. Z, v Krzywoblocki 0
A2 = 1. 37 ••• is a nume-
force and the reciprocal 5th power of distance, rical constant evaluated by Maxwell
[ 64
J . The
above iterate gives
the following system of equations: '\
ad (ntl) t
(1. 4.19)
(plntl)
Ui
(ntl)) i ,
=0 ,
(1.4.20) (n) 5 (ntl) (n) (ntl)E(n~ 2h(nt O 3 '"0t P(ntl) t 3p (ntl) Uo t p E t 2P.
(1.4.21)
0
,1
0
1
d p(ntl)t p (ntl) (n) p(nt l) E(n) t
"
0
1J
etc.(1.4.22)
0
1J,
t
k Uk
too
1J
+ 2p
(n)
,lJ
1J
0
0
-
1, 1
,
(n) Eo 0
1J
(n) (n) r / (n) P"k k -(4/15)h k ko,,+(4 5)h(o 0) IJ ,
00
1J
1, J
_ -0,
The Hilbert-Enskog- Chapman-Burnett method obtains the necessary relations in terms of a solution of the Maxwell-Boltzmann equation. It is assumed that the collision term is dominant and that the distribution function may be determined by successive iterations on the collision term. This is roughly equivalent to assuming that the distribution function is given in terms of a power series in the mean free path. Then the Maxwell-Boltzmann equation can be reduced to a sequence of linear integral equations, which are in turn solved by replacing them by systems of simultaneous linear equations. The first approximation to the distribution function is the equilibrium distribution, (1. 4.8) which gives the Euler equation after substitution into the Maxwell-Boltzmann equation.
25
- 17 -
M. Z. v. Krzywoblocki
The second and the third approximations respectively give the NavierStokes, and the Burnett equations. The Euler, Navier.Stokes and Burnett equations are all contained in the system of Eqs.(1.4.1) - (1.4.5) if it is assumed that 't .. , and lJ
' i can be given in the form of (1.4.11) and (1.4.12).
p(~) lJ
= q(o) = 0 (Euler equation) yields"t' ..(1) , q(.I) , which give the Na1
(2)
vier-Stokes equations.'):.. lJ
(3)
xwellian molecules) and 1:.. lJ
(2)
lJ
,q.
1
,q.
1
yield the Burnett equations (for Ma(3)
1
will furnish the so-called Thirteen
Moment equations, derived originally in 1949 completely independently by Grad [34]. 1. 5. Asymptotic Expansion.
Asymptotic solutions of Boltzmann equation are very thoroughly treated by Grad. The reader interested in the subject is referred to his works, (see [35, 36J ).
1,6, MHD And Plasma Actually both magneto-hydrodynamics and plasmadynamics in all their forms are used very extensively in the field of hypersonics. Below in chapter 2 and 4, we shall be barely able to consider a MHD flow. Plasmadynamics will not be discussed at all and the reader is referred to the enormous number of references on this subject. Needless to say, the dynamics of rarefied, ionized gas is today perhaps the most important aspect of the reentry hypersonics. Lack of space does not allow the author to discuss it as thoroughly as it should be done.
26
- 18 M. Z. v. Krzywoblocki
2. Hodograph Transformation and Integral Operators. 2.1. Preliminary Remarks. During reentry of a blunt body there appears in front a bow -shock. Behind it the flow varies from sub- through trans-, to super-sonic one. The subsonic domain must be determined very precisely. It furnishes the initial data for the transonic flow and finally through it one must obtain precise information about the supersonic region at Mach number greater than one so that one can apply the theory of characteristics which is so powerful tool in the supersonic domain. We shall not discuss the corelation between subsonic and transonic flows.
In this chapter we restrict ourselves to discussing one of the possible techniques in the subsonic domain . Due to the high temperature the gas behind the bow-shock may be ionized. The technique in question is the hodograph transformation combined with the integral operator technique. The method is applicable at the present status to an inviscid, non-heat conducting gas. In the case of an irrotational steady flow, the equations in the hodograph plane are linear. In other cases of flows like rotational, magneto-hydrodynamic flow, etc., the equations in hodograph plane are nonlinear. In some terms there appears the Jacobian of the transformation in coefficients in front of derivatives. In such cases, one can apply a limiting process (iteration, successive approximation, etc.) starting with the irrotational equations. In hodograph plane such a procedure is much more efficient and more strongly convergent than in the physical plane. The procedure carries with itself all the disadvantages of the hodograph transformation technique like lack
27
l
of
- 19 -
M. Z . v. Krzywoblocki
knowledge of the boundary conditions in the hodograph plane, a very time consuming returning process from the hodograph plane to the physical plane, etc, All these items must be solved by some sort of approximation techniques. The two curve boundary value problems must be solved by some sort of trial and error procedure. The technique was generalized to a three dimensional flow and to MHD flows in a two dimensinal case.
2. 2. Hodograph Transf ormation in Two Dimensional Flow To explain the fundamental aspects of the integral operator technique we begin with the simplest possible case. It is assumed that the flow of a perfect gas is steady, irrotational, in-
viscid, non-heat conducting, isentropic and two dimensional. The equations of conservations of mass, momentum are:
t: (p
(2.2.1)
u)
+
fy
(p v) = 2
(2.2.2)
j Ii d, = - dp ;
.p d (-%-
where the used symbols denote : city components in
.p
0
I
) =pd
2+ 2 ( T ) - - dp,
is a mass density u,
x-axis and y-axis, respectively,
ty along the stream line. Introducing the speed of sound in Eq. (2 1 2.2) we obtain: 2 2
(2.2.3) dr
u +v -pc -2 d (-2-)'
28
v are velo2
c = dp / d P
- 20 -
M. Z. v. Krzywoblocki
from which one can calculate the partial derivatives
'Op/ox, VprVy.
Eq. (2.2.1) may be written as :
Let
q=
grad ((J
;
r
oP ~ ox'ry
we substitute
calculated from
Eq. (2. 2. 3) into Eq. (2.2.4) ; then the equation for the potElntial function has the form: (2.2.5)
(1-~2\cDTxx +(1-..t2 )1DTyy -2'u2', xy =0. C
C
C
We may satisfy Eq. (2.2.1) by a stream-function,
(2.2.6)
pu =
*'
'Y
='r(x, y) :
()1jI
pv = - Ox '
or
d'/l
(2.2.6')
~
1
-- p
In a way analogous to the procedure for obtaining the potential function we get: (2.2.7)
u
(1 -
2
v
2
2" )" xx + (1 -2") 't' yy c
c
uv 2 -2 c
'1' xy
= O.
Following the standard procedure, the hodograph transformation from (x, y) - plane to (q, 9) - plane furnishes the general Cauchy-Riemann equations: (2.2.8)
,
29
u = q cos 9, v = q sin 9,
- 21 M. Z. v. Krzywoblocki
or Chaplygin's equations:
h
(2.2.9)
2
I_M 2
ll+ ~ +2
Oq2
~e2
q
(1
+ M2)
q
a~
Tci
i _ I_M 2 _ i'1l +I_M ..£.1.+ _2 -d
(2.2.10)
Oq2
q2
pq (-) dq 1- M2
yq
oe 2
let (2.2.11)
A=
J (I-M)2 1/2 q-1 dq, 0
o,
* =
o.
d). = (I_M2)1/2 -1 i. e., dq q ,
in Eq. (2.2.9) , then we have:
(2.2.12) where If we introduce
'I' = f it ~ -1/2 ,
in Eq. (2.2.12), we arrive at : (2.2.13)
2
* i
~
b. 'f'~ =tl +.-!
"Ot
ue 2
= F ()..)
,\,*',
where (2.2.14)
F (}.) = (2 pl/2) -1 d ~
For an incopressible fluid (M=O) , ~=
Y-1
= constant,
F = 0,
(~-1/2 dd: ) • A= log
q , we get:
/J. 'I' = constant,
In the compressible case, we attampt to find a solution in the form:
30
- 22 -
M. Z . v. Krzywoblocki
~ L n=1
+
\Jjk; f I 0
(2.2.15)
f g , n n
where fo' fl ' f2 ... are analytical functions of the complex variable z = ~ + i (} and
gi' g2 ' g3 .•• , are real functions of ). only.
If the relation (2.2.15) satisfies Eq. (2.2.14)
we can take its real part
I
as well as its imaginary part as a solution, each supplying one definite flow of a compressible fluid. From: (2.2.16) Mn = 0
A A oa db 7> a ~ b f:l (ab) = a ub+ b a+2( QA 'IT + ~ "'Si) ,
and
we find:
Of
6 (fnn g ) = f I:Jg + 2 "\~ n n
(2.2.17)
()I\
()f g' = f gil + 2 2 n n n OA
g' . n
Here of /?J).is the directional derivative of
f
real axis and equals the complex derivative
d f / d z of the analytical
n
function
f
n
n • Substitution of (2.2.15) into Eq. (2.2.14) gives the condition:
00
~
n=
in the direction of the
n
00
( f gil + 2 fl g' ) = F nn nn
(~) ( fo + ~ f g ). f-. nn n=1
or (2.2.18)
f
~
(fn (giln
We choose the functions
(2.2.19)
(-1/2)fn _1
f (z) n
=f~,
and the functions g ().) n
J
- Fg ) + 2 fl g' n n n
=F
f • o
so that
z i.e., f1 =(-1/2)/ fodz,
so that :
31
Z
f 2=(-1/2)jf 1 dZ, ... ,
- 23 M. Z. v. Krzywoblocky
(2.2.20)
gIl:: g" - F g ; n+ n n
n:: 0,1,2,3, ... ,
With these assumptions, the above condition can be written as: 00
(2.2.21)
L
n::1
(f gl - f gl) :: F f • n n+1 n-1 n 0
The left-hand side is the limit for n::
00
of the difference:
f gl 1 - f gIl' Since the sum in (2.2.15) must be supposed to be convern n+ 0 gent, f g goes toward zero. We also assume that f gl or f gl 1 n n n n n n+ has the limit zero for infinite n. Then Eq. (2.2.21) reduces to the simple condition: (2.2.22)
- fog'1 :: F fo
or
gIl :: - F,
gl:: -
J~ F d).
•
The recursion formula (2.2.20) can also be written as ~
g~-l
J F gn-1 d). , and this remains valid for n :;: °also if we introduce
(2.2.23)
gn::
-
The sequence of functions
p, p
o
. depends on F (~) and
gl' g2' g3"
this is completely determined by the
g :;: 1 .
-
relation. Thus the gn O,)
can be computed once for all and tabulated. The functions
f1' f2 ' f3 • . . depend on fo only and can be re-
presented explicitly in terms of f. By differentiation it can be seen o that (2.2.24)
fn (z) :: (-1)
n
n -1 (n! 2)
.£(z (z -~) n
flo (') d) ,
satisfies the condition (2.2.19). In fact, taking the derivative on both sides, n n-1 fl (z) :;: (-1) (nI2) n n
.
fZ (z -~) n-1 fl (~) dt:;: (-1/2) f 0
0
32
n-
1 (z) .
- 24 M. Z. v. Krzywoblocki
For n= 0 , the right-hand side of Eq. (2.2.24) becomes
f
o
,expect
for an additional constant which has no importance. Inserting Eq. (2. 2. 24) in (2.2.15) leads, with g = 1, to: 00
~o
yO"
=!
(2.2.25)
gPl z
o
(n! 2n l -1
f~ (~)
r(~
'In
o
r~ (~l d~
G (A, 9,' ) d"
where (2.2.26)
G(l, 9,~)
If we write
w(z) for the arbitrary function
f' (z), the final o
solution can be written in the form:
(2.2.27)
't'()., fJ)=f 1/ 2
Once the p, then
p - relation
G (A, fJ,
C) is
JZ G(~, fJ,~) w(~)
is known,
p
de
is a given function of A.
defined by Eq. (2.2.26) while for w(C) any analytic
function of one complex variable can be introduced.
Eq. (2.2.27) re-
presents the integral operator which transforms an analytic function into a couple of solutions of Eq. (2.2.9) • There remains to be seen that the infinite series included in Eq. (2.2.27) converges and that all solutions of Eq. (2 • 2. 9) can be represented in this form by an appropriate choice of f (~) . For an isentropic flow the quantities A, ~ ,F in terms of the Mach number
q2[(~_
1 ) /2
can be expressed
M. From the fundamental relation
+ M- 2] = c! ,where
6 33
is
the ratio of the specific
- 25M.Z. v.Krzywoblocki
c / c, we derive:
heats, 1 1\
p
v
jq (l-M)2 1/2 q-1 dq = (1/2) jq (l- M)2 1/2 d log (q2)
=
J
2 1/2 = (1/2) M (l-M) (2.2.28)
d [ - log { (t-l) /2 + M-2}]
= 2- 1 log ({ 1- {1_M 2// 2} l1+(1_M 2)1/2 r1[{1+h(1_M2)1/2) • 2 .{ l_h(1_M 2)1/2}J 1/ h )
,
where
-1
h=(t+1){t-1)
.
By similar computations the functions
~
and
Eqs. (2.2.28) and (2.2.30) combined determine therefore, the functions
Fare:
F as function
of ~ and,
gl (A) , g2(A) , g3(A) , ".
For the sake of practical computation an appropr,iate independent variable has to be chosen. The function pression;
F (A)
F can be roughly approximated by the simple ex-
=
From Eq. (2. 2. 20) the
C~
.2
(C
= constant) .
g can be computed : n
gO.) = 2n nl C ). -n with C 1 = -(1/2) C [1+n -1+ C(n+l)-2] . n+ n n . n
From a certain n = n
o
the bracket is smaller than 2, whatever
34
C is,
- 26 M, Z, v, Krzywoblocki and thus IC n+11 < ~ICnl with 1,,\1<1, The coefficients gn(nI2 n )-1 in Eq, (2,2,26) behave like the terms of a convergen geometrical series, For further details ,see
[66J
'
Once a solution in the hodograph plane has been found, one still has to transfer it to the physical plane, First assume that both stream functionr and potential rare known as functions of q,
e and
the lines
stant plotted with a certain increment
d (f
f
= constant and
= con.,.
=A. Let P be the
= d't'
point in the physical plane that corresponds to
'I'
pl. To find the stream-
110
1 pI and pp = llq-l: line element PP 1 mapping plpl we make pp _ -1 1 1 1 = A( OP I) ,In this way the entire streamline passing through P
can be plotted, The next streamline will be known if we have one initial point P 2' Such a point is given by PP 21. PP 1 and j5j5 2 =ll(pq) -1 where and
p
a known function of
q, depending on the p,
q = QT151
p - relation.
An analytic procedure assuming the knowledge of the function
'I' (q, e) alone dy = sin
~ 'I'
~
can be based on the relations Qf Ids
e ds
dq + d f ~
= q,
dx = cos
ed.s I
and runs as follows, Along a streamline we have:
de = 0 'oq 11. dq +
~e d e = d
Using Eq. (2.2,8) , we obtain:
(2.2.31)
35
- 27 M. Z. v. Krzywoblocki
For a definite streamline
'I' (q, 8) =c
we know
8 as function of q and
thus the factor of dq is a known function of q. From the above relad ~ we draw;
tions between dx, dy, ds, and (2.2.32)
x=
J q-1 cos
8 d
~
J
, y = q-1 sin 8 df.
If d ~ is substituted from Eq. (2.2.31) the integrands become functions
of q and the integrals are defined. They determine one streamline. To find the other ones, one potential line has to be computed which supplies the initial values for the integrals (2.2.32) . This computation can be performed in an analogous way if
f - line
in the x, y-plane will be given by
the conditions of the flow, e. g., if all streamlines are known to start with the same initial velocity vector. In practical problems boundary conditions will be given in the physical plane while the functions
'!l
(q,8) and
tp
(q, 8) can only be
determined by conditions in the hodograph plane. A tedious process of successive approximations has then to be applied . Assume that the flow around an elliptical profile
E with given velocity at infinity has to be
found. We would then start from the known solution of this problem for an incompressible fluid. This gives us a velocity distribution along the ellipse and allows us to map the ellipse onto the hodograph plane where it appears as a streamline problem for a flow around
along
E'. We try to solve the compressible fluid E' in the hodograph plane. This supplies a
E' and enables us, according to the first me-
thod described in this section, to plot the corresponding streamline E1 in the physical plane which will not coincide with E. A contour between E and
E1 might be assumed and the whole procedure repeated, and so
36
- 28 M. Z. v. Krzywoblocki
on.
2.3. Bergman's Operator Technique in General We shall briefly discuss the main results obtainable by means of Bergman's operators. The reader who is more deeply interested in the subject is referred to the original sources
[4 - 9 , 47, 48, 51, 60) .
We begin with the discussion of differential equations in two variables. In general, Bergman considers a differential equation of the form: (2.3.1)
"'~
:=
L(U) = U xx
=
~
+U
yy
+ aU
'Z
x
~
+ b U + c U = O. y
By means of substitutions : (2.3.2)
z=x+iy,
z
*' = x - ly. ,
Eq. (2.3.1) takes the form: (2.3.3.)
L(U)=U
-
zz
Let A= A(z,
1\0?<
!i'+AU +BU'!t+CU= 0, B=A,U(z,z )=U(x,y). z z
z*}, B:B(z, z",) , C::C(z,
z~,
for (z, zll-)J "£(0, 0),
be continuously differentiable functions and let (2.3.4)
D=n
z
-1~ z
dz*"+ B,
F=-A -AB+C, z
o
where n = n(z) is an arbitrary analytic function of a complex variable which is regular for Let further
z E "ij. 2( 0) .
,...E(z, z*' ,
'I\: 4 t), for (z, z )(11 (0,0),
\t
I ~ 1,
be twice
continuously differentiable solution of the equation: (2.3.5)
-
2 -
B(E)=(I-t )E*t-t z
-1~ .... r""E )~ 0 E~+2tz(E ~+DE ....... + z
37
zz
..
- 29 M. Z. v Krzywoblocki
which possesses the following properties
*
For (z, z }eU;4 (0, o) lim (1 - t 2// 2
(2.3.6)
E ~z,
(uniformly in t) . Further
z~)(
(z,
z*,t) = 0
z
t=+1
-u:
t
-1'E If is continuous for
z
It I
(0, O),
~ 1.
Let
f
U(z, z-)I. } = lE(Z, z1f.,t} f (z(l-t 2/ ) 2} (l-t 2)-1/2 dt,
(2.3.7)
s where
f
is an analytic function of a complex variable , regular at the
origin, (2.3.8)
z* E(z,
l, t} = exp[ - J Adzif + n(z}] E(Z,
1
and s is a path in the complex and
1 and omits the point
zit, t} ,
t- plane which connects the points - 1
t = 0 • Then Eq. (2. 3, 7) is a solution of the
equation: L(U}
(2.3.9)
= Uzz·..+ AU z + BU z~ + CU = 0 ,
which is twice continuously differentiable in
-Uf(o,
OJ. For the proof,
see [ 4.J . We call
E (see Eq. (2. 3. 8) } a generating function for the differen-
tial equation (2.3. 9) with respect to the origin. Eq. (2.3. 7) yields complex solutions of Eq. (2. 3. 9) . If B(z, z} = A(z, and C(z,
z}
is real, we obtain for
l = z real
38
solutions writing:
z)
- 30 M. Z. v, Krzywoblocki
where we denote by El the function (2.3.8) and by E2 an analogously formed expression such that E 2(z, z', t) = El (z', z, t) . To a given differential equation (2.3.9) there exist infinitely many generating functions E. It is of interest to investigate them, and to determine those which have some interesting properties. For harmonic functions we have the representation: (2.3.11) where
g is an arbitrary analytic function of a complex variable
We shall show that the generating functions E" i = 1, 2, in (2.3.10) can 1
be chosen in such a way that (2.3.10), after a slight modification, represents a generalization of the formula (2. 3. 11) . Bergmann distinguishes various kinds of operators. He calls the operator C2(z,
t; g) which transforms
g(z) into U(z, z>l') the integral opera-
tor of the first kind for the equation
L = 0 . An integral operator of
this kind can be obtained as follows. Let 00
(2.3.12)
g(z)
=
I n=O
and let
12.3.13) 11,/2)" -121r)-1[ gl'll_t 2))t- 2dt Inn+l) \nl/2)rln+l/21j"1 s If E,(z, z~ t) , i ;: 1, 2, are of the form: 1
z'* (2.3.14)
E 1(Z,z; t)=exp[-{ A(z,
Jo
(2.3.15) E 2(z, zl\', t) = exp[ -
rA(Z~
/)dt]~+tz:f(z,
z*,
til '
z) dzJ[ l+tzl ((z ,z, t )] ,
\,
39
A/.
- 31 -
M. Z. v.Krzywoblocki
then C2(z, z;~" g) :E 1(z, z ,t) f (z(l-t 2)/2) + 'sl '
=,'
(2,3.16)
will be an integral operator of the first kind. Here
f
is defined
by
(2,3.13) . For various applications it is convenient to write the integral operator (2.3.7) in a somewhat modified form and to derive for it different representations . We introduce the function
g(z) (inverse to
(2.3.13) )by the relation: (2.3.17)
('1 2 2 1/2 g(z) = f(z(l-t )/2) (l-t )dt. Jt =-l
I
If
1(-
E(z, z,If.,t) =exp[•
(z
-j
**JL 1+ ~
A(z, z ) di
o (2.3.18)
~
n
'If
n
(n)
=z Q
(z, z)
.,.,(z, z ) ,
then the integral operator
!
(2.3.19)
E(z,
z~ t) f(z(1-t 2)/2) (1_t 2)-1/2
dt
. . sl
can also be written in the form:
z*
exp[ -
~
(2.3.20)
~
f A (z, t) dz+]f g(z) + Lr(2n+1J/ 22n r (n+1) I -1 .
ti
n=l
(n) ',.. . Q (z, z ) ~
z1
'0' • •
40
\
J
('zn_1 g(z/ll,) dz n · .• dZ1 '
!~
- 32 M. Z. v. Krzywoblocki
[l
or
exp . -
-/I.
z
*'
",,'jr
A(z, z ) dz ,.Lg(z) +
L
z
(2.3.21)
•/. 0
Here (n)
Q
(2.3.22) (2n)
p(2) = _ 2 F
(2
2n
(n),l\.
B (n, n+1) Q (z, z ) .
n=l
(j
where P
-(A-
'z
I
if
=)'
(z, z)
(z -~)
n-1
.1
g(C) d~J'
*' p
~
(2n)
i't
(z, z ) dz ,
0
are given in the form:
,
*
l
12n+1) pI2n+2) " _ 2[ p;2n) +DpI2n)+F[ p I2n)dt
12.3.23)
We note that in addition to (2.3.13), f(z/2) function of
g in the form :
(2.3,24)
='IT
-1[ g(O)
n" 1,2, ...
can be represented as a
If/2 dg(z sin2~ ) z sin'l7' d ( z sin2"J .10
+ 2/
The operator of the first kind can be represented in terms of integrals.
'*-
-)f
Let D(z, z ) and F(z, z ) be functions of two complex variables
z
and
z;f,. which are regular in a domain "LL 4 of the space of two complex variables
z and z*', and Ie
g(z) be a function of a complex variable z ~ \2 "A .2 .. (4 which is regular in the domain eLL. (!Jv and 1..,- include the
I."
origin. )
41
- 33 M. Z. v. Krzywoblocki
We introduce :
where
F)/
~» =
dz .. dZ*Il dz 1
. dz
-v-
1\= F{z», zVi-1 ) ,
dz
j+1
dz. dz. J J-1
= D (z '1>-1 '
D )/-1
l+
z
-J+'-1
J\):: J~{g), we denote the some of the
and by
T (F», D\1_1' " ' , F 1 ; g) and
>\
j+1
Di
»-1
2
. dZ 1 ) , etc. ,
expressions
where all possible combinations of F,..
occur, except those for which we have D 1 in the last place.
Then (2. 3. 26)
is a solution of the equation V (z, 0) = g{z) ,
(2. 3. 27)
L (V) = 0 which has the property that 1t:
V{O, z) = g {OJ •
Various properties of the integral operator of the first kind are discussed
in
(4].
It may be valuable to present a solution of the differential equa-
tion:
f1 2 V+F{r 2 )V=0,
(2.3.8 ) where
2 is an entire function of F(t)
In the case
r 2 = x 2 + Y2 = z z*.
of the differential equation (2.3.28) , the generating function
E(z, z~, t) of the integral operator of the first kind is a real function
42
- 34 M. Z. v.Krzywoblocki
of r 2 = z z"'" and t . The generating function
L 00
(2.3.29)
E (t 2 , t) = 1 +
t 2n Q(2n) (!2)
,
n=1 where
Q(2n) is provided by :
(2.3.30)
(2.3.31)
is thus real and we may speak of the conjugate solutions of Eq. (2,3,28) whose developments at the origin are given by :
1
1
('.3.3') V { E(r'. t)Re [ flu)] (l-t ') -1/'dt • RelLE (t'. t)f(U)(I-t') -1/ 'dt}.
(2.3.33)
where
II
w=L1E(1'2,t)Im~f(U~ (l_t 2)-1/2dt = Iml1 E(1'2,t)f(U)(1_t2)-1/2dt}, (1
2
u = z ( 1 - t ) /2 .
A real solution, regular at the origin, can be written in the form: 00
(2.3.34)
V=
~[
an J(n)(l')
cos'1\.~
- bn J(n) (t)
sin"t\~J
while its conjugate is given by: (2.3.35)
~"(n)
w=\
Ib J tr=01.. n
(r)cosn~+a J n
43
(n)' (l'lSinn({JJ, I
'
- 35 M. Z. v. Krzywoblocki
where (2.3.36)
Integral operators can be of an expontial type like :
k m
(2.3.37)
E
= exp
-
Q,
jt-
-
Q - Q (z, z ,t) -
~
(z, z.J(-,., ) t .
,"'" Let u(z,
*
z) be the solution of Eq. (2.3.9) obtained by applying a gene-
rating function of the the form (2.3.37) to the function f(z) n
= 1,
2, .... Then the function U(zl' z2)
= u(z, l')
= zn,
(where z
= zl +iz 2,
z = zl - iz 2) satisfies for any fixed value of z2 an ordinary linear differential equation (in the variable z 1) : k
(2,3.38)
~
Br (z
1
r=O The order
ct"U = 0,
,z
) --
2
d zl' 1
(B k = 1 ) .
k of Eq. (2.3.38) is independent of the value of n appearing
in the function
f(z) = zn
,and depends only on the degree
m of
Q
in (2.3.37) , It is always possible to determine an equation (2.3.38) whose order is at most Eichler [14]
m+l.
considers another type of a differential equation, na-
mely, (2.3.39)
~2 ~ + N (x)
r
0,
where (2.3.40)
N (x)
The solutions (2.3.41)
~
of
f(,) -
= Co + C1 x + C2 x
2
+ , .. .
Eq. (2. 3. 39) are generated by integral operators:
f.'
S (x, y,
~ ) 1(0 d ~, 44
, . x + iy .
- 36 M. Z. v.
Krzywoblocki
satisfies:
where
s
(2.3.42)
xx
+S
yy
N (x) S = 0, S (x, y, z) x
+ is (x, y, z) = N(x)/2 . y
In analogy to Eq. (2.3. 17) ~I (z, z) can be represented in the form:
(2.3.43)
+ ... ,
2i
X
(2.3.44)
PI (x) '(1/
N(x)dx t
Kl' P2(x) • (1/ 2J.X(Pl' tN(x) PI (x))dx t K2 • "
where ~ n are integration constants (ascending series) . e 2 (z,
z,
g) also can be written in the form :
(descending series) where the
q (x) are connected by the recurrence n
formulae: (2.3.46)
q"
+ Nqo = 0 , q\ + Nq1 = - 2qb ' . . .
We may discuss briefly a certain class of fourth - order equations which will indicate how how the methods employed previously are extended.
Consider equations of the form:
(2.3.47) where
L(U) = U
'" ~+ MU + LU "f+ NU'II-"'+ AU zzz z zz zz z z z
+ BU
~+
z
CU = 0,
M, L, N, A, B, C are entire functions of the complex variables
z, z*. When written in terms of the variables : x = (z+z*:)/2 , y=(z-tj/2i, (which become real if the variable z is replaced of z ) the equation assumes the form:
45
by
z,
the conjugate
- 37 M. Z. v. Krzywoblocki
fl/J. U + aUxx + 2bUxy+ cUyy+ dU x+ c Uy+ fU =
(2.3.48) where
a, b, c, d, e,
°,
and f, are simply related to the coefficients of
Eq. (2.3.47) . There exist four functions
E(kr)(z, z*', t) , k = I, II,
which are defined for sufficiently small values, say \z\ \z'
< J2' and for
It I ~ 1, possessing the following property: if fr(C)
r= 1, 2, are any analytic functions of
and g" (~)
r= 1,2,
~
defined and regular
in a neighborhood of the origin, then U (z,
l)
=
L2 11 [E (Ir)(z, ttl fr (z(l_t 2) /2) +
n=1 -1 (2.3.49)
+E
(1If1)
>to 2 ~ 2 -1/2 (z, z ,t) gr (z· (l-t )/2 ~ (l-t ) dt,
is a solution of Eq. (2.3,47) . Coversely, if U(z, z*) is a solution of (2.3.47) defined in a neighborhood of the origin z =z* =
°
,then U can
be represented in the form (2.3.49) by meal'J.8 of suitably chosen functions fj-t
and
g""
p=
1, 2 . The functions
E(kj4)(z,
t,
t ) introduced above
have the property that : (Il)
(2.3,50)
E
(2,3, 51)
~\~) z
*' _
(12) _ (112) .... _ _ (Ill) (z, 0, t) - E (0, z ,t) - 1, E (z, 0, t) -E (0, z , t)-O, (z, 0, t) = E~II 1)(0, ...
Each of the four functions
z~
t) = 0, E(I!) (z, 0, t)=E(1I 2)(0, z z
z~ t)= 1 .
E (kfl) is required to satisfy the following
partial differential equation: L (E):z-l t -l(l_t 2)[E ....... +ME +LE +/2+AE /2] 1 zz z t tz tz t
46
+
- 38 M. Z. v. Krzywoblocki
- z-1 t -2[E
~+ ME
it
zz z
z
+ LE */2 + AE /2Jz
4 [ E ~ "'1-+ ME ] -(3/4) z -2 t -3 (1-t) tz z t
+(3/4)z -2 t -4 [ EIt"'l-+ME-I + L(E) =0, zz J
(2.3.52) where
+
L
is the operator defined
by
eq. (2.3.47) .
There were developped integral operators which transform functions of two variables into solutions of certain classes of partial differential equations in three variables. We consider a partial differential equation of the form : (2.3.53) We shall find an integral operator generating solutions of Eq. (2. 3. 53). Let H(t, 't) satisfy the equation: (2.3.54) for
2 (1- 't') Ht't' -
1'1'1"
't
-1
1 and 0 ~:r
2
(1+ 1') H/ l"t(Hl"l"+ 2 t
< 1"0
-1
H1" + BH)
= 0,
(where l"0 is any positive constant) and
where (2.3.55) Suppose
2 2
B= - (3/2) A - I'A /2 - r A /4 + C •
t
H / t't is continuous at ~ t
Let (2.3.56)
E('t,'t)
= exp (-2
-1
=r =0
.
(r
Jo A I'dt) H (r,l:) ,
and let f (w, ~) be analytic in the complex variables w, ~ for 2 W(U (0) and I~k 1 . Then the function ,\,(X) defined by
(2.3.57)
'Y(X) = (2'l1if 1{
(i
E(t, "C)
~(1:1 )t~-1
47
f(U(1-"C\~) d't~-ld~ ,
- 39 M. Z. v. Krzywoblocki
satisfies Eq. (2.3.53) in a neighborhood of the origin, where u =x
+ (iy + z)
/2 . ~ + (iy -
z)
/2 .
~
-1
.
A solution of Eq. (2.3.53) may be given by a series expansion. We shall say that function
g(B,~),
0 ~B~'lr,
0~f~2"lr
satisfies condition
L if it can be expanded into a uniformly convergent series of Legendre functions:
00
g(B,f) = r[A P (cos B) n=O no n,o
+
n
(2.3.58)
+ '[ (A m =1
nm
cos m f+ B
nm
sin'l1\,(g) P I
n, n
(cos B)J '
The coefficients A ,B are expressible as certain integrals which nm nm can be obtained by taking account of the orthogonality properties of the terms of the above series . Let
S be the spherical surface
r =
y and
~
the interior of S.
Suppose that there exists a positive functions A(r, B,
~
, such that every function
in':) and is continuous
Suppose further that
in
~
'I' (r,
) +S
(p' B ,
~
B,
f ) which
satisfies Eq. (2, 3. 53)
also satisfies the inequality :
) satisfies condition
L. Then
r can
be expanded into the following series , uniformly convergent in every compact subset of
~ 00
*" (f nI *' (Y)) -1 [ AnoPn,o (cos B) + f (r, B, ~) = k=o t n In("t) n "\
n
(2.3.60)
+ [ ( A nm cos mCl+ B sin m£D) P (cos B)] , ! nm I n,m m=l
48
- 40 M. Z.v. Krzywoblocki
1 1
where ~
(2.3,61)
I n (t)
=
2 n
E(l', 1') (1-'1:)
d"C,
-1 Bergman considers a partial
A3 ~ + F
(2.3.62)
(y, z) ~
differential equation of the form:
= O. ,If-
We introduce the variables: and express the function of Z and Z
'* ; we
X = x,
Z = (z+iy) /2,
Z = -(z-iy)/2 ,
F(y, z) appearing in Eq. (2. 3. 62) as a function
also use the symbol
F for this new function.
The equation (2.3.62) then assumes the form: (2.3.63)
'I' XX - 'Y zz it-
r
.lr
+ F( Z,
Z)
=0
We proceed to obtain particular solutions of (2.3.63) which are polynomials in X, as follows, Let
-lzt
(2.3.64)
+F
and let the polynomials
P
(2.3.65)
Let the functions (2.3.66)
-
~ Z~
(2.3.6'7)
N
== ( k-lI)' (
. ';
k-\)
flo
'I>
) Z
k=O, 1, 2,' . "
equation:
N-k+~ fl>
Z,
2N; \)=k, k-2, . . . , k-2
[~],
~
(Z. Z ) satisfy the equations :
p(N-l,k,k)
-
rr(N,k,k) it-
ZZ
P (N-l,k,\») N
N
- 'lz
(Z, Z ) be any solution of the
p(N, k/l!) be defined as follows:
IT (N, k,d.)
NN
~
~
¥ =0
(N k k 2») , ,-
N=O, 1,2"
~
-
rZ
+ ('ol+2)('ol+1)1T(N, k, ,,+2)_
49
+
F 1T(N,K,K) - 0
-
p(N-l, k-2, \»)
TT~z!'V)
+
+ F 1T(N, k,» )=0, ))
- 41 M. Z. v. Krzywoblocki
Then one finds by direct computation that the functions :
'I' (N, k)(X, Z, Z) = 1 (2'ITi)-I[ uN ~ -(N-k)~-1 de + ~~I=1 [k/2J
+;-
(2.3.68)
IT(N,k,k-2))) (Z,Z) xk-2~
»=0 ill
+X +Z
u = Z,
(where
~
-1
) are (complex) solutions of Eq. (2. 3.63) .
The following equations and systems are treated by Bergman in [4]: (2.3.69) (i)
1II 1x
+ 11.1 + I
yy
11/ I zz
+ F(y, z) ltJ = 0; T
(2.3.70) (ii) in the domain of the three-dimensional Riemannian space; (iii) system of equations :
()2'Y _ ~ ()}\: -F(z,z t/ z l zl 1 1
(2.3.71)
"\
where zl' z~ ,z2' z;, are independent complex variables and
F, G
are entire functions of the indicated variables ; (iv)
equations of mixed type : a special case
: 00
M(\Ij)
(2.3.72)
I
=\IJ
I
xx
+ I(x)
\IJ 1
yy
= 0, I(x) = \' a (-xt,a 1>O .... ~1 n
(v)
initial value problem in the large ;
(vi)
generalized Cauchy-Riemann equations
(vii)
the differential equation:
(2.3.73)
6
2
0/+
N(x)'I'=O,
with a new type of singularity of N ; (viii)
an integral operator
fOIl
equations with non-analytic coefficients.
50
- 42 M. Z. v. Krzywoblocki
2.4. Generalization of the Hodograph Technique to Diabatic Flow. This was done in
[ 60J
the continuity equation with tained for (2.4.1)
f
Starting with the Euler equation and dQ
f
0, Qf 0,
the following results was ob-
in the hodograph plane : a
\II
1 I'qq
~ (J I ' (J
-c'l' -b\ll +a \If -b 1 'q e H'IH) I, q I' q 1,
-c
III =0
1, (J II q
,
where
2.5. Generalization of the Hodograph Technique to Magneto-Gas Dynamics. It is assumed that a diabatic flow is inviscid, non-heat conducting,
obeying the perfect gas law, heat being added by means of sources distributed in the flow domain with an electro-magnetic field. Starting with the equations of continuity, momentum, energy and the pressure-density-entropy relation, accompanied by Maxwell's equations, we obtain (see details in
):
[ 61J
where
(2.5.2)
+ ('t-
1) c
-2
2 -1
Q, q - (pc)
51
F e 'q
]
,
- 43 M. Z. v. Krzywoblocki
b
= (pq)
(2.5.3)
-1 { g(sin
- c -2[ q - g(sin
e - cos e ) e+ cos e U[(t- 1) Q, e- fIFe, eJ},
(2.5.4)
c=
(2.5,5)
d = y -1[ q _ g (sine + cos eu '
(2.5.6)
e = q J1 6)
(pq)-1 g(sin
L!'"
e - cos e) ;
q - g ( sin
2
2-2
M =q c
,
e+ cos e)1 ,g, -g 1 x 'y
= w (x, y) ,
is the electromagnetic force potential and Q contains the e Joule heat. The Maxwell equations and Ohm's law are transformed dire-
and
F
ctly
into the hodograph plane by means of a simple transformation of
derivatives. In the work in question the author derived the canonical forms in the hodograph plane, discussed the general approach to solving equations and proposed a detailed table-procedure for particular steps of a solution. In particular, an inverse problem was solved. In it, in the physical (x, y) - plane the streamlines map into a pattern of concentric circles about the or igin .
2.6
Transonic Flow. Due to the importance of the transonic regime in the flow around
a blunt
body (Apollo shape), as furnishing the initial data for the
supersonic domain (characteristics), the transonic flow is treated in more details . Stark
[71J
uses the Bergman integral operator to generate fami-
lies of flow patterns which yield the transonic flow pattern(the solution due to Ringleb
[68J
is a special case ) . In order to obtain a flow
52
- 44 M. Z. v. Krzywoblocki
pattern of a certain shape by this method it is necessary to determine the specific associate functions of the operator for the subsonic region and for the supersonic region. First the Chaplygin equation was transformed into
=0
(2.6.1)
J
where M. Jf =
(2.6.2)
U.ingA."
J9q-1 d q,
defined by
I()4.)
~ +lA- f.tl('tl]
9
= -2 (1- M2) •
1/2 d1: we obtain equatlollll
of the form: (2.6.3)
~).).+'I' 99 + 4 N ~). = 0,
(2.6.4)
uJ -Ill
1M 199
+ 4N VI
J
0, for M > 1 (supersonic).
1M
Nl! N(A) , N1:: N1 (A)
Here
for M < 1 (subsonic)
and
(2.6. 5a)
~= (1/2) log[ (hT)(1+T) -1 {(1+hT) (1_hT)-1 }
(2.6.5b)
T= (1_M2)1/2 ,
1/hJ
- arctan (T), - -T = (M 2- 1) 1/2 ,h =[ (~- 1)(t+ 1).1J*. A=h -1 arctan (hT) Bergman integral operator for the subsonic region is given ~
P(f)~
1
=1m P
(2.6.7)
(2.6.6)
by:
(f),
(A 1E(1) +A 2(Z(1_t 2 )/2)2/3 E (2)) f(Z(1-t 2)/2J (1_t 2 )-1/2 dt , E(k)=Hl(k), E*
t
q(n,k)/(_t 2Z)n-(1/2)+(2k/3), k=1,2.
n=O Here P(f) is the Bergman integral operator of the second kind,
53
- 45 M . Z. v. Krzywoblocki
z='X+i8
, H = H (A) depends upon
N, and AI' A2 are complex
constants. C is a sUitably chosen curve in the complex t - plane connecting the points t= - 1 and t= 1 . In general C is the path of the upper
[3]). The functions q (n, k) are
half of the unit circle, (see (2) and
dependent upon the differential equation:they satisfy the system of ordinary differential equations:
q (0, k)+ 4 F q( 0, k)= 0
n
'
2(n+2k/ 3) q~' k) + q}. ~n+1, k) + 4F q (n+1, k) = 0 , where 2 F = - N - Nl,/2,
n = 1,2,' . "
k = 1, 2.
Correspondingly, in the supersonic region, we have
--
z = i(I\+ 8),
depends on N1" A1 and A2 are complex constants,
and H = H
(")
C is taken
along the upper half of the unit circle and the functions
q(n, k)
are certain functions which depend only on the differential
equation and they satisfy the system of ordinary differential equations, _ -(O,k)
+4 F
-q (O,k) = 0
1
q/\/I
2 i (n+2k/ 3)
qt,
'
k)+ q~n/l+1, k) _ 4F 1q(n+l, k) = 0 ,
n = 0,1,2, . . . . For the supersonic region, the integral operator is given by : ~ =
1m
P ( f) ,
J,
i'lij '" IA PI + A,[ '11-"I!'J '!3 ,,1 2111['11-,'1!2J 11-,'1- 1!'dL. "
54
- 46 M, Z, v, Krzywoblocki
(2,6,8)
E(k)=HE (k) , E*(k)= fq(n,k)/(_t2zt-1/2+(2k/3), k=l,2, n=O
We associate with every stream function '\' the functions: (269) ' ,
AX(9):'lim_'I'
~
M~ 1
when approaching from (2,6,10)
the subsonic region, and
-y 19)= lim
Ai
Xt 9)=lim_].iM M-+1 oM
'2
-M .... l+
~
X(9)=-M~1 lim + 1J ~M
'2
when approaching from the supersonic region . Stark proposes two methods for obtaining the associate functions. In method
I, there are used the expressions:
(2.6.11)
"V 1(8) = a (1) + fL')I)cos n (8-8 ) + b(l)sin n(Q A. 0 rl=in 0 n
(2.6.12)
X.
(8) = a(2) + 2
0
Qo~
~
,
f~1 la(2)cos n (8-8 ) + b(2)sin n(Q - Q )1. non o~
The associate function in the subsonic region of the desired stream function is given by :
(2.6.13) with respect to the operator (2.6.7) such that these associate functions satisfy the limit relations:
55
- 47 -
M. Z. v. Krzywoblocki
(2.6.14)
cos [p(8-8 0 )
-
'If: J'
lim l'1'1\.P (G(,If) I M-Ht
:: 0
A corresponding treatment in the supersonic case gives the associate function in the supersonic region. In method II, one considers the functions Xl (8),
X2(8)
expanded
in Taylor series instead of the Fourier series in Eqs. (2.6.11) and (2.6.12) used in method 1. Correspondingly different associate functions are provided which permit further treatment similar to the considerations of method I . We compute two stream functions 'X)8) , (2.6.16)
X2(8)
are of the form:
X1(8) =
a sin 8,
X. 2(8)
f
whose
conresponding
= b sin 8 ,
where a, and b are real constants . We follow streamlines which cross the sonic line , To pass from the hodograph (q,8) -paIne to the physical (x, y)-plane involves integrating equations:
l 2 2 -1 -1 dx = [ (M -l)(pq) cos 8· ~8 -(pq) sin 8 'rqJ dq (2.6.17)
-i
+[ 2
cos 8'
'P q - (y q) -1 sin
8·
f 8 1j d 8,
1dq
2 2 -1 -1 dy= [ (M-1)(Pq) sin8'~8+(pq) cos8'f q (2.6,18)
+[ f1 sin 8 'lI'q + (pq)-l cos
56
+
8'~ 8
1
d8.
+
- 48 M. Z. v. Krzywoblocki
Once we know the relation between the density can integrate Eqs. (2.6.17)
p and
velocity
q, we
and (2.6.18) to yield the sonic line in
the physical (x, y)-plane.
r defined by corresponding
For stream functions
'X 1(e),
x2(e)
given in (2.6.16), the sonic line in the physical (x, y)-plane is given
e by
in terms of the parameter
:
(2.6.19)
x = (1/2) (6/5)3[ a-(6b /5)J cos 2
(2. B. 20)
y = (B/5)3
e,
t (1/2)[at (Bb/5)] (e -'If/2) (1/4)la- (6b /5)J sin
t
2e} ,
provided we take the origin in the physical plane as the image of the point q=(5/6)1/2,
e ='tr/2
The value of
is calculated
q
on the sonic line in the hodograph plane. M2 = q 2/ [1 -
by
(¥ - 1) q 2/2 J with
t=1.4.
Example
(I) : sonic line in physical plane is circular.
If we consider the stream function
1
defined
by
(2. 6. 16)with
a, b related according to : (2.6.21)
a
+ (6b / 5) = 0,
then (2.6.19) and (2.6.20) yield for parametric equations of the sonic line in the physical plane : (2.6.22)
3
x = a(6/5) cos
Elimination of (2.6.23)
x
e
2
y = (1/2)a(6/5) sin 2 e
e,
gives:
223 + Y - a(6/5) x = 0 ,
57
.
- 49 M. Z. v. Krzywoblocki
which gives a circle through the origin with center on x-axis . If we set a = c (6/5)1/2, then (2.6.16) under the restriction (2.6.21) ,
becomes: (2.6.24)
X 1(9)
= C(6/5)1/2 sin 9,
X2(9)
= _C(5/6)1/2 sin 9,
which define the Ringleb solution:
'r = (c/q) sin 9 ,
(2.6.25)
of Chaplygin's equation. There are computed grid values and stream lines for the stream function defined by (2.6.24) with the explicit value 1/6 for the constant C=11(5/18) ~ 2,538. The associate function
fA
(~)
with respect to he (subsonic)
Bergman operator (2.6.7) is given by: 00
(2.6.26)
where
fA (
~ ) ~ ~ 7/ 61\~/r (2m+5/3) +
L/r
(2m+7/3) ] (2
l; = T + C3T 3 + c5 T5 + ... , (see [7]
(2.6.27) K = Z(6/5)1/2
r
(2/3), L = -i(1l/50)(2/15)1/6
), and
r
(1/3),
provided we take for AI' A2 the specific values given by = exp [ - 1f i( 1/2
+ k/6)l
~ )2m).
' k= 1, 2 . The associate function
Ak =
~( , )
with respect to the (supersonic) Bergman operator (2.6.8) is
where K, L are the constants defined by (2.6.27) . For the given coordinates (Mach number M and angle 9) of points in the hodograph plane, the stream functions are evaluated by applying the
58
- 50 M. Z. v. Krzywoblocki
operators in which the associate functions are calculated by Burrough B-5000 truncated (after the first ten non-zero terms). and tabulated in
[71J . Also
the Ringleb stream functions
l'
= (c/ q) sin
9 are tabula-
ted for the purpose of an investigation of the accuracy of the numerical operator methods discussed here . Example (II): sonic line in physical plane is not circular. We consider the stream function defined by the corresponding limit functions : (2.6.29) where
"X 1(8) = c sin c is given
8.
X2(9) = (-c/2) sin 8 •
='rr(5/18)1/6~
by c
2.538.
According to Eqs. (2.6.19) and (2.6.20). the sonic line in the physical plane for stream function is given by the parametric equations: (2.6.30)
x=(4c/5) (6/5)3 cos 2e. y=(c/5) (6/5)3 [ (9- 'If/2) +2 sin 29J.
The associate function
fB(~)
fB (C) is furnished by (2.6.13) :
= cFil)(O; ~) - (c/2) Gil)(O; ~) 00
= ~7/6
~J[X/f (2m + 5/3) +
)( = 2 f(2/3).
L = i(1/20) (18/5)2/3 r
(2.6.31) where (2.6.32)
the associate function
'"
fB (~) is given by
59
(1/3).
- 51 -
M. Z. v. Krzywoblocki 00
f (~) = ~7/6 ~o B -ell r
(2.6.33)
where
X ,rAJ.,
are given
by
t[ 'X Ir
(2m + 5/3) +
(2m+7 13)] (2.6.32).
To determine streamlines in the hodograph plane for Expample (II) , an ALGOL procedure was written which yields stream function value for given coordinates q, q and uses
e . This
procedure first computes M from
fB(~) in (2.6.7) if M < 1 , or uses ~(~) in (2.6.8)
if M > 1, (for M:: 1, stream function equals 'Xl (e) cedure, interpolations to find out in the q direction factor
10
-3
at
the path of
f
). Using this pro-
= constant were carried
5 degree intervals of
e using
tolerance
. To obtain the image in the physical plane, Eqs. (2. 6.17)
and (2.6.18) were integrated using the trapezoid rule. The path of integration was along q = (5/6) 11 2 from
e = 40 degrees,
e = 90
degrees clockwise to
then along de = 0 to a point on the stream line under
consideration, and then along this stream line to the point whose image is sought in the physical plane.
2.7. Gilbert's Technique. A lot of work on integral operators was done by [43 - 46J and Gilbert L1, 17 - 31, 32, 33J
Kreyszig
Below, we shall
briefly discuss some recent results obtained by Gilbert. His technique was used in the development of the generalized axially symmetric potential theory (abbreviated GASPT) and the GASPT differential operator is of the form :
60
- 52 M. Z. v. Krzywoblocki
For illustrative purposes we shall briefly present Gilbert's technique applied to a generalized axially symmetric potential (see
[17]
) :
,2
. . .2
1
L[U)=~ +~ +1<9OZ OJ
(2.7.1)
of the form
'" u
.;
=0.
~.P
Solutions to Eq. (2. 7.1) may be constructed by an operator which transforms functions of a single complex variable f( rt ) into axially symmetric potentials
u(r, 9), (r 2 = z 2
+
1 -1
u (1' 9) = A (f, ~, X ): n 0
(2.7.2)
l '
9 = cos z/ r) :
-1 k-1 1 f(rt) (u-U) u· du,
Tl
11= Z+i y(u+u· 1)/2
where
E>
0
is sufficiently small,
IX-Xol
l
<£,
X:(z,P), Xo::(zo,
po"
is a differentiable arc in the u-pla-
ne; here the arc is taken as the upper half of a unit circle.
If
f(Q") has a Taylor expansion convergent in some neighborhood of
tJ = 0 : (2.7.3)
f (cf)
L
=
an o-n ,
n=O
then the operator
A ( f,
n
L,
X) defines an axially symmetric poten0
tial for sufficiently smalll- : (2.7.4)
This may be seen by considering the identity: "I1'
(2.7.5)
n ). 1-2~ -1 2 t Cn(cos 9)=2 (nD r(2)+n) (r(),) )
J
n
[z+ifcos~J (sinf)
2A-l
o
It is convenient to continue the arguments of u(!o, 9) to complex values, if the followings are introduced:
t' = (z 2 +P2) 1/2,
61
~ = z1' -1
,
dr.
- 53 M. Z. v. Krzywoblocki
f
which reduces to Let
u(r, B)
=cos
be an a xially symmetric potential regular at the ori-
1
gin:
-1
(2.7.6)
z and p.
B for real
u(l', 9) = 1
,.1 k-l -1 f(o') (u - u) u du, 00
where
lui
=
1m u? 0,
1,
f(O')
=[
an (t'n ,
f(a') may be generated by the following integral transform:
then
. (1-2~(),l"
(2.7.7)
where
W(t'.
S ) ==
u(r, B) ,and
-1
2.2 .. k/2 .. 1
+0' t )
d~,
C is the real axis.
Let K
(O"r-~ E ) = (2i)-k • (1- 2 ~ tr
(2.7.8)
then
-1
k 1r- 1 (1-
r
1 +0' 2 1'- 2 ) - k/ 2-1 ,
K (0"1" ,e)
is analytic in
~
) is analytic in
is harmonic, W(t.
$ =± 1.
f 2)(k-1)/2(1_Q"2 t: 2) .
S S
and also, since utI', B) except for the points
Thus the integral involved in the expression for f (0")
is a Cauchy-integral and the restriction of the integration path to the real axis is not necessary. Gilbert
applies the similar technique to a generali-
Zed axially symmetric wave equation of the form (GASWE) :
62
M. Z. v. Krzywoblocki
,"\2u
",\2
1 "\ u
u + _v_u_ + 2 \l y~ x2
(2.7.9)
vi
Here solutions to (2.7.9) with
k
2 + k u
lJ
~
\J
0
\~y
,
1 = rea.
y
> 0 will be investigated by using
Bergman's operator method. Henrici [38J gave an integral representation for a complete system of solutions for (2.7.9) ; namely, for 2 2 2 -1 r = x + y , 9 = cos xl r , we have:
l ))
[kr (cos 9 + i sin 9 cos ~ ) ] -'JJ l>+n [k r (cos 9 + i sin 9 cos
)-'V+1 J (k' ' (k rsm(Jsm~ y_1 rsm(Jsm~
(2.7. 10 ). for
1r
[j
>
0 and n
•
= 0,
[j
•
)
(.
smlf
)2~1_1
f)] .
d~,
1, 2,' . • .
It is known that an arbitrary solution to (2.7.9), regular about the
origin may be expressed as a Bessel-Gegenbauer series (see,
[38J ):
-\}
w (r, 9) =1'(2);) (k r) . (2.7.11)
. fL annl' rh2 \.l+ ~
n)] -1 J"
v+n
(kr) CV (cos 9),
n
n=O and an arbitrary analytic function regular about the origin may be expressed as a Neumann series
(2.7.12)
f (0') =".
-" n-~
(16
a
i
J + (0') . n V n
It may be shown for r sufficiently small that the class of analytic
63
- 55 M. Z. v. Krzywoblocki
functions (2.7.12) may be mapped onto the class of functions (2.7.11) by an operator of the form :
Xk..[fJ
(2.7.13)
hk
"l.:' t
J
~-1 [k y (? - ~ -1) / (2 i)-'] , -1
. f(krt)(>-S)~
d~,
~
w(r,8)::u(x, y)= J{k[f] , ()=x+(i/2)y(~+~
(2.7,14)
t
where
-1
-1
),
is a differentiable arc in the w-plane , defined by :
1v=. .sIS=e .~ ;O~ri~1rJ'l h~=-(iky) 1- \)/2·rCY+ 1/ 2 )[r(l/2)] -1 . I
(
1
I
it is useful for us to consider replacing the contour 1., by an arbitrary smooth contour
t
which has + 1 and - 1 as end points .
Cauchy's theorem tells us that both representations are the same provided that we do not pas I> over a singularity of the integrand in the process of deforming
£'
into
1v
Gilbert and Howard consider the problem of obtaining an inverse operator of the form :
(2.7.15)
'JC~v [UJ{
K (0', r, g)
u (r f ' r(l-
~
21'/2) d
S~ f(k~),
where a, b are appropriately selected limits of integration and 2 2 1/2 r = (x +y) , $= x/ r. They take a = - 1, b = + 1 , and try to determine r(2)>)
K (IS, r
s)
such that:
(kr)-t+1K(~' q
Itt
an n!(r(n+2 ») )-\+n (kr)
64
c~ (S)} d~
- 56 M. Z. v. Krzywoblocki
= (kcr'(,)
(2.7.16)
f.
a J"
n=O
v+n
n
(ko').
This may be done formally by recalling the orthogonality relation for the Gebenbauer polynomials
[16]
:
( = On m
(2.7.17)
2
1-2 ~ [ 'I1T(n+2Y)
r 2(~) (V+ n) ]
the right-hand side may be written as :
by use of the Legendre duplication formula. Thus , if we define : K (~, r, ~ ):: a\) (1-
S2)'i)-1/2 (2 2~-1 )-1
(kr)
v
-\>
(kO') (l>+ m) .
(2.7.18)
we have: (kll')-\)a
(2;7.19)
1'1'1\
~+m
(kO') =r(2'Y) (kr)-» .
j
Km(O" r, ~ ) u
J
+l
-1
Consequently, if we define:
65
21/2
(rS' r (1- ~)
)d
S.
-1
;
- 57 M. Z. v. Krzywoblocki
K{cr, r,O:::
fL
K (tl', r, m
s)
=
m=O V = a.)r/cr) (l-
(2.7.20)
52)v-1/2
~ (v+m) J ,1
-1
'\l+m
(k(j'){J (kr)) )l+m
i=o then we get formally :
(2.7.21)
r(2~){kr)
-1[+1
21/2
K{~ r,;) u{r5,r{1- ~)
)dS= f{kO").
-1 K (~ r , ~ )
In order to show that the series expansion for
converges to a holomorphic function of the two complex variables ~
's
and that the representation (2.7.21) is valid, one considers
the asymptotic behavior of its Gegenbauer coefficients as
m~
00.
For the proof of this statement and the location of singularities and the growth of solutions, the reader is referred to [29]. [32J consider the gene~
As another example, Gilbert and Howard
ralized bi-axially symmetric Helmholtz equation( GBSHE) in the form:
r'
> 0, k>O.
\l
We shall consider these solutions of the GBSHE which are of the class C2 in some neighborhood of the origin and even in x and in y . Then we must have, for u{x, y) , a solution (2.7.23)
u
x
=0
on x = 0 ,
Henrici [39] furnishes
u y
=0
: on
y
=0
.
a complete set of solutions, regular about
the origin, as this particular class of solutions (denoted by S) cited
66
.~
C1'1\
j
- 58 -
M. Z. v. Krzywoblocki
below (2.7.22) ; ('V -1/2 ~ -1/2) -f-v f (r t) = P , (t)(kr) J (kr) n=O 1 ". n ' n 1'1+Y+2n J " ,
(2.7.24)
t = cos 20 , r2 =
where
Polynomials (see
i
[161
+
i
and P (cI., n
~ ) stands
for the Jacobi
);
=
p(V- 1/ 2, }1-1/2)(t) (kr~v- J (kr) f (r t) = n } jAH+2n - n '
•r
(~+n+1/ 2{r (~f (1/ 2)r (n+I)] -Ilr(k~)-~ -" J~ +~ +2n (k<1)l
.t
(O'/xt
T2 (~,
O'=X+iyCosP,
1.~,'V;$ 1.
x=rcosO,
1 2 2 . 2j, r~=_ky sID t /4,
",1)( sin+)2l>.1
y=rsinO, gl= ..
dfJ,
y2sin2~/(4Xa'),
k"''V'>OJ,.=O, 1, 2,,,,,
An arbitrary solution of the class S may be represented in a series
. P (V-1/2,/"\ -1/2)(COS 20) J (kr) n l'+})+2n ,
(2.7.25) and an
even analytic function regular about the origin may be expres-
sed as
Z 00
(2.7.26)(
a 2n Jfi+'V+ 2n (0") •
Hence for r sufficiently small it follows that the class of analytic functions (2.7.26) is mapped onto the class of solutions (2.7.25) by an operator of the form;
67
- 59 -
M. Z. v. Krzywoblocki ~
w(r,e)=u(x,y)~OLfJ=a
1 -1
-1 2'V-l
f(kO")('iJ4(~-~)
.
+1
(2.7.27) where
J. 1 1 -1 'i'2{r, 1-~,\);~, "1)~
. (J= x+iy
(~- ~-ll
/2,
~1 = i(~- ~-1)2/
I
d" (16 xC') ,
It is useful to continue the arguments of the solutions (2.7.22)
to complex values. A cpntinuation of r2
= x2 +
u(x, y) of
i ,
5= cos 2 e = (x 2-i) / (x 2_y2) to complex values allows one to obtain an inverse operator
0 -1 [ u ] which maps the class of solutions
S
back onto the class of analytic function (2.7.26) .
t
To obtain such an operator in the form : (2.7.28)
I( k'1"
a- I [,.]
0
K(a',
r, e)u[r ((1+i)/2)1/2, r((1-~)/~]cH
-1
we try determine (kG') -I'
K
such that, formally:
-> ~ '2n JW
.+
2n (k<)
0
"'-ik.,
(kr)
r, f,
).
This may be done by recalling the orthogonality relation for the Jacobi polynomials :
68
- 60 M. Z. v. Krzywoblocki
1
(v-l/2,;~\-1/2)
+1(1_I=)V-l/2(1+f) 11+1/2 P (V-l/2"M-l/2)(f) P
-1
~
n
>
=1
"
m
2r+~(n+l)+ 1/2f (n+f+l/2)
nm
. (L) d~
.
(2.7.30) Thus, if we define :
E)
Kn(G', r,
2
rH + 2n (kr))
.
(2.7.31)
where
-(~+I»)l / ~+I> 1)n(kr kit) J~+H2n(ka') (J
Pn(~-1/2,)A"1/2)(f)
bn = (2n +JA +») rtn +}I+)))
we have a 2n (kO")
-1
.
(1_E/- 1/ 2(1 +pJl-l/2,
(r(n +/A + 1/2))-1,
-(~+») J ~t\)+ 2n(kO')
f
+1
-
1
Kn (Ir, r,
S).
(2.7.32) Hence
i
-
+1
(2.7.33) f(kO") =
s) u l r((1+g)/2) 1/ 2,r((1-S)/2) 1/ 2
K
W,
=
2-(~\H)(r!d
r,
d~,
-1
where K(
~) ~
. ~(2n+,"+\)
Jr
/'+"(1-
S) P-1/2 (1+~ /1.-1/2.
(n+)) +1") (r(n+flt~))
-1
;1 -1 ('V- 1/ 2 jA-l/2) J tl +V+ 2n (ko') [ J f+ll+ 2n(kr)J Pn '
To verify that these formal calculations are justified and that
69
(S ). K
- 61 -
M. Z. v. Krzywoblocki
as a function of the complex variables,
(J
,1",
and
$
is
holomorphic, the reader is referred to the proof given by Gilbert and Howard
[32J.
Also the location of the singularities and the
growth of the solutions are given in
[321 . Application
rators in ordinary differential equations was achieved blocki [59J
of integral opeby
v. Krzywo-
.
3. Reduction of Independent Variables. 3. 1. Preliminary Remarka. The idea of the reduction of the number of independent variables in solving the partial differential equations was actually introduced into the field of incompressible fluid dynamics by L. Prandtl in 1904.
In the course of the years the technique was generalized to compressible fluids and to hypersonics. In this case we deal with a viscous , heat-conducting fluid . The main goal of the technique is to deal with ordinary differential equations in place of partial differential equations. The two-point boundary value problem in ordinary differential equations has a rigorous solution in some cases at least . But the two-curve boundary value problem in partial differential equations has up to now in general no rigorous solution. For this reason, ordinary differential equations are often preferable in handling some problems in mathematical physics.
3.2. Early Approaches. Assume a system of partial non-linear differential equations of arbitrary order and degree in n dependent variables
70
f.(x, y) 1
and
- 62 M. Z. v. Krzywoblocki
in two independent variables
[Dj : fl(X,
(3.2.1)
-
(x, y):
y, f" f,
1
,f.
1, X
1
1,
, . . . ) = 0, (i=l, 2,' . " n), Y
with the following boundary conditions superimposed on it :
which show that some functions
f. should fulfill the boundary value
problem only along the curve
{1x( 1) , / 1) whereas the others
J
should fulfill it only along the curve {x(2), y(2)}. We shall add to
[DJ
the system (3.2.4)
one more equation of the form:
11n+1 (x, y ; f(n+1)(x, y)) = 9n+1(x, y;z) ; z = f(n+1)(x, y) ,
f 1 (x, y) is some real function of x and y, n+ satisfying the following conditions : the independent variable y should where the function
be expressable explicitly in terms of x and z; any derivatives of f with respect to x and y must be expressable explicitly in terms n+1 of x and z only. The so obtained system, associated with the original system system
[D] [ D11
is denoted by the symbol is
[ D11
related to the original system
a particular solution of [ D11
. The associated
IDJ in the
is a particular solution of
sense that
[DJ., since
a particular solution of a system of (n+ 1) -differential equations is a particular solution of the reduced system of n - equations. The function f
~
1(x, y)= z -
will be introduced into
71
Q),.,
Pi
(i=1, "', n).
- 63M. Z. v. Krzywoblocki
Then, with the system D
(3.2.5)
2
lDJ '
we associate the system
:~.(x,z;F.(x,z), 1
2
dF· -d1
1
Z
[D2]:
'
d ~i, . . . ) = O,(i=l ''',n), dz
with the boundary conditions:
~(1) (1) (1)
(3.2.6)cUi (x
(3 2 7) ••
(1U~~
dF
I
iI
,z ;~(X,'II,dz\ (1) (1) ,"·)=O,(i,=I,".,j), I x, Z I
ca(2) ( (2) z(2)'F (x(2) z(2)) dF k \ . .. )=0 (k=J'+1 ... n) x, 'k ' 'dz (2) (2)' " '" x ,z ,
IJ k
In this we can associate the first original system with an ordinary differential equation at the end of our chain association. Hence, in particular, let us restrict ourselves to the system of the first
order:
(3.2.8)
[D]: ~1' (x, y;
f. ,f. 1
1, X
,f.
1,
i = 0,
Y
(i=I,"', n),
with the boundary conditions : (3.2.9)
where the functions
t i (1) ,
tk (2)
are of known , given forms in terms
of x, y, f ,etc. A special case of the above boundary conditions may k,x be:
with
d. i '
d. k
being known .
72
- 64 M. Z. v. Krzywoblocki
Let us introduce a parameter function
F(x, z)
u
and let us associate with any
of two independent variables a parametric function
y.(x; u) of one independent variable and of the parameter u ; then the 1
system
[Dl
(3.2.13)
Y.(x;u) =~.(x, Yl(x;u), . ", y (x;u); u),
of (3.2.8) is associated with the system:
1
(i = 1,"', n) ,
n
1
or with the system :
[D~: Yi = fi(x, y 1(x;u), . ",
+
Yn(x;u) ; u)
n
+ y.
(3.2.14)
J~
Ai .(x'Y1'· ", yn;u)y.,
J
(i=l,,,·, n),
J
subject to the following boundary conditions: (3.2.15)
y.(a~l)
; u) = r.(1)
11
(a(~)
;y.(a(I);u),
1111
Y.(a(~) ;u), "', u),
(i=I,"', j),
11
a special case of which will be (3.2.17)
y.(P) ;u) = d..(u) , 1
1
1
(i=l,"', j),
(3.2.18) and the dot over the symbol respect
to x, and
y
denotes
the ordinary derivative with
u is a parameter.
We are entirely working in the real domain and dealing with the system of ordinary, first order differential equations (3.2.14). v. Krzywoblocki
[49J
proves the existence and uniqueness theorems of
the system (3.2.14) first with
(3.2.17) and (3.2.18) and next with
73
- 65 -
M. Z. v. Krzywoblocki
more general boundary conditions by employing the integral system corresponding to (3.2.14), thus obtaining the existence theorem (but not uniqueness) for a system of the original form (3,2.8) , 3.3. Michal's Approach. Michal
considers an
r-parameter continuous transformation
group: Gr'. xi -- fi(x 1J'"
J
xm. } J
Y= ~(y;a1, ... ,
(3.3.1) where the
XiS
J
'"
J
=
a r )- f(x'a) J
J
at) =r(y;a), (i=l,"', m, m>l, r
(i~
t)
1),
and a's are numerical and y is a variable of calss
with n sufficiently large, in a Banach space, SG
.?
C(n)
B. The group
is assumed to possess S ~ 1 functionally independent
Fr~
r
chet differentiable absolute invariants:
where
s< m . It is further assumed that
Fr~chet differentiable, absolute invariant y for all admissable
y and
A B-valued function (xl"
G
r
g(y, xl, ... xm) solvable in
XiS.
