Communications in

Commun. Math. Phys. 84, 483-503 (1982) Mathematical

Physics
© Springer-Verlag 1982

Hydrodynamics in Two Dimensions and Vortex Theory*

C. Marchioro 1 and M. Pulvirenti 2
1 Dipartimento di Matematica, Libera Universita di Trento, 1-38050 Povo (Trento), Italy
2 Istituto Matematico delΓUniversita di Roma, 1-00100 Roma, Italy

Abstract. We consider the Navier-Stokes equation for a viscous and incom-

pressible fluid in R 2 . We show that such an equation may be interpreted as a
mean field equation (Vlasov-like limit) for a system of particles, called vortices,
interacting via a logarithmic potential, on which, in addition, a stochastic
perturbation is acting. More precisely we prove that the solutions of the
Navier-Stokes equation may be approximated, in a suitable way, by finite
dimensional diffusion processes with the diffusion constant related to the
viscosity. As a particular case, when the diffusion constant is zero, the finite
dimensional theory reduces to the usual deterministic vortex theory, and the
limiting equation reduces to the Euler equation.

1. Introduction
In this paper we deal with an incompressible, viscid or inviscid fluid in two
dimensions and study the connection between the equations governing the motion
of such a fluid and the vortex theory. Furthermore we investigate some aspects of
the hydrodynamical equations that, in particular, will suggest a quite natural
proof of the existence and uniqueness of the solutions for a wide class of initial
conditions.
It is well known that an incompressible and viscous two dimensional fluid,
under the action of an external conservative field, is described by the following
evolution equations

~(

du2 dtti
ω(x, t) = curl w(χ, t)=- — , (1.1)
oxί ox2

* Partially supported by Italian CNR

0010-3616/82/0084/0483/S04.20
484 C. Marchioro and M. Pulvirenti

where x = (x 1 ; x 2 )elR 2 , u = (u l 5 u 2 )eIR 2 is the velocities field, v^O is the viscosity

coefficient, V = \~—, - — , A = V V is the Laplace operator.
\dxί dxj
In the following we shall denote Eq. (1.1) as the NS equation (Navier-Stokes)
when v > 0 and as the E equation (Euler) when v = 0. Introducing the operator V1
d d \
- — , - — , by virtue of F w = 0, one obtains
7X2 OXJ

u(x, t) = j (Vλg) (x - j;) φ , t) dy, (1.2)

where g(r)= — —-Inr is the fundamental solution of the Poisson equation.

2π
The E equation is related to the so-called vortex theory [1]. Let us consider a
system of N particles (vortices). To each particle is associated a vorticity intensity
α elR, i=l...JV. Denoting by xt the position of the i th vortex, we consider the
following initial value problem:
N

(1.3)

lϊx1 ...xN^>x1(t)...xN(t) is the solution of the initial value problem (1.3), (assuming
that such a solution exists and is unique) then the signed measure ωf(dx)
N

= Σ a^^^idx) [here δJ^dx) denotes the Dirac measure based on x] satisfies the

following identity:

^ ω f ( / ) = ωf[(Mo-P)/] (1.4)

for all sufficiently regular /, where

uo(x,t)= $(VLg)(x-y)X({x*y})ω?(dy), (1.5)

otherwise.
In fact, by a direct computation,

i= 1

Σ aι Σ β/
i=l j*i

ωf[(M0 F ) / ] . (1.7)
Vortex Theory 485

Thus Eq. (1.4) is a weak version of the E equation, except that the self-energy factor
(that would give an infinite contribution) is neglected in the definition of the
velocity field.
The problem of finding a rigorous connection between the evolution problems
(1.3) and (1.4) in case v = 0 arises quite naturally. What is expected to be true is the
following. Let ω be a sufficiently smooth initial condition for the E equation and
N

suppose that ω is approximated, in some sense, by ωN = Σ aiδXι [xi initial

ί=i

conditions for the initial value problem (1.3)]. Then ωf would approximate ωp the
solution of the E equation with initial datum ω.
Although such an approximation problem has been widely investigated for
numerical purposes (see [2] for example), as far as we know it has not been solved
(see [3, p. 110]). In this paper we treat this problem, also in the presence of
viscosity, giving a precise connection between the NS equation and a generaliza-
tion of the vortex theory.
We describe the basic idea. The E equation in two dimensions is described by
characteristics. In fact, by a direct inspection of the equation, we realize that each
molecule of vorticity evolves under the force field generated by all other molecules.
Such an idea has been used by Kato (and previously by other authors) to obtain an
existence and uniqueness theorem for classical solutions of such an equation [4].
The presence of the Laplace operator in the NS equation suggests the use of
stochastic characteristics in place of deterministic ones. Such a natural idea has
already been used in [5] to give a derivation of the NS equation 1 . Other attempts
in this direction, different and independent from ours, are going to be developed
[6].
Our approach is the following. We first deal with a regularized version of the
initial value problem associated to (1.1), i.e. we replace g by gε, a smooth function
different from g in an ε-sphere around the origin. For such a new initial value
problem we prove existence and uniqueness of the solutions, by the use of the
contraction principle (Sect. 2).
The techniques used in Sect. 2 are inspired by the approach proposed by
Dobrushin in [7] for the Vlasov equation that is very similar to the E equation.
The only difference with our case is that we deal with stochastic differential
equations in place of ordinary differential problems, in order to take into account
the viscosity term. The main part of the results of this section have also been
obtained by McKean [8] in a different physical context.
In Sect. 3 we remove the cutoff, proving that the above solutions are a
convergent sequence when £->0. In particular we obtain, for a large class of initial
conditions, an existence and uniqueness theorem by means of a constructive
procedure.
In Sect. 4 we show that a system of N vortices, interacting via a potential g&
under a stochastic perturbation, has a solution which converges to the solution of
the NS equation in the limit N-+oo and ε = ε(JV)^0, once the initial datum of the
vortex systems approach the initial datum of the NS equation, at time zero.

