Brownian
Brownian
1. INTRODUCTION
and potential theory separately, we shall show that there is a natural way
in terms of Rrownian motion of obtaining the operator mapping charges
into potentials and that, conversely, from the potential operator it is possible
to recover the family {P’} of transition operators which define Brownian
motion. These facts will make it clear that in a technical sense the two theories
are identical. In the last section WC shall discuss what minor modifications
in the argument need to be made for dimensions greater than three and
we shall see in what sense the correspondence fails in dimensions one and two.
2. BROWWAX MOTION
PrU[xt, E E, , . .. . (.q, - xf.-,) E En] = Pr,[x, ,E E,] * **a . Pr,,[x, ,,- xIII_i~ E,,]
Equation (1) gives several definitions for Pr,[x, E E] and we must check
that they are consistent. For example, we must check that
and
%[xt (5El = j,. j,& e-‘u-y’*/2tdy d/L(u)= j,, jyu. dy) d/J(u).
determined by the facts that Pt is linear and that for measures p 2 0 with
p(P) -7 1, (f@)(E) = P r ,[ x 1 E E]. Moreover if ($‘l)~E denotes the measure
whose value on the set F is (@)(E n F), then
3. POTENTIAL THEORY
where x0 is the position of the charge and where the constant 112~ has been
fixed after a certain choice of units.
More generally one defines a charge distribution to be any finite signed
measure on the Bore1 sets of H3. The potential at x due to the charge distribu-
tion is again the work required to bring a unit charge from infinity to the
point x. Since force (and hence work) are additive, the potential due to a
charge distribution consisting of charges ql, .. .. qn at points x1, . .. . CT, is
332 KNAPP
1 . 44Y)
2x J Ra IA-y! .
PROOF. It suffices to prove the lemma for the case p 2 0 since the general
case follows by taking differences. Letting K,, denote the closed ball about
the origin of radius n, we have
The inside integral on the right is bounded by its value when y -= 0, which
is some finite number c. Thus the right side does not exceed cp(R3)/2a < 03,
and g must be finite a.c. in I;, . Hence g is finite a.c. in H3. Q.E.D.
1 1
44)
% I R3 Ix --yI
is called the potential of CL. The operator transforming a charge into its
potential is called the potential operator.
Potential theory is the class of theorems relating charges and potentials
and quantities definable in terms of them. It includes the subject known
in physics as electrostatics since the quantities commonly arising in electro-
statics are all definable in terms of distances, charges, and potentials. It is
readily verified, for instance, that capacity, energy, field, and force are all
definable at least dimensionally in terms of distance, charge, and potential.
As a further indication that the class of theorems relating charges and poten-
HROWNIAN MOT109 AND POTENTIAL THEORY 333
tials is large, we give four examples of how concepts may be defined explicitly
(and not just dimensionally) in terms of distance, charge, and potential.
1. The total charge of a charge distribution p is p(R3). The support of p
is the complement of the union of all open sets li’ with the property that
p vanishes on every measurable subset of U.
2. \Ve can reasonably ask what the total amount of work required to
assemble a charge distribution is if only an “infinitesimal” amount of charge
is brought into position at one time. The way to compute this amount of
work is to integrate the potential function against the charge distribution,
provided the integral exists. Thus we define the energy of a charge distribution
to be the integral of its potential with respect to the charge, provided the
integral exists.
3. If a total amount of charge q is put on a piece of conducting metal in
R3, the charge will redistribute itself in such a way that the potential is a
constant on the set where the metal is. The situation where the potential is
constant on the metal is the one which minimizes energy among all charges p
with total charge q and with support on the set where the metal is, and this
situation is referred to as equilibrium. \Ve define an equilibrium potential
for a compact set I:’ to be a potential which is 1 on E and which arises from
a charge with support in E. An equilibrium set is a set which has an equilibrium
potential. One can show that such a potential is unique and that the charge
producing it is also unique.
