1 95 lAp J. . .114 . .

HOC

THE THEORY OF THE FLUCTUATIONS IN BRIGHTNESS

OF THE MILKY WAY. Ill
S. Chandrasekhar and G. Münch
Yerkes Observatory
Received January 23, 1951
ABSTRACT
In this paper the integral equation (derived in Paper I) governing the fluctuations in brightness of
the Milky Way is explicitly solved for the case in which the system extends to a finite distance in the
direction of the fine of sight and when all the clouds are equally transparent. The derived theoretical
distributions are illustrated.
1. Introduction.—In the two earlier papers1 of this series we pointed out that the
theory of the fluctuations in brightness of the Milky Way leads one to consider the
probability distribution of the quantity

w= / TT Mr, a)

where qi ^ 1 is a chance variable occurring with a known frequency ^(<7) and the number
of factors—“clouds”—n{r) is also a chance variable governed by the Poisson distribution

(w = 0, 1, . . .) . (2)
n\ > > /
Further, in equation (1), £ is some fixed positive constant. If/(#, £) denotes the prob-
ability that u exceeds the specified value, then, as we have shown in Paper I,/(w, £) as
a function of the two variables u and £ satisfies the partial integrodifierential equation

From the definition of u as an integral over a quantity which must always be less
than 1, it follows that u can never exceed the value £. But it can take the value £ itself
with exactly the probability e~* (cf. the remarks in Paper I following eq. [18]). The condi-
tions on / resulting from this fact are
/U,£) =0 (w > £)
an
limit /(w,£) = e~^ . (4)
w->£ —0

Equation (3) should therefore be strictly written in the form

/+ (5)
dw d£

Further, the definition of / as a probability distribution requires that

/ (0,£) = 1 . (6)
This is the normalization condition.
1
AP. 112, 380 and 393, 1950. These papers will be referred to as ‘Taper I” and “Paper II,”
respectively.
110

© American Astronomical Society • Provided by the NASA Astrophysics Data System

 BRIGHTNESS OF MILKY WAY. Ill 111

If, as in Paper I, giu, £) describes the frequency distribution of w, then, in accordance

with conditions (4), we should define

/(«,£) = /"«(«,?) du. (7)

The discontinuity of / at w = £ now implies that g(u, £) has at this point a singularity of
the nature of a 5-function (with ^amplitude”
The solution of equation (5) with boundary conditions (4) and (6) presents a problem
of considerable mathematical interest. In this paper we shall show how the complete
distribution of u can be obtained for the case when all the clouds are equally transparent,
i.e., when
x//(q') = ô (q — q') (ç = constant < 1 ). (8)

In this case the equation governing / reduces to

(9)
du^ d£

2. Preliminary remarks on the structure of equation (9).—Since u/q exceeds £ when

qt; < u ^ % and / = 0 for w > £, we conclude from equation (9) that

t+a+u-0 («i«*««. »»

The general solution of this equation is

/(#,£) = e~t </> (£ - w),

where ^ is an arbitrary function of the argument. If cfr can be specified in some way,
then the solution can be continued into the region q2£ < u ^ q£; for the left-hand side of
equation (9) in the domain q2£ < u ^ q£ depends only on/in the domain w ^
And if the solution can be extended to the domain g2£ < u^ q%, then it can also be
extended to the next domain qs% < u ^ q2%, and so on. It thus appears that we must
consider equation (9), successively, in the domains
qn^<u^qn-i^ (w = l,2,...). (12)

The completion of the solution of equation (9) in the manner we have indicated requires
that we know the solution in the first of the domains (12), namely,
q£<u^£. (13)

And, to obtain the solution in this first domain, we must go back to the original defini-
tions and appeal to the problem which gave rise to equation (9).
A further consequence which may be derived directly from equation (9) may be
noted here. Since /(0, £) = 1 for all £, it follows, successively, that all the derivatives
of/ with respect to u must vanish at w = 0:

3. The determination of f(u, £) and g(u, £) in the domain q% < u ^ £ and the enumera-
tion of all their discontinuities.—We inferred that /(w, £) —» as w —* £ —• 0 and that
£) = 0 for u > £ by appealing to the physical problem, in particular, to the fact that
the probability that no cloud occurs in the interval (0, £), is e~*. The question now arises

© American Astronomical Society • Provided by the NASA Astrophysics Data System

112 S. CHANDRASEKHAR AND G. MÜNCH

whether further information concerning/^, f) cannot be similarly obtained by consider-

ing situations in which one or more specified numbers of clouds occur in the interval
(0, £). Actually, we shall see that this provides the key to the entire problem. Let
fn(u, Ç) and gn(u}£) U = 0, 1, . . .) (is)
govern the absolute probability distribution of u when it is known that there are n clouds
in the interval (0, £) (but it is not known where they occur in the interval). More precise-
ly, for every w = 0, 1, 2, . . . , /n(w, £) represents the probability that there will be
exactly n clouds in the interval (0, £) and that the random variable u will exceed the
preassigned quantity u. Further, let gn(w, £) represent the probability density of w, i.e.,
the derivative of /n(w, £) with respect to u at such points where this derivative exists. The
results concerning / and g obtained from the consideration of the case when there are no
clouds in (0, £) can now be expressed in the form
fo(u, ^ e~Z and go(u,£) = e~z ô (£ — u). (i6)
With fn and gn defined in the foregoing manner, we can write
00 oo
/(«,£)= 2/»(«, £) and £ (w, £) = ^ («,£) • o?)
n=0 7i=0
Now the probability that there are n clouds somewhere in the interval (0, £) is
e~*$;n/n \. Therefore, the probability that there are n clouds in this interval and that they
occur between (£i, (?2, & + ¿fe), • • • > and (£n, £n + d£n), where
, a«)
e~^d^id^2 • • • dl;n . (19)
2
The inequalities (18) define a simplex in the n-dimensional Euclidean space (£i, £2,
• ••>£«), whose vertices are at
, £),. . . ,(0, 0,. . . , 0, ?),
(0,0, . . . , 0, ^), and (0, 0, . . . , 0, 0).
We shall call this the ^fundamental simplex." Its volume is £n/w! .
Since the total probability of occurrence of n clouds in the interval (0, Q is e~^n/nl,
it follows that in the (¿1, £2, . • • , ?n)-space we have a uniform distribution of a priori
probability in the fundamental simplex with weight e~*. This is, of course, in accordance
with (19).
The value of u which would result from a distribution of n clouds specified by in-
equality (18), is
w = £1+ (£2 — £1) + <?2(£3 £2) + • • • + (T-1 (£ti £n—1) + (£ — £tí) • (21)
According to this equation, the smallest value which u can take, when it is known that
there are n clouds in the interval (0, £), is ^n£; this happens when £1 = £2 ='. . . = £n =
0. And the largest value which wean take is £; this happens when £1 = £2 = • • . = £n =
£. It therefore follows that /n(w, £) is defined only in the domain grn£ ^ w ^ £ and that
fn(u, £) for u> £ ;
= e~t£n/nl for w^çw£. (22
n
Since the total probability that n clouds occur in the interval (0, £) is e~^ /nl, we mus
also have
(2:

2
A simplex is the w-dimensional analogue of the three-dimensional tetrahedron.

© American Astronomical Society • Provided by the NASA Astrophysics Data System

1 95 lAp J. . .114 . .HOC

BRIGHTNESS OF MILKY WAY. Ill 113

From our earlier remark concerning the distribution of a priori probability in the
(in in)-space, it follows that the probability that u exceeds a specified value,
when it is known that n clouds occur in the interval (0, £) is e~* times the volume of the space
in which inequalities (18) and the further inequality,
Íl+?(Í2-Íl) +<72(Í3-Í2) +. . .+ <T(i-in)^, (24)
are simultaneously satisfied.
Now the shape of the solid into which the fundamental simplex is truncated by
hyperplane (21) depends on the specified value of In fact, we must distinguish the
n cases
ql£<u^ql~l% {l=\,...,n). (25)

