a

r
X
i
v
:
1
3
0
8
.
1
7
1
9
v
1


[
m
a
t
h
.
A
P
]


8

A
u
g

2
0
1
3
IMPROVED WELL-POSEDNESS FOR THE QUADRATIC DERIVATIVE
NONLINEAR WAVE EQUATION IN 2D
VIKTOR GRIGORYAN
1
AND ALLISON TANGUAY
2
Abstract. In this paper we consider the Cauchy problem for the nonlinear wave equa-
tion (NLW) with quadratic derivative nonlinearities in two space dimensions. Following
Gr unrocks result in 3D, we take the data in the Fourier-Lebesgue spaces

H
r
s
, which coin-
cide with the Sobolev spaces of the same regularity for r = 2, but scale like lower regularity
Sobolev spaces for 1 < r < 2. We show local well-posedness (LWP) for the range of ex-
ponents s > 1 +
3
2r
,
3
2
< r 2. On one end this recovers the best known result on the
Sobolev scale, H
7
4
+
, while on the other end establishes the

H
3
2
+
2+
result, which scales like
the Sobolev H
5
3
+
, thus, corresponding to a
1
12
derivative improvement on the Sobolev
scale.
1. Introduction
Consider the quadratic derivative nonlinear wave equation (NLW),
(1) u = (u)
2
in R
1+n
,
where u is the space-time gradient of u, and no special structure is assumed in the non-
linearity. We are interested in the local well-posedness (LWP) question for equation (1) for
initial data in Fourier-Lebesgue spaces

H
r
s
. More precisely, we consider the Cauchy problem
for (1) with initial conditions
(2) (u,
t
u)[
t=0
= (f, g)

H
r
s


H
r
s1
,
and wish to establish local well-posedness for a range of the exponents (r, s), which improves
on previously known Sobolev space results. The Fourier-Lebesgue spaces

H
r
s
have been
previously successfully used to achieve improved regularity results for a variety of equations.
See for example [10], [23], [6], [7], [2], [8], [9].
Equation (1) is invariant under the scaling
(3) (t, x) (t, x).
That is, if u is a solution to (1) in R
1+n
, then so is u

(t, x) = u(t, x). Under this scaling,

the homogeneous Sobolev norm of the initial data scales as
|u

(0, )|

H
s
(R
n
)
=
s
n
2
|u(0, )|

H
s
(R
n
)
,
and s
c
=
n
2
is called the critical exponent, since the

H
sc
norm of the initial data is preserved
under the scaling. Under general scaling considerations, one expects a local well-posedness
for data in the Sobolev space H
s
for s > s
c
(subcritical regime), global existence for small
data in

H
sc
(critical regime), and some form of ill-posedness for data in H
s
for s < s
c
(supercritical regime).
Consequently, local well-posedness for (1) is expected to hold for data in the Sobolev
space H
s
, s >
n
2
. This can be proved in dimensions n 4 with a Strichartz estimate
approach; however, it is known to be false in dimensions n = 2, 3.
2010 Mathematics Subject Classication. 35L70.
1
2 GRIGORYAN AND TANGUAY
In dimension n = 3, Ponce and Sideris [18] proved LWP for s > 2, which is sharp in
light of the counterexamples of Linblad [15], [16], [17]. If the nonlinearity has a null-form
structure, Klainerman and Machedon [12] improved the range for LWP to s >
3
2
, which is
almost critical in 3D.
In dimension n = 2, the best known result for the general quadratic derivative NLW is
for s >
7
4
, which can be shown using the Strichartz estimates approach. As we were unable
to locate a reference for this result, we sketch its proof in Appendix A. If the nonlinearity
has a null-form structure, the LWP results can again be improved. For the Q
0
null-form,
almost critical LWP was shown by Klainerman and Selberg [14] in the context of wave
maps. For the other null forms, Zhou [24] established LWP for data in H
s
with s >
5
4
, and
also showed that this is sharp.
Recently, Gr unrock showed in [9] that the gap to the almost criticality for the general
quadratic derivative NLW (1) can be closed in dimension n = 3 by considering initial data
in the Fourier-Lebesgue spaces

H
r
s
. These spaces are dened by their norms,
|f|

H
r
s
= |
s

f|
L
r

,
1
r
+
1
r

= 1,
where

f stands for the (spatial) Fourier transform of f, and
1
=
_
1 +[[
2
. The norm
of the corresponding homogeneous space is |f|

H
r
s
= |[[
s

f|
L
r

. Under the scaling (3), the

norms of the initial data in the homogeneous Fourier-Lebesgue spaces scale as
|u

(0, )|

H
r
s
(R
n
)
=
s
n
r
|u(0, )|

H
r
s
(R
n
)
,
so the critical exponent for these spaces is s
r
c
=
n
r
. In terms of scaling of the norms,
we therefore observe the following correspondence between the homogeneous Sobolev and
Fourier-Lebesgue spaces,
(4)

