The Annals of Probability

2021, Vol. 49, No. 3, 1402–1426

https://doi.org/10.1214/20-AOP1484
© Institute of Mathematical Statistics, 2021

GLOBAL WELL POSEDNESS OF THE TWO-DIMENSIONAL STOCHASTIC

NONLINEAR WAVE EQUATION ON AN UNBOUNDED DOMAIN

B Y L EONARDO T OLOMEO
Mathematical Institute, Hausdorff Center for Mathematics, Universität Bonn, [email protected]

We study the two-dimensional wave equation with cubic nonlinearity

posed on R2 with space-time white noise forcing. After a suitable renormal-
isation of the nonlinearity, we prove global well-posedness for this equation
for initial data in Hs , s > 45 .

1. Introduction. We consider the following stochastic nonlinear wave equation (SNLW)

with additive space-time white noise forcing
⎧
⎪
⎨utt = u − u + 3∞ · u + dW,
3
⎪
(1.1) u(0, ·) = u ∈ H s ,
0
⎪
⎪
loc
⎩u (0, ·) = u ∈ H s−1 ,
t 1 loc

where dW denotes a space-time white noise on R2 . The term “−3∞ · u” in the equation de-
notes a time-dependent renormalisation which has been first introduced by Gubinelli, Koch
and Oh in [11] for a family of stochastic wave equations with power nonlinearities. In this
work the authors prove local well-posedness for renormalised stochastic wave equations
posed on T2 with certain polynomial nonlinearities.
The necessity to apply such a renormalisation in order to get nontrivial solutions was
highlighted in a series of works by Albeverio, Haba, Oberguggenberger and Russo [1, 19,
20, 26] and, more recently, by Oh, Okamoto and Robert [21], who showed that, without the
renormalisation term, solutions to (1.1) must satisfy a linear wave equation.
The renormalisation that we apply in (1.1) is better described as follows. Recall that dW
is defined to be a distribution-valued random variable such that for every test function φ,
φ, dW  is a gaussian random variable with mean 0 and variance
  
(1.2) E φ, dW 2 = φ2L2 .
Ignoring the term with ∞ in (1.1), we consider a perturbative expansion u = v + ψ, where
t sin((t − t  )|∇|)
(1.3) ψ := dW t 
0 |∇|
solves the linear wave equation
ψtt = ψ + dW.
Here, we used the notation F (∇)u to denote the distribution whose Fourier transform is given
by

F
(∇)u(ξ ) := F (ξ )u(ξ ).

Received December 2019; revised August 2020.

MSC2020 subject classifications. 35L71, 60H15.
Key words and phrases. Stochastic nonlinear wave equation, nonlinear wave equation, renormalisation, Wick
ordering, Hermite polynomial, white noise.
1402
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1403

Formally, the term v would then solve the equation

vtt − v = −(ψ + v)3 = −ψ 3 − 3ψ 2 v − 3ψv 2 − v 3 .
However, because of the roughness of dW , it can be shown that the terms ψ 3 , ψ 2 do not
make sense as space-time distributions. Therefore, we introduce the Wick renormalisation
 
:ψ 2 : = ψ 2 − E |ψ|2 ,
(1.4)  
:ψ 3 : = ψ 3 − 3E |ψ|2 ψ
and define v = u − ψ to solve the equation
⎧
⎪
⎨vtt − v = −:ψ : − 3 :ψ : v − 3ψv − v ,
3 2 2 3
⎪
(1.5) v(0, ·) = u ∈ H s R2 ,
0
⎪
⎪
loc
⎩v (0, ·) = u ∈ H s−1 R2 .
t 1 loc
While both terms on the right-hand side of (1.4) diverge (for both definitions), it is actually
possible to give a meaning to the renormalised terms :ψ 2 :, :ψ 3 : by first taking a smooth
−ε,∞
approximation of the noise dW and then taking a limit in the space Wloc . This will be
carried out explicitly in Section 2. Denoting (formally) :u3 := u3 − 3E[|ψ|2 ]u, solving the
equation (1.5) for v corresponds to solving the equation
⎧
⎪
⎨utt = u − :u: + dW,
3
⎪
(SNLW) u(0, ·) = u ∈ H s ,
0
⎪
⎪
loc
⎩u (0, ·) = u ∈ H s−1
t 1 loc

in the variable u. Since E[|ψ|2 ] = +∞ for every t > 0, by inserting this into (SNLW) we
obtain the formula (1.1). This kind of renormalisation is exactly the same that appears in [11]
for the cubic wave equation on the torus.
Before stating our main result, we need to define what we mean by solutions of (1.1). As
we already discussed, we write u = ψ + v, and we require v to solve the mild formulation of
(1.5),
sin(t|∇|)
v = cos t|∇| u0 + u1
|∇|
(1.6)
t sin((t − t  )|∇|)
+ −:ψ 3 : t  − 3 :ψ 2 : t  v t  − 3ψv 2 t  − v 3 t  dt  .
0 |∇|

T HEOREM 1.1. Consider the equation (SNLW) on R2 , and let 1 > s > 45 . Then, (SNLW)
is almost surely globally well-posed. More precisely, for every (u0 , u1 ) ∈ Hloc
s , there exists a

unique ψ + C(R; Hloc (R ))-valued random variable u such that almost surely:
s 2

• for every stopping time T > 0, u|[−T ,T ] is the unique solution to (SNLW) in the space
ψ + C([−T , T ]; Hloc
s ),

• for every time 0 < T < ∞, u is continuous in the initial data (u0 , u1 ).

In the recent years there have been many developments in the study of global solutions
for parabolic stochastic SPDEs, both on a compact domain (see, for instance, [13, 16, 17]
for a study of 4d models) and, more recently, on the euclidian space [8, 10]. However, the
dispersive nature of the wave equation does not allow to extend the techniques developed for
the study of stochastic quantisation equation to the study of equation (1.1).
The probabilistic study of dispersive equation, on the other hand, has been mostly focused
on the analysis of deterministic equations with random initial data, often because of its links
1404 L. TOLOMEO

to the invariance of appropriate Gibbs-type measures. This direction of investigation was

started by McKean and Vaninsky [14] and Bourgain [2] and have seen a tremendous develop-
ment in recent years. For a study of wave equations with random initial data, see, for instance,
[4–6, 22–24] and references therein. In this setting it is possible to find global well-posedness
results that apply to noncompact domains, as for the one-dimensional wave equation on the
real line [14] or for Schrödinger equations posed on R [3, 7].
More recently, there has been increasing interest in the analysis of stochastic dispersive
equations. However, there are still not many global well-posedness results available, and this
result is the first one the author is aware of that deals with a noncompact domain. In [15], the
author and his collaborators showed global well-posedness for the nonlinear stochastic beam
equation posed on T3 . In [12], the author, together with the authors of [11], showed a similar
result to Theorem 1.1 by extending the solutions of the stochastic wave equation with cubic
nonlinearity posed on T2 built in [11] for infinite time. In [9], Forlano recently adapted the
globalisation argument in this paper for the study the BBM equation with random initial data
outside L2 (T).
The main difference between the result in [12] and Theorem 1.1 is that we do not have
a local well-posedness result available for (SNLW). Indeed, due to the unboundedness of
the domain, one can show that, for every t > 0, the stochastic convolution ψ(t) satisfies
ψ(t)X = +∞ for every translation invariant norm ·X defined on a subspace X of dis-
tributions. This, in turn, implies that, given σ ∈ R, t, R > 0, a typical solution of (1.6) could
satisfy a priori u(t)H σ (B) ≥ R on some ball B. Therefore, any perturbative argument for
local well-posedness (such as a Banach fixed point argument) is bound to fail.
We, therefore, follow a different strategy in order to show global well-posedness for the
equation (1.6) directly. First, we build local solutions (in space and time) to the equation (1.6)
by an approximation argument. We introduce a cutoff function ρ with compact support with
ρ ≡ 1 on a big ball B = B(0, R) and consider a localized version of the equation for v given
by
(1.7) ṽtt − ṽ + ṽ 3 + 3ṽ 2 ρψ + 3ṽρ :ψ 2 : + ρ :ψ 3 := 0.
Notice that in the ball B, this equation is exactly the same as (1.5). Sections 3 and 4 are
dedicated to showing local and global well-posedness for the equation (1.7).
Let ṽR1 and ṽR2 denote the solution to (1.7) corresponding to two different values
of R. Because of finite speed of propagation, we have that ṽR1 (t) = ṽR2 (t) in the ball
B(0, min(R1 , R2 ) − t); see Proposition 5.1 and Corollary 5.2 for a precise statement. There-
fore, we can take a limit, as R → ∞, and build a global solution v̄ to the original equation
(1.6). Moreover, we can use the finite speed of propagation again to show that any solution v
of (1.6) satisfies v = v̄ and obtain the existence and uniqueness results of Theorem 1.1. Lastly,
since the unique solution v coincides with the solution of (1.7) on bounded space-time do-
mains, the continuity part of the statement of Theorem 1.1 will follow from the analogous
result for the equation (1.7). This analysis is performed rigorously in Section 5.
The main technical difficulties arise in Section 4, that is, in showing global well-posedness
for the localized equation (1.7). In order to do so, we show an energy estimate. We would
like to estimate the functional
1 1 1
E ṽ(t), ∂t ṽ(t) := |∂t ṽ|2 + |∇ ṽ|2 + |ṽ|4 ,
2 2 4
since it is conserved by the flow of cubic wave equation with no forcing. However, as we will
see in Section 2, ψ ∈ / L2loc (R2 ), so we expect ṽ ∈ / H 1 , hence E(ṽ, ∂t ṽ) = +∞.
In order to deal with the lack of regularity of ṽ, we make use of the I-method developed
by Colliander, Keel, Staffilani, Takaoka and Tao in 2002 to show global well-posedness for
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1405

dispersive equations with initial data of regularity below the energy space. We consider an
operator I = IN given by the Fourier multiplier mN with
⎧
⎪
⎨1 for |ξ | ≤ N,
1−s
mN (ξ ) = N
⎪
⎩ for |ξ | ≥ 3N.
ξ
For every N , one has that I ṽ ∈ H 1 if and only if ṽ ∈ H s , and I is the identity at low frequen-
cies. Therefore, we can bound ṽH s by making use of the functional
F = E I ṽ(t), I ∂t ṽ(t) .
Of course, the functional F will not be time independent, but we can bound it by making use
of a Gronwall argument. We have that
d
F = commutator terms + forcing terms.
dt
The commutator terms are typical of the I-method, and we can estimate them by making use
of simple harmonic analytic techniques.
The terms coming from the forcing are more subtle to estimate, and, in particular, the term