1 It is also the basis of a numerical method due to Chorin for the study of slightly viscous flow: J.
Fluid Mech. 57, 785 (1973)
486 C. Marchioro and M. Pulvirenti

All the results we obtain hold for the NS equation as well as for the E equation,
all our estimates being independent of v, with the exception of Sect. 4 in which a
stronger result is obtained for the deterministic case.
We finally remark that from the point of view of our approach, we are
interested in weak solutions of our initial value problems. In the case of sufficiently
smooth initial conditions, the uniqueness guarantees that we actually obtain
classical solutions, which, as it is known, do exist.

2. Regularized Problem
Let M be a separable and complete metric space, with a uniformly bounded metric
d\M x M->IR"\ Let Jί(M) be the space of the Borel, signed measures on M and
Ji + {M) its positive part. For all με Ji(M) let us denote μ + and μ~ the positive and
negative part of μ according to the Jordan decomposition of μ and ||μ||
+ \dμ~ its total variation. We introduce the following spaces

Ji+(M;a,b)={μεMM)\\\μ+\\=a,\\μ-\\=b}.
A metric topology can be given on Ji{M,a,b) in the following way. Let
μ,veJi+(M, 1) and ^(μ, v) be the set of all joint representations of μ and v. We
recall that a joint representation P is a probability measure o n M x M such that
$P(dx,dy)f(x)=$μ(dx)f(x),
\P(dx,dy)f{y)=\v(dy)f{y),
for all Borel functions / .
We define:
Rϊ{μ,v)= pmiJP(dx,dy)d(x,y). (2.3)

Rλ is a metric on Jί + {\) and induces a topology that is equivalent to the weak

convergence topology [9]. The metric R1 is called the Kantorovich-Rubinstein
(KR) distance and will play a basic role in our analysis.
For all μ, veJί±(M,a) we define

,), (2.4)
\a a)
and if μ, ve Jί(M, a, b) we put

(2.5)
The topology induced by the metric (2.5) is equivalent to the weak convergence
topology on Jί(M, α, b) and makes it separable and complete. We specialize the

above considerations by putting M = R 2 , d{x,y) = min(\x — y\,ί\ | x | = / £ |xj2.

y α= l
Since the metric d is topologically equivalent to the usual metric, the topology
induced by £ in ^#(IR2 α, b) is equivalent to the usual weak convergence topology.
Vortex Theory 487

+ + + + 2
For notational simplicity we put Ji =Ji {β?\ Jΐ {a) = Jί (1R ,al Jί(a,b)

With the above tools we shall treat a regularized version of the NS equation.
2
Let g be the fundamental solution of the Poisson equation in 1R and gε defined as
1
gε(r) = g(r) if | r | ^ ε and arbitrarily extended to an even C^IR ) function such that
L 2
\gE(r)\S\g'(r)\, |ffe"(r)|g|0"(r)|. Defining Kε(x) = (V gε) (\x\), xeR we have:
+
Theorem 2.1. Let ωeJi{a ,a~) and vϊ O. There exists a unique function
R + Bt-*ωεteJ?(a+,a~) with the following properties: for all /e^^IR 2 ), such that
2
IF/I and ΔfeLJR )

(2.6)
ω'0
where
uε{x,t)=\Ke(x-y)of,{dy). (2.7)

Moreover, for all Borel sets ^4cIR2,

ω>(A)=\PεM\y)ω(dy), (2.8)
ε
where P t( \y) are the transition probabilities of the diffusion process, solution of the
stochastic differential equation
dx{t) = uε(x{t), t)dt + σ dw(t) (2.9)

with — = v.

Proof. Let T > 0 and Ji(a*\a T) the space of the continuous functions

[0,T]9i-^^e^(a+,a~). (2.10)

Jί(a+,a~ T) is a complete metric space with metric function given by

Rτ(μl9μ2)= sup R(μϊ,μf), (2.11)

te[0,T)

where ft = {μί} t e [ O f Γ ] .
Let (Ω, Σ, P) be a probability space (such space will not play any significant role
in the rest of the section) and w a standard Wiener process on it.
We consider the processes solutions of the following stochastic differential
equation

dxjίt) = u[(x (ί), ήdt + σ dw(t), (2.12)

where
uiε(x,t)=^Ke(x-y)ι4(dy) (2.13)

and μt = {μ\}tlBίo,T]eMa+,a~ T), i=ί,2.