Y(X 1,'"
,xm) of the numerical variables
. . ,xm) is called an invariant function
y is the same function of the y = H(x) then
y
possesses a B-valued,
y
as
XiS
under
is of the
Gr if, under Gr
x IS,
i. e., if
= H(x) .
If Band B1 are two reflexive Banach spaces (not necessarily di-
stinct, (3.3.3)
i. e., they may be the same Banach space) and if :
n (x,
y(x)
J
kOx
...
J
"ly -
dx g 74
)
- 66 M. Z. v. Krzywoblocki
is a B -valued
g-th order differential invariant
functions y(x 1, .. " xm ) under the group
of invariant b-valued
G, then r
0
is expressable
in the form: (3.3.4) where the corresponding B-valued functions
(,? i' "',
))
~
s) exist
and are determined by the formula:
m
i
1
(3.3.5) '1>( ~ i(x, "', x ), .. " "ls(x , . ", x
m
))
=
= g(y (xl, .. '/Xm),X l ,'" An outline of the proof of the above statement is given in
,x.~). [ 65] .
Conversely, if Bl and B are two Banach spaces (not necessarily distinct) and if the
s
»(II} i'
... , "Is) is an arbitrary B-valued function of
-invariants "ti' "', 1s ' then the corresponding
B-valued
function y(x 1, ... , xm) determined by Eq. (3.3.5) is an invariant function under
Gr' In addition, the B1- valued
expression (3.3.4) may be expressed as ferential invariant under
a B1-valued q'th order dif-
0 of invariant B-valued functions y(x 1, ... , xm)
G. r
If
(3.3.6)
y(x) is a solution to the partial differential equation: 0 (x, y(x)
'C}y 'h ' "', h
=0 ,
g)
"Ox
where
q'th order differential
0 is defined in the discussion following (3.3.3) and, in addition
y(x) is an invariant function under
G then y(x) is said to be an invar
riant solution of Eq. (3.3.6) . 1 m Hence , an invariant solution y(x , "', x ) of Eq. (3.3.6) deter-
mines, by means of Eq.(3.3.5), a solutiony(1Yl 1 ,"" "ls)to the
75
- 67 -
M. Z. v. Krzywoblocki
corresponding partial differential equation: (3.3.7)
w
(yt,» (,) ,
~)) il<1j "
w("t,'J(1)"") and
where y,
~gy
••
= 0,
''d\g
»(r1j) are defined by (3.3.4) and
(3.3.5) , respectively. To each solution 'J (1) of Eq, (3.3.7) corresponds an invariant solution y(x) of Eq. (3.3. S) . The correspondence is given by (3.3.5). Morgan
[siJ
deals with one-dimensional normed linear
spaces and continuous groups in only one numerical parameter a. 3.4. Examples of the Reduction to Parametric Functions. We restrict ourselves to two independent variables and for illustrative purposes assume a one - parameter continuous group of transformations of the form: _2
where
_
x == - Y
(3.4.1) f.(i 1
=
2 f (a) x :....;. f y
2
-"' 2 '
= 1, 2) are some functions of the numerical parameter a.
Then one must have : (3.4.2)
Y g (x)
= y g (x) ,
where g(x) is some function of x. Suppose, that the function g(x) can be represented in the form
g(f 1(a) ). g(x), then Eq. (3.4.2)
implies:
which may enable one to find out the form of the function consequently that
of g(x)
g(f 1) and
For example, if m, n, p, are real num-
bers , then one may choose :
76
- 68 -
M. Z. v. Krzywohlocki
f =a 1
m
n t mp = 0 ;
p=-nm
-1
;
P
fl= exp (ma) ; f2= exp(n a) ; g = x ; ntmp=O ; p = -nm
-1
fl= exp (ima); f2= exp (ina); g=xP ; ntmp=O; p = -nm -1 ;
(3.4.4)
where in the last proposition only the rea 1 parts Re
f fJ
should be
taken into account. One may construct easily more complicated funcf. ( i = 1, 2) and
tions
1
(3.4.5)
x= ma
+ x; y=yexp
which results in (3.4.6)
g (na); y=yexp (pi)
= yexp (px) ,
p = • n/ m, or :
x = mat x ; Y= yexp (a exp n) ; y = y exp (px) = y exp (px) ,
which results in
p = - m
-1
exp n. The absolute invariants of the group
are: (3.4.7)
~
p
"\ = y
=yx ;
exp (px) .
In a similar way we deal with the dependent variables; thus • as an example one may choose:
For illustrative purposes assume a function
g(x) equal to
r
x,
say . Then it should be : (3.4.9) The procedure is identical with the one, explained above. The absolute invariants of the group (3.4.8) are :
(3.4.10)
go
=
k(
r
x
Hence by virtue of equation :
77
- 69 M. Z. v. Krzywoblocki 1 m z( (x , . . . ,x )
(3.4.11)
= FJ'
(~),
the invariant solutions of equation
1 ~(x "
(3.4.12)
. . ,x
opy
m
n
·,~)=O,
; y l' .
~(x
f
must be of the form :
x
(3.4.13) with
~
given
-r
by Eq. (3. 4. 7) . Contrary to Morgan's approach, there
are no conditions superimposed upon the differential form or system. Hence the function
g(x) = x -r
a loss of generality
in Eq. (3.4.13)
one may assume it to be equal to unity .
Yo are equal to
Moreover, x is a constant and the functions
Fo
(x = const.
,~
)
= Fer
("( j
, which is correct.
In case of two independent variables
1
Nt = "l (x,
is arbitrary and without
1
2
x, x,
when the function
2
x) satisfies the usual conditions of analyticity, is expan-
dable in power series, etc., one can choose this invariant "\.
in an
arbitrary manner. It is usually possible to construct the one-parameter continuous group of transformation: _ i 1 m . xi = f (x '... J x J. a) J (1=1 J •.• J m', m
(3. 4. 14)
Y6 = fo (yI ; a),
(S = 1,
corresponding to the chosen
"t! '
~ ~
2);
"', n; n ~ 1 ) .
i. e., satisfying all the properties of
such a group. In practical applications, the function
"'l = y/ f(x)
is
often used. Applying the usual procedure from the theory of groups, one can easily verify whether the function in question satisfies the conditions of the one-parameter group of transformations. Let us choose :
78
- 70 M. Z. v. Drzywoblocki
(3.4.15)
1=y/f(x);
x=xexPP1;
Then, from the condition that (3.4.16)
Y / f (x) =
"I
is
PI = parameter.
an invariant, i. e., from:
y / f (x) ;
one has: (3.4.17)
Y=
yf(xexPP 1)/f(x).
This transformation is valid in the extreme case when
P = O.
In the next step one gets : (3.4.18)
x
= x exp P2
;
Y = y f (x exp P2) /f(x)
or inserting Eqs. (3. 4.15) and (3.4.17) into (3.4.18) :
Similarly, one can easily verify that Eq.(3.4.15) represents an invariant : (3.4.20)
Y / f(X) = [Y f(x exp P1)/f(x)] / f(x exp PI) = y/f(x) .
3. 5. Application of Theory to the Hypersonic Boundary Layer. Assuming that fundamenta system of equations governing the distribution of the veloc.ity components, density, pressure and temperature is given in form of the solution of Maxwell-Boltzmann equation as derived by Grad in Cartesian tensor notation one obtains a system of equations in case of a flow without external forces. After some elementary operations on tensor forms in the case of a two - dimensional Caru1== u,
tesian coordinates, the system with form (see
(49]):
79
u 2 ::. v, etc. takes the
.. 71 M. Z. v. Krzywoblocki -1
(3.5.1)
9(u u, x+ v u, y) + c(
(3.5.2)
f (uv, x +vv, y )+d.,
(p,
=v
M
-1
(p,
x
+p
+P ) xy, Y
xx, x
+p
y
+p
yx,x
yy,y
=0 ;
)=0
(3.5.3) (3.5.4)
01.
p=R9T;
eo
,2 . c2
0
'
:;: V R T Go
0
0
0'
"I D l-(C T), u + (c T), v] + p (u, + v, ) + p u, + .... I v x v y x y xx x
(3.5.5)
(p
xx
u),
x
+ (p
xx
v),
y
+ (2/15)(2 S
- S )+ y, y
x, x
+ (2/3) (2p u, + 2p u, -p v, -p xxx
(3.5.6) + (2 / 3) P (2u, (p
xy
u),
x
- v, ) + Ii y
x
+(p
xy y
xy
v),
y
+ (S
+ P (u, + v, ) + p u, yy,/ xy x y (3.5.7)
=0 ;
+ p (u, + v, ) + p v, + ( S + S ) /2 xy y x yy y x, x y, y
+ Ii (p
-1
yy
f.I 1-
u),
x
yxx yy
-1
.
+ (2/3) (2p
v),
y
y, x
)/5 + P v, + xx x
y
Vo
yy
+S
= 0;
+ p(u, + v, ) +
r-1 p pxy = 0, "X.= c + (p
r-1 p pxx
Q
f'
x, y
v,) + y
R
+ (2/15)(2
-1 0
x
,fA =. Re :;: L fo U0
r
s y, y - Sx, x) +
v, + 2p v, - p u, - p u,) + yy y yx x xy y xx x
(3.5.8) +(2/3)p(2v,
y
-u,)+ x
(S u), + (S v), + (11/5)S u, x x xy Xx
80
=0;
+ (2/5) S v, x
y
+
r' -1
- 72 M. Z. v. Krzywoblocki
+ (7/5)8 u,
Y Y
+ 7 Cl(, -1 [p
+(2/5)S (RT),
xx
_2d..- 1p - 1 [p
(P
xx
(3.5.9) + 5 Q,-1 P (RT),
x
Y
v,
+p
x
xx, x.
x
(RT),
xy
+P
xy, y
+ (5/2) i-I
-1
+2 c(,
RT(p
xx, x
+p
xy, y
)+
1-
y
)+p
~ A-I
xy
(P
Rp 8
yx, x
+P
= 0,
x
yy, y
)
+
)= R,..,
(S u), + ( 8 v), + (2/5) S u, + (7/5) 8 v, + yx yy xy xx + (2/5) S u, + (11/5) S v, + 2 0/.-1 RT (p +P )+ Y x Y Y yx, X yy, y +7c(,-I[p _ 2 d.. -1
O-l[p
I
(3.5.10) + 5 rJ. -lp (RT), (3.5.11)
P
xx
yx
=p
xx
yx
y
(RT), (P
x
xx, x
+ (5/2)
+p; P
xy
+p
yy
+P
(RT),
xy, y
~-1 ~ i = P
yx
y
J-
) + P (P +P ) + yy yx, x yy, y 1Rp8 y
=0 ,
= Pxy = Pyx;P
yy
=p
yy
+p.
This is a system of ten non-linear partial differential equations of the first order in the dependent variables and two independent variables. To transform the system of the above equations , one may choose the
following rules of transformations:
Q.(x, y) = Q(x, z); z = y / g(x); z, 1
(3;5 12)
z
'y
=g
-1
X
= - zg
-1
g'
. Q = z d Q/ dz' Q1 = z, dQ/ dz '1, x 'x ',y y
where all the higher order derivatives of
;
z with respect to yare
equal to zero and primes denote derivatives with respect to z. We introduce new
dependent variables:
81
etc.,
M. Z. v. Krzywoblocki
(3,5.13)
... v:;: g' v; p
xy
:;:gtp ,iP :;: gl(p g' xJ x, y xy
·1"'" ), :;: g'px,y , y
then we obtain, from Eq. (3.5.1 ) : (3.5.14)
.... ·1 r ... 9(-zu+v)u'+ i L-z P~x(x,z)+ pixy (x,z)
1 :;:
0,
where prime denotes ordinary differentiation with respect to Let u' (x, z):;: G, d z:;: G-1 du, z :;:
J G-1 du,
z.
then Eq. (3.5.14) can be
given in the form
o (-zu +
(3.5.15)
)
.v. ) u' +
01.. -ll~ .. zp ~
xx
""}
(x, u) + pI (x, u) :;: 0 , xy
where prime denotes ordinary differentiation with respect
to
With the new dependent variables:
;P
....
p:;: g' p
;p
....
yy
= g' p
yy
u . yy
=g'P
yy
one easily gets, from Eq. (3.5.2): (3.5.16)
o (-zu +v. . ) v'
J
(x, u) + cl.
-1["":Pyy(x, u) - zp'xy (x,u))
=0
.
After all the transformation in the equations of continuity and state, one obtains (3.5.17)
If we use a new dependent variable,
....
S :;: g'S , Eq. (3. 5, 5) takes the y y
form:
..
"" o(-zu + v) d (c T)/du - zp + ~} v xx (3.5.18)
+ ;, (x, u) P
yy
[(g')--zv'(x, 1] u) p
- -L zS' (x, u) -
x
With the use of new dependent variables one easily obtains :
82
S'y(x, u)l'J
xy
/2 :;: 0 .
S :;: g' S ,p X
+
X
xx
.... = gl P
xx
,
,
- 74 M. Z. v. Krzywoblocki
~~ _ z d(u pxx )/ du + d(; pxx )/ du
- (2/15)[2zS 1 (x, u) + 8' (x, u)l + x X 'j
+ (2/3) [ -2zPx;{2. (gl) -1 + ZVI(X, U)) Pxy - ;'(X, u) P (3.5.19)
fa,
.1
-1
"" gppxx
pf'
yy1}
G +
=0;
1 { -Zd(U Pxy )/du+d(;p xy )/du+[8x (x,U)-ZS'Y (x,U)] /5 -
_ZV'(X,U)P (3.5.20)
+ do.
-1
xx
_[Z_V'I(X,u)] p +(gl)-1p {G + xy yyj
-1 ~ ~ gpp
xy
= 0 ;
r1- z d (upyy)/du + d(;p yy ) /du + (2/15)[zSIx(x, u) + 28y' (x, U)] + + (2/3) [zp
xx
- ((g') -1 + 2 zv ' (x, u)) P + xy
(3.5.21)+2;' (X,U)pJ}
G+ c(l
p f.-I gp;yy
=0.
Components of the heat flow vector are given in the forms of:
t-,
d(uSx)/du + d(;-SX )/du + (1/5) [[ -11, + 2;-' (x, U)) SX +
+ [7 (gl)-1 _2ZV'(X, u)] S 1 + 2ct- 1[_zd (RTp )/du +d(RTp )/ xx xy yj + 5 ol.-1 (-zP
xx
+; ) d (RT)/du - 20(,-1 xy
0 -1f[-zPI
I
+ pi (x, u)l P + [ -z pi (x, u) + pi (x, U)] xy 'J xx xy yy (3.5.22)
~_
+ (5/2)rJ..,
-1
~
xx
d~J +
(x, u) +
PXyl) G +
JJ
A-1 gR p Sx = 0,
z d (USy)/dU + d (;Sy)/dU + (1/5)~~(gl)-1 - 7 ZV'(X, U)] Sx +
+[_2z+11;'(C,u)]sl+201.-1[-Zd(RTP )/du+d(RTp )/duJ+ y3 xy yy 83
- 75 -
M. Z. v. Krzywoblocki
+ 5cl- 1( -zp
xy
+P
yy
+ pi (x, u)] p xy
xy
+[ -
) d(RT)/du - 2 r( -lflf[_zPI (x, u) + ~I xx Zpl
xy
(x, u) + pi (x, yy
+ (5/2) ri -1 ~ A-I g R P Sy
(3.5,23)
l}
u)]
p G + yyJ
=0 •
Eqs. (3.5.15) to (3.5.23) jointly with the equation for G, i. e., ,etc., ~, etc. , xx represent the system of 23 ordinary, non-linear, first order differenG = 1/(dz/ du), and equations (definitions) for P
tial equations for 23 dependent functions,
S x
Pxx;
P , yy
p
Pxx; pxy ,
S y
P
Pyy;
P
xx
P
xy
"-
xy
P
z· v' J
P "-
yy
J
p . p' J
J
T;
"-
yy
S x
v
"-
S , y
G,
in one independent variable u with a parameter x. One may preserve the original forms of dependent functions without introducing the functions
v, pxx
' etc., thus obtaining a little more complicated
structure of the remaining equations . Basically, when all the des like
,etc., are inserted into the main xx system of equations, it reduces to ten equations for : z, v, PI p, T,
P
xx
,p
yy
v
,
p
magnitu~
, G ,P
,S S .
x y
3.6. Particular Gases of the Direct Reduction of Independent Variables (with no Parameters) . Below, we demonstrate how a group may be found such that a particular system of partial differential equations is conformally invariant under the group in question. The technique, shown below is a new one.
84
- 76 M. Z. v. Krzywoblocki
[ 62 J
Consider the system of equations
u = u (x, y) .
(3.6.1)
= 0
,
We wish find a
conformal invariance under a group
_
1 A 1 :x=f(x,y;a),
2
y = f (x, y ; a) ,
4 3 u = f ( u; a) :=. f
(3.6.2)
Al
5 (u;a)+f(a).
5
il~
Since
oy
= 0,
on Of
Eqs. (3.6.1) , (3.6.2) yield:
1..
=
vx
(f4 ) U
()x
"vJ
+ 1.. (f4 ) fl oy U oy
4
=u
(3.6.3)
A function the yls up to the der
r1Ek
4
(l.!.
'Ox
uy
'Ox r.l
+
2l) + f4 (uu ';)x 'Oy i) x ()y
~ ';)y
+ QU ]1.) oy Jy
.
of the XIS, yls, and the partial derivatives of
k - th order is said to be conformally invariant un-
(we drop E k
r)
and simply say 1 k 1
if :
t
- _1 -n uy U r.l (x.y (x), "',y (x), - 1 ' "', - - k ) U(xm)
ux
(3.6.4)
k n kn 1 n ()y 1 n ~y =F(x,y ,''',y , " ' , " ' , k;a)r.l(x,y (x),· .. ,y ,"":"'-k)' 0(Xm) U(xmJ k n 1 n uy . 1 where F(x, y , ... ,y " ' , " ' , U(xm)k ; a) IS an arbitrary
r.
under
function and
r.l is the same function of the original and transformed
variables. From Eqs. (3.6.3) and (3.6.4) , it is required that:
~ ux oJ
UU=U(u f4
uY
(3.6.5)
= F
+ ()f4
uy )
'Oy uy
+
f4(~ ~
'Ou UU ou (x, y, u, vy' Tx; a) ~ ,
85
'Ox
vy
+
~ OY) uy 'Of
- 77 -
lVI.
where the function
F
z. v. Krzywoblocki
has to satisfy some conditions due to the group
properties but is otherwise arbitrary, i. e., there are sought solution of Eq. (3. 6. 5) which satisfy the conditions of a one -parameter group of transformation. One such a class of solutions is obtained by equating on both sides of equation all terms in (3.6.3) containing
~~
explicitly and set-
ting all other terms equal to zero. Eq. (3. 6.3) then yields;
uf4
(3.6.6)
Ox oy
u~
of4
(3.6.7)
u~
k
oY
0, 0,
0,
(3.6.8)
which implies that
'Oil
= f4
'JJ
UU
.l1
()u
or
'Oy 0y'
Oy=F(x,y,u;a)
The system of partial differential equations
u~ =0.
G will be called
system corresponding to Eq. (3. 6.1) and to the group If
'";)
the
AI'
u = 0, Eq. (3.6.1) is satisfied. However, solutions will be sought
for which sformation with
ulo.
Since the definition of a group states that the tran-
Al has an inverse, the Jacobian determinant associated
AI:
(3.6,9)
i)
(x, y, ill I =
o(x, y, u)
Vx ax Jx ~
~ ~
oy 'OJ 'Ox
oy
~y
~
() u I oii OU --I ox
uy
UUI
I
86
- 78 M. Z. v. Krzywoblocki 4
cannot be zero . If in the last equation of (3. 6.2), If = 0, then o ii '0 u () ii . . (x y- ti) = - =m WhlCh c a s e ' , =0 0X uy vu = 0 u(x, y, u) , contrary to the original assumption. Therefore it is necessary that
lu
f4
f
0•
Similarly, the subgroup: (3.6.10)
SA: .1
x= f1(x,
y; a),
_
2
Y = f (x, y ; a) ,
must have an inverse and therefore neither the Jacobian determinant associated with SA 1
(3.6.11)i 0 (x, 51)1
I U(x,
=
y)
vX' uX" () x
() y
nor the Jacobian determinant:
Y)I-
U(X, (3.6.12) II uti, y) -
Uy vi
~y
-0y
associated with the inverse transformation may be equal to zero . From (3.6.12) , it follows that not both l)u equal to zero .. If ~
(3.6.13)
f
~;
and
~~
are
0, then from (3.6.8) :
Ox
uy = 0,
and therefore (3.6.14)
~~ f
O.
Eqs. (3.6.6), (3.6.7), (3.6.8) and the above remarks imply that:
87
- 79 -
M. Z. v. Krzywoblocki
ax '(Jy = 0,
(3.6.15 )
Hence, (3.6.2) satisfying conditions (3.6.15) furnishes the conformal invariance . We shall apply the above technique which may be called a technique of vanishing coefficients, to the time dependent Navier-Stokes equations in three dimension of motion of a viscous, compressible, perfect gas. The equations of motion are ;
.
(1)
Y(UU Tt""
JU
2
(3.6.16)
_uu ) + _ClP
UU
+ u ~ + v"'5Y + w () z 2
2
_
()x
2
2
2
_{1(1.:.:. + () u + ~) _ 3-1tl(~ +~ + "Zl w = ()x 2 uz 2 ()x 2 UXdy tlx()z) 0,
vi
Ov ()v ":-'v 'v "Ip (ii) 0 ( +u+ v _u- + w _u_ ) + _u_ J (It ox
uy
uz
_
uY
(3.6.17)
and (iii)
0 (
F
0 w + u J w + v "\) w + w Ox "\)y
ot
··,2w
?}
()X
uy
2
J w) 0z
+ () p z
•
r}
02
~)=O "'I 2 uz
u
_11(-101- +---.!.+ 0W).!.r(_u +_v + I ">. 2 "'I 2 ,2 3 uxi) z
(3.6.18)
()
vyuz
Z
-,2
.
In addition, the gas is described by equations of state, continuity and conservation of energy. They are, respectively (3.6.19)
(i v)
(3.6.20)
(v)
p =
9R T
(state) ,
J \~ +';) (pu) + ~) + ~
-'S"t
'Ox
-vy
88
()z
o
(continuity),
- 80 -
M. Z. v. Krzywoblocki
and
+ ( U w+ b)2 + (u u + dw)2 +
Uy
2 au "'uv SY
/-I,
\)w
+)z)
21
= 0 (energ~,
R
C and k are constants . v The above system of six partial differential equations in six un-
known functions of system that
Oz Tx
+ 3 ( T,; +
(3.6.21 ) where
uz
II
II
A
x, y,
z
and t will be
A
II
II
is conformally invariant under
•
referred to as the
In order to find the conditions on a group
find the system corresponding to
"A
II
such
A1 ' it is necessary to
and AI. Al is a transforma-
tion group defined as : _
I
-
2
_ 3
AI: x =f (x, y, z;a) , y=f (x, y, z;a), z=f (x, y, z;a) , _ 4 5 6
u =f (u;a)=. f (u;a) u + f (a) .
Since no new techniques are involved in finding the system in question, this problem will be left to the reader. The present paper will confine its attention to one such group under which mally invariant. It is the following group :
y = y - '( 2a ,
t
= t - ~ 4a ,
u = u,
z = z+ ~3 a ,
v = v, w = w ,
89
II
A
II
is confor-
~
81
~
M. Z. v. Krzywoblocki
(3.6.22)
-p=p,
T=T
where the r's are arbitrary constants. The conformal invariance of 1 the system of equations "A" under the group P is easily verified. The invariants of
pl
may be chosen as follows:
With the use of equation (3.6.23) the system
"A" is reduced to
the following system which will be referred to as the system" B " :
90
- 82 M. Z. v. Krzywoblocki.
(3.6.24)
" B "
(3.6.25)
(3.6.27)
V'V2
VY2
()'q3+K2))1~ + ... ) +... = 0 ,
(ii)V 5
(t3
(iii))5
()Y3 ( ~ 3013 +
(i v) ~ 4
= R\'\ \>6 '
+~2
(3.6.29) (vi)
C ))
v
U().\ V2)
U())5 VI)
i3~
(3.6.28)
+ ... ) +... = 0 ,
t2 t1 ~
~V5
(v)
()Y3
+t2
)"'l1
+t1 Oll{l
d P\ Y3)
d ()\ Y3)
oV 6
5
(t -0 +'6 t 3
()~3
+0 4
o"'t2
~3
=0 ,
0\16
2 1
~ + ... )+ ... = 0 . 'tJ.1
The system "B" is absolutely invariant under the group:
(3.6.30)
)5
j
=V
j
(j
= 1, .. " 6) ,
whose invariants , as in the previous case, may be chosen as follows:
91
• 83 -
M, Z. v. Krzywoblocki
The system
"B" may be reduced to a system
51
included) with independent variables n C"
is,
(3.6.32)
and
.5 2
11
C"
(not
. The system
similarly, absolutely invariant under the group: p(3) :
~ = f: + > 1 >1
1/
a
08
'
?)2
= F _y a )2 09'
'1' =t (i=l'" 6) ii'
whose invariants, as in the previous cases, may be chosen as :
p.( 0( )
(3.6.33)
J
(3.6.34)
92
, ,
- 84 M. Z. v. Krzywoblocki
+[r2 t9 (t l05 t 6~9+ ~16667aa+2t3t5(7ta) ~
0 t/
+ 2 t 3 (r5 2 9 2+ (3.6.35)
+ ¥1 t 4
~~
t t
'6 8 t 9 J
t
¥
(a 2 ) + 3i 46'6 6a
t t
+r.
2
d c-A 2
J
0,
i 4 t 6 t8
:l]
(t 2 t 3 ~ 7 + t4 t6)
i d
t
(9 ( Y1 6 +
~'"
-,,d~
'¥ 5t9 1~519 +'h!8 i + 't,r4'{ 6(8 It 5 ~ 9 93
(~5 r9+¥7~ a) +
i~ 31 =
+ t 2 t 3 5 9ta 5 t 9+ 2 7 08)+
+ ~ 3 5)+ 7 '18
+
+
t
+ K7 8 i +
- 85 M. Z. v. Krzywoblocki
(3.6.37) (iv)
(v)
(3.6.38)
P4 =R
~5 ~6
t 3 ~ 6 t8
[
+
d 1~5 d d..
6) ~ 8]
¥2 ~ 5 ~ 9 + ( ~ 2 17 + ~ 4 6
94
d( ~ 5 ~3) d
=
0,
- 86 M. Z. v. Krzywoblocki
where:
Al
~~
¥3 t 6 t,
5[
+¥2
t6 t9 i I
+[(¥1~6+t305)t9+ (3.6.40)
+[¥2
+
t3¥7 t 81
t5 69+(t2 t7+ 04
Y6)t8J~3
~2
+ '
and
+(\ I
2 2
2 2 2 + \' ) (y' \' + 2 '( Y (i 3 , 5 '. 9 .5 ,7
95
2
2
\<J 8 ,g \ + \1,7 )/ ) u8
+
- 87 M. Z. v. Krzywoblocki
The triply invariant solutions of the original system under p(l) (eq, (3.6.22),
p(2)
(3.6.32)) are determined
(eq.
I:
(3.6.30)) and p(3)
A II
(eq.
by the solutions of the system of ordinary
differential equations, "D
II ,
The correspondence is given by the rela-
tions :
d.. = '2 ': 6 \ 9 x
l(\
1 ') 6 t , 3 ) 5)
t
i9t
I
t
3 \ 7 \ 8] Y
t
,
~ i 2 j 5 a9 t
(3.6.42)
\
()
\;1 1
=u
( \ 2 :\ 7 t
,3 =v
'I
2
~
j
4 \ 6) ) 8
z
3
t
6' 8 t,
:!,
'
6
=T .
Although the present investigation considered a perfect gas, the reduction used can evidently be applied if equation (3.6.19) is replaced by a general equation of state, ~,
Cv
and
k (see eqs.
F (p,
f
,T)
=0
. Furthermore,
(3.6.16), (3.6.17) , (3.6.18) and (3.6.21)
could have been taken as arbitrary functions
of
p
and
T .
The technique discussed in this chapter can be applied to all the regimes
(sub - trans- supersonic) and to all kinds of fluids: inviscid,
viscous, MHD, etc. There are serious disadvantages of this technique: (i) it is very difficult to find invariant groups; (ii) it is certainly very difficult to satisfy the boundary conditions, particularly two-curve B, C. , in many practical cases .
96
- 88 M. Z. v. Krzywoblocki
4. Topological Technique.
4.1. Fundamental Concepts. Without explaining all the details of the topological technique, we are going directly to the final results. The technique is based upon the elementary fundamentals of homology . Assume a three - dimensional nonsteady rotational flow of a perfect, inviscid, non-heat-conducting fluid in an electromagnetic field. The fundamental equations governing the hydrodynamic phenomena are: equation of motion: ~ q, t
(4.1.1)
+(
... M)~ q. v q
+ rI'l-ln v
p
equations of continuity and state : (4.1.2) pressure -density- entropy relation:
-+
where
P
unit mass,
denotes the external (including electromagnetic) forces per
...q
the velocity, and
From the first law d Q = T dS with
h
S the entropy .
of thermodynamics: -1
= cv d T + P d (P ),
= cp T =cv
T
+
Pf
.. 1
'
c ,c v
p
= constant,
one obtains the vector equation: (4.1.4)
T
\J
S
- \]h +
P-1 \7p 97
=
0
T\JS = VQ
- 89 M. Z. v; Krzywoblocki
The expression d,Q contains the Joule heat as well as any other enrgy (heat) addition or subtraction from or to outside. Addition of the expression
~ x
(\l x ~ ) to both sides of Eq. (4.1. 1.)
combining terms on
the left-hand side and using Eq. (4.1. 4) furnishes the generalized Crocco equation: ~
-?
(4.1. 5)
~
-7
+ \J H - 'iJ Q = q x w+ p
q, t
;
2 -7 -t H=q 12+h; w=curl q ;
(4.1. 6)
ct
H = h = c T at the point where o 0 p 0 fined as:
= O. The velocity of sound is de-
(4.1. 7) which in a
steady flow can be obtained from the generalized energy
;:.u;::~n °:2';: ::~~ 1;-1 a2- t~. d~
-f
d Q' K
(r .J<)
where the integration is performed along a streamline, ~
,)-,
(~,
= streamsurfaces, from a point "j = 0 to a point
r
The function K(
'\' 'fl
• /, ) = const., II
<)
II
•
,j), ) assumes a constant value along a streamline,
= constant.
In the case of a unsteady flow one may calculate
the velocity of sound from the generalized equation of energy obtained by means of the combination of Eqs. (4. 1. 1) and (4. 1. 4) and integration
with respect to the running coordinate
1:
w along the particle line [(
A. XJ
V) = constant
(4.1.9)
2 fW q't' dW-t
q 12 ~
The function
( A' x.. , 'V
-7
0
L (A,
1 + (¥-1)
-1 a 2.('~/ow -7p.
X' »)assumes
(,t'X ,))).
a constant value along a particle line
) = constant. The particle surfaces
identically the continuity
/w
dW 1 -} dQ = L
equation (4.1. 2) .
98
A, X , \>
satisfy
- 90 M. Z. v. Krzywoblocki
Consider the following two fundamental equations (continuity and Crocco) : (4.1. 10)
div
(9 q) +9, t = 9d i v ~ +grad 9 . q+P, t
-+
q x
(4.1.11)
-7-7 w = q, t
+ \I (H
=0
~
- Q) - P .
From (4.1.11) we get:
-+qX(qxW)=(q'W)q-qw=qxLq't 7 -t ~ -7 -? 27 ~ root ~J + Mv(H-Q)-P .
(4.1.12)
We may propose a certain number of hydrodynamic systems: -+ -1 -i (I) div q = (grad q+ t) ;
curl~" (II)
div
(~ q)
curl (III)
div
q-2! = -
P,
p.
9
(~. ill
f' t
(9 ell = 9 q-2
+ I7(H-QI-ilJ)
;
t (~'~iq - qx[ ~'t +\7(H -
Q) -
pJ}+ gradfx~;
(q x~) = div q, t + ,l(H - Q) - div P;
curl
(et x ~)
cr. - curl
= curl
t
P.
For the purpose of the particular interest in an inviscid , non heat-conducting fluid, the system (I) is taken into account: (4.1.13)
~
div q = W; W = -
(4. 1.141 curl
Z:; t
0
9 -1 (grad p' ~q +
Wand
(q·;l11. -~x
Z being
r'
P't); j
J}.
+ 17 (H - QI - P
known and given .
The electromagnetic part of the set of equations governing the dynamic system in question consists of two groups. As the first group, we choose the pre - Maxwell system: (4.1.15)
~
curl B =
~
tl J, 99
- 91 M. Z. v. Krzywoblocki -4
(4.1.16)
div
(4.1.17)
curl E
(4.1.18)
div E
0,
B ~
~
= - V B/ () t
,
= k -lcl
-+ where the used symbols denote: E
ty;
= electrostatic
vector field intensi-4
k = permittivity of the medium (dielectric constant) ; H = magnetic
vector field intensity;
B = ft it ;
11 =
permiability of the medium
(assumed to be equal to unity); (1= charge density;
i' = "1 A;
1 current
(amp) ;
-; = conduction current density; A = cross-sectional area.
These are two systems of equations, Eqs. (4.1.15) , (4.1.16) and (4.1.17)
(4.1.18)
~
--i
in two unknown vectors
Band
E.
The second group consists of : a)
Ohm's law: ~
-I
(4.1.19)
J = ()1(E
"
+ ,~~ q x II )
~ 1= electrical conductivity;
b)
Joule's heat:
(4.1.20) c)
Q = J2 0" -1. l'
Q:; Q
1
+Q
ext
"
the forces arising from the charge density and from the induced
magnetic effect due to the motion of the electrically conducting fluid through the magnetic lines of force : (4.1.21)
4 P
...,
1
=(l.E
"4""
+ JxB'
~
~
4
P = PI + P
'
ext
.
The hydrodynamic system, Eqs. (4.1.13), (4.1.14) and the electromagnetic system, Eqs. (4.1.15) to (4.1.18) are interrelated. Moreover, the right-hand sides of each system contain the unknown functions from itself.
100
- 92 M. Z. v. Krzywoblocki -'t
Thus the functions
Wand Z in Eqs. (4.1.13) and (4.1.14) contain the
unknown functions
p
fJ,
~~
J in (4.1.15) contains
~
The function H, Q t' Q, the ex and the constant parameters k, jJ. , t:r 1 are known and
the unknown vectors vector
-)
-7~->-)
,q, J, E, H. The vector E, q, and
tl .
given. All the other scalar functions and vectors can be calculated from the equations given above. This suggests that one may apply some sort of successive approximation process, in which the right-hand sides would consist of functions known and given from the previous steps of the successive approximation procedure. In each step the study of the pre-Maxwell system or of the hydrodynamical system is reduced to the study of systems having the form: (4.1.22)
-+ div Z
= () ;
~
curl Z
;::t.
=;::::
4
div ~
= O.
By comparison with Eqs. (4.1.13), (4.1.14), both systems of equations, hydrodynamic one (4.1.13), (4.1.14) and electromagnetic one, Eqs. (4.1.15) to (4.1. 18) are actually representable in each step of the successive approximation procedure by means of the system (4.1. 22) . The system (4.1. 22) was extensively studied by Blank, Friedrichs and Grad in connection with application of it to the pre-Maxwell system. We shall use the most important results of those investigations in the generalization. The fundamental philosophy in the present generalization is the following: the system of equations of an irrotational flow of an inviscid, incompressible fluid is identical to the system of equations of electromagnetics in special conditions (div
It = 0 ; curll = 0 ; or
~
curl E = 0 ) .
101
div
Ii = 0 ;
- 93 -
M. Z. v. Krzywoblocki
The generalization to the electromagnetohydrodynamics of a rotational flow of a compressible fluid and embedding it into the entirety of the Blank-Friedr ichs -Grad theory is obtained by a successive approximation procedure, in each step of which the right-hand sides of equations are assumed to be known and given. The theorems derived in
[lOJ
are applied below to the electro-
magnetic equations. Blank et al. define the period as the
constant of
homology class in the vector field . These are:
[r}[r; z] · fr ;Z ~ ;
(4.1.23) (4.1.24) (4.1.25)
(4.1.26)
~J
· [s ,l·jsz, dS;
[~1·[2:
L= open surface ;
[c]. [c x} 15,·;ii;
closed surface
[IJ .
J ' open
{S
The (C, S) or
ctor ; its periods or with charges,
C = open curve ;
D. We have four homology groups in D :
curve
E
= closed curve ;
S = closed surface;
J~.d§
yJ.
in a three-dimensional domain
face
r
ij, C
closed curve
{rl
,open sur-
group is associated with the electric ve-
(E periods) may be associated with emf, V = [ C;E] Q
=
k[ S ;
E) . The
(r, L)
or 'h),
group refers
to the magnetic vectors; its periods ('YY\, periods ) are defined by flux
~
=
[L; B]
[r;
)A I =
and current
BJ
in external circuits A ~
well-posed problem gives boundary values of ~
set of 'Y'r\. pariods ; or of
Zt and a set of
Z and the values of a n E periods.
The system (4.1. 15) to (4.1.18) is transferred into the form (with -4 B
UBI Ut) :
102
- 94 M. Z. v. Krzywoblocki
.:..
l
= -B;
(4.1.27)
curl
(4.1. 28)
curl B =
.!.t
-4
J
with
and
q
-4
fJ
;
div
g
div
~
= k- 1 q;
o,
assumed to be given as functions of space coordina-
tes at a given time
-+
t. With curl curl
E
...E
unique solution of (4.1. 23) , (4.1. 24) for -+
trary boundary values sentation [S; E]
Et
E
and
~
=
-}It J , there exists a and
...B
possessing arbi-
periods in closed surface repre-
,compatible with
q
or
open curve
E
Blank, Friedrichs, and Grad show that
[C ;
EJ,
is given by minimi-
zing the expression:
(4.1.29)
F
[ill • Jo2- 11curl
E)2 dV
+
~ E\. /' 1 dV.
subiect to the admissibility condition: (4.1.30) Obviously (4.1.31)
div -4 E = k -1 q; ~
-i
E t given on
* S;
~
[ S ; E]
B [= _:1curjl E ~1 An:t:er SOlUtiJon
F
=D2
B
dV +
B
~
or
ri LC
]
E given.
o:ained by minimizing:
SltEt x B . dS ,
subject to the admissiblity condition: (4.1. 32)
curl
....
B
=
rJ
~
. ~
Only the given and known boundary values,
Et' appear in Eq
(4.1. 31) , (well-posed problem) . Also there exists a unique solution of (4.1. 27), (4.1. 28) with given boundary values
13n and ~
well as given boundary values
pari ods
([r ; B) or [r ;
En and "m. pariods
103
([L;
BJ
E) or
f;
) as E
~
- 95 M. Z. v. Krzywoblocki
4.2. Application to the System of Hydrodynamic Equations. Consider the system of equations(4.1.13) , (4.1.14) . The righthand sides of these equations are supposed to be known and given in each step of the successive approximation procedure. A solution for q
~
in the subsystem of Eqs. (4.1.13), (4.1.14) furnishes the vector q.
In each step of the successive approximation procedure this subsystem can be representable in a system of the form: (4.2.1)
curl
-+
-4
= N;
M
div ~
-+
=
M
R',
--+
N:::::Z,
(4.2.2)
R=W.
Then there exists a unique solution of (4.2.1) ( or of (4.1. 13) ,(4.1.14) ) ~
with given boundary values open surface periods
.,
(I;
A]
->--7
M (i. e., A = q ) and flY\. n n n or close curve periods
li;
pariods :
J'
compa-
[z::.;
A
A
tible with N.
q,
For the vector field of the velocity
the'h1. period
be generally assumed to be relevant to the
J may
physical applications
and notions . It may be referred to as the total (velocity) flux concept through the surface
I
.
The period
li ;
AJ can be interpreted as
the circulation. A unique solution of (4.2.1), or of (4.1.13), (4.1.14) exists with given boundary values
4
Mt (i. e. ,
-+
ther representations : open curve periods periods
rS ; AJ compatible
4
At = qt ) and f
[c;
periods in ei-
AJ or closed
surface
with R. The physical interpretation of
the periods is difficult, if in general possible; let us remodel the system (4.2.1) in the sense: -1
(4.2.3)
curl curl
div M
104
.-)
=R
div:='=;
=
0
J
- 96 M. Z. v. Krzywoblocki
which can be treated as a system consisting of two srbsystems: ;3
-4
curl Z
(4.2.5)
curl M =Z
, div Z = 0 ;
,...J
.-'
.~
-t
;:Z
~
=
(4.2.4)
-+
; div M = R ;
(div;:=
=0 ) ;
-t
( div Z = 0 ) .
Then we have a unique solution of the system (4.2.3) with arbitrary
~t (i, e., 1. =cr.) and tl;
boundary values [C ; MJ
1)
[8;
(or
M
arbitrarily prescribed periods
compatible withR .
In the approach accepted in the present work, each of the statements presented above is valid independently in each step of the chosen successive approximation procedure. Denoting particular steps in this procedure by the letter
"m II
,
we have the following system of equations (i)
hydrodynamical :
(4.2.6)
div
(~)m+1
= - [ p(grad
(4.2.7)
curl
(
=
(ii)
p . q+ P't) J m
q~2I (q. tJ)""q -q X[
+
;
'V
(H-Q) -
PJ) m ;
electromagnetic:
(4.2.8)
-+ curl B
(4.2.9)
-+ Curl E
~
-7
div
= ttJm
m+1
=-
m+!
UB m / 0t
~
div E
13m, ~, ~m' ~lm'
The quantities
=0 ; B m+!
m+!
=k
-!
~
appearing in the system of
equations are calculated directly from the following equations : (4.2.10) (4.2.11)
-+ curl q
m
~
J
m
= C5'
=
~
w m
~
(E
1 m
+ AA q r
m
---+ X H );
m
105
- 97 M. Z. v. Krzywoblocki
(4.2.12)
Qlm =
-10
J~ C' ~1
the velocity of sound,
a
P
2
1m
-t
-t
-1
=~E
m
+J)(B
m
m
can be calculated from Eq. (4.1. 9) .
Thus, the expressions (4.2.10) to (4.2.12) do not need to be considered The systems (4.2.6) , (4.2.7) and (4.2.8) are of the form; (4.2.13)
curl
-t
-t
-10
A =N m+l m
div A m+1
=(J
m
or of the form ; -4
(4.2.14)
= curl
curl curl A m+1
-t-+
N m
~
=S
div A 1 m+
m
= (tm
•
Blank et al. propose a variational solution of the system; -4
-+
--1
=S
div A
(4.2.15)
curl curl A
(4.2.16)
admissible ;
div
(4.2.17)
variational;
curl curl A
A=
(j"
-4
= 0-
,. -4
=S ,
which system admits an arbitrary specification of the boundary values At
with the given
the vector
(4.
periods [S ; AJ or [C ; Al . We propose that
....
Am+l
is then given by minimizing the expression;
'.18) F lAm+!l
'1,
,-11 ,urI 1m+!I'
dV -
J. Am+! . Sm
dV •
which in case of the present system and of the applied successive approximation procedure takes the form; (4.2.19)
)--'
-I
F ~ cl m+1 _ =
! -11 2
--l 12 curl (qm+l)
JD
106
- 98 M. Z. v. Krzywoblocki
(4.2.20)
S~ = q~2I (q'(;)) -q .-qx [
(4.2.21)
F[
(4.'.22)
F
m
= 1,
B.n+11 ~ ID 2- 11our1 tm+d'
lE'm+11
]:-ll curdm+11'
dV
Q).
-1
dV -
PJ}
m ;
Sm
dV;
J: s:"
dV •
Bm+1 . m+1 .
2, 3, .... The tangential values of the functions in question on the boundary:
~,
Bt ; Et are
given and arbitrary. The zero-approximation values,
"" ; t), qo (x -t t), Po (x
Wo
-I ; t), and Eo(x; "" t) are also known (x"+ ; t) , Bo(x
and given. The tool of the calculus of variation which has to be applied to minimize the expression (4.2.18) is a well-known one and does not need to be presented here. We may have to demonstrate that the sequence of functions
-+
Am+1
, obtained in the manner described above, converges -7
uniformly to a function A . Under some restrictive conditions , v. Krzywoblocki demonstrates in [57]
that it should be possible in some cases to construct such a
proof of the convergence. The technique presented in this chapter may be applied to all kinds of fluids. The disadvantages of this technique are : (i) it is relatively easy to construct the existence proof but it is difficult to find a formal solution at a point ; (ii) the boundary conditions on a given curve can be satisfied ; but two - curve boundary value problem is unsolved up to now. In all the techniques, cited above, it is very clearly seen that the application of the high speed computing machines is an imperative aspect of each of these techniques.
107
- 99 M. Z.v. Krzywoblocki
5. Sub-light Relativistic Hypersonics Based On The Relativistic Energodynamics 5. 1. Preliminary Remarks The upper limit of hypersonics was never uniquely defined . Actually, there was constructed the so-called relativistic hypersonics based upon the special theory of relativity in the Einstein formulation. But, the Einsteinian classical special theory of relativity should not be applied to describe the dynamic phenomena in a matterfull
medium. It
refers only to a matter-less medium (light, electromagnetic waves) . Hence, the approach to the sub-light relativistic hypersonics from the side of the "light-barrier" has to be based upon a more realistic approach to the description of the phenomena in the region of velocities below the velocity of light. As such one we have chosen the relativistic energodynamics, whose principal fundamentals are explained below. Recently, there appeared some objections against the special theory of relativity in the Einstein formulation which is based , between others, upon the hypothesis of the constancy of the speed of light. These objections mainly refer to the following items : (1) the constancy of the speed of light; (2) the so-called time-dilatation. The problem of the constancy of the light was discussed by the author in some of his previous papers (see references). In the
years 1959 - 61 the author of the present work proposed
(see references) a some sort of remodelling the Eistein special thory of relativity. The proposition referes to the following items: (1) the velocity of light is not a constant, but a function of the position and possibly time (space-time) ;
108
- 100 -
M. Z. v. Krzywoblocki
(2) The geometrical model of Einstein's metric of a four-dimensional space - time invariant under the transformations of the four space-time coordinates is substituted by the ener-
gy model of the (author's) metric of a four-dimensional space-time invariant under the transformations of the four space-time coordinates. The stipulation is that the total energy in the system in question is invariant under the transformations of the four space-time coordinates; (3) all the above given assumptions clearly demonstrate that the Einstein special theory of relativity is a particular case of the generalized special theory of relativity proposed by the author. As the first points we discuss the fundamental equations and the transformation of coordinates. It will be clearly seen that actually the relativistic energodynamics is a some sort of an extension (into the domain of the sub-light velocities) of the classical special theory of relativity in the Einstein formulation.
5. 2. Fundamental
Equations
Assume a compressible medium of the density motion with the velocity
~
q
P
being in
The following equations are taken
into consideration: (1) The generalized first law of thermodynamics in the form convenient for our purposes : (5.2.1)
d U
= dQ-dW-dE +dE 1
where:
109
2
- 101 M. Z. v. Krzywoblocki
d U = variation of the internal energy of a unit mass of the medium that takes place when it is undergoing a change from the initial to the final state (denote it by the symbol dQ
[r - FJ ) ;
= elementary amount of energy which flows into (or from) the unit
mass of the medium from (or to) the surrounding space-time when when it is in
~ - FJ ; here one may include the addition of the
energy of an electromagnetic origin (joule heat) ; d W
= elementary work done by a unit mass against the surrounding pre~ sure in
d E1
[r - FJ ;
= elementary energy decrease of a unit mass that is caused due to
[r - F]
the gain of mass in dE
2
(conversion of energy into mass) ;
= elementary energy increase of a unit mass that is caused due to the loss of mass in
fr-Fl (conversion L
of mass into energy) .
"
The momentum equation for an inviscid non-heat-conducting fluid is of the form: (5.2.2)
where; p ~
n1
=pressure; = momentum source (or sink) due to the mass gain per unit volu-+
me due to the conversion of energy into mass (units of
n1
are
force per unit volume) ; ---t
n2
= momentum source (or sink) due to the mass loss per unit volu-
me due to the conversion of mass into enrgy ; -t
F
= the sum of the extraneous body forces per unit volume including the force of an electromagnetic origin .
Equation of state: (5.2.3)
p = R PT
, R
f
const.,
110
- 102 M. Z. v. Krzywoblocki
where:
T= temperature. Let: -1
d W = P d v; v = F
(5.2.4)
;
d U =do d (p v) ,
where:
d.. = coefficient function of (possibly) arguments (p, v) . Introducing Eqs. (5.2.3), (5.2.4) into Eq. (5.2.1) furnishes:
(5.2.5)
d Q = (~+1) d (RT) • vdp + dEl· dE 2•
We refer the differential d Q in Eq. (5. 2.5) to the element of the path of a particle (particle line) in an unsteady motion or of the streamline in a steady motion, by
q
Us,
with
q
"Up/u.
=
ct· V'p
; multiply Eq. (5. 2. 5)
and rearrange it, so as to obtain the expression for
this expression is equated to the expression for vq. Eq. (5.2.2) by multiplying it scalarly by
q . Next
'i/ p
v
q. VP ;
obtained from
we integrate the so-
obtained result along a particle line or along a streamline at any time t
= t 1, i.e., between
s = So and s
,with q = 0 for
s = so' and
with R T = pv
f (5.2.6)
q. t ds
so(S
}sQ with
,
i
P =E - E - Q 1
+2
1 q2 +
2
pI:.
/I
'
+
q=q
-I--t q
f(~
+1) d ( pv ) -
1 Fe' d1 = F (t) = constant = C (s ) ,
ads = dst, F(t) being an arbitrary function of integration, which
- without the loss of generality - may be assumed to be equal to a constant. The constant C (s) is a constant only along a particle-line or a streamline, but it may vary, and actually varies in a rotational flow, from a particle-line to a particle-line.
111
- 103 M. Z. v. Krzywoblocki
At this point we shall propose two basic assumptions. First is that the potential energy of the dynamic system in question may be transformed - by a process which is unknown, as yet - into the energy of light. Second, that the velocity of light is not a constant magnitude, but is a function of the position (and even possibly of time) . Briefly, in our case.: F.E.}. _s (d.+ 1) d (pv) = Ac 2 /2,
(5.2.7)
s~
where the symbol "c" denotes the velocity of light and A is a proportionality coefficient. The constant on the right-hand side of Eq. (5.2.6) may be
chosen so as to represent the potential energy of the system
at rest (i. e., q = 0) , without energy addition from the outside (Q = 0) and without any work done on the system by the extraneous forces
(£0 s~ -1 Ite . d~ = 0 ) . Or,
taking the full spectrum from q
= 0 upward
into account, it may represent the maximum velocity obtainable in the = 0) and all of the terms
system when the conditions are stationary (q, t
2
on the left-hand side of Eq. (5.2.6) are equal to zero except q /2 . Hence, on ther right-hand side of Eq. (5. 2. 6) we can write: C(s)
(5.2.8)
=A c~
/2
= q2 max /
2
2/ = cm
2.
We can apply Eq. (5.2.6) to a matter-less form of enrgy. In this case the potential enrgy
Jrs Y
-1
So
-t
-t
Fe' ds
fS (d.. + 1)
d (pv) , the work of the extraneous forces
lso
,E 1 and E2 are identically equal to zero. The motion
of the matter-less energy takes place in form of a wave propagation; hence, the symbol -+ q denotes now the velocity of the wave propagation From Eq. (5.2.6) one gets (the corresponding functions are denoted by the subscript
"l"):
112
- 104 M. Z. v. Krzywoblcki
(5.2.9)
(S q~,t lSLO
For a steady-condition motion with no energy exchange one has q
l,t
= 0, Ql=
ql = q lm = const. Extenging the range of the pos-
0,
sible values of
q t up to the value of the velocity of light , we see
that this extreme case corresponds to the Einstein special theory of . . -1 2 -1 2 . 2 2 2 2 relatlvlty. 2 qt = 2 c t wlth qt = (dx/ dt) +(dy/ dt) +(dz/ dt) or 2 2 2 2 (dx) + (dy) +(dz) = cmt(dt) .
r
In a stationary motion Eq. (5.2.6) can be written in the form: (5.2.10) where the term
r2 refers to all the remaining terms. By the physical
definition of the velocity one has q2
= (dx/dt)2 + (dy/dt)2+(dz/dt)2
and Eq. (5.2.10) can be remodelled into the form: (5.2.11)
5. 3. Transformation Equations Assume a coordinate system (x, y, z, t) and the metric: (5.3.1)
2 2 2 2 (dx) + (dy) + (dz) - f(dt)
= 0, f = f(x, y, z,
t),
which suppose to be invariant under the transformation of coordinates of the form: (5.3.2a)
Xl =~(x
(5.3.2b)
yl
=Y
~
- Ut)= A
,
Zl
=Z
;
113
=c:J,.
(x, t) ;
- 105 -
M. Z. v. Krzywoblocki
(5.3.2c)
t =~
tl=p+ ~t=B;~=~(x,t);
(x, t) ,
U = U (x, t) is a given function. From the condition:
where (5.3.3)
(dx)2 - f (dt)2 = (dXl)2 - f' (dt,)2 ,
inserting Eqs. (5.3. 2a, b, c) into Eq. (5.3.3) with: , etc.,
(5.3.4) we get the relations : (5.3.5a)
(A, )2 _ fl (B, )2 = 1 ;
x
x 2
2
(5.3.5b)
(A't) - f'(B't)
(5.3.5c)
A, A, - fiB, B, = 0 , txt x
= -f ;
which imply that : (5.3.6a) (5.3.6b) f, fl being known and given, Eqs. (5. 3. 6a, b) allow
With the functions
one to find solutions for A and B . Consider the quadratic form: (5.3.7)
2 (d"l;') =a .. dx.dx.=O, (ij=l, 2, 3, 4), lJ 1 J
and a transformation of C'oordinates : (5.3.8)
T : (x.) = x.( x!) ; 1
1
1
T
-1
: x! = x! (x.) , 1
which implies that : (5.3.9)
dx~ =( ox'./OXj) dx .. 1
1
J
J
114
1
1
- 106 M. Z. v. Krzywoblocki
Suppose: (5.2.10)
a
dx dx = a I dx l dx l ijij ijij
=0 .
Inserting Eq. (5.3.9) into Eq. (5.3.10) furnishes:
'0 x!
() X.I
(5.3.11)
(alt~ -a~j ()x~
()x:)dX( dxl'
=0,
which implies that: (5.3.12) with
ad.~
a ' ij = a lji , Ad..p = A~cl .
= a \'101. ,
The system of partial differential equations (5.3.12) contains tions and 24 functions (
ad..~
= 10, a\j = 10, xli = 4
10 equa-
). The additio-
nal equation is given by relation (5.3.7) or (5.3.10j. Assume that: (5.3.13)
all
= a 22 = a 33 = 1 ; a 44 = knwon function; a ij = 0; if j ;
also: (5.3.14)
a l =a l =a l =0 12 13 23 .
Espressions (5.3.10)
, (5.3. 12) give us a system of 11 equations in
11 unknowns (a I . a I . a I . a I . a I . a I . a I . Xl . Xl . Xl . Xl ) 11' 22' 33' 44' 14' 24' 34' l' 2' 3' 4 . The corresponding expressions A .. are given in the Appendix. IJ We may consider some special cases:
(I)
Let x = Xl . X = Xl which implies that: 2 2' 3 3'
(5.3.15)
()X'2/ihd.
(5.3.16)
OXd,. /Ox 2
= 020. ;
=6~2
;
ox
OXr}.,/
l /()
()x 3
115
x~
=
= 00.3
03~; .
c:A.= 1,2,3,4;
- 107 M. Z. v. Krzywoblocki
Eqs. (A 1) to (A 10) give:
-a ' (Vxl/ox)2 44 4 1
(5.3.17)
(5.3.19)A 33
=
1 - a l33
A44= a 44 - a~l
= 0,
a'33 = 1 ;
('0 xII / OX 4)2
(5.3.20)
- 2 a l14 ( ()
-a ' ('()x'/ux) 44 4 4
(5.3.21) A12
=
A14=
A13
= A23
= 0;
2
x~/
() x4 ) (
ux'4/ d x4 )
=0;
= 0;
- all (u x \ / () xII) (0 x\
/ () x4 ) -
(5.3.22)
(5.3.24) A3: - 2 a l34 (ux'4 / UX 4 ) = 0, a 34
= O.
Eqs. (5.3.10), (5.3.17), (5.3,20), (5.3.22) (call it system I) are four equations in five unknowns: a ' . a ' . a ' . x' . x' 11' 44' 14' l' 4' Obviously in this case we have : (5.3,25) Eq. (5.3.10) in system I has the form:
116
- 108 M. Z. v. Krzywoblocki
(5.3.26)
+ 2 a l14
(dXI) (dx l
1
4
)
=0 .
The Jacobian of the transformation:
o
(5.3.27)
(x~) J =()(x\
= (vx l / ~Xl) (0 xli ()x 4) - (vx l/'()x 4)(ox 4/{)x l ) f o.
1
Let us consider Eqs. (5. 3. 17), (5. 3. 20), (5. 3. 22) as a system of algebraic equations in three unknowns: all' 2a{4 ; a44 ; then the determinant of their coefficients cannot vanish. In. reality, we have:
(Ox D
l / U\)2
=(i)xl/ {)x4)2
1/() xl )( C)x4/ Uxl)
(0 xli '0 Xl )2
(C)xl/'O x4)( (jX I4/ Vx4)
(ox 4/0X 4)
(Jx
(;)xl/ Vx l )( Vx l/ ()x4)
2
(0 x l/ Ux l )( 0x4/ () x4) (0 x4/U xl)(o xi()x 4)
(5.3.28) = - J2 (c)xl/axl) (ux 4/0 x4 )
f
O.
5.4. Transformation Based Upon The Principle of The Invariance of The Total Energy Space-Time Metric Below, we derive the transformation equations, which may be considered to be a some sort of an analogy to the well-known Lorentz transformation. They are based not upon the principle of the invariance of the light space-time metric, like in the Einstein special theory of relativity, but upon a completely new principle, i. e., the principle of the invariance of the energy space-time metric, expressed in four dimensional space
117
- 109 M. Z. v. Krzywoblocki
- time coordinate system. This principle states: Principle of The Invariance of The Energy Space-Time Metric: "let us express the energy equation, Eq. (5.2.11), in the form: (5.4.1)
(dx)
2
2
+ (dy) + (dz)
2
2 2 - I (dt) = 0 .
Then the energy space-time metric (5.4.1) is invariant under any group of transformations of (x, y, z, t) . " Assume the transformation equations: (5.4.2)
Xl =~(x-Ut) = A ; yl= Y ; Zl= z;
tl =
rx+pt=B,
where, in general U=U(x,t). Inserting Eqs. (5.4.2) into Eq. (5.4.1) furnishes the result : (5.4.3) (dX,)2 _ p2(dt,)2 = (dx)2 _ I2(dt)2 , which gives:: (5.4.4a) (5.4.4b) (5 . 4 . 4) c
(A )2_ p2(B )2 = _ 12 't 't .
Since the function
U is known and given, the system(5. 4. 4a, b , c ,re-
presents a system of three equations in three unknown functions d,.,
~
'/. Recombining the above system furnishes : (5.4.5a) (5.4.5b) Eq. (5.4. 5b) gives the equation for
~
118
which may be solved by some sort
- 110 M. Z. v. Krzywoblocki
of an iteration process:
~2[[(Ut)'tJ2 - r2 [(x-Ut),J 2J= (c{'x)2 (x_Ut)2 r2 -
(~'t)2(X-Ut)2 + 2c\GI.'t(Ut),x(x-Ut) +
(5.4.6) + 2 d..
r+-, xr
2
(x-Ut)(x-Ut),
2
- r ; ~= ~(x,t).
x
Eqs. (5.4.4a) and (5.4.4c) provide a simultaneous system of equations for ~
and '( 2 l2 2] -2 ~ = (A) - I (1') - (
1, t) 2 x 2 -
-2~~'t-(~'tt)2;
(5.4.7)
~
2
2 ] -2 = l(A, x) - 1 (1') - 2
2
t, t
x (~
t), t -
~=~(x,t);
rr 'xx-( ( 'x) 2 x 2
(5.4.8) The inverse transformations can be calculated from Eqs. (5. 3. 2a, b, c. ). One gets:
~x,+~Ut')((1.~+~~U)-l
(5.4.9a)
x=(
(5.4.9b)
t:: (~t-r't'Hck~+-~r\J)
·1
Assume the forms :
~I[ Xl
(5.4.10a)
x::
(5.4.10b)
y::yl
(5.4.10c)
t:: -
+ UI(XI, tl) tl] ;Z::ZI
t' Xl + ~I t l
,
with:
119
- 111 M. Z. v. Krzywoblocki
(5.4.11)
d.'=cI..'(x',t');
t'
(3'=P'(x',t ' );
A comparison of Eqs. (5.4. ga, b) (5.4.12)
(5.4.13)
t=~[~(~+~V)-1
=t(x',t').
and (5. 4.10a, b, c) furnishes:
r V)r 1
~' = (~
V'
~ V)
+
=d.~-1
-1
V .
When applying the above transformation equation to the special theory of the relativistic energodynamics, one has to assume that the velocity
= const. In many problems of the relativistic character the both
V
axes,
x
and
x' , may coincide and the velocity
V
may have only
x-component. This simplifies enormously the formal calculations, given above. 5.5. Generalized Transformation Equations For Velocity From Eqs. (5.4.2) one gets the transformation equations for the velocity components: ~,
=
dre( (x-Ut)] _ r.' d[yx + ~t] - L'\x
,
d
+ III x - tit (~Vt)
[rc + yx +
(5.5.1)
(5.5.2)
y'
= dy/dt' = Y(dt/dt' )
(5.5.3)
i'
= dz/dt' =
(5.5.4)
dt/dt' = dt/d(
z(dt/dt' Yx+ Pt)
ddt
J.
(~t) J -1
) =[
t x+ i x+d( ~ t)/dtJ -1,
where dot over an unprimed symbol denotes the total derivative with respect to time
t and dot over a primed symbol denotes the total
120
- 112 M. Z. v. Krzywoblocki
derivative with respect to time t ' , i.e., d/dt and d/dt ' , respectively. From Eqs. (5. 4.10a to c) we obtain the reciprocal
~ = d [d. 1 (x' t Ultl)J/d[ -
=[~lxltJ..IXI
(5.5.5) (5.5.6)
Y= dy'/d(•
,[
(5.5.7) z=z' (5.5.8)
t d(
txlt
•
t x' t
~ Itt)
iIUltl)/dt')/dt~[ - ~'X'_i'x'td(
~'t')
= yl[
-1
Wtl)dt']
_1'~1 - ~ 'x't d(~ 't')/dt'J
•
-l'x'-l'x'+d(~'t')
dt'/dt = [ -
equations:
-1 ;
/ dtJ 1,-1 ;
t 'x' -('x' + d(~ 't')/dt'] -1
5.6. Generalized Transformation Equations For Acceleration We easily get: (5.6.1)
x= dx/dt = d[
/dt '
dt'/dt,
and
x' = d 2x/dt 2 = d \d ~'(XltUltl)J /d~.
dtl/dtl /dt =
= dt l/dt'di(dt/dt l )-1 d[d..I(XltU'tl)] /dtl)
/dt ' =
t
= dt ' / dt· (dt/ dt') -2 dt/ dt ' . d 2[ d... ' (x' t Ultl)] / dt ,2 -(d[cil(xl t Ultl)] /dtl) d 2t/dtI2} = = (dt/dt l)-3{ dt/dt ' . d 2[ c{l(XltUltl)] /dtI2_ (5.6.2)
- d[d..'(xl+Ultl)] /dt ' . (d 2t/dt I2 )}.
121
- 113 -
M. Z. v. Krzywoblocki
Now:
=~'(x'+ U't') + d..,(x'+iJlt'
(5.6.3)
dl<:\.'(X' + U't')] /dt'
(5. 6.4)
d2~,(X' + U't')] /dt,2=d..'(x' + U't') + 2d.'(x'+U't'+U')+o..'(~'+U't+2iJl).
+ U' ) ;
From Eq. (5.5.8) : dt/dt'=-
(5.6.5)
¥'x'-
~'x'+ ~'t'+~'
2 2 " d t/ dt' = - 'x' -
t
=-l 'x'
(5.6.6)
It...
tx' - fx'
.,.t
•
•
- ~ 'x' + ~ It' + P' + ~' ;:
- 2 ~ 'x' - t Ix' + ~ It' + 2 ~ I
.
Inserting Eqs. (5.6.3) to (5.6.6) into (5.6.2) furnishes:
x =~(- ¥'x'-
¥IXI +
~
It I +
i3'f*'(X'+U't')+2~'(X'+U't'+U')
+ ol'(x'+ U't ' +2U')] - [d..'(XI+U't') + oL I
(x l+ Ult'+ UI)] .
(5.6.7) Similarly: (5.6.8)
Y= dy/dt = dy'/dt' . dt'/dt
y = d2y/dt~ d(dy'/dt'.dt'/dt)/dt = = dl(dY'/dt')(dt/dt,)-l] /dt' . (dt'/dt) = = dt'/dt· (dt/dt ,)-2 { (dt/dt ' )(d 2Y'/dt I2 ) - dy'/dt" d2t/dt ,2 } (5.6.9)
= (dt/dt ,)-3{ dt/dt " yl - Y'd 2t/dt I2 }.
122
- 114 -
M. Z. v. Krzywoblocki
Inserting Eqs. (5.6.5), (5.6.') into Eq. (5.6.9.) furnished:
y =[(- ~ 'x' _~/X' +~ It' +~ ') y'- y' (- ~'x' - 2 ~ ,~, - t'x' +~ It' + 2 ~'~
(-
tx'- t'x'+~'t'+ ~'f3
And:
z=l(-
t'~'- ¥'x'+~'t'+ 3')z'-
- z' (- ~'x' - 2 ~'x' t ~'t' + 2 ~J_~I)(I)] (5.6.11)
. (- tx' -
. ~'x'
+
. ~Itl
+13 ')
-3 •
5,7. Generalized Special Relativity and Mechanics. The Dynamics of a Particle In this section we discuss the dynamics of a particle, Assume two particles moving in the system
S' before a head-on collision
with the velocities + u' and -u' parallel to the x-axis in such a way that a head-on encounter can occur, By hypothesis the two particles are perfectly similar and elastic; it is evident that they will first be brought to rest after the collision and then rebound under the action of the elastic forces developed, moving back over their original paths with the respective velocities
-u I and +u I of the same magnitude as
before but reversed in direction, In this system of coordinates the collision is obviously such as to satisfy the conservation laws of mass and momentum, Let us now change to a second system of coordinates relative to the first in the
S moving
x- direction with velocity (-V). From the
section 5,4 we have the following results :
123
- 115 M. Z. v. Krzywoblocki
(5.7.1)
T: x'=d.,(x-Ut); T
-1
: x=
'x' + ct' U't';
;t'=tx+
~t
Y = y'
z = z'
(5.7.2)
where
01..
y'=y; z'=z
~,
d..',
,
r' , U' are given by :
(5.7.3)
(5.7.4)
~'= ((3+~U)
(5.7.5)
t
=
-1
;
td..-1 (~+ y U) -1 ;
(5.7.6) Assume now that: (5.7.7)
U=
tv;
U' =
J' V, I
where
V is the velocity of 8' with respect to
8
and
0 , 6
functions of (x, t) . Then we get: (5.7.8)
x'=.i(x-OVt); y'=y
z'
= z; t'
=Kx+~t,
which furnishes: (5.7.9)
x=ci.'(x'+ o'Vt'); y=y';z=z'
and
~o:C1(~+X6v)-1,
(5.7.10)
1:1(.'=
(5.7.11)
O'v=d.~-16V , 124
;t=- ~'x'+~'t',
are
- 116 M. Z. v. Krzywoblocki
or: (5.7.12)
.r
~ 1 =( ~ + 0u V)
-1
;
t
r
-1
= K~ (~ + K0 V)
-1
;
JI
r:t-1J
= rJ.. r"
•
For velocities we get from Eq. (5.7.9) :
x = d[ ~1(XI+
(IVtlij /d(- "IXI + ~Itl) =
=[ d (ct' x')/dt' + d( Q\.' ,
= (~'x' +ci'x'
0 'Vt')/dt'~
'!