Except in the two cases

2Í<^Í (26)
and
qnï<ufi,qn-l%, (27)

it is not simple to evaluate' the volume of the truncated solid. But in the two cases (26)
and (27), one of the two solids into which the hyperplane (21) cuts the fundamental
simplex is again a simplex. Hence in these two cases the volume of the required solid
may be found simply.
Considering, first, case (26), we observe that the volume delimited by inequalities (18)
and (24) is a simplex, the co-ordinates of whose vertices are given by
= =: == =
?2 = . • • = , £z+l £z + 2 • • • = £n £ {l — , n) (28)
and
==
£l = £2 = £3 = • • • = £r» £ (29)
Using equation (21), we readily find that the n vertices defined by equations (28) have
the co-ordinates
u-qli
£1= £2 = ... = £ £z+i — £z+2 — . • . — £„—£. (30)
l-ql '

The volume of the simplex whose vertices are given by equations (29) and (30) is3
1 1
n
u- q £
1 — qn ?

u — qn£ u /n—1
,n—1 £

(31)
n\

u - qn£ U -qn-^ U-
1 — qn 1 ,n-l 1 - Ç2 ?

u — qn% « - qn-^ K-
1 —^ 1 — Ç"-!' 1 - g2 1 -q

Cf. D. M. Y. Sommerville, Geometry of n Dimensions (London: Methuen, 1929), p. 124.

© American Astronomical Society • Provided by the NASA Astrophysics Data System

114 S. CHANDRASEKHAR AND 0. MÜNCH

By subtracting from each column of this determinant the column immediately preceding
it, we can make all the elements above the principal diagonal zero and leave along the
principal diagonal the elements
. £ —u % —u £ —u % —u
and (32)
’ l-qn' 1 - ’ * * ’ ,
1 — </2 ’ Ï-
The volume of the simplex is, therefore,
1 (£-«)"
n\ n (33)
n
?‘=1

The probability that u exceeds a specified value in the interval qi; ^ u ^ £ when it is
known that there are n clouds in (0, £) is therefore given by

(£-u)n
/»(«,?)= —
n\ —n w=i,2, ...)•
(34)
n
1=1

The corresponding expression for gn(u, £) is

gn(u, Ç) = ^, — (q£<u<£; « = 1,2,...);

(35)
n
3=1

in particular,

gl(M, O =-y=Y (ç^<«<i)(36)^

and
g^u, O = {qi<u^!;) . W)

Considering, next, case (27), we observe that now the solid cut ojf from the funda-
mental simplex is a simplex. The co-ordinates of its (n + 1) vertices are given by
=
£i £2 = . • . = £z = 0 and £z+i = £¿+2 = • • . = £n (^ = 0, 1 , . . . , w — 1 ) (38)
and
(0, 0, , 0, 0). (39

From equations (21) and (38) we readily find that the co-ordinates of the n vertices
defined by equations (38) are

£1 = £2 = . . • = £z = 0 and £i+i = £z+2 = . . . = £n =

H K J
y (40)
(/ = 0, . . . ,rc-l).
4
Actually, in this case, gi(u, £) = 0 ior u < q%.

© American Astronomical Society • Provided by the NASA Astrophysics Data System

1 95 lAp J. . .114 . .HOC

BRIGHTNESS OF MILKY WAY. Ill 115

The volume of this simplex is clearly

1 (u-qn£)n (u-qn£)n
fl\ «-"1 n (41)
P[g'(l—5”_i) wlçM"-1)/2 (1 — g"')
J=1

The probability that u exceeds a specified value in the interval qn£ ^ u ^ qn~l£ is
therefore given by
{u-qnH)n
/»(«, £) = ?” « = l,...) •
qn(n—l)/2 (J _ gj) (42)
1=1

The corresponding expression for gn(u, 0 is

e-t (u-qn£)n-1
gn(u, = (qnt<u<qn-1t; n = l, . . .) ■
(w — 1 ) ! 1 (43)
Çn(»!-l)/2 Pj" (1 — q ')
)=1
In particular,
u — O^1 P
gAu, $) = e~{ --(p_g) (1 _g2) (q2t^u^qi;) . (4*)b

According to equation (43),

¿)n-l
?)= ^
qti(n 1 )/2 1 _ qij (45)
y=i
= constant (gn£ <u< qn~l^; w = 1, . . .).