H
r
s


H

, if = s +n
_
1
2

1
r
_
.
Gr unrock established local well-posedness for data in the space

H
r
s
for s >
2
r
+1, 1 < r 2.
Thus, his range of exponents almost reaches criticality at the endpoint r = 1. His approach
relies on free wave interaction estimates of Foschi-Klainerman [4] that come with a factor of
[[[[[[
n3
2
, which becomes unbounded near the null cone [[ = [[ in two dimensions. Thus,
Gr unrocks result cannot be directly generalized to a LWP result for the general quadratic
NLW (1) in dimension n = 2. However, if the nonlinearity has enough cancellation along
the null cone to oset this factor, then his arguments can be salvaged. This approach was
explored by the rst author and A. Nahmod in [3], who achieve almost critical LWP for the
null-form problem in 2D.
The main result of this paper is the following.
Theorem 1.1. Let
3
2
< r 2, s >
3
2r
+ 1, then the Cauchy problem (1)-(2) is locally
well-posed for data in the space

H
r
s


H
r
s1
.
Remark 1.2. Well-posedness here is understood in the sense of Theorem B.1, with the
solution found in the space Z
r
s,b
, which will be dened in the next section.
Remark 1.3. Notice that at one extreme, (s, r) = (
7
4
, 2), this result coincides with the local
well-posedness for data in H
7
4
+
, while at the other extreme, (s, r) = (2,
3
2
), we obtain local
well-posedness in the space

H
3
2
+
2+
. By the scaling correspondence (4), this last space scales
1
Alternatively, one can use = 1 + ||

1 + ||
2
.
IMPROVED LWP FOR QUADRATIC DERIVATIVE NLW IN 2D 3
like the Sobolev space H
5
3
+
, which gives a
1
12
derivative improvement on the H
7
4
+
Sobolev
data result.
Remark 1.4. An analogous result to Theorem 1.1 can be proved for the equation
(5) u = (u
2
).
This equation is invariant under the scaling u u

, where u

(t, x) = u(t, x). And the

critical exponent on the Sobolev scale is s
c
=
n
2
1, while on the Fourier-Lebesgue scale it
is s
r
c
=
n
r
1. The best known result for (5) in dimension n = 2 for data in H
s
is for s >
3
4
,
which, again, can be shown using Strichartz estimates. A similar argument to the one we
use for equation (1) will show that the Cauchy problem for (5) with data in

H
r
s
is locally
well-posed for s >
3
2r
,
3
2
< r 2.
The region for the the parameters (s,
1
r
), for which Theorem 1.1 holds is shaded in Figure
1. Notice that the bottom and right edges of the region are not included. If our estimates did
not break down for r
3
2
, the s-line s =
3
2r
+1 would have picked up the point (s, r) = (
5
2
, 1),
which corresponds to the Sobolev space H
3
2
. This is still above the critical regularity, as
can be seen from the picture, where the solid line represents the critical relation s
r
c
=
2
r
.
1
r
s
s =
3
2r
+ 1
s =
2
r
1
2
1
2
3
1
7
4
2
(s, r) = (
5
2
, 1) (
3
2
, 2)
Figure 1. The shaded region represents the range of indeces for which LWP
for data in space

H
r
s


H
r
s1
holds.
The rest of the paper is organized as follows. in Section 2 the solution spaces are in-
troduced, and Theorem 1.1 is reduced to a bilinear estimate. This is done via the general
LWP theorem, which is stated in Appendix B. In Section 3 we establish bilinear Fourier
restriction estimates in

L
r
spaces. These estimates are used in Section 4 to establish the
main bilinear estimate (8).
As was mentioned earlier, we also sketch the proof of the LWP for equation (1) with data
in the Sobolev space H
7
4
+
H
3
4
+
by the Strichartz estimates in Appendix A.
Acknowledgments: The authors would like to thank Andrea Nahmod for helpful dis-
cussions and her encouragement during this project.
4 GRIGORYAN AND TANGUAY
2. The X
r
s,b
space, reduction to a bilinear estimate
The local in time solution is obtained via a contraction principle in a suitable solution
space. For this we will use time restricted X
r
s,b
space. The X
r
s,b
space is a Fourier-L
r

analogue of the wave-Sobolev space X

s,b
, and is dened via the norm
|u|
X
r
s,b
= |
s
[[ [[
b
u|
L
r

,
,
where u denotes the time-space Fourier transform of u. The time restriction space is then
X
r
s,b;T
=
_
u = U[
[T,T]R
n : U X
r
s,b
_
,
with its norm dened as
|u|
X
r
s,b;T
= inf
_
|U|
X
r
s,b
: U[
[T,T]R
n = u
_
.
If b >
1
r
, the so-called transfer principle allows one to transfer multilinear estimates for
free solutions of the homogeneous equation u = 0 with data in