[d(- tx')/dt'
+d(~'t')/dt'l1
l
+ ct' u 'Vt' +cL'a'Vt' +
(5.7.13) K'x ' + ~'tl) = yl(- ('x' -
t'xl + ~ It' + ~')
-1
(5.7.14)
Y= dy'/d(-
(5.7.15)
• / ,.., (.1-1 z=dz l d(_Y'X'+(3lt')=Z'(- r'x'- tx'+~'tl+ I-I)
;
, et c.
o
We proced now with the head-on collision. Let u1 and
u2 be the ve-
locities of the two particles in the S system before the collision, and let m 1 and m 2 be the masses of the two particles before the collision. Furthermore,
let us denote by M the sum of the masses of the two
particles at the instant of the collision when they have come to relative rest, and are hence both moving with the velocity + V with respect to our present system of coordinates} S .
In accordance with the conservation laws, which must also hold in this new system of coordinates, the total mass and total momentum of the two particles must be the same before collision and at the instant of relative rest, so that we can evidently write.
125
- 117 M. Z. v. Krzywoolocki
(5.7.16)
m +m =M' 1 2 '
and (5.7.17)
m 1u 1 +m 2u2 =MV.
From Eq. (5. 7.13) we have: (5. 7.18)
x=[{ ~IXI+ ~IXI+ d{ ~I 6 'Vt')/dtj {_ ~ IXI_ txl+ ~ It I + ~,)-1;
hance, for our problem, we have for the particles 1 and 2: (5.7.19)
u 1 = f~'x'1+a(lu'+[d(ol'
oIVtl)/dt~l}(- fIxl- tl UI + ~ltl+ ~1)-1
(5.7.20)
u2=1~;x;-",;u'+[d(oI.' O'Vtl)/dtl)~{-
i;x;+ r2u'+~lt+
~;)-1,
where: (5. 7. 21a)
0(1
(5. 7. 21b)
~
(5. 7. 21c)
1
=
I = 1
(\,1
(Xl
tl)
0(.
I = o(.l{XI , t;) ; 2 2
1),
~
I = fl'(x ' , t;) ; 2 2
l' 1 '
~I{XI
1
t 1 = XI{XIl '
,t
tl) 1 '
¥;=
t{x; , tl)2 '.
J =[ d{~' o'Vt')/dt') XI=XI
(5.7.22) [d{oL ' o'Vt')/dt ' 1
tl= til
1
(5.7.23)
[d{d,'6IVt')/dt']2=[d{o(,'6IVt')/dt~
xl:x l • t'=t f
2
Rearranging we get: (5.7.24)
u l{- vlu -d- I ) -[d(ct.'OIVt')/dt,l =d..IXI+U (VI Xl- B't ' - ~'1); ~11 1 J I l l 1:1111"11
(5725) • •
u'{Y'u +,J,)_fd(,.J '("Vt')/dt'] 2=,j'xl+u 0 2 2 "I. 2 l "\ a \1\.2 2 2(VIX' () 2 2 _~'t'_r.!') : 2 2 \J 2 .
126
- 118 M. Z. v.Krzywoblocki
Let
V = Const. then Eqs. (5.7.24), (5.7.25) can be written as:
(5.7.26)
u l(-
~lU(d..l)-v[d(o..'6It')/d.~1ci:lxl+u1(~ l x l- PIt l- ~1);
(5.7.27)
u l( t
~u2 +d, ~)- V[d( ~ 16' 't')/dt ']
2= ~
t
~x~ +u 2( ~x2- F2t 2- ~2)'
Consider Eqs. (5.7.26), (5.7.27)as a system of linear algebraic equahons in the unknowns
u I , V, and solve for
V:
_~I
c(,'xltu (V-IXI _~Itl _~I)
\flu + ~I
CA.lxltul(~IXI - Rlt l -RI)
_VIU
alII
o2 2
111°11
11
1
2 2 2 2 0 2 2 '·2 2 \- 2 V=r-------------------~~--· - V I U - 0(..1 - [ d( Q,. I rJ It 1)/ dt I11 o1 1 1 VIU+d.. 1
02 2
=f (-
2
-[d(d.,'C't')/dt '] 2
VIU _ti,I) L{,X' +u (Ylx l • Altl 1 L22 2°22 r21
tOll
.~I)J
'-2
•
_( vlu +cLl)rd.'xl+u (VIXI - 13 It I _AI)] 622 2L11 1'11 111"1 1
(5.7. 28)
.~( ~1 u 1+ ~ 1) [d(
ci,
I O't')/dt '] 2+(
or 2~+ :t 2)[d( ~ I 6Itl)/dt~lrl
Next, from Eqs. (5.7.16), (5.7.17) we get: (5.7.29)
m u + m 2u 2 = (m + m ) V , 1 1 1 2
or dividing by (5.7.30)
m 1m 2
m and rearranging:
2
-1
= (V-u 2 ) (u 1 - V)
-1
.
Inserting Eq. (5.7.28) into Eq. (5.7.30) furnishes:
127
- 119 -
M. Z. v. Krzywoblocki
m /m 1 2
~~
~
~
~ 2'J
= (- V I U - I )[ci.. I XI +u ( VI X - It I - 1)1 - ( V I U +,tl). DIll
222(12222
.[d.11Xl1+u 1('Ii 1I Xl1- '"ill1t 1l - r 1.I );j lf ( A
0
+(Vlu +<1 I )[d(cA,I 022 2
J}-l
Olt l)/dt I 1
02.22
I·_d( 01" I 6' It 1)/ dt I) 2+
V I U + /1,1) 01 1 1
-uJ[u _[(_VIU ~ 1 011
_d-I)r~,X + 2~22
+u (YIXI-Bltl_AI)]_(v'lu +c\,I)rd,lxl+u (VIXI-P.ltl_~I)l}. 2 0 2 2 ,- 2 ,- 2 0 2 2 2 L 1 1 1 0 1 1 ,- 1 1 '-I 'J [('/lu +0(..1 dd(i'o't')/dt,l +(tlu +d,' )
(5.7.31)
[ °1 I l L
J2 02 2
2
rd(~"hl)/dt'J(-lJ-1 ~
l'
We may try to obtain the results of the classical special relativity for m/m2' let: (5.7.32)
<1..1 = c:I.. 2 = (1- V2/ c 2)-1 / 2 = const. =
(5.7.33)
~1
" t2 = - Vc -2 (l-V 2/c 2)-1/2
(5.7.34)
(3~
=
~2
2 2 -1/2
= (l-V /c )
= ct l = const.
Eqs. (5.7.19) and (5.7.20) become:
-2
(5.7.36)
u 1 = (u'+ V)(Vulc
(5.7.37)
u2 = (-u'+V)(- Vulc
-2
= -«"'Vc
+ 1) ; +1).
128
-2
= const. ;
- 120 M. Z. v. Krzywoblocki
From Eqs.(5.7.36)and (5.7.37)we get: (5.7.38)
(I-U: c- 2// 2 = (I_V 2c· 2)1/2 (l.u ,2 c- 2// 2 (1+Vu ' c- 2 )-1 ,
(5.7.39)
(l_u 2 c -2//2 = (1- V 2c -2//2(I_U,2c -2)1/2 (1- VUIC -2)-1. 2
From Eqs. (5.7.38), (5.7.39) we get: (5.7.40)
2 -2 1/2 2 -2 -1/2 -2 -2 -1 (l-u 2c ) (l-u 1c ) =(l+Vulc HI-Vu 'c )
Inserting Eqs. (5.7.36) , (5.7.37) into eq. (5.7.35) gives: (5.7.41)
-2 -2-1 m / m = (1 + VUIC ) (I-Vu lc ) .
1
2
Again, inserting Eq. (5.7.40) into Eq. (5.7.41) furnishes (5.7.42)
2 -2 1/2 2 -2 -1/2 m/m 2 = (l-u 2c ) (l-u{) ,
which is the result of the class·ical theory of relativity. To express the moving mass in terms of the rest mass we set: (5.7.43)
u 2 =0,
m 2 =m O '
m =m 1 '
u =u 1 .
Eq. (5.7.31) becomes:
[ft (-~ ;uHnu+. . ;i[
m/m = o
.[u (5.7.44)
•
V
I)
-~ 2I [d.. 11Xl1 +u( aVIlXI.Qlt~. l "'1"1 d( ,l.' 6't ')dt~, +
I u- Q,1)(d.. I Xl) 1 1 2 2
+~; U-o.;)(.).,X')-
h;u+o.;)
0.' ,
[d.-;x;+U( (;x; -~; t; - p;)n·
[d(q,' 6 't'r t'],
129
+d.,[ d(eV 6't'~t'1 J-] -1,
- 121 -
M. Z. v. Krzywoblocki
or
Next, omit the subscripy 1 remembering that the quantities without subscript belong to particle I, and replace the subscript 2 by the subscript 0, remembering that hese quantities belong to particle 2 :
1ci";[
m " mo[u [( tj u+d.') [d( d.' 6:")/ dt (5,7.46)
.!!-!
'u- d. ')("-; x;)- d,; [d..'x'+u
d (,,' 6
",)/~,~},
(r 'x' _p"'_ ~ '~-1_1]
-1
Special Cases: We try to investigate Eq. (5.7.46) for the case (a)
Let
eX. , ;::
f\. '0
;::
const.
p' ;:
~'0
;::
const.
t' ;: t 0
6' ;: 6'0
when~u-10:
'" const. ;::
const.
Eq. (5.7.46) becomes:
m" mo[u[( tu+ .. ')0[' 6' +
130
- 122 M. Z. v. Krzywobloeki
(5.7.49)
ct. 1 = (I-V 2-2-1/2 e)
( ,Vi = 1,
tl= _IX,IVc
-2
~I
,
= c{ I
•
Inserting Eq. (5.7.49) into Eq. (5.7.48) gives: (5.7.50)
-1
m=m (uct l (_t(.IVC
-2
o
)+2C(.'e(.'
-1
-1)
-1
-2-1 =m (I-Vue) , 0
and (5.7.51)
lim m = mo' u~O
(b)
This result can however, be found to be true under more general
assumption, i. e., we assume (5.7.47) and : (5. 7 . 52)
~I
0I
= ~I .
With Eq. (5.7.52) inserted into Eq. (5.7.48) one gets: (5.7.53)
r -1.1 m=mo (1+utu l ( ~I) )
or (5.7.54)
m
= mo
(1+u «.1
-1
~
I)
-1
,
and (5.7.55)
lim u~O
m = m
o
5.8. Three Kinds Of Fundamental Laws Conservation considerations, even if not expressed in the forms of final laws, used to play an important role in the early physics. It is sufficient to mention the names and works of Galileo, Newyon. The beginning of the 20th century witnessed the appearance of the Einstein
131
- 123 -
M. Z. v. Krzywoblocki
postulates about the symmetry of space (equivalence of directions and different points in space) . The problem of stability and optimization of dynamic systems played undoubtedly an important role in the thinking of the physicists in previous centuries. However, those considerations were evidently either not yet precisely elaborated or not thought to be particularly important. Actually, a great impact towards considerations of this character was furnished by the development of the calculus of variations in last centuries. It seems that today it should be firmly established that actually every energetic system we deal with should be in equilibrium (stable)/ The recent broad application of the calculus of variations (or of the Pontryagin maximum principle which is equivalent) in engineering, mechanics, physics, etc., seems to indicate that the optimization requirements in problems in which we are looking for an analytical representation of the energetic system in question (by means of solving a differential system which describes mathematically tre energetic system) are lot less important than the requirements superimposed upon the system by the symmetry and conservation laws. Thus, it seems that the optimum energy principles , as we, shall call them, which
require'
that a certain kind of energy ( for illustratuve pllrpose we may consider the potential energy) of the dynamic system in question, which is in an
equilibrium state, or at least in a quasi-equilibrium state, be
a minimum (optimum for the stability conditions), represent a third kind (with symmetry and conservation laws) of laws which have to be used to determine analytically the status of the energetic system in question. There arises the question of the examples of the energetic systems
132
- 124 -
M. Z. v. Krzywoblocki
to which one can and should apply all three kinds of laws, discussed above. We may briefly mention three such cases: (1) Assume a relativistic multi-fluid magneto-hydrodynamics (hypersonics) ; it consists of at least two fluids; electron-fluid and ion-fluid. It may consist of three
electron -, ion -, and neutral
- fluids. In the
case the velocity range would be of the order comparable to the order of the velocity of light, the relativistic phenomena should be considered. In this case the equilibrium state of the three -or multi-fluid of the magneto-hydro-dynamic character could be considered as a superposition of three one-fluid systems, each being in an equilibrium state in itself. Obviously, the above three kinds of laws, discussed above, could and should be applied to such a complex system. (2) A possible model of the structure of the universe. (3) In the case one neglects the transformation of the time-coordi-
*'
nate,
i. e., t=t , one deals only with the space coordinates. But obviou-
sly , in such non-relativistic system one can preserve all three kinds of laws mentioned above. It seems that they should be preserved in all the complex systems, the "classical"
which interfere one with each other. Thus,
energodynamics can be applied to hypersonic multi-
fluid magneto-hydro-dynamics in a nonrelativistic region. Assume in the space-time R an energetic system in a closdd domain D. Actually, one may consider open infinite domain, of
L
is a
D
00
I
contained
L referring to an
, under the condition that the total energy If
finite one. Similarly, in the space-time R there is located
*"
\~ L- in D . We seek a correspondence between
\" L-
and
L* . To this
end we require the following set of laws to be preserved: (1) The conservation laws (Newton) referring to the conservation
133
- 125 -
M. Z. v. Krzywoblocki
*'
of momentum, mass and energy, in Rand
R,
i. e., expressed in
four dimensional space-time (or only in space) ; (2) The symmetry laws (Einstein) expressing the independence of the equations, governing the dynamic system in question upon the tran.
'1t
sformations of the space-time coordinate system from R to R and vice versa; (3) The stability laws expressing the equilibrium state of the dynamic systems in
Rand
R* .
We require that the set of laws (1), (2), and (3) be satisfied for any system of space-time coordinates under any arbitrary group of transformation of the four coordinates stant quantity, where respect to
R H R* with
U being a con-
U denotes a relative velocity of
R* with
R.
The principle of the invariance of the total energy space-time metric furniches the transformation equations between
Rand
*
R . The
remaining conservation laws, i. e., these of momentum and mass, may be transformed from R to R* by using the transformation of coordinates obtained above. The main reason why the equation of the conservation of energy is used as the tool for furnishing the transformation of coordinates is that this is the only equation which contains expressions for all the kinds of energy in the system in question . Equations of the conservation of momentum and mass are not of this character. Moreover, in the case of a transformation of mass into another kind of energy (light, say), the equation of the conservation of mass loses its meaning, unless we mtroduce the concept of some sort of souces or sinks of energy (mass). The same remark refers to the equation of the conservation of momentum (sources of momentum) .
134
- 126 M. Z. v. Krzywoblocki
5. 9.0ptimum Energy Principle We may discuss more thoroughly the third kind of laws governing the phenomena of the mechanical nature in a dynamic sub-system, i. e. , the stability laws expressing the equilibrium state of the dynamic subsystem in question. Actually, very often in the classical fields of the mathematical physics (fluid dynamic including gas-dynamics, aerodynamics, kinetic theory of gases, etc.) the third kind of laws is not taken into account at all. It is usually tacitly assumed that the dynamic system in question is stable. Moreover, it is often assumed that the systems in question extend up to infinity (for example, in the case of a flow of a gas around a body the usual assumption is that the incoming stream extends in all directions up to infinity). This implies that there does not exist the problem of the influence and of the interference of the neighboring sub-systems with the sub-system under consideration. But, if one wants to consider an ensemble of sub-systems, interacting one with each other, then actually the stability conditions should be considered. To different particular problems, different criteria may and should be applied, always expressing the condition that the energy distribution is such as to assure the equilibrium of the sub-system. For illustrative
purpose, one may require that the potential energy of
the sub-system should be a minimum. In the case we consider a system composed of a few sub-systems an analogous variational principle should be proposed . The optimum energy principle leads to a variational principle of the form (this should be (5.9.1)
r jI(x,
P. E. dv = I = min., or y, z, t)
135
treated as an example only):
¢ 1=
0.
~
127 M. Z.. v. Rrzywoblocki
The author presented above the principles of the relativistic (and classical) energodynamics as applied to the
multi~fluid
relativistic
hyper~
sonics of an ionized gas. The fundamental equations and concepts were derived and the three kinds of laws governing the behavior of a dynamic system of a continuous medium were discussed . A particular attention was paid to the optimum energy principle, so important in composite systems for the stability considerations. Mathematically, it is expressed in form of a variational principle. 6. Concluding Remarks The author presented some techniques of solving differential equations occuring in the field of hypersonics of both neutral and ionized gases. But there are many indications that the practical aerodynamics may follow a different path from those described above. Let us discuss briefly a hypersonic flow around a thick, blunt body. Concerning the suband tran-sonic regions, it seems justified to assume that the best methods available at the present time (for the flow around a blunt thick body) are those which deal directly with the original forms of the equations of conservation of momentum, mass and energy. Thus, these equations should be subject to direct programming without any previous significant remodelling them. This implies that the process of programming becomes the most important one . In the supersonic domain the application of the theory of characteristics seems to be generally acceptable (with a combination with numerical methods). The problem of programming and computing of a general three-dimensional flow-pattern around finite bodies seems to be a good example demonstrating the necessity of organizing large centers of numerical analysis
136
and
- 128 -
M. Z. v. Krzywoblcoki
and computing. The author would like to emphasize that at the present status of affairs a derivation of new equations describing the hypersonic phenomena is not important. We have derived in the past a considerable amount of such equations which we are unable to solve up to now. As a matter of fact, we did not solve, as yet, Euler's equations of motion in a three-dimensional flow around Apollo shape with both angles of attack and yaw. The most important analytical aspect of the hypersonic aerodynamics today is the creation of a methodology of solving the differential equations occurring in this field. The problem of importance or unimportance of particular aspects of hypersonics is somewhat similar to the question of the importance or unimportance of a value-free science. A value -free science is considered by many scientists to be an absurd in a strict sense. The science has its own standards by which its statements are tested or evaluated. On the other hand, it is not enough
to state that scientific values are merely methodologica, con-
cerning with means and not ends. An exception to this is if they aid in an objective pursuit of truth. But the truth has many faces. A concept can be true or false in the degree to which it corresponds to the standards of the science itself. The same concept may be good or bad, correspondingly, if it contributes more or less to the basic needs of human life ( or to any other aspect on the earth) . Some scientists believe that science is an idle curiosity; but some believe that it has a responsiblity to serve these basic needs, cited above. From this point of view it seems obvious that the primary analytic problem today in hypersonicaerodynamics is the problem of solving equations occurring in this field. When this cannot be achieved in a purely analytic form, the tool of the highspeed computing machines must be applied.
137
- 129 -
M. Z. v. Krzywoblocki
Acknowledgement
The author expresses his deep thanks to: the Department of the Navy, Office of Naval Research, Washington, D. C"
in particular to
Dr. Ralph D. Cooper, Head, Fluid Dynamics Branch, ONR, for providing the transatlantic transportation; to Bing Fund, Los Angeles, California, in particular to Mrs. Anna H. Bing Arnold, for a grant covering the remaining travel expenses, which grants facilitated the author the trip to Italy to deliver the summer course in Varenna. His deep gratitude is due to the Engineering Research Division of the Michigan State University, East Lansing, Michigan, in particular to Mr. John W. Hoffman, Director, and to the Chairman of the Mechanical Engineering Department of Michigan State University, Professor Dr. Charles R. St. Clair, for providing an assistant for preparing the present paper. Finally, his thanks are due to Mr. H. R. Kim, a Ph. D. candidate in the Space Program of MSU, Instructor in the M. E. Department of MSU, for his invaluable help in preparing the present paper. To Mrs. Edna Harney, Secretary, M. E. Department of MSU, the author is deeply indebted for patience in taking care of all the correspondence and innumerous changes in schedule, etc. associated with the accomplished trip to Italy.
139
- 130 M. Z. v. Krzywohlocki
Appendix A =1 - a l (0 Xl I i) x )2 - 2a l ( () Xl I d x )( () Xl I 1111 1 1 14 1 1 4
(A 1)
-a~2
(0
-a l
( 0 Xl I i) x )2 - 2a l (0 Xl I 3 1 34 3
33
-a 44 (() xli 0 Xl)
-a 22 ( () xV
2
0
x 2)2 - 2a 124
i oxV () x 2 )2 -2a 34
-a4/o x4/0 x 2)2
-ab( (A 3)
2
(()xll1
ox)( 2
oxll ux 2 )4
(0
x21 () x 2)( d xli () x 2) -
(()
x31 0 x 2)( 0 x4/0
2a 14 ( 0 xII
- 2a 24 (
'0
x3)(
"0
2
-a l 33
xii 2
xl41
() x 3) 0 x 3) -
=0;
A ::a - a l (c Xl Iv x )2 - 2a l (0 Xl lox )( 44 44 11 1 4 14 1 4 -a ln (0 42
x41 0 x 3) -
0 x21 0 x3)( () x41
"0 xV V X3)2 - 2a 34 ( () x~1 '0 x3)( ()
-a I ( ) Xl I "0 x) 44 4 3
x 2) -
=0 ;
x~1 0 X3)2 -
-a~2( ox~1 0 x 3)
u x1)( () Xl 4lox)1 -
=0;
-a 3
A33::1-a l1 (0
x )1
x~1 V X1)2 - 2a~4( ~ Xl/U x 1)( 0x~1 "Ox 1)-
A =1- a l (0 xliv x)2 - 2a l 22 11 1 2 14
(A 2)
d
'Ox )2 - 2a l (0 xii 4 24 2
(; Xl4I () x4)-
ux)( 'Oxll 4 4
Ox )4
n' xii3 eX 4)2 - 2a 34l (u xii3 ux 4)( Qxl/ox )4 4 141
- 131 -
.
M. Z. v; Krzywoblocki
(A 4)
(A 5)
-a' (0 x' / () x) 44
-a
44
4
2
4
= O· '
(0 x'4lox 1)( () x'4I U x2) = 0 '.
A =-a' (u 13 11 -a 22 (()
x'/o x )(0 x'l 0 x ) - 2a' (~x' 1"0 x )( (. x'l () x )1 1 1 314· 1 1 4 3 xV 0 x1)( ') xlI
() x3) -
2a 24 ( ()
xV
UX1)( () X.V () x3)-
-a'33 (ox'IOx)(ox'/Ox)-2a' (bx'/Ox)(Ox'/ux)3 1 3 3 34 3 1 ~ 3 (A 6)
- a 44 ( () x
41ax1)( 0 x4/ d x4)
- a' ( Q x' I 22 2 - a' ( 33 (A 7)
= 0;
() x1)( "0 x'2/ '0 x4) - 2a'24 ("0 x'2I '0 x1)( "0 x'41"0 x4 ) -
"0 x'3/ 'Ox 1)( 0 x'3/ 0 x. 4 ) - 2a'34 ('0 x'3/ -0 x1)(0 x'4/ 0 x4 ) -
- a' ('Ox'/ 'Ox 44
A23 -=-a 11 (()
4
(Ox'l 4
1
x~j'o x2)(o
-a 2la x'21
0 x2)(
'Ox ) '" O· 4
'
xl/"() x3) - 2a 14 (o xV d x2)( ()x4/
0 x21 0 x3) -
142
2a 24 (
()x 3)-
()x~ / () x2 )( 0 x4/ () x3)-
- 132 M. Z. v. Krzywohlocki
-a' (~x'I()x )( 44' 4 2
(A 8)
-a' ( ';.
33
x'i t'
. 3
-a'44 (~x'i 4
(A 9)
I
"'\x'i 4 v
x )( (,
2
J
x'i 3
(.x) - 3
= O·'
7' x ) -2a' (-;--' x'i \.0 4 34 'oJ 3
'( x 2)(-(',~ x'i ) 4 \..
x )-
4
0X )( '(.Ix l /-·x ) = O· 2 4 ,. 4 '
-~ 1""1"x)(ox' "" I...."x)-2a'('.:-x' , 1:-',X)(0X' ,~, I '"r/x)--a'(ox' 34 11 1 v 3 1 v 4 14 1 ~ 3 4' 4
A
143
- 133 M. Z. v. Krzywoblocki
List of References 1. Aziz, A. K. , Gilbert, R. P. , Howard, H. C.
: On a Non-linear Elliptic Boundary Value Problem with Generalized Goursat Data. Tech. Note BN-377, Univ. of Maryland, Sept. 1964.
2. Bergman, S.
: Two-Dimensional Transonic Flow Patterns. Amer. J. Math. 70, 1948, pp.856-891.
3. Bergman, S.
: On Solutions of Linear Partial Differential Equations of Mixed Type. Amer. J. Math. 74, 1952, pp.444-474.
4. Bergman, S.
: Integral Operators in the Theory of Linear Partial Differential Equations. SpringerVerlag, Berlin-Goettingen-Heidelberg, 1961.
5. Bergman, S.
: On Value Distribution of Meromorphic Functions of Two Complex Variables. Studies in Mathematical Analysis and Related Topics. (Ed. by Gilbarg, D., Solomon, H" et al.), Stanford Univ. Press, 1962.
6. Bergman, S.
: On Distortion Theorems in the Theory of Quasi-Pseudo Conformal Mappings. Applied Mathematics and Statistics Laboratories, Stanford University, Feb., 1963.
7. Bergman, S.
: On Integral Operators Generating Stream Functions of Compressible Fluids, NonLinear Problems in Engineering. Academic Press, Inc., 1964 (to appear).
8. Bergman, S., Bojani~, R.
: Application of Integral Operators to the Theory of Partial Differential Equations with Singular Coefficients. Archive for Rational Mechanics and Analysis, Vol. 10, No.4, 1962, pp. 323-340.
9. Bergman, S., Schiffer, M.
: Kernel Functions and Elliptic Differential Equations in Mathematical Physics. Academic Press, Inc., New York, 1953.
145
- 134 M. Z. v. Krzywoblocki
10. Blank, A. A., Friedrichs, K. 0., Grad, H.
: Notes on Magneto-hydrodynamics, V. Theory of Maxwell's Equations without Displacement Current. Physics and Mathematics, NYO-6486, AEC Computing and Applied Mathematics Center; Institute of Mathematical Sciences, New York University, Nov. 1, 1957.
11. BUrnett, D.
: The Distribution of Molecular Velocities and the Mean Motion in a Non-uniform Gas. Proc. London Math.Soc., Vol. 40, 1935, pp.382.
12. Chapman, S., Cowling, T. G.
: The Mathematical Theory of Non-uniform Gases, 1st ed. Cambridge University Press, London, 1939.
13. Courant, R .. , Friedrichs, K. O.
: Supersonic Flow and Shock Waves. Interscience Publishers, Inc., New York, 1948 .
14. Eichler, M. M. E.
: On The Differential Equation u +u + + N(x)u = 0. xx yy Trans. Amer. Math. Soc., Vol. 65, 1949, pp.259-278.
15. Epstein, P.S.
: On the Resistance Experienced by Spheres in their Motion Through Gases. Phys. Review, Vol. 23, 1924, pp. 710.
16. Ergelyi, A. et al.
: Higher Transcental Functions, Vol. I, II. McGraw-Hill Pub. Co., New York, 1953.
17. Gilbert, R. P.
: On the Singularities of Generalized Axially Symmetric Potentials. Archive for Rational Mechanics and Analysis, Vol.6, No.2, 1960, pp.171-176.
18. Gilbert, R. P.
: Singularities of Three-Dimensional Harmonic Functions. Pacific J. Math. , Vol. 10, No.4, 1960,pp.1243-1255.
19. Gilbert, R.P.
: On the Geometric Character of Singularity Manifolds for Harmonic Functions in Three Variables, I. Tech.
146
- 135 M. Z. v. Krzywoblocki Note BN-256, AFOSR-1045, Univ. of Maryland, Sept., 1961. Also Archive for Rational Mechanics and Analysis, Vol. 9, No.4, 1962, pp.352-360. 20. Gilbert. R.P.
: A note on Harmonic Functions in (p+2) Variables. Archive for Rational Mechanics and Analysis Vol. 8. No.3. 1961, pp.223-227.
21. Gilbert. R. P.
: On Harmonic Functions of Four Variables with Rational p4 - Associates. Tech. Note BN-274. AFOSR-2252, Univ. of Maryland. Jan,. 1962.
22. Gilbert. R.P.
: Poisson's Equation and Generalized Axially Symmetric Potential Theory. Tech. Note BN-283, AFOSR-2450, Univ. of Maryland, March, 1962.
23. Gilbert, R. P.
: A Note on the Singularities of Harmonic Functions in Three Variables. Proc. Amer. Math. Soc .• Vol. 13, No.2, April 1962, pp.229-232.
24. Gilbert. R.P.
: Some Properties of Generalized Axially Symmetric Potentials. Amer.J. Math. , Vol. LXXXIV, No.3, July, 1962, pp. 475-484.
25. Gilbert, R.P.
: Harmonic Functions in Four Variables with Algebraic and Rational p4 Associates. Tech. Note BN-294, Univ. of Maryland. July, 1962.
26. Gilbert, R.P.
: Composition Formulas in Generalized Axially Symmetric Potential Theory. Tech. Note BN-298, Univ. of Maryland, Oct., 1962.
27. Gilbert. R. P.
: Operators Which Generate Harmonic Functions in Three Variables. Tech. Note BN-306, Univ. of Maryland, Dec., 1962.
147
- 136 M. Z. v. Krzywoblocki
28. Gilbert, R. P.
: Bergman's Integral Operator Method in Generalized Axially Symmetric Potential Theory (u) NOLTR 63-124, United States Naval Ordnance Laboratory, White Oak, Maryland, June 14, 1963.
29. Gilbert, R. P.
: On Solutions of the Generalized Axially Symmetric Wave Equation Represented by Bergman Operators. Tech. Note BN350, Univ. of Maryland, March, 1964.
30. Gilbert, R.P.
: On Generalized Axially Symmetric Potentials Whose Associates are Distributions. Tech. Note BN-356, Univ. of Maryland, April, 1964.
31. Gilbert, R. P.
: Composition Formulas in Generalized Axially Symmetric Potential Theory. J. Math. and Mech., Vol. 13, No.4, 1964, pp.557-588.
32. Gilbert, R.P., Howard, H. C.
: On Solutions of the Generalized Bi-Axially Symmetric Helmholtz Equation Generated by Integral Operators. Tech. Note BN-352, Univ. of Maryland, April, 1964.
33. Gilbert, R.P., Howard, H. C.
: Integral Operator Methods for Generalized Axially Symmetric Potentials in (n+ 1) Variables. Tech. note BN-366, Univ. of Maryland, July, 1964.
34. Grad H.
: On the Kinetic Theory of Rarefied Gases. Comm. Pure Appl. Math., Vol. 2, 1949, pp.331-407.
35. Grad, H.
: Asymptotic Theory of the Boltzmann Equation, 1. Phys. Fluids, Vol. 6, pp.147.
36. Grad, H.
: Asymptotic Theory of the Boltzmann Equation, II. Rarefied Gas Dynamics. Proc. of the Third International Symposium on Rarefied Gas Dynamics, Held at the Palais de l'Unesco, Paris, in 1962, Vol. I; Supplement 2, Academic Press, 1963. pp. 26-59.
148
- 137 M. Z. v. Krzywoblocki 37. Heinema, M.
: Theory of Drag in Highly Rarefied Gases. Comm. Pure Appl. Math., Vol. 1, no.3, 1948, pp.259-273.
38. Henrici, P.
: Zur Funktionentheorie del Wellengleichung Commentarii Mathematici Helvetici, Vol. XXVII, 1953, pp.235-293.
39. Henrici, P,
: Complete Systems of Solutions for a Class of Singular Elliptic Partial Differential Equations. Boundary Problems in Differential Equations. Univ. of Wis. Press, Madison, Wis., 1960, pp. 19-34.
40. Ikenberry, E., Truesdell, C.
: On the Pressures and the Flux of Energy in a Gas according to Maxwell's Kinetic Theory, I, II. J. Rational Mechanics and Analysis, Vol. 5, No.1, Jan., 1956, pp.1-54, 55-128.
41.
Jaff~,
G.
42. Keller, J.B.
: Zur Methodik der Kinetischen Gastheorie. Annalen der Physik, Ser.5, Vol. 6, 1930, pp.195-252. : On the Solution of The Boltzmann Equation
for Rarefied Gases. Comm. Pure Appl. Math., Vol. 1, No.3, 1948, pp 275-285. 43. Kreyszig. E.
: Coefficient Problems in Systems of Partial Differential Equations. Arch. Rat. Mech. Anal., Vol. 1, 1958, pp.283-294.
44. Kreyszig, E.
: On Singularities of Solutions of Partial Differential Equations in Three Variables. Arch. Rat. Mech. Anal., Vol. 2, 1958, pp. 151-159.
45. Kreyszig, E.
: On Regular and Singular Harmonic Functions of Three Variables. Arch. Rat. Mech. Anal., Vol. 4, 1960, pp. 352- 370.
46. Kreyszig, E.
: Kanonische Integral Operatoren zur Erzeugung Harmonischer Funktionen von vier Veraenderlichen. Arch. der Math., Vol. 14, 1963, pp.193-203.
149
- 138 M. Z. v. Krzywoblocki 47. Krzywoblocki, v., M. Z.
: Bergman's Linear Integral Operator Method in the Theory of Compressible Fluid Flow. oesterreich. Ing. -Arch. , Vol. 6, 1952, pp. 330-360; Vol. 7,1953, pp. 336-370; Vol. 8, 1954, pp.237-263; Vol. 10, 1956, pp. 1-38.
48. Krzywoblocki, v., M. Z.
: On the Generalized Integral Operator Method in the Subsonic Diabatic Flow of a Compressible Fluid. Proc. of the IX International Congress of Applied Mechanics, Brussells, 1956, Paper I-II, 1957, pp. 414-419.
49. Krzywoblocki,
: On the Generalized Theory of the Laminar Two-Dimensional Boundary Layer Along a Flat Plate in Continuum and Slip Flow Regimes, I, II. Bulletin de la la Soci~t~ MatMmat. de Grece 1, Vol. 29, 1,2,3, 1954,pp.34-74; III. Vol. 31, 1,2,3, 1959, pp.41-68.
V.,
M. Z.
50. Krzywoblocki, v., M. Z.
: On a Method of Solving the General System of Equations in Magneto-Gas-Dynamic.s. Bull. de la Soc. MatMmatique de Grece, Nouvelle Serie, Tome 1, Fasc. 1, 1960, pp. 63-97.
51. Krzywoblocki, v .• M. Z.
: Bergman's Linear Integral Operator Method in the Theory of Compressible Fluid Flow. (With an appendix by Davis, P., Rabinovitz, P.). Springer-Verlag, Vienna, 1960.
52. Krzywoblocki, v., M. Z.
: On the Fundamentals of the Theories of Relativity. Research Report, R-61-33, The Martin Co., Denver, U. S. A., October 1961, pp. ii + 26.
53. Krzywoblocki.
: On the General Form of the Special Theory of Relativity. Parts I to IV. Acta Physic a Austriaca, Vol. 8, No.4, 1960, pp.387-394; Vol. 14, No.1, 1961,pp.22-28; Vol. 14, No.1, 1961, pp. 39-49;Vol.I4, No.2, 1962, pp.239-241.
V.,
M. Z.
150
- 139 M. Z. v. Krzywoblocki 54. Krzywoblocki, v., M. Z.
55. Krzywoblocki,
V.,
M.Z.
: Special Relativity- A Particular Energy Formulation in Newtonian Mechanics. Part I, II. Acta Physica Austriaca, Vol. 15, No.3, pp. 201-212; Vol. 15, N.3, 1962, pp.251-26.1.
: On the Fundamentals of the Relativistic Theories. Acta Physica Austriaca, Vol. 15, No.4, 1962, pp.320-336.
56. Krzywoblocki, v,, M. Z.
: Ergodic Problem in the Theory of Vibrations and Wave Propagation. Zagadnienia Drag1in Nieliniow., Vol. 4, 1962, pp. 53-76.
57. Krzywoblocki,
: Generalization of Topological Aspects in Existence Proofs to Magnetohydrodynamic s. Fundamental Topic s in Relati vistic Fluid Mechanics and Magnetohydrodynamics. Ed. by Wasserman, R., Wells, C. P. Proc. of a Symposium Held at Michigan State Univ., Oct., 1962. Published by Academic Press, New York, 1963, pp. 91-124.
V.,
M. Z.
58. Krzywoblocki, v., M. Z.
: Some Ergodic Problems in Physical Mathematics. Proc. Mountain State Navy Res. and Devel. Clinic, Raton, New Mexico, Sept. 28-29, 1961, Published by Commun. and Electronics Foundation, Raton, New Mexico, 1963, pp. F35-F66.
59. Krzywoblocki, v., M. Z.
: Operators in Ordinary Differential Equations. J. fuer Reine Angew. Mathematik, Vol. 214/215, 1964, pp.137-140.
60. Krzywoblocki, v., M. Z. Hassan, H. A.
: Bergman's Linear Integral Operator Method in Diabatic Flow. J. Soc. Indust. Appl. Math., Vol. 5, 1957, pp. 47-65.
61. Krzywoblocki, v., M. Z. Hassan, H. A
: On the Limiting Lines in Diabatic Flow. Seorsum Impressum Ex Tom. VI Commentariorum Mathematicorum Universitatis Sancti Pauli, Tokyo, 3, 1,1958, pp. 115-139.
151
- 140 M. Z. v. Krzywoblocki 62. Krzywoblocki, v., M. Z. Roth, H.
: On the Reduction of the Number of Independent Variables in Systems of Partial Differential Equations. (To be published).
63. Loeb, L. B.
: Kinetic Theory of Gases. McGraw-Hill Pub. Co., Inc., New York, 1934.
64. Maxwell, J. C.
: On the Dynamical Theory of Gases.
Phil. Trans. R. Soc. London, 157 (1866), 49-88. (With corrections) Phil. Mag. (4) 35, 129-145, 185-217 (1868). Papers, 2, 26-78. 65. Michal, A. D.
: Differential Invariants and Invariant Partial Differential Equations in Normed Linear Spaces. Proc. Natl. Acad. Sci., U.S. 37, No.9, 1952, pp. 623-627.
66. Mises, v' l R., Schiffer, M.
: Advances in Applied Mechanics, Vol. 1, 1948.
67. Morgan, A. J. A.
: The Reduction by One of the Number of Indepentent Variables in Some Systems of Partial Differential Equations. Quart. J. Math. [N. S] 3, No. 12, 1952, pp. 250-259.
68. Ringleb, F.
: Exakte Loesungen der Differentialgleichungen einer adiabatischen Gasstroemung. Z. Angew. Math. Mech., Vol. 20, 1940 pp. 185-195. Also, Aero. Sci J., 1942.
69.
S~nger,
E.
: Gaskinetik sehr hoher Fluggeschwindigkeiten. Deutsche Luftfahrtforschung, Bericht 972, Berlin, 1938.
70. Schaaf, S. A., Chambre, P. L.
: Flow of Rarefied Gases. High Speed Aerodynamics and Jet Propulsion. Vol. III, Fundamentals of Gas Dynamics (Ed. Emmons, H. W.). Priceton Univ. Press, 1958.
71. Stark, J.M.
: Transonic Flow Patterns Generated by Bergman's Integral Operator. Dept. of Math. Stanford University, Aug., 1964.
152
- 141 M. Z. v. Krzywohlocki 72. Tsien, H. S.
: Superaerodynamics, Mechanics of Rarefied Gases. J. of the Aeronautical Sciences, Dec. , 1946, pp. 653-664.
73. Zahm, A. F.
: Superaerodynamics. J. of the Franklin Institute, Vol. 217, 1934, pp. 153-166.
153
CENTRO INTERNAZIONALE MATE MATICO ESTIVO (C. 1. M. E.)
J. KAMFE: DE FERIET
,
LA THEORIE DE L'INFORMATION
,
ET LA MECANIQL'E STATISTIQUE CLASSIQUE ,
I
DES SYSTEMES EN EQUILIBR.E
Corso tenuto a Varenna (Como) dal 21 al 29 agosto 1964
155
LA THEORlE DE L'INFORMATlON ET LA MECANlQUE STATISTlQUE CLASSlQUE DES SYSTEMES EN EQUILIBRE par ,
,
Joseph KAMPE DE FERIET (Universite - Lille) I - lNTRODU CTlON. Le but de ces lecons est modeste; elles ne pretendent aucunement attirer votre attention par leur nouveaute et leur originalite; elles se proposent seulement de vous montrer, aussi simplement que possible, comment les progres du Calcul des Probabilites et specialement d'un de ses nouveaux chapitres, la Theorie de l'lnformation, permettent d'unifier et de clarifier l'expose de la Mecanique Statistique c1assique, tout en lui donnant, - au moins aux yeux du Mathematicien, - un peu de la rigueur qui lui fait parfois defaut. En effet la Mecanique Statistique, - sous la forme primitive et rudimentaire de la TMorie Cinetique dans les travaux de J. C. MAXWELL (1859-1879) et de L. BOLTZMANN (1872-1898) et sous la forme plus m11re et plus generale de J. W. GIBBS (1902), - est nee
a une epoque
ou Ie Calcul
des Probabilites, sur lequel elle devrait sans cesse s'appuyer, etait encore encombre par les battages de cartes et les jeux de des. Les theoremes limites, outils indispensables de la Statistique,poin-
a peine sous ce fatras, herite de la question du Pascal a propos du jeu de "passe-dix".
taient
chevalier de Mere
a
Commel'a note N. WIENER ~O]P. 897 : "It (la Mecanique Statistique) developpeu without an adequate armory of concepts and mathematical technique, which is only ~ (ceci est ecrit en 1938 ) in the process of development at the hands of the modern school of students of integral theoryll ; elle s 'etait donc developpee I gra:ce
a de geniales
intuitions physiques, sans
Ie support mathematique indispensable "putting the cart before the horse II •
157
- 146 -
J. Kampe De Feriet
A.1. KHINCHIN, (dans Ie premier ouvrage qu1un matMmaticien puisse lire d1un bout ges
~lectriques
a lIautre
sans ressentir, en tournant les"pages, des
d~char
qui vont du picotement agaGant au choc violent) est encore
plus pr~cis et plus s~vere~ [IS] p. 2. II In the first investigations (MAXWELL, BOLTZMANN) these applications of statistical methods were not of a systematical character. Fairly vague and somewhat timid probabilistic arguments do not pretend here to be the fundamental basis ... The notions of the theory of probability do not appear in a precise form and are not free from a certain amount of confusion which often discredits the mathematical arguments by making them either void of any content or even definitely incorrect. The limit theorems of the theory of probability do not find any application as yet. The mathematical level of all these investigations is quite low ... " 11 n1est pas rare, aujourd'hui encore, de retrouver dans des livres d'enseignement la p~riode
r~p~tition
pure et simple des arguments de cette
premi~re
: IIMany of BOLTZMANNls arguments have been repeated by, almost
all, textbooks and,Iam sorry to add, not always in a critical fashion II R. KURTH ~O]p. 76. Apres Illes vues profondes (de BOLTZMANN) probablement claires dans son esprit, mais dont GIBBS parait d lune
lI
l'expos~
admirable
reste si confus ll (p. VIII) Ie livre de
clart~ II
: II comme tout est en ordre et s len chaf
a la traduction francaise
ne" , nous dit Marcel Brillouin dans sa prMace t6bJ. Mais R. KURTH
[20] p. S7 qui admire aussi la clart~ logique (IIGIBBS
book is still distinguished by the clarity of its arguments and presentation II) doit
n~anmoins
conStater"GIBBS ... hardly touched the problems of probabili-
ty connected with the foundations of Statistical Mechanics ". KHINCHIN tIS} p.4 note avec s~v~r'it~ :
158
II
The mathematical level of
- 147 -
J. Kampe De Feriet
the book is not high; although the argument are clear from the logical standpoint, they do not pretend to any analytical rigor II cision la lacune fondamentale
II
•
Il pointe avec pre-
The limit theorem of the theory of probabili-
ty does not find any application (at that time they were not quite developped in the theory of probability itself)". Cette derniere remarque est d'une grande importance et transforme la critique en louange: la notion d'ensemble representatif (deja utilisee par J. C. MAXWELL) joue un rtlle fondamental dans l'expose de GIBBS; elle ne trouve sa justification mathematique que dans la loi forte des grands nombres ; or GIBBS ecrivait en 1902 et la premiere demonstration de cette loi n'a ete donnee, pour un cas tres particulier, par Emile BOREL,qu'en 1909 ; il a fallu attendre jusqu'a 1930 pour la demonstration generale de A.N. KOLMOGOROV. Aux regrets de l'absence d 'un cadre mathanatique rigoureux se
m~le
donc notre admiration pour Ia
clarte des intuitions de GIBBS. Le noeud de la Mecanique Statistique des systemes en equilibre statistique reside dans la distribution canonique de GIBBS; Ie but de ces
le~ons
est de vous montrer comment la Theorie de l'Information, a la fois sous I'aspect que lui a donne Sir Ronald FISHER (1925) et sous celui de Claude SHANNON etNorbert WIENER (1948), eclaire d'un: jour nouveau tout Ie probleme; nous developperons surtout Ie second point de vue, parce que dans nos essais d'une extension de la Mecanique Statistique aux milieux continus, c'est lui qui nOUE; fourni un fil conducteur precieux.
[151, [16J .
II - MECANIQUE STATISTIQUE. Nous allons rapidement rappeler les traits essentiels de la Mecanique Statistique sous la forme que leur a donnee J. W. GIBBS [6]' On considere un systeme materiel holonome ayant
159
k degres de liberte, c'est-a-dire
~
148 J. Kampe De Feriet
que la position de ce systeme dans l'espace est dMinie par les valeurs de k
coordonn~es:
q l' '"
qk ' L'ensemble des points
Wc = (q1""
qk)
compatibles avec les liaisons constitue l'espace de configuration E
d~signant l'~nergie cin~tique
c sont dMinis par :
du systeme,les
OE c Pj =Oqj
ou
k
moments
n
c ;
conjugu~s
dq, q'
j
=~
dt
l'ensemble des points Wv
= (PI'"
"Pk)
n ; un ~tat
constitue l'espace des vitesses
ou phase du systeme est -- ---
v
dMini par Ie couple:
w = (w, w) c v l'ensemble des points w constitue l'espace des phases (1)
n = nc
X
nv '
L '~nergie potentielle du systeme ~nergie
~tant d~sign~e
par E , soit son p
totale : E=E+E, c p
Le mouvement du systeme doit satisfaire aux
2k
~quations
differentiel-
(l)L'espace des phases n a 2 k dimensions, mais du point de vue m~trique on ~e peut Ie consid~rer comme un sous ensemble de l'espace euclidien R2 ; on trouvera dans [2] de tres importantes remarques sur les propri~tes metriques de n,
160
- 149 J. Kampe De Feriet les de HAMILTON-JACOBI
(~)
dq. uE _J = _ (2: 1) dt Op. J Nous ne consid~rons que des l'~nergie
tot ale E ne
dp. _J dt syst~mes
d~pend
j = 1. 2•... k .
-~.
J conservatifs. c 'est-a.-dire que
pas explicitement du temps: E = E (w)
(2.2)
Sous des conditions de int~ressant
d'unicit~ - 00
=
la
des
r~gularit~
M~canique.
int~grales
on
de E •
d~montre
v~rifi~es
alors Ie
pour tous les
tMor~me
syst~mes
d 'existence et
de (2.1) : a. tout w£ n correspond pour
< t < + 00 un et un seul point j
L'ensemble des points Ttw. quand
t
= 1.2 .... k],
varie de
n l'orbite (unique) passant par w; nous
-00
a. + 00. dMinit dans
r~serverons
Ie nom de trajectoi-
re a. l'ensemble des points =1.2 ...
Ttwc =[qj(t). dans I' espace de configuration
kJ
n
c
L 'ensemble des Tt dMinit un groupe aMlien de transformations
cette
propri~t~
constituant Ie principe de HUYGHENS pour les
syst~mes
conservatifs. Une propri~t~ fondamentale est exprim~e par Ie Th~or~me de LIOUVILLE : Soit
'f
la
(J' -
alg~bre (corps bor~lien) des parties de n
(2.)Nous avons utilis~ pour repr~senter I'Hamiltonien ( = ~nergie totale) du syst~me la lettre E au lieu du H classique en M~canique Analytique. parce que l'usage s I est ~tabli de d~signer par H la mesure de l'information qui jouera un r6le important plus loin.
161
- 150 -
J; Kampl! De Fl!riet
mesurables par rapport a. la me sure de LEBESGUE de
l'~l~ment
m dMinie a. partir
de volume a. 2 k dimensions:
= dq1
dw
.. , .. dqk dP1'" dPk
La me sure de tout ensemble A E
Jf est invariante par Ie groupe des tran-
sformations T t :
-co
(2,3)
Pour construire la
M~canique
des phases une mesure de
Statistique classique on introduit sur llespace
probabilit~
P
,absolument continue (3) par
rapport a. la mesure de LEBESGUE: P«m Clest la. une
diff~rence
essentielle avec la
ou on suppose au contraire la meSllre de par rapport
a
M~canique
probabilit~
Statistique quantique P
singllli~re
(4)
m, P -l- m
la
probabilit~
P
brable de surfaces
~tant concentr~e, d'~nergie
par exemple, sur une famille
constante
E(w) = d..
i
1
Du point de vue moderne la l'~tude
du mOllvement
pose que
l'~tat
d~nom
al~atoire
du
M~canique
=
1, 2""
Statistiqlle consiste dans
syst~me mat~riel consid~r~,
initial est choisi au hasard selon la loi de
(3) c'est-a.-dire que la condition m(A)
probabilit~
= 0 implique PIA) = 0
(4) clest-a.-dire qulil existe une partie A de n telle que: m(A) = 0 , PIA) = 1 m(AI) = min) , P(AI) =·0
162
si lIon sup:
- 151 -
J.
Prob
(2.4)
[WE
La loi de variation de la satisfaire
De Fe'riet
AE~
AJ = PtA)
probabilit~
Kamp~
en fonction du temps doit
~videmment
a la condition:
(2.5) qui exprime simplement que la dans
l'~coulement
probabilit~
se conserve (comme la masse
d'un fluide) lorsque lIon suit Ie point Ttw Ie long de
son orbite. On dit que Ie mouvement probabilit~
al~atoire
est stationnaire si la me sure de
est invariante
(2.6)
-oo
consid~rerons
dans cette
~tude.
Au langage purement probabiliste que nous venons d 'employer on pr~fere
presque toujours, dans les
image, dont ment
l'id~e
employ~e
trait~s
premiere remonte
de
M~canique
a MAXWELL
mat~riel donn~
pies al'instant initial dans I'espace des phases de des points wI"'" wN
situ~s
n'exercent aucune action
a chaque instant
de cet ensemble paratt, du mouvement
al~atoire
; on distribue ces co-
fa~on
dans 1a partie A de
= NP(A) ; puis on laisse ces systemes
definit
~t~ syst~matique
par GIBBS: on postule l'existence d 'un nombre "tres grand"
N de copies identiques du systeme
nA
et a
Statistique, une
~vo1uer,
n
que Ie nombre nA soit
pr~cis~ment:
en supposant qu'ils
m~caniqueles
uns sur 1es autres;1e nuage des points
t
repr~sentatif;
l'ensemb1e
a beaucoup d'esprits
1a
consid~ration
, moins abstraite que celle
du systeme ; c 'est ainsi qu 'a la notion de mouve-
163
- 152 -
J.. Kampe De Feriet
ment
al~atoire
stationnaire correspond ici celle
d'~quilibre
statistique:
dans la partie A il entre et il sort constamment des points w., mais Ie J nombre nA de ces points est ind~pendant du temps. Le nuage est constamment en mouvement, mais pour un observateur qui compte seulement les pdn1s il semble en En fait,
~quilibre.
gr~ce
(5)
a la loi forte des grands nombres 1'ensemble
repr~
sentatif de GIBBS est seulement une image approcMe de la description probabiliste. Soit en effet
IA (w) l'indicateur de la partie A; considE!rons la
suite infinie de variables alE!atoires
ind~pendantes
(5) Il faut souligner, - car la tentation de confusion n'est pas imaginaire, que m~me si Ie systeme ·mat~riel se r~duit a une mol~cule (poss~dant k degr~s de libert~, k ~ 3) I 'ensemble reprE!sentatif de GIBBS ne doit a aucun prix Mre confondu avec un gaz de MAXWELL-BOLTZMANN; dans I 'ensemble gibbsien les N mol~cules s'ignorent completement les unes les autres et n'exercent les unes sur les autres aucune action; en outre leurs orbites individuelles sont des courbes continues; dans Ie gaz maxwellien, au contraire, les molE!cules agissent les unes sur les autres (tout au moins quand leur distance n'est pas trop grande); leur action mutuelle est port~e a son paroxysme au moment d'un choc , qui d~truit la continuit~ de l'orbite. Oserons-nous avouer que, convaincu de l'abtme s~parant l'ensemble gibbsien et Ie gaz maxwellien, nous avons toujours admir~ avec surprise que sur deux modeles aussi diff~rents, on ait pu fonder la m~me Thermodynamique !
164
- 153 -
J. Kampe De Feriet
w1' ••• wN' •••
Ies points
ind~pendantes,
~tant
r~sultat
Ie
d'une suite infinie
faites chacune selon Ia Ioi de
probabilit~
d'~preuves
P(A). La loi forte
des grands nombres probr Iim k(X 1+... +X N) =X]=l - N-'+co nous donne:
~
r lim
Prob
l
= p(A)l = 1
'J
N->too N
d'~preuves ind~pendantes
pour presque toutes les suites infinies
Ie rapport
nA
N
tend vers P(A) ; quand Nest "tres grand" rna is fini on peut donc
distribuer "approximativement"I'ensemble nombre des points Dans la
situ~s
dans A
M~canique
ment continue par rapport
repr~sentatif
a I'instant initial soit donn~ par nA= NP(A).
Statistique classique la mesure P
a Ia
presque partout dans r(w)
~tant
mesure de LEBESGUEjd'apres Ie
de RADON-NIKODYM il existe une fonction r(w) , d~termin~e
de faGon que Ie
~
densit~
de
absolu-
th~oreme
probabilit~,
0, satisfaisant les trois conditions: 0
r(w) E L(O)
(2.7)
iOr(W)dW = 1 telle que pour tout (2.8)
A£
:r on ait
P(A) =ir(W)dW.
Pour que Ie systeme soit en chaque
t
~quilibre
statistique il faut et il suffit que pour
donn~
(2.9)
r (TtW) = r(w)
sauf peut-Mre sur un ensemble At Comme la condition
de mesure nulle.
m(\) = 0 n1entratne pas
165
- 154 -
J .. Kampe De Feriet
m
[V < t < A-]= t -00
n'~tant
(I 'union
pas
d~nombrable),
satisfaite pour presque tout
w
temps. Mais, pratiquement, en ~cart~e
0
+00
on ne peut pas en conclure que (2.9) est
simultan~ment
M~canique
parce que lIon n 'envisage que des
pour toutes les valeurs du
densit~s
continues en
plus discontinues sur un nombre fini de surfaces dans ~quilibre
voit que: pour que Ie systeme soit en fit que la
densit~
souvent: une
de
probabilit~
int~grale
~quations
les 2 k
~quations
a aborder un sujet oil.,
taine confusion; d 'apres la
n.
est
wou au
Dans ce cas on
statistique il faut et il suf-
r(w) soit un invariant (ou comme 1'0/1 dit
premiere des
Ceci nous amene
difficult~
Statistique cette
th~orie g~n~rale
de HAMILTON-JACOBI). semble-t-il, regne une cer-
des
~quations diff~rentielles,
de HAMILTON-JACOBI peuvent admettre
invariants ind~pendants :o/l(W)""
2.k-l
'Y2k_1(w) au plus:
-oo
r~ponse
paratt donc simple : on obtient
nant pour la r(~ 1""
densit~
de
probabilit~
l'~quilibre
statistique en pre-
une fonction quelconque
'Y2k-l) satisfaisant les conditions (2.7). Or pratiquement, -
sauf dans des travaux r~cents, par exemple H. GRAD [9) en 1952, - Ie seul invariant que 1'on voit apparattre est
1'~nergie
E(w) du systeme; 1'on
suppose toujours :
r
(2.10)
La raison terpr~tation M~thodes
invoqu~e
= r(E) pour ce choix repose, croyons-nous sur une in-
,
confuse d'une th~oreme ~nonc~ par H. POtNCARE dans "Les
nouvelles de la
M~canique
C~leste".
En effet, on affirme en
s'appuyant sur Ie titre du Chapitre V [24] p. 233 "Non-existence des int~-
,
grales uniformes" que H. POINCARE
a d~montre
166
qu'il n'existe qu'un seul
- 155 J. Kampe De Feriet
invariant fonction uniforme de
l'~nergie
UJ;
E.
Mais si on se donne la peine de lire Ie Chapitre V, on constate que I
POINCARE envisage un probleme
tr~s
particulier, puisqu'il suppose essen-
tiellement que E est une fonction analytique d lun para metre I" E
2
= Eo(wc ) + rEI (w) + I" E 2(w) + ...
w ; il ~crit (p. 233) : IIJe me propose de d~ v montrer que, sauf certains cas exceptionnels que nous ~tudierons plus loin,
ou
Eo est
ind~pendante
~quations
les
de
(de HAMILTON-JACOBI) n'admettent pas d'autre
analytique et uniforme que l'integrale
int~grale
E = const. II
La demonstration ne s'applique done qulaux invariants qui sont
a
la fois "analytiques et uniformes ll• II est clair, d'apres Ie contexte, que par Ilanalytique ll POINCARE" entend lIanalytique par rapport nage de
t' = 011;
il resulte aussi de l'examen de sa
quand il parle d'une
int~grale
a l'
au voisi-
d~monstration
anqJytique, cette fonction doit
~tre
que,
une inte-
grale pour toutes les valeurs de ,u (voisines de 0) et non seulement pour une valeur
num~rique
donnee de
III
en effet il utilise
;
ses dans ses calculs des developpements en me les coefficients des puissances tage
a egarer
,~'
s~rie
a plusieurs
et identifie terme
repri-
a ter-
. Ce qui a peut-Mre contribue davan-
certains auteurs, c lest Ie titre de Pavant dernier paragraphe
du Chapitre V: (p. 259)
IIInt~grales
non holomorphes en
f
II ; or des les
3 premieres lignes , on voit que POINCARE" rejette bien l'hypothese que ses
int~grales
sont analytiques en
F = F'(W) +/"F FII (w) OU FI non-holomorphes pour /-l
=0
f
,mais qu'illes suppose de la forme
et FII sont analytiques en ~ ; elles sont bien mais developpables selon les puissances de
vJi! .... Ces hypotheses avaient un sens dans Ie contexte de ses recherches de
M~canique C~leste
; elles nlen ont aucun dans Ie cadre general de la
167
- 156 -
J.
Kamp~
De Feriet
Mecanique, ou l'on n'a aucune raison d'introduire un paramHre et de considerer des invariants
= ~ (w, f)
t·
(w,
r)
qui satisfassent la condition 0/( Tt w,)4) =
pour toutes les valeurs de
l.l.
Nous nous rangeons donc entieremeht a 1'opinion exprimee par C. TRUESDELL dans les belles leGons qu'il a professees ici
m~me,
ala
Villa Monastero, en 1960 : (28)p. 43 "This Theorem (Ie theoreme du Chapitre V) is sometimes regarded as showing that, apart from exceptional cases, a Hamiltonian system has no time-independent integrals other than its energy. This would be a happy result for Statistical Mechanics, but it is certainly not what POINCARE claims, nor can I see any reason to expect it" , En realite la raison pour laquelle on suppose toujours que la densite de probabilite ne depend que de 1'invariant E ne resulte pas du tMore/
me mal compris de POINCARE sur l'inexistence d'autres invariants fonctions uniformes de
w ,mais plutbt de la coutume bien etablie de
melanger, deux theories differentes : on sert au lecteur une sorte de cocktail ou, a la liqueur pure et transparente des idees de GIBBS, on ajoute un peu des alcools puissants mais mal decantes de BOLTZMANN; ala Mecanique Statistique exposee au debut de ce paragraphe, ou les orbites sont essentiellement continues (6)
on melange de la Theorie cinetique :
on admet que des chocs peuvent se produire a certains instants t , au ---n cours desquels les q .(t) restent continues mais les p .(t) presentent des J J discontinuites de premiere espece, p ,(t-O) et p ,(HO) prenant des J J valeurs diff~rentes. En d'autres termes la trajectoire Ttwc reste une
(6) Les fonctions q .(t) et p .(t) sont non-seulement continues mais admettent, par 1'hypothese Jm~me J des equations de HAMILTON-JACOBI, des derivees q.(t) et p,(t) continues. J J
168
- 157 -
J. Kampe De Feriet
se
d~compose
n,
mais llorbite T w c t en une suite dlarcs de courbes ; chaque arc est continu dans
courbe continue dans llespace de configuration
un intervalle ouvert (tn' t n+1) , mais les
extr~mit~s
des deux arcs corre-
spondants aux deux intervalles (t
I' t ) et (t ,t 1) ne se raccordent n- n n n+ pas; a cause de la discontinuit~ des p ,(t) aux instants t un invariant ne J n prendra pas, en g~n~ral, une seule valeur constante, quand t varie de - 00
a + 00, mais une suite de valeurs constantes \lJ (T w) =
1
t
diff~rentes:
C(,
n
sur chaque arc de llorbite ; il y a un seul invariant, qui conserve la valeur sur tous les arcs de Porbite, c lest
pr~cis~ment ll~nergi'e
m~me
:
-oo
~lastiques.
(7)
(\l) Dans [2)P. 19-22, A.BLANC - LAPIERRE, P. CASAL et A.TORTRAT mettent en relief les difficult~s du concept de choc purement ~lastique de 2 mol~cules. liEn effet dans ll~tude du choc, on ne peut plus assimiler les corps qui y participent a des corps solides indMormables; on admettra qu I ils sont parfaitement ~lastiques, clest-a-dire que ll~nergie emmagasin~e par chacun dleux, lorsqulil subit une d~formation, est irit~gralement restitu~e lorsque cette dMormation cesse .Mais il est alors possible de montrer que, Iorsque deux corps de cette esp~ce entrent en collision, Ie choc provoque en chacun dleux des vibrations qui, en g~n~ral, subsistent apr~s leur s~paration,de sorte qulapr~s Ie choc on ne peut plus consid~rer ces corps comme des solides indMormables ... On voit dans quel1e inextricable situation on se trouve si, des milliards de fois par seconde, il faut cesser de consid~rer une mol~cule comme un corps solide et la regarder comme un corps ~lastique ... La M~canique Statistique, el1e, ne cherche pas a r~sou dre ces difficult~s qui ne proviennent peut-~tre en definitive que d lune idee un peu trop anthropomorphique des pMnom~nes mol~culairesll .
169
- 158 -
J. Kampe De Feriet
persuad~s
Nous sommes, pour notre part, par l'invariant
E
(jusqu'aux travaux
r~cents
que Ie rt>le exc1usif joue
de H. GRAD, de R. M.LEWIS
et de C. TRUESDELL) dans la Mecanique Statistique, - est une trace, peuttltre enfouie dans Ie subconsciert, des vieilles idees sur llinvariance de E et de
E
seul dans les chocs elastiques. (8) .
Ces points bilit~
dans
~tant pr~cises,
l'~quilibre
que lIon admette pour la densite de proba-
statistique une fonction ne
d~pendaht
que de
l'~nergie:
r = r(E)
ou une expression plus generale
d~pendant
de
m invariants
il nlen reste pas moins vrai quIa chaque fonction r dra une Mecanique Statistique
m~canique
ete compl~tement
Statistique sur une fonction
la distribution canonique: -
(2.11)
r(E) =
correspon-
diff~rente.
Or, en fait, cet arbitraire a fonde toute la
particuli~re
~limine ; GIBBS
r
[61
bien determinee,
J
~E
~(@)
ot
Z(~) ~ e -@Edw
~ ~tant un param~tre qui, lorsque llensemble representatif est susceptible dlune
interpr~tation
temp~rature
thermodynamique, joue Ie rtle de llinverse de la
absolue.
(Il Une expression claire des
id~es
que nous venons dlesquisser se trouve
d~ja dans ces lignes de MAXWELL (cherchant a justifier Ie theoreme ergod ique dans son c~l~bre memoire de 1879) "each encounter will introduce a dis-turbance into the motion ci. the system, so that it will pass from one undisturbed path into another. The two paths must both satisfy the equation of energy and they must intersect each other in the phase for which the condi .. tions of encounter with the fixed obstacle are satisfied, but they are not subject to the ~quations of momentum" (cit~es par C. TRUESDELL [28J p. 22)
170
- 159 -
J. Kampe De Feriet
De m~me dans Ie cas g~n~ral [28] p. 41 la distribution pantacanoniexp( -~1 'l'l- .. ·-(am 'I'm} (2.12)
r('Vl""
Z(
apparatt comme Les
~"
...
~m)::
Z(~l"" ~m)
~m) ~ Lexp(- ~,'I',- ... -~m 'fm)dw
priviligi~e.
consid~rations
par lesquelles GIBBS introduit la distribution
canonique n font guere qu 'une valeur heuristique ; la premiere justification, dont la rigueur satisfasse un matMmaticien, a
~t~ donn~e
par A. I.KHINCHIN
~8) p. 84-93. - Il d~montre que la distribution canonique est une cons~ quence du TMoreme Central limite du Calcul des
Probahilit~s
lorsque Ie
nombre . k des degr~s de libert~ tend vers l'infini ; il pousse m~me pr~cision
la
jusqu 'a donner l'ordre de grandeur de l'approximation en fonction
Nous contentant de renvoyer nous aborderons
l'~tude
a cet ouvrage,
aujourd'hui classique,
de la distribution canonique d 'un tout autre point de
vue; d'abord au • III, nous pla~ant dans Ie cercle d'id~es de B. MANDELBROT,
(221 nous montrerons
de quelle lumiere la th~orie de
l'estimation statistique de Sir Ronald FISHER peut (\clairer la signification de la distribution canonique ; ensuite au § IV , d~veloppant une id(\e de E. T. JAYNES
[13) nous (\tablirons,
comment dans Ie cadre de la Th(\orie de l'In-
formation de Cl. SHANNON et N. WIENER, les distributions canoniques ou pantacanoniques sont une
cons~quence imm~diate
de l'Information.
171
du principe du Maximum
- 160 -
J.
Kamp~
De FE!riet
III - ESTIMATION D'UN PARA METRE ET STATISTIQUES EXHAUSTIVES. L'un des problemes fondamentaux de la Statistique matMmatique est Ie suivant : on sait qu 'un point probabilit~,
al~atoire
X = (Y I , .•. Ym ) dans R donn~e
appartenant a une famille
Prob[X~A1 =P(A, 9)
(3.1) on fait
Xl' ... x N les points metres 8 qui
AER m ,
al~atoires ind~pendants
observ~s
s
J de RS
repr~sentent
8E.J
Math~matique
variable
Xl""
XN ; soient
Ie mieux ces observations ?