−3 I vt (I v)2 I (ρψ)

requires a very sharp large deviation estimate for the Lp norm of the term I (ρψ).
Putting the estimates together for fixed N , allows us to show existence of ṽ for a random
time O(1). Iterating this procedure on a sequence N0 N1 N2 · · · , we get global
existence for ṽ.
We would like to point out that, as far as the author is aware, this is the first instance of
an application of the I —method that requires to change the cutoff parameter N in a time-
dependent way.

R EMARK 1.2. The argument in this paper is completely pathwise, and, after the analysis
performed in Section 2, it relies only on deterministic estimates. In particular, we can replace
:ψ j : by stochastic processes Xj and consider the equation
(1.8) vtt − v = −X3 − 3X2 v − 3X1 v 2 − v 3 .
As long as the processes X1 , X2 , X3 satisfy the same estimates as in Proposition 2.1(iv)–(v),
by the same argument as in this paper, we can show an analogous of Theorem 1.1 for the
equation (1.8).

R EMARK 1.3. One could wonder if it is possible to improve the result of Theorem 1.1 by
extending the range of s for which the theorem holds. Analysing the proof of Theorem 1.1,
one can see that the restriction s > 45 derives from the application of the I-method and, in
particular, from the estimate (4.13) for k = 3. While it does not seem to be possible to improve
this particular estimate (see Remark 4.6), there are examples in which a more sophisticated
application of the I-method can yield better results; see, for instance, [25] in the context of
wave equation on the two-dimensional torus.
However, pursuing this kind of improvement would substantially complicate the argument
in this paper, which is already quite involved. Since our main goal in this paper is to present
the first example of a dispersive SPDE with space-time white noise forcing which is globally
well-posed on an unbounded domain, we decided to keep the most technical part of the proof
as simple as possible.
1406 L. TOLOMEO

R EMARK 1.4. As one can observe by analysing the proof of Lemma 4.11, the energy
estimate that leads to global well-posedness for the equation (1.7) relies heavily on the loga-
rithmic divergence of ψ and, more specifically, on the estimate
 
IN (ρψ)  log N.
Llog N
At the moment, we do not know if it is possible to extend the argument in this paper to
equations that present a stronger singularities. For instance, if we consider the equation
⎧
⎪
⎨utt = u− :u : +∇ dW,
3 ε
⎪
u(0, ·) = u ∈ H s ,
0
⎪
⎪
loc
⎩u (0, ·) = u ∈ H s−1 ,
t 1 loc
our argument does not apply.
However, it is still possible find examples in the literature of equations that satisfy a similar
logarithmic divergence. One example is given by the BBM equation with random initial data,
as explored by Forlano in [9]. Another example is given by the wave equation on T3 with
random initial data considered in [6],
⎧
⎪
⎨utt = u − u ,
3
⎪
(1.9) u(0, ·) = u ,
0
⎪
⎪
⎩u (0, ·) = u ,
t 1

with
   
u0 (x) =  aj gj e in·x
, u1 (x) =  bj hj e in·x
,
n∈Z3 n∈Z3
where (gj )j , (hj )j are i.i.d. standard complex-valued gaussian random variables and

|aj |2 n2s + |bj |2 n2s−2 < +∞
n
) ∈ Hs .
so that (u0 , u1 In [6], the authors show that (1.9) is locally well-posed for s ≥ 0, and
it is globally well-posed for s > 0. The techniques developed in this paper should be able to
extend the global well-posedness result to the full range s ≥ 0.

2. Stochastic convolution. In this section we establish relevant estimates on the stochas-

tic convolution
t sin((t − t  )|∇|)
ψ(t, ·) := dWt .
0 |∇|
We recall that, in this section and throughout the paper, the notation F (∇)u denotes the
distribution with Fourier transform given by
F
(∇)u(ξ ) := F (ξ )u(ξ ).
 t sin((t−t  )|∇|)
Let ψN (t, ·) := 0 |∇| χN (∇) dWt , where χN is the indicator function of the ball of
radius N in R . Following the construction in [11], we define :ψNl : for 0 ≤ l ≤ 3 in the
2

following way:
:ψN0 : (x, t) := 1,
:ψN1 : (x, t) := ψN (x, t),
 
:ψN2 : (x, t) := ψN2 (x, t) − E ψN2 (x, t) ,
 
:ψN3 : (x, t) := ψN3 (x, t) − 3E ψN2 (x, t) ψN (x, t).
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1407

This corresponds with the Wick renormalisation obtained through the Hermite polynomial
described in [11]. We will prove the following.

P ROPOSITION 2.1. Let 0 ≤ l ≤ 3, let ρ : R2 → R a C ∞ function with compact support,

let ε > 0 and let T > 0. Let 1 ≤ p, q, r < ∞:
∞ (R × R2 ) a.s.
(i) For every N , ψN ∈ Ct,x
(ii) The sequence ρ :ψN : is almost surely (uniformly) in LT Wx−ε,r . More precisely,
l q

  
ρ :ψ l :  ≤ CT ,p,q,ε,r .
LT Wx−ε,r Lp (
q
N )

(iii) The sequence ρ :ψNl : is Cauchy in Lω LT Wx−ε,r .

p q

−ε,r
q
Because of the arbitrariness of ρ, (iii) implies that :ψNl : has a limit in Lp ( , LT Wloc ). We
call the limit of this sequence :ψ :. We have that:
l

(iv) ρ :ψ l :L∞ Wx−ε,r < ∞ a.s., and ρ :ψ l : (t) is a.s. continuous in t with values in Wx−ε,r .
T
(v) Let m : R → [0, 1] be an even smooth function such that m(r) = 1 for every
|r| ≤ 1, and ∂rn m(r) n r −ε−n for r ≥ 1, n ≥ 0. Let IN be the operator associated
to the Fourier multiplier mN (|ξ |) := m(|ξ |/N), that is, I N φ(ξ ) := m(|ξ |/N)φ̂(ξ ). Then,
1 1
p 2 2
p
P(IN (ρψ)Lt,x ([0,T ];R2 ) > λp log N) ≤ CT ,ρ,m λ . −p

In order to prove this proposition, we need some tools. Recall the Hermite polynomials
generating function
∞ k

1 2 t
(2.1) F (t, x; σ ) := etx− 2 σ t = Hk (x; σ ).
k=0
k!
For simplicity, let F (t, x) := F (t, x; 1), and let Hk (x) := Hk (x; 1). Fix d ∈ N. Consider the
d
Hilbert space L2 (Rd , μ), where dμd := (2π)− 2 exp(−|x|2 /2). Then, Hermite polynomials
satisfy

Hk (x)Hm (x) dμ1 (x) = k!δkm

R
m ∈ N. Define the homogeneous Wiener chaos of order k to be an element of
for all k, 
the form dj =1 Hkj (xj ), where k = k1 + · · · + kd . Denote by Hk the closure of homoge-
neous Wiener chaoses of order k in L2 (Rd , μd ). Then, from the property L2 (Rd , μd ) =
d 2
j =1 L (R, μ1 ), we have the Itô–Wiener decomposition
∞

L2 Rd , μd = Hk .
k=0

Consider the operator L := −( − x · ∇) (the Ornstein–Uhlenbeck


operator). Then, any
element in Hk is an eigenvector of L with eigenvalue k, so ∞ k=0 H k is the spectral decom-
position of L2 associated to L.
Moreover, we have the following hypercontractivity of the Ornstein–Uhlenbeck semigroup
U (t) := e−tL due to Nelson [18].

L EMMA 2.2. Let q > 1 and p ≥ q. Then, for every u ∈ Lq (Rd , μd ) and t ≥ 12 log( p−1
q−1 ),
we have
 
(2.2) U (t)u ≤ uLq (Rd ,μd ) .
Lp (Rd ,μd )
1408 L. TOLOMEO

Notice that the constant of the inequality in (2.2) (i.e., 1) and the range of p, q, t do not
depend on the dimension d. As a consequence, the following holds.

L EMMA 2.3. Let F ∈ Hk . Then, for p ≥ 2, we have

k
(2.3) F Lp (Rd ,μd ) ≤ (p − 1) 2 F L2 (Rd ,μd ) .