4. The capacity of a conductor in R3 is defined as the total amount of
charge needed to produce a unit potential on the set where the conductor is.
We thus define the capacity of any equilibrium set to be the total charge
of the charge distribution which produces the equilibrium potential.
,I- 1
= -
o (2&p’
e-l~-vl*/*t dt
1
dp(y).
du dp(y).
i x - y 1-l emu2jz
1
By monotone convergence
Q.E.D.
(QtfXY)
= jR3& e-12-ui’r2tf(x)
dx.
Moreover, Qt satisjies the ~following properties:
(1) I;Q”ii = 1
(2) p-t = QSQ’
(3) strong lim Q’f = f
t10
(4) strqntlim Q!“f = 0
(5) Iff 5 0, then Qy > 0.
R3
(27rt)-3/* e--lz--yIei*tf(x)
1
dx dp(y).
Hence (Q”f)(y) equals the integrand on the right side. Then (I) and (5)
are obvious.
Next we prove that Q” has range in Co , We have
As y --t y. , the right side tends to zero by Lebesgue’s Theorem, and hence
Qtf is continuous. In addition,
The first term is small for large N since f vanishes at infinity. For fixed N
the second term tends to zero as y + 00. Hence Qy vanishes at infinity.
336 KNAPP
We still have to prove (2), (3), and (4). For (2) we have
‘s -y’2
X exp
[
-
2r 1f (4 dx dr
= (2xs)-3i2 ('&4-3;2 j,, j,, exp [ _ 'y - kf + xs)@ + ')I* ]
2st,/(s+ f)
For (3) let l > 0 be given. Choose 6 > 0 so that supV: f(x + y) -f(y)/ < E
for all x with 1 .r / < 6. Then
< E + 2 11
f Ilrnj,r,,s,z (27~t)-~/~e-lz12~2L
dx
/
For small enough r, the right side is <2c. In (4) let c > 0 and choose M
so that 1f(x)] < E when x is outside the ball B =: {xl I x 1 < M}. Then
From now until after Proposition 4.13, we assume that {Q’} is any family
of operators mapping C,, into C,, and satisfying statements (l), (2), and (3)
BROWMAN MOTION AND POTENTIAL THEORY 337
Af = strong lim
t10 (
defined on the vector subspace of all f E Co for which the limit exists.
PROOF.
PROOF. For h > 0 we have by Lemma 4.3 and property (2) of {Qt}
II
Q”+“f - Q’f - Q’Af
h
ii = 11Q”-” ( QYk-f) - Q’(Af) 11 by (2)
The first term on the right tends to zero as k JO since f E domain A, and
the second term tends to zero by (3). Q.E.D.
If h = -k < 0,
<I 0’ I QW + h) -fWll dt
T
<
i -fW
” ‘if(y + h) dt bY (1)
= T IlfbJ + h) -f(r)ll,
and the right side tends to zero as h -+ 0 by uniform continuity off. Next
we prove that j’Q”fdt vanishes at infinity. Let l > 0. Choose by (3) a
6 > 0 such that’if 0 < t < 6, then )I Q,“f -f ‘I <:I c. Then for every n and
for 0 < f < 6,
For each n such that nS < T, let Y,, be a real number such that if 1y / > Y,, ,
then I(Qndf)(y)I < E. Let Y be the maximum of the Y,‘s. Then 1y I > Y
implies 1 Qtf(y)l < 2.~ for every f E [0, T] and hence I S,‘Qtf(y) df I < 2rT.
Q.E.D.
LEMMA 4.1. If f E Co , then
Q”f ds = Qtf.
PROOF. Let E > 0 and choose h > 0 by (3) so that Y < h implies
IIQ’f -f II <E. Then
t+h
<$
I t
IIQY-QYllds
<; I :+A
t IIQ”ff-fllds
-=c 6. Q.E.D.