[51
que sont dus les principes g~n~raux
de la solution de ce probleme et la terminologie
employ~e
en Statistique
; nous nous bornerons au cas ou m = I (X est alors une
al~atoire
consid~r~e
sur R), ou s = 1 et ou les lois de
probabilit~
sont absolument continues, admettant une
p(x, IJ) ; cette fonction est
suppos~e
de la fa-
densit~
connue pour tout
de pro-
Xe R
et
pour En que :
d~signant
N par B un ensemble mesurable de R il est clair ( I
(3.2)
:
; comment estimer les valeurs des s para-
C'est a Sir Ronald FISHER
babilit~
donn~
de
N observations inctependantes de X , c 'est -a-dire que I 'on consi-
dere la suite de N points
mille
oMit a une loi de
d~pendant
de lois
IJ = (6 1, .•. 8s) parcourant un intervalle
parametres
m
= I L(x l ,···, x N' 8)dx 1, .. dX N JB
OU L(x 1, ... x N' 8) = p(xl' 8)x .. , -------
172
X
p(x N, 8)
- 161 Kamp~
J.
De
F~riet
dMinit la fonction de vraisemblance (likelihood function). En
g~n~ral
op~rent,
les statisticiens
non sur les nombres
observ~s
Xl' ... xN eux-mt!mes, mais sur certaines fonctions de ces nombres qu'ils appellent des IIstatistiques II ou des lIestimateursll. Consid~rant
e=
la variable aleatoire.
on prendra comme estimation de 0 la valeur observ~es de
e
= La valeur moyenne de la statistique
par :
=
8(x 1,··· xN) L (Xl' .,' xN' O)dxl' '. dX N ~ RN
~videmment
une fonction de
la fonction b(
or
b(O)
I'~crit
dMinit Ie biais de la st'atistique
sous Ia forme
m~me
e
si
=0
9
on dit que Ia statistique
(3.5) (
0; si on
9= 0 + b(O),
(3,4)
De
donn~e
ef
(3. 3)
c1est
8 est
Ia variance
- 2 e - a)
jde
est sans biais (unbiased),
Ia statistique a pour v:Ieur : [ 9(x 1", .x N) -
=
e}·.
-
L(xl'" 'X N' 0) dx l ", ,dx N
RN On de
p(X, 0)
d~montre
qu'elle satisfait sous certaines conditions de
a l'in~ga1it~
de CRAMER-RAO
(1 N
+ b I (0)) 2
j::P(X, 6) [~ log pIx, 6)) 2 dx 173
r~gularit~
• 162 J. Kampe De Feriet propos~
FISHER a m~tre
9) contenue dans une observation xl de
r
I
de prendre comme me sure de l'information (sur Ie paX la quantit~
+oo
= p(x, . 9)r.~()8 log p(x, 9)) 2 dx
J
-00
La
quantit~
d'information contenu dans
N observations
ind~pendantes
(xl' ... xN) est alors Ie nombre NI qui figure au d~nominateur de (3.6) : plus l'information est grande, plus la borne inf~rieure des fluctuations de
9 est petite.
la statistique
Lorsque, dans
l'in~gaIM
on dit que la statistique
9
de CRAMER-RAO,
l'egalit~
est
r~alisee,
est efficace (efficient).
Mais Ia notion la plus importante, toujours due 1l FISHER, est la suivante : soit 't'( 9)
a(Xl' ...
la
densit~
de
probabilit~
de la variable
al~atoire
XN), parfaitement determin~e en fonction de
p(x, 9), par la
formule:
ott
A designe Ie domaine contenu entre les deux surfaces: 9
et
Considerons Ia probabilit~ conditionnelle du point al~atoire N (Xl' ... XN) dans R , quand on sait que (Xl"" XN) a pris la valeur
e
9
(c'est.1l-dire que Ie point est sur la surface
stique
S9); on dit que la stati-
9 est exhaustive (sufficient) si cette probabilite conditionelle est
174
- 163 -
J. Kampe De
ind~pendante
F~riet
de B ; en dlautres termes .toute l'information que llobserva-
tion de (xl' ... xN) nous fournit sur
Best contenue dans la
d~termination
de la surface particuliere SB qui passe par ce point ; la
~onnaissance
de la position du point (xl' ... ' xN) sur cette surface ne nous apporte aucune information
suppl~mentaire
A titre d 'exemple
sur
consid~rons
par -x
1
p(x,B) = - h'ttB Des calculs
~l~mentaires
B .
la famille de lois normales
2
2B
e
d~finie
B >0
montrent que la statistique:
est sans biais, efficace et exhaustive;apres l'observation (xl'··· xN) nous estimerons done Ie parametre par la for mule
a = !N
r x~ N 1
J
On constate d 'ailleurs que cette valeur est ceUe que lIon obtient par la thode du maximum de vraisemblance (maximum likelihood) consistant
m~
a
chercher la valeur B qui rend maximum L(x l , ... xN' B) pour les valeurs observ~es
(xl' ... xN) . Ces quelques notions
~tant rappel~es,
sultat fondamental de la TMorie : si la que il nlexiste pas en
g~n~ral
nous pouvons aborder un
densit~
de
probabilit~
r~
est quelcon-
de statistiques exhaustives ; celles-ci nle-
xistent que pour des lois tres particulieres du type exponentiel. B. O. KOOPMAN (1936) a
donn~
une premiere
d~montration bas~e
sur la remarque que, s lil existe au moins une statistique exhaustive
a(Xl' ... XN)
alors l'egalit~ 175
• 164 -
J. Kampe De Feriet
entratne pour tout
e: L(x l , ... xN,ct) =
L(Y l ,··· YN'cC.) Mais dans sa demonstration il fait intervenir des conditions d 'analyti-
cite des fonctions,superflues comme l'a montre E. B. DYNKIN qui a non seulement resolu Ie
probl~me
[4}
dans un cas plus general, mais en-
core a introduit des idees nouvelles d'une grande importance, en particulier la definition d 'une statistique exhaustive necessaire (necessary and sufficient statistic).
n dit que la fonction
F(lt l , .•. x N) est dependante de la fonction
= G(x l ' ...' XN) entratne
G(x l '· .. x N) si l'egalite G(yl'· .. YN) F(y I' ... yN)
= F(x l ,···
xN) ; les fonctions F et G sont equivalentes
si chacune est dependante de l'autre. La fonction
e (xl' ... xN)
est dite
une statistique necessaire si elle est dependante de toute statistique exhaustive. Toute fonction dependante d'une statistique necessaire est elle
m~me
une statistique necessaire. Deux statistiques necessaires et exhaustives sont equivalentes. Le rang
r
de la famille de lois de probabilite correspondant est dMini par
sion de l'espace fonctionnellineaire G(x, e)
l
e
=s-
I , s etant la dimen-
contenant toutes les fonctions
= log p(x, e) p(x,
ou
r
a
varie dans l'intervalle
eo )
[e l , e2 ]
et
eo
est une valeur arbi-
traire dans cet intervalle ; on aura bien entendu en general r = + 00 176
•
- 165 -
J.. Kampe De Feriet
Sous une condition t~
de la
1L
d~rive
oX
g~n~rale
<'J
r~sultats
rest fini, alors la
p(x, B)
r~gularit~
(existence et continui-
p(x, B) sur un ensemble dense sur
DYNKIN (p. 25 et 26) prouve les a) Si Ie rang
de
l.;
= exp [
les systemes de fonctions
suivants :
densit~
de
probabilit~
c j (e)tr/ x ) + cote) +
(1"Vl""
R), E. B.
a la forme:
~/c (X)J
~Ir) (qui forment une base pour~)
et (1, c ,... c ) etai:J.t lineairement independants. 1 r b) Pour tout N
~
r Ie systeme de fonctions :
forme un systeme de statistiques necessaires et exhaustives, les fonctions
8.
etant independantes fonctionnellement. J Considerons par exemple la famille de lois de probabilites defi-
nies par: -x
p(x, B) =
ee e 1
x ) 0
x
=0
L 'espace lineaire G(x, e)
£ 1
11 est done de rang Par consequent pour
<0
eontient toutes les fonetions 1
= ( - - -)
eo
B> 0
B
x
eo e
+ Log-
1 et admet eomme base N
~
(1, x)
1 la fonction
est une statistique necessaire et exhaustive
177
toute autre statistique
- 186 -
J. Kampe De Feriet
n~cessaire
et exhaustive lui est
~quivalente
et nous prendrons selon l'u-
sage
fa qui est
~galement
~
=
(xl +... + xN)
sans biais.
IV. - LA DISTRIBUTION CANONIQUE ET LA THEORIE DE L'ESTIMATION
Revenons aux
syst~mes m~caniques
§n ; nous supposerons que peut toujours prendre
~gal
E
l'~nergie
conservatifs
consid~r~s
au
a un minimum isol~, que lIon
E
a: = 0
puisque l'on dispose d'une constante arbitraire dans
l'~nergie
potentielle.
Nous introduirons les notations suivantes :
D = x
{w: O~E~x).
tw: Quand phases
n
x croit
0
a + 00,
E=xJ. les surfaces
~x
dans I' espace des
s 'enveloppent l'une l'autre comme une famille de spMres
(ce sont les hypoth~ses m~mes de KHINCHIN Designons Ie volume du domaine (4.1)
7
V(x)
=
(18)
p. 33) ;
D par
x
f
; dw ,!
Dx
Nous dMinirons la fonction de structure du
178
syst~me
par
- 167 J. Kampe De Feriet
dO'
(4.2)
Sex)
=1 Ix
x
Sex)
\grad E I
()E 2
= + [(-)
dV
(18]
p.
36
revenons
a la
~l~mentaires ~tant rappel~es,
Statistique et supposons Ie systeme en probabilit~ ~tant
r~sulte
M~canique
statistique, la
densit~
l'~nergie
= reEl imm~diatement
alors
nergie du systeme E(w) probabilit~
~quilibre
fonction seulement de
r 11
que:
= Sex).
dx
Ces dMinitions
de
et (9)
+
Jqj
On d~montre tres facilment (4.3)
0
Ix
dG' d~signe l'~l~ment d'aire de la surface x
ou
>
Igrad E \
des
est une variable
est dMinie par la
densit~
r~sultats pr~c~dents
al~atoire
X
que l'e-
dont la loi de
:
p(x) = r(x)S(x)
x ) 0
(4.4) =
On a
~videmment
1
x <0
0
la condition :
+00
(4.5)
r(x)S(x)dx = 1
Q
(9) On a evidemment grad E \ > 0 , puis que par hypothese les ~quations de HAMILTON-JACOBI n'admettent pas de point singulier dans
I
n. 179
- 168 -
J.
Kamp~
De
F~riet
La valeur moyenne de 1'~nergie a donc pour valeur :
(+00
E=
(4.6)
X
i
r(x)S(x)dx ,
I
)!;
et sa variance est (4.7)
donn~e
(E - E)
Ceci
~tant,
2
par
+00
=
] C:
consid~rons
(WI' ••• , wN) nous en
-2
(x-E) r(x)S(x)dx.
un ensemble
d~duisons
les
repr~sentatif
N variables
de GIBBS
al~atoires ind~pendan
tes : (4.8) ayant toutes la
m~me
loi de
probabilit~
1'on a toujours par hypothese syst~me ~tant
~
XJ
dMinie par (4.4) ; (notons que
0) :r;emarquons en passant que Ie
conservatif, on n'est pas
sterne au m~me instant pour tous les
oblig~
de
consid~rer l'~tat
du sy-
w. [l'instant initial dans (4.8)], J
puisque
pour tout
t
Admettons que la loi de rametre
repr~senter
9 , qui doit
gie du systeme (de
probabi1it~
m~me
ne
d~pende
que d 'un seul pa-
la valeur moyenne
(4.6) de
l'~ner-
que les Thermodynamiciens nous affirment
que la seule grandeur macroscopique que 1'on puisse mesurer sur un gaz est sa (4.9)
temp~rature)
; la
densit~
de
p(x, 9) = r(x, 9)S(x)
=0 la fonction de structure, qui ne
probabilit~ S'~crira
alors :
x?O x<0
d~pend
180
que de Ia dMinition m~canique du
• 169 -
J. Kampe De Feriet
systeme, devant Mre
ind~pendante
du para metre introduit dans la Stati-
stique. Nous voulons que
soit une statistique semble
et exhaustive ; dans I 'observation d 'un en-
n~eessaire
repr~sentatif
partieulier :
nous prendrons done eomme valeur du parametre
e'est-a-dire que nous estimerons
E par la moyenne
Si no us admettons que la Statistique stive quel que soit Ie nombre N doit
~tre ~gal
pel~
au
a I
et
d'apr~s
~
9
est n~eessaire et exhau-
1 , Ie rang de la loi de
Ie tMoreme fondamental de DYNKIN rap-
§ III la densit~ de probabilit~ doit ~tre de la forme
x :;. 0
(4.10)
Dans eette formule nous avons pris
'Y 1(x) = x
statistique n~eessaire et exhaustive ~tant par hypothese tre statistique
doit lui
probabilit~
n~eessaire
paree que la
9 ,
toute au-
et exhaustive
~tre ~quivalente .Comme
nous pouvons disposer de la fonetion
181
- 170 -
J.
',\!(x)
Kamp~
De Feriet
nous prendrons:
].
t (x)
= log S(x).
Faisons un changement de paramMre en posant :
f= - c 1 (8) et introduisons la fonction de partition
,.teo
Z(~) = I e -P S(x)dx
(4.11)
lc
transform~e
qui nlest pas autre chose que la
tion de structure ; nous supposons pour tout
p>
telle que cette fonction existe
0 ; on sait qu'elle estalors analytique en
v~es
'+eo
-J 0 x
(4.12)
Z'(3) =
(4.13)
ZIf(?) =
existent et sont continues Ceci
S(x}
de LAPLACE de la fone-
~tant
-~xS(x)dx
e
pour que la
.; ses
d~ri-
< 0
JJ+~x e - ~x\"' S(x}dx pour~>
~
> 0
O.
densit~
de
probabi1it~
(4.10) satisfasse la
condition (4. 5) il faut et il suffit que e -co(8)
= reoe
c 1(8)x
S(x)dx
JO ou avec Ie nouveau
param~tre
~
:
-c o(8) e En conclusion la toire
X = E(w)
densit~
= Z(~}.
de
probabilit~
doit avoir la forme :
182
(4.10) de la variable
al~a
- 171 J.
F~riet
- ~x
p(x,~)
(4.14)
De
Kamp~
= eZ(~)S(X)
x
=0
x
ce qui donne pour la
densit~
de
~
0
<
0
r
:
probabilit~
- ~x
(4.15)
r(x,
c'est-a-dire
=_e__ Z(d)
~)
pr~cis~ment
x
~
0
la distribution canonique.
La distribution canonique de GIBBS appara!t ainsi comme la seule loi de pour
probabilit~
l'~nergie
ne
d~pendant
que d'un seul parametre et admettant
la statistique exhaustive suffisante
e
(4. 16)
1
= -
(E1 +••. +E N)
N
D'apres B. MANDELBROT [22) p. rait
~t~ d~montr~e
dans un
m~moire
m~me ~poque
1021, cette propri~t~ au-
pour la premiere fois par
L~o
SZILARD, en 1925,
ou il aurait, sans employer le terme,
d~couvert,
a la
que Sir Ronald FISHER,la notion de statistique exhaustive.
Est- il besoin de souligner l'importance de cette
interpr~tation
qui replace la tMorie de GIBBS dans le cadre de la Statistique
math~
matique moderne. Notons, en terminant, deux
r~sultats
se
d~duisant
immMiatement
du simple rapprochement de (4.6) et (4.7) avec (4.12) et (4.13) : (4.17) (4.18)
d E= - -
df
(E_~)2
=
d2
d~2
log
Z(~)
log
Z(~)
183
> 0 >0
- 172 J. Kampe De Feriet
ces formules nous serviront au
§VI.
V - LA THEORIE DE L'INFORMATION. Nous avons vu au § III que des 1925, Sir Ronald FISHER avait propos~
quantit~
une me sure de la
vation faite en vue d 'estimer Ie
babilit~. se
d'information contenue dans une obser-
param~tre
d rune famille de lois de pro-
En 1948 Claude SHANNON [25] et de Norbert WIENER ~o),
pla~ant
dans un cadre beaucoup plus bas~e
le tMorie,
sur des
id~es
tout
g~n~ral,
a fait
FISHER; cette tMorie apris rapidement un conduisant
a des
ont construit une nouvel-
diff~rentes
de celles de
d~veloppement consid~rableJ
applications de grande importance dans l'etude de la
transmission de l'information(codage), de la linguistique, de l'automation ... R~cemment
E. T. JAYNES
[13]
a montr~ que Ie principe du maximum
de l'information pourrait jouer un rtlle unifiant et simplifiant en que Statistique; de d'~crire
m~me
que les
automatiquement les
nique soumis
~quations
~quations
a des liaisons holonomes
ximum conduit automatiquement
a la
M~cani
de LAGRANGE permettent
du mouvement d'un systeme
donn~es,
m~ca
ainsi ce principe du ma-
loi de probabilite correspondant
a
l'equilibre statistique sous des contraintes statistiques donnees. On considere des systemes (beaucoup plus gene raux que les systemes
mat~riels
divers etats
n;
de la Mecanique)
susceptibles de prendre
w; l'ensemble de ces
~tats
definit l'espace des phases
dans une experience on observera la realisation d'un w bien
d~
termine ; une suite d 'etats consecutifs (obtenue par l'evolution du systeme selon ses lois propres), definit une orbite
dans
n
Les exemples suivants illustrent les diverses possibilites
184
- 173 -
J. Kampe De Feriet
1 - Ie
syst~me
est une machine
a ~crire
(lettres majuscules et minuscules n
sage, la
s~quence
2 - Ie
~tats
+ signes divers) ; l'espace des
comprend done 88 points ; une orbite constitue un mes-
phases
langage
; elle peut prendre 88
des points
~tant d~termin~e
par la structure du
employ~.
syst~me
est la roue d 'une loterie ; l'aiguille tourne devant une
circonf~rence
(cadran) ; les
~tats
w sont les points de la circon-
f~rence
ou s'arrNe l'aiguille, l'espace nest alors un intervalle
continu
Co,
2 l't
J;
une orbite est constitu~e d 'une suite arbitraire
de points. 3
-Ie systeme est une corde vibrante, de d'ALEMBERT-EULER; un Ie
d~placement
~tat
consid~r~e
avec l'approximation
w de la corde est dMini par
transversal u(x) et la vitesse transversale
v(x);
nest alors l'ensemble des couples de fonctions [U(x) , v(x)] continues dans ~O, f et s'annulent aux extr~mit~s de la corde x
= 1 ; une
orbite est
d~finie
x, d~pendant du temps t
~t u(x, t)
= v(x, t)
x
dans n par la suite de fonctions
= 0 et de
[U(X, t), v(x, t~, integrales des equations:
utd v(x, t) =i-2 u(x, t) Ox
Supposons que 1bn connaisse a priori la probabilite que les differents
~tats
wont de se manifester dans une observation du
syst~me
cela signifie que l'espace des phases est un espace de probabilite (n,~, P), ou
~
est une G'-alg~bre de parties de
ction d'ensemble (j'-additive, definie sur
185
'F ,
n
telle que
et
P
une fon-
- 174 J. Kampe De Feriet
o
~
PtA) ~ 1 p(n) = 1
(n,1/,
avant d'avoir effectu~ aucune observation sur
a priori
~prouvons
Pouvons-nous mesurer l'incertitude que nous
P) ? Pouvons-nojls
mesurer l'information que nous tirerons a posteriori de l'observation d'un ~tat wE(n, 1(, P) ? Nous admettrons que ces deux mesures doivent Mre apport~e
formation
par une observation
~tant
identique
~gales
l'in-
a l'incertitude
avant toute observation. Les
consid~rations
les les expressions
heuristiques suivantes rendront plus naturel-
math~matiques
introduites plus loin: supposons que, l'~tat
dans une observation, nous sachions seulement que se trouve dans une partie la
quantit~
PtA)
A de
n ; il
d' information ainsi fournie
et soit
ind~pendante
pIe de sa position si valeur de cette
seulement de la mesure
propri~t~
de
A, par exem-
est un espace m~trique : soit
n
quantit~
semble naturel d'admettre que
d~pend
de toute autre
F [P(AN la
d'information. ind~pen,.
Supposons maintenant que l'on fasse deux observations dantes de
l'~tat r~alis~
sique on emploie deux
W ,
(par exemple dans une
m~thodes
observation nous apprendra que en dMuisons
r~alis~
W
~videmment
W€
de mesure W€
exp~rience
diff~rentes)
; la
A et la seconde que
An B ; il
par~it
premi~re
III
E B ; nous
logique d'admettre que
l'information obtenue par ces deux observations
ind~pendantes
me des informations fournies par chacune d'elles
186
de phy-
:
est la som-
- 175 -
J.
OU,
Kamp~
De
F~riet
a cause 'de I 'ind~pendance des observations -:
(5.1)
Si done nous admettons que la fonction
F
est continue sur l'in-
... -, tervalle : 0, Ij , cette condition Mtermine (a une constante multiplicative
pr~s)
la fonction
F
:
F LP(A)]
(5.2)
Notons, en pa,ssant, ces
- ,- log PtA)
r~sultats
conformes
w€ n (condition toujours
1 - si lion sait seulement que s~e
a notre intuition : r~ali
par dMinition), notre information - est nulle
2 - si nous
consid~rons
une suite
)A de parties de
n ,
d~croissante
(au sens strict)
n
la suite des informations correspondantes va en
croissant:
l'information augmente avec la Quand on passe de
A, J
a
pr~cision
de I 'observation.
A, 1 Ie gain d'information est J+ PtA,) J > log P(Aj +1)
~gal
a
°
Ceci
~tant consid~rons
un nombre fini de parties
une partition finie de
AI'" . An de
187
n
deux
n , c 'est-a.-dire
a deux
disjointes :
... 176 .~ J. Kampe De
: ,n A.
J
'1
=
F~riet
0 quantit~
Evaluons maintenant la
d'information que nous recevrons
en moyenne, lorsque nos observations nous permettent seulement de sal'~tat
voir que
w se trouve dans l'une des parties A., mais que nous J
sommes incapables de distinguer l'un de l'autre deux
~tats
w' et
w"
qui appartiennent au m~me A.; l'information fournie par l'observation J
w~
A. J P(A .), J
~tant
la
-log P(A.)
et la
J
quantit~
probabilit~
moyenne de l'information,
fournie par une observation a pour valeur
=-
H
(5.3)
c'est
que
w~A. ~tant
J
d'autre part
esp~rance math~matique
:
L.3 P(A.)J log P(A.)J
cette for mule qui est la base de la tMorie de l'information. Remarquons que la partition [AI""
probabilit~s
sont seulement soumises aux conditions J
Si l'une des
I 3 P(A.)J
>;, 0
P(A.)
(5.4)
en posant
AJ ~tant quelconque, les
probabilit~s
=1 .
est nulle la formule conserve son sens,
:
x log x Nous ferons,
~
=0
pour
x
=0
plusieurs reprises, usage de
l'in~galit~
~l~men
taire log x (5i 5)
Cette (5.6)
< x-I
log x = x in~galit~
permet de H
,~
x
f
1
si et seulement si x = 1 .
1
d~montrer imm~diatement l'in~galit~
log n
188
..
- 177 J. Kampe De Fhiet
~tant
Ie maximum
atteint si et seulement si 1
P(A.) = J n Consid~rons
[AI' ... Anl
j = 1, 2.•. n
une partition plus fine que la partition
c'est-4-dire supposons que chaque partie
m~me partitionn~e
en un nombre fini (variable avec
J,
~tant
la fonction
x log x
classique de
convexit~
J
convexe dans l'intervalle
montre que la
Aj soit elle de parties
B. kll B. 1 = ~ J, J,
II B. k = A.
'R"
j)
quantit~
r
LO,
donn~e
k -,
I)
f
1 ;
l'in~galit~
moyenne d'information
fournie par la partition fine '\ -
j, k
satisfait
P(B. k) log P(B. ) J, J,k
~
(5.7) Plus la partition est fine plus l'information que lIon en tire D'autre part, si au lieu de l'observation d'un seul nous
r~alisons
sant les
~tats
une suite de
WI' • ••
N
WN,
~preuves ind~pendantes,
e~t ~tat
grande. W
,
nous fournis-
en operant sur I 'espace produit
nN = n \ ...
x
n
avec la mesure produit P
N
=P
.( ... '/.. P
nous obtenons par les m@mes raisonnements qui nous ont conduits (5.3)
189
~
- 178 -
J.
=- N
HN
(5.8) v~rifie,
ce qui
attachons
a la
Ces
-j
Kamp~
De
F~riet
P(A.) log P(A.) = NH J
J
encore une fois, une des
propri~t~s
intuitives que nous
notion d'information.
propri~t~s ~l~mentaires ~tant rappel~es consid~rons
Ie cas
ou l'espace des phases est un ensemble fini , ... n = \W • 1
La loi de p. J
= Prob
probabilit~
W \
n
est alors dMinie par n nombres
r;.. w=w.lJ.;
L
(5.9)
j
La partition la plus fine de un point quantit~
Aj
n
p. = 1 J
est celIe ou chaque partie se
r~duit
a
= {Wjl; la formule (5.7) no us donne alors cornme
moyenne dlinformation susceptible dletre fournie par une obser-
vation du
syst~me
: H =
(5.10)
-1j p.J log p.J
Clest la formule qui sert de point de
d~part
a la
Th~orie
de
SHANNON 1-25: p. 19. ... ..J On peut
d~montrer
que cette expression est la seule
tion de SHANNON, mise au point et satisfasse aux syst~me
propri~t~s
compl~t~e
(d~monstra
par KHINCHIN) qui
suivantes. Nous voulons dMinir pour chaque
fini l'information par une fonction ne
d~pendant
que des proba-
bilit~s ; soit fl (1), f2(Pl' P2)'''' fn(Pl'''' Pn)'''' la suite de fonctions
ainsi obtenues devant satisfaire aux axiomes suivants AI)
fn(Pl""
Ie domaine :
:
Pn) est une fonction ~~n negative d~finie et continue dans Pj
~ 0
T Pj 190
=
1
- 179 Kamp~
J.
A2 )
fn(Pl""
Pn)
est une fonction symE\trique de ses
A3)
f n+k (P1'''' Pn'~'
\)
fn(Pl""
g)
= fn(Pl''''
~
f (P " " P ) + LP. f n 1 n. J m
arguments
Pn)
~ 0
r. k J,
~, •.•
I-
p. J
J
a une
n
1 1 fn(u' ... Ii)
Pn ) ~
p.=~r·k J J,
Alors,
De F~riet
r j , m) p. J
constante multiplicative
pr~s
(qui revient
a fixer
la
base du logarithme) on a f (p l' . .. p ) n n tMor~me
Ce axiomes
A.
un
syst~me
et spE\cialement
a l'entropie
J
A
5
log P. J
sont prE\cisE\ment ceux que lion
en Thermodynamique, lorsque lion
donnE\ comme fractionnE\ en
Dans ce qui litE\ donnE\e
~ J
d 'unicitE\ doit sa grande importance au fait que les
1
doit imposer
= - \. p.
prE\c~de,
syst~mes
;
partie Is .
nous sommes partis d rune loi ··de probabi-
P pour en dE\duire la quantitE\ moyenne d'information
que lIon peut dE\duire d'une observation de l'E\tat dE\rE\
consid~re
nous avons admis que
H
H
w du syst~me consi-
reprE\sente aussi la mesure de no-
tre incertitude avant toute observation . Nous nous proposons de retourner en quelque sorte la situation: supposons qu'on nous fournisse a priori, avant toute observation, certaines informations sur Ie peut-on dE\terminer la loi de probabilitE\
191
PtA),
syst~me:
compatible avec les
- 180 J.
Kamp~
De
F~riet
informations donnees, qui corresponde au maximum de notre incertitude H ? . .• Bien entendu, si cette loi existe, c' est avec elle que nous obtiendrons egalement Ie maximum d'information, quand nous ferons a posteriori une observation de
w •
Ce principe du maximum
d~ncertitude
(appele plus volontiers
dans la litterature maximum d'information) est,
a notre
avis, Ie coeur
mtlme de la theorie '; nous pensons que quand il aura ete pleinement assimile il fera apparattre la notion d'information comme plus primitive que celle de probabilite : au lieu de deduire, comme on Ie fait, la mesure de l'information de la Theore des probabilites, nous sommes convaincus 'que l'axiomatiq)le de la Theorie de l'information soigneusement mise au point servira de base
a l'introduction
des probabilites.
Nous ne considerons actuellement que des
syst~mes
prenant un
nombre fini d'etats. Supposons que la seule information fournie a priori soit Ie nombre
n de ces etats
n'ayant aucune raison d'attribuer
; depuis LAPLACE on admet que
a un
etat une probabilite differente 'de
celle d'un autre etat, on doit prendre p, J
=
1 n
Or nous savons, d'apres (5.6), que parmi tous les choix possibles pour les
p.
J
c 'est precisement ce choix qui correspond au maxi-
mum de l'incertitude a priori ou de l'information a posteriori H ;: log n . Nous allons etendre cette methode, chant
a determiner les p, de J
a des
mani~re
cas plus gene raux, en cher-
a maximiser
H
en tenant
compte des contraintes imposees par des informations supplementaires
192
- 181 Kamp~
J.
donn~es
res se
a priori sur Ie systeme. En pr~sentent
al~atoire
U(w)
g~n~ral
ces
De
F~riet
donn~es suppl~mentai
de la maniere suivante ; on considere une variable
n
prenant sur
les valeurs
U(W 1), ... U(wn ) et l'on
nous donne sa valeur moyenne
I
(5.11)
p. U(w.) J J
~
On la suppose, bien entendu non moins un Si n
=2
j
tel que
U(w.)
n
>2
d~g~n~r~e,
c 'est-a.-dire qu'il existe au
f U
J (5.9) et
(5.11)
il Y a une
infinit~
les conditions
1M ; mais si
=U
bles; nous choisirons celle qui rend
d~terminent
de lois de
la loi de probabi-
probabilit~
possi-
H maximum
Introduisons la fonction de partition (5.12)
qui est dMinie pour tout
~
r~el
elle est strictement positive.
On a :
-L U(w.)e J
Z"( \~)
= [U(w.) e
2 - ~U(w.) J
j
L'in~galit~
- ~U(w.) J
Z' (~)
J
de SCHWARZ montre que
droll
193
• 182 • J. Kampe De Feriet
zn
Zl 2 (-) Z
= -Z La fonction
0
est donc convexe ; il en est de
log Z(S)
ction log Y( ~)
~
m~me
de la fon-
011:
Y(~)
= e
~U Z(~)
e ~[U-U(Wjil
=[ j
comme il Y a au moins un jll
IT -
pour lequel
jl
J
Y(+oo) = Y(-oo) = +
et de
U(w)
; il en r~sulte que
<0
U(W'II)
IT -
pour lequel
J
>
0
et un
00
m~me
log Y(+oo) = log Y(-oo) = + II
r~sulte
de la
et une seule valeur
"
~
convexit~
de
~
00
et de cette remarque qu1il existe une telle que
d
d~ log Y(~)
:
= 0
c lest-a-dire teUe que d
d ~ log Z( ~ )
(5.13)
- -
IT D).
(queUe que soit la valeur numerique donnee Ceci etant cherchons Ie maximum de contraintes
(5.9)
et
H
-pIT = Lp.[log _1 ,J p,
-tJ.,
H,
en tenant compte des
(5.11) par la methode des multiplicateurs:
J
:: [p, log p~ JJ
-cl. -
J
J
194
-d,- ~U(w.)
J
~U(W,)] J
- 183 -
J.
l'in~galit~
En nous appuyant sur
Kamp~
(5.5), nous obtenons :
@U ~ ~(e-~~-"aU(Wj)_Pj)
H -cJi-
De Feriet
=
e-~Z(~)-l
J
Si nous prenons;
= log
Q,
Z( (3 ) nous avons :
H ~ log Z(@) et d lapr~s I 'in~galit~ (5.5)
-
ment si
l'~galit~
ne peut
d~termin~
pour j = 1,2 ... n
par la condition que
probabilit~
L U(w.) e
Z(~)
~
.
J
J
doit
(5.11)
soit
v~ri
:
\"
1
c'est-a.-dire que
atteinte que si et seule-
J
p.= - - J Z (S)
Me avec cette loi de
~tre
\~U(w.)
e
Le parametre (3 est
+ (aU
-~U(w.) J
~tre ~gal
£ru- = -Z((3)-U
a. la racine unique
, ~
de
l'~quation
(5.13) ; dlou la conclusion
IT
lorsque la valeur moyenne a priori est (5.14) ou
...
~
donn~
est donn~e Ie maximum de llincertitude
par
HMAX
= log
est la racine unique de
seulement si la loi de
e
1\
Z(~)
+" ~ U
(5.13); ce maximum est atteint si et
probabilit~
est
d~finie
par:
"~U(w.)
p. = " J Z (~)
J
pour tout
195
j
•
- 184 -
J. Kampe De Feriet
Ce principe s'etend au cas OU lion connaUra!t les valeurs moyen-
Ul' ...
nes
Ur
de
ment independantes
r
variables aleatoires non degenerees et lineaire· ~
r
n - 2 . En designant par
:
I -~IUl(w.)-J ...- ~ r Ur (w.)J
Z(~I'''' ~r ) = . e J la fonction de partition, on a : (q.16)
ou
A
•
~ 1''''
I"r
(,l
sont les racines' uniques des
r
equations
- - U.1
(5.17)
Ie maximum etant atteint si et seulement si
"
p. = J
(5.18)
e
Bien entendu si
...
-~lU1(w.)- ... -~ U (w.)
pour tout
r
= n - 1, les p. sont completement
determines par la solution de d "t:slgne . I 'on par
aux cas
r r J
J
J
n equations lineaires compatibles
H(l) MAX' H(2) MAX' . . .
r = 1,2,3 ... n - 2,
Comme nous pouvions nous
si
l ' e maXlmum correspon dant
on obtient la chaine d'inegalites :
y attendre
intuitivement l'informa-
hon a priori contenue dans la donnee de la valeur moyenne
U, dimir
--
nue l'incertitude maxima que no us pouvions avoir quand nous connaissions seulement les valeurs moyennes
196
Ul' . .. ITr-l
.
- 185 J. Kampl! De
Les
r~sultats
F~riet
relatifs au cas ou l'espace des phases
nest fi-
ni, que nous venons de rappeler sommairement, ne sont pas suffisants pour les applications
a
a la
M~canique
Statistique ou
n
est un continu
2 k dimensions. pr~sente
Le premier cas qui se des phases
nest la droite
r~elle
a l'esprit
est celui ou l'espace
R; ne traitant que Ie cas, ou
il existe une densit~ de probabilit~, N. WIENER
[30]
p.
76
se contente
d'~crire
"Thus a reasonable measure of· the amount of information associated with the curve
f1(x)
is
The quantity we here define as amount of information is the negative of the quantity usually defined as entropy in similar situation . The definition here given is not the one given by
R. A. FISHER for statistical
problems, although it is a statistical definition and can be used to replace FISHER's definition in the technique of statistics" . L'allusion
a
la possibilit~ de remplacer la dMinition de FISHER rappeMe au § III n'est pas Ie seul mystere qui se cache pour nous dans ces lignes. En d~finition
effet appliquons la
de WIENER
a une loi de
probabilit~
unifor-
me sur un irttervalle (a, bJ f (x) = 0 1
x
< a
x > b
1 b-a
f (x) = -
1
Nous obtenons pour la d 'un
~tat
du
quantit~
d'information fournie par l'observation
syst~me:
197
- 186 -
J. Kamp~ De
F~riet
log (b-a) c 'est-a-dire que cette tive selon que m~me
b - a
quantit~
n~ga
d'information est positive, nulle ou
est sup~rieur, ~gal ou inf~rieur a l l .... De
Claude SHANNON
"g~n~ralisell
Ie cas fini
I J pJ log PJ
H
=-
H
= -l.ooP(X) log p(x)dx
par
:+00 (5.19)
p(x)
~
0
d~signant
la
densit~
de
probabilit~
.
Dans les deux cas Claude SHANNON appelle sterne ; nous nous sommes soigneusement
gard~s
H 1"'entropie" du syde 1'emploi de ce mot,
jusqu'aux applications oil. l'on pourra constater que, pour certains systemes
mat~riels,
H coincide effectivement avec 1'entropie
les Thermodynamiciens ; nous
esp~rons
ainsi
~viter
calcul~e
par
autant que possible
les reactions en chaines que produit I 'apparition du mot
"entropie" dans
un texte ; sur beaucoup d'esprits il produit Peffet de la cape rouge du matador sur un taureau ; il a des
[11]
r~sonances m~taphysiques [H. GRAD
va jusqu'a lui trouver un parfum IItMOlogiqUerr]. 11 est piquant de noter que la definition de C. SHANNON est mani-
festement
inspir~e
finissant
la "negentropie II
de la fameuse fonction
H
H de
L. BOLTZMANN,
du gaz de la tMorie cinetique par :
'1f(U,
v, w,) log flu, v, w) du dv dw
R3
198
de-
- 187 J. Kampl! De Fl!riet
f(u, v, w)
ou
d 'une
est la
mol~cule.
pe bleue du
densit~
de
probabilit~
- En vous parlant, je vois briller devant moi, la nap-
Lac de Ctlme; si on y versait toute l'encre
puis 70 ans pour
~crire
sur la fonction
Quels sont les rapports de la Claude SHANNON et de la
de-
!-
nquantit~
"n~gentropien
d'information"
H de
H de BOLTZMANN? Peutr~pondre
pour Ie moment, serait-il plus sage de
crivait Henri POINCARE'" en 1912,
d~pens~e
H de BOLTZMANN, Ie lac se
transformerait en une immense tache noire
~tre
(u, v, w)
de la vitess.e
a l'~poque
en citant ce
qu'~-
ou les quanta faisaient l'ob-
jet des discussions du monde scientifique ; "M. SOMMERFELD a propos~
a celle
une tMorie qu'il veut rattacher
de M. PLANCK; bien que Ie
seul lien qu'il y ait entre elles, c'est que la lettre
h figure dans les
/
deux formules"
(oeuvres d'Henri POINCARE t IX, p. 667) .
Le paradoxe
signal~
plus haut,
lit~ uniforme sur un intervalle
la
quantit~
quantit~
[a,
d'information dans Ie
bl
a l'occasion
de la loi de probabi-
est dtJ. essentiellement au fait que
cas continu n'est pas la limite de la
d'information pour des partitions de plus en plus fines
par (5.3), chaque
A.
J
tendant vers un point
d~finie
; en effet, dans l'exemple
consid~r~, si on divise La, bj en n intervalles ~gaux, on aura P(A.) = 1., d lOU H = log n et lIon trouve comme limite, non pas J n log (b-a) mais '+ ,< , Pour lever ces qu faux
consid~rations
difficult~s
heuristiques qui nous ont servi d'introduction ; au
lieu de parler de l'information l~e,
fondamentales, il faut remonter jus-
d~duite
il faut comparer les informations
d'une mesure de
a l'int~rieur
probabilit~
iso-
d'une classe de me-
sures. Consid~rant l'espace de mesure
199
(0,1)
nous choisirons une me-
- 188 -
J. Kampll De Fllriet
sure de rM{!rence t ,
qui n 'est pas n{!cessairement une mesure de
probabilit{! : elle est seulement assujettie a Mre G"-finie (10) ; cette g{!n{!ralisation est absolument n{!cessaire parce que lorsque
n = R la
me-
sure de comparaison naturelle est la mesure de LEBESGUE ; on a alors
m (R)
= + co
•
Ceci (!tant, quand on remplace la me sure de rM{!rence "t' par une mesure de
probabilit~
P la variation de la
nue dans l'observation que
a, d 'apr~s
WE. A
- log P(A)
quantit~
(esp~rance
(5.2), la valeur : P(A)
+ logt'(A) = - log't'(A)
Si l'on consid~re une partition finie [AI"" moyenne
d'information conte-
calcul~e
matMmatique)
An] de dans
n ,
l'hypoth~se
la valeur oll
Pest
la vraie mesure, de cette variation d'information a pour valeur:
L'in~galit~
P(A.) log _ J J 't"(Aj)
= -LP(A.)
I (P 11') n
(5.20)
j
(5.5) permet d'{!crire :
~ L~A.)
I (P It') n
.
J
J
- P(A.)] J
d'oll
I (P 11: )
~ t(n) - 1
n
en particulier, dans Ie cas oll t(n) = + co , on en dMuit seulement:
I (P \ 1') n
(10) c'est-a-dire que
n
est
< + co d~composable
U :COAj =
telle que
n
't( Aj)
par une partition
ApAk = ~
j
Fk
soit finie par tout
200
j ..
d~nombrable;
~ 189~
J. Kampe De Feriet
Lorsque la mesure
a la
est singuli~re par rapport
P
mesure t
leur comparaison n'a plus de sens ; en effet il existe, par dMinition ne partition de
0
compl~mentaires
en deux ensembles
A et
u~
AI
tels que:
par
cons~quent,
1:(A)
P(A')= 0
l(A') =1"(0)
pour cette partition :
=~
12(PI1') si
=0
PIA) = 1
l' (0) est finie
+ 0 x logt(O);
= ~ O(); si 1: (0) = +
12
Au contraire si
0()
0()
,
12 est ind~termin~e.
Pest absolument continue par rapport
a 1:'
P«'"t' il ne se
pr~sente
plus aucune
difficu1t~
; en effet pour un ensemble
-C (A) = 0 implique PtA) = 0 ; par
la condition correspondant
a une valeur bien
~ PtA) log ~~!l = ~
Si l'espace des phases
0
Ie terme
PtA) log PIA) + PtA) log 1; (A)
a0
d~termin~e ~gale
cons~quent
•
est infini on peut former une suite
infinie de partitions de plus en plus fines, alors sous certaines tions sur la mani~re
dont 1:(A.)...), 0 quartd
J
on a : (5.21)
lim
In(PI!)
~
I(Pil)
~
'I' (w)
d~signe
-1
condi~
A. tend vers un point
J
f(W) log
f (w)
w
d 1:
0
n~+O()
OU
A
la
d~riv~e
de RADON-N1KODYM de
201
P
par rapport
- 190 -
J. Kampe De Feriet
a "t'
(dMinie
a un
ensemble de 1: -mesure nulle pres, (11)).
On comprend alors pourquoi (5.19) n'est pas la limite de (5.10); on ne peut comparer que des mesures absolument continues I June par rapport
a I 'autre.
Par rapport
a
Rune mesure de probabilite sur un
ensemble fini peut toujours t!tre consideree comme constituee par un nombre fini de masses PI' ... Pn ~ placees en des point toute me sure de probabilite
[Xl'···
xJ ;
P admettant une densite de probabilite
a cette distribution de masses. Les deux cas sont irreductibles l'un a l'autre. p(x)
est singuliere par rapport
Si lIon ecrit (5.10) sous la forme '\ p. H = - L p. log..:.1 . J 1 J
on voit qu'elle represente la variation d'information quand on passe de la mesure l' , uniforme sur I 'ensemble fini {WI' ... W n) (tous les points ont la mt!me masse) notons que l' (f2) son
1.n l' ,
a la
mesure definie par les masses
p.
J
= n, on pourrait prendre comme mesure de comparai-
qui est une me sure de probabilite. Mais ce choix est im-
possible des que I 'on generalise Ie cas de I 'ensemble fini, mt!me par exemple pour un ensemble ayant une infinite denombrable de points
La mesure uniforme l' do "'"ant
:
a chaque point la mt!me me sure 1
est telle que
(11) L'ouvrage de KULLBACK [19J est Ie premier, a notre connaissan ce, qui ait systematiquement adopte ce point de vue; nous y renvoyons pour un expose detaille des demonstrations,
202
- 191 -
J.
't'(O) = +
00
Kamp~
De Feriet
;
la formule
_ \"j=+oo Pj H - - Lj=l Pj log '1 repr~sente
encore ici la variation d'information quand on passe de 1:'
a
s~rie
P
; la
ne converge pas
n~cessairement
d'ailleurs, bien que
j=+oo _ Lj=l Pj - 1
p. ~O J
de telle sorte que lIon peut avoir Dans Ie cas 011 d'elle-m~me
0 = R
H =
+ 00
•
• la mesure uniforme qui s'impose
est la mesure de LEBESGUE
m; on ne peut donc lui com-
parer que des mesures absolument continues par rapport
«
p d~riv~e
La
simplemente la p(x)
:
m
de RADON-NIKODYM de
densit~
a elle
de
probabilit~
P
par rapport
a
m est
:
!:~{XI dx "1,
~O
On voit donc que la formule de WIENER-SHANNON
1 +00
H=repr~sente,
non pas, une
_:(X) log p(x) dx
quantit~
d'information, rna is la variation d'in-
formation quand on passe de la me sure de LEBESGUE probabilit~
leurs de me une
-
P
; Ie fait que cette
00
a + 00
quantit~
,
qui
~tait
int~grale
a la
mesure de
peut prendre toutes les va-
absurde quand on la considerait com-
d'information, n'a donc plus rien de choquant.
203
- 192 J. Kampe De Feriet
VI - LE PRINCIPE DU MAXIMUM DE LIENTROPIE EN MECANIQUE STATISTIQUE. Reprenons les definitions et les notations du § IV ; nous considerons un systeme materiel en equilibre statistique ; la loi de probabilite, definie sur une C' -algebre de parties de par rapport
a la
me sure de LEBESGUE
nest absolument continue m
dans
2k
R ;
nous supposons
que la densite de probabilite (derivee de RADON-NIKODYM de
a
rapport
m) ne depend que du seul invariant
E(w)
P
par
(energie totale du
systeme conservatif.) . Nous appellerons entropie du systeme materiel la variation de la quantite drinformation quand on passe de la mesure de LEBESGUE a la mesure de probabilite
m
P I
H = I(Pim) = -/
(e.1)
r,-E(w)!
In -
log rl-.E(w)l dw
~
droll, en appliquant a cette integrale, la formule du changement de variaE(w)
ble
=x
on tire de (4. 3)cette expression de lrentropie : i+Oo
H
(6.2)
= -.
r(x)
10-
log r(xfl
S(x)dx
./
Oll
S(x)
est la fonction de structure du systeme definie par
En appliquant l'inegalite gral nous obtenons
(5.5) ala fonction sous Ie signe inte-
,"+00
,
H.~
r- 1 r(x)Lr(x)
- 1] S(x)dx.
JO d'oll en tenant compte de
H
~
(4. 2) .
(4.3) et (4.5) :
V(+oo) - 1
204
- 193 -
J.
KHINCHIN
[18J
fait itnplicitement
V(+oo) = +
(6.3)
Kamp~
De
F~riet
l'hypoth~se
00 ;
dans ce cas on obtient seulement pour l'entropie (~.
H ~
4)
+ 00
;
il serait in:teressant de rechercher s'il n 'existe pas des
lesquels ces
V(+oo)
syst~me
(volume de l'espace des phases
n)
syst~mes
pour
est fini; pour
l'entropie serait bornl:\e supl:\rieurement.
Nous allons prouver la proposition suivante : La distribution canonique de GIBBS rend maximum l'entropie
H lors-
que lIon suppose donnl:\e la valeur moyenne de lll:\nergie
E =
(4.6)
l+~
r(x) S(x)dx;
.10 elle est la seule
a possl:\der
cette propriete.
Pour tenir compte des conditions (4.5) r(x)
et
(4.6)
imposl:\es
introduisons les multiplicateurs de LAGRANGE fV et \3 H -
-:l, -
;~ E = 10t:(X)'-. lOg~( ) -·Iv r x
;xJ S(x)dx
. S(x)dx
En appliquant l'inl:\galitl:\ H _,
-.,E
(5.5)
~
(+oo;- e-·\ - r,x - r(x)]
./0'
Ie second membre a pour valeur :
205
S(x)dx
.
a
- 194 J. Kampe De Feriet
en introduisant la fonction de partition nous
d~terminons
Z(\3)
dMinie par
(4.11)
si
:'\, par :
cV = log Z(G) I
(6.5) nous obtenons
H ;S; log Z(~) + ~E l'~galit~ ~tant
satisfaite si et seulement si pour tout
x
- ~x
(6.E)
r(x)
= eZ(p)
(4.6) soit satisfaite il faut et il suffitque
Pour que la condition
+00
x
(6.7)
o
Si nous faisons (comme KHINCHIN) de partition,
transform~e
existe pour tout ~ (pour
de LAPLACE
> 0,
(1 > 0) et ses
par
cons~quent
l'hypoth~se
que la fonction
de la fonctibn de structure
c lest alors une fonction analytique de (3
d~riv~es.
Z'(P)
.-1+:e-~XS{X)dx
Z"(~)
=}o
( +00 2 -r;\.x
existent et sont continues pour tout Llin~galit~
= E
S(x)dx
Z (~)
x e \ S(x)dx
@> O.
de SCHWARTZ nous donne
:
206
<0
>0
- 195 -
J.
Kamp~
De
F~riet
Z' 2 -Z" -(-)
Z Done
log Z((3)
Z
est une fonetion eonvexe de
~
dans
(0, +00) .
Introduisons la fonetions: Y(!3)
= e ~E Z{(3);
puisque
> 0, la fonction
log Y(~)
est ~galement eonvexe
= log Y(+oo) = +00
log Y(O)
(6.8) En effet
d'autre part
Y(O) • Z(O) • /:"'S(X)dX • V(+",) • + •
si lion admet llhypothese de KHINCHIN
(6.3).
En outre: Y(::l)
> -
IE/2 I
~2 E
)0
>e dlou il
r~sulte
n r~sulte
:
Y(+oo)
1"/2 E
S(x)dx
=Ke
= + 00
imm~diatement
des
propri~t~s
de
log
Y(~)
A
existe une et une seule valeur
(-1 pour laquelle
d
d~ log Y((3)
207
::;
E
+
d
:
~ log Z(~)
=0
qu lil
- 196 -
J.
Par
Kamp~
De Feriet
cons~quent
HMAX = log Z( ce maximum
~tant
p)"
+ ~E
atteint si et seulement si
.
r(x)
= e - '~ Z( '3) \
~ ~tant
la racine de
l'~quation
- - E
qui existe toujours et est unique si Nous retrouvons ainsi
m~thode
•
§ IV
bas~e sur la notion de
11 faut souligner que contraire-
statistique necessaire et exhaustive.
a la
0
a nouveau la distribution canonique, par
une m~thode toute differente de cene du
ment
>
E
de KHINCHIN, ou la distribution canonique n'appa-
rait que comme une limite lorsque Ie nombre de degres de liberte augmente
indefiniment, les deux methodes que nous avons decrites sont
compl~tement ind~pendantes
si bien
a
Exemple A:
attir~
k
k = 1
que
a
de
k
k =
l~O.
et la demontration s'appUque aus-
oscillateur lineaire .
Considerons un point materiel se mouvant sur un axe 2 vers l'origine par une force - m ~ x . Po sons q
p ::
=x
l'espace de configuration
nc
est l'axe
n
I 'axe Op , l'espace des phases v gie a pour valeur :
208
Ox,
dx dt
0'f' n
l'espace des vitesses Ie plan Oqp , L 'ener-
- 197 J. Kampe De Feriet
E
la II surfaee II
m
=2" (p
2
+»2 q2)
~ est done repr~sent~e par l'ellipse 2 .2 2 2x P + v q =m
Ie IIvolume II
V(x)
int~rieur
a l'ellipse est
~gal
a
2 ITx
V(x) = - -
m ""
d I ou la fonetion de structure (6.9)
S(x)
On en dMuit
imm~diatement
Z(r.:l)
(6.10)
\~
L'~nergie moyenne
E
:
21T
= mV
la fonction de partition
=!1T .!. m),l ~
~tant donn~e la densit~ de probabilit~ qui rend
l'entropie maxima est dMinie par la distribution canonique
mv
~:l
r(E) = -211 i'". e
(6.11)
..
~E
'i
~
;)U
~ est la racine (unique) de l'equation
elest-a-dire (6.12)
Llentropie maxima a pour valeur (6.13)
HMAX
21r
Si lion eonsidere l'ensemble re un nombre
-
= 1 + log -;-'i> E repr~sentatif
N tres grand dloscillateurs
209
de GIBBS, elest-a-di-
1in~aires,
~voluant
sans
- 198 Kamp~
J.
agir les uns sur les autres , l'ensemble
~tant
De
F~riet
~quilibre
en
statistique
avec la distribution (6.11), l'entropie telle que la dMinissent les Physiciens sera
pr~cis~ment
:
S = NH
Exemple
B.
L 'exemple nous parait b~tard,
MAX
pr~c~dent
int~ressant
id~es
est une illustration des
d'en esquisser un autre, appartenant
ou l'introduction d I un choc
~lastique,
de vue de GIBBS et de BOLTZMANN ;
m~lange
Consid~rons
que fois qu 'il atteint Ie point Z. = 0 . Si nous
un point
OZ
l'~nergie
(6.14) Par
E
hypoth~se
nergie
E
n
c mouvement
~lastique
cha-
par W la vitesse du
w2
= m(gZ~)
chaque fois que Ie point
mat~riel
atteint le point
instantan~ment remplac~e
par
0, la
+ W1 :
l'~
reste donc invariante
L'espace de configuration dans
mat~riel
a pour valeur
- W1 (W 1 > 0) est
vitesse
d~signons
a ce type
les deux points
pes ant, mobile sur une verticale OZ, subissant un choc
point sur
de GIBBS ; il
nest Ia verticale OZ (0 ~ Z < + 00);
c
Ia trajectoire est un intervalle p~riodique
0
~
Z
~
H parcouru d'un
; I'espace des vitesses est I'axe OW(- 00
I'espace des phases est Ie demi-plan OWZ, confond avec Ia IIsurface'1.
x
dMinie par W2
m(gZ~)
= x;
210
Z
~
< W <+00)
0 . L'orbite
se
;
- 199 -
J.
,-
t-, \
Kamp~
De
F~riet
c 'est un arc de parabole AHB oil.
DB = + W:+
~
1
"
~\
I
I
OA =
-
V2Xm'
WI
L'orbite est discontinue, Ie point·
B
0
repr~sentatif
squement de
TtW passant bruA enB
a
chaque
choc. On obtient facilement :
3/
2 V(x) =- (2x)
2
m
3g
d 'oil. la fonction de structure : .!..(2x) S(x) =mg m
1/
2
et la fonction de partition :
-
-3/
Z(~) =~g \/2;: ~ 2 E >0
La valeur moyenne de l'~nergie t~
de
probabilit~
~tant donn~e, la dertsi-
qui rend l'entropie maxima est la distribution canoni-
que.
= mg
(6.15)
r(E)
oil.
l'~quation
@ est la racine de
rm V21T
£.@l=_~ 1 Z (8) \
2 ~
211
A
\3
'E A
3/ 2 e-
-A
=- E
~=
3
iE'
• 200 J.
Kamp~
De
F~riet
on a H
MAX
3 2
1 mg
= - + log(-
Si dans (6.15) on remplace voit que les deux variables
211 Vi -) m E
al~atoires
3 2+ - log - E 2 3
par Z
son expression (6.14) on (cote) et
W
(vitesse) sont
des variables independantes ; leurs lois de probabilite ont respectivement pour densites:
Z>O
(6.16) et (6.17) Z et
C! (W) =
\1 ~,
W suivent donc respectivement la loi de LAPLACE et la 10i
normale . Si nous
consid~rons
"tres grand" nombre
maintenant un ensemble representatif d1ull
N de ces points materiels, Ie nombre des
points compris dans la couche
(ZJ Z+dZ) sera:
ce qui est precisement la distribution de la masse specifique dans une atmosphere en equilibre a temperature constante selon la loi barometrique de LAPLACE, la temperature absolue To
~tant
definie par :
1
= --A mR~
Du fait que notre modele, melange hybride de GIBBS et de BOLTZMANN, nous conduit ala loi barometrique de LAPLACE auronsnous la naivete de croire qu1une atmosphere en equilibre est effectivement constituee par un nombre tres grand de billes tombant en chute
212
- 201 -
J. Kampe De Feriet
libre au dessus d lun plan
~lastique
parfaitement poli ? ..
Dans d'autres problemes de la Physique tMorique, dont
ll~che
a d~brouiller que celui-ci, combien de fois la
veau est moins ais~
r~us
site d lun modele n lest-elle pas prise pour une preuve que la Nature a llextr~me bont~
de se conduire effectivement selon les regles de notre
modele ? •. C.
GAZ de KNUDSEN. On dit qulun gaz est un gaz de Knudsen lorsqu lil est suffisamment
rarMi~
mol~cules
pour que les
ais~
m~me
de montrer que,
lIautre aucune force, des
~tre consid~r~es
mol~cules
si elles nlexercent llune sur
ne peuvent pas
les l.lnes des, aut:r:es (au sens du Calcul des se reduisent pas
a des
Soit en effet l~cules
;
d~signons
position de la mol~cule
Probabilit~s)
par·
S(P;()
une sphere de centre
r~servoir
mol~cules
D . La
et de rayon
conditionnelle de la
de centre P 2 quand on connait la position de la
PI doit
ind~pendants
P
de forme spMrique et
probabilit~
~videmment
satisfaire aux conditions
Prob
r
Prob
[p 2 ~ D - S(P l' 2 E~ = 1
P2 f
.
213
c )J
S(P l' 2
ce qui montre clairement que les deux points sont pas
, si elles ne
D le domaine de llespace Oll se meuvent les mo-
mol~cule
de centre
~tre ind~pendantes
points .
(. Supposons qu lil n I y ait que deux de rayon E dans le
ind~
comme
(voir H. GRAD [1~ p. 242).
pendantes les unes des autres 11 est
puissent
:
=0
al~atoires
PI et
P 2 ne
- 202..-
J. Kampe De Feriet
Pour nous Ie un ensemble de
t~ C
maine
v~ritable
n points
mat~riels
PI""
constitu~
par
P n conte nus dans un do.
R 3 ; l'espace de configuration est done Ie domaine n
~,
si nous
gaz de KNUDSEN est done
d~signons
par
c
= (.I,;,n =. _~1\ ~, , .. \
VI""
uA L,.
R3n
Vn les vitesses des points
mat~riels
l'espace des vitesses est ; nV
= R3 l(
...
)(
R3 = R3n
d'ot! l'espace des phases
n = Ll n x R3n Aucune force ment
ext~rieure
cin~tique
n'agissant sur ces points leur
; elle a pour valeur
~nergie
est pure-
;
m,
222. E = -2 L (u.+v.+w,J J J J
(6.18)
si nous supposons que taus les points ont la mt!me masse et si (u. v. w.) d~signent les projections de la vitesse V,. Nous admettrons J J J J que E est Ie seul invariant du systeme, ce qui exprime que les chocs des points avec la paroi du vent mt!me avoir deux
r~servoir
a deux
sont
des chocs
~lastiques
~lastiques
; les points peu-
(notons cependant
que la probabilit~ de pr~sence simultan~e au mt!me point de instant
t
donn~
de deux
mol~cules
P, et J
P
Le volume de l'hypersphere ; -j=n 2 2 ) , ' l(u,+v,+w, -J= J J J
214
~
2x m
k'
A , ,a
est nulle).
un
- 203 Kamp~
J.
De
F~riet
3n
etant .
(2Y;x) 2 m
(Vb.
on obtient donc
designant Ie volume du domaine aCR3)
N
V(x)
= VII
1
- rc
dloit la fonction de structure du
h
211 3n
- 3 (-::i"2 2n +'1) m
2.
Xi
syst~me mat~riel
S(x)
et par consequent la fonction de partition a pour valeur :
3n
Z(~) = V;(2:)"2
(6.19)
-3n
~2
Supposons donnee la valeur moyenne me
soit
~
E
de l'energie du syst~-
la racine de llequation
~=
3n 2
Z (~)
1 ~ = - E
A
~=
3n
2E
La loi de probabilite qui rend l'entropie maxima est donne par: A
r(E) ::
-(aE e Z(~)
et
r(E) = 0
En designant par
1.1 (x, y, z)
si au mains pour un j l'indicateur
215
de t::. :
P.¢ /). J.
- 204- Kamp~
J.
I~(x,
la
densit~
probabilit~
de
~quilibre
riel en
y, z)
=1
(x, y,
= 0
(x, y, z)
De Feriet
z)€b.
1A
qui rend maxima l'entropie du systeme
mat~
prend la forme :
" -m~\-=" "2 L ( u.2+ v.2+ w.2) J
r=e
(6.20)
J J
J
"IT I (x., y, z.) . fJ. J J J J
Le principe du maximum de l'entropie nous a done permis de prouver (puis que
r
d~eompose
se
1 - la position de ehaque les autres 2
mol~eules
- la position de ehaque
mol~eule
~
la vitesse de ehaque tes les autres
4
- pour une se sont
4n
produits) que
est
ind~pendante
de la position de tilutes
est
ind~pendante
de sa vitesse et de
. mol~eule
mol~cules
la vitesse de toutes les autres 3
en
mol~cule
est
ind~pendante
des vitesses de tou-
mol~eules
mol~eule
les trois composantes
ind~pendantes
et ont la
m~me
u., V., w. de la vitesJ J J loi de probabilit~ (isotropie)
Notons que 5
11 (x, un
y, z)
ne se d~eomposant pas (sauf dans les cas OU t.:\
parall~l~pipede
tions de x,
de y
parallele aux axes) en un produit de et
eule ne sont pas, en
de
z
g~n~ral,
, les trois
eoordonn~es
ind~pendantes
.
Le maximum de l'entropie a pour valeur
216
est
3 fone-
d'une
mol~
- 205 -
J. Kampe De Feriet
3n
(6.21)
HMAX =
211
'2 (l+logm) + n
• -3/ 2 log (V[\ 3 )
Si nous admettons que notre systeme de repr~sente
effectivement un gaz de KNUDSEN Ie
de ce gaz sera
mat~riels
temp~rature
absolue
par :
3 E=-nkT
(6.22)
ou
d~finie
n points
2
k est la constante
de BOLTZMANN
k = 1,380
x
10
-16
c.g.s.
On en tire et l'expression de I 'entropie prend la forme : 3n HMAX = "2 (1 + log
c 'est-a-dire
pr~cis~ment
l'entropie
2~k
~)
+nlog (Vb T
calcul~e
par les
3/ 2
)
m~thodes
habituel-
les de la Thermodynamique. Ne devrions-nous pas tltre presque surpris qu'un scMma aussi simplifi~
donne un
r~sultat
en parfait accord avec la Pysique
exp~ri
mentale? .. Nous nous contenterons de souligner que, pour nous, cet accord justifie l'usage du mot entropie en d~signer
M~canique
Statistique, pour
la mesure H de l'incertitude ou de l'information telle qu 'elle
a ~t~ dMinie au
§ V pour les systemes les plus
217
g~n~raux.
- 206 -
J. Kampl! De Fl!riet
VII
-
CONCLUSION.
le~ons
Si ces M~canique
attirer votre attention sur Ie profit que la
Statistique classique peut tirer des progres du Calcul des
Pr6babilit~s,
est loin
ont pu
manqu~
elles n'auront pas
d'~tre
leur but; l'exemple choisi ici
unique; pour n'en citer qu'un seul autre n'est-il pas cu-
rieux que des outils puissants comme la raissent jamais dans les
expos~s
th~orie
actuels de la
des martingales n'appa-
M~canique
Dans plusieurs directions il y aurait sans doute a m~thodes
consacr~es
un peu archai4ues ,
attitude n'est nullement
motiv~e
op~rer
Statistique ? une refonte des
par un long usage. Une telle
par un manque de respect pour les pion-
niers, qui ont eu Ie courage, avec leurs htiches en silex,de s 'attaquer
a des
diplodocus ; bien au contraire , clest en profitant de nos outils
matMmatiques modernes que nous les honorerons, en les selon la
pens~e
de PASCAL:
d~passant,car,
"si nous sommes grands, c'est parce que
neuS semmes montes sur les epaules de nos ancIHres".
218
- 207 J. Kampe De Feriet BIBLIOGRAPHIE
1. -B.P. ADHIKARI, , D.D. JOSHI - Distance, discrimination et sum~
r~
exhaust if . Pub. Inst. Statistique Un. Paris - 5, 1956, p. 57-74.
2. -A. BLANC-LAPIERRE, P. CASAL, A. TORTRAT tMmatiques de la
M~canique
M~thodes
ma-
Statistique. Paris, 1959.
3. -L. BRILLOUIN -
Science and information Theory. 2 nd ed.
4. -E. B. DYNKIN -
New York, 1962.
Necessary and sufficient Statistics for a family
of probability distributions.
Selected Transl. in Math. Statistics and Probability. 1. M. S. - A. M. S. I - 1961, p. 23-40.
5. -Sir Ronald FISHER -
Theory of Statistical estimation. Proc. Cambrige Philosophical Soc. 22 - 1925, p. 700-725.
6. -J. W. GIBBS (a) -
Elementary Principles in Statistical Mechanics. New York, 1902.
(b) -
Principes
~l~mentaires
de
M~canique
Statistique.
Paris - Hermann, 1926. 7. - H. GRAD -
On Kinetic Theory of Rarefied Gases. Comm. Pure and Applied Math. 2; 1949, p. 331.
8. -H. GRAD·
Statistical Mechanics of Dynamical Systems
with Integrals other than Energy.
Jour. Physical Chem. 56, 1963, p. 1939.
219
- 208 J. Kampe De Feriet
9. -H. GRAD -
Statistical Me.chanics, Thermodynamics and
Fluid Dynamics of Systems with an arbitrary number of Integrals. Comm. Pure and Applied Math. 5, 1952, p. 455. 10.- H. GRAD -
Kinetic Theory of Gases. Handbuch der Physics XII, p. 205 Berlin 1958
11 -H. GRAD-
The many faces of Entropy. Comm. Pure and Applied Math. 15, 1962, p. 325.
12 • R. JANCEL-
Les fondements de la Mecanique Statistique Paris, 1963.
Classique .et Quantique. 13.
~.
T. JAYNES-
Information Theory and Statistical Mechanics. Phys. Rev. 106, 1957, p. 620, 107, p. 171.
14.--D. D. JOSHI -
L'information en Statistique MatMmatique et
dans la tMorie des communications. Pub. Inst. Statistique Un. Paris, 8 - 1959, p. 83-161. 15.. -J. KAMPE DE FERIET - Statistical Mechanics of continuous Media. Proc. Symp. Applied. Math. 13 - 1962, p. 165-198.
" lET 16. -J. KAMPE" DE FER
Les
int~grales al~atoires
des equations
aux derivees partielles et la Mecanique Statistique des milieux continus.
Atti 2 Reunione Math. expression latine Firenze • 1961 - p. 152-166.
220
- 209 J. Kampe De Feriet /
~
17. - J.
KAMPE DE FERIET - Information theory and Statistical MeBangalore, 1963.
chanics. 18. - A.!. KHINCHIN-
Mathematical Foundations of StatistiNew York, 1949.
cal Mechanics. 19. - S. KULLBACK-
Information Theory and Statistics. New York, 1959. Axiomatics of Classical Statistical Me-
20. - R. KURTH-
New York, 1960.
chanics. 21. - B. MANDELBROT -
An outline of a purely phenomenological
Theory of Statistical Thermodynamics.
1. R. E. Trans . Information Theory 1. T. - 2, 1956, p. 190. 22. - B. MANDELBROT -
The role of Sufficiency and of Estima-
tion in Thermodynamics.
Ann. Math. Statistics, 33, 1962,
p. 1021. Die Herleitung der Grundgleichungen
23. - W. NOLL-
der Thermomechanik der Kontinua aus der Statistichen Mechanik. J. Rat. Mechanics.Analyses
- 4 - 1955 - p. 627-646. ~
24. - H. POINCARE nique
Les Methodes Nouvelles de la Meca-
O~leste.