This estimate follows simply by applying (2.2) to F , setting q = 2, t = 12 log(p − 1) and

recalling that F is an eigenvector of U (t) with eigenvalue e−kt . As a further consequence,
we obtain the following lemma.

L EMMA 2.4. Fix k ∈ N and c(n1 , . . . , nk ) ∈ C. Given d ∈ N, let {gn }dn=1 be a sequence
of independent standard complex-valued Gaussian random variables. Define Sk (ω) by

Sk (ω) = c(n1 , . . . , nk )gn1 (ω) · · · gnk (ω),
(k,d)

where (k, d) is defined by

 
(k, d) = (n1 , . . . , nk ) ∈ {1, . . . , d}k .
Then, for p ≥ 2, we have
√ k
(2.4) Sk Lp ( ) ≤ k + 1(p − 1) 2 Sk L2 ( ) .

sin(t|∇|)
P ROOF OF P ROPOSITION 2.1. (i) Consider the operator AN (t) := |∇| χN (∇). We
 sin(t|ξ |) 2π iξ ·x
have that AN (t)φ = φ ∗ aN (t, ·), where aN (t, x) = BN |ξ | e dξ . Moreover,

aN ∈ H 2m [−T , T ] × R2 for every m ∈ N,

since
   
aN 2Ḣ 2m ∼ ∂t2m aN 2L2 + m aN 2L2
T
=2 sin t|ξ | 2 |ξ |4m−2 dξ
−T BN

< +∞.
Therefore, for any ball B ⊆ R2 ,
   
EψN 2Ḣ 2m ([−T ,T ]×B) ∼ E∂t2m ψN 2L2 ([−T ,T ]×B) + Em ψN 2L2 ([−T ,T ]×B)
and
 
E∂t2m ψN 2L2 ([−T ,T ]×B)
 t 
     
2
= E ∂t2m aN t − t , x − x dW t , x  dt dx
[−T ,T ]×B 0 R2

≤ aN 2Ḣ 2m ([−T ,T ]×R2 ) dt dx

[−T ,T ]×B

= 2T |B|aN 2Ḣ 2m ([−T ,T ]×R2 )

< +∞
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1409

and, similiarly,
 
Em ψN 2L2 ([−T ,T ]×B)
 t 
     
2
= E  aN t − t , x − x dW t , x  dt dx
m
[−T ,T ]×B 0 R2

≤ aN 2Ḣ 2m ([−T ,T ]×R2 ) dt dx

[−T ,T ]×B

= 2T |B|aN 2Ḣ 2m ([−T ,T ]×R2 )

< +∞,

∞ a.s.
so EψN 2Ḣ 2m ([−T ,T ]×B) < +∞ for every m a.s., therefore ψN ∈ Ct,x
(ii) We have that, for s < t, by Plancherel,
 
E ψN (s, x)ψN (t, y)
 s t 
       
=E aN s − t , x − x dW t , x aN t − t , y − x dW t , x
0 R2 0 R2
(2.5) s
= aN s − t  , x − x  aN t − t  , y − x  dt  dx 
0 R2
s sin((s − t  )|ξ |) sin((t − t  )|ξ |) −2π iξ ·(x−y)
= e dξ dt  .
0 BN |ξ |2

Define γ (t, ξ ) by

t sin((t − t  )|ξ |)2 

(2.6) γ (t, ξ ) := dt .
0 |ξ |2

By applying the Bessel potentials ∇x −ε and ∇y −ε of order ε and setting t = s, x = y, we
get
  
E ∇−ε ψN (t, x)2 = ξ −2ε γ (t, ξ ) dξ
|ξ |<N
1
 t3 + t
|ξ |>1 ξ 2+2ε
 t3 + t

for any ε > 0, x ∈ R2 , and uniformly in N . In particular, by hypercontractivity (Lemma 2.4)

we have that
  
E ∇−ε ψN (t, x)p p,ε,t 1

and, thus, since ρ ∈ Cc∞ ,

    p  
E ρ(·)ψN (t, ·)W −ε,p = E ∇−ε ρ(·)ψN (t, ·)Lp < +∞
p

for any ε > 0, t > 0 and p ≥ 1 uniformly in N ∈ N.

1410 L. TOLOMEO

By the properties of Wick products ([27], Theorem I.22) we have that

    l
E :ψNl (t, x)::ψNl (t, y): ∼ E ψN (t, x)ψN (t, y)

l
= γ (t, ξj )e−2π iξj ·(x−y) dξj
|ξ1 |,...,|ξl |≤N j =1

l 
l
= e−2π i( 1 ξj )·(x−y) γ (t, ξj ) dξj .
|ξ1 |,...,|ξl |≤N j =1

Therefore, proceeding as before,

   
l
E ∇−ε :ψNl (t, ·): (x)2 = ξ1 + · · · + ξl −2ε γ (t, ξj ) dξj
|ξ1 |,...,|ξl |≤N j =1


l
t ξ1 + · · · + ξl −2ε ξj −2 dξj < ∞,
ξ1 ,...,ξl j =1

for any ε > 0, t > 0 uniformly in x ∈ R2 and N . Hence, we have

  
ρ(·) :ψN (t, ·)l :2 ρ,t 1
H −ε

and, by hypercontractivity,
  
E ρ(·) :ψN (t, ·)l :W −ε,p p,ρ,t 1.
p

Integrating this in time, we obtain

  
E ρ(x) :ψN (t, x)l : p
p
(2.7) −ε,p p,ρ,T 1.
Lt Wx ([−T ,T ]×R2 )

Moreover, since ρ(·) :ψNl : has compact support, T < +∞, if q, r, s ≤ p, it follows that
  
ρ(x) :ψN (t, x)l :  p,ρ,T 1.
Lrt Wx−ε,s ([−T ,T ]×R2 ) Lq ( )

(iii) Following (ii), we have that, for N ≤ M,

 
E ψN (s, x)ψM (t, y)
s sin((s − t  )|ξ |) sin((t − t  )|ξ |) −2π iξ ·(x−y)
= e dξ dt 
0 BN |ξ | 2
 
= E ψN (s, x)ψN (t, y) .
Therefore, using the properties of Wick products ([27], Theorem I.22) again,
 
E:ψM
l
(t, x): − :ψNl (t, x):2
     
= E:ψM
l
(t, x):2 − 2E :ψM
l
(t, x)::ψNl (t, x): + E:ψNl (t, x):2
  l   l   l
∼ E ψM (t, x)2 − 2 E ψM (t, x)ψN (t, x) + E ψN (t, x)2
  l   l
= E ψM (t, x)2 − E ψN (t, x)2
l 
l
= e−2π i( 1 ξj )·(x−y) γ (t, ξj ) dξj .
N≤|ξ1 |,...,|ξl |≤M j =1
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1411

Similarly, we have
 
E∇−ε :ψM
l
: − :ψNl : (t, x)2

l
−2ε
= ξ1 + · · · + ξl  γ (t, ξj ) dξj
N≤|ξ1 |,...,|ξl |≤M j =1


l
t ξ1 + · · · + ξl −2ε ξj −2 dξj  N −2θ
|ξ1 |,...,|ξl |≥N j =1

for every 0 < θ < ε. Integrating in space and time, we obtain

  
E ρ(x) :ψM (t, x)l : − ρ(x) :ψN (t, x)l :2L2 Hx−ε p,ρ,t N −2θ ,
t

from which, by hypercontractivity,

  
E ρ(x) :ψM (t, x)l : − ρ(x) :ψN (t, x)l :W −ε,p p,ρ,t N −pθ ,
p

and, arguing as in (ii),

  
ρ(x) :ψM (t, x)l : − ρ(x) :ψN (t, x)l : p
Lt Wx
−ε,q 
([−T ,T ]×R2 ) Lq ( )
max(p,q,r),ρ,T N −θ .
(iv) Using the formula (2.5), we have that
 
E :ψN (s, x)l ::ψN (t, y)l :
 l
∼ E ψN (s, x)ψN (t, y)

s∧t sin((s − t  )|ξ |) sin((t − t  )|ξ |) −2π iξ ·(x−y) 
l
= e dξ dt .
0 BN |ξ |2
Therefore, calling
s∧t sin((s − t  )|ξ |) sin((t − t  )|ξ |) 
γN (s, t; ξ ) := dt s,t ξ −2
0 |ξ |2
and, proceeding as in (iii), we have that
 
E ∇−ε :ψN (s, ·)l : (x) ∇−ε :ψN (t, ·)l : (x)

l
∼ ξ1 + · · · + ξl −2ε γ (s, t; ξj ) dξj .
|ξ1 |,...,|ξl |≤N j =1

Therefore,
 
E ∇−ε :ψN (t + h, ·)l : (x) − ∇−ε :ψN (t, ·)l : (x) 2

∼ ξ1 + · · · + ξl −2ε
|ξ1 |,...,|ξl |≤N
 l 
 
l 
l
× γ (t + h, t + h; ξj ) − 2 γ (t, t + h; ξj ) + γ (t, t; ξj ) dξj .
j =1 j =1 j =1