BROWNIAS MOTION AND POTENTIAL THEORY 339
PROOF. Set g - strong lim Af,, . Since IjQ”(Af,,) - QLgjj < 11Afn -g;l,
QlAf,, converges in norm uniformly in t to Q’g as n + CO.Therefore for
fixed h we have by I,emma 4.4 and by (1)
h = - j’Q”+tfdt - $ /‘Qydt
h o 0
1
= -lTfhQffdt -;(-Qydt
h h
1
= - lTthQffdt - ; J;QYdt.
h T
PROOF. Let h,, JO and let f be given. Set fn = h;’ s,““Vydt. Then
f,, E domain -4 by I,emma 4.9 and strong lim fn = f by Lemma 4.7.
Q.E.D.
340 KNAPP
exists for each /\ > 0 and for all Jo Co , and RA is a bounded linear operator
from Co into Co of norm < 1 /A.
IS T
e-““Q”j dt
I
<
s T
cAt I (?‘I! IIjil dt < il j,l ce-Af dt + 0 as 7’ + CO,
=- 1 cce-Aft-h)Q!Lf dt - k [ e-AtQfj dt
h I A
,y+h- 1 m .A
=- e-htQtj dt - k e-W-h)Q’f dt
h I o J0
+hR,j-jashJ0.
for all f E Co .
BROWNIAN MOTION AND r'OTENTIAI. TIIEORY 341
.4(/\ - A)-’ :- (A -. h)(h -4)“ -{- h(X - A)-’ : h(il .-- A)-’ -- I, (4)
and the operator on the right is bounded; hence Q: is defined and bounded.
Its norm is
For the rest of the proof it is sufficient to prove Eq. (3) for the dense subspace
domain A (see Lemma 4.10). In fact, letj be given and choose jn E domain A
with j = strong lim jn . Then
Since Q”Aj = AQ’j for Jo domain A (Lemma 4.3) and hence since
(A - A)QL = Qt(h - A) on domain A, the relation range (A - A)-’ =
domain (A - A) = domain A implies that
We use Eqs. (4) and (6) together as follows. For jrz domain A
Hence I; h(X - :I)-*j -f,l --r 0 for f~ domain XI. The same argument as
in Eq. (5) except with Q,t and Qf replaced by h(h -- .4)-l and I, respectively,
shows that consequently ! X(h -. .q)--tf ---f 1’ --f 0 for all f E C’, . Applying
this result to .4f for ,f~ domain -4, we have
limAsup 1 /If - M(X .-- .4)-‘.fl : lim;up 1;Af - X(X - A)--‘(Af)l: 0. (8)
Sow we can prove Eq. (3). By Eq. (6) the second step of the following
calculation is justified:
The content of Proposition 4.13 for the present purpose is that -4, the
three formal properties of {Q1}, and the definition of A in terms of {Q’}
completely determine {Qt}.
For the fourth step in the recovery of {P”} from the potential operator,
we introduce an operator G and prove that its inverse is ---4. We define
and the right side converges uniformly to --f by (4) of Proposition 4.2.
Hence by Lemma 4.8, c;fe domain il and A(C;f) -..: -J Q.E.D.
\Ve shall now identify .1 as a constant multiple of the Laplacian operator
‘2
y’” & : 3- ..- __3
/. ‘* 2Xz2 2x,2
when each is restricted to a suitable domain. This result is due to Yosida [ 161.
LEMMA 4.15. If p,,If~ domain .q for mery t >, 0 und q :I(Qy) = (,)‘g,
then f E domain A and ‘3f g.
PROOF. \Vc have strong lim,l, Q’f .: f and strong lim,,, .-l(Qy) = g.
Apply I,emma 4.8. Q.E.D.
\Ve use the notation Lr to denote the space of all Lebesgue integrable
functions on H3, and we denote ‘;f j:, = s If(x)! d.r.