I - Paris, Gauthier-Villars, 1892.
25. - Cl. SHANNON, W. WEA VER - The Mathematical Theory of ComUniv. lllinois Press, 1949.
munication. 26. - M. TRIBUS-
The Maximum Entropy Estimate in Re-
liability. In: Recent Developments in, Information and Decision Processes.
New York, 1962.
221
- 210 -
J. Kampe De Feriet
27. - M. TRIBUS, R. B. EVANS - The Probability Foundations of Thermodynamics.
Appl. Mech. Rev. 16, 1963, p. 765
28. - C. TRUESDELL -
Ergodic Theory in Classical Statistical
Mechanics. in: Ergodic Theories. Proceed. Int. School of Pysics, course 14, Varenna - New York, 1961.
The Homogeneous chaos.
29. - N. WIENER-
American Journ. of. Mathematics - 49 - 1938, p. 897-936. 30. - N. WIENER -
Cybernetics. Paris, Hermann, 1948.
~:
(ajoutee
a la
correction des epreuves, 1
Mai 1965).
Le ProfesseurHarold Grad a bien voulu attirer notre attention sur un enonce du Chapitre XI : "Maximum and minimum properties of various distributions in phase" [6aJ p. 130, OU Gibbs demontre "Theorem II : if an ensemble of systems is
canonically distributed in phase , the
average index of probability is less than in any other distribution having the same average energy". 11 n1a donc manql.le
a
J. W. Gibbs, pour donner
a ce
resultat ma-
thematique (qu1il demontre d1ailleurs en quelques !ignes, en usant une inegalite equivalente
a (5.5)
) toute sa signification profonde, que Ie
cadre d1une theorie generale, comme celui qui nous est fourni aujourd Ihui par la theorie de lIinformation.
222
CENTRO lNTERNAZlONALE MATEMATlCO ESTlVO (C.l.M.E.)
M. LUNC
EQUATIONS DE TRANSPORT
Corso tenuto a Varenna (Como) dal 21 al 29 agosto 1964
223
EQUATIONS DE TRANSPORT PAR Micha~
LUNC
(Academie polonaise des Sciences)
Introduction. Cet expose aura pour but de montrer comment
a partir
de la
theo~
rie cinetique des gaz on peut obtenir les equations macroscopiques qui decrivent llevolution du fluide reel, en llocourence dlun gaz rarefie. Le probleme general est
extrem~ment
vaste car il englobe la
cine~
matique et la dynamique du mouvement mais, aussi, Ie probleme de radiation, de diffusion de matiere, la propagation de la chaleur et une quantite d lautres choses. 11 est bien evident que lIon ne pourra faire ici qulun essai de presentation de quelques problemes choisis un peu arbitrairement et subjectivement. Mais les problemes qui seront
abor~
des ici ne seraient pas tous menes jusqu Ill. la solution. Dans certains de cas
celle~ci
ble, peut
~tre
existe, dans dlautres elle reste
a trouver.
11 nous sem-
a tort, que des problemes poses et non encore resolus
peuvent Mre plus interessants pour beaucoup dlauditeurs que les blemes llfinis II. On
t~chera
pro~
donc de faire ici quelques ouvertures sur
Pavenir et de susciter, nous Pesperons, des recherches nouvelles. Grandeurs moleculaires. Le concept fondamental qui est theorie cinetique des gaz
a la base des applications de la
a la mecanique des fluides : c lest llexisten-
ce dlun systeme materiel bien defini-Ia molecule. A chaque instant t ce systemes peut
~tre
parfaitement defini par deux ensembles conjugues
des grandeurs : les coordonnees generalises et les vitesses generalises. Ce dernier des ensembles peut
~tre
remplace par un autre
celui des moments conjungues.
225
ensemble~
- 216 M. Lunc
L'ensemble des
coordonn~es g~n~ralis~es
sera
ou, d'une manH~re abr~g~e
q. /i=l, 2, .,' s/
par
d€sign~
..... q
par II est bien
1
~vident
que
qi
ne possMe pas,
rielles . L 'ensemble des vitesses
g~nE!ralement, gE!nE!ralis~es
de propriE!tE!s vecto-
sera dE!signE! par
.
qi
~/ q cette grandeur, non plus, ne possede pas de carac-
ou bien par
tere vectoriel/ . L'ensemble des moments conjuguE!s sera dE!sign~ par p. 1
...l.
ou par
p .
Vne remarque importante
la connaissance des moments
s'impose des cet instant. On sait que gE!nE!ralis~s
d 'un systeme matE!riel dE!-
pend essentiellement de la connaissance de la fonction de Lagrange dE' de ce systeme. Or la fonction de Lagrange ne peut
~tre
connue que si
l'on connatt l'action globale exerc~e par tout ce qui se trouve en dehors de notre systeme matE!riel. En l'occurrence, la fonction de Lagrange pour notre molE!cule doit se dE!duire des forces que toutes les autres molE!cules exercent sur la ntltre . Comme la fonction de Lagrange ne contient d'autres informations sur ces forces que les valeurs ,
des coordonnees IE!cule il en
g~nE!ralis~es
r~sulte
et des vitesses gE!nE!ralisE!es de notre mo-
que toutes les autres molE!cules n'interviennent que
collectivement. Clest, bien entendu, une excessive simplification. Entre autres dMauts une telle maniere de
proc~der
fait fi des rencontres de
notre molE!cules avec d'autres , La fonction de Lagrange devrait contenir les informations sur toutes les molE!cules du gaz. Mais Ie nombre de molE!cules E!tant toujours tres E!levE! la chose est tout
a fait
impos-
sible. Ainsi on ne peut pas trouver, en toute rigueur, les moments conjuguE!s des coordonnE!es gE!nE!ralisE!es de la molE!cule. Et pourtant clest ce qu'on fait couramment. Autrement dit on attribue cules extE!rieures la
propri~tE!
a toutes
les molE!-
de crE!er un champ des forces local. On
226
- 217 M. !'unc
suppos~s tr~s
excepte, toutefois les intervalles de temps, squlune
mol~cule
proximit~
quelconque se trouve dans la
a la
force collective des autres
de notre
mol~cules.
Pendant ces
courts intervalles de rencontre la fonction de Lagrande doit ment
d~pendre
ses de deux
coordonn~es g~n~ralis~es
des
mol~cules.
~tre
~quations
de
La distinction entre les
quelqulutilit~
~videmment
doit
n~cessaire
et , peut Mre, des vitesp~riodes
ou la fonction compliqu~e
de Lagrange peut Mre connue et celles ou elle est trop pour
mol~
~tre n~glig~e
cule qui fait que son action individuelle ne puisse plus par rapport
courts, lor-
~cho
trouver son
dans les
de mouvement. Souslignons, une fOls de plus, la brievete des
periodes de renco'ntre - condition gaz est plus
rar~fi~
d lautant mieux satisfaite que Ie ~
. Lleffet des rencontres doit figurer dans les mani~re
quations uniquement de la
statistique.
Pas sons maintenant au sujet prope de cet
expos~.
Si les coordon-
n~es g~n~ralis~es et les vitesses g~n~ralisees d rune mol~cule sont con-
nues qui
aun
certain instant
d~pendent
attach~es
~
de
t
~
alors nous pouvons former des fonctions
q , q . et
a chaque
mol~cule
t.
Ces fonctions sont en quelque sorte
individuelle. Soit ~
G = G ( q,
....4, t
)
(1)
une telle fonction. Lorsque Ie mouvement de la chaque instant la valeur de
mol~cule
est connu , on connaitra a
G qui devient ainsi une fonction de
seulement . Or la connaissance de mouvement de chaque G
me on lla vu, impossible. On devra donc ction de
2s+1
et
a chaque instant est li~e a la conLagrange a chaque instant ce qui est, com-
par la suite la connaissance de naissance de la fonction de
mol~cule
t
variables ind~pendantes
227
consid~rer
qi'
qj
et
t
G comme fon-
- 218 M. Lunc
La fonction IE:lculaire
attachee
G
/propriE:ltE:l
a la
molE:lcule sera appellE:le grandeur mo-
molE:lculaire selon certains auteurs/. Parmi des
innombrables grandeurs molE:lculaires certaines possedent des propriE:ltE:ls physiques particulierement simples . Telles seront la masse, la quantitE:l de mouvement, l'E:lnergie d'une molE:lcule et certaines autres. Souvent il est plus facile de dMinir une quantitE:l par ce qu 'elle n1est pas, plutot par ce qu'elle est. Prenons, comme exemple une moIE:lcule ponctuelle et supposons que donnE:les cartE:lsiennes et (La position
i
q.1=c.1
/i=1, 2, 3/ soient ses coor~
q.= X. 1
1
ses vitesses.
et la vitesse
c sont
maintenant des vrais vecteurs~
ConsidE:lrons l'entourage du point de l'espace de phase stant
t. Une fonction quelconque de
x,"t
et
t
ne sera pas une gran-
deur molE:lculaire si cette fonction n 'est pas attachee particuli ere, mais
a l'ensemble
Ct, c) a l'in-
a une
molE:lcule
des molE:lcules ayant les memes coor-
donnE:les dans l'espace de phase. Ainsi, la fonction de distribution -> ...\
f(x, c, t)
ne sera pas considE:lrE:le comme une grandeur moleculaire ,
bien que possE:ldant certaines de ses proprietes . La fonction de distribution est, en somme, une propriete collective et non individuelle. Cette distinction entre les proprietes collectives et individuelles tient, en fin de compte, a la difference entre la partie previsible du mouvement entre les rencontres et Ie mouvement imprevisible pendant celles-ci L 'observation macroscopique ne permet de mesurer que les valeurs moyennes des grandeurs moleculaires. L'E:lvolution de ces valeurs moyennes qui est Ie sujet principal de la mecanique des fluides generalisE:le-ce sont prE:lcisement les equations de transport qui vont Nre etudies ici. La definition de la valeur moyenne d'une grandeur moleculaire G (q., 1
q. t) J
est la suivante:
228
21-9 -
M. Lunc
= g(q.,1 t) =J G (q.~., t) lJ
f (q., 1
q.,J t)
d
"4/
s
j
f(q.,
q., t) dS-c;
(2)
1 J
-'"
d q d~signe un ~lement s-dimensions nel de l'espace de vitesses g~n~ralis~es et l'int~gration est effectu~e
Dans cette ~quation Ie symbole
dans l'espace entier de vitesses leur moyenne
donn~e
par
g~n~ralis~es
. La dMinition d'une va-
(2) peut, bien entendu, etre
.......... les fonctions de q, q et t et qui ne sont pas,
~tendue
a toutes
n~cessairement,
des
grandeurmoMculaires. Il est evident que dans certains cas cette dMinition peut conduire aux integrales divergentes et deporvues, alors, de sens physique. Attirons ici l'attention sur un probleme que nous croyons fort ressant et qui, nous semble -t-il Na pas nous choisissons pour
r~pr~senter
la place des vitesses
g~neralisees
G(q.,p.,t) 1
J
a
conjugu~s
. On dMinira dans Ie nouvel espace g~n~ralisees
p. une nouvelle moyenne de la grandeur
pa~=
Supposons que
notre gaz les moments
de phase comprenant les coordonnees conjugues
~te r~solu.
int~
q. et les moments 1
mol~culaire
g(q.,t) = jG(q.'P.,t) f (q.,p.,t) d Plff(q.,p.,t) d P 1
1
J
1
J
s)
1
J
s
(3)
Il n'est pas du tout certain que cette nouvelle moyenne qui, evidemment, est differente, dans Ie cas l'~quation
(2)
se comporte d'une
g~n~ral,
mani~re
de la moyenne dMinie par independante. Voici done
un interessant sujet de recherche. /
Etat d'equilibre d'un gaz Si f ( q., ci.., t) r~pr~sente la fonction de distribution 1 J des vitesses g~n~ralis~es du gaz, et si
of
<;Df
=it
(4)
229
- 220 M. !'unc
repr~sente ll~qllation de BOLTZMANN
Chapman~Cowlingl
que de
l!op~rateur
ou'Dest un
Ion utilise les notations op~rateur diff~rentiel et
Nous disons que Ie gaz se trouve dans
of e
L'effet des rencontres distribution des vitesses
de
~
mol~culaires.
des rencontres
ut
classi-
lI~tat d'~quilibre
lorsque
(5)
:;: 0 mol~culaires
g~n~ralis~es
ne modifie pas la fonction de
lorsque Ie gaz se trouve dans
l'etat d'~quilibre. II nous semble utile de dire
a cette place, et bien que cela ne
concerne pas directement Ie sujet central de cet ner des preuves stribution
a l'appui
a lI~tat
expos~,
et sans don-
de ce qui va Hre dit que la fonction de di-
d '~quilibre peut toujours
~tre exprim~e
tion des
int~grales
premieres de mouvement. On
diverses
int~grales
intensives et par E
n
les
comme fonc-
d~signera
int~grales
par
I
m extensives.
Les unes, comme les autres restent constantes pour une particule dans llintervalle de temps La fonction de distribution iO)(q., 1
s~parant deux rencontres mol~culaires a lIetat d 'equilibre sera d~sign~e par
.*)
q.) et elle ne depend pas explicitement de temps. On peut monJ
trer qu'elle peut
~tre exprim~e
:;: exp
par la relation
L n
~Nous
A (q.,I ) E . n 1 m n
(6)
avions admis la possibilit~ d lexistence des int~grales I , bien que nous ne connaissons pratiquement aucun cas ou ces int~graTes existent r~ellement.
230
- 221 M. Lunc A sont certaines fonctions qui doivent etre d~dermin~es par la connaisn sance des valeurs moyennes des E . Remarquons que les integrales pren
mieres Im et En d~pendent, elles, de vitesses g~n~ralisees,. Si l'on a, en particulier, a faire aux molecules ponctuelles, il n'existe pas d'integrales premieres intensives, quant aux integrales extensives elles se reduisent aux sUlvantes/ on se place dans Ie cas de la m~canique classique nonrelativistej: 1. La conservation de masse ou, ce que revient au sid~rons
un gaz sans
r~actions
m~me
si nous con-
chimiQues, conservation du nombre des
molecules. Cette integrale peut toujours Hre
~crite
comme une constante,
par ex. comme unite. 2. L'integrale premiere exprimant la conservation de mol~cule.
~mc2 3.
l'~nergie
d 'une
On peut l'ecrire sous forme de la somme de l'energie cinetique
et de l'energie potentielle U(x). L'int~grale
mouvement. Cette
premiere exprimant la conservation de la quantite de int~grale
3 int~grales premieres tes de vitesse
doit
~tre consid~r~e
ind~pendantes
pouvant etre
r~duites
aux 3 composan-
c .• 1
Il est evident que la forme lineaire des sives est egalement une d'ajoutter
comme se composant de
int~grale
inh~grales
premieres exsten-
premiere extensive. On a aussi Ie droit
a chaque int~grale extensive une constante quelconque.
Ainsi la fonction de distribution
a l'~tat d 'equilibre pour un gaz com-
pose de molecules ponctuelles est
231
- 222 M. Lunc
Les coefficients A, AI' B I , B 2, et B3 sont certaines fqnctions de position qui doivent etre
d~termin~es
a llaide
des valeurs moyennes,
consid~r~es
comme connues. On montre que si Pon connatt la densit~ numerique n, la vitesse moyenne du gaz
v et la temperature T , la fonction de distri-
bution slecrit f (0)..... (e, x, t)
-" t) ] 3/2 exp [ ='lI-3/2..... n{x, t) [m/ 2kT (x,
m{ ......... c-v) 2/ 2kT]
(8)
d I~tre des foncti ons de lieu et
Les quantites n, T et v. sont supposees 1
de temps. On verra plus loin qu'elles ne sont pas du tout arbitraires. Le contraire serait dlailleurs bien etonnant:il faut, tout au moins, satisfaire les equations generales de mouvement du fluide. En fait, les conditions suffisantes vont Hre trouvees comme etant beaucoup plus strictes.
,
Etats slecartant de l'equilibre A premier abord on serait tente de mesurer l'ecart de !letat de
uef
l'equilibre par la valeur de l'expression
¥.
Pour de tres nombreu-
ses raisons qui ne seraient pas donnees
a cet
endroit, cette maniere
de
proc~der
nlest pas indiquee. Des resultats plus consistants peuvent
etre trouves lorsqu'on procede de la maniere qui sera indiquee et qui est aisement applicable aux equations de transport . Soit f la fonction de distribution vraie et soit
f{ 0) la fonction de
distribution qui serait celle de l'etat de l'equilibre correspondant aux valeurs locales des moyennes de toutes les integrales premieres extensives de mouvement. On peut toujours poser f
= f(O)
(I
+~)
ou ,l; est une certaine fonction des q.,
'f
tion de distribution
(9)
q.
et de t. Du fait que la fonc-J /0) donne les valeurs correctes de la moyenne de 1
232
- 223 M. Lunc
chaque fonction
E
J
n
',it
En f
et, par
, on a 0
JE n riO)
+P) ds4
(1
cons~quent
JEnP riO) ',it Cette derniere
~quation
peut
P sont
(11)
0
0
~tre interprN~e
tes les moyennes des produits des
L (q., t) 1
a l'~tat
dans l'eq. (12) a
Par
<
p2)(O)
consid~rer
l'expression
~o,
(12)
d'~quilibre.
~t~ appel~e
La fonction
L(q., t) apparaissant 1
par nous la mesure de
l'~cart
L'~tat
nous disons que est tres
~loign~
l'~tat
de
est voisin de
l'~quilibre
l'~quilibre
g~n~rale
des
~quations
L'~quation
lorsque
~tats.
de transport. ind~pendantes
Passons au systeme des variables
qi' Pj .
fondamentale de Boltzmann s '~crit alors sous la forme C\ f
JJ
ou
l'~quili-
lorsque L»1. Dans la
suite nous allons nous occuper du premier de ces deux Forme
de
Cette quantite ne peut E!tre nulle que lorsqu'il y a ~quilibre.
d~finition
L «1.
=
a
<)(0) d~signe la moyenne prise avec la distribution
ou Ie symbole
[1].
premieres extensives par
a l'iHat d'~quilibre.
nulles
correspondant
comme signifiant que tou-
int~grales
Ceci nous conduit tout naturellement
bre
(10)
Uf () f = "'\7" + (H f} =_e ut '()t'
(13)
H est la fonction de Hamilton et f lIes parentheses de Poisson
M et
finies pour deux fonctions quelconques
ON -aM {M, ~} =oM () Pi Oqi - () qi
IN () p.
1
233
N par (14)
d~-
- 224 M. !:'unc
Exprimons aussi la grandeur moleculaire en fonction de t, multiplions les deux cotes de l'eq. (13) par
q., p. et 1 J G et integrons dans
tout l'espace des moments conjugues. On obtient, alors, une nouvelle equation qui ne contiendra que les fonctions des variables
q. et 1
t
et
qui est
JG[ ~:
(15)
Dans la suite l'operateur moyenne (3)
< >sera pris
dans Ie sens de l'eq.
, en se servant de l'integration d'apres l'espace des moments. De-
veloppons Ie cote gauche de l'eq. (15). On aura
/
G
~f dS) P =/' G ~ tf u
P .;
d s
G
() H op.
1
P jG
df d Jq. s 1
UH Qq.
1
~ d t= up. 1
S
~
...lo
d P
s
(16)
Si lIon tient compte de la definition de Ia moyenne, alors on transcrit (16) sous la forme
Le dernier terme de ( 17) peut etre transforme par Ie theoreme de
234
- 225 M. Lunc
int~grale
GA USS-OSTROGRADSKI en une m~e
Ie long de l'hypersurface fer-
enveloppant Ie sous-espace entier des moments
dans l'espace cntier des
conjugu~s,
et p .. Si l'expression ( 0
q.
J
1
plong~
~uq.H f) tend
as-
sez rapidement vers z~ro lorsque Ie point figuratif se ra~proche de I' hypersurface d 'int~gration,
l'int~grale
Ie long de cette hypersurface
disparait. Le regroupement des autres termes nous conduit
~/o)+
n 2.. (G)-
at \'
at
n~OG>+ ~ (0 UH)+ n ~ /G OH \at Op. ()p. Uql' \' ;-'p.
~quations
() ([H, G}) j G 1
- n
Les
a l'~quation
U 1
1
=
()~
>-
f
(18)
dsp
canoniques de mouvement de Hamilton-Jacobi
nous don-
nent
UH D~signons
OH () qi
= 4i
() Pi
() G
par (
V >
!J
l'expression
De
G
.
< q.) = v. , 1
1
-'>
f
at
n
et posons
(19)
= - Pi
V. = q. 1
1
d p
(20)
V.
(21)
s
-
l'op~rateur d~riv~e
Introduisons aussi
d
dt
= at
+
1
substantielle
g~n~ralis~e
(22)
V, 1
Transcrivons maintenantl'eq. (18) en tenant compte des (21) et (22), Alors
235
~qs,
(19). (201
- 226 -
M. Lunc
n
~ (G)+ (~~
~i)(G)- n <~~}- n([H,Gl>i-
+ n
o oq.
+
ueG
(n
(23)
1
Si nous posons la forme
ctt~
G=l, Ie
g~n~ralis~e
1 dn n dt
de
droit de
l'~quation
C)vi
+ ""'\""':'= IJ
de
(23) s'annulle et lIon obtient
continuit~
:
0
(24)
qi
ce qui permet de simplifier
l'~q.
(23) . On obtient alors la forme ge-
nerale de I' ~quation de transport d 'une grandeur
mol~culaire
i.. (G) + .!.n dt
(25)
, Equations de transport des grandeurs moleculaires additivement inva.riantes. Parmi les diverses grandeurs
mol~culaires
il en existent qui jou-
issent de la propriete de se conserver dans une paire de dant leur rencontre. de mouvement et
De cette
l'~nergie
propri~t~
totale
mol~cules
jouissent la masse, la
repr~sent~e
pen-
quantit~
par la fonction de Hamil-
ton. Si la conservation additive de deux premieres de ces grandeurs ne presente aucun probleme, la conservation de la fonction de Hamilton peut,
a premiere
vue, sembler
~tonnante.
Nous avions bien
insist~
sur Ie fait que la fonction de Hamilton ne peut pas t!tre connue pendant les periodes de rencontre. L'explication est pourtant physiquement simple: pendant ces
p~riodes
lecules varient, en effet, tres
les hamiltoniens de chacune de deux moconsid~rablement
236
et de maniere difficil-
- 227 -
M. ;tunc
N~anmoins
ment calculable.
la somme de ces deux variations est ri-
goureusement nulle en raison de la loi de conservation de l'energie dans un systeme materiel
isol~ constitu~
par la paire des molecules.
L'equation de transport de chaque grandeur moleculaire additivement po~sede
invariante ne
pas de second membre. L'equation de transport
de masse nous a donn~ (24) et nous n'y reviendrons plus. , Equation de transport de l'~nergie. Si nous posons celle-ci se
r~duit
d
()
dt
H
G=H dans l'equation
aussitot
d~signat
I UH) -"ut
par
E la
de transport (25),
a 1
+ -n
Transformons Ie dernier terme de
et en
g~nerale
(26)
l'~q.
(26) en
~crivant
diff~rence
(27)
E=H-(H). Alors
(28)
est
l'intensit~
quantit~
du flux
g~n~ralis~
pourrait etre appelee
de l'energie interne du gaz. Cette
"intensit~
de flux generalise de chaleur!'
Calculons les termes successifs de l'eq. (26) 1/
2/
\)(> dtd(H)= Ot H
u + vi ()qi
<> H sera utilis~ sans changement;
/ ~)= Q(H> _(.E.!).
\0t,
()t
ut'
237
- 228 M. [June
Ie troisieme terme sera L'~quation
de transport de -
1
Cas particulier:
1
,
1
mol~cules
eompos~
1
.e.
De
l'~q.
= my.
() t
Le flux
Ri " Dans eette
1
()qi
(29)
= 0
ponetuelles. On a mainte-
+ mC.1
,,< ~m \+ U(xi,t)
potentielle de la
+ E, (30)
mol~cule.
d~duisons
\ 2m
g~n~ralis~
Ri
.l
() t
" (~ 2 + ~) =(~> 2
(31)
"0 t
2
sera
n (Cl.H)= nm + 2 (2V.C.C. J 1 J ~quation
-
2
" 2 ~ = - l/_p_)= _ () t
n
mol~eules
+ U (xi,t)
l'~nergie
(30) nous
de
2
2 m
ou U(x., t) est
1 ORi -
ponctuelles dans un champ de potentiel.
= mel'
p.
1 H=-
1
prend alors la forme
"0
Prenons un gaz q.= x.
l'~nergi.e
.) . JE >+ uqi
v·
nant
VR
1 i par - - n Oqi
remplae~
figurent une
c.c 2 )
quantit~
1
= V.p .. + Q. J lJ
tens orielle
1
Pij
(32)
dMinie par
dite tenseur de pression et Ie veeteur (34)
qui est L'~q.
llintensit~
de flux de chaleur.
(29) devient maintenat
238
- 229 M. I:.unc
() [ mv V· 1 JXi 2
2
mv
+
2
+ 3kT ) +
2
2
A _v_ ( + Q) = i.. (.!!!Y.2 + 3kT) + v dU _ + 1 'Ql' () i dt 2 n JXi vl ij 2 i () xi n Q xi
.!.
+
+
(35)
n ;
Equation de transport de Revenons
a l'~quation
quantit~
de mouvement
g~n~rale
(25) et posons pour pour
j= 1, 2, 3 j
>3
En raison de la conservation additive de G, Ie
(25)
est nul. Les termes successifs de (25)
1/
d dt
G
2/
() -;;,1 aq.
3/
JG
=
0
(n
)
cOt~
droit de
l'~q.
sont les suivants:
= n1
1
(Tt >=
()v. - m
'J/
pour j=l, 2, 3 et = 0 pour j
4/([H, GJ)=
>3 ;
j=l, 2, 3; i=l, 2, .. , s;
Si, maintenant, on fait usage des ~qs. (19) Ie dernier terme devient
({H, G}) = Mais il est bien
my.
1
~vident
o_v._J o qi que Ie dernier terme de cette expression est
239
- 230 M. -IJunc
~gal,
selon la seconde loi de
l~cule
que nous
a la
m~canique,
force agissant sur la mo-
d~signerons
par mF. . Rassemblant J termes nous obtenons l'~quation de mouvement dv. J dt
alors tous ces
1 OPij = F. J
-Y
qui ne differe pas par sa forme des d~riv~e
(36)
o~ ~quations
de mouvement habituelles
p .. dont Ie sens ne differe du tenlJ et j non sup~rieurs a 3. seur habituel de pression que pour
mais qui contient la
Dans Ie cas
g~n~ral
de
nous avons
= OPij
+
'Ox.
k
> 3 ; i, j = I, 2, 3.
(37)
1
Ainsi
l'~quation
de mouvement prise dans Ie cas Ie plus
contenir des termes
compl~mentaires
qui ressemblent aux
composantes du tenseur de pression et qui peuvent cas ou il existe des quantit~s
Vk
corr~lations
/k>3/ et Ci
Forme dMinitive de
g~n~ral
diff~rer
peut
d~riv~es
z~ro
de
des au
non-nulles entre les composantes des
/i=l,2,3/.
l'~quation
de transport de
l'~nergie
uP ..
__ lJ calcul~ d'apres (36) dans l'~q. (35) on OXi obtient successivement, apres avoir divis~ l'~q. (35) par m, Substituons
d dt
2
(!... + 2
3kT 2m dv.
+
.!. oU
m
()Xi
vi
+
1
OQi
9
'Ox. 1
P ..
+
.!.
avJ.
J> PWbx'1
av .
+
1 OU) = + vi (F.1 - -dt) = - - + - - +-2!. _J + v. (F. + -1 Jl OXi 2m dt .P 'Ox1. 1 1 m OXi 3k dT
1 OQi
= O.
240
(38)
- 231 M. Lunc
Or Ie dernier terme est nul puisque lion a dit que les
mol~cules
vent dans un champ de potentiel. De cette maniere on obtient de transport d '~nergie pour un gaz dT 2m dt
3k
1
oQi
+ - ax. p 1
nit~
y
de
mol~cules
a
v. p.. IJ_-J Ox.
l'~quation
monoatomiques
(3~
0
1
Le sens physique de cette
~quation
r~pr~sente
Le premier terme
1
+
compos~
se meu-
est fort simple.
la variation de
l'~nergie
interne d'une u-
de masse de notre gaz. Le second terme-c'est l'apport de
l'~ner
gie par conduction thermique. Le troisieme terme de (39) donne la transformation en
~nergie
interne du travail
effectu~
sur Ie gaz par les
forces dissipatives. 11 n'y a aucune mol~cules
difficult~
a
g~n~raliser
(39) aux gaz composEls des
plus complexes. 11 faut alors prendre a la place de (30) une
expression
appropri~e
pour
Transport des grandeurs
H et utiliser
mol~culaires
dans
l'~q.
(36) complete.
l'~tat d'~quilibre.
L '~tat d '~quilibre est, comme on l'a deja dit, 11 faut, en cons~quent,
m~me
temps, satisfaire aussi
l'~quation
d~fini
par
l'~q.
(5) ..
de Boltzmann et, par
il faut qu'en meme temps
;Vf =
0
(40)
Or cette derniere equation fournit une fort large classe de solutions qui ne sont pas toujours compatibles avec (5). Nous savons, par ailleurs, que la solution forme (6), et se reduit pour
mol~cules
g~n~rale
de (5) est de
ponctuelles a la distribution
maxwellienne (8). Nous allons nous limiter ici au cas de ponctuelles et, par
cons~quent,
done agir l'operateur
J)
mol~cules
a la distribution maxwellienne. Faisons
sur f(O).
241
- 232 -
M, t-unc
On obtient ...."
fDf(O) =(_(/ () t
+ c. ~ + """:\
1
'?II)
F, ::;;-)
uX.
(
f 0
) ::I
1.) C. 1
1
En raison de la forme particuliere de la fonction de distribution nous aurons ici
a.
partir de
l'~q.
(36)
(42)
Puisque. d'autre part, dans 3k dT 2m dt
+ E.
'()v i
p 0\
l'~q,
=_
K
m
(39) on aura
(l1P __ 3 _
.p
dt
2 T
= --mk -dtd ( In )'0 T -3/2 )
Q.= 0 celle-ci conduit 1 dT) =
dt
= 0
(43)
Nous voyons que Ie long de la trajectoire moyenne Ie produit'p T- 3/ 2 reste constant. (42)
et
(43)
L'~volution
du gaz est donc adiabatique. Substituons
dans l'~q. (41). On obtient
242
-. 233 M. Lunc
~f
(0)
() [ ( ) =fO Ci U· x .
-3/2 (lnyT
aj V
mCiCj
1
1
L
~X.
(lnT- 5/ 2) +
u 1
~ kT
+
a
2kT
C)
Vj dT + C. C.(~ -- + 2TdT .. 1
J
i
~
VXi
+ ~ C C2 OT ~quation
(
1
= f(O) l·-C.
Cette
dT
-lnpT)+~(~+2TdTOij)
~.
}= °
(44)
1
doit etre satisfaite quelque soit la valeur de C .. Il en 1
r~sulte
que les coefficients de C., de C.C. et de C.C 2 doivent etre nuls. 1 1 J 1 Le premier et Ie dernier de ces coefficients seraient nuls lorsque Ia temperature est uniforme. On a donc
(45)
T = T(t) . Ie deuxieme coefficient est nul si
(46)
et
o
v. () x.l = 0, lorsque i J
f
(47)
On doit donc avoir v
i
d -1/2 = - (In T )x dt i
(48)
Le seul mouvement permis est donc une expansion ou compression adiabatiques. 11 convient ici d1attirer llattention sur des conclusions
~
trangement similaires/ mais non-identiques/ obtenues assez r~cemment par A. A. NIKOLSKI
(z1.
Il a montr~, notamment, la possibilit~ de
distribution maxwellienne pour un mouvement
243
d~crit
par
ll~quation
Ia
.• 234 -
M. Lunc X.
v
qui
r~pr~sente
1
=-+
i
une dilatation ou contraction uniformes de l'Univers. Cel-
le·ci correspond donc
a la loi
T (t) =at
et p(t) L'~quation L'~q.
(49)
t
=
particuli~re
:
;t2
(50)
b(x) t+ 3.
(51) occup~
(50) est valable dans l'espace entier
par Ie gaz.
(51.), par contre, est vaiable Ie long de chaque trajectoire, seu-
lement. Revenons au cas plus d~quations ~quations
l'~quilibre d~crit
de
par Ie
syst~me
(48). De deux premi~res de ces
et
(43), (45), (46), (47)
nous tirons
?Jp
i~x.
1
avec
~quation
Uy
+ v.
put Cette
g~n~ral
1
=
°.
(52)
48. nous donne
aT
'Oy
-'-
y"0t - 2T'\)t
11 est certai qu'
-
()T 2 Tot
-3
(x
up
+3)
i F7Jx i
a cM~ de solutions
=0
(50) et (51)
(53),
peuvent exister
de nombreuses autres et ce sujet, nous semble, est digne de recherches futures. Les
~tats
voisins de
l'~quilibre.
Reprenons pour les
mol~cules
_ (0)
f (c.,x.,t) - f lJ
ou
i
l ) est
conditions
d~crite
par (8)
d'~quilibre donn~es
ponctuelles
l'~q.
(9)
,-r
-:
(c.,x.,t): 1 +Q(c.,x.,t), lJ
..
~lJ-
mais ne remplit pas, par les
244
~qs.
(43)
a
n~c~ssairement, (48)
les
. Par contre ,
- 235 M. Lunc
nous allons exiger de cette fonction f(O) exactes de
n, v. et 1
T
queUe nous donne les valeurs
en chaque point et
a chaque
instant. De la
fonction ~ nous exigeons donc que • (0)
_ (0)
\ l' )
=" Ci
En plus de ca, nous voulons que la
<
j) >
,2' (0) = ,C ~ = 0
>
(55)
in~galit~
1
soit satisfaite.
11 est bien
~vident
par la relation (41)
qu'exprim~
que ,Q)f(O) ne sera plus nul, bien
(sans, bien entendu, quleUe soit ~gal~e au z~ro) .
On aura, apres avoir tenu compte de (36)
~f(O) =f (0)[t ..!. (InpT- 3/ 2 ) + C. l2 (In OT- 3/ 2) _! ~Pij J + dt Jx. r p ;:lx, r
1
J
1
mCiCj () Vj
+ - - (-kT
+
2 _1
~x.
2T
1
dt
mCiC
'-' ) + - ij
2kT
l,1)p) ~quations
d/:
=
0.
(56)
1
(57)
de conservation additive exigent que
j
:JJr d3 -e • 0
I
Ci:{)f d 3
.Jc1Jr Puisque
1 '0T} Tux.
imm~diatement
On trouve
Les
dT (
-
d3
C·
(58)
= 0,
(59)
C = o.
(60)
sous forme
f
(61)
245
- 236 M. !'unc
Les eqs. (59)
f
c. 'J) iO)d
)1
,(60)
s'ecriront
!
C+/ C. ~Q)f(O) d3 C+ C.~~ f(O) 3) 1 )1
......
d C
3
=0
(62)
(63) L'integration montre que lion a 1
/
'7'1 (0)
mC . .vf 1
'0
--> _
("
d 3 C - - ::;:- (p .. - o.. p) U X. 1J 1J J
(64)
et - P-
d dt
(In9 T
-3/2
).
(65)
Il est bien evident que ces expression s 'annullent lorsque Ie gaz est en etat d 'equilibre. Ainsi les equations (59) et (60) s'ecriront
~
C=.L. ! . 'ox. JmC.[J,~f(O)+(:;)J.)f(O)Jd 1
3
'f
(p .. -
1J
~.1J
p)
(66)
J
J
m;2[q;.l) flO) + (21
~) flO)] d/'
P
:t
(In y T- 3/ 2 ) ,
(67)
Pour toute autre grandeur moleculaire il faut, evidemment, prendre en consideration l'equation complMe de transport. Choix de la fonction
qi
Notations
p
Pour choisir une fonction mise au point par H. GRAD
r31 L.
_
appropriee on utilisera. ici la methode
et qui fait usage de polyne,mes orthogo-
naux semlables aux polyne,mes d'Hermite (et que lion designera ici
246
- 237 M. £unc
par Ie nom de polynemes de Grad). La dMinition de ces polynemes sera
donn~e
plus bas. Mais tout dlabord on introduira ici une notation
approprit'!e, bast'!e essentiellement sur la notation de Grad, avec de petites modifications de dt'!tail. On se bornera dans la suite aux seules coordonnt'!es cartt'!siennes et lion utilisera les notations tensorielles habituelles. 1. Un tenseur de n-me ordre sera dt'!signt'! par un indice inft'!rieur pris entre parentheses indiquant son ordre, suivi par les indices (ou par un indice) relatifs aux diverses composantes. On t'!crira donc
T(n) = T(),
, , = T(n)..:' n 11' 12"" In
2. L 'ensemble des indices a
(68) ~t~ d~signt'!
dans la formule prt'!ce-
dente, et sera aussi dt'!signt'! dans la suite, par la lettre sousignees. Notons donc (69)
pour un ensemble de
a lettre
k indices
pilote
3. Le tenseur, originellement de llordre contracte s
n+2s
et qui a t'!tt'!
fois sera dt'!signt'! par un indice supt'!rieur, indiquant son
ordre primitif et par llindice inft'!rieur, comme prt'!cedemment , pour son ordre apres contraction. Ainsi:
4. On introduira la vitesse sans dimension
c,
U=~ i 2kT
Vrn
c =
(71 )
n 247
- 238 M. !.une
5. On d~signera par
n
form~
g(n)!. Ie tenseur
par Ie produit de
compos antes de la vitesse sans dimension . Cette grandeur
laire est done
d~finie
mol~cu
par (72)
6. On introduira deux
a) La finie par les
~quations
esp~ees
de tenseurs unitaires
premi~re esp~ce
de tenseurs unitaires sera de-
sueeessives (73)
(74)
D'une
mani~re g~nerale
2s,
design~ par
de s 0(2)
Ie tenseur unitaire de
0(2s)i
sera
premi~re esp~ee
d'ordre
form~ par la somme de produits
dont les indices ne se rep~tent pas et qui, ensemble, con-
stituent !. tout entier . Comme on voit sur l'exemple fourni par (74) les r~petitions des
6
ne sont pas prises en eonsid~ration et Ie
6(2s)
nombre de termes pour
l'~q.
est done ~gal
a
1..3.5 ••• (2s-1) = (2s-1) II
b) La seeonde m~e
par la
r~gle
esp~ee
de tenseurs unitaires sera for-
que lIon peut aisement diviner
a partir
de l'exemple
suivant
+
6
+... +6
+...
i i
1m
On voit done que
d (m)(2)
est
+0 i
i
(75)
m-1 m
eonstitu~ par la somme de tous les
248
- 239 -
M. t.unc
,
o(2)
diff~rents form~s
a partir de l'ensemble i contenant m
~lements.
'":\(m)
Lo tenseur suivant
(.) (4)
sera
etc.
7. Une troisieme espece de tenseurs, un peu form~e
lorsqu 'on dispose de
lIon prend la
moiti~
2 ensembles des indices
Cette expression contient
J.. et
n!
J..
que
(76) termes
diff~rents.
des polyntlmes de GRAD
Les polyntlmes de GRAD seront dMinis par H(
et
pard (2n).!.,
Propri~t~s
i
sera
des indices dans chacun de deux ensembles. Cette
design~e
espe.ce sera
diff~rente,
.
n)~
l'~quation:
=
(77).
Les produits tels que (n) O(2S).!. g(n-2s).!. sont form~s de
n
~l~ments
dont
ments qui restent les
6:
a la maniere des
diff~rents
2s
a la
sont
grandeur
produits ainsi
Les polyntlmes
on prend l'ensemble .!. compos~
attribu~s g(
n~
au
2)" s
1
()
~~~).!.
et les n-2s
~le
apres quoi on ajoutte tous
constitu~s.
de Grad
cons~cutifs
seraient donc: (78)
249
240 -
~
M.Lunc ,-.
= f2
H(1)i
Le polyntlme peut prouver
H(2)~ =
2g(2)~ - 6'(2)~
H
2 3/2
H(n)~
(3)~
est
ais~ment
=
g(3)~
(2, Dans cette i
et par
~quation l'~lement
(2)~
U,J H( n,:, )'
sup~rier
= H( n+ 1)",:,+J
Ie symbole nouveau
r~gle
~
+ j
de
=
g(1)~
par rapport
qu'iI existe une
de former des polyntlmes d'ordre
(80)
_ 2J/ l a(3)
symm~trique
(79)
1/2 U" 1
=
g(1)i
a tous ses indices. On
r~currence
par la
permettant
r~gle
(82)
+0(2)'':" J,H( n-,:, 1)' d~signe
I 'ensemble
form~
par
j.
Les polyntlmes de Grad forment, comme iI l'a
d~montr~,
un
en-
semble complet de fonctions orthogonales dans I 'espace entier des U, et avec une fonction de poids
exp (- U2 ). On
d~montre
1
que I 'on a une
relation fondamentale de normalisation et de I 'orthogonalit~
1'(
-3/1
(83)
H(n )'1
et une autre qui enjd~COUle :
2
~
H(m)~ H(n)I. exp(-U ) d3 U = 0 ; si
mFn
ou si ~
F
I., ou si les deux in~galMs se trouvent simultan~
ment remplies. Notons que les deux ensembles d~r~s
comme
diff~rents
lorsqu'au moins un
250
~
~lement
et I. seront conside l'un des ens em-
- 241 -
M, Lunc
bles ne se trouve pas dans l'autre. Les ensemble que l'on obtient par un changement quelconque de l'ordre des tre, consideres comme
~gaux,
~lements
seraient, par con-
Une fois de plus, notons que Ie cMe
droit de l'equation (84) contient, en principe, n! termes, dont certains, ou meme tous, peuvent
~tre
nuls.
Le developpement de la fonction ~ en une serie des polyntlmes
de
GRAD. Puis que les polynomes de GRAD forment un ensemble complet des fonctions, nous pouvons representer chaque "bonne" fonction par Ie
de-
veloppement en serie 0<>
~(U"x"t) 1 J
1 [
;;:!
= \
L
~ ~(2n) A(n)~ (xj' t) H(n)~ (U )J(O)
(84)
YI.: 0
Dans ce developpement les rtlles de la variable
U, et des variables 1
x, et t se trouvent nettement separes. J Si la fonction
A
,par une formule d'inversion
(n)~
qui resulte de l'eq. (83). On trouve: 3 2 A( )' -_11- / n~
f
J
H( )'
n~
(La propriete de symmetrie de doit
~tre
1exp (-U2 ) d U. ~
(85)
3
H(n)~
par rapport
a tous
se indices
exploitee pour la demonstration de la relation fondamentale
(85) . crest gr~ce dans l'eq. (84)
a cette
propriete que Ie coefficient
l/n!
figurant
disparatt lors de la contraction du cMe droit avec
() (2)' , • Les equations de normalisation (55) nous conduisent n
~,I.
maintenant aux equations consecutives
251
- 242 M. l,unc
r
A =1(3/2 i (0)
(86)
I
.J
(-
_ -3/2 A(l)i- lr La troisieme l'~nergie
.,. 2 ~ H(1)d exp (-U)d 3 U=0
i
J
~quation
de normalisation , exprimant la conservation de
r~duit
se
(87)
a
22->/ U exp (- U ) d3U = 0
T
.J
et puisque d'apres (80) lion a 2 U =U.U. = 1 1
j'-
(2) = l'B: (2) g(O)!, 2· (0)
+ r(2)J= !r1+H(2) 1 0(0).:, 2 t:
(88)
(O)!}
on obtient aussittlt l1
(2)
2-""
T
(89)
+ H(O)!,J exp ( - U ) d 3U = 0 .
Si lion tient compte de l'eq. (86), on trouve: A (2) (O)!,
=
0
(90)
Nous avons, par ailleurs,
rt/
=!
U.
P.. mC.C.(1+d)f d 3C = 11- 3/ 20 u.U (1+,h) exp (-U 2)d 3 1J) 1 J • )- ./ 1 j ! Puisque d'apres (80) on
a
1 g(2)!," 2"(1+H(2)!, ) , on trouve
(91)
et, finalement : p .. - p ()..
1J - 1J A (2)i j p Le coefficient successif A(3)!, peut
(92) ~tre
trouve par un procede a-
nalogue qui fait intervenir les moments du 3-me ordre
252
g(3)!..
- 243 M. Lunc
On a
(93) L'intensit~
li~e
de flux de chaleur est
au coMficient
contract~
de
troisieme ordre. En effet
Q.
1
2) =1.2 Jon 3 /U.U \ 1
Dlautre part (3) ( (3) )_ A (3) ( (3) \+ A (3) ,. (3) A (I)!. g (I)!. (1) ijj g (l)ikW (l)jij~ kik
>+
+ A~~~jji (g (tl\kki) =3A~~~i(g~~~i> En omettant les
d~tails
de calcul faisant
~tat
r~petition
de la
taines composantes et en faisant remarquer que Ie polyntlme
(94) de cer-
contract~
(3)
de GRAD
H (1) i s I exprime par la formule
= 23/2 U (U 2 - 5/2) (95) (l)i i _ lion obtient finalement pour Ie d~veloppement de arrM~ aux termes H(3)
T
contract~s
du troisieme ordre l'expression
H(2)ij
p .. - p
y2 + 5
6'
=l.J _ _ _..... lJ"UU+ i j
Qi H (3) = (1) i
pQ
4
Q.
'5
_1
p
PQ
2
U.(U _ 5/2) 1
(96)
, Equations de transport de la quantit~ de mouvement et de ll~nergie
Le probleme qui est certainement essentiel (et qui, soit dit ~ntre parentheses, semble avoir une lorsque on ment de
arr~te,
p a un
comme
r~ponse
cl~tait
n~gative)
fait dans
ll~q.
certain ordte de moment (les 13
253
est de savoir si, (96), Ie developpepremiers moments
- 244 M. ;t:.unc
l'~quation pr~c~dente) n~glige-t-on
dans
aux termes pris en
consid~ration,
des termes petits par rapport
ou laisse-t-on en dehors des termes
importants? Bien que, comme nous l'avions ses, tout porte
a croire
entre parenthe-
que des termes essentiels aient t~chons
soyons ici illogiquement optimistes et ti du d~veloppement de ~ exprim~ par
H~~~.
remarqu~
~M n~glig~s,
de tirer Ie meilleur par-
(96) , coup~ aux termes en
Une maniere simple d'ecrire les equations de transport consiste
dans l'usage des polynOmes primer {j) f(O)
a l'aide
de GRAD. A cet effet il serait utile d'ex-
de ces polynOmes . On aura d'apres l'eq. (56)
et en tenant compte des
~qs.
(78), (79), (80) et (95) 1
(1(0) .... 1r- 3/ 2 I 2 .... [_.- / d r ruf d 3C = (~ m) exp(-U ) d 3U 2 2( n p) OX. (PifP 0ij)H(l,/ J
+
~ J. ) H
(ovi + ox. 2Tdt J
ij
+ 2- 3/2 n aT H(3) T~x.
(2)ij
Lorsqu'on multiplie (97) par
f exp:rim~
(97)
(l)i
a l'aide
de l'~q. (96) et lor-
squ'on integre Ie produit ainsi obtenu on verra disparaitre tous les termes qui contiennent des polynOmes de Grad d'ordre
1(
diff~rent
(83) nous donne, apres quelques courtes transformations, Ie
( ID f
e)
-" OVi p .. )~ d C = (0/ m) ( lJ I
3
L'~quation
de
j Ce
continuit~ (61)
~}
f(O):Dfd C 3
VXj
J
= - ~~f
(0)
.
m~me r~sultat
d
-Pb.. P
lJ
nous donne
+ Q.
1
la fonction sous Ie signe de
).
(98)
av. P.. -PO..
C = -(9/ m ) (~ 3 ux.
~tre
TOXi
r~sultat
imm~diatement
J
peut
OT
et l'eq.
obtenu par une
l'int~grale.
254
lJ P
aT lJ +Q~) (99) Tux.
int~gration
1
directe de
Les equations de la conserva-
- 245 M. :tunc
quantit~
tion de la
l'~nergie
de mouvement et de
int~gration
(66) et (67), respectivement se retrouvent par
n 'apportent aucune information
compl~mentaire
qui s'expriment par directe et
concernant Ie mouvement
de notre fluide . Equations de transport
mol~culaires
des grandeurs
du type (100)
Ce
probH~me
intimement tance
li~
vue, peut parattre fort abstrait est
au transport de l'entropie du gaz et possede une impor-
consid~rable.
gaz est
a premiere
qui,
a.
~gale,
unit~
On sait que l'entropie d'une
un coefficient de
I f logf
d3
I
proportionalit~
de volume de
pres
a.
lIint~grale
C = n \logf~
(101)
Or nous avons In f = In f tol + In(1+p) = In[11- 3/2 et, par
cons~quent,
n,lnf)=
(p I m)~_,-
on a
-3+ln
(1'1'
Si nous admettons que certaines conditions dans
l'~q.
I
+)
, In( 1+ ! .
-31 2 f/m) L
(cf
suppl~mentaires,
(103) en une -;
9l m )n -3)+ In(1+ P) f.[
i->'
-3InfJ+jln(1+ 11fd 3C.
~q.
(12)
s~rie
(103)
, nous pouvons, sous
d~velopper
I 'int~grale qui figure
f d C
00
J
= >---
+1 I ( _l)n,l
n=l
i
-
?
n
-~
f d C
3'
Prenons pour la valeur de G les deux premier termes du pement en
(102)
s~rie d'int~grales
.'-
3
-~
_U 2
(104) et posons, donc,
255
(104) d~velop
- 246 -
M. !:.unc
G =
tr~s
La raison de ce choix la suite. Ecrivons laire. Le
2
.r
~~ -
particulier va apparaitre rapidement dans
l'~quation
cM~ diff~rentiel
(105)
2
de transport pour cette grandeur
de
l'~quation
mol~cu
de transport sera maintenant :
group~
Nous avons, comme on l'aperpoit rapidement,
dans
1'~q.(106)les
termes d'apr~s l'ordre croissant de la fonction ~ . Cette raison est, bien entendu, de
l'~cart
par les
motiv~e
consid~rations
L , bien que la
l'~quilibre
avec la croissance de llordre en d'autres conditions doivent
~tre
1
concernant la mesure de
d~croissance
des
ne soit pas tout
satisfaites
int~grales
a fait certaine et
simultan~ment
Le premier terme de l'expression (106) vient d'Hre
avec calcul~
L<1. et
s'exprime par (98) . Nous allons calculer les termes du second ordre et nous allons soient
n~gliger
inf~rieurs
aux
les termes d'ordre 3, en souhaitant qu'ils
pr~cedents,
ce qui, en principe, devrait Hre
vraiment Ie cas : tout au moins pour les sure de
petites valeurs de la me-
l'~cart d'~quilibre.
Le calcul effectif des termes de assez
tr~s
consid~rable
et
donnant directement Ie v~rification
deuxi~me
d~pourvu dlint~ret
r~sultat
ordre exige un travail
. Nous l'omettrons ici, en
final et en engageant les auditeurs
individuelle. On a
256
a la
- 247 -
M. !'unc
k/(:
2 ',)
<)+f(J)~f)
,- d ~ dt
2..3.2 ( 3
= (i/m)! .I
-, d3 C =
p .. P .. 2 + .2L..Jl ) + - J
+
2 5
ux.J
5
4p2
5p
(p, .-po .. ) Q. ~
J
lJ
p
Qi{Pij
1
2
-P~j)
2 P
1
"Jp
eo .
+
(107)
X . ..J
J
n convient de noter, en passant, une relation fort utile p .. p ..
.2L.E p
2
= p
Pour calculer Ie
ct>t~
2
droit de I '~quation de transport on doit con-
naitre la loi de l'interaction entre les
mol~cules
pendant leurs rencon-
tres. Toutefois certaines conclusions peuvent Mre consid~rations
tres
g~n~rales,
trouv~es
car ces lois d'interaction apparaissent ,
en fin de compte, sous la forme de coMficients purement qui ne
d~pendent
par des
pas des conditions des
~coulements
num~riques
particuliers.
Nous allons introuire un certain nombre de notations usuelles. Les vitesses de deux ient
d~sign~es
mol~cules
-,
par
Les vitesses des
a un instant
pr~c~dant
leur rencontre sera-
-,*
C et C respectivement, or par
m~mes
mol~cules
-"
U et
--\ 'ItU
apres la rencontre seront ~t
..),.l4-,
d~sign~es
_'I
par les m~mes lettres accentu~es, c. ad . par C et C ou par TJ ._, ... et U . Toutes les fonctions qui d~pendent de la vitesse mol~culaire seraient
d~sign~es
de la
E les parametres
m~me
maniere. On
g~om~triques
d~signera
par b
et par
de la rencotre {b-distance assymp257
- 248 M, Lunc
E -angle
totiquedes trajectoires mol~culaires, du plan de
rM~rence),
On d~signera par mol~culaires
deurs
c,a
du plan de mouvement et
(G]
la variation totale subie par les deux gran-
attacMes aux deux
mol~cules
pendant la rencontre,
d,
[G1
=
(GI
+
+ G'I()
GlI-l) - (G
(109)
On aura avec ces notations
()eG
J
n<~) = ut
G
Qe f ":"
~ ut
~I·
. 1 1< "t . .-, --'·r d C = -- f.:.1iKf'f*"' -f f ) \C-C Ib db cIt d CdC
3
I
-,
4j.J
3
3
(110)
(Cette forme du terme des rencontres peur Mre ouvrage de la tMorie G est
donn~e
cin~tique
par la formule
trouv~e
dans chaque
des gaz , par ex, dans [lJ)
Lorsque
(105) on aura (111)
On a d lautre part
flf~1 - ff~ = f( 0) i 0)"1- [( 1+ ~ I ) (1+ ~ It-I
<{ J + (~I ~*I
) -
-
~
(1+
-
i
(1+
f) }
4"')
La fonction qui se trouve sous Ie signe de (en y incorporant le coMficient de
1)
(112) llint~grale
dans (110)
se trouvant devant) est
-~ f(o) /o)~ Id-c~' b [ [ ~ 2] + [~1 (f~j,' - ~ ~i - ~(~] [p2J + (113)
Cette expression contient des termes du second,
tri~me ordre en
troisi~me
et qua-
4 ' Seuls les termes du quatri~me ordre seront n~gli258
- 249 -
M.l:..unc
g~s
dans nos calculs. Pour rendre ceux-ci plus simples posons encore, abr~ger
pour
(114)
4 B. = -5 (Q./pS1) . 1
La fonction
;f
S I~crira
-
...: =
alors sous la forme
p .. U.U. IJIJ
On choisira un nouveau
t S1
figurer la vitesse
(96 1)
11
des
coordonn~es
dans lequel va
...).~
ZS1 et
Ie par rapport ;\ leur centre de
-
zn
les vitesses de chaque
gravit~
mol~cu-
avant leur rencontre . On aura
~
Z
et
.
U.
= gi + Zi '
..... U.
= gi
Z.
(116)
U'
= gi + Z!1 ,
U<:·
= gi • Z! 1
(117)
1
i
1
1
Par suite de la conservation de
1
l'~nergie cin~tique
pendant Ie choc
Z ne subissent pas de changement
les modules des vitesses relatives et, par
B.U. (U -5/2).
syst~me
...).
-''I- = Z
2
+
du centre de gravit~ de la paire des mol~cu,.
les. On appellera aussi
alors
(115)
1
cons~quent,
(118) .,.l.
Le jacobien de la transformation des variables variables dlint~gration
I
D(U,
t
et
>'\'
U, U en nouvelles
.-\
Zest
ul = 8
(119)
D(g, Z') ~
259
~
250 M, !:'unc
Un calcul fort simple montre que I 'on a (120)
f .. (Z! Z! -Z.Z.) lJ 1 J 1 J
avec
+ 2P ..
f .. = 4 B.g.
lJ
On
d~signera
par
1J
M2 et
(121)
,
lJ
M3 respectivement les termes que lion ob-
tient par I 'int~gration des expressions d' ordre deux et d'ordre trois en
y.
On a donc M 2
~
-
l~)~ !C--'.C'tl
.!.jr/r"i: 4
:.-2 f(:) ' .. _" Ii: -_
b db d d C.
! ,
Cd
3
3
c'~.
(122)
;:,
Par suite de relations (119) . (120) et (121) on aura M2 = _(1/4,],;3) n 2n
(i
exp { _(U 2+U-lt2
)1[f] 21 U. U*\b db O,t.d 3Ud3u*=
i
= -(4/'l'r 3)n 2n :":, exp I:: • 1/: ;
11 .2(g2+ Z 2)}!r..(Z! Z! '
- J_J
llJ
1
J
g
_Z.Z.) 12b db d[d 3Z d 3 (123) 1 J i .'
(Pour ~viter la confusion on employera Ie symbole[] uniquement dans le sens de
l'op~rateut
.. ordinaire par celles L'int~gration
(112) , en remplacant les parentMses dans Ie sens de I 'ordre
sup~rieur)
.
de l'expression (123) par rapport
aE ,
~
~-~
get
Z se
fera selon la m~thode bien connue, en faisantintroduire l'angle -"
entre la direction des vecteurs
Z et
--"
Z', A cet effet on peut
der de la maniere suivante. Prenant Ie vecteur des
coordonn~es
on
~crira
X. proc~-
Z pour l'un des axes
: ~
Z=Z(1,0,0)
260
(124)
- 251 M. i:.unc
et
zi =
Z(cos'X. ,sin'X
L 'int~gration par rapport
a c se
COSt ,sin};.sine,)
fait entre les limites de
211 et fait disparattre tous les termes de en cos t res. Le
degr~
l'int~gration
final de
0
a fait
L'int~gration par rapport
l'integration par rapport rapport
a la
aE et g ...,
a
Z
a
ou
~lementai,..
se presentera comme suit
15
-12 2 M2=-(2/'tr) 2n n (B /3 + P .. P .. /5) 1J J1
a
impair en sine
. Les calculs sont assez longs mais tout r~sultat
(125)
2 ~,,.: <~ x exp (-x )d3Xfsin~b db .
( 126)
y ont ete effectu~s , par c'ontre ~t~
remplacee par
l'int~gration
par
nouvelle variable d'integration ~
1:= (2pour des raisons de
l'abbr~viation
Z
d'ecriture.
L 'expression (126) contient une integrale double ou figure Ie sinus de l'angle
X. .
Cette
quantit~
depend de
b
et de
x 2 . Elle peut tltre
consideree comme connue lorsque la loi de l'interacticin est connue explicitement. Alors l'integrale double ,(>:"
K5 ' (2 / '1f)Jx 5 e _.2 d/ / ' in2X b db • (32111,X 7e _x2 dx!Sinhdb se trouvera complMement
Jd~terll1in~e
quelque
'~OOit
(127)
le mouvement du
gaz . Cette quantite possede la dimension de la surface et l'on peut consid~rer
cette
int~grale
comme la valeur de la section efficace
de
rencontres moleculaires. K5 est une constante physique pour chaque espece moleculaire. Nous verrons que les termes d'ordre superieur feront apparatter de nouvelles constantes de la mtlme espece.
261
.. 252 -
M. Lunc
Les termes du troisieme ordre apparaissant dans 1'expression (113) sont, ,
M3 •
i j~j [~] [Ir u: H~ 1YQ r!~)~-
• (l/a 3)
n2nJJ~exp \ -
C'lb db dE d i d3c't
+v"2I}[<ji 12(H':..~'+ f\\U-U'l'bdb dEd3Ud31J~
(U 2
(128)
La tres grande simplification intervenue dans la fonction df1e
a la
a int~grer
est
forme tres particuliere de (105) que nous avions choisie
pour la grandeur nous conduit au
mol~culaire.
r~sultat
Le cal cuI long mais toujours tres banal
suivant ;
ou
Tous les coMficients
K sont des
s
con~tantes
physiques
r~pr~sentant
diverses sections efficaces de rencontre. 11 convient de noter que 1'ordre de grandeur de K
s
ne
d~pend
pas beaucoup de
toute vraisemblance, doit Mre celui de la section de la
s
et, selon
mol~cule.
Pour
les gaz tels que O2, N2 etc il doit Mre de l' ordre de grandeur de 10 .16 cm 2 . Equation compl~mentaire de mouvement. Nous pouvons maintenant
~crire l'~quation
262
de transport pour notre gran-
- 253 M. !:.unc
deur
mol~culaire
G.
A cet effet il faut que nous assemblions les et qui sont
repr~sent~s
dMinition (130). Nous
par les
~crirons
~qs.
r~sultats d~ja
obtenus
(98). (107). (126) et (129) et par la
tout d'abord Ie
cet~
droit de
l'~quation
de transport :
~ e
n
<
~ Ji 2
(0-2' ) ) at
+( Ceci
~tant;
transport de
2 ~ B2 I 1 =-n rltK5{T+'5Pitji)+'5K5PjkPktij+
:~
K5-
~~ K7 + ;0 K9) BiBjP ij 1
nous pouvons
~crire l'~quation
(131)
compMte abbregee de
G. On a :
(p .. -pd.) Q. } lJ
lJ
P
1
2
=
(132)
Nous avons ainsi obtenu une nouvelle l'~volution
du gaz et qui doit remplacer
263
~quation gen~rale
~ ~quation
decrivant
empirique dans Ie
- 254 -
M. Lunc
systeme des
~quations
habituelles. On pourrait, par ex. rennoncer
aux relations empiriques qui relient les dans les de
~quations
diff~rents
de Navier-Stokes .
2
L'~q.
ordre en Q. et p ... 11 serait 1
co~fficients
de viscosite
(132) contient des termes ~videmment indiqu~
lJ
de voir
ce que lIon obtient lorsque 1'on conserve dans (132) les termes les plus grands et ce que les termes dlordre connaissance du mouvement
r~el.
264
sup~rieur
apportent
a notre
BIBLIOGRAPHIE
1
-
M. LUNC
2
-
A: A. NIKOLSKI
3
-
H. GRAD
Rarefied Gas Dynamics (third Symposium) v. 1 ; p. 95 Accademic Press, 1963 Engineering Journal - Acad. De Sciences USSR, 1964 CPAM, 2,1949, pp 331-407
References generales
1. - S. CHAPMAN and 1. COWLING
The Mathematical Theory of non uniform Gases - Camhrige University Press 1952
2. - H. GRAD
Principles of the Kinetic Theory of Gases - Encyclopedie of Physics, vol XII - Springer Verlag 1958
266
CENTRO INTERNAZIONALE MATEMA TICO ESTIVO (C.I.M.E.)
I. ESTERMANN
1. APPLICATIONS OF MOLECULAR BEAMS TO PROBLEMS IN RAREFIED GAS DYNAMICS 2. EXPERIMENTAL METHODS IN RAREFIED GAS DYNAMICS
Corso tenuto a Varenna (Como) dal 21 al 29 agosto 1964
267
APPLICATIONS OF MOLECULAR BEAMS TO PROBLEMS IN RAREFIED GAS DYNAMICS by
I. ESTERMANN Chief Scientist, U. S. Office of Naval Research, London.
1..
INTRODUCTION In classical aerodynamics the air is considered as a continuum
and its flow characteristics are described by the equations of fluid mechanics, From the standpoint of kinetic theory, the validity of these equations is based on the physical assumption that the behavior of the fluid is determined almost exclusively by the interactions between the individual molecules and only to a very minor, generally negligible, extent by the interaction between molecules and solid boundaries . Another expression of this physical situation is contained in the statement that under the conditions encGuntered in classical fluid mechanics, the mean free paths between individual molecules is very small compared with the dimensions of the bodies in contact with the fluid. If, however, the density of the gas is gradually being decreased to reach such values as, for instance, exist in tr,e upper atmosphere, the mean free path which is inversely proportional to the density will gradually increase until the condition mentioned above is no longer valid. We are then reaching the regime known as rarefied gas dynamics, which is characterized by a mean free path
A. between
the molecules of the gas of the same order or
even larger than the critical dimension d
of the system, The ratio be-
tween this mean free path ). and the characteristic dimension K = ">i d , commonly called the Knudsen number, divides gas dynamics into the various flow regimes, values of those of
K < 0,01 represent continuum flow,
K>10 free molecular flow, and intermediate values the re-
gions commonly called slip and transition flow. In this series of lectu.
269
- 260 I. Estermann
res we will be concerned only with the regions of very large K's, that is with free molecular flow, also called Knudsen flow. In this flow regime individual molecules will make many collisions with the solid walls bordering on the system before they will have a chance to collide with one another. As a result, the flow characteristics are determined by the interactions between solid surfaces and gas molecules while the continuum quantities such as viscosity and heat conductivity lose their importance. For the kinetic understanding of the free molecular flow regi .. me, it is , therefore, important to determine the characteristic
inter~t
-tions between a free molecule and a solid surface. These interactions involve the transfer of energy and momentum. Among the various experimental methods for the determination of these transfer properties, the molecular beam method has the unique advantage of simplicity, at
j
least in principle. Although this simplicity is largely or possibly even over compensated by experimental difficulties, the molecular beam met\'-od is becoming more and more accepted as an aerodynamic tool and
the following lectures will be devoted to an explanation of the principles of the method and to a discussion
of its applications to aerodynamic
problems. 2.
MOLECULAR BEAM TECHNIQUE In the development of modern physics the study of corpuscular
beams has played and is playing a very important role. The investigation of streams of moving particles is
one of the most direct and, at least
in principle, simplest methods for obtaining information about the pro; ... erties of elementary particles. Beams of charged particles have been
270
- 261 1. Estermann
studied since 1879 when Sir William Crookes discovered the cathodic rays during his experiments with electric discharges through gases at low pressures. His observations led ultimately to the discovery of the electron, one of the most fundamental particles in modern physics. Related studies by Goldstein led to the discovery of beams consisting of positively charged ions. Since the beginning of the century, experiments with charged particles have played an ever-increasing role in the devel-opment of modern physics. It was not, however, until 1911 (6) that beams of neutral particles, namely ordinary atoms or molecules, moV-ing in straight lines with thermal velocities, were used in a laboratory experiment, and another ten years elapsed before these beams were applied to investigations of fundamental problems in modern physics. 2.1
Thermal Molecular Beams According to the kinetic theory, a gas consists of a very large
number of
dis~rete
particles which move in random paths, each section
of these paths being a straight line terminated by a collision either with another particle or with a wall of the containing vessel. If a vessel A
,(Fig. 1 ,)
ted space
containing a gas is brought into a highly evacua-
B and a small orifice
sel, the molecules inside
A
0
is cut into the wall of the ves-
whose velocity vectors are pointing to-
wards the orificie will effuse through it and continue to fly into the va'-uum along a straight path. Since the gas molecules inside the vessel have velocity vectors in all directions, the molecules effusing from the orifice 0
will fill a solid angle of 211'. By arranging a screen with
a small opening
C inside the evacuated space, it is possible to
se-
lect from the effusing molecules those whose velocity vectors lie in the general direction of the straight line connecting
271
0
and
(
. (It
- 262 1. Estermann
would be more accurate to say that the molecules selected are those whose velocities fall within the solid angle determined by the limits of the two orifices and the distance between them.) If the orifice 0 small enough to be regarded as the point source, the molecules -ing through
C
is pasJ.
form a molecular beam of particles moving on al-
most parallel trajectories. Since their paths do not intersect, collisions between them will occur only if a faster molecule overtakes a sloW- er one moving along the same path. Such collisions will take place only very rarely. Since we have
taci~ly
assumed that the whole process
takes place in a good vacuum, collisions between the molecules forming the beam and residual molecules in the vacuum space can also be neglected. We may, therefore, define a molecular beam as a stream of molecules moving collision-free in a highly-evacuated space in straight and almost parallel trajectories within the limits of a geometrically defined beam. As far as beam properties are concerned, there is no difference between atoms and molecules. We will, therefore, avoid the term atomic beams ; such beams are simply molecular beams composed of monoatomic molecules . For the purpose of our discussions, which are related to applications to aerodynamics, we will normally consider only beams of molecules that are common components of air, namely oxygen and nitrogen , or occasionally if
for some reason
a mo-
noatomic molecule is desired, argon or helium. In a few exceptional cases we might also consider beams composed of other molecules su"h as atomic potassium which, as we shall see later, makes the perform- ance of certain experiments
much easier.