Interpolating between
γ (t + h, t + h; ξ ) − γ (t, t + h; ξ ), γ (t, t + h; ξ ) − γ (t, t; ξ ) t |h|ξ −1
and
γ (t + h, t + h; ξ ) − γ (t, t + h; ξ ), γ (t, t + h; ξ ) − γ (t, t; ξ ) t ξ −2 ,
1412 L. TOLOMEO

we get that, for every 0 ≤ θ ≤ 1,

γ (t + h, t + h; ξ ) − γ (t, t + h; ξ ), γ (t, t + h; ξ ) − γ (t, t; ξ ) t min |h|ξ −1 , ξ −2
t |h|θ ξ −2+θ .
Choosing θ < 2ε, by the discrete Liebnitz formula we obtain
 
E ∇−ε :ψN (t + h, ·)l : (x) − ∇−ε :ψN (t, ·)l : (x) 2

l
t,l ξ1 + · · · + ξl −2ε |h|θ ξ1 −2+θ ξj −2 dξj t,l |h|θ .
ξ1 ,...,ξl j =2

Since this inequality passes to the limit, we have that

 
E ∇−ε :ψ(t + h, ·)l : (x) − ∇−ε :ψ(t, ·)l : (x) 2 t |h|θ
for a.e. t. Integrating and using hypercontractivity, we get
   pθ
E ρ(·) :ψ(t + h, ·)l : − :ψ(t, ·)l : W −ε,p t,p |h| 2 .
p

Moreover, if |t| ≤ T , the implicit constant C can be chosen as C = C(T , p). Therefore, by
Kolmogorov Continuity Theorem, for |t| < T , we have that ρ :ψ l :∈ C α W −ε,p , for every
pθ
α < 2(p−1) , so ρ :ψ l :∈ Ct W −ε,p a.s.
(v) In order to prove this, we just need to prove
 p   p p
E IN (ρψ)Lp ≤ C(T , ρ, m)p p 2 log 2 N,
then (v) will follow from a straightforward application of Chebishev inequality.
Proceeding as in (2.5), we have that
  
E IN (ρψ)2 (t, x)
 2 
 i2π(ξ +η)·x 
= E  mN (ξ + η)ψ̂(ξ )ρ̂(η)e dξ dη

= mN (ξ + η1 )mN (ξ + η2 )γ (t, ξ )ρ̂(η1 )ρ̂(η2 )ei2π(η1 −η2 )·x dξ dη1 dη2 ,

which is the inverse Fourier transform of the function F : R2 × R2 → C,

F (η1 , η2 ) = ρ̂(η1 )ρ̂(η2 ) mN (ξ + η1 )mN (ξ + η2 )γ (t, ξ ) dξ,

restricted to the plane {(x, −x)}. We have that γ (t, ξ ) t ξ −2 and that, for n ≥ 1, ∂ηn mN (ξ +
η) m,n ξ + η−n . Therefore, calling

φ(η1 , η2 ) := mN (ξ + η1 )mN (ξ + η2 )γ (t, ξ ) dξ,

we have that

 n  log N if n = 0,
∇ φ n,m
η1 ,η2 L∞
1 if n ≥ 1,

hence φC k t,k log N . Since ρ̂(η1 )ρ̂(η2 ) is a Schwartz function of η1 , η2 , this implies that
   log N
(2.8) E IN (ρψ)2 (t, x) T ,ρ,m .
x4
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1413

By hypercontractivity, since IN (ρψ) is Gaussian, one gets that

p
   p log 2 N
(2.9) E IN (ρψ)p (t, x) ≤ C(T , ρ, m)p p 2 ,
4p x
hence, integrating,
 p  p p
E IN (ρψ)Lp ≤ C(T , ρ, m)p p 2 log 2 N. 
t,x

3. Local well-posedness for the localized equation. Take a compactly supported ρ ∈

Cc∞ (R2 ), and consider the equation

vtt − v + v 3 + 3v 2 ρψ + 3vρ :ψ 2 : + ρ :ψ 3 : = 0,
(3.1)
v(t0 , x) = u0 (x)∂t v(t0 , x) = u1 (x).
Notice that (at least formally), whenever ρ = 1, we have that
v 3 + 3v 2 ρψ + 3vρ :ψ 2 : + ρ :ψ 3 : = :(v + ψ)3 :.
Consider as well the mild formulation of (3.1),
sin((t − t0 )|∇|) t sin((t − t  )|∇|)
v(t) = cos (t − t0 )|∇| u0 + u1 +
|∇| t0 |∇|
(LSNLW) ⎛ ⎞

2 
3 j
× ⎝v 3 + v ρ :ψ 3−j :⎠ .
j
j =0

The following local well-posedness result holds.

P ROPOSITION 3.1 (Local well-posedness). Let 1 > s ≥ 23 , and let (u0 , u1 ) ∈ Hs . Then,
there exists a time τ = τ (t0 , ω, (u0 , u1 )Hs ) > 0 a.s., nonincreasing in (u0 , u1 )Hs , such
that the equation (LSNLW) has a unique solution in the space C([t0 − τ, t0 + τ ]; H s ). More-
over, a.s. in ω, we have that inf|t0 |<T τ (t0 , ω, (u0 , u1 )Hs ) > 0 for every T < ∞.

R EMARK 3.2. For the sake of simplicity, in this work we will use just Sobolev embed-
dings to get the required estimates, obtaining the local well-posedness result in the space H s
for s ≥ 23 . Since the constraint coming from Section 5 is stronger than this one, it is enough
for our purposes. However, improving the proof by making use of the Strichartz estimates
for the wave equation, it is possible to relax the condition to s > 14 ; see [11] for this analysis
which can be applied with very little modifications to this problem.

P ROOF. This proposition will follow by a standard fixed point argument. Consider the
map
sin((t − t0 )|∇|)
(u0 ,u1 ) v := cos (t − t0 )|∇| u0 + u1
|∇|
(3.2) ⎛ ⎞

t sin((t − t  )|∇|) ⎝ 3  2
3 j
+ v + v ρ :ψ 3−j :⎠ ,
t0 |∇| j =0
j

2
defined on functions v ∈ Ct ([t0 − τ, t0 + τ ]; H s ). By the Sobolev embedding H s → H 3 →
W 0+,6− → L6 and using that s − 1 < 0 and the fact that ρ :ψ k : is compactly supported, we
1414 L. TOLOMEO

have that

2 
    3  j 
(u0 ,u1 ) vH s ≤ (u0 , u1 )Hs + 2τ v 3 L∞ Hxs−1 + v ρ :ψ 3−j : L∞ s−1
t j t Hx
j =0


2 
  3  
 (u0 , u1 ) vL∞ W 0+,6− ρ :ψ 3−j :L1 W 0−,6+
j
Hs + 2τ vL∞ +
3
6
t L j t t x
j =0


2 
  3  
 (u0 , u1 )Hs + 2τ v3L∞ H s + vL∞ H s ρ :ψ 3−j :L1 W 0−,6+
j
t x j t x t x
j =0
   
 (u0 , u1 )Hs + τ v3Ct H s + 1 + v2Ct H s max ρ :ψ j :L1 W 0−,6+ .
1≤j ≤3 t x

By Proposition 2.1, ρ :ψ j :L1 W 0−,6+ is finite a.s. (locally in time), so it is possibile to find
t x
τ 1 such that max1≤j ≤3 ρ :ψ j :L1 W 0−,6+ 1. Therefore, for τ small enough, we have
t x
that if vCt H s ≤ 2(u0 , u1 )Hs , then u0 ,u1 vCt H s ≤ 2(u0 , u1 )Hs .
Proceeding similarly, we have that
u0 ,u1 v − u0 ,u1 w  τ v − wCt H s v2Ct H s + w2Ct H s
 
+ v − wCt Hs 1 + vCt Hs + wCt Hs max ρ :ψ j :L1 W 0−,6+ ,
j =1,2 t x

for the same reason, choosing τ 1, we have that (u0 ,u1 ) is a contraction on B2(u0 ,u1 )Hs ⊆
Ct H .
s

In order to finish the proof of this proposition, we just need to show that, for every T > 0,
 
inf τ t0 , ω, (u0 , u1 )Hs > 0.
|t0 |<T

Notice that, by the previous argument in order to have that u0 ,u1 is a contraction, we
just need that τ, ρ :ψ j :L1 Hxs−1 < δ for a certain fixed δ. By Proposition 2.1 we have that
t
ρ :ψ j :L2 W 0−,6+ ([−T −1,T +1]×R2 ) < +∞ a.s. As a consequence, we have that (when τ < 1)
t x
  1 
ρ :ψ j :  τ 2 ρ :ψ j :L2 W 0−,6+ ([−T −1,T +1]×R2 ) .
L1t Wx0−,6− ([t0 −τ,t0 +τ ]×R2 ) t x

Therefore, for fixed T , τ can be chosen independently from t0 , and we have inf|t0 |<T τ (t0 , ω,
(u0 , u1 )Hs ) > 0. 

By this local well-posedness result, we obtain the following blow-up condition.

C OROLLARY 3.3 (Blow-up condition). Let T ∗ be the maximal time for which the so-
lution to (LSNLW) exists in the interval [0, T ∗ ), in the sense that (v, vt ) ∈ C([0, T ∗ ); Hs )
and, for every δ > 0, there is no function (ṽ, ṽt ) such that (v, vt ) ∈ C([0, T ∗ + δ); H s ) which
solves (LSNLW). Suppose that T ∗ < +∞. Then, we have
 
(3.3) lim∗  v(s), vt (s) Hs = +∞.
s→T

4. Global well-posedness for the localized equation. In this section we establish global
well-posedness for the equation (LSNLW). In particular, we will prove the following.