488 C. Marchioro and M. Pulvirenti

Denoting by x-(ί, x) the unique process solution of (2.12) starting almost surely
from x at time zero, and by E the expectation with respect to P, we have the
following estimate

^t, x), x2{t, x))] ^ (aLε exp2αLεT) J ds R(μl, μ 2 ), (2.14)

o
where L ε = max(Lε, 2maxi£ ε ), Z ε is the Lipschitz constant of Kε and

Proof of (2.14). With the above notation we have \Kε(x1 — y) — Kε(x2 — y

^Lεd{xvx2). Hence, <P=1>,

d(xx(t, x), x 2 (ί, x)) ^ |x x (ί, x) - x 2 (ί, x)|

Kx^s, x), 5) - wε2(x2(5, x), s)\

0
ί

g(xt(5, x), 5 ) - ul(x2(s, x), 5)| + \ul(x2(s, x), 5)-t/ε2(x2(5, x), s)\}
0

l(dy) [Kε(Xl(s,x)-y)-Kε(x2(s,x)
0

+ If Kε(x2(s, x) - y) [μl(dy) -

0
x 1 (s, x), x2(s, x))
ί

j ds φc^s, x), x2(s, x))

0

+ s
\ds{a lP +(dy1,dy2)\Kε(x2(s,x)-y1)-Kε(x2(S,x)-y2)\
0

,2± ,,2±N

for all

ί, x), x 2 (ί, x))] g 2αLε (f d s E E φ ^ ί , x), x 2 (ί, x))]

lo

(2.16)
b J
that implies (2.14).
+ +
We define the following map Jί{a ,a" T)^Jί{a ,a~ T),

A={ft} f e [ o,n- > S / ϊ ω = {SίΊω}fg[o,τ] :(Sfω)U)= $P*(A\y)oj(dy), (2.17)

Vortex Theory 489

where Pf( \y) are the transition probabilities of the process solution of (2.12) with
+ 2
μf replaced by μ. Obviously S^ωe^(a , a"). Moreover, for all fe C2(IR ) such that
\Vf\ and Δf are bounded, the following holds [10]:

~ (Sfω) (/) = (Sfω) M 'F) Π + ^ (S?ω) ( A f), _

(2.18)

(2.19)
+
We prove that the map μ-+S^ω is a contraction in Jί(a , a~ T) for small T as
consequence of (2.14). Putting ω\ = S^ω, we define Q^(dxvdx2) via the relations:

±
j Qf{dxv dx2) f{xv x2) = —^ J ω (Jx) E^fix^t, x), x 2 (ί, x)] . (2.20)
a
ι± 2± 1
ω ω \ _ . . . * , , ω _ ,
Then Q^eΉl—r-,—+- . Thus, integrating both members ofΛ (2.14) with
\a- a~ j a
and summing
R(ωl,ω?) ^ (2αLε exp2αLεT) j ds ^(μ,1, μ s 2 ). (2.21)
o
From (2.21) we obtain the existence of a unique fixed point of the map μ-+S^ω
for small T and hence a unique solution of the initial value problem (2.6). Such a
solution may be extended to arbitrary times so the proof is complete. •
Remark. Theorem 2.1 deals in particular with the regularized Euler problem. In
this case
x)dx9 (2.22)
where y-*Tfy is the solution of the initial value problem

d _

(2.23)

Proposition 2.1. Lei μ e ^ ( α + , α ~ T), ω e ^ ( α + , α ~ ) , i = l , 2 . T/zeπ

R(Sfav S^ω2) ^ (exp2αLεί) JR(ωl5 ω 2 ). (2.24)
Proof. Denoting by x(ί, x) the process solution of

dx(t) = <(x(ί), ί) dt + σ dw(ί) (2.25)

starting almost surely from x at time zero, we have, <P= 1),

d(x(t, x), x(ί, j/)) ^ |x - y | + J ds\u*(x(s> *)>s) ~ UMS> y)> s)\

S \x - y\ + I ds\$ μs(dz) {Kc(x(s, x)-z)- Kε(x(s, y) - z)}\

0

ύ\x-y\ + 2Lεa } ds d{x{s, x), x(s9 y)), (2.26)

o
490 C. Marchioro and M. Pulvirenti

and hence
£[d(x(ί, x), x(ί, y))] ^ |x — y\ exp 2αLεί, that implies
(2.27)
E[φc(t, x), x(ί, 3;))] ^ d(x, y) exp2αLεί
because E[d(x(ί, x), x(ί, 3;))] ^ 1.
We define P^ by
J P^ίdx, dy) /(x, y) = J P * ^ , dy) £[/(x(ί, x), x(ί, y))] (2.28)

for P ± e ( ^(—ψ-,—1-|. By direct inspection

Thus the thesis follows by integrating (2.29) with P ± . •

The next theorem describes the continuity property of the regularized solutions
with respect to the initial conditions.
Theorem 2.2. Let ώε and ξεeJi(a+,a~\T) be two solutions of the initial value
problem (2.6) with initial conditions ω and ξ respectively. Then
Rτ(ώ\ ξε)^R(ω, ξ)exp[2αL ε (T + e 2 α L ε T )] . (2.30)
Proof.
R(ω*v ξl) S R{ωεt, Sfω) + R(Sf'ω9 ξεt). (2.31)
By virtue of (2.21)

R{ωBt9 ξε) S (2aLε exp 2aLεt)} ds R(ωεs, Q, (2.32)

0

and by Proposition 2.1

R{S?ω9 ξε) g(exp2αL ε ί) R(ω, ξ). (2.34)