A molecular beam apparatus, schematically shown in Fig. 1,
272
- 263 1. Estermann
consists at least of the four following components : (1)
A source
A
,in general a vessel filled with a gas or
vapor and equipped with an orifice
0
for the effusion of the molecu-
les. (2) ragm
C
A collimating system, in the simplest case a single diaphwhich together with the source orifice determines the geo-
metrical form of the beam. However, a more elaborate system of diaphragms
is frequently necessary. (3)
A detector for the beam, and
(4)
A vacuum envelope
B in which all the other compo-
nents are arranged and where the pressure can be maintained at such a low level that the mean free path is very large compared to the distance the molecules have to travel. (This means that the Knudsen number in the apparatus must have a large value. ) Other components, which are employed only if required by specific experiments, are velocity selectors or mass filters, which may be of a magnetic
,electrical or mechanical nature, and other special
devices needed for the problem to be attacked. For the purpose of the following discussions, these special devices will be restricted to solid (or occaSionally liquid) surfaces with which the beam molecules may exchange energy and momentum. 2.1.1.
Molecular Beam Sources Since the earlier molecular beam experiments were carried out
with metal atoms which were produced by evaporation from a solid or liquid supply contained in the vessel A
in Fig. 1 under the effect of
elevated temperatures, the source is commonly called an "oven ll
,
e-
veR if no heating is involved. For beams of common gases, the source
273
- 264 -
1. Estermann
may be an orifice or slit connected by a tube to a reservoir containing a gas such as nitrogen or argon, whose pressure is kept constant. The temperature of the source can be varied by external heating or cooling. (More recently, supersonic jets have been used as beam souces). These sourqes differ fundamentally from the first type because the beams emitted by them are not in thermodynamic equilibrium. They will be discussed in section since the following discussions will not involve experiments requiring complicated oven constructions, no further details will be given here; they may be found in references(2) and (3). 2.1. 2.
Molecular Beam Intensity Probably the most important factor for the design of a mole-
cular beam experiment is the obtainable beam intensity . In the case of an ideal orifice used as source, the beam intensity depends on the rate of effusion. For true molecular flow, that means under conditions where the mean free path in the gas or'vapor behind the orifice is large compared to the dimensions of the orifice and where the thickness of the wall is small compared to its width or diameter, the rate of effusion can be calculated easily from the elementary kinetic gas theory . It is simply equal to the number of molecules (N) which strike a wall area equal to the area (a) of the orifice per unit time,
N = 1/4 nca where
n
p,
molecular weight
c
their average
N can be expressed in terms of the pressure
the absolute temperature T, and the mass m
tions
(1)
is the number of molecules per cm 3 and
velocity . The number
namely
of a molecule or the
M of the beam-forming substance by using the relaN = p/kT and
c = (8kT /1T m) 1/2 274
- 265 I. Estermann
where
k
is Boltzmann's constant. The result is
(2) Introducing the numerical values of the universal constants and measuring
p
in mm
Hg, we obtain for the number of mols effusing per
unit time Q
= 5.83>«0
Equations 1, 2 and
-2
pa(MT)
-1/2
mol sec
-1
(3)
3 give only the total rate of effusion of
a gas through an aperture. For the formation of a molecular beam it is, however, important to know the angular distribution of the molecules. According to Knudsen, the number of molecules dN leaving a surface element dO" of an aperture
and contained in an element of solid
angle dw which makes an angle
9 with the normal to dO' is given by
dN
= const.
cos9~w
d.r
(4)
This equation is known as the Cosine Law of molecular effusion and is illustrated in Fig. 2 • It has been verified experimentally and is completely analagous to Lambert's Law for the intensity of radiation emitted from a light source. The intensity of a molecular beam is defined as the number of molecules passing per second through a unit cross section of the beam or impinging upon a unit area of a detector. From Fig. 3 it follows that the number of molecules originating from a surface element dO" of the source
P
and arriving at surface element
dO'l
of the detector
P'
is given by nE 4
dG" do" cos 91 cos 92
2("7'2(21T
r}o
275
)0
cos
.
e sm
(5)
9 d
ed ~
- 266 -
1. Estermann
where d
81 and 82 are the angles between the line connecting dO' and and their respective normals . The intensity of the beam at pi
can be calculated by integrating
equation (5) over dO'. In most molecu-
lar beam arrangements, cos 81= cos8 2 = 1 and the result of the integration is 5.83 x 10
d Ncr'
'i'r
-2
a p dO-'
I -1 mo sec
(6)
r2 (MT)1/2
where a is the area of the source aperture, p the pressure in the source measured in mm Hg, r the distance of the target surface d I from the oven slit and I the intensity of the beam at d (jl.
2. 1. 3
Collimating Systems Referring to collimation, the essential difference between mole-
cular beams and beams of charged particles , such as electrons or ions, lies in the electric neutrality of the former. As a result, they cannot be focussed by electrostatic or magnetic lenses and their optical behaviour is , therefore, more like that of a beam of light in a pinhole caln-era. The simplest collimating system is a screen equipped with a single orifice or slit
x.
as shown in Fig. 4
which, together with the source slit, defines a beam This beam consists of a central region of constant
intensity (umbra) and a region on either side (penumbra) where the intensity drops linearly from the maximum value to zero. If we plot the intensity in the center plane of the beam against position, we obtain, therefore, a trapezoid (Fig, 5). ~ In many applications, narrow rectangular slits are used as orifices; it is, therefore, customary to use the terms source or oven slit, colli-
mating slit, and detector slit independent of the geometrical form of the orifices. 276
- 267 -
1. Estermann
In most practical applications, the collimating system will con .. sist of several slits. The employment of an additional slit is sometimes governed by a desire for a more precise alignment, but in most of the cases of interest to us in connection with aerodynamic applications of molecular beams, the reason for the use of several slits is differen . Contrary to ordinary ligth beams, molecular beams are attenuated dUI"-ing their paths because they travel through a scattering medium: the residual gas in the apparatus. This gas originates in part from the imperfections of the pumping system and from small leaks but to a major extent from the molecules effusing from the source slit -ted by the collimator.
which are reje(-
If these molecules, which may be a thousand
times as numerous as those passing through the collimating slit are not condensed by
strongly-cool
surfaces, they will be reflected from the
walls and result in a residual pressure which decreases the mean free path in the apparatus.
The use of additional slits or canals makes it
possible to partition the apparatus into several sections which may be evacuated by separate pumps. If, for the reason mentioned above, the pressure in the vicinity of the source has a value p, the pressure in the next section separated from the first one through a solid wall containing only a small opening where
ran be lower by a factor R
= W/W f
Wf is the flow rate thr ough the connecting orifice and Wp the
effective pumping speed of the pump evacuating the second section. The same reduction will
apply to the third and fonowing sections. Since
it is easy to make the factor
R
= 100, the SUb-division of the apparatus
into several separately pumped sections (differential pumping) can reduce the pressure in the second and third sections to tolerable values even if the pressure in the first section remains relatively high. If, as
277
- 268 1. Estermann
a practical example, we assume that the pressure in the first chamber surrounding the source is one micron, the pressure in the second region will be. 01 micron and in the third region. 0001 microns, provid..- ed that the pumps employed are capable of maintaining their capacity at these low pressures. To obtain large values of R, it is necessary to employ canals instead of simple orifices; if the alignment is good, the flow resistance of a canal for a molecular beam is the same as of an orifice while its flow resistance for the molecules of the residual gas, which move in random directions, is larger by a factor I( which may be of the order of 50, as explained in section 1. 4. The great advantage of this procedure for minimizing beam attenuation can be shown by the following considerations. The effect of the scattering gas on the intensity I of a molecurar beam per unit length of passage is given by I
= 10
e
-lit..
(7)
where \. is the mean free path of the beam molecules in the scatter-ing gas and 10 the intensity in the absence of a scattering gas.
(A
is
inversely proportional to the pressure.) If we divide the apparatus into several chambers, the intensity at the end of the
nth chamber will be
given by (8)
where L1 ,L 2 andL n are the lengths of the path of the beam mole cu· les in the various chambers and
A. 1 'A 2 and A. n the mean free paths
corresponding to the pressures in the various chambers. If the pressure ratio between two successive chambers is given by the factor defined earlier, the equation modifies to
278
R
- 269 1. Estermann
(9)
Because of the exponential character of the attenuation of the beam, most of the weakening takes place in the first chamber; the beam path in this chamber should, therefore, be made very short. The distances in the second and following chambers can be made considerably longer without any appreciable weakening because the corresponding exponentials approach unity very rapidly if
R is large enough . A suitable
arrangement for differential pumping is shown in Fig. 6. Equation 7 also shows that the mean free paths
A in
the apparatus should be se-
veral times as large as the distances travelled if a good beam intensity is to be obtained. It is easily recognizeable that for
A=
L the beam
will be weakened by a factor of e or to about 1/3 of the original intensity. 2. 1.. 4.. Detectors
While the discussions of source.s and collimating systems in the preceding sections apply more or less to molecular beams of all kinds of substances, the technique of detection depends to a large degree on the physical and chemical properties of the molecules forming the beam. Since we are
at the moment concerned only with aerodynamic appli-
cations, we will confine our discussion of detectors to those which are applicable to beams composed of permanent gases such as the major components of air . As an exception we will also discuss briefly a detector suitable for alkali atoms, not because they are directly employed in experiments concerned with aerodynamic problems but because they play an important role in certain model esperiments which have indirect relations to one of the problems in which we are interested.
279
- 270 1. Estermann
2. 1. 4. 1. Manometer Detectors The most common detectors for non-condensable materials use as a sensitive element a low pressure manometer which registers a pressure increase in an otherwise closed vessel equipped with a narrow entrance slit or canal through which the beam molecules can enter . While such a slit or canal, if properly aligned, does not offer any impedance to molecules moving in the proper direction, it has a significant flow resistance for molecules with random velocities.
Since the
velocity vectors of the beam molecules entering the vessel will become randomized after collisions with the walls, their admission will cause a pressure increase in the vessel until equilibrium is reached, that means until the number of beam molecules entering per unit time becomes equal to the number of random gas molecules leaving in the same interval. From the equations given in Section 2.1.2. the number of incoming molecules through a slit of the cross sectional area a I is given by (10)
where p
is the pressure in the oven, a the area of the oven slit
o
and r the distance between the oven slit and the detector slit. The number of effusing molecules is given by equation
where
tz.
= p a l I (21lmkT' e D PD is the equilibrium pressure and
n
(11 )
TI the temperature inside
the detector vessel. The equilibrium pressure is given by n.
1
= n , or
2 = (p al1lr ) (T' IT) . o
e
(12)
If a canal is used instead of a simple entrance slit, the flow impedance
for the randomized molecules is increased by a factor
280
k which
de-
- 271 -
1. Estermann
pends primarily on the ratio between the width and the depth of the canal (for details see reference (2), page 312). This factor has in actual practice been made as large as 50 or even more. Taking as an example a helium beam with a = 10
-3
cm
2
r = 20 cm, Po = 1 mm, we obtain for PD approximately 10- 6mm Hg if an ideal entrance slit leads to the detector, or canal having a
5 X 10- 5 mm with a
k - factor of about 50.
Since beam intensities should be measurable to better than
0-
ne per cent and deflected or scattered molecules amounting to possibly only one percent of the parent beam should be measurable with fair accuracy, the registration of such small pressure increments requires a
8
very sensitive manometer (sensitivity of 10- to 10
-9
mm Hg or better.)
Moreover, the total pressure in the manometer chamber will be of the same order of magnitude as the pressure in the collimating chamber, namely between
10- 6 and 10- 8 mm Hg and may fluctuate by as much
as a few per cent because of uneven performance of pumps. It is, therefore, necessary to use
a manometer which is so designed that it -9 can indicate pressure changes of the order of 10 mm Hg in a background or ambient pressure of 10
-6
mm Hg . Moreover, the effects of
random fluctuations in the apparatus must be eliminated or at least strongly diminished. The first requirement has been met by the construction of hot wire (Pirani) and ionization gauges of the necessary sensitivity, the second by the employment of two identical gauges, one of which is exposed to the beam and the other pointing in the opposite direction. The two gauges are connected to an electrical network which registers only the difference in pressure between them. More details about the construction of such manometers are found in references (2)
281
- 272 1. Estermann
and (7), but it should be mentioned here that sensitivities of the order of 10- 10 mm Hg have obtained in practice. Although the manometer detectors have been developed to an acceptable performance level, they have a number of inherent shortcolrI- ings. They suffer, for instance, from lack of zero stability but an even more serious defect is their inability to distinguish between pressure changes produced by a signal and those due to noise . Since most of the experiments are performed in a noisy environment, namely the imperfect vacuum in the apparatus, the compensation of pressure fluctuations remains a difficult problem. Although the application of a pair of gauges as explained above provides some improvement, the most commonly employed technique for
the impovement of signal to
noise ratio, namely modulation ,is difficult to apply. That is due to the relatively large time constant of the manometer detectors which in practice is of the order of 10 to 60 seconds and which does not, therefore, permit modulation by chopping the beam with a reasonable frequency. 2.1.4.2.Beam Ionization Detector.(8) This device, which overcomes the last problem mentioned in the preceding paragraph is also called IIcross-fire ll detector. The beam is exposed to a transverse electron bombardment by which the beam atoms or molecules are transformed into electrically charged particles. These ions are accelerated and directed towards a surface which emits secondary electrons under ion bombardment. The secondary electrons are converted into measurable currents with the aid of conventional amplifiers. Since the associated processes have very short time constants,
282
- 273 -
1. Estermann
modulation with frequencies of the order of 10 to 100 hertz is quite feasible and the arrangement is inherently very sensitive. If an amplifier tuned to the modulation frequency is employed, the noise is suppressed except for the very small component which has the same frequency as the modulating frequency. The application of a mass filter makes it also possible to distinguish between signals due to beam molecules and those caused by the molecules of the residual gas in the apparatus. On the other hand, the efficiency of the ionization by electron bombardment is only of the order of 10- 3 to 10- 4 so that the over-all sensitivity of the equipment is probably just competitive with that of the manometer detector. 2. 1.4. 3. Surface Ionization Detector. A detecting device which is both very simple and permits moJ-alation is the surface ionization detectoA9)
It operates on the follow-
- ing principle : If a molecule of low ionization potential I impinges on a surface with an electron work function
¢
larger than I, there is a
certain probability for the impinging atom to lose its valence electron and, if the surface temperature is high enough, to evaporate as a pus-itive ion.
The ratio of the number of re-evaporating ions
number of re-evaporating neutral atoms
+
N to the
N is given by
~ /N = exp (¢ - I) /kT where k
= 8.5 x 10
-4
0
ev/ K is the Boltzmann constant and I and
¢
are
measured in electron volts. An examination of the values of the physical constants involved shows that potassium or cesium atoms impinging o on a tungsten surface of 1200 K have a very high probability of ionization. For these and other alkali atoms as well as for certain molecules
283
- 274 -
1. Estermann
containing alkali atoms, the surface ion detection works very satisfa'-torily . When extreme sensitivities are needed, it can be combined with a beam modulator and a mass filter and achieve an excellent sig-nal to noise ratio.
For the beams composed of the common compo-
nents of air, such as nitrogen or oxygen or even monoatomic rare gases such as argon or neon, the ionization potentials are unfortunately far higher than the work functions of any suitable electron emitter and this technique is, therefore, not applicable. It can, however, be utilized in model experiments where alkali atoms can be substituted for tho-
se which are of real aerodynamic interest. 2.1. 5.
Vacuum Envelope. Since, as pOinted out in the preceding section, a considerable
pressure is produced in the apparatus by the molecules effusing from the source which are not accepted by the collimating slit, a large pumpwing capacity is needed to maintain the mean free path at a sufficiently low level. Further improvement of the vacuum conditions can be obtaihwed by differential pumping, that means by a subdivision of the apparatus into several compartments with separate pumping systems. As a rough practical estimate, the pumps attached to the first or oven chamber will need a capacity of
100 liters / sec or more while those evacua-
ting the subsequent chambers should be selected more for low final pressure and freedom from fluctuations than for extreme pumping capac-ity. For experiments where surface contamination by the pumping fluid may be a serious problem, the use of oil diffusion pumps may not be advisable. Mercury diffusion pumps, although somewhat slower in
pum~
wing capacity, are much better in this respect but even cleaner conditions
can be obtained by the use of ion pumps ("Vacion" pumps) or by
284
- 275 I. Estermann
cryopumping, that means condensation of the residual gas by surfaces cooled with liquid hydrogen or helium. For the beams of alkali metals which can be used for model experiments related to aerodynamic problems, the pumping situation is usually much simpler. The excess molecules can be removed by
con-
densation on a liquid air or liquid nitrogen cooled surface and the required pumping speed is, therefore, considerably less. However, more emphasis should be placed on freedom from fluctuations and on the ultimate vacuum obtainable. Protection of the surfaces to be tested from contamination is also an important factor. In every case the judicial selection of the proper pumping system can facilitate experimentation to a large extent. 3.
INTERACTION OF ATOMS AND MOLECULES WITH SURFACES For the calculation of the transport properties of gases at low
densities, a knowledge of the interaction between individual molecules and solid surfaces is essential. The conventional methods for the determination of the energy and momentum transfer are usually expressed "accomodation coefficients" and depend on measurements of viscosity and heat transfer between two solids surrounded by the gas under inve'-tigation. One of the difficulties associated with these methods
lies in
the fact that, as the density of the gas becomes lower and lower, the rates of transfer become so small that their accurate measurement is very difficult. On the other hand, if one increases the density to the point where the measurements become easier, the conditions for ideal molecular flow are no longer maintained. Moreover, all these measurements produce only a value for the transport coefficients but do not permit an investigation of the phenomena on a molecular scale. 285
- 276 I. Estermann
The aim of the molecular beam method is to determine the change in the energy and momentum distribution of individual molecules due to their interaction with a surface. For this purpose it is necessary to measure this distribution before and after the impact and to correlate the results with the energy and momentum transfer suffered by the surfaces. If such measurements are successful, the
transport prop-
-erties of a rarefied gas can be computed with the aid of the kinetic theory which describes in an acceptable way the interactions of molecules with each other. In the case of extremely low densities, the latter interactions can be completely neglected. 3. 1
Historical Review The earliest clues to the nature of molecular scattering at the
surface were obtained from experiments on the damping of vanes and discs in rarefied gases and in the observations of the deviations from hydrodynam- ic theorv in viscous flow through capillaries. For the explanation of these observations,Maxwell(lO\ntroduced a coefficient of momentum transfer f or, as it is now called, ... He interpreted this coefficient as representing the fraction of molecules reflected by the surface according to the cosine law and the complementary value
1-f as the fraction reflected specu-
larly. In an analogous fashion, Knudsen (5) defined an accomodation coefficient for the energy transfer between a gas and a surface which is commonly designated by the symbol cl and which was originally thought to represent the fraction of the impinging molecules which "accommodate" , that means which leave the surface after impact with a temperature corresponding to that of the surface while quantity
the complementary
1 - 0\. represents the fraction which leaves the surface after
impact without changing their temperature. More recently Schaart ll )
286
- 277 I. Estermann
proposed the introduction of a third transfer coefficient
(J"
representing
the efficiency of normal momentum transfer between the gas and the surface while the coefficient 0' represents the efficiency of tangential momentum transfer and the coefficient
0("
the efficiency of
~
transfer.
Inspection of more recent measurements shows that the coefficient 0" is usually close to unity, giving strong support to the belief that in gene.. _ -al diffuse or cosine scattering prevails while the reported values of 0(. are frequently rather small and, for light gases and very clean systems, often approach zero. Very few experimental values for the coefficient 0'1 are as yet available. 3.2.
Momentum Transfer The measurements of viscosity and heat conductivity referred to
earlier permit at best the determination of an overall coefficient f or ex. but do not give any details about the process on a molecular scale. If, however, a beam of unidirectional molecules is directed onto a surface and the spatial distribution of the scattered molecules is measured,
one
obtains considerably more detailed information. The first experiments of this sort were carried out by
Wood(12) and Knudsen(13) who used beams
of mercury and cadmium reflected from glass surfaces. They appeared to verify the cosine scattering law to a very large degree, that means gave values of f close to 1. (14)
,and Estermann and Stern
Later experiments by Knauer and Stern (15).
with hellum and hydrogen beams scat-
tered from alkali halide surfaces produced evidence of a considerable amount of specular reflection. Their results can be interpreted by the assumption that for systems of this sort the coefficient f may be very small or even zero. Experiments with zinc, cadmium and mercury
287
- 278-
1. Estermajln
atoms scattered from alkali halide cry s tal surfaces by Ellett(16) and his collaborators and by Josephy (17) showed a rather different behavior. Typical results are shown in Fig. 7, and they are qualitatively similar for all reported cases . An examination of these results shows
imme~
diately that the scattering is neithe!' specular nor cosine and that a simple resolution of the observations into a fraction f undergoing cosine reflection and a fraction 1 ~
f
undergoing specular reflection as propo~
sed by Maxwell is not obvious. The results can best be described as quasi~specular
scattering with the maximum not pointing exactly into
the direction of the specularly reflected ray but lying somewhat closer to the normal. Comparison of the results obtained with different beam and crystal temperatures shows the following general behavior: (1) The position of the maxima shifts towards the specular ray with decreasing crystal temperature and with an increase in beam temperature and (2) the relative number of atoms scattered according to the cosine law increases with a decrease in crystal temperature. More extensive results were reported recently by Datz, Moore and Taylor (18) and by Smith and Fite(19). A portion of Datz's results are reproduced in Fig. 8, which shows the scattering of helium atoms
(at source temperatures up to
approximately 1900 0 C ) from platinum surfaces whose temperature was varied between 50 and 1200 0 C . The striking feature of these tests is the reversible transition from
quasi~specular
to entirely diffused
scat~
tering upon passage of a broad threshold in the temperature of the
sur~
face in the region of 100 to 200 0 C. This transition seems to be connected with the absorption of oxygen; the conditions for manolayer formation being energetically correct in this temperature region. Once again the maximum of the lobe is shifted towards the specular ray at high beam
288
- 279 -
1. Estermann
temperature and in the direction of the surface normal at low beam temperatures. Similar but more complex observations were made by Smith and Fite on the scattering of H2 molecules and argon atoms from nick...el surfaces. It was found that certain N. scatterers bearing a trace of 1
carbon produced marked specular maxima after bake-out to 600 0 C in the presence of H2 . The specular maxima, shown in fig. 9, were observed only in the surface temperature range from
200 to 300 0 C. At
lower temperatures the surface was presumed to be obscured by condensed gases but the return to diffuse scattering at temperatures bet¥een 300 and 1000 0 C is not yet understood. It should be noted that Fite's apparatus permits a crude determination of the evergy accommodation experienced by the particles at the surface. From
his estimate
it is found that the portion of the reflected beam near the maximum of the quasi-specular lobe undergo almost no energy adjustment while portions in directions corresponding more to the surface normal suffer almost complete adjustment. Experiments by Estermann, Frisch and Stern(20) on the reflection of hydrogen and helium from lithium fluoride surfaces has shown in 1931 already that the specularly reflected molecules do not suffer any energy accommodation at all. Experiments with the normal components of air, such as oxygen, nitrogen and argon, on glass and metal surfaces were carried out extensively by Hurlbut
(7,21) and his collaborators. In most cases
they found perfect cosine scattering*
* For
as shown in Fig. 10 and only if
pure cosine scattering, the end points of the scattered beam vectors are located on a sphere (Fig. 11).
289
- 280 1. Estermann o the angle of incidence measured from the normal is close to 90 , an
appreciable amount of quasi-specular reflection was observed . These observations agree with earlier but not quite so elaborate measurements by Knauer and Stern who also found a predominant scattering in the forward direction at very large angles of incidence. Regarding the normal momentum transfer represented by the coefficient
0': an attempt has been
made by Stickney (22) to obtain a meas-
-urement of this quantity. In his studies, beams of noble gases and of N2 ' CO 2 and H2 were directed against surfaces of tungsten, platinum and aluminum
in a torsion balance apparatus. The results may be summa-
rized as follows : The transfer of normal momentum at normal incidence is about equally complete .for argon and nitrogen, more complete for
CO 2, and less complete for neon, N2 and helium. No difference in behaviour among the three surface materials tested was observed, suggestIng that surface contamination played a major role in these measurements. To summarize the results obtained to date, one can only say that the situation is rather complicated and does not yet permit a simple explanation of all the observed results.
One might, however, state
with a fair amount of certainty that for light gases and technical surfaces, scattering according to the cosine law is almost always prevalent, with small deviations occurring only at very large angles of incidence. For atomically better defined surfaces, namely cleaved crystal surfaces, the situation is far more obscure. Here quasi-specular scattering is most frequently encountered, and purely specular reflection is limited to light gases and very good crystal surfaces, such as those of LiF. The experiments have also shoWn that the condition of the surface, particularly the possible absorption and desorption of gases, has a very
290
- 281 I. Estermann
large effect on the observations . Reffering in particular to some of the results obtained by Fite, the possibility of chemical reactions occurring between the molecules of the beam and those of the surface may further obscure the elementary phenomena. With reference to possible aerodynamic applications, one may expect that good crystal surfaces would probably suffer less drag in a rarefied gas flow than ordinary technical surfaces, but it seems very improbable that they may lend themsel ves for practical applications. 3.3.
Energy Transfer Measurements of the thermal accommodation coefficient which
represents the efficiency of heat transfer between a gas and a solid surface are numerous, but unfortunately the agreement between results obtained by different investigators is rather poor. In a survey paper, J. P. Hartnett (23) has flatly come out with the statement that all the published measurements are unreliable. As pointed out earlier, one of the reasons for this state of affairs may be due to experimental difficulties, particularly to the fact that if the pressure regime is so chosen that true molecular flow exists, the magnitude of the heat transfer through the gas is of the same order or sometimes even smaller than the unavoidable heat leaks and is, therefore, difficult to measure. Another difficulty arises from the condition of the solid surface. It is very difficult to obtain reproducible, clean surfaces and one frequently does not know whether one measures the accommodation on the material supposedly under investigation or the accommodation on a layer of condensed or adsorbed material . There is, therefore, good reason to question the validity of many published results. Among the most reliable measurements are those reported by
Wachman (24)
291
- 282 -
1. Estermann
The second shortcoming of the traditional method for the meaS-urement ·of the energy exchange between solids and gas molecules arises from the fact the scattered molecules may not be in thermodynaWl- ic equilibrium and that their energy
con~ent
may not be describable by
a temperature as is usually the c'Jstom. It is, therefore, interesting to design a molecular beam experiment for the measurement of thermal accommodation . For this purpose the energy distribution in the parent beam was measured and this measurement repeated after the scattering. Work begun by Bennett and Estermann (25) and continued by Marcus and McFee (26) has not solved this problem in all details but has identified an experimental technique which might ultimately answer some of the important questions. For reasons of experimental convenience the measurements were carried out with beams of potassium atoms reflected from various surfaces at various temperatures. Potassium is, of course, far different in behavior from the consituents of the air and it is, therefore, impossible to translate the results directly to aerodynamic problems. The choice of this material was governed by the existence of the surface ionization detection which, as pointed out in section 2. 1.4.2., is far superior to the manometer detectors but unfortunately only applicable to atoms or molecules with a low ionization potential . The results can be summarized as follows : In every case over a temperature variation of the reflecting surface from 350 to over 2000 0 K, the beams of potassium reflected from tungsten or other metallic targets exhibit a velocity spectrum characteristic of the temperature of the target surface, irrespective of the velocity distribution in the parent beam. The velocity analysis of the parent and reflected beams was made by means of a mechanical filter. The beam source in this
292
- 283 -
I. Estermann
case is a small oven in which metallic potassium is heated to a tempe,._ -ature between 250 and 300
o
C; the beam molecules effuse through an
"ideal" slit into a chamber maintained at a very good vacuum. The velocity selector consists of two slotted discs driven by two synchronous motors supplied from the same oscillator rotating with approximately 8000 rpm. A phase shifter varies the phase of the voltage fed to one motor relative to the other. The change in phase results in an angular displacement between the slots on the two discs. For each phase angle, only a narrow velocity region is transmitted through the selector . If the intensity of the beam is plotted as a function of phase angle, one obtains directly a reproduction of the velocity distribution. By coupling the motor driving the chart in a strip chart recorder with the phase shifter, this velocity distribution can be recorded automatically. Curves obtained in this form are shown in Fig. 12. One of the important results obtained with this apparatus was that the parent beam has a Maxwellian velocity distribution so low
only if the pressure in the oven is
that the mean free path of the molecules in the oven is large
compared to the width of the oven slit (curve a in Fig. 12) . If the oV-en pressure is increased, the velocity distribution becomes distorted in the sense that the intensity of slow molecules falls considerably below the expected value (curve b in Fig. 12). For the scattered beams (Fig. 13) it has been shown that in all conditions investigated the veloc-ity distribution measured after scattering corresponded exactly to the temperature of the target irrespective of the distribution in the parent beam, i. e. thet the acommodation coefficient cds equal to one. The accuracy of the measurements was better than 17.. It had been hoped to find a system or conditions in which the accommodation coefficient was
293
- 284 I. Estermann
less than one, but the only indication of this sort was obtained in scattering of K - atoms from a lithium fluoride crystal. The results obtained in this case were, however, not explainable on the basis of a simple theory. An indirect way of obtaining information about accomodation coefficients results from the work of Stickney (22) mentioned earlier. If the observations on the normal momentum transfer coefficient
()"I
are
examined under the assumption that only cosine scattering takes place, the results can be related to accommodation coefficients for translation of energy . The values obtained in this form for a tungsten surface vary between D.55 for helium and 1.0 for nitrogen. In conclusion it may be said that the investigation of the energy transfer between surfaces and molecules of thermal velocities have not yet resulted in a.'1y significant answers to aerodynamic problems. It is hoped that the work can be extended to systems which permit simultaneous measurement of
(j
and d.. so that a correlation between these two
quantities can be obtained. 4.
NON -THERMAL BEAMS.
4.1
Intensity and Energy Limitations of "Classical" Molecular Beams. The application of classical molecular beams is subject to two
fundamental limitations, one relating to the maximum intensity, the other to the maximum energy attainable. The first limitation is based on the equations given in Section 2.1.2, because some of the parameters appearing in these equations cannot be set arbitrarily without destroying some of the beam properties, and because others must be adjusted to the conditions of specific experiments. To obtain numerical values for beam intensities, the geometric
294
- 285 -
1. Estermann
parameters of Eq. 6 must be chosen first. Following a recent comprehensive review of this subject (27), and keeping aerodynamic applications in mind! we chose for the distance the point at which the intensity
I'
between the beam source and
I is to be measured, the value of one
meter. We also retain the assumption that cos 81 = cos 82 = 1. The value a, however, cannot be selected arbitrarily without a correspond.-ing selection of the source pressure p. In effect, the requirement for molecular effusion that the mean free path of the molecules in the source should be of the same order of magnitude as the characteristic dimension of the beam source, imposes a fixed value on the product a· p for any given source geometry. If one tries ,for example, to increase p
at a
given value of a, or a
at a given value of p, one finds that
the intensity does not increase linearly with either of these variables. Moreover, as discussed in Sect. 3.3. and shown in Fig. 12, the
velo~
-ity distribution undergoes a more or less severe disturbance which mayor may not interfere with the purpose of the experiment. Although an ingenious choice of the shape of the source may ameliorate the problem, the sky is definitely not the limit. The other two variables in Eq. 6, namely M and T, are usually determined within narrow ranges and allow only little flexibility. Assuming an ideal circular opening of the diameter d as beam source and setting the product ap at the maximum value permissible by the condition d= A , we obtain for the intensity of a nitrogen o. 13 2 sec. beam at 300 K at 1m distance the value 5x10 molecules / cm At first glance J this seems like a big number. In order to bring it into proper perspective from the standpoint of experimental measurability , we will translate it into other types of flux assuming a detector a-
295
- 286 I. Estermann
rea of lcm 2 . Using the proper conversion factors, we find that 10 13 -6 molecules / sec mean about 10 amps (if all particles carry a single
-5
electrical charge) a momentum flux of 3 x 10 dynes, an energy flux -10 0 -6 of 10 watts/ (T t t- T ) C, or in terms of mass, 10 arge source grams/hour or the formation of one monolayer per minute . The usa.· ble values of this flux are, however, frequently much smaller, e. g. when velocity filters or large spatial resolution are required. The real issue is, of course, not the actual magnitude of the signal produced by the flux, but the ratio of signal-to-noise. The major noise source in molecular beam experiments is the residual gas pressure in the vacuum envelope, and in particular its fluctuations. Choice of good pumps and an absolutely air-tight and well outgassed system are, therefore essential conditions for successful experiments. It is itnportant to realize that even at pressures of 10
-7
mm Hg, the number of
residual gas molecules striking a target area of lcm 2 is of the same order as that of the molecules of our assumed beam; thus improvement of vacuum technology to permit operation under ambient pressures of -9 -10 10 or 10 mm Hg may substantially improve the signal-to-noise ratio. other techniques, such as differential detection or beam modulation, are also very effective in improving this ratio. Choice of more sophisticated shapes of beam sources is another approach to higher signal-tonoise ratios
but in each case , one runs into the inevitable law of di-
minishing return, and radically new concepts are necessary to permit a large step forward. 4.2
High-Intensity Beams The most promising approach so far is the use of a supersonic
nozzle as beam souce which may ultimately permit an increase in usa. -
296
~
287 1. Estermann
ble beam intensity by several orders of magnitude over even the more sophisticated sources of thermal beams. This idea was first proposed in 1951 by Kantrowitz and Grey (28), but not successfully applied until 1954, when Becker and Bier (29) demonstrated the feasibility of this method. Even today, although a number of groups of experimenters are working on this problem, "nozzle" beams are still in an early stage of development. More details are to be found in reference (27) and will not be given here since applications to aerodynamic problems have not yet been reported. It should, however", be pointed out that nozzle beams are not in thermal equilibrium and offer molecular velocity distributions quite different from those obeying Maswell's law. It should also be
no~
-ed that the average energy of the molecules in these beams is higher than in thermal beams, but not by a significant factor. 4.3.
High-Energy Beams The average energy in thermal molecular beams is controlled
by the source temperature T. - it the value of
It is obvious that material problems liM-
T to the order of a few thousand degrees K, alloW-
-lng at best an improvement of a factor of 10 over room temperature sources, A further augmentation of the molecular energy may be obtai\'\.-ad by selecting only the atoms in the high-energy tail of the Maxwell distribution. These two measures may yield beams containing a substantial fraction of molecules with energies up to 1 e V, but it seems unprof"2"
-Uable to try to go beyond this range without recourse to fundamentally different concepts. The application of shock waves seems to point to a way out of this dilemma, but has the great disadvantage of permitting operation only in the time range of milliseconds. Electric arc heaters, as well as mechanical acceleration, do not appear to provi-
297
- 288 1. Estermann
de a substantial improvement, although the production of beam energies up to leV has been reported (30). A far more promising approach is the electrical acceleration of charged atoms or molecules (ions) with subsequent neutralization. Charged particles have been accelerated to energies of millions and billions of electron wolts and the technique of acceleration is at present a highly developed art. For the purpose of aerodynamic applications, energies between land 100 electron volts are, however, of real concern. The lower limit of this range is given by the desire to make a connection with the maximum energy range obtainable by thermal techniques, the upper by the energies which correspond to the relative velocities of gas molecules striking satellites or space vehicles in the upper atmosphere. Beams of the order of 10eV would be most desirable . It is fortunate that the collision cross sect ions for charge ex-
change between an ion and its neutral counterpart are several times larger than those for momentum transfer. By passing a beam of ions through a chamber filled with a gas composed
of similar molecules
at a proper pressure one obtains a beam containing a large fraction of neutral atoms having essentially the same velocity as the parent ions. Remaining ions can easily be deflected by electric or magnetic fields so that a completely neutral beam is the final product. This technique has been pioneered by Amdur (31) and his cowork-ers since 1940 , but applications gun only recently by Devienne
to aerodynamical problems were be-
(32). His experiments, however, are
still of a preliminary nature and while some results were given during the lecture course, it seems premature to include them in this review.
298
- 289 -
I. Estermann Bibliography A
General References 1.
R. G. J. Fraser, Molecular Rays, Cambridge Univ. Press, 1931
2.
I. Estermann, Molecular Beam Technique, Rev. Mod. Phys. 18, 300,1946
3.
N. F. Ramsey, Molecular Beams, Oxford Univ. Press, 1950
4.
I. Estermann (ed), Recent Research in Molecular Beams, Academic Press, New York, 1959
5. B
M. Knudsen, Kinetic Theory of Gases, Methuen, London, 1934
Individual Papers 6.
L. Dunoyer, Le Radium, 8,142,1911; 10,400,1913
7.
F. C. Hurlbut, Univ. of Calif. Eng. Proj. Rept HE-150-118, 1953
8.
G. Wessel and H. Lew, Phys. Rev. 92,641,1953
9.
J. B. Taylor, Z. Physik 52, 846, 1929
10.
See, e. g. , J. C. Maxwell, On the Dynamic Theory of Gases, Cambridge Univ. Press, Cambridge, 1960
11.
S. A. Schaaf and P. L. Chambre, Flow of Rarefied Gases in High Speed Aerodynamics and Jet Propulsion, Vol. III, Part. H, Princeton Univ. Press, 1958
12.
R. W. Wood, Phil. Mag. 30, 304, 1915
13.
M. Knudsen, Ann. Physik 34, 593, 1911
14.
F. Knauer and O. Stern, Z. Physik 53,779,1929
15.
I. Estermann and O. Stern, Z. Physik 61,95,1930
16.
A. Ellett and H. Zahl, Phys. Res. 38, 977, 1931
17.
B. Josephy, Z. Physik 80, 733, 1933
18.
S. Datz, G. E. Moore and E. H. Taylor, in Rarefied Gas Dynamics, 3rd Symposium, (J. A. Laurmann, ed.) Vol. I, p. 347 Academic Press, New York, 1963
299
- 290 J. Estermann
19. J.N. Smith and W.L. Fite, ibid, p. 430 20. I. Estermann, R. Frisch and 0 . Stern, Z. Physik 73, 348, 1931 21. F.C. Hurlbut and D.E. Beck, Univ. of Calif. Eng. Proj. Rept. HE-150, 166, 1959 22. R.E. Stickney and F. C. Hurlbut, in Rarefied Gas Dynamics, 3rd Symposium, (J.A.Laurmann, ed ) Vol. I, p. 454, Academic Press, New York, 1963 23. J. P. Hartnett, in Rarefied Gas Dynamics, 2nd Symposium,(L. Talbot, ed) , p.1, Academic Press, New York, 1961 24. H. Wachman, Ph.D. Thesis, Univ. of Missouri, 1957 25. A.I. Bennett, Ph.D. Thesis, Carnegie Inst. of Tech. 1953 26. P. M. Marcus and J. H. McFee, in Recent Research in Molecu.lar Beams (1. Estermann, ed), p. 43, Academic Press, New York, 1959; J.H. McFee,Ph.D. Thesis, Carnegie lnst. of Technology, 1959 27. J.B. Anderson, R.P. Andres, and J.B. Fenn, High Energy and High Intensity Molecular Beams. To appear in Advances in Atomic and Molecular Physics (D. R. Bates and 1. Estermann, eds). Vol. I, Academic Press, New York, 1964. 28. A.
Kantrowitz and J. Grey, Rev. Sc. Inst. 22, 328, 1951
29. E. W. Becker and K. Bier, Zeits. f. Naturforschung 9a, 975, 1954 30. P.B. Moon, Brit. Journ. Appl. Phys;.i 97; 1953 31.
I. Amdur, J. Chern. Phys .
.!..!.'
157, 1943
32. F. M. Devienne and J. Souquet, in Rarefied Gas Dynamics (L. Talbot, ed) p. 83, Academic Press, New York, 1961.
300
B
au Fig I· Molecular Beam Apparatus (schematic).
Fig. 2· The Cosine Law.
p
Fig. 3 • Intensity Calculation of Molecular Beams.
o
C
Fig 4 - Beam Formation by Two Silts
Fig. 5· Ideallntensily Distribution.
301
Incident beam,'
Beam room
I
, ,,
ID.!'
F
I
" l,"'"
", ")
'OI.~ O~.::...J..-e,-:.-'1....
- - - - - 35 c m - - - -
Fig. 7 - Reflection from Crystal Surfaces.
Fig. 6 - Apparatus with Differential Pumping.
NORMAL tOO
,,
Plane 01/ test surlace
'--",,,",,::.J...~
C
8
"L" •
..., 5cm I
~
"';B
NORMAL 80
60
la
lb
THe
25°C
THe
•
TpI
it 80·e
TPI =
•
Tpi
25°C
•
Tpi
= 1870°C !080·e
= 25°C
Fig. 8 - Scattering of He from Pt (Datzj.
302
Fig. 9 . Scattering of H2 from W(Smith and Fite).
Fig. 10 Air molecules scattered from glass surface. Inclination angle 25.11·, azimuth angle 0.00· o Feb•. IS, 1953, /:). Febr. 17, 1953. Inclination angle 25.11·, azimuth angle 19.57°. V Febr. 21, 1953.
Fig. II . Cosine Scattering.
303
12
II
10
8
9
7
5
6
o
2
3
4
Transit time (proportional to l/v) Fig. 12 - Velocity Distribution in Molecular Beams.
TUNGSTEN REFLECTOR
IZO
110
100
90
80
70
60
50
40
30
20
Fig. 13 - Velocity Distribution in a Reflected Beam; Source temperature 530oK. Refl ector TE'mpf'J ..1lure ,)J)I, <'~.
304
10
o
EXPERIMENTAL METHODS IN RAREFIED GAS DYNAMICS by 1. Estermann
1.
Definitions For the purpose of the following discussion, we define as rare-
fied gas dynamics a flow regime where the pressure of the air and its constituents is so low that conventional aerodynamics in which a gas is treated as a continuum fluid is no longer applicable. For explicit characterization of this regime, we use a dimensionless number, call· oed the Knudsen number, K
= >.. I d, in which>... is the mean free path of
the molecules of the gas at the particular altitude or pressure, and d a "'-
characteristic dimension of the moving object.
The latter may be the
diameter of a sphere, the radius of curvature of the leading edge of an airfoil, or in some cases, the thickness of a boundary layer. The mean free path is defined in the kinetic theory of gases as the average Ie; ;.;; it
rf the path travelled by a molecule between successive collisions,
lL,J~~lds
on the nature of the gas molecules, expressed by a collision
cross section cr, and is inversely proportional to the number of particles per unit volume. Since this quantity is , at constant temperature, proportional to the pressure, the mean free path is normally inversely proportional to the pressure .
..
The Knudsen number is related to two more conventional aerodynamic parameters, the Mach number M and the Reynolds Number Re, by the relation K=t MIRe for small values of Re and K:1MI Vife for values of Re large enough to indicate the existence of a boundary layer.
305
- 292 I. Estermann
The Knudsen number as defined in the preceding paragraph permits a division of aerodynamics into several flow regimes (1). For small values of K, conventional fluid mechanics is applicable, while for large values, the so-called free molecular flow conditions prevail. The limits of these flow regimes are somewhat arbitrary, but it is custo\'\')- ary to set the upper limit of the Knudsen number for continuum flow at O. 01 to
O. 05 and the lower limit for free molecular flow at 5 or 10.
The region between .these limits is generally divided into two parts, with that
or KO.l the if.
transition flow regime.
Slip flow is sometimes treated by applying
corrections to the equations of continuum mechanics, but transition and free molecular flow require a different approach which has led to new branch of science called rarefied gas dynamics. The current interest in this new scientific area has been stimulated by missile and satellite technology, which is concerned with flight conditions in the upper atmosphere. Because of the exponential decrease of the atmospheric pressure with altitude, the mean free path of the air molecules increases rapidly with altitude as shown in Fig. 1. It should be noted that the mean free path at sea level is of the order of -6 6 x 10 cm, but that at altitudes above 100 km, the mean free path becomes comparable to the dimensions of man-made objects which may be sent into or through these altitudes and whose aerodynamic behavior is of interest. In· Fig. 2, we show
the Knudsen
~
number relating to a
The boundaries between these regimes are not only somewhat undefined, but depend also on other parameters. It was recently shown, e. g., that at M = 2, a Knudsen number of 5 will be sufficient to provide free molecular flow conditions, while at M = 5, K values as high as 30 or 40 will be required.
306
- 293 1. Estermann
length of 1 cm as function of altitude and a division of atmospheric conditions into the various flow regimes. (In this figure, slip and transition flow are treated together.) It should also be noted that the composition of the atmosphere changes with altitude, and that in the higher regions, the ionosphere, electrically charged particles are also present. A good approach to the flow problems in the upper atmosphere can be made by the design and development of laboratory tools which permit the simulation of some of the most importatlt factors. The combination of the partial results obtainable with these tools with one another and with theoretical considerations can go a long way towards the solution of these problems. The most important of these tools are lowdensity wind tunnels, revolving arms, discs and cylinders, and molecular beam arrangements. Representative examples of these categories will be discussed in the following sections of this paper, as well as some of the important measuring techniques involved. 2.
Low-Density Wind Tunnels These instruments permit the reproduction of flow patterns over
models of tractable size in the slip and transition flow regimes and with some strain, into the beginnings of free molecular flow. The first tunnels of this kind were buill at the Ames Laboratory of the U. S. NASA at Moffet field (2) and the University of California in Berkeley (3) about 1947. .A few years later" a similar installation was started in Toronto,Ca•• -ada, and in the last few years, a number of other insitutions on both sides of the Atlantic began to construct similar facilities. The writer became associated with the Berkeley project in 1947 and will, therefore, use the Berkeley Wind Tunnel as a general example. Details of other tunnels which differ in important features, concepts or specifitations
307
- 294 -
1. Estermann
will be given later. 2.1
The Tunnel at the University of California, Berkeley A schematic drawing of this tunnel, which is very similar to the
Ames tunnel, is shown in Fig. 3. The major components of this, as of other conventional wind tunnels, are a gas supply, a settling chamber a nozzle, an observation section, a pumping system, and control equiDmenL The difference from normal pressure tunnels lies in the operat· -jng pressure in the test chamber, which for Berkeley is from 50 to lOOr Hg. No diffuser is used, as attempts to obtain pressure recovery did not have promising results. The gas supply consists of either atmospheric air or compressed gas cylinders (nitroEien or other gases) , reducing valves, dryers and control valves, all of conventional design. The settling chaml:)er is a cylindrical steel vessel, about 2 m long with a diameter of about 1 m; it is equipped with a side arm of pyrex glass which permits partial optical excitation
of the gas for the purpose of flow visualization (see
section 5. 1). It is separated from the observation section by a steel plate into which various expansion nozzles for different Mach numbers can be inserted. These nozzles are axially symmetric and are made of a plastic material, typical exit diameters are 9" for a M = 2 nozzle and 5" for M = 6 nozzle. The observation chamber is equipped with a three, dimensional traverse system which is driven by electric - motors; the position of models, probes, etc., attached to it can be read on Root- Veeder counters. Large plate glass windows permit optical and photographic observations . The most interesting and unique features of low denSity wind tunnels are the pumping systems. The desired test section pressure, noz-
308
- 295 -
1. Estermann
zle size, and Mach number together determine the volumetric capacity (pumping speed) and the equilibrium pressure requirements of the pump-i.ng system. It is obvious that compromises must be made in order to remain within acceptable cost limitations. The Berkeley and Ames tunnels use a mUltistage steam ejector system, consisting of three stages operating from the tunnel into a condensor where most of the steam is removed, and two stages, which compress the air to atmospheric pressure and exhaust., it together with the remainder of the steam. Fig. 4 shows a typical ;, stage ejector. The compression ratio per stage is approximately 1 to 10, permtting an ultimate vacuum of about 10 ~ Hg. This pressure, which is much lower than the vapor pressure of water at the ambient temperature, can be reached because the low pressure stage contains a supersonic section. The steam requirements are about 700'0 lbs /hour at 150 psi for removai of 60 Ibs/hr of air. The volumetric pumping capacity varies; it amount to 60,000 liters/sec at 100~Hg, and about 30,000 l/sec at 50~Hg, permitting the use of a M = 6 nozzle of 5" diameter. The mean free path at
50~
is approximately 1 mm, which is not quite enough for the establi
shment of free molecular flow conditions over models of reasonable size. As a recent addition, the tunnel has been equipped with an electric heater upstream of the nozzle; this permits reaching stagnation temperatures of about 1000 OK. In another modification, a plasma jet heater has been added (Fig. 5) . With this equipment, stagnation tempe ... -atures of 5000 0 K have been obtained at M = 6 and static pressure of 100~
Hg. Under these conditions, the gas leaving the nozzle is partly io-
nized (about 0.2'.) .
309
- 296 1. Estermann
2. 2
The tunnels at Royal the University of Toronto Institute of Aerodynamics, Armament Research and Development Establishment, Fort Halstead, and National Physical Laboratory, Teddington These tunnels are in many respects similar to the Berkeley tun-
nel. The main difference is in the pumping system. UTIA (4) and RARDE (5) use 6 three-stage oil diffusion booster pumps of the type shown in Fig. 6 , giving a total pumping speed of 70001/ sec. The pumps are backed by two 200 cu
ft/ min
mechanical pumps. The operating pres-
sure range is from about 10 to 70
~
Hg. NPL has just completed a so- ..
mewhat larger facility of the same general design, using five four-stage booster pumps with a total pumping speed of20,OPO 1/ sec, backed by 5 Roots blowers and 2 mechanical pumps. The test section pressure ran,· -ed from 10 to 10014' Hg, nozzles for Mach numbers from 2 to 10 are contemplated. The complete arrangement of the NPL tunnel is shown in Fig. 7. 2.3
The Tunnel at the Laboratoire Mediterraneen, Nice The first tunnel for pressures from about 0.1 to
10~Hg
was de-
signed by Devienne (7) at the Laboratoire Mediterraneen in Nice. The pumping system consists of
4 three-stage oil diffusion pumps with a
total capacity of 30,0001/ sec at 0,1 I""Hg, backed by booster and me chanical pumps. It has been reported that Mach numbers of
4 have
been obtained with a nozzle diameter of 4 cm, producing a uniform flow over a core of about 1 cm diameter . A schematic view of the installation is given in Fig. 8. This tunnel did not produce the desired results and is no longer in operation. 2.4. The Two-phase Tunnel at the University of Southern California Engineering Center, los Angeles (8)
310
- 297 I. Estermann
In this tunnel, a novel and radically different pumping system, na., mely the condensation of the tunnel air on a surface cooled to T
< 20 ~
(" cryopumping ") is applied. If every gas molecule striking such a surface is condensed or trapped, each cm
2
provides an equivalent pumping
speed of 10 literl sec. It is, therefore, feasible to reach pumping spe~ eds from 10 5 to 10 6 liter/sec at pressures of 1r Hg or less in the test section. The condenser used in the USC tunnel has six plates, each having an area of 10 ft 2, which are cooled to about 15-20 OJ( by means of helium gas precooled in a helium refrigerator placed outside the tunnel. The latter is a Collins type cryostat employing two reciprocating expansion engines capable of removi ng 350 watts at 20 oK, with a rated power consumption of 50 hp. The tunnel itself consists of a large steel tank
measuring approximately
9 fe et in diameter and 35 feet in
length , with a sump-like appendage at the downstream end housing the condenser (Fig. 9 ). Before entering the nozzle, the nitrogen gas used for tunnel operation is heated by means of a graphite heater to 800 oK; it then passes through a large settling chamber before entering the nozzle (Fig. 10) . The conical nozzle has an angle of 40° ; to exercise control over the boundary laver developing inside the nozzle with high Mach numbers at low pressures, the nozzle walls can be cooled with liquid nitrogen. After about 10 hours of operation,the thickness of the solilj nitrogen deposit on the condenser plates becomes so large that thawing becomes necessary. The nozzle has an exit diameter of 19" and is 12-1/2 11 long, giving a Mach number of 8. It is interesting to note that only a core of 2" diameter is filled with uniform flow (See section 2.5). 2. 5
General Comments on Low Density Wind Tunnels Low density wind tunnels as described
311
in the preceding sec-
- 298 1. Estermann
tion have made it possible to study the basic flow and heat transfer prob-lems in the slip and transition range, with a few stabs into the direction of free molecular flow, for flows up to about M:: 8 . ( Fig. 11 shows the flow regions attainable with the Berkeley tunnel.) The application of plasmajets has made it possible to begin simulation of the temperature and ionization conditions of high altitude flight. Extension of this work into the free molecular flow regime will require much larger installations than those presently in existence or in the construction stage. These, however, do not appear to be impossible. Diffusion pumns with speeds of 50,000 l/sec (32" diameter, 4 stage) are becoming available, and it is to be hoped that cryopumping will be developed sufficiently to permit reliable and continuing performance below the
If Hg pressure ran-
ge. One of the inherent difficulties is the growth of the boundary layer thickness inside a nozzle with decreasing pressure, which reduces the diameter of the uniform core of the airstream flowing from the nozzle to a small part of the nozzle diameter (as shown in Fig. 12 for the UTIA tunnel) . Attempts have been made to reduce the boundary layer thickness by cooling of or by suction through slots in the nozzle walls, but
it ap-
pears certain that complete exploration of the free molecular flow regime requires additional equipments. Another fundamental problem concerning simulation of flow conditions by wind tunnels refers to actual velocity of the air stream with respect to the model. The high Mach numbers in super and hypersonic tunnels do not necessarily indicate high speeds of the air flow, but only a large ratio between this speed and the local velocity of sound. The latter, however, is proportional to the square root of the absolute temper_ -ature. The expansion of air through a nozzle, being largely isentropic,
312
- 299 -
1. Estermann
produces a very strong cooling of the effluent gas, and a large part of the high Mach numbers produced is due to the corresponding reduction of the speed of sound and not to the acceleration of the gas flow. Thip effect can be compensated, at least in part, by the installation of plas-roajet or other heaters upstream of the nozzle, but together with the growth of boundary layer thickness mentioned earlier, it severely limits the usefulness of wind tunnels for the simulation of high altitude flight conditions to relatively low speeds and altitudes. In the following paragraphs, we shall give examples of other equipments which are potentially useful for the solution of this problem for higher speeds and altitu~es.
3.
The "Molecular Gun" of the Laboratoire Mediterran~en (9) This newly installed instrument is a combination of a molecular
jet and a wind tunnel and has as its objective the simulation of the intelt.actions between a body moving at a very high speed in a rarefied gas in the free molecular flow regime. While a " normal" molecular beam employs molecules of thermal velocities, which are approximately equal to the speed of sound, this apparatus produces particles of much higher speeds by acceleration of electrically charged particles and their subsequent neutralization. The general design of the apparatus is shown in Fig. 13. The experimental gas, e. g., argon, is admitted through a control valve into a supersonic nozzle where a molecular jet is formed. From there, it passes through a quartz tube, where it is partly ionized by a high-frequency discharge, into the first vacuum chamber. The resulting mixture of electrons, positive and negative ions, and neutral atoms is passed through an electrostatic accelerator and lens system from which a reasonably homogeneous positive ion beam emerges. This
313
- 300 -
1. Estermann
beam is deflected through an angle of 90 0 and thereby separated from the neutral and negative particles by means of an electromagnet. A sell-ond electrostatic lens and
decelerator system refoc1.\l>ses the positive
ions and reduces their velocity to the desired value, corresponding to el1\-ergies in the range of 20 to 100 ev.
The ion beam is then intersected
by a second beam of neutral argon atoms. Since the effective cross section for charge exchange is much larger than that for momentum tranl-fer, a fair portion of the ions will be neutralized without ani appreciable change of velocity. A second magnetic field removes the remai,,· ~ing
ions, and the final beam entering the test section consists of near--
ly mono-energetic neutral atoms. Beam densities of 3 x 10 6 molecules per square cm per sec have been reported and their application to aeIP-odynamic measurements is in prospect. Several modifications of this equipment have been constructed recently and have been described at the Fourth International Conference on Rarefied Gas Dynamics in Toronto. 4.
Revolving Arms, Discs, and Cylinders (10) These instruments have been built in various laboratories and are
useful for aerodynamic measurements in the free molecular flow regime under certain precautions . Their main advantage is that there is no fundamental limitation to the Knudsen number which can be obtained since they are, from a vacuum technology standpoint, static systems which can be pumped down without much effort to pressures of 10 and below,
r Hg
-2
providing mean free path lengths of the order of meters.
Their main drawback is their inability to reach high linear velocities , 800 m/ sec being about the top speed that has been reached in actual use. Revolving cylinders are useful for drag and heat transfer measurements in rarefied gases, and revolving arms and discs allow the investigation
314
- 301 -
1. Estermann
of flow patterns over models in various media including ionized gases in the free molecular flow regime. As an example, we show the revolll-ing arm of the Laboratoire Mediterraneen de Recherches Thermodynamiques (Fig. 14), (11) which is arranged for operation at pressure ley.,.els
between O. 25 and 5
r- Hg.
The vacuum tank has a diameter of 1. 5
m and can be evacuated by means of an oil diffusion pump. The arm, made of high-tensile strength duraluminum, has a diameter of 1.25 m and is driven by an external motor at speeds up to 9000 rpm. Ionization up to 10" is obtained by means of an electrodeless high-frequency discharge in a quartz or pyrex tube attached to the top of the tank. Gas is admitted through a controlled leak which also provides pressure regulation. The arm may be used to carry models for aerodynamic tests or Langmuir probes for the exploration of charge exchange. 5.
Measuring Techniques The aerodynamic quantities which are of interest in rarefied gas
dynamics are fundamentally the same elementary physical quantities, such as pressure, force, density and heat transfer rates, etc., which are measured in conventional wind tunnels. Since their magnitudes, however" are much lower than in the conventional case, special techniques and instrumentation are required. Moreover, the interpretation of the measured quantities in terms of the desired information is not always as direct as in the case of higher pressures. In the following sections, we shall give a few examples of the most important procedures which are currently in use. 5.1
Flow Visualization (12,13,14) At normal densities, the various techniques for flow visualization
315
- 302 -
1. Estermann
have been extremely valuable. The schlieren method provides a survey of the flow pattern and indicates the areas where more detailed measurements are required . The interferometer methods allow the determina,tion of density distribution, frequently with a high degree of accuracy , over large parts of the flow. With diminished gas density, however, the optical density becomes so low that these methods fail. On the other hand, the increasing mean free path and lifetime of optically excited molecules make it possible to use afterglow phenomena for flow visualization. In the technique used at Berkeley and elsewhere (15, 16) the gas is admitted through the side tube shown in Fig. 3, where it is excited in an electrodeless discharge. The excited stream moves through the nozzle into the test section where a chemical reaction produces lutpinescen(t. Of various afterglows which have been used, the airglow has been found most useful.
It is caused by the reaction
for which the 0 atoms are produced by O2 dissociation in the discharge tube and the NO molecules by collisions between these atoms and N atoms which are also produced in the discharge. The glow is enhanced if NO is introduced upstream of the nozzle as shown in Fig. 3 . A representative example of the results attainable by this method is shown in Fig. 15. If a plasmajet is used, no further excitation is required since the gas stream is hot enough to become luminescent by temperature excitation. Attempts to visualize low density flow patterns through light absorption in the U. V. have met with only limited sucess (13,15). 5.2
Force (2) Under conditions existing in most low density wind tunnels,
316
- 303 -
1. Estermann aerodynamic pressures are of the order of gr / cm 2 and their measurement requires more sensitive balances than are usually employed. If one wants to reach the free molecular flow regime, targets should have dimensions of the order of A, i. e., approximately 1 mm or less. As a result forces of the order of milligrams will have to be measured. A small torsion balance capable of the proper sensitivity was used in Berkeley (17) for this purpose. It consisted of a tungsten torsion wire supported by a movable frame which carried a quartz fiber to which the target plate was attached . The free end of this fiber served as a
poin~
-er for indicating the angular twist of the wire. The support fiber was protected from the air stream by a shield which also restricted its motion to a small deviation from the vertical. Fqrces exerted by the gas flow on the target plate were compensated by applying a twist on the torsion wire. The assembly of the balance is shown in Fig. 16, a re,-resentative calibration curve in Fig. 17. For the proper interpretation of the results, knowledge of the accommodation coefficients is necessary. 5.3
Density The most direct approach to the measurement of this quantity is
the attenuation in beams of photons or particles according to the diffell!-ential equation. dI/dX = - P r X where I is the intensity of the beam at the point X inside the gas stream,
t'
the mass absorption coefficient, and p the local density.
None of the gases used in wind tunnels have a suitable absorption coefficient in the visible, and as mentioned before, attempts to use the absorption band of oxygen in the vacuum ultraviolet have only been partly successful. Better results have been obtained with electron beams
317
- 304 -
1. Estermann
of about air is
10 kV energy (18) for which the mass absorption coefficient in 7.4 x 10
5
2
cm / g . It has also been shown that the differential
equation listed above can be integrated for the traverse of an axisymetric air stream. In Fig. 18, we show a schematic drawing of the apparatus, in Fig. 19 the results of a traverse along the axis of the supersonic air flow around a sphere from a point upstream to the stagnation point. It can be seen that the electron current at the collector decreases sharply when the shock wave region is reached, then more slowly to the stagnation point of the model. One
wo~ld
expect this behavior since the
beam must traverse an ever-increasing thickness of relatively dense gas as it moves along the stagnation line. The beam intensity drops sharply to zero as the model intercepts the beam. It appears possible to use the atterruation of a soft X-ray beam, for which
r-:: 10 3 cm 2/ g,
in a simi-
lar way. In another variation the intensity of the light emitted by the atoms which are excited by the electron beam (19) is used for local density determinations. 5.4
Pressure The main difference between static pressure measurement at nor-
mal and low densities lies less in pressure-measuring instruments, which differ mainly in their sensitivity, than in certain peculiar properties of a low pressure system. Among the factors to be considered are the response time of the measuring system (20) , thermal transpiration, and adsorption and desorption of gases. A more fundamental difficulty affec.-ing dynamic measurements arises from viscous effects which require the application of substantial corrections to both static pressure and to impact pressure measurements.
318
- 305 -
1. Estermann
5.5
Free Molecule Probes These instruments have a limited value for measurements in the
free molecular regime. These probes must be so designed that their characteristic dimensions are small compared to the mean free path in the gas flow. For the usual test section pressures of 1 - 100
r Hg,
this
requirement restricts the diameter of a probe to less than one mm. The most important probes of this kind are pressure orifice, (21) temperature, (22) and heat transfer (23) probes. The first type are made from hyperdermic needle tubing, having a hole in the side, which is covered with a very thin foil through which a very small hole has been pierced (Fig. 20) . For equilibrium conditions, the number of molecules emerging from the probe per unit time, 1/4 nc, is equal to the number entering. For a measurement of the molecular speed ratio S which is for all practical purposes approximately equal to the local Mach number, the probe is arranged in the gas stream in three positions as shown in Fig. 21. The molecular speed ratio is then p S=
- pi 0
2f\Y Ps
o=~
C
where Po' plo and ps are the measured equilibrium pressures when the orif'ice faces into the mass flow, away from the flow, and perpendicular to the flow,
v
the speed of the mass flow and
c the mean
mo-
lecular. The thermal probes consist of a thin wire which can be easily so dimensioned that the Knudsen number is large with respect to its diameter.
The
theory shows that a circular cylinder which is
a perfect heat conductor and is protected from heat conduction and radiation losses and is
placed with its
319
axis
perpendicu-
~
306
~
I. Estermann
lar to a uniform stream of gas, is heated to an equilibrium temperature mich is a function of only the Mach number , the stagnation tempe,. -'lture, and the specific heat of the gas. These equilibrium temperatures become insensitive to changes in flow velocity at Mach numbers of about 2; beyond this range, up to about M = 3, the heat transfer chara«-teristics of the probe whith can be measured with the same instrument are more responsive to Mach number changes. Figs. 22 and 23 illustrate this effect. 6.
Summary and Conclusion While complete simulation of high altitude flight conditions may be
considered as impractical, many of the important factors can be reproduced with reasonable approximation in the laboratory. Rarefied gas flow over models can be examined in low density wind tunnels for subsonic, supersonic and low hypersonic speeds (up to Mach number 8) in the slip and transition flow regimes, and the formation and structure of shock waves may be made visible by various techniques. These wind tunnels are also useful for the determination of aerodynamic forces and heat transfer characteristics. Direct extension of these investigations into the free molecular flow regime is still in an esploratory stage, but a good start has already been made. Where direct experimental me .. thods are not yet available, useful results can frequently be obtained by combining theoretical calculation based on kinetic theory with expelP-imental data obtained from the application of molecular beam and other indirect methods.
320
- 307 -
1. Estermann Bibliography ~,
L
H. S. Tsien, Journ. Aero. Sci.
2.
J.R.Stalder, in Rarefied Gas Dynamics, First Symposium, (F.M. Devienne, ed), p.1, Pergamon Press, London, 1960 (Review)
3.
E. D. Kane and R. G. Folsom Jour.Aero. Sci., l.§., 46, 1949
4.
K. R. Enkenhus, UTlA Report No ; 44, 1957
5.
W.A. Clayden, in Rarefied Gas Dynamics, First Symposium, (F. M. Devienne, ed) p. 21, Pergamon Press, London, 1960
6.
D. W. Holder and L. Bernstein, Symposium on User Experi( nce of Large Scale Industrial Vacuum Plants; Paper 8, lnst. of Mech. Eng. London, 1961
7.
F. M. Devienne, G. M. Forestier, and A. F. Roustan, Laboratoire Mediterran~en de Recherches Thermodynamiques, Report, 1958
8.
R. L. Chuan and K. Krishnamurty, Univ : of South. Calif. Engineering Center Research Report No 42-201, 1955
9.
F. M. Devienne and J. Souquet, in Rarefied Gas Dynamics, 2nd Symposium (L. Talbot, ed) p. 83, Academic Press, New York, 1961
10.
A.R. Kuhlthau, in Rarefied Gas Dynamics, First Symposium (F.M. Devienne, ed) p. 192, Pergamon Press, London, 1960 (Review)
11.
F. M. Devienne and B. C. Crave, Lab. Med. Rech. Therm. Techn. Note 1961
12.
W. B. Kunkel and F. C. Hurlbut, Jour. Appl. Phys. 28, 827, 1957
13.
1. Estermann, in Rarefied Gas Dynamics; First Symposium; (F. M.
653, 1946
Devienne, ed) p. 38, Pergamon Press, London, 1960 (Review) 14.
F. C. Hurlbut, ibid, p. 55 (Review)
15.
R. A. Evans, Jour. Appl. Phys. 28, 1005, 1957
16.
P. M. Sherman, Jour. Aero. Sci. 24, 93, 1957
17.
I. Estermann and E. D. Kane, Jour. Appl. Phys. 60, 608, 1949
18.
F. C. Hurlbut, WADC Techn. Rep. 57-644, 1957
19.
E. O. Gadamer, UTlA Report No 83, 1962
20.
S.A. Schaaf and R.R. Cyr, Jour. Appl. Phys. 20,
321
~60,
1949
- 308 -
1. Estermann
21.
K. R. Enkenhus, UTIA Report No 43, 1957
22.