P ROPOSITION 4.1. Let s > 45 . Then, the solution to (SNLW) with (u0 , u1 ) ∈ Hs can be
extended a.s. to a global solution u : R × T2 → R.
More precisely, for every T > 0, δ > 0, there exists a set T ,δ such that:
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1415

• P(( T ,ε )c ) ≤ δ,
• For every ω ∈ T ,δ , there exists a unique solution u : [−T , T ] × R2 → R to (SNLW)
such that v(0, x) = u0 (x), vt (0, x) = u1 (x). Moreover, this solution satisfies the estimate
vL∞ s ≤ C(T , s, δ).
t ([−T ,T ];Hx )

Let m : R2 → R be a smooth radial function, 0 < m ≤ 1, such that

⎧
⎪
⎨1 for |ξ | ≤ 1,
(4.1) m(ξ ) = 1
⎪
⎩ ε for |ξ | ≥ 3.
|ξ |
Let mN (ξ ) := m( Nξ ). This way, mN will satisfy
⎧
⎪
⎨1 for |ξ | ≤ N,
ε
(4.2) mN (ξ ) = N
⎪
⎩ for |ξ | ≥ 3N.
|ξ |
Let IN the operator corresponding to the Fourier multiplier mN , that is, I N f (ξ ) =
mN (ξ )f (ξ ). By the definition of the multiplier, for every σ ∈ R, 1 < p < +∞, 0 ≤ δ ≤ ε,
IN will satisfy
(4.3) IN f W σ +δ,p  N δ f W σ,p ,
(4.4) f W σ,p  IN f W σ +ε,p .
One can establish these estimates by showing the analogous ones for I1 , and then the N -
dependence will follow from a simple scaling argument.
From (4.3) and (4.4), one has that, for fixed N , ·Hs is equivalent to IN ·Hs+ε . Therefore,
by the blow-up condition (3.3), in order to show existence of solutions up to time T , it is
enough to show that
 
(4.5) sup (IN v, IN vt )Hs+ε < +∞.
|s|<T

By taking ε = 1 − s,
1 1 1 1
(4.6) E(v, vt ) := vt2 + v2 + |∇v|2 + v4,
2 2 2 4
we clearly have that
 
E(IN v, IN vt )  (IN v, IN vt )2H1  v, vt 2Hs .
The goal of this section will henceforth be showing finiteness of E(IN v, IN vt ). In the fol-
lowing we will abuse of notation and omit the subscript N whenever it is not important in
the analysis, writing I instead of IN . Similarly, we will write E instead of E(IN v, IN vt ) and
E(s) instead of E(IN v(s), IN vt (s)).

L EMMA 4.2.
d
E(t)
dt
(4.7) = −3 I vt (I v)2 I (ρψ)

(4.8) −3 I vt I vI ρ :ψ 2 : − I vt I ρ :ψ 3 :
1416 L. TOLOMEO


+ I vt (I v)3 − I v 3 + 3 (I v)2 I (ρψ) − I v 2 ρψ
(4.9)

+ 3 (I v)I ρ :ψ 2 : − I vρ :ψ 2 :

(4.10) + I vt I v.

P ROOF. We will show this proposition by a formal computation using (3.1). This com-
putation can be made rigorous a posteriori by using the estimates of this section. We omit this
part of the argument.
By (3.1),
d
E(t)
dt
= I vt I vtt − I v + (I v)3 + I v

= I vt −I v 3 + 3v 2 ρψ + 3vρ :ψ 2 : + ρ :ψ 3 : + (I v)3 + I v .

The lemma follows by adding and subtracting the terms 3(I v)2 I (ρψ) and 3(I v)I (ρ :ψ 2 :).


We will now proceed to estimate the various terms of the time derivative of E(I v, I vt )
with the goal of applying a Gronwall argument. The terms (4.10) and (4.8) are relatively
harmless. Estimating the commutator terms in (4.9) is the core of the I-method and will take
most of this section. However, from a technical point of view, the hardest term to estimate
will be (4.7) which will also give the main contribution to the estimate on the growth of E.
This term is also what makes the iteration of the I-method with varying N necessary.

L EMMA 4.3.
(4.11) (4.10) = I vt I v ≤ E(I v, I vt ).

P ROOF. By Hölder,

I vt I v ≤ I vt L2 I vL2 ≤ E(I v, I vt ). 

L EMMA 4.4. For every γ > 0,

3  1 
(4.12) (4.8)  N γ E(I v, I vt ) 4 ρ :ψ 2 :W −γ ,4 + E(I v, I vt ) 2 ρ :ψ 3 :H −γ .

P ROOF. By Hölder and (4.3),

−3 I vt I vI ρ :ψ 2 : − I vt I ρ :ψ 3 :
   
 I vt L2 I vL4 I ρ :ψ 2 : L4 + vt L2 I ρ :ψ 3 : L2
1 1  1 
 N γ E(I v, I vt ) 2 + 4 ρ :ψ 2 :W −γ ,4 + E(I v, I vt ) 2 ρ :ψ 3 :H −γ . 

L EMMA 4.5. Let k ≤ 3. Then,

 
(4.13) (I v)k − I v k  2 s N −(1−k(1−s)) I vk 1 .
L H
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1417

P ROOF. Let vN = |ξ |<N/3 v̂(ξ )eiξ ·x , and let vN := v − vN .
N (ξ ) = 0 only if |ξ | < N/3, by definition of I we have that I vN = vN . Sim-
Since v
ilarly, v
N (ξ ) = 0 only if |ξ | < kN/3, so I (vN ) = vN . Therefore, we have the following
k k k

cancellation:
(I v)k − I v k
= I (vN + vN ) k − I (vN + vN )k

= vN + I (vN ) k − I (vN + vN )k

(4.14) 
k−1 
k k−l−1
= vN
k
− I vN
k
+ l
(I vN )vN (I vN )k−l−1 − I vN vN
l
vN
l
l=0


k−1 
k k−l−1
= l
(I vN )vN (I vN )k−l−1 − I vN vN
l
vN .
l
l=0

We proceed to estimating the elements of this sum individually. In particular, (4.13) follows
if we prove that, for every l ≤ k − 1,
 k−l−1 
(4.15) (I vN )v l
N (I vN )
 2
L  N −(1−k(1−s)) I vkH 1

and
 
(4.16) I vN v l v k−l−1  2  N −(1−k(1−s)) I vk 1 .
N N L H

Let ε := 1 − s. By Sobolev embeddings, we have that f  2ε  f H 1−ε . By Hölder, we

L
have
1−kε (k−1)ε 1−kε (k−1)ε
f  ( 21 − (k−1)ε −1  f L1−ε
2 f 
1−ε
2  f L1−ε
2 f H 1−ε .
1−ε
L 2 ) Lε

Therefore, again by Hölder, we have

 k−l−1 
(I vN )v l  2
N (I vN ) L
 
 (I vN ) ( 21 − (k−1)ε −1 I vN l 2 I vN k−l−1
2
L 2 ) Lε Lε
1−kε (k−1)ε
 I vN  1−ε
L2
I vN  1−ε
H 1−ε
I vk−1
H 1−ε
1−kε 1−kε (k−1)ε
 N −(1−ε) 1−ε I vH1−ε k−1
1−ε I vN H 1−ε I vH 1−ε
1−ε

 N −(1−kε) I vkH 1−ε .

Proceeding similarly and using (4.4), we have

   
I vN v l v k−l−1  2   vN v l v k−l−1  2
N N L N N L

 N −(1−kε) vkH 1−ε

 N −(1−kε) I vkH 1 . 

R EMARK 4.6. We do not expect any extra cancellation in the summands of (4.14). To
see this, let us writing one summand in Fourier transform, corresponding to the case k = 3,
1418 L. TOLOMEO

l = 1,
∧
(I vN )2 vN
l
− I vN
2 l
vN (ξ )

= m(ξ1 )m(ξ2 )m(ξ3 ) − m(ξ1 + ξ2 + ξ3 )

|ξ1 |,|ξ2 |>N/3,|ξ3 |≤N/3,ξ1 +ξ2 +ξ3 =ξ

× v̂(ξ1 )v̂(ξ2 )v̂(ξ3 ) dξ1 dξ2 .

In the regime |ξ1 |, |ξ2 |, |ξ1 + ξ2 |  N , we have that
1−s 1−s 1−s
N N N
m(ξ1 )m(ξ2 )m(ξ3 ) − m(ξ1 + ξ2 + ξ3 ) = −
|ξ1 | |ξ2 | |ξ1 + ξ2 + ξ3 |
1−s
N
∼−
|ξ1 + ξ2 |
∼ −m(ξ1 + ξ2 + ξ3 )
which corresponds to the “worst” term of the difference that appears in (4.14).

L EMMA 4.7. For every γ > 0, 0 < s̃ < 1, there exist p(γ ) > 1, η(γ ) > 0 such that
  1−s̃
(4.17) (If )(Ig) − I (fg)
L2 γ ,s̃ N γ − 2 f H 1−s̃ gW −η(γ ),1 ∩W −η(γ ),p(γ ) .