The thesis follows combining the above estimates. •

3. Construction of the Solutions

In this section we shall obtain weak solutions of the NS and E equations as limit of
the regularized solutions as ε-»0.
Theorem3.1. Let ωeZ^nL^QR 2 ), ωdxeJf(a+,a~), v^O. There exists a unique
solution ω-±ωteL1nLoo(]R2\ ωtdxeJΐ(a+,a~) of the initial value problem (weak
form of the NS and E equations)

jtωt(f)=ω,[{u-V)n+vωt(Af)
(3.1)
Vortex Theory 491

2 2
for all /eC2(IR ), such that |F/|, ΔfeLJJR ), where
u{x,t)=\K{x-y)ωt{y)dy
λ 2 (3 2)
κ{x)={V g)(\x\), xeR . '
Moreover
ωt{f)=\imω*{f) (3.3)

for all fe C n L ^ ϊ R 2 ) and the above limit holds uniformly in v and t on compact sets.
Finally, there exists a unique solution of the following stochastic differential
equation
dx(t) = u(x{t\ ήdt + σ dw(t) (3.4)
once specified the distribution of the random variable x{ ), and denoting by Pt(dx\y)

the transition probabilities associated to the solution of (3.4), it results v = —

lω1{x)dx=\Pt(A\y)ω{y)dy. (3.5)
A

We need a preliminary lemma (essentially Lemma 1.4 in [4]), whose proof will
be postponed to the end of the section.
Lemma 3.1. Let t->ωt(x) be a family of LίnLao(ΊR2) functions such that

sup H ω J I ^ ^ , sup \\ωt\\ίSA2. (3.6)

ίe[O,T] te[O,T]

Then, for all 8^0 and £e[0, T],

ϊ\ωt(y)\\Ke(x-y)\dy^c1{Ai+A2), (3.7)
J|ω f ()0|\K ε (x-y)-K ε (x'-y)\dy ύ2Cl{Ax+ A2)φ(x9x'), (3.8)
where
ψ{x,x') = φ{\x-x'\),
ίr(l-lnr) if 0<r<l
( 1 9 )
if r i s
i.
Proof of Theorem 3.1. We first observe, by general properties of the diffusion
processes with smooth drift, (see e.g. [10]) that, if ωeL1nLOD(ΊR2), then

Let xε(ί), xε'(t) be two processes constructed in Theorem 2.1, with ε > ε ' > 0 , and
starting from xelR 2 at time zero almost surely, then <P—1),

d(x\ή, x°'(ή) g } ds\uε(xe(s), s) - uAxE'(s), s)\

0

+ \uε(xε'(s),s)-uE,(xε'(s),s)\}. (3.10)
492 C. Marchioro and M. Pulvirenti

By virtue of Lemma 3.1

ε ε ε ε
\uε(x (s\ s) - uε(x \s\ s)\ S c{ω) φ(x (s\ x \s)), (3.11)
where
c(ω) = 2c 1 (||ω|| 1 + | [ ω | | J . (3.12)
Furthermore
ε ε
\uε(x Xs)9s)-uε,{x '(s),s)\
ε ε
= If dy ω%y) Kε(x \s) -y)-\dy ωi{y) Kε{x \s) - y)\
ε ε
ύ\\dy{ωly)-ω ;{y))Kε{x Xs)-y)\
ε ε
+ lί dy ωΐiy) {Kε(x '(s) -y)- Kε,(x '(s) - y)}
ε ε ε
ύ\ldy(ω s(y)-ω ;(y))Kε(x \s)-y)\
+ c2\M^ f dy\y\~\ (3.13)
e e
for some c 2 > 0 . Hence, using the notation -E x [/(x (i))]=£[/(x (i,x))], for some
c 3 >0, we have

S } ds\c31|co||roε + c(ω) f^-^ dx £ x [φ(x ε (s), xε'(s))]l

α
O l J

c*'(s)-J')l] (3.14)

Since
j dy ωl(y) Kε(xe'(s) - y) = j dy ω(y) £,[K e (x e '(s) - /(s))], (3.15)
we obtain
J5*D J dy(ω%y) - ofs{y)) Kε(x°'(s) - y)\]
= EXU j dy ω(y) Ey[Kε(x*'(s) - /(s)) - Kε(x°'(s) - / ( s )

|^ j y
. (3.16)
Thus
ω~(y)

• £ Jj dx ~β EMXΛs) ~ f(s)) - Kε(x*\s) - /(s))\)

:-/(s))-Xε(x-/(s
J

(3.17)
Vortex Theory 493

Defining, finally:

1
ί ^ ^ } ϊ ( 3 . 1 8 )

Y2(ε,e',ή= \dx^~β + ^-^Exlφ(x\t),x^tm, (3.19)

we obtain
f
]sY2{εiε 9s)9 (3.20)
where

c(ω) = c(ω) + a\c\-~r +c\~^ \ (3.21)

[ \a j \a I)
We now want to estimate Y2 in terms of Yv Since
φ(x^ y) = φ(d(x, y)) (3.22)
by the concavity of φ and Jensen inequality,
7 2 (ε,ε / ,ί)^2φ(y l (ε 5 ε',ί)). (3.23)
Let h(x0, t) be the solution of the following initial value problem {B = 2c(ω)):
z(t) = Bφ(z{t))\
r (3.24)

Then if x o < 1,
-e~Bt) if h<\
= l+B{t-t0) if fe^l,
where

and if x 0 ^ 1,
h(

From (3.25) we have lim h{xo,t) = 0 and since

ω|| oo ίε,ί) (3.26)

by (3.20) and (3.23) we obtain

lim sup R(ωεnωεt) = 0 (3.27)

ε-+0 te[O,T]
ε' <ε

because R(ωεt, ωf) ^ Y^ε, ε', ί).