LEMMA 4.16. IffEC,,nLl, then $Y~E domain .,I for f > 0, and
1 -
I [1 g(t $- h, x - y)f(x)
Fl R”
dx
I,et S be the set of testing functions of Schwartz; that is, the set of real-
valued infinitely differentiable functions on R3 which, together with their
partial derivatives of all orders, remain bounded when multiplied by any
polynomial. WC have S C C0 r\ L1 . The Fourier transform is a one-one
map of S onto itself.
= 2q(2Tt)-3/2e- Is-Yl*jet].
A(~tf)(r) = j 2V2[(2nt)-3~2e-‘2-y”~2t]
f(x) dx.
R2
Applying Green’s identity twice and using the fact that f E S, we get
Choose 6 so that the first term on the right side is < 4/2 and then choose y
(by Lebesgue’s Theorem) close enough to y0 to make the second term
< E/2.
It remains to be proved that g(y) --f 0 as y --f cc. Choose R large enough
so that (27~)-~R-~/lf& < c, and choose Y < R small enough so that
(1) ilQtll = 1
(2) Qstt = QSQt
(3) stro:lgolim Q”f = f forallfEC,
Both Brownian motion and potential theory are often studied in other
domains than all of R3, such JS all of Rn or some open set of R”. In this
section we shall point out what parts of the preceding carry over to the
n-dimensional theory, but we shall treat only the case where the underlying
set is all of R’J.
The n-dimensional Brownian motion transition operator Pt is defined by
- jR,Ix-Y~4(y) in dimension 1
2jR~Wx-y144y) in dimension 2
where
c, = 1/&r”‘T(‘/*(n - 2)).
BROWNIhU ?dOTION AUD POTENTIAL THEORY 347
of Theorem 4.1 is still the heart of the connection, and the theories are still
equivalent.
But the situation is different in dimensions 1 and 2. Proposition 5.1, whose
proof is omitted, is what replaces Theorem 4.1, and it is not sufficient to
demonstrate that the potential operator can be obtained in a natural way
from the family (P}. It asserts that the exact analog of Theorem 4.1 holds
for a charge precisely when its total charge is zero and its potential is finite a.e.
?f PF) # 0, then
!E J“ (@‘)‘(x) dt = + oo OY ---co a.e.
0
Ifp(R”) = 0, then
g(x) = vz I’ (pP)‘(x) dt
0
exists a.e., is finite a.e., and satisjies
if n=l
if n=2.
348 KNAPP
In the converse direction most of the results of Section 4 are still valid
after changes in certain constants. Everything from Proposition 4.2 through
Proposition 4.17 goes over with little change. Lemma 4.19 is still valid, and
only Lemma 4.18 and Theorem 4.20 break down. The result that replaces
Lemma 4.18 is Proposition 5.2, and it is insufficient to prove directly any
analog of Theorem 4.20, even for charges of total charge zero.
Or
if n=2.
I R2 If(Y)1 11% I X-Y I Id’< 00
- f R,f(Y)I x -Y I dY if n=l
g(x) =
2~R,fWWx-yldr if n=2.
REFERENCES
11. G. A. HINT, Markoff processes and potentials I, II, III, Illinois J. Math. 1
(1957), 44-93, 316-369; ibid. 2 (1958), 151-213.
12. S. KAKIJTANI, On Brownian motions in n-space. Z’roc. Imp. Acud. Tokyo 20
(1944), 648-652.
13. S. KAKLTANI, Two-dimensional Brownian motion and harmonic functions. Proc.
Imp. Acad. Tokyo 20 (1944), 706-714.
14. P. LBVY, “l’heorie de I’addition des variables aleatoires,” Vol. 1. Gauthier-
Villars, Paris, 1937.
15. K. YOSIDA, On the differentiability and the representation of one-parameter
semigroup of linear operators, J. Math. Sot. Jafian 1 (1948), 15-21.
16. K. YOSII)A, An operator-theoretical treatment of temporally homogeneous
>,larkoff process. J. Math. Sot. Japan 1 (1949), 244-253.