P ROOF. As in the proof of Lemma 4.5, let us define

uM := u(ξ )eiξ ·x

|ξ |<M/3
and uM := u − uM . Writing f = f 1 +f 1 and g = gN + gN , we have that
N 2 N 2

(If )(Ig) − I (fg) :

(I) = (If 1 )(IgN ) − I (f 1 gN ),
N 2 N 2

(II) + (If 1 )(IgN ) − I (f 1 gN ),

N 2 N 2

(III) + (If 1 )(Ig),

N 2

(IV) − I (f 1 g).
N 2
We have that:
• I = 0, since If 1 =f 1 , IgN = gN by definition of I , and (f 1 gN ) (ξ ) = 0
N 2 N 2 N 2
1
only for |ξ | ≤ (N + N )/3 < N , so I (f
2 1 gN ) = f 1 gN as well.
 N 2 N 2
• IIL2 can be written as suph =1 R2 h · (II). Calling f 1 = a, gN = b, expanding II
L2 N 2
in Fourier series and using Plancherel, we have to estimate

1 a(ξ1 )b(ξ2 ) mN (ξ1 )mN (ξ2 ) − mN (ξ1 + ξ2 ) ĥ(ξ1 + ξ2 ) dξ1 dξ2 .

ξ1 <N 2 /3,ξ2 >N/3
Using the fact that on the considered interval m(ξ1 ) ≡ 1 and that, by the mean value theo-
rem, |m(ξ1 + ξ2 ) − m(ξ1 )| s N 1−s |ξ1 ||ξ2 |−2+s , we have that

h · (II)
R2

= 1 a(ξ1 )b(ξ2 ) mN (ξ1 )mN (ξ2 ) − mN (ξ1 + ξ2 ) ĥ(ξ1 + ξ2 ) dξ1 dξ2

ξ1 <N 2 /3,ξ2 >N/3
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1419
  
 a(ξ1 )  b(ξ2 )  |ξ1 |1+s̃+δ  
N (1−s)     ĥ(ξ1 + ξ2 ) dξ1 dξ2
1    
ξ1 <N 2 /2,ξ2 ≥N/2 |ξ1 |s̃+δ |ξ2 |δ |ξ2 |2−s−δ
  
−( 1−s̃ − 32 δ)  a(ξ1 )  b(ξ2 )  
N    ĥ(ξ1 + ξ2 ) dξ1 dξ2
2
1    |ξ2 | 
ξ1 <N 2 /2,ξ2 ≥N/2 |ξ1 | s̃+δ δ
   
1−s̃ 3  a(ξ1 )   b(ξ2 ) 
 N −( 2 − 2 δ)   
 
 ĥ 2
 2 L
|ξ |
1
s̃+δ 1 |ξ |δ
L 2 L
−( 1−s̃ 3
2 − 2 δ)
N f H 1−s̃ gH −δ hL2 ,
1−s̃ 3
therefore, IIL2  N −( 2 − 2 δ) f H 1−s̃ gH −δ .
• By Hölder, Sobolev embeddings and (4.3),

IIIL2  If 1 L2 IgL∞

N 2
1−s̃
δ N − 2 f H 1−s̃ IgW 3δ,δ−1
1−s̃
δ N − 2 +4δ f H 1−s̃ gW −δ,δ−1 .

• By duality (like for II), selfadjointness of I , fractional Liebnitz inequality, Sobolev em-
beddings and (4.3), we have

h · (IV)

=− hI (f 1 g)
N 2

=− I (h)f 1 g
N 2
 
 I (h)f 1 
W 2δ,(1−δ)
−1 gW −2δ,δ−1
N 2
 
 gW −2δ,δ−1 I (h)H 2δ f 1  ( 21 −δ)−1
N 2 L
 
+ I (h) ( 21 −δ)−1 f 1 H 2δ
L N 2
 
 gW −2δ,δ−1 I (h)H 2δ f 1 H 2δ
N 2
1
 N 2δ N − 2 (1−s̃−2δ) gW −2δ,δ−1 hL2 f 1  1−s̃
N 2 H
1−s̃
 N −( 2 −3δ) gW −2δ,δ−1 hL2 f 1  1−s̃ ,
N 2 H

1−s̃
so IVL2  N −( 2 −3δ) gW −2δ,δ−1 f H 1−s̃ .

Therefore, by choosing δ such that γ ≥ 4δ, η = δ and p = δ −1 , we obtain (4.17). 

L EMMA 4.8. Let 0 < k ≤ 2. For every γ > 0, there exist p(γ ) > 1, η(γ ) > 0 such that
 
I v k ρ :ψ 3−k : − (I v)k I ρ :ψ 3−k :  2
L
(4.18) 1−k(1−s)  
s,γ N − 2 +γ I vk 1 ρ :ψ 3−k : H W −η(γ ),1 ∩W −η(γ ),p(γ ) .
1420 L. TOLOMEO

P ROOF. We have that:

 
I v k ρ :ψ 3−k : − (I v)k I ρ :ψ 3−k :  2
L
 
(I)  I v k ρ :ψ 3−k : − I v k I ρ :ψ 3−k : L2
 
(II) +  I v k − (I v)k I ρ :ψ 3−k :  2 . L
• By Sobolev embeddings and fractional Liebnitz, we have that
 k  
v   v k  2  vH s vk−12  vkH s .
H 1−k(1−s) s, 1+(k−1)(1−s)
W L (1−s)
Therefore, by (4.17),
1−k(1−s)  
IL2  N − 2 +γ
vkH s ρ :ψ 3−k :W −η(γ ),p(γ ) .
From (4.4) we have that vH s  I vH 1 , so
1−k(1−s)  
IL2 (1−s),γ N − 2 +γ
I vkH 1 ρ :ψ 3−k :W −η(γ ),p(γ ) .
• From Hölder, (4.13) and Sobolev embeddings, we have
   
IIL2  I v k − (I v)k L2 I ρ :ψ 3−k : L∞
 
s,δ N −(1−k(1−s)) I vkH 1 I ρ :ψ 3−k : W 3δ,δ−1
 
s,δ N −(1−k(1−s)) N 4δ I vkH 1 ρ :ψ 3−k :W −δ,δ−1 .

Choosing δ small enough, we have that 1 − k(1 − s) − 4δ > 1−k(1−s)

2 − γ , so the main
contribution comes from I. We get (4.18) by taking γ =≥ 4δ, p(γ ) = max(p(γ ), δ −1 ),
 

η(γ  ) = max(η(γ ), δ) and then renaming γ = γ  . 

L EMMA 4.9. There exists c > 0 such that, for every 0 < η < 18 ,
T T 
(4.7)(s) ds = −3 I vt (s) I v(s) 2 I ρψ(s) ds
t0 t0
(4.19) 
T
 1+ E 1+cη
I ψ η−1 .
t0 Lt,x

P ROOF. Let 0 < θ < 1. Since we have

1
I vH 1  E 2 ,
1
I vL4  E 4 ,
1+θ
by Gagliardo–Nirenberg we have I v 4
θ, 1+θ E 4 . Therefore, by Sobolev inequality, we
1+θ W
have that I v 4  E , and the implicit constant is uniform in θ as long as 0 ≤ θ ≤
4
L 1−θ
θmax < 1. Take θ = 4η. Therefore, by Hölder,
 T 
  T
 I vt (I v) I ψ  ≤
2
I vt L2 I vL4 I v I ψ
 4 η−1
t0 T 2 t0 L 1−4η Lx
T 1 1 1+4η
 E2E4 E 4 I ψ η−1
t0 Lx
T
 E 1+η I ψ η−1
t0 Lx
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1421

T 1−η
1
 E 1+η 1−η I ψ η−1
t0 Lx,t
T 1+η

 1+ E 1−η I ψ η−1 .
t0 Lx,t

Therefore, choosing c = maxη∈[0, 1 ] η−1 ( 1+η

1−η − 1), we have
8
 T  
  T
 I vt (I v)2 I ψ   1 + E 1+cη I ψ
 η−1
t0 T2 t0 Lx,t

which gives (4.19). 

L EMMA 4.10. Let T > 0. For every |t − t0 | ≤ T and for every 0 < η ≤ 18 , we have that,
for every γ > 0,
E(t) − E(t0 )
t 
(4.20) s,γ ,T 1+ E 1+cη I ψ η−1
t0 Lt,x
t
(4.21) + N −(1−3(1−s)) E 2
t0


2 t 1−k(1−s) k+1  
(4.22) + N− 2 +γ
E 2  :ψ 3−k :L∞ W −η(γ ),1 ∩W −ηk (γ ),p(γ )
t
k=0 t0
t 3  1 
(4.23) + N γ E 4 ρ :ψ 2 :W −γ ,4 + E 2 ρ :ψ 3 :H −γ
t0
t
(4.24) + E(s) ds,
t0
where c is the one given by Lemma 4.9 and η(γ ), p(γ ) are the ones given by Lemma 4.8.