Thus we obtain the existence of a family of weakly continuous limiting
+ +
measures {ωt{dx)}te^eJf (a ,a~). By the estimate ||co^{|00 ^ IMI^ and Kω^l^
494 C. Marchioro and M.Pulvirenti

^\\ω\\v we conclude that ωt(dx) is absolutely continuous with respect to the

Lebesgue measure and denoting by ωt its density,
IKUiHIωlU, KIL^UωlL. (3.28)
We claim that the map ω-+ωt is a weak solution of the NS equation. We first
observe that u(x, t) = J K(x — y) ω(y) dy exists because (3.28) and Lemma 3.1.
Furthermore
u(x,t)= limκfi(x,ί). (3.29)
ε->0
In fact

|u(x, t) - ue(x, ί)| = |f {K(x - y) ωt(y) - Kε(x - y) ω%y)} dy\

||ω||00+ J K(x-y) {ωt(y) - ω*t(y)} dy. (3.30)

\χ-y\>η

The last term in (3.30) goes to zero as ε->0 by (3.27) and standard arguments. Thus
(3.29) follows since η is arbitrary.
Moreover

ωί[(« F ) / ] = l i m ω ? [ ( M ί - F ) / ] (3.31)
ε->0
2
for all feC^lR2) such that
Proof of (3.31).

|ωf[(M P) / ] - ω?[(uε V)Π\ S \ω&u V) / ] - ωfc[(M • F) / ] |

+ |ωtε[(u V) Π ~ ωεtl(ue V)/]|. (3.32)

The first term on the right hand side of (3.32)—»0 because of (3.3) since (u- V) f is a
continuous and bounded function by virtue of (3.28) and Lemma 3.1.
The second term on the right hand side of

(3.32) ύ\\Vf\\J dx\ω\{x)\ |M(X, t) - uε(x, t)\

f |tt(χ,ί)-«β(χ,ί)l
\x\<r

C(ω) f dx|ω;(oo)| X({|x| ^r}), (3.33)

where X({ }) is the indicator of the set {}. Then, j ...dx~+0 because of (3.29)
\x\<r
and the dominated convergence theorem. Hence by (3.3),

Iimsup2 t h term of (3.22)^ WVfW^φ) jdx|ω t (x)|

(3.34)
X({|x| >r})^O(r),
where
limO(r) = 0 (3.35)
r-> oo

because ω ί eL 1 (lR 2 ). This proves (3.31).

Vortex Theory 495

By the uniform boundness (in ε and t on compact sets) of ω\[(u V) / ] and

ωε( A f) we obtain that ωt satisfies the NS equation in the integral form and hence
by weak continuity in the differential form (3.1).
Let t-*ωt a weakly continuous family of L1nL00QR2) functions such that HωJ^
^llωol^JωJL^IKIL.
Defining
u(x, ί) = J K(x - y) ωt(y) dy, (3.36)
we prove that the process solution of
dx(t) = u{x(t\ t) at + σdw{t) (3.37)
may be defined. To this purpouse let us introduce the Banach space 3d of all
processes ξt defined for ίe[0, T] such that \\ξ\\ < oo where

| | £ | | = sup E(d(ξ{t))), d{x) = d(x,O). (3.38)

ίe[O,Γ]

The processes starting almost surely form x at time zero and satisfying
dx\t) = ΰε(xε(t\ t) dt + σdw(ή, (3.39)
where
ΰε(x, t) = J Kε(x - y) ωt(y) dy (3.40)
are a Cauchy sequence in 8$. By Lemma 3.1 we have <P = 1>,

d{x\t) ~ xε\t)) ^\ds\ ωs(dy) \Kε(xε(s) -y)- Kε(x\s) - y)\

o

^ } ds{\ ωs(y) \Kε(x\s) -y)- Kε(xε\s) - y)\ dy

0

+ j ωs0>) \Kε(x*'(s) -y)- K,(x?'(s) - y)\ dy}

t

^jdsc{ω0)φ(\xε{s)-xε'(s)\) + c2\\ω0\\O0 j dy\y\~x. (3.41)

o |y|^fi

Proceeding as above we get E(<i(xε(ί) — xε'(t))) ^ fi(O(ε), ί) and hence the existence of a
ε
process such that x(t) = limx (ί) in J*. Moreover the differential of x(t) is w(x(ί), t) dt
+ σdw as follows by the use of the same arguments leading to (3.41). To complete
the proof it remains to show the uniqueness of the solution and of the process we
have constructed. Let ω-^ώt be another solution of the NS equation such that
ώtEL1nLa0{]R2). Proceeding as in the existence part one can prove that
R(ώvωε)->0 and hence the uniqueness follows. In the same way one proves the
uniqueness of the process. If x(t) is another process satisfying (3.37), by (3.41) it
ε
follows that || 3c — x ||—>0 and this concludes the proof. •