It follows from this last equation that at w = 1,2,...) the complete frequency
function g(u, £) must have a jump in its (n — l)th derivative of amount given by the
right-hand side of equation (45) ; the function and all its lower-order derivatives (if they
exist) must, however, be continuous at w = q11^. Thus, at w = #£, the function itself
suffers a jump of amount e-V(l ~ q)',aXu = q2\ there is a jump in the first derivative
of amount e~*/q{\ — q){\ — #2), but the function itself is continuous here; at w = q*%
there occurs a discontinuity in the second derivative, but the function and its first de-
rivative are continuous, and so on.
Returning to equation (34), we can now write down, in accordance with equations
(17), the complete probability distribution of u in the domain (13). Thus, combining the
results expressed by equations (16) and (34), we have

f(u,£)=e * i+ S
-n — 1 (46)
1
»¡H (1 -q ')
3=1
5
It is readily verified that the solutions given by eqs. (37) and (44) are continuous at « = q%; but
there is sudden change of slope. Also, in this case, g2(w, £) = 0 for w < q2%.

© American Astronomical Society • Provided by the NASA Astrophysics Data System

 116 S. CHANDRASEKHAR AND G. MÜNCH

and
00
(£ —u)n 1

g(u, £) = e-f 5 ( £ — w) +
n=l (47)
(«-1)!]^ (1-<z0
y=i
for q% < u ^
4. alternative form of f(u, tj) in the domain q% < u ^ £.—Solution (46) for/(w, f)
in the domain included between the straight lines u — í and u = q£m the (w, f)-plane
can be rewritten in the form

/(w, £) P[(l_çi+») (ç£^m<£), (48)

n=0 n\ y=i
where

(49)
^ 3=1

Equation (48) can be transformed into a more convenient form by making use of the
identity

(50)
j=l &=0 3=1 ^

established in Paper II (eq. [18]). Letting

00=1, &= (-1)*]^.^ (M)

2 3=1 '
we can rewrite identity (50) in the form
00 oo
n (i - ?n+í)= (52)
3=1 fc = 0

Two particular results which follow from equation (52) and which we shall find useful in
the subsequent work are

and 0 (/= 1, 2, . . .). (53)

K § g*

Combining equations (48) and (52), we can now write

m m
(£—«)’*
j{u, £) ^Qk<ikn-, (54)
n=0 ^1 fc=0

or, inverting the order of the summations, we have

00 oo
/(«, £) (55)
¿=0 n=0

© American Astronomical Society • Provided by the NASA Astrophysics Data System

 BRIGHTNESS OF MILKY WAY. Ill 117

Hence
CO
f (u, £) = Ke^'^Qk exp [qk{£ —u)] (qÇ^u < £). (56)
k=0
This is the required form of the solution.
5. The completion of the solution for f.—With f(u, £) now known in the domain
q% ^ u < %, the solution of equation (9) can be completed in the manner suggested in
§2*.
First, it is convenient to use
F(u, £) = e*f (u, £), (57)

instead of /. The equation satisfied by F is

dF dF
(58)
da

As we have already explained in § 2, the solution of equation (58) must be carried

out, successively, in the domains
qn£^u< qn~l£ (rc=l, 2, ...). (59)

We shall refer to the region included between the straight lines u = #n~1£ and u = qn£
as the wth domain; let Fn(u, £) denote the required solution of equation (9) in this
domain.
According to equation (58), the equation governing Fn+i in the (n + l)th domain is