H. Wong, Univ. of Calif. Eng. Res. Proj. Rep. HE-150-143, 1956
23.
J.A. Laurmann and D. C. Ipsen, WADC Tech. Rep. 57-440, 1957
322
A/Illude
10'
r-i
l
111--I
la'
V
I
! A'
A
./
!~ I
/ 10'
I
I
I
I
,/
!
!
+-H I
:
II
/
I
j
I 10'
10 1
la'
PartIcle mean Iree palh
Fig. 1. Mean free path as function of altitude.
A.~dud~
1+0
I \
120
I
I '\+---+---'----1--.-""."I°O -t-I_-+-_ - - H --i
ree Molecul
IFlow
I
100 I-t---+--*--+---+--~-+-~
I
.O~--;
I
___--+____~~
Conventlonol
I
I
2O ~FrTr o C--.J._--'-_ _ -+
. - - - - -___
-2
LOQIO~ M F'iQ. 2 • Al ti tude variation of aerodynamic parameters .
323
II
nagnatlM rtglon
Photo rtlt
Fig.
I ""!It= =from supply olldt
V
nltfIC
C, mpra Bpllows
~
~
Pyre I pipe and omllator
"-
3 Schematic of wind tunnel of the University of California, Berkeley.
(Dolin g wat"
our
I
Exhaust stac' and /luff{tr
SubsOniC srctum
(K'f'" BI '
SuptfJonJr urtlon
{Rtglon AJ
~.
W,tp r tubp Inttr rondrnur
6"
Inlt' .....
Stum
Fig. " Three-stage steam ejector system.
-... Wo'~r Row - -- Gas flow
Fig. 5 Plasma generator (schematic diagram).
324
6'$
In
I
Thermal Insulation
Immersion heater unit Fig.
6 Booster
pump.
Fig, 7 . Low·density wind tunnel at National Physical Laboratory, Teddington (Crown copyrigh t · reproduced wi th permissi on of H.M. Government),
Buty' (Jllta/att manomttrr
Fig.
8
Vo lt'm.!rl' mtt"
Schematic view of the low-density wind tunnel of the Laboratoire Mcditerraneen in Nice.
325
; - - - - 2711 - - --'
Fig. 9 Schematic layour of the low-density wind tunnel at the University of Southern California, Los Angeles. 1 = jack, 2 = stainless bellows seal, 3 = modulating valve 10" travel, 4 = floor line, 5 = cryostat, 6 = condenser, 7 = plaIt coils, 8 = heater plug in, 9 = aftereooler, 10 = access PO", 11 = probe mount, 12 = nozzle, 13 = valve.
Fig. 10 Nozzle assembly of the University of Southern California wind tunnel. 1 = probe mount, 2 = nozzle, 3 = stagnation chamber, 4 = valve.
o~
__~~~~~~~~~~
10- 1
10' Reynoids number leading egde lIat piale Base pressure Cone pressure Cone drag
Fig. 11 Mach and Reynolds number range, wind tunnel in Berkeley. Characleristic dimension = 1 ft.
326
• Olslance off axis
Fig. 12 Mach number profiles at exit plane at various operating pressures of the Toronto wind tunnel. Operating static pressure [I'Hg]
o
45 41.1 40.0 34.8 28.8
t::. 'V
!
Fig. 13 Molecular gun at the Laboratoire Mediterraneen. 1 = argon intake, 2 = nozzle, 3 = RF coil, 4 = accelerator, 5 = magnet, 6 = valve, 7 = pump, 8 = decelerator, 9 = magnet, 10 = observation chamber, 11 = detector.
6 o
0
10
Fig.14 Revolving arm at the Laboratoire Mediterraneen. 1 = quartz tube, 2 = RF coil, 3 = pyrex tube, 4 = model, S - JeYOlving arm, 6 = probes, 7 = gas inlet, 8 = pump, 9 = shaft, 10 = seal,
327
Fig. 15 - flow Visualization by Nitrogen Afterglow.
Fig. 16 - Torsion Balance for Low Density Hind Tunnel.
328
06 /
fltilll-
;I
gralllmS
/ /
02
o~
o
__ ____ ____ __ ____ ~
~
10'
40'
~
60'
~
80'
~
100'
Angle 01 twist
Fig 17 Null balance calibration. Tungsten torsion fiber diameter = 0.00064 in. increasing weights, • decreasing weights.
o
._!_ . ,
-- it1t-----jj,I~:,-=-_1l1
n
Fig. 18 Smematic drawing of the electron beam apparatus. A = tuppons. B = cathode, C = heat shield, D = leads, E = anode, F = defining orifice, G and H = aperture plates, I = electron collector.
10
~
- ' -- - 1
I
x ~
~
I
I
I
:
I
~
I
08
C>.
I; ~
~
c
0.6
~ ~
" ~ '"
a
0.4 02 o~~
o
__~__~~~~
86100 86200 86300 86400 86500 Allat position coordinates Ilast figuft ~D.101 inch}
Fig. 19 Density distriburioo ill froot of a sphere.
329
Fig. 20 Orifice probe. 1 = orifice, 2 = cylinder, 3 = end sealed with drop of de KhociaIky cement, 4 = file cut, 5 = 0.00031" thick hard aluminium foil cemented to cylinder.
~Q~---t---t---t---+---+~~---+--~---4---4---4
o
ID
4
Fig. 21 Orientation of orifice probe. • INCHES F"ROM LEAOING EDGE
Fiq. 22 - Lines of constant teD)pel'oture raUo in the vicinity of a wedge.
~~--~r-~~~~~~~'---'-r---+~~I~---+---t---4
o
'"
;0
",Qj---4---4---4---4---4~~~~---4---4~-4~--+
> o
ID
'Q~----~--+---+---~~~~~~~~,i'&II----+---+---~
::: I
u
~Ol~~~~~~~~-T-r~
I
INCHES FROM LEADING EDGE
Fig. 23 - Lines of constant heat transfer ratio in the vicinitY of a wedge.
330
CENTRO INTERNAZIONALE MATEMATICO ESTIVO
SILVIO NOCILLA
SULL'INTERAZIONE TRA FLUSSI DI MOLECOLE LIBERE E SUPERFICIE RIGIDE
Corso tenuto a Varenna (Como) dal 21 al 29 agosto 1964
331
SULL'INTERAZIONE TRA FLUSSI DI MOLECOLE LIBERE E SUPERFICIE RIGIDE di SILVIO NOCILLA (Scuola di Ingegneria Aerospaziale - Politecnico di Torino) 1- Interazione superfidale ed adsorbimento: noz;ioni introduttive. Consideriamo un missile mobile negli alti strati dell'atmosfera terrestre, a quote variabili tra i 150 e i 300 Km circa, doe dove esiste un altissimo grado di rarefazione dell'aria con densite. dell'ordine di 10- 8,
10 -1 0 volte minori di quelle normali, Sappiamo dalla teoria cinetica dei gas che illibero cammino medio
A
delle molecole costituenti l'aria
e da-
to dalla formula: (1. 1)
- 1- .•
V21T
con: m = massa di una molecola
f
= densite.
T = temperatura assoluta KS= costante di Sutherland "j 110 0 K ~ = sezione efficace per gli urti molecolari (diametro di a:z;ione di
una molecola, tale doe che si ha un urto quando la distanza tra i centri di due molecole ~ minore 0 eguale a
f
be),
indipendente da
e da T.
In base alla (1. 1) ed alla cleterminazione di ~ con proceciimen e ti che qui non inciichiamo, si hanno per l'atmosfera tipo internazionale i s~ guenti valori, validi all'equatore (v. Nobile [1
333
J ; parte II):
- 314 S,. Nocilla
quota in km 0 50
pin ~di
Hg(x)
A in
m
T in oK
760.10 3
0,000, 000. 5
293
733
0,000.914
350
100
2, 74
0,021
302
150
0,0856
1,36
528
200
0,0108
17
782
250
0,0025
100
1037
goo
0,0008
392
1291
(X)l~ldi Hg = 10 -3 mm di Hg = 10 -3 torr = 1,3157 10 -6 atm Poiche i1 missile,
0
satellite artificiale che consideriamo, ha
dimensioni d dell'ordine del metro', il numer6 di Knudsen: (1. 2)
K = n
AId
assumendo, tanto per fissare Ie idee, proprio d = 1 m, viene ad avere gli stessi valori numerici indicati nella colonna dei liberi cammini medi
A,
variabili cioe da 1 a 400 circa per quote da 150 a 300 Km. Cia porta alIa fondamenta1e conseguenza, posta a base dello studio delle correnti di mole cole libere, e che ne giustifica la
denominazi~
ne, che agli effetti della interazione superficiale tra molecole e solido si trascurano gli urti tra molecole e molecole, mentre invece si tiene conto degli urti tra molecole e superficie. Ne consegue che i metodi classici del la meccanica dei fluidi, fondati sull' assul'lzione che ogni elemento di fluido di dimensioni piccole fin che si vuole rispetto all'ostacolo contel'lga un numere enormemente grande di molecole urtantesi continuamente S1 da costi tuire un gas soddisfacente alle consuete leggi termo-gas-dinamiche, non sono piu applicabili nelle condizioni presenti.
E' qui necessario impiega-
re dei procedimenti di studio del tutto diversi. Per orientarsi verso questi 334
- 315 -
S. Nocilla
procedimenti si immagini in un primo momento di seguire nel suo movimento una molecola del gas nel quale il missile si muove. Innanzitutto
e bene osse!:.
yare che nonostante il valore del numero di Knudsen enormemente piu
elev~
to rispetto a quello dei gas in condizioni normali di temperatura e pressione, e pur sempre lecito pensare che il gas rarefatto considerato sia in equilibrio Maxwelliano, perc he tale equilibrio e venuto a stabilirsi nel corso di secoH e millenni, indipendentemente dalla presenza
0
meno del satellite artificiale.
Supponiamo poi che la molecola presa in considerazione venga ad urtare i1 missile, e chiediamoci cosa puo capitare di essa dopo l'urto. Evidentemente una penetrazione profonda della molecola nel corpo non e possibile perche questa e allo stato solido. D'altra parte la fisica e la chimica hanno dimostrato 1'esistenza sulla superficie dei solidi di campi di forza superficiali che se pur non ancora conosciuti in tutti i dettaglr ci permettono tuttavia di farci un'idea per 10 meno di prima approssimazione delle condizioni ambie!! tali fisico-chimiche sulla superficie. Non e il caso qui di indagare a fondo su queste condizioni, per Ie quali oltre tutto sarebbe necessaria la competenza di un fisico-chimico; basti soltanto segnalare fondamentalmente Ie fo!:. ze di Van der Waals che Ie molecole costituenti il solido si scambiano tra di loro e scambiano con Ie molecole del gas rarefatto incidenti su di esso. In virtu di queste condizioni ambientali fisico-chimiche vi e una certa probabilita che la molecola prima presa in considerazione venga "catturata" dalla superficie del solido, e si trovi a soggiornare su di essa. Ma durante questa periodo di soggiorno altre molecole del gas colpiscono la superficie, seguono vicissitudini analoghe alla precedente, e si trovano
COS!
a soggior-
nare tutte insieme sulla stessa superficie del solido. Ora e chiaro che questa continua cattura di molecole da parte della superficie del solido non puo proseguire indefinitamente, perch?! altrimenti il gas incontrato nel suo mov..! 335
- 316 -
S. Nocilla
mento nell'atmosfera rarefatta si verrebbe continuamente a depositare su di essa. Si
e indotti ad ammettere,
e la esperienza 10 conferma, che Ie molecQ
Ie, dopo un certo periodo di soggiorno sulla superficie, vengano da questa
ri~
messe nel gas circostante, e che in condizioni stazionarie, come supponiamo di trovarci, si venga a stabilire un equilibrio statistico tra ·molecole incidenti, molecole che soggiornano, e molecole che vengono riemesse ( 0 rievaporate) dalla superficie.
n fenomeno che abbiamo ora descritto qualitivamente non differisce sostanzialmente dal ben nota fenomeno della "adsorbimento", e pili pr~ priamente dell'adsorbimento fisico,
0
fisisorpzione, a proposito del quale ci
limiteremo, tra la vastissima letteratura esistente, a segnalare il De Boer [2]. Osserviamo subito che, come appare dalla descrizione precedente, nel fenomeno dell 'adsorbimento tre grandezze intervengono anzitutto nel suo equi librio statistico: - il numero di N di molecole che nell'unita di tempo colpiscono l'unita di superficie rigida
x
- il tempo di soggiorno ""(; delle molecole adsorbite sulla superficie
x
- il numero N di molecole adsorbite sull'unita di superficie.
Queste tre grandezze fisiche in condizioni di equilibrio statistico non sono tra di loro indipendenti, rna legate dalla relazione fondamentale:
(1. 3)
Inoltre, se per Ie molecole adsorbite adottiamo 10 schema di Frenkel [3], secondo il quale esse si comportano come degli oscillatori armonici in moto secondo la normale alIa superficie adsorbente, il tempo di soggiorno
336
e dato
- 317 -
S. Nocilla
dalla formula:
(1. 4)
-r;
if*-
e
Q/RT
w
dove Q e il calore di adsorbimento, RIa costante del gas incidente, T
la w temperatura del solido adsorbente e -r:;** un parametro con la dimensione di un tempo, avente una diretta relazione col periodo di vibrazione delle
m~
lecole od atomi costituenti la superficie adsorbente, ed il suo stesso ordine di grandezza, cioe di 10- 12 , 10-1\ec. L'ordine di grandezza del tempo di soggiorno
"'t:
*
e fortemente variabile, a parita. di temperatura, a seconda
del gas adsorbito. Cosi ad esempio (v. De Boer [2] pag. 35) per temper~ tura ambiente esso e dell'ordine di 10-12sec. per l'idrogeno su varie super-10 fici . e di 10 sec. per l'argon, l'ossigeno, l'azoto, il mossido di carbonic su varie superfici (dunque circa mille volte il periodo di oscillazione
-,;*\
Per gas formati da molecole piu pesanti si possono avere tempi di soggiorno di un or dine di grandezza molto maggiore, nno a 10
-2
sec.
Cib premesso possiamo porci la seguente domanda. Visto che la descrizione qualitativa del fenomeno dell'interazione tra flusso di mole cole libere e superficie non differisce sostanzialmente dal fenomeno classico dell'adsorbimento, quali sono gli elementi comuni ai due fenomeni e quali invece differiscono? E' chiaro che nello studio del fenomeno che ci interessa giovera valersi del bagaglio di conoscenze, invero notevole, sia teoriche che sperimentali oggi acquistate sull'adsorbimento, e su di esse inserire i nuovi
pr~.
blemi che il fenomeno aerodinamico pone. Riguardo agli elementi comuni sono da annoverarsi Ie condizioni fisico-chimiche superficiali, come i campi di forza superficiali, Ie costanti geometric he dei reticoli cristallini costitue!! ti il solido, i moti vibratori termici di tali reticoli, Ie proprieta. delle mole-
337
- 318 -
S. Nocilla
cole adsorbite assimilabili, in certe condizioni, ad un gas con struttura bidimensionale, anziche tridimensionale come di solito. Tali condizioni superficiali molto verosimilmente non sono influenzate dalla funzione di distribuzione delle velocita delle molecole incidenti (v. prossimo numerol e quindi dal fatto che la superficie adsorbente sia in quiete oppure in moto rispetto al gas rarefatto circostante. Tra Ie grandezze fisiche che intervengono in entrambi i fenomeni vanno evidentemente annoverate Ie variabili di stato T,
J' ' p del
gas e Ie tre grandezze gia ricordate N, -r;* e N* e la temperatura T della . w superficie solida. Particolare importanza hanno poi in entrambi i fenomeni" anche se in misura diversa nell'uno
0
nell'altro, i seguenti altri elementi, e
cioe la legge di distribuzione spaziale delle molecole riemesse da
o~mento
di superfi-
..
cie e gli scambi di quantita di moto e di energia tra superfi-
normale es lerna a
U= velocH"
dA
del missile
cie del solido e gas, nonche la loro dipendenza in modo
part~
colare dalla temperatura T
w del solido. Viceversa un elemento del tutto nuovo rispetto ai fenomeni classici di adsorbimento
superficie esterna
del missile
e l'influenza suI feno-
-
meno dell' inter azione della velocita U del missile (v. fig. 1), o meglio del suo rapporto s= UIc
Fig. I
con la velocita piu probabile
338
- 319 -
S. Nocilla c = V2RT del gas rarefatto, e della sua direzione rispetto all'elemento superficiale dA di missile che si considera, che caratterizzeremo mediante
~ da esso formato con la normale esterna a dA. Per questa moti-
I'angolo
vo dovremo prendere in considerazione anche Ie seguenti grandezze, tutte funzioni note di s e
~
, come verra precisato nel n. 2:
N. = numero di molecole che nell'unita di tempo colpiscono I'unita di 1
superficie rigida = portata numerica incidente
- ...... Q.
1
= p. n + 'to t = quantiU, di moto complessiva da esse posseduta = 1
1
= portata di quantiU, di moto incidente
E. = energia cinetica complessiva da esse posseduta = portata di ener 1
-
gia cinetica incidente
nonche Ie seguenti altre, a -priori sconosciute, tranne Nr = Ni .. N = numero di molecole che nell'unita di tempo sono riemesse dalla r
unita di superficie = portata numerica riemessa
Qr = -p r rr +L'r "1 = quantita di mota complessiva da esse posseduta = = portata di quantita di mota riemessa
E = energia cinetica complessiva da esse posseduta = portata di ener r
-
gia cinetica riemessa Prima di procedere oltre nello studio dell'interazione sara pero opportuno soffermarsi su alcuni particolari aspetti dei gas in equilibrio maxwel. Hano alle bassissime densita. Nella teoria cinetica dei gas perfetti si assu-
339
- 320 -
S. Nocilla
mono solitamente come variabili di stato la temperatura assoluta, T, la densita
9 e la pressione p,
(1. 5)
tra loro legate dalla equazione di stato:
p =
fRT
Ora di tali tre grandezze Ie prime due mantengono un preciso significato anche se la densita
.f
e piccola fill che
si vuole, purche naturalmente il
num~
ro di molecole per unita di volume sia ancora tale da poter applicare i metodi statistici, il che nei problemi in studio
e senz' altro vero.
11 concetto di
pressione invece perde in parte il suo consueto significato, fO:J.dato sull'assunzione che gli urti tra molecole del gas e parete siano perfettamente elast!. ci di guisa che la pressione p. dovuta aHe molecole incidenti su di un elemen 1
-
to di superficie dA e quella p dovuta aIle stesse molecole dopo l'urto siano r
tra di loro eguali, ed eguali aHa meta della pressione del gas:
(1. 6)
p. = p = 1/2 P 1
r
p = p. + P = 2 p. r
1
1
Ora alle bassissime densita che noi consideriamo questo schema di interaziQ ne superficiale
e da rimettersi in discussione, anzi e proprio uno dei princi-
pali problemi aperti del fenomeno stesso. Per superare questa difficolta ci pare spontaneo proporre di assumere come terza variabile del gas, in luogo della pressione p, il numero N di molecole incidenti nell'unita di tempo sull'unita di superficie del recipiente contenente il gas, numero che in equilibrio statistico
e anche eguale a quello delle molecole riemesse dalla stessa super-
ficie. Tale numero da anche la portata numeric a da una faccia all'altra di un qualunque elemento di superficie interno al fluido e puo valutarsi
indipe~
dentemente dal concetto di pressione in funzione della temperahra e della 340
- 321 8. Nocilla
densit9. del 'gas. Introducendo per maggiore omogeneit9. di formule la dens ita numerica
V anziche la densita
j,
ossia:
)) =
s> 1m =
numero di mol~
cole contenute nell'unita di volume del gas, coi metodi della teoria cinetica dei gas si ricava:
N :;
(1. 7)
~a
v ~ RT 21T
:;
).I . c
2{rr
(1. 7) pub essere assunta, come nuova equazione di stato nelle variabili T
10ppure c), V ed N in luogo della (1,6). Poiche Ie quantitA N e V sono dell'ordine di grandezza del numero di Avogadro NAv :; 6,023 x 10 23 mQlecole per mole, potra essere piu comodo ai fini pratici considerare in luogo di N e
))
i loro rapporti con detto numero di Avogadro. Anche con questa m£
difica la (1. 7), data la sua struttura, continua a valere immutata. 8i osservi poi che dalla (1. 7) risulta che essendo:
(1. 8)
R = RM
1M,
con RM = costante universale dei gas = 1,9864 cal
10 K
M = peso molecolare del gas la quantita N, a paritA di temperatura assoluta T e di dens ita numerica
V,
risulta inversamente proporzionale alla radice quadrata del peso molecolare M.
A titolo di esmpio riportiamo alcuni valori numerici delle quantita so-
pra considerate. In condizioni normali (temperatura di 20 0 C e pressione
760 mm. di Hg) abbiamo (De Boer idrogeno (H 2) : N :; 11,0 x 10
N/N Av = 1,82
23
[2]
molecole moli
pag. 7):
I cm.2
I cm 2 sec.
341
sec.
- 322 -
S. Nocilla
azoto
2 molecole / cm sec.
N = 2,94 x 10 23
(N 2):
. 2 moll / cm sec.
N/N Av = 0, 487
. (° ) : osslgeno 2
2 molecole / cm sec.
N = 2,75 x 10 23
moli / c m 2 sec.
N/N Av = 0,456
Alla quota di 250 Km. alliequatore, assumendo per la temperatura T i1 valore di 1036 oK e per la densita
.9 un valore 0,79 x 10- 9 volte minore di quello
corrispondente alle condizioni normali di cui sopra (valori dati in [1] , pa£ te II, per 11 atmosfera tipo internazionale a quella quota) si ottiene che i valori pitt sopra riportati devono essere molpiplicati tutti per i1 fattore numerico:
0,79 x 10 -9 xtf2 93 I 1036
= 0,42 x 10 -9
Si ha cosi in definitiva, limitandoci a riportare Ie frazioni molari: -9
idrogeno (H 2)
: N/N Av = 0,76 x 10
(N 2)
: N/N Av =0,20x10
ossigeno (°2)
: N/N Av =0,19x10
azoto
moli / cm.
2
-9
"
"
-9
"
"
-
sec.
Accanto alla portata numerica N converra associare Ie
corrisponde~
ti portate di quantita di moto Q e di energia cinetica E, che valgono:
-+.. ... Q = 1/2 P n, con p dato dalla (1. 5) ed n normale alla superficie attraversata
342
- 323 -
S. Nocilla
E = NkT (2 + ~ ), con mono- , bi- ,
0
~ = 0,
1, 3/2 rispettivamente per gas
pluriatomici
La considerazione delle grandezze T, V ed N come variabili di stato permette di dare in modo semplice e completo il concetto di rieniissione diffusa per qualunque valore di s e
S.
Precisamente diremo che Ie molecole
cadute sull'elemento dA ne sono riemesse in modo diffuso se esse si
comport~
no come la parte che attraversa dA in un solo verso di un gas in equilibrio maxwelliano in cui Ie due variabili di stato T ed N valgono rispettivamente:
(1. 9)
Tw = temperatura della parete {
N i
= numero delle molecole incidenti nell'unitA di tem-
po sull'unitA di superficie
Le altre grandezze che caratterizzano il gas di cui sopra, che contraddistingueremo con l'indice w, valgono:
V2 R T w
c w
()
)W
"'t"
w
E
w
=0
=
N~k T
w
(2 + d) 343
=m.)}
w
- 324 -
S. Nocilla
,-c.
,E abbiamo indicato rispettivamente la pressione normadove con p w w w Ie, 10 sforzo tangenziale e la portata di energia cinetica dovute aIle sole molecole riemesse in modo diffuso. Vediamo aHora come Ie prime idee che furono poste alIa base della teoria dell'adsorbimento aIle bassissime pressioni possano essere utilizzate ed estese allo studio dell'interazione di un flusso di mole cole libere con una superficie rigida. Anzitutto osserviamo che data la estrema rarefazione del gas possiamo riferirci senz'altro all'adsorbimento fisico (e non chimico ) unimolecolare, con superficie adsorbente lontana dalle condizioni di saturazione. Usando illinguaggio delle isoterme di adsorbimento di Langwluir (v. fig. 2) in cui
S*
e riportato il rapporto
tra superficie ricoperta da gas ad-
sorbito e superficie esposta al gas in funzione della pressione p di questo ultimo, ci troviamo in corrispondenza della parte iniziale di tali isoterme, dove la relazione tra p e
V* e
fl'1
con ottima approssimazione lineare. Un'idea introdotta gia fin da Maxwell
e,
come ben noto, che tra tutte Ie
m~
lecole incidenti sulla superficie una
o
certa frazione, che chiameremo a,
1 mm al Hg
sia trattenuta per un certo periodo di tempo su di essa, per venirne poi
Isoterma ai. aasorbimenlo di LangMuir
riemessa in modo diffuso nel modo precisato precedentemente, mentre la rimanente frazione 1 - a subisca
Fig 2
344
p
- 325 -
S. Nocilla
un urto elastico sulla superficie, ossia subisca quella che viene denominata riemissione speculare, mantenendo proprieta identiche a quelle del gas incidente Secondo questa modello di interazione, che chiameremo brevemente maxwelliano, Ie grandezze p , "V ,E definite a pag. 1, relative a tutte Ie r r r molecole riemesse, essendo funzioni additive delle molecole stesse, in virtu di tutto quanta detto precedentemente hanno Ie espressioni seguenti, valide sia per s= 0, sia per s
f0:
p = (1 - a) r
p. + a p 1 w
'"t' = (1 - a) "'C.
(1. 11)
r
E
1
= (1 - a) E.
r
1
+a E
w
dove p., 1:., E. sono definite a pag. 1
1
1
t.
Tenuto conto delle (1. 11) si ricava Ia conseguenza degna della massJ:. ma attenzione che secondo questa modello sia il coefficiente di accomodamento per l'energia (X, introdotto da Knudsen [4 j, sia gli altri due coefficienti di accomodamento 6' e 6", introdotti piu recentemente da Bell e Schaaf[ 5 ] e cioe:
a
E. - E =
1
E.
1
- E
r w
(1. 12)
~=
Ti -1r
ff' =
1:.1 risultano eguali tra di loro ed eguali ad a: 345
Pi
- Pr
Pi
- Pw
- 326 -
S. Nocilla
(1. 13)
Infine, sempre secondo questo modello di interazione, il tempo di soggiorno
-.;
:If
relativo a tutte Ie molecole incidenti si calcola nel modo
seguente. Per Ie molecole riemesse specularmente esso
e ovviamente nullo.
Per queHe riemesse in modo diffuso si pub adottare 10 schema di Frenkel, descritto a pag. 4, e quindi applicare per esse la formula (1. 4) che ora scriveremo:
dove
Q/RT
• e
(1. 14)
"'C'
*w
w
rappresenta il tempo di soggiorno delle sole mole-cole riemes-
se in modo diffuso. Appare poi verosimile che, d'accordo con quanto gia de,! to a pag. 5, tanto Q che
-r;llf.'1f
solido, e non dai parametri
dipendano solo dalle condizioni superficiali sul s e
~
. Al calcolo del numero N* di moleco-
Ie adsorbite per unita di superficie contribuiscono soltanto Ie molecole
rieme~
se in modo diffuso, e pertanto applicando la (1. 3) a queste ultime soltanto abbiamo:
(1.15)
:If'
N
=
aN.
1
-r: :Ifw
a N. "t" 1
:If*'
e
Q/RT
w
Applicando invece la stessa formula (1. 3) a tutte Ie molecole incidenti, e riemesse in parte in modo diffuso, in parte in modo speculare, abbiamo:
(1. 16)
346
- 327 -
S. Nocilla
Confrontando le (1. 15) e (1. 16), si ricava in definitiva:
(1.
17)
-,;
*"
=a-c K*
e
che stabilisce una relazione lineare tra il tempo di soggiorno medio complessivo
-,;* ed i coefficienti di accomodamento (1. 13) , d'accordo con quanta
gia avemmo occasione di mettere in luce in [8], valendoci di risultati sperimentali. Dalla (1. 17) consegue pure, in base alle (1. 12) ed (1. 13), una rela E. - E ,-C. - "C ,p. - p ed il 1 r 1 r 1 r il che presenta un espressivo significato fisico.
zione line are tra le differenze tempo di soggiorno "C *,
In conclusione il modello di interazione maxwelliano, adattato "eli fenomeni aerodinamici che ci interessano, riconduce tutto il problema dell'interazione alla determinazione del solo parametro a , di cui interesserebbe cQ noscere in particolare la dipendenza dalla temperatura del solido T , dal pa w rametro di velocita s e dall'angolo di incidenza ~ formato tra la velocita del missile e la normale all'elemento di superficie dA:
(1. 18)
a = a(T , s, w
e)
La determinazione teorica di ques1adipendenza,
COS!
come il perfezionamento
del modello stesso richiederebbero uno studio molto pili approfondito e diffici. le, che dalla letteratura a noi nota non risulta sia ancora stato affrontato in modo convincente. Pensiamo che uno studio di questa genere, se davvero deve costituire un sostanziale passo avanti rispetto al modello di prima approssimazione maxwelliano, dovrebbe intanto conservare quelli che riteniamo si.a:. no i suoi caratteri qualitativi fondamentali, e cioe: 347
- 328 -
S. Nocilla
10 ) prevedere uno scambio di quantita di moto e di energia tra molecole incidenti e riemesse e Ia superficie rigida; o 2) prevedere un fenomeno almeno parziale di adsorbimento delle molecole del gas incidente da parte del solido, con relativo tempo di soggiorno.
Inoltre in armonia con una proprieta chiaramente messa in luce dai risultati sperimentali ottenuti coi raggi molecolari, dovrebbe soddisfare ad una terza esigenza, e cioe:
30 ) prevedere una riemissione in tutte Ie direzioni rispetto alla normale a dA anche se tutte Ie molecole incidono secondo la stessa direzione, come avviene appunto coi raggi molecolari.
Naturalmente tutti questi aspetti parziali del fenomeno dovrebbero essere as: cessibili ad una valutazione quantitativa, ed essere collegati logicamente tra di loro. 11 che ci pare impresa tutt'altro che facile.
348
- 329 -
S. Nocilla
2. Calcolo delle grandezze fisiche fondamentali che intervengono nei fenomeni d'interazione.
Per dare piu solido fondamento allo studio del problema dell'interil.zione introduciamo, accanto allo spazio fisico
(Oxyz) nel quale si muove
il missile (v. fig. 3), anche 10 spazio delle velocita assolute delle molecole costituenti l'atmosfera rarefatta, che chiameremo (Ou ..
0
v w ), essendo u , 0
0
0
v , w Ie componenti della velocita assoluta V di una molecola secondo gli o
0
0
(~ =
w
z
o
'_.'--y
~------~--~----v
x spazlo lislco
u spazio aelle velocita
Fig. 3
349
versore ai
V
- 330 -
S. Nocilla -+
assi Oxyz; ed inoltre 10 spazio (Auvw) delle velocita V relative al missile. Porremo cioe:
V
=
velocita assoluta di una molecola
D
=
velocita assoluta del missile
o
-
o
V
(rispetto all' atmosfera) II
II
velocita relativa della stessa molecola rispetto ad una terna di assi che si trasli con velocita D.
In base al teorema di composizione delle velocita risulta ovviamente:
(2. 1)
- -Vo
=
V +D
-
-
V = -.. V - D o
ossia
Se in particolare ci riferiamo aIle molecole incidenti suI missile, ponendo: -+ D =- D i
~
la (2. 1) diventa: (2.2)
-- V = V + Do, o
1
Essa mostra come nel moto relativo al missile Ie molecole incidenti sono a-+
nimate da una velocita d'insieme Do comune a tutte Ie molecole, pili la velo1
cita "termica"
Vo , variabile da molecola a molecola.
una superficie elementare del missile ed
rt
Sia poi (v. fig. 4) dA
la sua normale rivolta verso 10
interno del missile. Limita.tldoci allo studio dell'interazione col solo elemen to dA assumeremo l'asse z con direzione eguale rna verso contrario ad
...,.
...,.
It
e l'asse x, di versore t, nel piano formato tra Do e l'asse z, orientato in mo 1
350
- 331 -
S. Noci1la -+
do che U. abbia componente 1
positiva secondo esso. Essendosi supposta z
maxwelliana la distribuzione delle velocita delle molecole costituenti l'atmosfera rarefatta, la probabilitA 1Ti( u , v • w )
o
0
0
che una molecola di tale atmosfera abbia velocita assox t
luta contenuta nell'elemento di volume du dv dw della o
0
n =nor'nale
Interna al missile
Fig. 4
0
spazio delle velocita assolute vale: d
(2. 3)
0
0
Y.
1
TT.(u ,v ,w ) 1
(u ,v ,w )
,I
0
0
0
v
0
1
du dv dw
o 0 f.(Q)-1 3 c
0
dove: _V2 / c 2 (2. 4)
f.(Q) = e
0
1
e la ben nota funzione di distribuzione di velocita maxwelliana,
definita in tut
to 10 spazio delle velocita. e c la velocita pili probabile legata alIa temperatura assoluta T dalla relazione: (2.5)
c
2
= 2R T
con R = costante del gas considerato =
~/m i
.~ = costante di Boltzmann ed m massa di una molecola. 351
- 332 -
S. Nocilla
Nella (2. 3) Vela dens ita numeric a, cioe il numero di molecole per unita di volume dell'atmosfera ambiente, e d'1(u O' vo, wo) e il numero di molecole con velocita nell'elemento duo dvo dw 0' tra Ie V sopra considerate. Tenuto conto delle (2.1) e (2.2) la (2.4) diventa:
(2.6)
-
dove V ha,come si disse, componenti u, v, w, e ....... Ui si pub esprimere nel modo seguente:
~ i essendo l'angolo di incidenza formato tra u;. ed n.
Da tutto quanto sopra
detto consegue che, essendo 1'elemento di volume dS Q relativo ad un medesi mo punto Q 10 stesso sia che 10 si consideri appartenente allo spazio Ouovowo oppure allo spazio Auvw:
(2. 8)
dudvdw
con evidente significato dei simboli si ha pure: d ' •. (u, v, w)
(2.9)
(
IT.1(u,v,w ) = _"I _))_ _ -- 1T1. uo 'vo,wo
)
1
-
11' con fi(Q) dato dalle (2.6) e (2.7).
352
f (Q) ddd u v W 3 c
-~.
- 333 -
S. Noeilla
In virtu della (2. 9) si possono esplicitamente ealeolare Ie grandezze
-
fondamentali relative al gas incidente, e cioe la portata numeriea Ni , la portata di quantita di moto Qi e la portata di energia Ei relative all'elemento dA, gia definite nel numero preeedente. Infatti se tutte Ie moleeole avessero egual velocita si avrebbe ad esempio: Ni dA = - w V dA Poichl! Ie" molecole non hanne tutte egual velecita, bisogna sostituire nella formula precedente d Vi (u, v, w)
al paste di
)J,
e poi sommare i1 contribute
di tlltte Ie molecele cui corrisponde -w ~ 0. Si deve doe calcolare l'integrale:
(2.10)
N. =j(_W)d)).(U'V'W) 1
1
-"'~O
=_v_j 11"3/2
r'"
(-w)f.{Q) 1
dS Q e3
t-
ehe deve essere inteso nel modo seguente:
(2.11)
j(. . )
dSQ =
j"'dU
~
-ro
dvl O(... )
-~
dw
-00
In modo perfettamente analogo si rieava:
(2. 12)
(2. 13)
-Qi --
E. = 1
J ()
-w m
j
9 = m)/
{..l 2
(-w)
J
3/2
11" con
(u, v, w, )
-w~ 0
-T/l~o
=
V d Vi
j
m V2 +
~ ~ T)d)l. (u,v,w) 1
V2 dS Q (-w) (-2- + ~ 'R T) fi (Q) - 3
J..
e
e [=0, 1, 3/2 rispettivamente per gas mono-, bi-
353
0
pluri-
- 334 -
S. Nocilla
atomici. Sostituendo Ie espressioni (2. 6) e (2.7) ed effettuando Ie
integraz~
ni in coordinate cartesiane secondo Ia (2. 11) si ottiene (v. ad esempio [6) e
[7 ]):
(2.14) (2. 15)
con:
Pi = 1/2'JRT [1 +erut+ 2.frr (2. 17)
{
"'[;. = _1_
lfi
(2. 18)
J RT I (&") • s·
A. ( fI)l
sin Q
E. =NkTf(s,Q) 1
e con:
(2. 19)
(2.20)
Tutti gli altri simboli furono gia definiti nel numero precedente. Per quanta concerne Ie molecole riemesse
si presenta Ia sostan-
ziale difficolta che non si conosce a priori Ia Iegge di riemissione, che anzi 354
- 335 -
S. Nocilla
costituisce l'incognita principale del problema. Sono stati proposti diversi modelli di riemissione che consistono nell'assumere certe leggi di riemissio. ne, in modo tale da lasciare liberi, possibilmente, certi parametri da determinarsi mediante confronto con risultati sperimentali. Tra questi modelli di riemissione ci sembrano degni di particolare menzione quelli in cui Ie molecole riemesse sono considerate come facenti parte di opportuni gas in equilibrio maxwelliano, in opportuno movimento rispetto adA. Questi modelli, su cui naturalmente si possono fare molte riserve, presentano il pregio essenziale, a parte motivi di carattere storico, di permettere di adottare per Ie molecole riemesse gli stessi metodi di calcolo, se non addirittura Ie stesse formule, impiegate per il gas incidente. Infatti, detta V r la dens ita numerica del gas fittizio in discorso, abbiamo: d (u,v,w) (2.21) rTr (u, w) = _Y....;..I'f._ __ 'J r
v,
dSQ _1_ rr 3/ 2
3
f (Q)
r
c r
con:
f
(2.22)
r
(Q) = e
,- - I r
- V -U
2 c2
/ r
--.
e dove Ur e c r sono parametri da fissarsi di volta in v olta. La costante Yr si determina in funzione di essi e delle grandezze relative al gas incidente mediante la condizione: (2. 23) Nr = Ni che essendo, in base aIle (1. 7) e (2.14) : ,\ .c N. =Y. _c X(G"); Nr = vr, __r 1
2;;
2
\/tr
355
X(,'l) ,
- 336 -
S. Nocilla con fi
U /e . cos r r
r
(2.23')
3r
(cfr. (2.19)), d lventa : =
YCA,(C')
• e ; 'Y ((i' ) r /"r
)}
r
-
Le grandezze fisiche fondamentali relative al gas riemesso, e cioe Nr' Qr ed E , si esprimono per mezzo della funzione f (Q) nel modo seguente: r
(2.24)
N = r
\V~o
r
(2. 26) E = r
dove si
j
w dlJ
w~O
=j
Q
(2. 25)
j
r
r
=
j'
)l3/r 2
1J'
wmV'd Y
wf (Q) dS Q /c 3 r r
rtf
v+ =
1
~
w
'It
1T'3/2
r
w(mV 2/ 2+ dkT)d))
w~o
=
r
Vfr (Q)
dS Q /c!
~3r/2 J W(V 2/2+ rRT)fr (Q)dSQ/C r3
11'~
v+
e introdotta la dens ita del gas fittizio
Jr = m Vr , e dove gli integrali
tripli devono essere intesi nel modo seguente:
(2.27)
J(. . ~+
I dSQ '
J:
j+CO dv
-00
-co
j
(Xl (. •• )
'dw
0
~
Anche qui poi eonverra seomporre la Qr nelle sue due componenti normale e tangenziale a dA nel modo seguente:
(2. 28)
Riguardo alla (2. 26) si osservi ehe il termine
dRT che vi compare implica (he.
I' assunzione, invero discutibile, che I' energia cineticarcorrisponde ai gradi di liberta rotatori delle molecole non vari durante durante il fenomeno d'interazione. Non vogliamo qui diseutere tale assunzione: ci limitiamo a dire che per Ie molecole monoatomiche
e ~ = 0 e quindi 356
la difficolta non sorge,
- 337 -
S. Nocilla
mentre per Ie molecole bi-
0
pluriatomiche l'assunzione stessa pUG essere
considerata come una prima approssimazione. A titolo di esempio indichiamo come vengono trattati i parametri
ITr e cr
nei modelli di riemissione se-
guenti: a) riemissione speculare
(2.29)
{
~
q . . . . q~ Ur = -U i cos lJin + Ui sm Vi t
-
-
(U r e il simmetrico di U rispetto aUa normale a
cr = c
dA)
b) riemissione diffusa
(2.30)
(T
w
= temperatura della parete)
c) riemissione maxwelliana Delle Nr = Ni molecole riemesse si ammette che una parte (l-a)N r sia riemessa specularmente, secondo Ie (2. 29), e la rimanente parte a Nr sia riemessa in modo diffuso, secondo Ie (2. 30) d) riemissione come parte di un gas in equilibrio maxwelliano con opportuno moto di insieme (v. Nocilla [9 ]). I parametri Ur'
~r'
c r non sono fissati a priori. 11 calcolo mostra
che secondo questo modello intervengono in modo essenziale solo due parametri, e cioe sr = ur/c r
e
~r'
Le formule finora riportate, anche se di fondamentale importanza,
357
- 338 -
S. Nocilla
non si prestano al confronto con la maggior parte dei risultati sperimentali oUenuti con la tecnica dei raggi molecolari, che costituiscono oggi una delle fonti sperimentali pili importanti per 10 studio dei fenomeni dell'interazione. Infatti tali formule danno, per ogni elemento superficiale dA, il valore globale delle diverse grandezze, senza precis are quale sia il contributo ad esse apportato dalle molecole incidenti
0
riemesse nelle diverse
direzioni,
mentre i risultati sperimentali sopra ricordati fanno intervenire in modo essenziale tali direzioni. Per poter confrontare la teoria con queste esperienze ~
bisogna riprendere il calcolo delle stesse grandezze N., Q., E 1· facendo uso 1
1
nella spazio delle velocita (Auvw), anziche delle coordinate cartesiane, delle coordinate polari V,
(2. 31)
r~
e,
V sin
~ cosi definite (v. fig. 3): o~8~;r
e cos ~
v = V sin
G sin
w = Vcos
e
_7r~¢S,+n-
~
In coordinate polari l'elemento di volume dS Q vale:
(2. 32) ~
e inoltre la velocita V si pub esprimere nel modo seguente:
(2.33)
V
V W. con
0 = W(Gl, ~)
: (sin (2) cos ~, sin f) sin~, cos te)
La formula (2. 9) diventa: dN.(Q) (2.34)
11,(Q) _ 1
-
1 '\ I
.I
1 n3/ 2 358
f.(Q) 1
V2 sin (9 dV d Gl d@ c3
- 339 -
S. Nocilla
data dalla (2. 6) deve essere pensata come funzione di dove la funzione f.(Q) 1
v, 6l, ¢: (2.35)
fi (Q)
¢)
= fi (V /c,@,
Gli integrali tripli del tipo (2. 11) si calc olano nel modo seguente: (2. 36)
j (. ..
=
)dSQ
$_
1 11
d
¢
[0 h
-1T
sin
e
d 0'
jO.:l
(... ) v2 dV
0
che introducendo l'angolo solido dSl relativo all'origine A: (2.37)
dJ2. = sinGld (i) d¢
si possono scrivere:
(2. 38)
j
I... ) dS Q
dfl
Ir-
=1
lif
7r
d¢ F( -1T lTifL
e,
¢ ) sin g
d 6:
con
Di conseguenza Ie formule (2. 10), (2.12) e (2. 13) diventano rispettivamente: N·1 =
J
con:
N.(.I1)
N. (Jl)
dJ2
~_l
(2.40)
-j Qi =
(2.41)
Vc
=
1
- Tr3/2
6J
{ilQI
V3dV 4 c
Q.(Jl) dil
U'2-
1
-(.Q)
con: Qi
cO'
.f c -
CO
2
11 3/ 2
cos @ ( fi(Q) "'0
359
V4dV
C5
LJ
- 340 -
S. Nocilla
E
i
J
E.(..i2) dJ2 :.;1- 1
=
(2.42)
con Ei Ul)
= --
.f c 3
cos
g
11 3/ 2
fl( y22 + ~ RT) f (Q) r
"0
In modo perfettamente analogo si possono trattare Ie molecole riemesse secondo i modelli precedentemente descritti, ottenendosi Ie formule seguenti, analoghe alle (2.40), (2.41), (2.42):
N
r
J
=
N (J2) d.1l
~+ r
(2.43)
con: N
W..l
r
(2.44)
1 (XJ
(2.45)
con: E
y2
(iZ) r
cos
ltV 0 (-2-
dove gli integrali rispetto all' angolo solido devono essere calcolati nel modo seguente:
360
- 341 -
S. Nocilla
(2.46)
j
FI-") dJl 'il.+
~
1i ( d,
1f/t
f FIISI, ;)
J-lT)o
Tra gli integrali sopra riportati finora
sine d lSI
e stato calcolato in modo esplicito, in
[ 9 ] , soltanto quello che nella (2. 43) de. la funzione Nr (ll) nel caso che la fr (Q) sia data dalla (2. 22). E' interessante osservare che, come risulta dalle ultime formule scritte, nel caso della riemissione diffusa nel senso da noi precisato in precedenza, tutte e tre Ie funzioni Nr (11) a cos
e,
,Qr (il),
Er Cal risultano proporzionali
con costanti di proporzionalite. esplicitamente indicate nelle for mule
stesse. Cib giustifica la denominazione di "riemissione secondo la legge del coseno" che si attribuisce a tale legge di riemissione, anche se solitamente nella letteratura quando si parla di legge del coseno ci si riferisce solo alIa funzione Nr (Jl.) e non aIle altre due. Prima di lasciare l' argomento dei modelli di z
riemissione vogliamo ri-
asse riel cone di riemissione
cordare un modello molto dO
interessante proposto recentemente da Schamberg
[101, che non rientra ne..! 10 schema di quelli sopra discussi(v. fig. 5). Secondo esso si ammette
Schema ai riernissione ai Schamberg [10]
che Ie molecole siano riemesse entro un opportuno Fig. 5
cono di semiapertura ~o'
361
- 342 -
S. Nocilla
entr~ il quale i1 numero di molecole Nr (Q) e proporzionale a cost rr/2. tI 'fc> ). con
contenute nell'angolo solido dll
r ~ 'Po.
Mediante opportune ipo-
tesi sulle velocita delle molecole riemesse vengono calcolati i coefficienti aerodinamici per corpi di varia forma. Quanto abbiamo finora esposto costituisce un contribl;.to preliminare allo studio teorico dell'interazione. Per procedere ulteriormente
nell'a~
profondimento di tale studio conviene passare da quelli che abbiamo
deno~
nato "modelli di riemissione" a quelli che possiamo definire "modelli di interazione", ossia a teorie che spieghino anche la genesi fisica delle funzioni di distribuzione di velocita delle molecole riemesse, tenedo conto delle CO!! dizioni fisico-chimiche superficiali e delle proprieta delle molecole incide!! ti. Studi in tal senso furono gia effettuati da diversi Autori, come LennardJones e sua scuola
[14]
[l1J,
Bonch-Bruevich
[12J,
Zwanzig [13), Erofeev
ed altri. Ci pare perC> importante sottolineare che per orientarsi ve!'.
so la scelta di un modello di interazione atto a calcolare in modo convincen te Ie grandezze fondamentali dell'interazione si debba innanzitutto fare in modo che siano soddisfatte Ie condizioni generali indicate alla fine del nume1'0
precedente. In secondo luogo ci pare sia tuttora da risolvere il problema
pregiudiziale, che riteniamo di fondamentale importanza, di deCidere se nell'equilibrio statistico che regge i1 fenomeno sia sufficiente considere separatamente ogni singola mole cola che interagisce con la superficie per poi venirne riemessa, oppure se sia necessario tenere anche conto dell'azione prodotta sulle molecole adsorbite dalle molecole successivamente incidenti. In altri termini se, sempre statisticamente parlando, la riemissione delle molecole adsorbite avviene indipendentemente dall'arrivo di nuove molecoIe e quindi i1 tempo di soggiorno
-c+ non dipende 362
sostanzialmente da N, 0E.
- 343 -
S. Nocilla
pure se sono Ie nuove molecole arrivate che cacciano via Ie precedenti e
qui~
di il tempo di soggiorno dipende da N. 11 fatto che nella adsorbimento fisico alle bassissime pressioni Ie isoterme di adsorbimento mostrino una dipen
*' -
Cj~
denza lineare tra la quantita di gas adsorbito /), proporzionale ad N , e la pressione p, proporzionale ad N , alla luce della (1. 3) appare una prova in favore della prima tesi. Non vogliamo qui spingerci in una discussione
pili approfondita della questione che, come abbiamo detto, riteniamo tutt'ora aperta tanto pili se si tiene anche conto del moto d'insieme delle molecole if.!. cidenti. Per terminare, un breve cenno alla letteratura sperimentale, che negli ultimi anni ha ricevuto un impulso assai notevole sviluppandosi a terra con la tecnica dei raggi molecolari e con la tecnica del braccio ruotante per Ie quali rima ndiamo alle conferenze del Prof. Estermann, e nell' alta atmosfera coi missili e coi satelliti artificiali. Dal complesso dei risultati sperl. mentali ottenuti a terra ci pare concordemente messo in luce dai vari sperimantato ri il fatto che alle temperature superficiali ambienti ( N vi
e la tendenza alIa riemissione
diffusa indipendentemente
0
300 oK)
quasi dalle
pr~
prieta del raggio incidente, mentre per temperature maggiori (ad esempio dell'ordine di 1000 oK) vi ne di tipo speculare,
0
e una molto pili spiccata tendenza ad una riemissi~
pili precisamente, usando la terminologia del nostro
modello di riemissione [ 9 ] (caso d; pag.2.51, nel primo caso si hanno valori di sr all'unita
0
=
Uri c r prossimi allo zero; nel secondo caso val~ri pros simi
maggiori. Si tenga perb presente che tutto cib
e fondato soltanto
. d·I NUL). .. t 0 ch e per mIg . 1·lOrare 111 . mo d0 sosu mlsure r ; t: nos t ro conv111clmen
stanziale Ie nostre conoscenze in merito sarebbe misura non solo di Nr
Ln.) ,rna anche dl. -. Ln.) Q r
363
neces_~~.Lo
10)
ed Er
procedere alIa
,se non addiritura
- 344 -
S, Nocilla
della funzione di distribuzione della velocita delle molecole riemesse, Queste misure dovrebbero fornire il criterio fondamentale per decidere quale dei diversi modelli di riemissione giormente al vero,
364
0
di interazione si avvicina mag-
- 345 -
S. Nocilla
BIBLIOGRAFIA
[1]
U. Nobile "Elementi di aerodinamica" [Libreria dello Stato, Roma, (1954)j.
[ 2]
.J.R. de Boer "The Dynamical Character of Adsorption" [Oxford
Univer. Press, London and New York (1953)]. [ 3]
.J. R. Frenkel "Theorie der Adsorption und verwandter Erscheinungen"
[Zeitsch. fUr Physik, ~ (1924) , pag. 117 J. [4)
M. Kundsen [Ann. Physik,1£ (1911), pag. 593]
[ 5]
S. Bell e S. A. Schaaf "Aerodynamic forces on a cylinder for the free molecule flow of a non-uniform gas" [Jet Propulsion, ~ (1953) pag. 314).
[6J
S. Nocilla "Sull' interazione tra un corpo rigidoeduna corrente di m£ lecole libere " Parte I - Scambi di energia [Atti Acc. Scienze Torino, 94 (1959-60), pag. 445J.
[7]
idem, Parte II - Scambi di quantita di moto [ibidem, 94 (1959-60) pag.595}.
[ 8]
idem, Parte III - Relazione tra i coefficienti d'interazione ed il tempo di soggiorno delle molecole sulla superficie [ibidem, 94 (1959-60), pag. 782].
[9]
S. Nocilla " The surface Re-Emission Law in Free Molecule Flow" [3rd Int. Rarefied Gas Dynamics Symposium, vol I (1963), Academic Press Inc., New York, pag. 327]'
[10]
~ Schamberg "A new analytic representation of surface interaction
for hyperthermal free-molecu Ie flow with application to satellite drag" [Heat Trans. and Fluid Mech. Inst., University of California, Los Angeles (1959), pag.
1). 365
- 346 -
S. Nocilla
[11]
J. E.Lennard Jones eA. F. Devonshire
li!! [12]
[Proc. Roy. Soc., vol. A
(1937), pag. 894).
B. L. Bonch-Bruevich "Quantum Theories of adsorption 11 [National Res. Council of Canada, Tech. Translation TT-509, (Ottawa 1954) da Upsekhi Fiz. Nauk. 40 (3) (1950), pag. 369].
[13 ]
R. W. Zwanzig "Collision of a gas atom with a cold surface" [JOll!:. nal of Chemical Physics, 32 (1960), pag. 1173J.
/14 /
A. I. Erofeev
" A proposito dell'azione reciproca tra gli atomi a
superficie di un corpo solido (titolo tradotto in italiano dal russo a cura del nostro Istituto)" [Inzhenernii Zhurnal, Torno IV (1960), pag. 36]'
366
eO.
CENTRO INTERNAZIONALE MA TEMATICO ESTIVO (C.I.M.E.)
F. SERNAGIOTTO
"SOLUTION OF RAYLEIGH'S PROBLEM FOR THE WHOLE KliNGE OF KNUDSEN NUMBERS"
Corso tenuto a Varenna (Como) dal
367
21 al 29 agosto 1964
"SOLUTION OF RAYLEIGH'S PROBLEM FOR THE WHOLE RANGE OF KNUDSEN NUMBERS" by Franco Sernagiotto (Universita- Milano) The aim of my talk is to give an exposition of a paper presented by Dr. Cercignani and me at the Forth International Symposium on Rarefied Gas Dynamics, held at Toronto, Canada, last july
(..!.l .
Time-dependent problems have been scarcely investigated in Rarefied Gas Dynamics for an arbitrary Knudsen number. In fact, only a problem not spatially homogeneous appears to have been considered: the Rayleighls problem. Also, for this typical problem, only approximate solutions have been given, in the frame of the kinetic theory of gases, by Yang and Lees in 1956 and 1960 1958
(~).
and by Gross and Jackson in
Exact solutions have been given in the limiting cases of con-
tinuum theory velocity
(~),
(Rayleigh, 1911 (!) ) , also with the correction for slip
(Schaaf, 195 0
(~)).
The classical Rayleigh IS problem is as follows: let a half space be filled with a gas of density
~o
and temperature To ' and bounded
by an infinite plane wall; the gas is initially in absolute equilibrium and the wall is at rest. Then the plate is set impulsively in motion in its plane with uniform velocity
Uo
. The propagation into the gas of.
the disturbance produced by the motion of the plate is to be studied. There are two independent variables, the time
t
and the ordi-
nate x, which measures distance from the plate. Now I want to say some words about Knudsen number in this problem. As usual, Knudsen number (1)
K
h
(~)
is defined in the following manner,
.1d 369
- 352 F. Sernagiotto
where
?.
is the
mean free
path and
d
is a characteristic
length. However, in this problem, because the plate is infinite there is no fixed characteristic length, as e. g. in plane Couette flow the diA definition of Knudsen number must necessa-
stance from the plates. rily be based on time. Calling
8 the
mean free time, i. e. the average time elapsed
between two successive collisions, we have: 2 2 out a To 0 0 t = (2) 8
( )A. is viscosity coefficient, and a o
is sound
tubed gas). From Eq. (2) one sees easily that
velocity in the imper-
..!
gives the degree of 8 rarefaction of the gas. This is quite obvious, if one thinks that at the
start of the motion,
the collisions between the gas molecules and
the plate surface predominate over those between gas molecules themselves. This regime is essentially the free molecules flow, no matter what the density of the undisturbed gas is . Let us now pass to give a sketch of our technique of solution . We have used Boltzmanp equation, with the only restrictions of linearization and use of the B. G. K. model
en to
des.cribe collisions.
Owing to the linearization we can write for the distribution function (3)
f(x, t,
.~)
= fo [1
3 2
where
fJ£.)
=~Q 1T
-c e
+ h(x, t, £.)J
2
is the
unperturbed Maxwellian state, and c
is the molecular velocity.
370
- 353 F. Sernagiotto
Boltzmann equation takes up the form
+ c
(5)
=
x
(..!.):
L (h)
with the initial and boundary conditions:
°
h(x, 0, ~ ) =
h(O, t, c ) = 2 u c
(6)
-
Besides, when and
0
z
x ..... 00, h(x, t, .£) must be limited for every fixed t
c. Now, if the B. G. K. model is introduced to describe collisions,
the Boltzmann equation becomes:
-dh + c
at
(7)
where
e is
2c z
ah 1 [ " = -
-
e
xox
ill
I+
OO
c
-00
_c 2 ] h e 1 dc - h zl l 1
the mean free time. It is easily seen that the solution
has the following form:
(8)
where
h(x, t, c ) = 2 c
-
z
Y(x, t, c )
x
Y satisfies the following integro-differential equation +00
(9)
dY
dY
~ + cx
~=
rrr J 1
-00
with the initial and boundary conditions:
(10)
Y ( x, 0,
C
Y ( 0, t,
C
x) =
x
)
°
= uo
371
2
-~1
e
.J Y (x, t, Cxi\ Y
- 354 -
F. Sernagiotto
Taking the Laplace transform of Eqs. (9) and (10') we get:
(st1)Ytc
(11)
Y( x,s ,c
x
Y(O,s,c)=
(12)
x
where s
xl
) dc
xl
s
is the Laplace parameter. In order to solve this equation,
we have followed the method of the elementary solutions (!.Q), (.!.). This method, firstly introduced by Case
(~)
in 1960, was used by Cer-
cignani in the solution of stationary problems in 1962 and 1964
(~).
Recently this method was extended to time dependent problems by Cercignani and me
(~).
Briefly described it constists in finding separate-
variables solutions of Eq. (11) and then constructing the general solution by superpos i. hon. It is easily seen that separate-variables solutions of Eq. (11) can
be written as follows: -(stl) (13)
Y (x, s, c )
x
where
f (ex' s )
e
v
f (c ,s) satisfies the equation: v x c
(14)
=
vx
(1 -
x
v ) fv (
Cx
, s)
1 (s t l)rrT
(
_c fv(c x ' s) e 1
2 xl
dc x
1
A careful discussion of this integral equation leads to the following results (10) 1)
For every s
there is
a continuous spectrum of values of v conve-
ring the full real axis. The corresponding solutioljs of
Eq. (11) are not
ordinary functions but generalized functions or distributions :
372
- 355 -
F. Sernagiotto
f (c , s) v x
(15 )
= P _v_ + P (v, s) v -
~
S(v - c x )
where pry, s) is given by :
2
v p(v, s) = lIT [(s+l) e - 2 v
and the symbol P means Cauchy principal value when integrals involving
f (c ) v x
are considered .
2) Besides, for complex values of s inside a curve
0 having
the fol-
lowing parametric representation
Re s
= - 1 + 2v e
(16)
1m s
= - lIT
ve
-v
-v
2
2
IV eu2
du
0
there are two complex values of v , opposite
± v(s), such that Eq. (14) is satisfied by: o ( 17)
f (c ,s) = v x
± Vo
(s) - Cx
(s) is fixed by the following relation: +OO _u 2 • 1 Vo e --- - - - du = 1 (18) (S+1){rT -00 Vo - U Vo
J
The
t· curve is sketched in Fig.!.
373
to each other,
- 356 F. Sernagiotto
-1
Res
Fig. 1 It is an easy
matter to prove that the set of the elementary solutions
has properties of full and partial range completeness. Particular enphasis is to be given to the half-range completeness, since it is well known that boundary conditions for the Boltzmann equation are given only for molecules leaving a physical wall which bounds the gas. I\ccording to the method of the elementary solutions, the general solution. of Eq. (11) and (12)will be:
x
e-(s+1)
Uo
(19) Y(x, s, c)=
x
for
s
( Vo
) ~( -c xB ",
inside the region
Instead, line
s
Res
for
s outside
= -1,
x
A of the complex s-plane (see Fig. 2) ;
A and at the right of the straight
the region
the general solution will be : co
(20) Y(x, s, c )
v:(s)
=
u
0
fTf
f o
x
fv(cx' s) XA(- v, s) e
-(s+I)- + v v
2 dv
[P(V,
s~2 + n2 v2
374
3~7
-
F. Sernagiotto
In Eq. (19) v 0 ding to:
Re
(s) is selected between the two possible values accor-
r.~]
~
>
vo(s)
0
t;Im s
:1
(~\
1-. . - - . . . . ., -.--"~..
: "'-
/t
.
I
B
Re ,
1--1
Fig. 2 XA(-v, s) and XB(-v, s) in E qs.(li)and2
For the explicit expressions of
IT
(20) see ref. (10). By integration with respect to the weights '2 e- c x .L - 2 and 0 TT -~ c e -c x we find u(x, s) the Laplace transform of the mass 10
x
velocity, and f(x, s) the Laplace transform of the shearing:stress :
_
-(s+l) ~ u(s+1) e Vo 0
00
(21) u(x, s) = --~~-~ s Vo XB(vo ' s)
J
o
= u.vi (s+l)
J o
S)]
x
00
(22) u(x, s)
x 2
-(s+1)-+v (vo tv) XB( -v, s) e v 2 2 2 dv lp(V" +TI v
e
-
-(s+l)- + v
Lp (v, s)_12 + 1T 2 v2
375
2
v
dv
- 358 F. Sernagiotto
(23)
T(x,s)
=
x -(s+1)e Vo ~.uo XB(vo ' s)
- PU
1.
0
x 2 co -(s+1)- +v fV(VO+V)XB(-V, s)e v .r;; Syrr dv p V,s 2 + 112 v 2
[()J
o
co
(24)
t(x,s)=
~oUosrrrf
x
-(s+1)- + v vXA(-v,s)e v
2
--------~--~~
"o
[P(V, s)
J2 + 112 v 2
dv
Eqs. (21) and (23) are valid for s in the region A of Fig. 2; Eqs. (22) and (24) are valid for s in the region
B of Fig. 2.
According to well known theorems on Laplace transform, the z-component of the mass velocity and the xz-component of the stress tensor are given by the integrals: a + i co
(25)
u(x, t) =
2rri
f
est ii (x, s) ds
a - ico a + ico
(26)
1 t(x, t) =2T1i
r
e
st
t(x, s) ds
.I
a - i co where ii(s, x) and 1:'(x, st are given
py Eqs
(21) and (23,), and the path of
integration is a vertical straight line at the right of
376
f.
(see Fig. 3)
~
359 F. Sernagiotto
i
1m s
A
AI
.,- - ..... \
I
,
\
i
---------
-1 ,
A
:a.
,
,!
Re s
I
\
Ii; '\
t . . --'"
I
I
Fig. 3 Owing to the analyticity properties of u(x, s} and
'r(x, s} , the previous
path of integration in the s-plane can be deformed to a path indented on the segment (-1, O) of the real axis and along the vertical straight line
= 0 (see Fig.
Re(s+l}
3) • Therefore we shall have:
(27)
u(x, t}
=
2 TT i
r
J
y 0
-1 +i co 1 sL e u(x, s} ds - - - . 2 IT 1
-l-i co
eD(x, s} ds
-1
-1 +i co
(28)
t'(x, t}
2 TT i
J
(I
eS'
'i: (x,
-l-ico where D(x, s} and
t
IJ
211' i
f
e
st
Ll (x,
s) ds
-1
(x, s) are the jumps respectively of ii(x, s} and
(x, s) through the segment ( - 1, O) of the real axis in the s-plane. For the explicit expression of
ref.
s) ds
C!J. 377
D(x, s} and of /:). (x, s), see
- 360 -
F. Sernagiotto
By the method I have described, the solution of the problem was reduced to quadratures. For general
x and t
however, the solu-
tion cannot be reduced to a simpler form. Instead, for
x=O, i. e. at
the plate, it was possible to get simpler expressions for mass velocity and shearing stress, quite suitable to numerical calculations.
In fact we have: t
u(
(29)
0,
t)
Uo
=
1
- + -1 2 2
r
-x e 11 (x) dx x
0
and
co
t( 0, t)
(30)
~o
1
uoTT-
J
[1 -
~(x)
(1+2X2)]
~ fo(x)t/e
dx
0
where
fl (x)
(31)
Here, as usual,
=1-{rFx e
x
2
erfc (x)
I (x) denotes the modified first kind Bessel funtion of 1
order one, and erfc (x) is defined by:
2 erfc (x) = fiT
(32)
The behavior of the velocity and the stress at the plate is shown in Figs. 4 and 5 respectively. /10
1. - -- - - -u(o,t) - -- _____ ~.
'r:(
0,
tt)/( q.u
Uo \
0.'S
o
/2fTf)
,
~~-- -
. ' - .. ~':''':.-:-:
--,·------.. --T'--.. ----" 7J
t
Fig.4
T7r
Fig. 5
378
- 361 F. Sernagiotto
For general
x and
t we have given asymptotic expansions, both for
large and small Knudsen numbers .. Agreement was found with Yang and Lees (1960) (~) in the range of small t/ e and with Gross and Jackson (1958)
(~)
in the range of large t/ e . Besides, for large
~
we have
found: a) a slip correction to the expression of the mass velocity, with
a
slip coefficient equal to that derived from the solution of the stationary slip flow problem (l..!.) . b) A boundary layer term which has not the simple exponential decaying form, as in Gross and Jackson (see ref.
(~)),
and which is exactly desc-
ribed by the same function as in steady problems. ::::::::::::::::::::::::::::::::::::::::
References
UJ
C. Cercignani and F. Sernagiotto : " Rayleigh's problem at low Mach numbers according to kinetic theory" (paper presented at the IV Internaz. Symposium on Rarefied Gas Dynamics, Toronto, Canada, july 1964).
(~)
H. T. Yang and L. Lees (1956) J. Math. and Phys.
~,
195 ; and in
" Rarefied Gas Dynamics" (F. M. Devienne, ed.) p. 201 , Pergamon Press, London (1960) . (~)
E.P. Gross and E.A. Jackson (1958) Phys . Fluids .!,,318.
(~)
Rayleigh, J.W. Strutt, Lord (1911), in "Scientific Papers" , Vol. VI, p.29, Cambridge University press, Cambridge.
(~)
S. A. Schaaf (1950) Univ. of Calif. Inst. of Eng. Research, Rept. NO HE-150-66.
(~)
M. Knudsen (1950) "Kinetic theory of gases" , Methuen, London.
379
- 362 F. Sernagiotto
en
P. L. Bhatnagar, E. P. Gross, and Krook, Phys. Rev. 94, 511,
(.!!)
K. M. Case, Ann. Phys. (N. Y.)~, 1, 1960.
(~)
C. Cercignani, Ann. Phys. (N. Y.) 20,219,1962. C. Cercignani, (1964a) "The Kramers problem for a not completely diffusing wall "(to appear in the J. of Math. Anal. and Appl. ) C. Cercignani, (1964b) "Plane Couette flow according to the method of elementary solutions" (to appear in the J. of Math. Anal. and ApI. ) C. Cercignani, (1964c) "Plane Poiseuille flow according to the method of elementary solutions" (to appear in the J. of Math . Anal. and Appl.).
(.!,Q) C. Cercignani and F. Sernagiotto (1964)
"The method of elementary solutions for time-dependent problems in linearized kinetic theory" Ann. Phys (N. Y.) 30,154,1964. (~)
S. Albertoni, C. Cercignani, and L. Gotusso, Phys, Fluids 2,.933,1963.
====================
Acknowledgment The author is indebted to Dr. Cercignani for many suggestions in preparing this seminar.
380
CENTRO INTERNAZIONALE MATEMATICO
EST~VO
(C.I.M.E.)
GINO TIRONI
LINEARIZED RAYLEIGH'S PROBLEM IN MAGNETOGASDYNAMICS.