P ROOF. By Lemma 4.2,

t
E(t) − E(t0 ) = (4.7)(s) + (4.8)(s) + (4.9)(s) + (4.10)(s) ds.
t0
We have that:
• From (4.19),
t T  T 
(4.7)(s) ds = −3 I vt (s) I v(s) 2 I ρψ(s) ds  1 + E 1+cη I ψ η−1 .
t0 t0 t0 Lt,x

• From (4.12), tt0 (4.8)(s) ds  (4.23).
• From (4.13) for the fist term and (4.18) for the second and third terms, respectively, and
Hölder inequality, tt0 (4.9)(s) ds  (4.21) + (4.22).
• From (4.11),
t t  t
(4.10)(s) ds = I vt (s)I v(s) ds ≤ E I v(s), I vt (s) ds = (4.24).
t0 t0 t0 

L EMMA 4.11. Let T > 0, and let

IN ρψLlog N ([−T ,T ]×R2 )
(4.25) A(N) = .
log N
1422 L. TOLOMEO

For γ > 0, M ≥ 1,  ≥ 1, define

$   %
:= max:ρψ 3−k :L∞ ([−T ,T ];W −(max(ηk (γ ),γ )),max(pk (γ ),2) ) ≤ M ,
γ
(4.26) M k t
 
(4.27)  (N) := A(N) ≤  .
γ
Then, for γ = γ (s) small enough, α < 1 − 3(1 − s), δ < β < α, ω ∈ M , there exists
γ
τ = τ (s, M, , α − β) such that if ω ∈ M ∩  (N), E(t0 ) ≤ N β /2, |t0 | ≤ T , N ≥ N0 =
N0 (s, T , M, ), then E(t) ≤ N α for every t such that |t| ≤ T and |t − t0 | ≤ τ .

P ROOF. By Lemma 4.10, as long as E ≤ N α , since α < 1 − 3(1 − s), for N big enough

we have that (4.21) + (4.22) ≤ 1 + tT0 E. Similarly, from Young’s inequality, for some uni-
versal constant C, we have
T
(4.23) ≤ E(s) ds + CN 4γ M 4 .
t0

Choosing η = (log N)−1 in (4.20), as long as E ≤ N α , we get

T  T 
−1
(4.20) ≤ A(N) log N 1 + E 1+c(log N) (s) ds ≤  log N 1 + ecα E(s) ds .
t0 t0

Therefore, as long as E(t) ≤ N α , for N big enough (depending on s, T , M, ),

E(t)
T 
≤ E(t0 ) + C(s, γ , T ) log N 1 + E(s) ds
t0
T T
+1+ E(s) ds + CN 4γ M 4 + E(s) ds
t0 t0
(4.28) 1
≤ N β + C(s, γ , T ) log N + C  N 4γ M 4
2
T
+ 2 + C(s, γ , T ) log N E(s) ds
t0
T
≤ N β + C  (s, γ , T , ) log N E(s) ds.
t0

Let t¯ = max{s : t0 ≤ s ≤ T , E(s) ≤ N α }, t˜ = min{s : t0 ≥ s ≥ −T , E(s) ≤ N α }. Then, the

lemma is proven if we show that if t¯ = T , then |t¯ − t0 | ≥ τ (s, γ , T , ) and, similarly, if t˜ =
−T , then |t˜ − t0 | ≥ τ (s, γ , T , ). For t˜ ≤ t ≤ t¯, by definition, (4.28) holds, so, by Gronwall,
(4.29) E(t) ≤ N β exp |t − t0 |C  (s, γ , T , ) log N .
Suppose that t¯ = T . Then, one must have E(t¯) = N α . Therefore, by (4.29), |t¯ − t0 | ≥ τ :=
(α−β) ˜
C  (s,γ ,T ,) . The same holds for t , and the lemma is proven. 

P ROOF OF P ROPOSITION 4.1. Let ε > 0, T > 0, let 2(1 − s) < β < α < 1 − 3(1 − s)
and let γ as in Lemma 4.11.
By Proposition 2.1(iv), :ρψ 3−k :L∞ ([−T ,T ];W −(max(ηk (γ ),γ )),max(pk (γ ),2) ) < +∞ a.s., so there
t
γ γ
exists M = M(T , ε) such that P( M ) ≥ 1 − 2ε , where M is defined in (4.26). Moreover, by
Proposition 2.1(v),
≤ CT ,ρ,m − log N = N log CT ,ρ,m −log  .
c log N
P  (N)
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1423

If  = CT ,ρ,m e2 , this expression is summable in N , so, for N big enough, N ≥ Ñ = Ñ(ε),

& γ &
P ( N≥Ñ  (N)) ≥ 1 − 2ε . Let T ,ε := M ∩ N≥Ñ  (N). By inclusion-exclusion we
have that P( T ,ε ) ≥ 1 − ε.
For this choice of M, , γ , α, β, let N0 and τ be the ones given by Proposition 4.11,
and take (u0 , u1 ) ∈ Hs . Define a sequence Nk of integers recursively. Take N1 such that
N1 ≥ max(N0 , Ñ) and
 
N12(1−s) (u0 , u1 )2Hs + u0 4H s
β
N1 .
By Sobolev embeddings, (4.3) and (4.4),
E IN1 v(0), IN1 vt (0) = E(IN1 u0 , IN1 u1 )
 
 (IN1 u0 , IN1 u1 )2H 1 + IN1 u0 4L4
(4.30) 2(1−s)  
 N1  (u0 , u1 )2H s + IN1 u0 4H s
 
 N12(1−s) (u0 , u1 )2H s + u0 4H s ,
β
so we will have E(IN1 v(t0 ), IN1 vt (t0 )) ≤ 12 N1 ; therefore, by Lemma 4.11, one has that
(v(t), vt (t))2Hs  E(IN1 v(t), IN1 vt (t)) ≤ N α for t ≤ T , t0 ≤ t ≤ t0 + τ , and similarly back-
ward in time.
Then, take Nk+1  Nk such that
2(1−s) β
(4.31) Nk+1 Nkα + Nk2α Nk+1 .
Take k ∈ N such that t0 + (k − 1)τ ≤ T . If one has that, for every such k,
1 β
(4.32) E INk v t0 + (k − 1)τ , INk vt t0 + (k − 1)τ ≤ Nk ,
2
then, by Lemma 4.11,
 
(v, vt )2 ∞
L ([t0 +(k−1)τ,min(t0 +kτ,T )];Hs )
(4.33)
 sup E INk v(s), INk vt (s) ≤ Nkα ,
t0 +(k−1)τ ≤s≤min(t0 +kτ,T )

and similarly backward in time. Therefore, because of the blowup condition (3.3), Propos-
tion 4.1 is shown if we show (4.32). Proceeding inductively, we know (4.32) for N1 and,
proceeding as in (4.30),
E INk+1 v(t0 + kτ ), INk vt (t0 + kτ )
2(1−s)     
 Nk+1 v(t0 + kτ ), vt (t0 + kτ ) 2H s + v(t0 + kτ )4H s
2(1−s) 2
 Nk+1 E INk v(t0 + kτ ), INk vt (t0 + kτ ) + E INk v(t0 + kτ ), INk vt (t0 + kτ )
2(1−s) β
 Nk+1 Nkα + Nk2α Nk+1 ,
β
so E(INk+1 v(t0 + kτ ), INk vt (t0 + kτ )) ≤ 12 Nk+1 , and we have (4.32). 

5. Independence from the cutoff and global well-posedeness for the global equation.
In this section we prove that, on appropriate space-time regions, the solution to (LSNLW)
does not depend on the particular choice of the cutoff ρ and proceed to the proof of Theo-
rem 1.1.
1424 L. TOLOMEO

P ROPOSITION 5.1 (Finite speed of propagation for SNLW). Let R, T > 0, x0 ∈ R2 ,

t0 ∈ R. Let u1 , u2 be solutions to (SNLW) on B(x0 , R) for a time T , in the sense that
uj |B(x0 ,R) = ψ|B(x0 ,R) + vj |B(x0 ,R) , vj ∈ C([t0 − T , t0 + T ]; Hloc
s ), s > 2 , and, for every
3
|t − t0 | ≤ T ,
sin((t − t0 )|∇|)
vj (t)|B(x0 ,R) = cos (t − t0 )|∇| vj (t0 ) + ∂t vj (t0 )
|∇|
(5.1) 
t sin ((t − t  )|∇|) 
+ :(ψ + vj )3 : t  dt   .
t0 |∇| B(x0 ,R)

Suppose, moreover, that

v1 (0) = v2 (0) and ∂t v1 (0) = ∂t v2 (0) on B(x0 , R).
Then, v1 (t)|B(x0 ,R−|t−t0 |) = v2 (t)|B(x0 ,R−|t−t0 |) for every |t − t0 | ≤ T .

P ROOF. Without loss of generality, assume that t0 = 0, x0 = 0. Let Dt = BR−|t| . Recall-

ing that the kernels of cos(s|∇|), sin(s|∇|)
|∇| are distributions supported on Bs , from (SNLW)
we have that
v1 − v2 |Dt
t sin ((t − t  )|∇|)  
= (v1 − v2 ) 3 :ψ 2 : +3ψ(v1 + v2 ) + v12 + v22 + v1 v2 D  dt  .
0 |∇| t

Proceeding as in Proposition 3.1, we obtain that

 
v1 (t) − v2 (t)
H s (Dt )
t
R v1 − v2 H s (Dt ) 1 + v1 2C([−R,R];H s (BR )) + v2 2C([−R,R];H s (BR ))
0
 
× max ρ :ψ j :L1 W 0−,6+
1≤j ≤3 t x

for any ρ ∈ Cc∞ with ρ ≡ 1 on BR . Therefore, by Gronwall,

   
v1 (t) − v2 (t) ≤ C(R, v1 , v2 , ω)v1 (0) − v2 (0)H s (BR ) = 0. 
H s (Dt )

From this proposition we obtain immediately the following.