(s+ÁK'(“-í)=''-(íí'{>
for, when (u> £) lies in the (n + l)th domain, (u/qj Ö lies in the wth domain.
Now, in the first domain included between the straight lines u — ^ and u — q^ (see
Fig. 1), the solution is (cf. eq. [56])
00
Fi{u, Í,) = K'^2/Qktxy[qk — u)\. (6i)
. i=0
We shall now show in detail how this knowledge of the solution in the first domain can
be used to extend the solution into the second domain included between the straight
lines u = q% and u = q2£.
Consider a point P, (u — ç£o; £ = £o), on the straight line u = q%. Since (d/du +
d/d£) is proportional to the directional derivative along a straight line of unit (positive)
slope, we draw through P a line PPf inclined at 45° to the co-ordinate axes. The co-
ordinates (u, £) of a point N on PP', which is at a distance 5 from P, are
s = s (62)
and ^

Conversely,

and £o(l-g)=£-M. (63)

In terms of the variable s, the equation governing P2 can be expressed in the form

© American Astronomical Society • Provided by the NASA Astrophysics Data System

1 95 lAp J. . .114 . .HOC

118 S. CHANDRASEKHAR AND G. MÜNCH

or, substituting for Fi according to equations (61), we have

V2
S=“i:|j&eXp[?i(i“1)v2]- (65

Integrating this equation from 0 to 5 and remembering that the solution must be con-
tinuous along the lines u = qn%, in = 1, 2, . . .), \ve obtain

F,< S; £,) - F. (1Í. ■ £•) - i: Hg I j«P [i‘ 0- 9 H ‘ ■ “S>

Fig. 1.—Illustrating the manner of integrating equation (60), successively, in the domains u ^
qn-l^n = 1, 2, . . .).

The term

Ki Vs ^ (67)
i-?à
which occurs on the right-hand side of equation (66), vanishes in virtue of the second of
the identities in (53). Equation (66), therefore, reduces to (cf. eqs. [51])
OO CO
E2 ( í ; £ 0) = K _ 1 (68)
jöo^g&expt^oO-g)] +öi^^exp [ ?* (^ ) [•

Reverting to the variables (u, £) in accordance with equations (63), we can rewrite the
foregoing solution in the form
0° °o
F2(u, £) =A: je„2öfcexp [qHt-u)} +QiS§exp Hh]}- <*»
{ L X J
k=Q k=0 V

This is the required solution for F in the second domain.

© American Astronomical Society • Provided by the NASA Astrophysics Data System

1 95 lAp J. . .114 . .HOC

BRIGHTNESS OF MILKY WAY. Ill 119

We shall now complete the solution by proving by induction that, quite generally,
the solution Fn(u, £) in the domain included between the straight lines u = qn^ and
u = qn~l% is given by
n—1 f j j f 4/ \ i
ex
Fn(u, ?) =^Ee¿% p[^(^-^)] 0=1,2,...). (70)
¿=o a=o II I
L \x yII'' / l
;=0 ^
Solutions (61) and (69), valid in domains 1 and 2, verify the truth of equation (70) for
n = \ and w = 2; to establish its general validity, we must prove that if the truth of
equation (70) be assumed for a general value of n, then its truth for (n + 1) can be
derived. For this latter purpose, we shall integrate equation (60) along a straight line
of unit slope in the domain (n + 1) and passing through a given point (qn¡to, £o) on the
line u = qn%. The co-ordinates (w, £) of a point on the line of unit slope distant s from
(Ho, £o) are
Ä7 and
£ = £o — (71)
V2 V2-
In terms of s, equation (60) becomes
dFn+l (72)
V2 £o
ds qV2’ V2 )

Substituting for Fn according to equation (70), we have

n—1 “
(73)

Integrating this last equation from 0 to 5 and remembering that Fn+i (0; i) = Fn(u =
£ = £o), we obtain
n —1
Fn+As ; ¿o) = [?*£o(l - qn l
)
¿=0 A=0 "

gi+1Q; Qk (74)
exp [^^(l —

-exp[3^o(l - H'-1) ^(hm- 0] Í •

After some further reductions, in which we make use of equations (51) and (53), we find
that we can simplify equation (74) to the form
CO
Fn+l{s ; £0) = *|>g^expM ^ +^2(7“ OH • (75)

Now, reverting to the variables (w, £) in accordance with the transformation formulae
(71), we have

Fn+1{u, £) =tf¿&¿%exp[Ws-5yi- ™
L X
Z=0 Ä;=0 ^ ^ /J

This agrees with equation (70) for (n -\- 1). This completes the proof of equation (70).