Corso tenuto a Varenna (Como) dal 21 al 29 agosto 1964
381
LINEARIZED RAYLEIGH'S PROBLEM IN MAGNETOGASDYNAMICS . by Gino Tironi (Universita- Milano) 1. -
Introduction and position of the problem This paper presents some results obtained on Rayleigh's problem
in Magnetogasdynamics. A more comprehensive and full treatement of this subject will follow elsewhere. Rayleigh's problem is a standard one in the theory of incompressible viscous fluids. The problem is related to the evaluation of the unsteady motion of a semi -infinite fluid, when a plate, submerged in it and originally at rest, is set impulsively in motion in its own plane with constant velocity. For incompressible viscous fluids the problem was first formulated by Stokes in
1850 [12J ,and generalized by Rayleigh
(8)
The problem which is easily solved for an incompressible viscoud fluid greatly complicates when compressibility is taken into account. Various approximate solutions of the problem were obtained with the aid of the boundary - layer theory by Illingworth Stewarts on
[1 ~
[7J
,Van Dyke
, or by linearization: Howarth
[6]
[13J and Hanin [4J.
Recently numerical calculations were performed by Harlow and Meixner
In the last years, owing to the increasing interest in plasma physics, extensions of the problem have been made to Magnetohydrodynamics. Rossow wski
[1 OJ
,Chang and Yen
[3]
Bryson and Roscisze-
[2J, have studied the problem for an incompressible conducting
fluid, when a constant magnetic field is applied perpendicularly to the plate. We will examine here the above problem for a compressible viscous fluid of finite conductivity in which a perfectly conducting plate 383
- 368 G. Tironi
is submerged. By li!learization, an equation is obtained for the pressure behaviour, which was previously obtained by Cole-Largerstrom-Trilling G. A. L. C. I. T. Report
(1949), and Howarth. General expressions
for the solution in operational form are given. However owing to the fact that these expressions are quite intractable for obtaining numerical results, the equation is solved by a finite-differences method. The existence and unicity of the solution in a non-cylindrical domain (the same we used for numerical integration of the equation) together with a detailed discussion of the stability of the finite-differences scheme was given in a recent paper by Albertoni and Cercignani .
[lJ .
Basic equations are the ordinary Navier-Stokes equations and Maxwell equations, where the force term is of electro-magnetic nature and the displacemer.t current is neglected according to ordinary magnetogasdynamics approximation.
%t
(1)
(2)
(3)
(4)
o D.!.l r Dt
CPI
ot
Dh PDt
+ div (
~
~!:!.)
\l ) H =
_
(H. (e -
=
rot ( .!:!!\
_0 =
o
-
m- rot
d
~ + at
- grad ( p +
di v (.!:!
( "VH rot
.~) + di v ( l:?
384
r-e
H2
2) + \l. ·t
m grad
T)+
- 369 G. Tironi
~ H rot!! - f-e
+ rot H· [ fA' e
J!
AH ]
(5)
These are respectively:(l) the equation of continuity, (2) the equation of momentum, (3) the equation of magnetic field (displacement current has been neglected), (4) the equation of energy and (5) the equation of state.
f- e
is the magnetic permeability of vacuum . ~ H
so called "magnetic viscosity" of the medium, and
1
=5'" fe
(5
is the
is its specific
conductivity. p,
p,
T, J! indicates respectively pressure, density, temperature, and
velocity of the medium.
!! denotes the magnetic field. 1 2 is the "stagnation enthalpy" and 't' is the stress p 2 tensor related to the strain tensor by ordinary Navier-Stokes assumpho
=C T +- u
tions :
t"
dUo
ij
oU
2
j =~(_1 + - ) _-
I
OXj
0\
au c 0 aXil ij
lot - .
3(
choice of coordinates is specified in the following figure
1~
y
z
f
plate
•
x
/-y 385
velocity of the plate.
- 370 -
G. Tironi
Symmetry properties show that the applied magnetic field does not z ~ component of induced field is zero. Moreover
change and that
spa~
tial dependence is only through the distance from the plate, y. Thus, considering temperature in place of enthalpy in equation (4), Eq. s (1)
~
(5) become;
aP ~ = IT + o y
( 6)
(7)
P(~ ot
~) (; Y
+v
=
0
~e Hy
"0
Hx
" y
(8)
(10)
p
(;
ay (-tH
(11)
where
RpT
u
is the
dHx
a
~) + ~ (u Hy - v Hx)
x~component
of velocity and
v the
y~component.
To the above equations boundary and initial conditions must be attached. These can be divided into electromagnetic and
fluid~dynamic.
In order to obtain proper boundary electromagnetic conditions at the plate, the electromagnetic problem must be solved both in the plate and in the fluid ; the two solutions must then be matched at the plate by ordinary, well known conditions for the electromagnetic vectors. Let us have perfectly conducting plate moving with a constant velocity
386
- 371 G. Tironi
U in the x -direction. At the plate (y = 0) the following conditions must be satisfied :
(12)
Byly=o+
By/y=O-
( 13)
= Ezlz=o+
E/ z
z=O
-
(In fact the only non-vanishing component of the electric field can be shown to be E ). The plate is supposed to extend from 0 to -
z
Maxwell equations
(14)
inside the plate will be :
OB
ot
E.
= - rot
I = rot !!
(15)
But for
00 •
a perfect conductor we have
(16)
Substituting (16) into equation (14) we can conclude that inside the plate we have: (17)
B x =0 •
B z =0 •
B =B
y
0
Then the electric field will be : ( 18)
E
z
=-
UB
Y
Into the fluid the electric field is given by
1 0 Bx E =- - - - - - -u B +vB z y y x
re5' ()
387
= const.
- 372 G. Tironi
As
+ y-- 0
u_u
E
(19)
v -+ 0
so that
- -1-
=
I
z
and
"Bx
oy
~e ()
y=O+
I
y=O +
U B
Y
Comparison of (19) and (18) gives the required boundary condition:
OB x
(20a)
~
=0
y=O
Analogously it can be shown [lOJ that for a perfectly insulating plate with a magnetic field fixed with the observer we have
(20b)
oRx '0
= 11.
y y=O
rYe
Rxly=O
fS U R y
=0
For a perfectly conducting plate and an applied field fixed with it, we find:
(20c)
Hx
ORx
o Y t=O
I'
=0;
y=O
=
-0'101- U R Ie
y
Gasdynamic boundary conditions at the wall are
(21)
vI
; ul y=O = U
=0
y=O
The heat flux is assigned:
(22)
q
= o
~T
R (~) y
= const. y=O
388
- 373 G. Tironi
(possibly
~ TI = 0) u y y=O
. This condition will further be investigated in the
following section. Initial conditions are:
(23)
~ = Po
u
=0
= Po
v
=0
T = To
H
P
x
=0
at t = 0 • The standard coordinates tranformation for this problem is one of the Von Mises' type. We put:
f
y
(24)
i
=
dy
o so that operations of partial differentiation become :
(25)
9
"0
To ~
Oy (:)
~
()
=
0
ot
L
~D
v
0
o'f
The total derivative is (26)
D = Dt
0
ot
in the new set of variables. Introducing the following dimensionless variables
389
- 374 G. Tironi
P= Po pi
(27)
=
a o v'
=
H hi Y
p
= P0
pi
H
T
= To
TI
'\f = '1,101 ao'V' I
=
u where subscript
V
X
t
Uu l
zero
= -v.o /a02
denotes values of quantities in the unperturbed
standard state and U is the plate velocity; a ~ = 0Pol 90 = '6 R To
tl
~ = /A-,,jpo
is the sound velocity , we finally obtain the
dimensionless form of the equations of motion:
o
(.!.)
"Ft?
=
p =
av
o"f
PT
1 ah
pTI" OU '0 ~ = 0"/1
(28)
-ov '0 t
4 3
au
(fP Of a
)+
OV
1
'0 h
R'H
01
o~ (ff~)
1
1 "0 h 2 '0 p . -(-)
~ ~f
RHO~ 2
Various characteristic numbers have been introduced:
x = "HI VO
: ratio of magnetic to kinematic viscosity ( Chang and Yen number)
390
- 375 G. Tironi
: Mach number M = U/a 0 2 2 RH= ~ a /~eH : Ruark magnetic number (the inverse of Cowling
number)
RH = Po ao U/~ H2 = ~ • M Pr =
II / ~ C:p : Prandtl
3. -
Initial Behaviour.
number
FUrther discussion of the linearized problem will show that a relevant choice of indipendent variables is :
(29)
Then, following Howarth, we will develop the general quantity
A in
a series of the type : (30)
A
We will consider
f"
and
k as constants (not depending on temperature)
and solve "by such a series expansion the full system (28) . This will give rise to the solution of an infinity of ordinary differential equations systems. We will find that, owing to their nature of induced quantities, the series for
v and h will not have the zero-th term. However we
cannot, as Howarth did, separate quantities such as u, p, T which are given by even power series, by v and
,p ,
h given by odd power
series. In fact the presence of an applied magnetic field mixes up terms of different parities, and the simmetry in series is no more so simple. Space and time derivatives now become:
391
- 376 -
G. Tironi
.l.
(31)
1
2 ~2
'0 t
o 0 (~"ij-f~);
1 0 =?J"{J ry
-'0
?J'5
Substituting in system (28) appropriate series expansions of type (30) we find at the lowest order:
Po
=
1
Po
=
To
111M
+ ~ hI -
hI
2t hI
- -
'i
u~
(32)
-83
VII
1
2
Tn
pr
+
$vI - v1 =
+
.!.
0"6
5 TI
0
2 - P~ 0 2
= - 2(0' -1) M (u~
)2
Condition (20. a) written for the new coordinates gives :
~ ~;
=
0
so that we get h~ = 0
hI
1
=
0
Having assigned the heat flux (k constant) we have :
aT = - ~ ClY
(constant)
at
y =0
Then
o (..£...:!' }
_
O/;.yo
"'I))) ,1, = v""- - rr -a
() I
r
Lo
- ~ (constant)
0
392
- 377 -
G. Tironi
And finally
=
T'0
0
= -~ T' = 0 2 T'
1
Taking into account the obvious condition at the infinity we get:
e
2 - 3z /16
-"2 z
These results are equal to the ones given 'by Howarth
[6J
For the initial behaviour of the magnetic field we find:
(33)
In particular initially, at the wall, the magnetic field grows as : h (0)
"'-
2
(t + ...
M
m(l +{I)
Lastly we give the limit for
t ~ 0 of
(~!)
the next paragraph. It is easily seen that ( 34)
((} p~
- 1
~Jt=() - ~
which will be requested in
I,/,= 0
po T~ )~ =0 + ( po T1 + p1 T6 + PI T6 )~ =0 + 0 ( 7)
= T~ 393
(0)
- 378 G. Tironi
( In fact solving the equation of continuity to higher order shows that
91 ==
0).
Then we have
lim
(35)
t --- ~'
4.·
(0 p)
- TI ( ) 1 0
~ '?=O
- - (3
Linearization of the problem. We now put:
p= (36)
1+
P
I
v = VI
P =1 +pl
=1 + T'
T
where primed quantities represent small perturbations of standard conditions . After dropping primes we get:
12+ OV =
at
01jl
p =f ah t
(37)
+ T 02h
'0 lL 'tv d~2 + M o'/J
o
=
ou
= (}2U
(It
?y2
~
--oaT t
oh
+ RI
H
o~
2 OV • .!0p 3 II t
ov =i
ot
0
'(- 1 =--
'6
a
df
oP
+ p1 ~ r
RH
1 2
at£... (h ) 2
;~;: + 13" -1) M2 # 2 + 1~ -1) ~H~)' 394
- 379 -
G. Tironi
The Eq. s of magnetic field and of x -component of velocity are coupled each '. with the other but not with remaining Eq. s of the system. Solving them corresponds to solving the ordinary Rayleigh's problem in magnetohydrodynamics. Operational solutions for
u and
h are easily obtained
(see for ex. [3J)
X-I - S/RH 2 2
(eri( /2~
r1 - r2
(38)
e r
q'P
- e
2
- r
r 2;f
)
1
Where we denote by a tilde image quantities in the Laplace transform space and
s
is the transformation parameter.
~
_: lU~'
+, ~ IfRH) , + "f.!,
{(is +'+,~RHl' -,
(39)
] 1/'
,Ii f/')
The antitransformation of expression (38) is not possible in the general
"t = I,
case but it is rather simple when
as wa's first noted by Chang
and Yen [3J In what follows the condition
X= 1
will be considered fullfilled.
This correspond to a gas highly conducting ; such conditions are easily obtained in shock tubes . For
'X = 1
we have:
1 [ 1 .Ii {t ) + erfc (-1 ...I"Ii IT )~ u = - erfc (- .L- + - 2 2 2lRH 2 {t 2~RH
rr
(40)
h = ~ M/R, ferfc 2 HL.
(~ :L + 2
{T
rr )-erfc (.!.2 1_ ~)J {t 2{RH
2/RH 395
- 380 G. Tironi
From the remaining of system (37) we obtain in the case Pr = 3/4 the pressure equation:
.i 3
'O3 p
0r,;t
(41)
1
-
o2p
7f o~2
02 .!. ~ 0
+ ( i . 1) .
+
M
'0 t 2
20 -
0t
Qu2 (iJ)
0
-1
RH
02
oy2
+ (6 ·1) -
h2 (- ) + 2
1 RH
£..
(0 h)2
o t of'
In the general case a fifth order equation could be obtained
We will discuss Eq.
(41). This corresponds to choose Pr = 3/4 .
The particular value of the Prandtl number is not without interest sin· ce for air we have Pr = 0.72 • 0.73. Eq. (41) is a well known equation first proposed by Stokes for describing the motion of sound waves in a viscous fluid. The boundary condition at
y=
0 is
(42) where (35) has been taken into account. Regularity conditions are assumed at infinity. It is not difficult to write down the operational solution of Eq. (41) with
boundary condition (42) . If we call
term in Eq. (41)
- 381 -
G. Tironi
we have to solve the following equation:
Taking the Laplace transform with respect to the time and denoting by a tilde the transformed quantities we have:
However this' expression is very difficult to handle; in fact it is even impossible to give explicit expression for the whole term ~ (s,
y ).
Then we would be confronted with a four- fold integral giving the analitical expression for
p(
1 ' t) .
We then turned to a numerical solution of Eq. (41)
5. -
Numerical calculation and discussion of the results.
In this section we will only give a brief account of numerical solution. Further discussion of it will be given elsewhere. Let us rewrite Eq. (41) in the form
_!o
d3 p
3
'Of 20t 397
- 382 G. Tironi
where
'6
¢ is as follows :
1 )
(44)
+
2fR,
H
t
e
-4T-'4RH [erf ('I/J-
2 {T
,/I
It
rc
J-
- - - ) - erf ( - + - ) x 2 fR. 2 (T 2m
H
H
""2t Now from formula (43) it can be noted that for the problem to be consistent p-> must be a small quantity of the type
~ = M2 b.
Besides,
owing to the linearity of the equation, as the source term (44) and the 2 boundary condition are proportional to M , then also the solution will proportional to M2. So that we will write in whllt 2 (The true pressure is indeed: P=Po (1+M p)). be
The equation for P is :
(45)
where
4
'03 p
3
~yh
02p
--0-
t
01
2 + Cl P
o
t2
2
=OH
H = M2¢ .
The boundary condition is : (46)
Now with the substitution (47)
oP
u
= ?J'/l
v
= 398
op
ot
follqws p= M2p.
- 383 G, Tironi
we get the following system:
ou ov
(48)
~
01
t
ov ot
4 '(} v -0- +
ol
3
au oy;
oH
+
Initial and boundary conditions are: _ b e- 3t / 40 ,.
(u)
'1=00
ov
3b
=(O'/J) V' =0 40
(49)
(u)
=0
(u)
t=O
==
e
-3t!41 ;
(v)
~=oo
0
o
0
t=o
In order to numerically calculate the solution, system (48) is put in the following finite-differences form
(50)
n+1 n u. - u J j
=
n n v j+1/2 - v j-l/2
~'f
At v
n+l j+l/2
- v
6t where
n j+1/2
n
4
=-
3
u
+ j+l
'6
AY;
is the space subscript and n
n
-u
j + oHn j+l/2
the time superscript. This sche-
me is an explicit one and emploies a "double - network"; u is calculated at points having an integer index and v at points denoted by fractional subscripts [9], This choice gives the scheme a better symmetry and elegance, A look to the source term (44) shows that a relevant parameter is RH' It represents the
square
399
of the ratio of the sound speed
- 384 G. Tironi
to the
Alfv~n
speed. Then it gives a measure of the importance of the
magnetic effects on the phenomenon. RH = 00 corresponds to a vanishing magnetic field, i. e. to purely gasdynamic situation. Calculations performed for such a case can be checked with graphs given in the same case by Hanin [4J, who numerically inverted Fourier transforms of pressure. This check reveals a very good agreement with Hanin's graphs. (See Fig. 1 ) . Fig. s 2,3 and 4 show the behaviour of P for different values of RH (=100, 2, 1) . All this calculations refer to the case b=O. Fig. 2 (R H=100), Shows a very little difference with Fig. 1 (purely gasdynamical case). However it seems likely that if one lets
enough time go by
the same
particularities shown by Fig. s 3 and 4 will appear. These are: the impulse of pressure (the compression wave) which develops also in the gasdynamical case, now seems to appear earlier as the applied magnetic field is stronger. Further more whilst for RH = 00 the impulse after its appearence soon begins to decrease, now up to the time reached in calculations the compression wave presents an increasing amplitude. Common sense would suggest that after having increased the amplitude would finally decrease. However this point has not yet been fully investigated. Finally we can note a last difference with Hanin's graphs. Whilst in the gasdynamical case P is always positive (also in the depression wave facing to the plate ), now a negative value of P appears at the plate. That is to say, in the magnetogasdynamical case, the pressure in the depression wave is below the undisturbed value. (Remember that the true pressure is given by p = Po (1 + M2p)). Probably this is the effect of the Maxwell stress tensor.
400
- 385 -
G. Tironi
It is shown elsewhere , that, as time progresses, a negative contribution to the (51 )
pressure given by 1
--IJJ
2
r-
e
H
2
x
appears. At a given station the asymptotic value of the H component is given x (expressed in dimensional form) by : (52)
Then at a given station the final contribution to the pressure by the Maxwell stress tensor is :
(53)
1 2
When dimensionless form is used one finds that the magnetic contribution to the pressure at any fixed station is
(54)
1 2
~-
D
401
- 386 G. Tironi
Acknowledgment The author wishes to thank prof. Sergio Albertoni for his valuable suggestions. He is grateful to dr. Carlo Cercignani for having suggested the problem and followed the research. Is also indebted to the C. N. R. for having partially supported this work. Calculations, if not otherwise stated, were performed by an IBM 7040 computer at the Centro di Calcolo dell'UniversitA. di Milano.
403
- 387 BIBLIOGRAPHY
G. Tironi
[1]
S. Albertoni and C. Cercignani
: "Su un problema misto in fluidodinamica" - Tamburini Editore Milano (1964
[2J
A.E. Bryson and J.Rosciszewski
Phys. of Fluids 5, 175(1962)
[3J
C.C. Chang and J.T.Yen
Phys. of Fluids 3,395(1960)
[4]
M.Hanin
Quart. J. Mech. App. Math. 13, 184 (1960
~]
F. H. Harlow and B.D. Meixner
Phys. of Fluids 4,1202(1961)
~J
L. Howarth
Quart. J. Mech. App. Math. 4,157 (1951)
~]
C. R. Illingworth
Proc. Camb. Phil. Soc. 46,603(1950)
l~l
Rayleigh, J. W. Strutt Lord
Scientific Papers, 6 vols., Camb. Univ. Press. 1899-1920 Vol. 6 p. 29
~] R. D. Richtmyer
"Difference Methods for Initial Value Problems" Interscience Pubbl. s Inc. (1962)
[lOJ ~n
K. Stewartson
~2J
Stokes, Sir George
Camb.Phil. Trans. Vol. IX(1850) Math. and Phys. Papers - Camb. Univ. Press, 1880-1905 - Vol. III p. 1
~3]
M.D. Van Dyke
Z. angew. Math. Phys. (ZAMP) 3,343 (1952)
Phys. of Fluids 3,395 (1960)
W.J.Rossow
. Proc. Camb. Phil. Soc. 46,603(1950)
405
CENTRO INTERNAZIONALE MATEMA TICO ESTIVO C.I.M.E.)
DARIO GRAFFI
ALCUNI RICHIAMI SULLA IONOSFERA
Corso tenuto a Varenna (Como) dal 21 al 29 agosto 1964
407
ALCUNI
RICHIAMI SULLA
IONOSFERA
di Dario Graffi Universita di Bologna 1.
Nell'introduzione a questo corso sulla dinamica dei gas rarefatti il
prof. Ferrari ha esposto notevoli caratteristiche dell'alta atmosfera. In particolare, ha osservato come ad altezze comprese fra 70-80 chilometri e qualche migliaio di chilometri esista la cosidetta ionosfera,perche a quelle altezze (sia pure con intensita divers a con I'altezza) i gas che costituiscono l'atmosfera sono notevolmente ionizzati.
Poich~
anche il
prof. Krzywoblocki, nelle sue Iezioni, ha accennato aHa ionosfera, ho accettato il cortese invito del prof. Ferrari (che vivamente ringrazio) di richiamare, sia pure brevemente, alcuni metodi radioelettrici con cui sono state messe in evidenza qualche proprieta di quella regione ; soffermandomi in particolare su questioni che presentano un certo interesse anche matematico. Comunque prego i miei benevoli ascoltatori di scusarmi se dirb cose, sostanzialmente, ben note.
2.
Comincerb con I'esporre, sia pure in maniera rapidissima, la
storia dellaionosfera. Nel 1884 uno scienziato inglese, Balfour Stewart, per spiegare alcune anomalie del magnetismo terrestre avanzb, in sostanza, l'ipotesi della ionosfera. Ma questa ipotesi sarebbe perb rimasta poco conosciuta se , nel 1901, Marconi, con Ie sue celebri esperienze, non avesse provato Ia possibilita di trasmissione mediante Ie onde radio fra 11 Inghilterra e 11 America, ossia Ia trasmissione fra due localita anche molto lontane. Ora poiche il successo di Marconi non era spiegabile (come del resto fu confermato da precise ricerche matematiche compiute in seguito) ammettendo llatmosfera omogenea, fin cial 1902 llinglese Heaviside e l lamericano Kennelly avanzarono di nuovo llipotesi 409
- 392 D. Graffi
dell 'esistenza di una regione ionizzata nell 'alta atmosfera, regione che avrebbe avuto 1'ufficio di rinviare a1 suo10 Ie onde radio menti si sarebbero disperse nello spazio)
(che altri-
rendendo cosi possibile Ie
trasmissioni a grande distanza. Ne1 1913 l'ing1ese Eccles sviluppo e applica alIa ionosfera una teoria della propagazione delle onde elettromagnetiche in un gas ionizzato ; di questa teoria, riporteremo fra poco alcuni risultati. Nel 1924, 10 scienziato Appleton di Cambridge inizia con diversi metodi sperimentali Ie sue ricerche sulla ionosfera (ricerche che pili tardi gli meritarono il premio Nobel) e che proseguite da Lui, dalla sua scuola e da un gran numero di studiosi di tutte Ie nazioni (fra gli italiani ricordera il Ranzi) dimostrarono direttamente 1'esistenza della ionosfera e permisero di conoscerne varie proprieta. E ormai, come ben not 0, sono disposte in diversi luoghi della terra stazioni trasmittenti e riceventi attrezzate appunto per misure ionosferiche. Piu di recente I 'uso dei razzi e dei satelliti artificiali ha contribuito ad allargare Ie nostre conoscenze sulla ionosfera. Pera, per brevita, in seguito mi limitera alle misure ionosferiche piu comuni cioe eseguite con trasmittente e ricevente alIa superficie terrestre (1) .
(1) Per ricerche sulla ionosfera mediante satelliti artificiali si veda: B. Rossi - Risultati e prospettive delle ricerche scientificp,e nella spazio-Supplemento al vol. XIX serie X del Nuovo Cimento (I sem. 1961) pag. 194, F. P. Checcaa::i - Ricerche ionosferiche a mezzo dei satelliti artificiali - idem serie I vol. I (1963) pag. 253.
410
- 393 D. Graffi
3. E' opportuno, prima di procedere,richiamare Ie formule usuali che esprimono la costante dielettrica E e la conduttivita zato (la permeabilita
~
r di
un gas ioniz-
del gas si puo supporre uguale a quella del vuo-
to) qualora sia attraversata da un campo elettromagnetico sinoidale di frequenza
f e quindi di pulsazione w = 2Tr
l-
Con alcune approsimazioni si trova ( 1) :
£ = Eo -
(1)
dove
£0
Ne
2
2 2 m(w +U)
(2)
Y=
e la costante dielettrica del vuoto, e ed m rispettivamente la
carica e la massa dell 'elettrone, N il numero di elettroni contenuto nel gas per unita di volume (2),
V
la frequenza colli sion ale cioe il numero
degli urti compiuti da un elettrone nell'unita di tempo, s'intende che questa ultima nozione ha carattere statistico. Ora Ie misure ivnosferiche si svolgono spesso con onde corte cioe con onde di lunghezza eompresa fra 100 e 10 metri ossia con frequenza com6 7 presa fra 3.10 e 3.10 hertz, talvolta si usano pero anche frequenze inferiori, rna non di molto, a 3,10 6 .
( 1) Cfr. per esempio l'articolo di H. Bremmer inserito in S. Fltlgge -Handbuch del' Physik vol. 16(Springer Verlag, Berlino, 1958) pag. 546. Si noti che Bremmer usa Ie unita di Gauss mentre noiusiamo"unita Giorgi razionalizzate . Percio nelle formule del testo la cost ante die lett rica del vuoto non ha valore unitario ed e indieata con Eo, inoltre, sempre nelle formule del testo, non compare il fattore 47r (2) Se nel gas sono presenti anehe ioni, per fissare Ie idee uno sola specie di ioni di massaM,N sarebbe, a rigQre, la somma degli elettroni per unita di volume col numero di ioni, pure per unita di volume, moltiplicato per;' . Poiehe 111/11 e al massimo. il contributo degli ioni al valore di N si puo ritenere trascurabile.
411
- 394 -
D. Graffi
Si pub allora studiare la propagazione nella lonosfera di tali onde applicando i procedimenti dell 'ottica geometrica.
Cio~,
in sostanza, si
considera I 'antenna trasmittente (per Ie nostre ipotesi nella Qassa atmosfera) come una sorgente luminosa che emette raggi in tutte Ie direzioni comprese nella atmosfera, raggi che poi si rifrangono
0
si riflettono
nella ionosfera secondo Ie leggi dell 'ottica geometrica. Ora per applicare tali leggi ricordiamo che se (come nel nostro caso) (" ~ abba stanza piccolo rispetto a £W si pub scrivere l'indice di rifrazione n di un gas ionizzato mediante la nota formula di Maxwell :
,n=
(3)
it
poich~
nella ionosfera U risulta dell' ordine di 10 4 mentre W, come 6 7 s1 ~ visto varia nel nostro caso fra 2 3. 10 e 2 O. 3. 10 si pub tra2 2 scurate /) rispetto awe si ha : Ora
n.
(4)
n
- J V1-kN
(4')
=
dalla formula (4) I in sostanza dovuta a Eccles, segue che l'indice di rifrazione di un gas ionizzato la frequenza f. del vuoto ed
~
~
minore dell 'unita e varia con w
In altre parole un gas ionizzato
~
cio~
con
meno rifrangente
un mezzo dispersivo.
4. Nelle misure ionosferiche la trasmittente T
~
la ricevente R (che
conviene considerare puntiformi) sono di solito molto vicine anzi spesso nella stesso luogo. Si pub allora, per semplificare, ritenere la terra piana e la ionosfera stratificata orizzontalmente parallelamente aHa
terra~l)
(1) Intendiamo per strato una regione in cui n(oN) ~ costante.Se n varia con continuita con l'altezza z, uno sirato generieo si riduce ana regione compresa fra i piani z e z + dz. 412
- 395 D, Graffi
In altre parole si suppone N (e quindi n) funzione solo dell 'altezza z dal suo10, cioe si considera la ionosfera formata da una successione finita 0 1 , f'm 't m L a d'1 s t ra t"1 orlzzontaI'1 ( )•
J'
Cib posto consideriamo (Figura 1 ) un raggio emesso da T e sia i 1'an-
:s~
-p
\
"
go10 che il raggio ( 0 meglio 1a sua tangente) forma in suo punto generico P, con 1a
\
verticale ossia con 1a norma1e agli strati ; diremo poi inclinazione iniziale del raggio
\. ________ <-
'it
\
-L
~,
il valore di i in T , Per la legge di
Cartesio, 1ungo il raggio, il prodotto nseni e costante e poiche nella bassa atmosfera,
cioe in T, si pub ritenere , N = 0 cioe n= 1, si ha (5)
n seni
= sen io
Ora nella ionosfera, n < 1 , dalla (6) segue seni
> senio cioe
il raggio
procedendo nella ionosfera si allontana da1lavertica1e, (2) e continua ad allontanarsi fino ache n diminuisce ossia fino ache N cresce, Ora pub accadere che il raggio in un suo punto S diventi orizzonta1e e, con
(1) Si noti che 1a struttura della ionosfera (cioe la distribuzione di Nodi neon 1'altezza) varia con Ie ore del giorno, con Ie stagioni e anche da luogo a luogo, Quindi 1'ipotesi della ionosfera stratificata orizzontalmente e valida solo in prima approsimazione, (2) In sostanza il raggio passando dalla bassa atmosfera alIa ionosfera passa da un mezzo piu rifrangente a uno me no rifrangente e percib si allontana dalla normale alIa superficie di separazione dei due mezzi cioe dalla verticale,
413
- 396 D. Graffi con un fenomeno analogo al miraggio, torni al suolo. Poiche, in S, i = ]I dalla (5) si ha, detta h I' altezza di S, n(h) i1 valore di
n alla
2-
altezza h: (6)
n(h) = sen io
La (6) e condizione necessaria affinche i1 raggio torni al suolo e se nell'altezza h, dn
dz.
r Ola
(6) diventa anche condizione sufficiente. Come
vedremo, si pub ritenere che la funzione N(z) abbia un solo massimo N M,O meglio sia crescente con derivata positiva fino all'altezza hM dove N e massimo. Allora n
(7)
m
=
n diminuisce fino al valore:
(1- ~!f' <..;
Poiche n pub variare da 1 a nm si ha che la (6) e soddisfatta dai lori di io compresi fra
4-
e. l'angolo
v~
im definito da sen im = nm'
e solo per quei valori. Percio i raggi per cui i o > im tornano al suolo (1), gli altri per cui io Ora se
~
cresce
< im
attraversano la ionosfera.
n tende aI, cioe per CV abbastanza grande
prati-
camente tutti i raggi attraversano la ionosfera. Al contrario se W diminuisce nm , che finora e implicitamente supposto reale e positivo diminui see, anzi per GJ = lAl, = y{~M' im =0 e tutti i raggi emessi. nella ionosfera,
compresi queUi di direzione vertic ale/ tornano al suolo; a maggior ragione tutti i raggi tornano al suolo se
cv < tv.e
Per la (6) si ha che i rag-
gi vertic ali cioe tali che io =0 tornano al suolo dopo aver raggiunto l'altezza
h dove
n=O.
(1) E' ovvio che la'massima altezza di un raggio che torna al suolo deve
essere minore di hM; infatti da quanti> si e detto nel testo segue subito che la massima altezza da cui i1 raggio torna al suolo e i1 val£re re piu basso di h che soddisfa la (6),
414
- 397 D. Graffi
5. Cio premesso, passiamo aHo studio dei metodi di misura della ionosfera. 11 pili comune e il cosidetto "pulse method" cioe la trasmittente emette un breve segnale costituito da un treno d'onde di pulsazione w. Di regola, il segnale giunge aHa ricevente per due vie lungo la superficie terrestre (cioe ,come si suol
dir~J
0
propagandosi
per raggio diretto)
o attraverso la ionosfera cioe per raggio riflesso. Si misura la differenza di tempo t' fra Ie due ricezioni dei segnali e poiche si puo ammettere la velocita del raggio diretto uguale aHa velocita c della luce nel vuoto, detta 2d la distanza tra Ie due stazioni, il tempo 2 t
t't
2d c
sara
impiegato dal segnale per percorrere la ionosfera. Ora
se la ionosfera fosse (fig. 2) uno specchio orizzontale e 10 strato fra la terra e 10 specchio orizzontale fosse occupato dal vuoto
(0 dalla
aria), la distanza TM dall trasmittente
1'1
/1'\----I
j 'T'
--
\
:he \ ~
-_ ...
(0 dalla ricevente) dal punto di
ne M sarebbe uguale act: za h
\_--
e
riflessio~
e l'altez-
della ionosfera sarebbe, per il
teorema di Pitagora ( si ricordi TR =2d):
-"R
(8)
Ana grandezza h, espressa da (8), si da il nome di altezza equie valente 0 effettiva della ionosfera. Si noti che in molti casi trasmittente e ricevente sono nella stesso luogo (cioe d =0) quindi he = C t
,
percio non e necessaria alcuna ipo-
tesi sulla velocita del raggio diretto. Ora, conforme l'esperienza e la teoria he, per w < w , una funzione di w; per w > w come si e vie c c sto la ionosfera e trasparente per i raggi vertic ali. Con opportuni
415
- 398 D. Graffi
artifizi si determina questa funzione ottenendo in tal modo
cosidetti
ionogrammi. Tornando al caso generale di trasmittente e ricevente poste in localita diverse cerchiamo qualche relazione fra I' altezza equivalente, definita da (8) e il raggio TSR che attraverso la ionosfera congiunge T con R. E' assai interessante a questa proposito il seguente teorema di Breit. L'altezza equivalente coincide con I' altezza del punto (che chiameremo aneora M) in cui si incontrano Ie tangenti tirate al raggio nei suoi estremi T e R 0, che e
/ / i
/
stesso, dove si incontrano i tratti rettilinei del raggio
10
~1
nella bassa atmosfera(l) (fig. 3) • Per di-
\\
mostrare il teorema enunciato osservia-
t\ .
~\
·1--';:,
I.S '\-
mo
che i1 segnale si propaga con la
"
velocita di gruppo, uguale, nel nostro caso, a nc(2) quindi , mentre i1 segna-
:/
Ie percorre nel tempo t: il tratto TS del
-1. __,._
'-
R
'T'
Fig. 3 (C punto di mezzo fra
T
raggio la sua proiezione sull 'orizzontale percorre ogni elemento del segmento TC
ed R) con velocita cnseni = cseni . Quindi
(1) Qualora si ammetta nella bassa atmosera n= 1. (2) Infatti la velocita di gruppo u e data dalla formula (cfr. E. PeruccaFisica generale e sperimentale - VIII edizione vol. II pag. 213-215, al posto della frequenza abbiamo la pulsazione w) : 1 U
d
W
ct; (-;-)
dove vela velocita di fase, uguale a ~ . Nel nostro caso ricordando . n (4) Sl ha . 1 1 . .!. =.!. d(nw) = .!..i. (, w2 _ k N = w U c dw c dw i' kN =~
r--'-'- -----
-; - (Zt-
quindi u = cn conforme al teste . 416
- 399 -
D. Graffi
t-
d , c semo
e di conseguenza per la (8) h
e
=
d cot io
= TC cot io = CM
conforme al teorema di Breit. Diremo altezza vera, l'altezza h del punto S pili elevato del raggio; dal teorema di Breit si ha che l'altezza vera e inferiore all 'altezza equivalente. E' bene notare che variando la distanza 2d fra trasmittente e ricevente potrebbero variare h e h, perb, se rimane costante h (il che come vee dremo fra poco si ottiene variando w con legge opportuna), anche h
e
rimane costante. Piu in breve, si pub dire che ad uguale altezza vera corrisponde uguale altezza equivalente. Per provare il teorema ora enunciato scriviamo la (6) ponendo in luogo di nth) il suo valore dato da (4) : (9)
Ora se un raggio verticale giunge fino all'altezza h deve essere n(h) cioe la sua frequenza
w1'l
vale
rk N(h)·.
=0
Se invece 1'inclinazione ini-
ziale e io dalla (9) si ha subito che giunge fino all 'altezza h il raggio di pulsazione wtale che : , 1 - sen 2 10
cos
2
io
ossia vale la relazione :
wn
(10)
w = --, cos 10
Cib posto, consideriamo il moto della proiezione P, del segnale sulI "asse z asse che supporremo verticale e nel piano del raggio. La
417
~
400 D. Graffi
'1
velocita
di), vale ( ricordando che i e l'angolo fra il raggio e
l'asse z) tenenclo presente (10) v
2
1
, 2-----:;---KN :),. = c,! n -sen" io = c 1 - - -sen 10 " w2
2·
2
= n c cos i = c" n - n sen ( i
= c cos
=c
. /'1 10 ,. -
KN 2 ,J w cos
,
I
•
10
c cos io 111 V
KN
- ,.. wh..
quindi 1 c cos it>
(11)
Ora nel caso dell'incidenza verticale i
o
=0
'
h
e
=h
en
= c t; quindi da
(11) (12)
Cioe in generale : h
elJ'L.
ct =
cos
io
e sostituendo nella (8) poiche (fig. 2) d = TM sen io h -,'
e -
J I
h2
en. cos'l.io
h
2
sen e II cos io
= c ~sen iD si ha :
.10
= he1'l.-
cioe he. e costante conforme al teorema enunciato.
6. Proseguendc Ie nostre considerazioni osserviamo anzitutto che nota e nota senio e percio per la (6) si puo ricavare il valore di n ossia e di N all'altezza vera h. Ma, come ora vedremo, e possibile, almena in h
linea concettuale, ottenere da misure di h
418
e
0
di
t
la legge di variazione
- 401 -
D. Graffi
di N con l'altezza zein particolare per la (11 ),la relazione fra h e
e
I
h, perb e necessario notare che Ie onde arrivano fino ad altezza h
0
che ora esporremo, cioe con trasmit-
tente e ricevevente al suolo, e possibile avere informazioni sulla ionosfera fino ad altezze inferiori ad hlVI,/in particolare N(z) pub essere nota solo per z < hM . Per altezza superiori i dati che si possiedono sono stati ottenuti coi satelliti artificiali. Per determinare la funzione N(z) consideriamo) per semplicita un raggio verticale e an cora per semplicita di esposizione ammettiamo N diverse dN da zero e anche per altezze molto piccole e Nb) crescente (con dZ fO) porremo w in luogo
fino Ii. z = hM . Moltiplichiamo la (12) in cui ora
d1 w (h r\.
(13)
e
in luogo di h )per
en.
'IXw , otteniamo :
w 2
w = kNlI dove Nv e
Come si e vis to all'altezza vera n= 0 quindi
ora il numero degli elettroni all'altezza vera. Dunque il primo membro di (13) funzione di w pub considerarsi una funzione f (Ny) di Nv e si pub scrivere, assumendo poi come variabile indipendente
J
h
(14)
f (N ) v
N=v=-N=
rr=,
e questa un 'equazione
dN
rr
=
~dN dN
.J v~-
N
integrale nell'incognitl dz gia risolta da Abel. dN
Si ha infatti: dz _ 1
IN
i v
dz
N:
d
dN
iN
f( N ) v
()
419
- 402 D. Graffi
Integrando e supponendo N = 0, per
z = 0
e
prendendo poi per
variabile indipendente w e ricordando il valore di f (Nv) =
~
e che
w2 Nv = k si ha: (15)
z (N) =
f (Nv) d Nv :!.
YN - NV
)0
w
o anche posta (16)
(N
1
z (N) = iT
=
:r
17
j'tkN n
r . fhe( ~ sen'r) dr
ykN
he (w) dw
I
kN -
~
w
sen
Ie (15) e (16) permettono di conoscere la funzione z
= z(N) per ogni valo-
re di N < NM . Invertendo questa funzione si ottiene N = N(z) fino ache z
< hM . Inoltre la
valore di
(16) esprime l'altezza vera corrispondente al generico
N, in funzione dell' altezza equivalente.
E' da notare che per eseguire gli integrali di (15)
0
(16)
~
necessario
conoscere (si tenga in particolare presente l'ultimo membro di (15)) he (w) per wvariabile da 0 a~ Ora/ per wmolto piccolo." non si conoscono i valori corrispondenti di h e del resto ricordiamo che Ie nostre e formule sono valide solo per w abbastanza grande. La difficolta si potrebbe superare nel modo seguente . Supponiamo che per mezzo dei razzi sia conosciuto la N(z) fino all'altezza h~1 e che corrisponda a valori
w,f
<~
. Allora la
> hI:
fi
h,
(Nt
= N(h.t)
di w nel campo delle onde corte ; in parti-
colare potrebbe essere N(z) = 0 per h scrivere se h
(kN,,',
dz dz..
dN
420
dN
(14) si puo
- 403 D. Graffi
Poiche h > h1 N/
N1 e quindi w> w1 f(~), e nota con misure eseguite
solo con onde radio di pulsazione w > w1' Inoltre poicM N(z)
e nota
fino all'altezza h1 anche l'integrale al primo membro di (17)
e noto e
10 indicheremo equazioni
(18)
per brevita con f ~ (Nv) , Si ha cosl per proprietA delle
di Abel :
N
=~
dz dN
dN
i
J
,/ Nl
f(Nv ) - ~(NV)
"IIN - N
r
V
e integrando e ricordando che per z = h~ N = N4 si ha :
(19)
z-h
~
e questa formula permette di conoscere z(N) anche per h1
per N>
~,ossia
la N(z)
< z < hM
7. Poic he la conduttivita
Y
della ionosfera non
e nulla Ie onde
che
la percorrono subiscono un 'attenuazione. Se, come abbiamo ammesso, V
e piccolo rispetto a
V~
w ,trascurando
rispetto a
w2 si ha per
il coefficiente di attenuazione 1'espressione ( si ricordi (3) e (4 ') ) :
(20)
.( =
~y
Ora considerando, per
-~
brevi~
//
'---' t(.~ ;
(
I
fro
/Vv j
il caso di un raggio verticale il rapporto fra
1'intensita del campo elettrico emesso E e e riflesso dalla ionosfera vale: (21)
421
E~
- 404 D. Graffi
Supposto noto, per ogni valore di
w il primo membro di (22) resta noto )
il secondo membro come funzione di
w ossia, per quanta si
~
detto,
come funzione di NV . Scriveremo percib :
a(Nvl ~
(23)
f'
dj,
Ricordando (21) e moltiplicando (23) per
I h
2
cVNVQ(Nv )
=
YN dz.
2 ,ew:. 2 c
Ik (Nv i N dz .\ dN
=J
-
r(, v
si ha :
iJ
dN
VNV-N () VNV-N-
·JV
Si ottiene ancora un'equazione integrale di Abel che permette il calcolo di
NV
in funziol1e di N
collisione tezza
z
V , e-
ossia, nota N in funzione di z, la frequenza di
percib la conduttivita della ionosfera, in funzione dell'al-
fino a. che z
< hM .
Anehe in questo caso si deve osservare che tl(N)
v
ossia w
sufficientemente grande . La difficolta
s~
~ noto solo per
N
v
pub perb superare in
modo analogo al procedimento indicato nel' numero precedente per il calco-
10 di N(z), rna su cib non insistiamo (1) .
(8) Esponiamo ora, brevemente, i principali risultati ottenuti con Ie misure ionosferiche ; considereremo per semplicita il caso deU'inciden:,o;a verticale a cui del resto, per quanto si Poich~
~
visto, ci si pub sempre ridurre .
la struttura della ionosfera varia con Ie ore del giorno e in special
modo varia col passare della notte al giorno riferiremo anzitutto sulle esperienze eseguite nelle ore notturne. (1)
Per maggiori dettagli si veda L. Caprioli " SuI calcolo della frequenza collisionale nella ionosfera" Rend. Lincei 1954-II pag. 365-370 .
422
- 405 D. Graffi
Come si e detto si ha riflessione dalla ionosfera solo se w > w • Ora c per w poco diverso da w la riflessione avviene ad un laltezza equic valente di circa 300 chilometri. E diminuendo w 11 altezza equivalente diminuisce dapprima lentamente poi bruscamente passando a valori intorno ai 100 chilometri . In altre parole le riflessioni avvengono per la maggior parte delle frequenze in due regioni dette F ed E rispettivamente ad altezze di 300 chilometri e di 100 chilometri. Questi risultati sperimentali sono, in sostanza, conformi all'ipotesi N = N(z) funzione crescente con llaltezza fino a z = h
• Piu precisamenM te ricordando che la riflessione avviene alllaltezza piu piccola h per cui 2 N(h) = ~ i pisultati sperimentali si possono interpretare ammettendo che N(z) presenti un massimo per z= hM (corrispondente alla regione F (Fig. 4) ~ diminuisca dapprima lentamente poi
e che al diminuire di N
/~
bruscamente fino a raggiungere quel valo-
F
re che le compete nella regione
E per
poi dimiI).uire ancora lentamente. Si noti che i risultati sperimentali si possono interpretare (Fig. 5) anche ammettendo in corrispondenza della regione Fig.4
E
un secondo massimo dell N(z) minore di NM . Questa ipotesi, accettata fino a qualche anne fa sembra ora esclusa in seguito a ricerche compiute coi satelliti artificiali .
>
Notiamo poi che di giorno la regione
Fig.5 F si suddivide in due Fled F 2 la prima sopra
300 chilometri, llaltra F.; intorno ai 200 chilometri cioe come
423
" 406 D. Graffi
nella E anche a partire da F 1 al crescere di N si ha prima un Iento poi un rapido aumento di he. Inoltre, sempre di giorno, dai 70 ai 90 chilometri si ha un'altra regione ,detta regione D, alla quale sono dovuti fenomeni di assorbimento che si manifestano di giorno specie con Ie onde medie. Notiamo che talvolta un'onda di pulsazione tre riflessioni dalla regione E
0
w subisce due
0
anche
F. Questi fenomeni sono dovuti
all 'azione del campo magnetico terrestre che rende anisotropa la ionosfera. Ma su cib non possiamo insistere .
9. Come si e visto) e possibile mediante Ie onde radio procurarsi dati sulla ionosfera utili anche per la dinamica dei gas rarefatti. Si ha cosi: un esempio di proficua relazione fra questioni. apparentemente diverse come la propagazione delle radioonde e la dinamica dei gas rarefatti. Poiche esempi di questo genere sono molto comuni concluderb ripetendo I 'opinione di distinti studiosi e cioe che I 'eccessiva specializzazione e, in definitiva, dannosa al progresso scientifico e tecnico.
424
CENTRO INTERNAZIONALE MA TEMA TICO ESTIVO C.I.M.E.)
CATALDO AGOSTINELLI
LE EQUAZIONI DELLE ONDE D'URTO IN UN GAS RAREFATTO ELETTRICAMENTE CONDUTTORE SOGGETTO A UN CAMPO MAGNETICO
Corso tenuto a Varenna (Como)
425
dal 21 al 29 agosto 1964
LE EQUAZIONI DELLE ONDE D'URTO IN UN GAS RAREFATTO ELETTRICAMENTE CONDUTTORE SOGGETTO A UN CAMPO MAGNETICO di Cataldo Agostinelli Universita di Torino 1. - Introduzione Un'onda d'urto
e,
come si sa, generalmente costituita da uno
strato di transizione, detto strato d 'urto, molto sottile, dell'ordine di pochi cammini liberi medi, tale da poterlo considerare una superficie di discontinuita attraverso la quale sono discontinui la velocita , la pressione, la densita,la temperatura, il flusso magnetico, ecc., oltre che Ie lore derivate. Le equazioni che definiscono Ie dette discontinuitl attraverso 10 strato d'urto si ottengono generalmente applicando i principi di conservazione della massa, della quantita di moto e dell'energia, nonche 1'equazione del campo magnetico . Qui stabiliremo Ie dette equazioni riferendoci al caso piu generaIe di un gas viscoso, di conduttivita elettrica finita , in cui il calore si propaga sia per conducibilita che per irradiazione, e supponend6 che la superficie d 'urto sia di forma qualsiasi. Nella valutazione dei salti
chiameremo fronte dell'onda la fac-
cia che limita la regione verso la quale il gas fluisce, mentre l'altra faccia sara quella rivolta verso valle dell'onda d'urto,
Aglielementi
relativi a queste due facce opporremo rispettivamente gli indici.t e 1 Per stabilire Ie equazioni di discontinuita attraverso 10 strato d'urto ricordiamo intanto che se F(x, y, z,
t )e
una funzione deriva-
bile definita in una regione dello spazio contenente il dominio limitato dalla superficie
S(t), variabile col tempo t si ha :
427
D (t ),
~
412 .
C. Agostinelli
~ jF d'l' = dt
(1)
D(t)
..
~
dove n
~ ~
il
..vxn.dU; ...
f-;) t
D(t) versore
della normale esterna alla superficie
S(t), e
la velocita dei punti di S(t) . (1) Se E(t)
r (t), (1')
j JF
~
un aUro dominio variabile, limitato dalla superficie -+
i cui punti si spostano con velocita
j ~F I
I
V,
d F d~ = r;-:- d 'Z"+ ,.. F· '/), (t) d t \ E(t) E(t) . tJ t
avremo analogamente
V·'·x .~n
' d~' ,
e supposto che i due domini D(t) , E(t), abbiano, all'istante sa frontiera, che
cio~
t , la stes-
in quell'istante coincidono, sottraendo membro a
membro si ottiene
(2)
~/ t d
D(t)
Fd T =
~J. dt
Fdt
E(t)
Cib premesso, consideriamo in un'onda d'urto una PO r"l.ione arbitraria s(t) della superficie d'urto e sia c(t) la curva situata su questa superficie che delimit a s(t), Orientiamo la normale alIa superficie d'urto nel senso in cui essa avanza e sia il dominio E(t) e la superficie P
(1)
I
n il
suo versore . Forniamo quindi
(t) nel modo seguente: per ogni punto
di s(t) conduciamo la normale e su di essa consideriamo, da parti
Cf~.
E, GCURSAT, Course dlAnalyse, t.I, p. 665 (Paris, GauthierVillars, 1933), I vedi anche: C, JACOB, Introduction mathematique a la M~canique des Fluides" Chap. XII, n, 11, Paris, Gauthier - Villars, 1959)
428
- 413 -
C. Agostinelli
opposte di s(t), due punti PIe P 2 tali che P - P =-h1t P -P 1 '2 con h > 0, costante.
=h:» '
n punto PI genera la superficie St (t), e il punta P 2 la superficie
s2(t) ; infine Ie normali ad art) condotte per i punti di crt) generano la terza superficie s3(t). II dominio E(t) in questione) e la sua frontiera
eI
(t)
e limitato
dalle tre superficie
= sl(t) + s2(t) + s3(t) •
Supponiamo ora che la funzione F sia dis continua attraverso 10 strato dlurto e i suoi valori limiti sulle due facce di s(t) siano F 2(P, t) , F 1(P, t) . In questo caso, essendo ~ molto piccolo, trascurando i termini
di ordine superiore al primo rispetto ad (3)
/
~
, si pub scrivere
F dZ'"=
E(t) Dlaltra parte la formula (2) resta ancora valida. Infatti in questo caso nei secondi membri delle equazioni (1) e (11) va aggiunto il termine
j
(F -F ) .
2
~
yx ~ .
d (j
s(t)
j
il quale svanisce nel fare la differenza Facciamo ora
tendere h -> 0
. Allora,supposto che la
resti finito , abbiamo dalla (3) lim h-)O e la formula
-d / F(t) dt. E(t)
dr
=
(2) conduce aHa
429
ddt
(F 1 + F 2) d,r s(t)
- 414 -
C. Agostinelli
~.! F d't
lim
h~O
dt JD(t)
poiche per il teorema della media risulta
lim h-'l> 0
J
~ jl) x ~d~=
F(
S3(t)
0
Segue pertanto lim hotO
(4)
~
....
-+
dove vIe v 2 sono i valori di v sui due lati dello strato d'urto,
.
Una relazione analoga vale se in luogo di un campo scalare F(P t), si ha un campo vettoriale F( P t) che presenta una discontinuita attraverso 10 strato d 'urto. 2. -
Le equazioni fondamentali delle onde d 'urto Per stabilire Ie equazioni fondamentali della discontinuita attraver-
so 10 strato d'urto nel moto di una corrente gassosa elettricamente conduttrice, in cui si genera un campo magnetico, utilizzeremo i risultati del numero precedente, applicando il principio della conservazione della massa, il teorema della quantita di mota (
0
del momento), quello della
energia, nonche l'equazione del campo magnetico. a) Principio di conservazione della massa contenuta in un dominio variabile D(t):e espresso dall'equazione
~
(5)
dove
f
Ir
D(t)
d'"t
= o,
e la densita del gas. In virtu della (4) avremo
430
~
415 -
C. Agostinelli
j
f,
-Y I x:'. dO.! r1 (:;I· VI xil.
(:;2
s(t)
d'" • 0
s(t)
da cui, per l'arbitrarieta della porzione s(t) della superficie d 'urto considerata , si ha, in ogni punto di questa superficie, ~
che
\\(~1
Y
12(;2- ')xJ -
(6)
e l'equazione
-V)xn
= 0,
di discontinuita derivante dal principio di conservazio-
ne della massa.
11 teorema della quantita di moto ( 0 del momento), si deduce facil-
b)
mente dall'equazione del moto, che, nel caso pii'l generale di un gas viscoso e radiativo, risulta
dove
p
e l'energia
all'unita di massa ;
potenziale dovuta
aIle azioni esterne, rife rita
h e la pressione totale,
del gas e della pressione di radiazione
,~
somma della pressione p 1 4 :: 'i' aT, essendo a la 4
OJ
.t,
costante di Stefan - Boltzmann e T la temperatura assoluta ; mentre I
A
7
e ~ sono i coefficienti di viscosita, che supponiamo cost anti, B
il flusso magnetico
e
r-
e
la permeabilita magnetica.
Integrando ambo i membri della (7) rispetto al dominio variabile D(t), si ottiene
I
i
(8)
I ,:D(t) \
+
d --v d,~= dt '
1
'.(' D(t)
rot.~
, f grad
,
t _
G. d<> -
D(t)
d ,: +
r./
I'
A (.I
D(t)
431
->
j
grad , D(t) ,
2 v. d
-,> c:>
\'
f t' I
d;-+
+ (, + ti ),' grad div v d D(t)
- 416 C. Agostinelli
Ora, tenendo conto dell'equazione di continuita, si ha
tJ(f ~)
-t
(r v,~)
dove grad l~
dV
-.
+ div (\' v), v +
= IJ t
.,.
d? (I'v)
~ i1 gradiente della diade
JG (l~'
si indica con S(t) la superficie che limita il dominio
I
"'D(t) essendo D(t),
ri
dv,/0 d't' =
r
dt
11""1~ ," '~ t
, D(t)
"
f
l
. Se percib
D(t), si ha :.-+
..,
\. v x n. v d Go' ,
S(t) , diretta esternamente a
nella quale si ponga
F
= \ V, si ha
,~
dv db dt
r D(t)
I
+
~)
, S(t)
il sersore della normale ad
cioe per l'equazione (1),
(9)
d~
. V
'I.I
=
d
j
dt ;' jv , d
,.,.
G
D(t)
il cui secondo membro rappresenta la derivata rispetto al tempo della
quantita di moto del fluido contenuto nel dominio D(t) . -'.
Tenendo conto che div,l1 ._\.
~
roF \~i.'
d ... : = dP -f'
'III
~
= 0, si,Jnoltre
1 T.) 2 ,r"'" + _1 grad;-.v = grad /:, ({.' ,]) - grad 2
-L3 J:J
e quindi ~
(10)
/ J D(t)
.. ~
~
"
rot],\ 'f'
xri. '. ;
: S(t)
1
d
-2
S(t)
432
B 'n 2,~ dC"
.
2
- 417 C. Agostinelli
Infine risulta.
..
\
'
,
dv
tl' b. 2 v + ( l'+ (' ') grad div v= grad ( ,\ div ~ + 2 (t 'D d P ) , dove
D
dv
~
dP
la dilatazione
dell'omografia vettoriale
Ne segue ,2v
(11 )
+ ( \' +, ') grad div v
D(t)
i
d';
=
-)
.
d ":-
(" - S(t) \
In virti:l delle relazioni (9). (10). e (11) • l'equazione (8) diventa
(12)
d
dt
j
D(t) v
d -r,. = - 'D(t)
1 }
grad ..--1. dr, -. 'S(t) (Pt +2(M.·:·
I
d'~
,. x n
)
( 'div v+ 2
+
I
S(t)
S(t)
che esprime il teorema della quantita di moto . Supponendo ora che il dominic D(t) alll'istante
precedente, passando al limite per
dominio E(t) considerato ne n. h-.O, e applicando la (4) , si ha d (
lim - h")O
_,
I
d t ,: D(t) \
Inoltre, essendo la forza esterna continua.
,
I
lim h-tO ID(t)
f gradp ,
t. coincida col
d~
0,
433
risulta
- 418 C. Agostinelli
e la (12) porge ."
"
,.
.;,
v (v - V) x n d
2 2
)
, 1 vI (v!- V) x n d
2
, s(t)
=
s(t)
J
+
i
2
2
= - s(t) l. Pt 2" . 1 +
1 i .;) ,> +-1 ])xn I t ( .' s(t)
~
nd
2 -
d
1
+
.,
\
div v + 2 ,'l'D
1
s(t) dove col simbolo
~
F
~
si
indicato il salta
F 2 - Fl'
Per l'arbitrarieU della superficie s(t) si deduce
(13)
...
...
-->
.•_>
p\ 2 v.2 (v 2- V) x n -
, 2.2 2 -+ -, > I .~ + .,; I v' (v - V) x n = .' P + - ,+ -.Bx n ..f;: + 1 I l L t 21 \ , J . -I
II"'"
j
I
• -t
.' I -, dV + ~ f\.' div v+ 2 riD d P
'I
2
J
n, 1
e questa
e l'equazione
di discontinuita derivante dal teorema della quan·
tita di moto. c)
II teorema dell'energia. Ricordiamo ora che trascurando l' ener-
gia dovuta alle reazioni chimiche, l'equazione dell'energia, nel movimento di un gas, risulta
(14)
d
1
Z
1.
E,
P -dt \ C T + _VH7 + -" ) 2' i \ v
= div ('/
+.!. rot [-l dove (15)
\,' div v + 2 ., D
~~~
dV dP
434
v-p
'.
t
x('1 roU
V+
c:. k grad Tt ,.- grad p ) +
- ::'//1 B )
-Xi
~
- 419 C. Agostinelli
1
e l'omografia degli sforzi dovuti alla viscosita, coefficiente di diffusivita magnetica /
k
=
0f
0') e il
e il coefficiente di conduci-
bilita termica, E = S, P = (l.. T4 e l'energia di radiazione, c la r r r velocita della luce e :til cosidetto coefficiente di opacita. Integrando ambo i membri della (14) rispetto al volume D(t), con calcoli analoghi a quelli precedenti si ottiene
f
rJ
E -+... p) d'6 = - ('I' v-p v.t;*grad T+ I 'S(t) t I.
(16)
+ ~ grad p ) x ri.d co" +
«.f
'1 f
"?
+-
r-. D(t)
1
--
+-
rot,Ex rot B.d'C-
f'l
1 ." .,
r
rot,Bi\_Hx';, de.
D(t)
ora risulta
.
rot _R x
rot'S
= rot
,',
rotBx"~: -diV(rot.BII,B)=-j:?x,b
-div
(rot"~
e
Ne segue
1rot J3-+
~
x
rotjS
...?...
[. .
~
+ rot]3 AB x v = div :BA -~
+ rot ( i"
:\
,..'
(BA v) - ~ rot)-: 'I"j+
~") x ,~,
I
-1.!.12 '-,
••, '
~ 1-: ,
cioe, per l'equazione )
,,', B
,7t + rot
~
(.B A v)
-,~
=
i IJ 2 R ,
cui soddisfa il campo magnetico, si ha ~
1/ rot}3
X
B+
~
rotr~,\ i
';,.
':1
x; = div;"
I
435
. (/,v)- ',rot' ,\',,_ ..
<';1!
,t
x
"),
- 420 C. Agostinelli
e pertanto
I ( J: rot I; x 'D(t)
+ rot
rot
=
x v) d
I
I
=
I
-4'
u')
-:jo.
• .0;'
I
,..
v) - y, rot B (\ H i
[ 13 II (.13 1\
_!
1
_.i
-(-
x n. db
2 )d
'D(t)t 2
, S(t)
Ma per l'equazione (1)
e d
1,.2
d t D(t)
'2:
d.,
2
r
1
d?:"
=
~ :S(t) '2
vxnd
percio sostituendo nella (16) si ottiene
(17)
d~ D(t)[ t' ( Cv T +i v 2 + P) + Er +2~ 13 2 ] (f
=f
• • ;.
-.\;.
v - pv t ,
" S(t)
i + -:-: I
+ f, grad I
1
e l'equazione
+-
C
ir
" (
,
i',
..
... ~,
grad p ) x n d
.
I" , S(t)
che
T
v) -1} rot;
'I
I,'
+
r 1
i·
2 . .
!;.
.;" + - .' v' x n 2:J
che esprime il teorema dell'energia.
Applicandola al dominio E(t) considerato nel n al limite per h ~ 0, e ricordando
precedente, passando
la (4), si ha
\"\rp (Cv T + -21 v 2 +P)+ Er + 21IVI] 2]2 (~-2 V) x n 1 s(t) .
- :i \'
1 2
I ['t' ~ - p ; + k
F .'
1 '..
T
"
grad T +
t
+ 1 ('S(t)! ;;; i(;~
\
~)
-
,rot f\
436
./0 rl ',,!
grad p
+
'2
I
21 (v - V) x i'~) d',
+ -2 v +,;'• ) + Er + -21'\ n J1
(C T v
=
's(t)
dC,
1
'
1 2 x r;- d G"' + rJ 1 ,~ 'J'~1 x ~ . d C'
2
,
- 421 C. Agostinelli
da cui, per l'arbitrarieta della porzione s(t) della superficie d'urto, si deduce
J
-~ .')0 [ , (C T + -1 v 2 +'j17 ) t E + -1 .8 2 (v~ - V) xn v 2 r 2 tl 2 2
(18)
= Iv-
Pv
+ k grad
T
t
1 [ ..
+ ~ grad p ] ~i
;to,
y....
1
Questa
~
l'equazione
~
xn+
r
1
2
:J J
+ - B I\( ] "v) - ~ rot» fiE +'2.tS .
t
2
2
"
xn,
1
di discontinuita derivante dal teorema dell'ener-
gia. In essa avendo riguardo alla (6) si possono trascurare i termini dipendenti dall'energia potenziale ~
,poich~ questa si ~ supposta
continua. d)
Equazione del campo magnetico, Consideriamo ora l'equazione
del campo magnetico ..,
\ I: '. . 2
(19)
e integriamo rispetto al volume D(t) . Abbiamo cos! .
(20)
1
. D(t)
-lo
i'):R
~ /:'
d z" +
f J
Per la formula (1) risulta -;-
I
\.' D(t)
t
(B II -;)
d Z"' =
dt
~/ il 28 d ~ , 0
D(t)
j
,
'dB d -dt"=r;)
rot
D(t)
~
J)
ll d o~ -
D(t)
inoltre , per il teorema del rotore si ha
437
~ ~ -+ !B·vxn J S(t)
- 422 C. Agostinelli
rot (
,v) d
,v) d
S(t)
D(t)
e
,J 2
rot rot
d
=
n
( rot
d
S(t)
D(t)
D(t) L lequazione
d
(20) diventa pertanto
(21)
I'
X
n,v
, S(t) e passando al limite si ottiene
j
! H2,(v2 - V)xn - "1(vl- V)xn.
s(t)
,
2
Hx~,\,. 'I
=
d,
,2
(
dG"'+1J
[rot.B/ I A s(t)
s(t)
n.
=
d~
da cui, per llarbitrarieta di s(t) , si deduce
che
e llequazione
di discontinuita del campo magnetico attraverso il
fronte d lurto. Riunendo i risultati ottenuti si ha che Ie equazioni fondamentali delle onde d lurto in una corrente di gas elettricamente
conduttore, in
cui si tenga conto della viscotita, della conducibilita termica e dello effetto radiativo sono : -~
~
~
~
(i
f2(v 2-V)xn - \'I(v 1-V)xn=0, ....
..,
->
>
,,_,
,~
•
,
.,
'12 v 2 (v 2-V) xn - (I V l'(v 1- V) x n = - ~ p 1 ~ t
438
B2 +-
]2
2i', 1
1::
n
+ -I' /{ I
.' I 2 x ri ,1,,1
1
+
- 423 C. Agostinelli
f div v+ 2tl ' +./l' l
,>
2
dv'/' D d l:' _ 1 n,
= , I V - Pt; + k grad T +.-S grad PrJ 21 x ~ ~ 1, '" f"
+
1\ -+ ,~'~ 1 ,2 .... Hxv /'--2' v q
-i
?rotE... 13J1 x ..n ~)'12
A
I
2
2
dove i simboli hanno i1 significato stabilito precedentemente. Osserviamo che moltiplicando scalarfnente per ii ambo i membri dell'ultima delle equazioni (23), e indicando semplicemente con Via componente normale della velocita dei punti della superficie diurto, si ha, con evidente significato dei simboli
e quindi
o la quale mostra la proprieta ben nota che la componEmte normale del campo magnetico
e continua
attraverso il fronte d'urto.
II sistema di equazioni (23) si puo scrivere ora pil:i semplicemente
2 [
(v - V)] \
,l
1
=
0 ,
439
- 424 C. Agostinelli
(23')
.~
Se i1 campo magnetico e longitudinale , ciol! prime tre
B - .B'1I it
i1 sistema delle
equazioni precedenti si riduce al seguente
n ,
(23")
J
\"(v - V) (C \" T + -1 \)v 2 + E ) + p v 2 II. V 2 r t ~ 1
-) + k grad T +c grad p ~r r
=[tv
che
e esattamente
"12
...
x n,
1
10 stesso di quello che si ha neU;ordinaria gas
dinamica quando si tenga conto ancora dell 'effetto radiativo. Se il campo magnetico e trasversale, cioe 13'Y\.. (23) si riducono aIle seguenti
440
= 0 , Ie equazioni
- 425 C. Agostinelli
(23 "')
c '1'> "~12 x .n', =[,,/,v + k grad T...grad p • - rotJJ;A.P ~r
I (Vol.
•
2 V) i; 1 1
r
r
,2
= I"J.! rotn I, I
-1
n.
1
Osserviamo infine che se si ammette che nello strato d'urto tutti gli elementi variano con a
continuit~
a partire dai valori a valle fino ~
quelli a monte, e intendiamo ora che n-sia il versonre tangente
delle linee ortogonali aHe superficie che limitano 1'onda d 'urto, dalle equazioni (23 ') si deducono Ie seguenti equazioni per 10 strato di transi:done ( (v I
h
-V)
=M
t'
)
,
.~ !~ 2 -> dv'I' j(v'V\.-V)v+(Pt+'2r) n-,\ divv+2i' D dP.Jn = AI'
(V
1\
12 1,\2 ,(32 l' V) (Cv \ T "" -2 \. v + E +........ ) +(p + ) v - - !-: x r 2« t 2( " 1"
c - ./ v + k grad T +........ grad p r " j ~.(
;'.
'L tt
-.'?
V
,
rotB '\ \.: x n = N
"
(v"Yt - V)J3 - J; v - n rot.i) i\ n = A 2 , ~11
'?>
dove M, N sono quantita scalari costanti, ed AI' A2 vettori costanti dipendenti dai valori dei primi membri suI fronte d'urto.
441