Proof of Lemma 3.ί. By definition of gε there exists a constant c > 0 such that

dxf x
496 C. Marchioro and M. Pulvirenti

Hence

(3.43)
To prove (3.8) we put r = | x - x ' | and Λ = {y\\x-y\^2r}. If r>ί then (3.8)
follows by (3.7). Otherwise

left hand side of (3.8)= J + j dy{\ωt(y)\\Kε{x-y)-Kε{x'-y)\). (3.44)

A ]R?\A

We have

u u
j i c y

/. (3.45)

On the other hand, for all x" in the segment (x,x;) one has \x" — y\>^\y — x|, and
hence

I ...dy. (3.46)

The last integral is bounded by constrllωl^ and finally

llJ*. D (3.47)
2r Q

4. The Mean Field Limit

As we have seen in Sect. 1, the solutions of the vortex dynamics may be thought of
as weak solutions of the Euler equation. This allows us to investigate the
approximation problem stated in the Introduction as a continuity property with
respect to the initial conditions. In presence of viscosity the situation is more
complicated. This case will be discussed later.
We consider a system of N vortices of intensity a1,a2...%, interacting via a
regularized potential:

(4Λ)
Vortex Theory 497

Defining

μ?{dx)= Σ « Λ - ( t ) (4-2)

±
with Hμ^H = α , we have:
2
Theorem 4.1 (Mean field limit). Leί v = 0 αnrf ωeL1nLO3(ΊR ) such that
+
ωdxeJi{a ,a~). For all μ^ such that

lim μN0(f) = ω(f), /e^nLJR2), (4.3)

N->oo

and all sequence ε = ε(N) such that

limε(N) = 0

lim R(μlω)exp[_2aLε(T + e2aL*τ)~]=0, (4.4)2

N-> oo

t/zen

limμ?(/) = ω ( (/), (4.5)

ω, Z5 the solution of the Euler equation with initial datum ω.

Proof. Let ε = ε(JV) be a sequence satisfying (4.4). Defining ωf Ξ ω^(N) the solution
of the regularized Euler problem with initial datum ω [see (2.6) for v = 0], we have:
R{ωt, μ?) ύ R(ωv ωf) + R(ω», μ»), (4.6)
R(ω?, μf) g R(ω^ S^co) + R(Sf V ^ ) . (4.7)
Furthermore by (2.21) and Proposition 2.1

£(ωf, S f ω ) ^ 2αLε exp2αL ε T J d5 JR(ωf, μf), (4.8)

b
S ? V μf) ^ R(ω, μ%) Qxp2aLεT. (4.9)
By (4.7)-(4.9)
f, μ?) S R(ω, μN0) exp [2αLε( T + β 2 f l L ε Γ )] (4.10)

and hence Theorem 4.1 is proved by virtue of (3.3), that implies R(ωt, ωf) N_>oo >0,
and assumption (4.4). •
We consider now the viscous case. To simulate the solution of the Navier-
Stokes equation in terms of a finite dimensional motion, a natural idea would be to
study the vortex dynamics perturbed by a white noise.
More precisely, for all positive ε>0, we consider the following stochastic
differential equation

dX + {t)=UΪQCMdt + σdW(t)9
(4.11)
498 C. Marchioro and M. Pulvirenti

where

i=ί...N,

(4.12)

2
σ
σ
— = v,σ>0;
2iV
and W is the standard Wiener process in IR .
Equation (4.11) describes the motion of N identical vortices with positive
1
circulation given by ^ i V " and N identical vortices with negative circulation
γ
given by —a~N~ . (We assume all vortices of the same intensity for the sake of
simplicity.)
We fix a sequence of positive numbers ε = ε(]V) such that lim ε(]V) = 0. The
N-> oo
process XN(t,XN) = {xf±(t,XN)}^1 solution of (4.11), starting almost surely from
XN = (Xχ,Xΰ) is uniquely determined by a±, N, a,ndXN.
We define the following family of measures in Jί(α+,α~) for ίe[0, T] :

a+f(xΓ(t,XN))-a-f(x?-(t,XNm, (4.13)
Λ= 1

/ continuous and bounded.

In analogy with the Euler case the measure μt(- ;XN) is expected to be an
approximation of every solution of the NS equation with a suitable choice oϊXN.
The main difference with the v = 0 case is that μt(- ;XN) is not a weak solution of
the NS equation. This because a tagged vortex moves in a velocity field that is a
random variable given by the random positions of all other vortices, while in the
NS equation any path is determined by a velocity field that is a sure variable
obtained by the solution of the same equation. Nevertheless one can hope that a
phenomenon (known in kinetic theory as "propagation of chaos") could occur in
this case also. This means that when N is large and the intensity of any vortex very
small, the velocity field generated by typical vortex paths is very similar to the
velocity field computed on the average positions.
McKean proved in [8] propagation of chaos and mean field limit for a system
of particles interacting via a smooth and bounded potential. He assumed that at
time zero, the random variables xx ...xN are identically distributed. Thus his result
cannot be directly applied to our purpose.
We introduce the space Γ = (IR2 x R 2 ) N together with the σ-algebra 38^ of all
Borel sets with respect to the product topology. Given ωeLίnLoo(K2) such that

dx = ω(dx)eJf(α+9α ), we define the measure ωQ0 = ( - τ - x - ^ - on (Γ,

Finally, givenXeΓ we denote by X^ =(xf ...x^) its first N components.