© American Astronomical Society • Provided by the NASA Astrophysics Data System

1 95 lAp J. . .114 . .HOC

Fig. 2.—The probability distribution/^, £) for q = 0.75 and £ = 2, 4, and oo {full-line curves). The
solution for f{u, £) in domains ^n£ ^ w < qn~l£ (« = 1, 2, . . . ) has been extended ton
« < qn£ {dashed
curves) to illustrate the discontinuities of the function and/or its derivatives at = q %.

Fig. 3—The frequency function g{u, £) for q = 0.75 and £ = 2, 4, and œ {full-line curves). The soli
tion for f{u, £) in domains qnf <u qn~^ {n = 1, 2, ... .) has been extended to w < g"£ {dashed curve:
to illustrate the discontinuities of the function and/or its derivatives at w = gn£.

120

© American Astronomical Society • Provided by the NASA Astrophysics Data System

1 95 lAp J. . .114 . .HOC

Fig. 4.—The probability distribution/^/,n-1 9 for q = 0.80 and £ = 3, 6, and °o (full-line curves). The
solution for/(w, £) in domains ^ u < g £ (« = 1, 2, . . . ) has been extended to nu < qn£ (dashed
curves) to illustrate the discontinuities of the function and/or its derivatives at u — q %.

Fig. 5.—The frequency function g(u, £) for q = 0.80 and £ = 3, 6, and œ (full-line curves). The solu-
tion for g(u, £) in domains çn£ < u ^ qn~l% (n = 1, 2, . . .) has been extended to u < q n
£ (dashed curves)
to illustrate the discontinuities of the function and/or its derivatives at w = qn%.

121

© American Astronomical Society • Provided by the NASA Astrophysics Data System

1 95 lAp J. . .114 . .HOC

122 S. CHANDRASEKHAR AND G. MÜNCH

According to equation (70), we may now write the complete solution for/(w, £) in
the form

/(«, £) (77)

(qn^u< qn-^ ; «=1,2,...).

It is of interest to verify that solution (77) is properly normalized. Thus, considering

00 oo
/(O , ¿) = Ke-Z^Q^Q^exp qk% , (78)
Z=0 ^

we expand the exponential and invert the order of the summations; we find

/(o, « -*«-‘2 g (79)

W=0 7._n
fc=0 ¿=0 H

In equation (79) the summation over l leads to a nonvanishing result only for Æ = 0,
and the summation over l then gives \/K (cf. eqs. [53]). Thus

/(O, É) =e-f)£) (80)

Similarly, it can be verified that solution (70) also satisfies the conditioñs expressed by
equation (14).
Finally, the frequency function governing u can be written in the form
n—1
Qi Qk
j £) — Ke * 2 nl jLmJ exp [?4(ê--^)] + e n(t.-u)
z=n i A: = 0 ^ (81)

Again it can be verified that this solution has the discontinuities we have enumerated
in § 3.
6. Illustrations of the derived distribution and frequency functions.—Using formulae (77)
and (81), we have computed the functions/(w, £) and g(u, £) for the cases
g = 0.75 £ = 4 and 2 ,
and (82)
g = 0.80 , £ = 6 and 3 .

It was found that, by retaining twelve to fourteen terms, we can preserve sufficient
accuracy (three to four significant figures) in the numerical calculations. Also it was
found that it was not necessary to go beyond the fourth or the fifth domain.
The computed solutions are illustrated in Figures 2-5. For comparison, the cor-
responding solutions for the infinite case, £ = œ (taken from Paper II), are also shown.
This comparison confirms the danger (to which we have already drawn attention in a
different connection6) of applying the theory valid for £ = œ even to relatively low
galactic latitudes.
6
S. Chandrasekhar and G. Münch, Ap. 113, 150, 1951.

© American Astronomical Society • Provided by the NASA Astrophysics Data System