C OROLLARY 5.2. Let T > 0, let t0 = 0 and let ρ1 , ρ2 be two cutoff functions such
that ρ1 (x) = ρ2 (x) = 1 for every x ∈ B2T . Let s > 45 . Let (u0 , u1 ) ∈ Hloc s , and let v , v
1 2
be, respectively, the solutions to (LSNLW) with cutoff function ρ1 and ρ2 and initial data,
respectively, (ρ1 u0 , ρ1 u1 ) and (ρ2 u0 , ρ2 u1 ). Then, v1 (t, x) = v2 (t, x) for every |x|, |t| < T .

P ROOF OF T HEOREM 1.1. Let ρn be a cutoff function such that ρN (x) = 1 for every
|x| ≤ n. Let (u0 , u1 ) ∈ Hloc
s , and let v be the solution of (LSNLW) with cutoff function ρ
n n
and initial data (ρn u0 , ρn u1 ). By Proposition 4.1, we will have vn ∈ C(R; Hs ). By Corol-
lary 5.2,
v := lim vn
n→+∞

is well defined, and we have that v|[−T ,T ]×BR = vT ∨R |[−T ,T ]×BR . By Proposition 4.1 again,
v will also be a continuous function of (u0 , u1 ) with values in C([−T , T ]; Hloc s ). There-

fore, the theorem is proven if we show that every solution ũ = ψ + ṽ of (SNLW) with
 GWP FOR STOCHASTIC WAVE EQUATION ON R2 1425
s ) satisfies ṽ(t) = v(t) for every t ≤ T . Let φ ∈ C ∞ ((−T , T ) × R2 ) be a
v ∈ C([−T , T ]; Hloc c
test function. Let n ∈ N be such that supp(φ) ⊆ [−n, n] × Bn . By Proposition 5.1, we have
that
ṽ|[−n,n]×Bn = vn |[−n,n]×Bn = v|[−n,n]×Bn .
Therefore, ṽ, φ = v, φ, so ṽ = v as space-time distributions. Since they both belong to
the space C([−T , T ]; Hloc
s ), the equality must hold in the space C([−T , T ]; Hs ) as well;
loc
hence, ṽ(t) = v(t) for every t ≤ T . 

Acknowledgements. The author would like to thank his PhD supervisor Tadahiro Oh for
suggesting the problem and his constant help and support. The author was supported by the
European Research Council (grant no. 637995 “ProbDynDispEq”), by The Maxwell Insti-
tute Graduate School in Analysis and its Applications, a Centre for Doctoral Training funded
by the U.K. Engineering and Physical Sciences Research Council (grant EP/L016508/01),
the Scottish Funding Council, Heriot-Watt University and the University of Edinburgh and
by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Ger-
many’s Excellence Strategy-EXC-2047/1-390685813 through the Collaborative Research
Centre (CRC) 1060.

REFERENCES
[1] A LBEVERIO , S., H ABA , Z. and RUSSO , F. (1996). Trivial solutions for a non-linear two-space-dimensional
wave equation perturbed by space-time white noise. Stoch. Stoch. Rep. 56 127–160. MR1396758
https://doi.org/10.1080/17442509608834039
[2] B OURGAIN , J. (1994). Periodic nonlinear Schrödinger equation and invariant measures. Comm. Math. Phys.
166 1–26. MR1309539
[3] B OURGAIN , J. (2000). Invariant measures for NLS in infinite volume. Comm. Math. Phys. 210 605–620.
MR1777342 https://doi.org/10.1007/s002200050792
[4] B URQ , N. and T ZVETKOV, N. (2008). Random data Cauchy theory for supercritical wave equations. I. Local
theory. Invent. Math. 173 449–475. MR2425133 https://doi.org/10.1007/s00222-008-0124-z
[5] B URQ , N. and T ZVETKOV, N. (2008). Random data Cauchy theory for supercritical wave equations.
II. A global existence result. Invent. Math. 173 477–496. MR2425134 https://doi.org/10.1007/
s00222-008-0123-0
[6] B URQ , N. and T ZVETKOV, N. (2014). Probabilistic well-posedness for the cubic wave equation. J. Eur.
Math. Soc. (JEMS) 16 1–30. MR3141727 https://doi.org/10.4171/JEMS/426
[7] C ACCIAFESTA , F. and DE S UZZONI , A.-S. (2015). Invariant measure for the Schrödinger equation on the
real line. J. Funct. Anal. 269 271–324. MR3345610 https://doi.org/10.1016/j.jfa.2015.04.021
[8] C HANDRA , A., M OINAT, A. and W EBER , H. A priori bounds for the 4 equation in the full sub-critical
regime. Preprint. Available at arXiv:1910.13854.
[9] F ORLANO , J. (2020). Almost sure global well posedness for the BBM equation with infinite L2 initial data.
Discrete Contin. Dyn. Syst. 40 267–318. MR4026960 https://doi.org/10.3934/dcds.2020011
[10] G UBINELLI , M. and H OFMANOVÁ , M. (2019). Global solutions to elliptic and parabolic 4 mod-
els in Euclidean space. Comm. Math. Phys. 368 1201–1266. MR3951704 https://doi.org/10.1007/
s00220-019-03398-4
[11] G UBINELLI , M., KOCH , H. and O H , T. (2018). Renormalization of the two-dimensional stochastic non-
linear wave equations. Trans. Amer. Math. Soc. 370 7335–7359. MR3841850 https://doi.org/10.1090/
tran/7452
[12] G UBINELLI , M., KOCH , H., O H , T. and T OLOMEO , L. Global dynamics for the two-dimensional stochastic
nonlinear wave equations. Preprint. Available at arXiv:2005.10570.
[13] H OSHINO , M. (2018). Global well-posedness of complex Ginzburg–Landau equation with a space-time
white noise. Ann. Inst. Henri Poincaré Probab. Stat. 54 1969–2001. MR3865664 https://doi.org/10.
1214/17-AIHP862
[14] M C K EAN , H. P. and VANINSKY, K. L. (1995). Statistical mechanics of nonlinear wave equations. In
Stochastic Analysis (Ithaca, NY, 1993). Proc. Sympos. Pure Math. 57 457–463. Amer. Math. Soc.,
Providence, RI. MR1335489 https://doi.org/10.1090/pspum/057/1335489
1426 L. TOLOMEO

[15] M OSINCAT, R., P OCOVNICU , O., T OLOMEO , L. and WANG , Y. Well-posedness theory for the three-
dimensional stochastic nonlinear beam equation with additive space-time white noise forcing. Preprint.
[16] M OURRAT, J.-C. and W EBER , H. (2017). Global well-posedness of the dynamic 4 model in the plane.
Ann. Probab. 45 2398–2476. MR3693966 https://doi.org/10.1214/16-AOP1116
[17] M OURRAT, J.-C. and W EBER , H. (2017). The dynamic 43 model comes down from infinity. Comm. Math.
Phys. 356 673–753. MR3719541 https://doi.org/10.1007/s00220-017-2997-4
[18] N ELSON , E. (1966). A quartic interaction in two dimensions. In Mathematical Theory of Elementary Parti-
cles (Proc. Conf., Dedham, Mass., 1965) 69–73. M.I.T. Press, Cambridge, MA. MR0210416
[19] O BERGUGGENBERGER , M. and RUSSO , F. (1998). Nonlinear stochastic wave equations. Integral Trans-
forms Spec. Funct. 6 71–83. MR1640497 https://doi.org/10.1080/10652469808819152
[20] O BERGUGGENBERGER , M. and RUSSO , F. (2001). Singular limiting behavior in nonlinear stochastic
wave equations. In Stochastic Analysis and Mathematical Physics. Progress in Probability 50 87–99.
Birkhäuser, Boston, MA. MR1886565
[21] O H , T., O KAMOTO , M. and ROBERT, T. (2020). A remark on triviality for the two-dimensional stochastic
nonlinear wave equation. Stochastic Process. Appl. 130 5838–5864. MR4127348 https://doi.org/10.
1016/j.spa.2020.05.010
[22] O H , T., O KAMOTO , M. and T ZVETKOV, N. Uniqueness and non-uniqueness of the Gaussian free field
evolution under the two-dimensional Wick ordered cubic wave equation. Preprint.
[23] O H , T. and P OCOVNICU , O. (2017). A remark on almost sure global well-posedness of the energy-
critical defocusing nonlinear wave equations in the periodic setting. Tohoku Math. J. (2) 69 455–481.
MR3695994 https://doi.org/10.2748/tmj/1505181626
[24] O H , T., P OCOVNICU , O. and T ZVETKOV, N. Probabilistic local Cauchy theory of the cubic nonlinear wave
equation in negative Sobolev spaces. Preprint. Available at arXiv:1904.06792.
[25] ROY, T. (2016). On the interpolation with the potential bound for global solutions of the defocusing cubic
wave equation on T2 . J. Funct. Anal. 270 3280–3306. MR3475458 https://doi.org/10.1016/j.jfa.2016.
02.018
[26] RUSSO , F. (1994). Colombeau generalized functions and stochastic analysis. In Stochastic Analysis and
Applications in Physics (Funchal, 1993). NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci. 449 329–349.
Kluwer Academic, Dordrecht. MR1337971
[27] S IMON , B. (1974). The P (ϕ)2 Euclidean (Quantum) Field Theory. Princeton Series in Physics. Princeton
Univ. Press, Princeton, NJ. MR0489552