Vortex Theory 499

2 +
Theorem4.2. Let v>0, ωeZ^nL^fΊR ) such that ω(dx)eJί(a ,a~). Then, for
ω^-a.a. XeΓ, there exists a sequence such that

1 +
Hmμ o (/;X^) S ΠmJV- £ {a f(x + )-a-f(χ7)}=ω(f) (4.14)

and
lim μt(f ;XN) = ωt(f), t>0 (4.15)

/or αl/ / e ^ n L ^ J P R 2 ) , where ωt denotes the unique solution of NS equation with

initial datum ω.
Proof. Without loss of generality we put σ= 1.
Let Ω = C([0, T]) be the set of all R 2 -valued continuous functions defined in
[0, T]. Ω is a metric space with metric function given by d(ξ,η) = sup d(ξ{t),η{ή).
te[0,T]
We denote by Σ the σ-algebra of the Borel sets in Ω and by Γ 00 the σ-algebra of the
Borel sets in Ω°°=(ΩxΩ) N .
Consider now XN(t,X;ρ) = {xf ± (ί,X;ρ)}f =1? ρeΩ^, the solutions of the in-
tegral equation:

[XN{t,X;ρ))i9 (4.16)

where X = {x* }t°L x e Γ. Then XN(t,X ρ) is, for all ί, a family of random variables in
(Ω00, Σ 0 0 , P$), where P^3 = ( P 0 x p o ) N a n d p o i s l h e Wiener measure (supported on
the trajectories of Ω starting from 0). Obviously xf ± (ί,X;ρ) depends only on XN
and ρ^ ...ρN. The process (4.16) is a realization of the solution of (4.11). Such a
solution may also be thought of as realized on (Ω00, Σ°°, P£), where
00
P ^ = ]^J (P x + x PxΓ) and where j Px(dη) f(η) = J P0(dη) f(η — x). Denoting it by
±
{xf (ί, I)}, then

Let us define a family of Borel measures on IR2 indexed by ξe Ω 0 0 :

i= 1

Then
Λ ί / , ^ ) - J P£(d|) /if(/ ξ). (4.19)

Finally we introduce the two dimensional process xN(t,ξ;η) defined on

(Ω, Σ, Px\ satisfying

xN(t, ξ;η) = η(t) + j ύχxN(s, ξ η), s) ds, (4.20)

o
500 C. Marchioro and M. Pulvirenti

where
ug(y, s) = J Kε(y - x) μ»(dx ξ). (4.21)
We have to estimate R(ωt,μt( ;XN)). Then
R(ωp μt( ;XN)) ^ R(ωt9 ω?) + R(ω?, μt( ;XN)), (4.22)
where ω^ = ωεt(N\ lΐε = ε(N)->0 when N->oo the first term in the right hand side of
(4.22) goes to zero by (3.3). To estimate the second term in the right hand side of
(4.22), we try an estimate for each trajectory.

R(co?, μ?{ •, ξ)) g R(ω?, ώ f ( ,ξ)) + R{&?{ • ξ), μ? ( • ; ! ) ) , (4.23)

where
ώf (/ ξ) = \dx ω(x)} P x ( ^ ) f(xN(t, ξ η)). (4.24)
By virtue of (2.21) we have

2aUT
R(ω?,ώf(-\ξ))ύ2aLεe JdsK(ω* μf( ξ)). (4.25)
o
Furthermore, realizing that xN±(t, ξ;ξ^) = xf ± (ί, ξ), we have

μf (/ ξ)= f V j v ( ^ I) /(x*(ί, I 17)), (4.26)

where

for all continuous and bounded F.

Thus, to estimate the second term in the right hand side of (4.23), we have to
compare different trajectories of the process (4.20). We have
N N
d(x (t,ξ;ηι),x (t,ξ;η2))^\η1(t)-η2(t)\

(4.28)

This implies
^(t,ξ;)7l),xΛ'(ί,|;^2))^|)7ι(ί)-)72(ί)[exp2LεαΓ, (4.29)
and hence
d(xN(t,ξ;ηi),xN(t,ξ;η2))Sd(ηι,η2)e2^τ. (4.30)

Denoting by v^( ξ) the positive and negative part of vw( ; | ) , putting λ±(dη)
= \ω±(x)dx § PJdη) and λ — λ+—λ~, we define joint representations
Q±e(£{{a±)~ιμ^±{- ξ), ( α * ) " 1 ώf*(• ξ)) via joint representations
Q ± 6«ίf((a ± Γ 1 v±( > | ) > ( α ± Γ U ± ) = ^ :

$Q±(dxJy)f(x,y)=$Q±(dη,dQ)f(xN(t,~ξ;η),xN(t,ξ~;Q)). (4.31)
Vortex Theory 501

Hence, still denoting the KR distance in Jί{Ω 1) by Rx and in Ji(Ω a+ , a") by R,

we have:

&N±
— i h μ'+~ (
a~
S inf [Q±{dxjy)d(x,y)

= inf lQ±{dη,dQ)d{xN{t9ξ;η\xN(t,ξ;ρ))

S inf ΪQ±{dη,dρ)exp2LεaTd(η,ρ)

(^#^) (4.32)

and hence
2 L T
^ ? v N ( - , ξ ) , λ ) . (4.33)

Defining λ00 = —:Γ x -3- as a probability measure on (Ω00,1"00) we prove that for

λ^-almost all ξ

weaklimv^(^ ξ) = λ±(dη). (4.34)

N-+00

To prove (4.34) we introduce the countable family of random variables on

(Ω,2;,λ ± (α ± )- 1 ):
Xt:ttk = X({η\η(t1)eAίJ=l...k}), (4.35)
\m-η{t)\\ (4.36)

where ί are rational instants in [0, T] and τ4f = [αf, b ), α , b^elR2 α?, b", α = 1,2 are
rational numbers.
By the strong law of large numbers, there exists a set Ω°° C Ω°° such that
^ ( Ω 0 0 ) - ! and for all ξeΩ™
± N χ±
ς i V A > Λ
± ^ ί i ί k ' ^ Δ ί i ί k vsί J V ^ ^ + \ t l t k
h v+ Ji)
α
ί=1

)=^~ 1 Σ iα
^ - * , )i4) = const | ί Λ - ί / . (4.38)

Combining Theorem 2, p. 514 and Remark 1, p. 513 of [10], we conclude from

(4.37) and (4.38), that the sequence v^( ξ) is relatively weakly compact for ξeΩ"0
and from (4.37) that it has a unique limit point, i.e. λ±. Thus (4.34) is proved.
Furthermore, defining for XeΓ

+ (0), η7(0)}°ί 1 = X } , (4.39)

502 C. Marchioro and M. Pulvirenti

there exists a set Γ such that ω°°(Γ)= 1 and if XeΓ:

P%(Ω(X)) = ί. (4.40)
(4.40) follows by the identity (it can be verified by a direct computation on a
function F depending only on a finite number of trajectories):
I λ»(dή) F(ή) = J ω">{dX) J Px(dή) F(η). (4.41)
In particular
\ \ (4.42)
that implies (4.40).
Then by (4.23), (4.25), and (4.33)

|(VW( , ξ\ λ) + 2Lεa j ds R{ωNs, μ?( •, | ) ) } . (4.43)

Hence

R(ω»,μ?( ξ))^R(vN( , hλ)exp2Lεa(T+e2aL*τ). (4.44)

N± N±
u \
^ - , ^ - ( . , Q J , then
N± N±
u

hence, taking the expectation with respect to P£, f o r Z e Γ , we have

R(ω»9 μ?( ;*)) ^ f P ? ( 4 ) ^(ωf5 μf( |))

2L τ
* )}, (4.45)
where O^ίX) = j Px(dξ) R{vN{ , ξ), λ). By the dominated convergence theorem and
(4.34) ON(X)-^0 and this achieves the proof for a suitable choice of ε = ε(N) by
virtue of (4.22). •
We conclude this section with some comments.
We have not fully exploited the stochastic nature of the NS equation, since we
were interested to give an unified approach to the NS and E equations. One would
expect that the regularizing properties of the diffusion processes would improve
the estimate in a v dependent way.
It would be interesting to obtain the mean field limit starting directly from the
unregularized problem (without the selfenergy factor) and by means of a con-
structive procedure in the case v > 0. As a first step one needs to study the N vortex
process. It has to be proved that collapses between two vortices of opposite sign
(that do happen), are exceptional in the sense of the Lebesgue measure, so that one
can hope that the finite dimensional processes make sense. A result in this
direction has been obtained for bounded domains [11].
Other kinds of systems may be treated with techniques similar to that used in
this paper. For example a three dimensional fluid with particular symmetries
Vortex Theory 503

admits a vortex description. A fluid in which the velocities field has a cyclindrical
symmetry and lies in a plane containing the symmetry axis preserves the same
structure at any time. Thus this system may be described in terms of a two
dimensional fluid in half plane.
Finally, an interesting problem is the study of the vortex theory and the mean
field limit in bounded regions. In this case the Green's function has to be modified
and the right boundary conditions for E and NS equations have to be taken into
account. The Euler case follows from our analysis with minor modification while
the Navier-Stokes case seems to be more complicated. This is because the right
boundary conditions on the velocity (i.e. (7 = 0 on the boundary) imply rather
complicated boundary conditions on the vorticity.

References
1. Friedrichs, K.O.: Special topics in fluid dynamics. New York: Gordon and Breach 1966
2. Hald, O., Mauceri Del Prete, V.: Math. Comput. 32, 791 (1978)
3. Saffman, P.G., Baker, G.R.: Ann. Rev. Fluid Mech. 11, 95 (1979)
4. Kato, T.: Arch. Rat. Mech. Anal. 25, 95 (1967)
5. Inue, A., Funaki, T.: Commun. Math. Phys. 65, 83 (1979)
6. Martinelli, F., Micheli, L.: Private communication
7. Dobrushin, R.L.: Sov. J. Funct. Anal. 13, 115 (1979)
8. McKean, H.P.: Propagation of chaos for a class of nonlinear parabolic equations, Lecture in
differential equations, Vol. 2, Aziz, A.K. (ed.). Princeton, NJ: Van Nostrand 1969
9. Dobrushin, R.L.: Teor. Veroyatn. Priloz. 15, 469 (1970)
10. Gihman, 1.1., Skorohod, A.V.: Introduction a la theorie des processus aleatoires. Moskow: MIR
1980
11. Diirr, D., Pulvirenti, M.: On the vortex flow in bounded domains. Commun. Math. Phys.
(in press) (1982)

Communicated by J. Glimm

Received November 8, 1981