H
r
s
spaces to general
elements of X
r
s,b
spaces. A trivial consequence of the transfer principle is the following
crucial embedding
X
r
s,b
C(R,

H
r
s
), thus, also X
r
s,b;T
C([T, T],

H
r
s
),
which guarantees that the solutions found in the X
r
s,b
spaces are the proper continuations
of the initial data in

H
r
s
spaces. For details on the transfer principle see [19] for L
2
based
spaces, [6] for the linear case on the X
r
s,b
spaces, and [3] for the X
r
s,b
multilinear case.
As the wave operator is of second order in time, we also need to separately estimate the
time derivative of the solution. Hence, we dene our solution space, Z
r
s,b
, by the norm
|u|
Z
r
s,b
= |u|
X
r
s,b
+|
t
u|
X
r
s1,b
.
The time restriction space Z
r
s,b;T
and its norm are dened as before.
In the sequel we will also make use of the notation

L
r
t,x
= X
r
0,0
, and

L
r
=

H
r
0
, where the
last norm can be taken either with respect to the time, or space variables.
By the general local well-posedness, Theorem B.1, Theorem 1.1 will follow from the
following two estimates
(6) |uv|
X
r
s1,b+1
|u|
Z
r
s,b
|v|
Z
r
s,b
,
and
(7) |(u)
2
(v)
2
]|
X
r
s1,b+1

_
|u|
Z
r
s,b
+|v|
Z
r
s,b
_
|u v|
Z
r
s,b
.
The second estimate will trivially follow from the rst one due to the quadratic nature of
the nonlinearity. Thus, we only need to prove estimate (6).
If b + 1 < 0, it is easy to see that estimate (6) will follow from the estimate
(8) |uv|
X
r
,0
|u|
X
,b
|v|
X
,b
,
where = s 1, and u, v are now general elements of X
r
,b
. The rest of this paper is
dedicated to proving (8) for >
3
2r
,
3
2
< r 2, and some b, , with
1
r
< b < 1, (0, 1b).
Remark 2.1. In [9] Gr unrock relies on the general local well-posedness theorem for the rst
order equation
t
u i(D)u = N(u), which he proved in the earlier paper [6]. For this,
he reformulates equation (1) as a system of rst order equations for u

= u iJ
1
x

t
u. If
we were to use this approach in our context, and the nonlinearity contains time derivatives
IMPROVED LWP FOR QUADRATIC DERIVATIVE NLW IN 2D 5
of u, we would have to place the left hand sides of estimates (6)-(7) in the space Z
r
s1,b

.
These estimates in addition to (8) would also require the estimate
(9) |uv|
X
r
1,b

+1
|u|
X
,b
|v|
X
,b
.
This is due to the fact that the absolute value of the symbol of the rst order operator,
[ ()[, does not control [[. However, Gr unrock uses the transfer principle and does
his estimates only on free solutions, which allows an easy control of the time derivatives by
the spatial derivatives, and essentially produces the three dimensional analogue of (9) for
free from (8). As we mentioned before, this approach fails in two dimensions. Nevertheless,
one can indeed show (9) with the arguments we employ here and a hyperbolic Leibniz type
rule. See [22] for the details, where also more general estimates of the form (6) are proved.
3. Bilinear Fourier restriction estimates
By duality, we can reformulate the multiplicative estimate (8) as the trilinear integral
estimate
(10) I |F
0
|
L
r
,
|F
1
|
L
p
,
|F
2
|
L
p
,
,
where
I =
___
F
0
(X
0
)F
1
(X
1
)F
2
(X
2
)(X
0
+X
1
+X
2
) dX
0
dX
1
dX
2

[
1
[ [
1
[
b
[
2
[ [
2
[
b
,
X
j
= (
j
,
j
) for j = 0, 1, 2, and F
1
, F
2
stand for
F
1
(
1
,
1
) =
1

[
1
[ [
1
[
b
u(
1
,
1
),
F
2
(
2
,
2
) =
2

[
2
[ [
2
[
b
v(
2
,
2
),
and to simplify notation, p = r

.
Since
0
+
1
+
2
= 0 in the above integral, by triangle inequality, [
j
[ [
k
[ + [
l
[
for all permutations (j, k, l) of (0, 1, 2). Hence, also,
j

k
+
l
. This implies that
only low-high-high (LHH), high-low-high (HLH), and high-high-low (HHL) interactions are
permitted. We then split I = I
LHH
+I
HLH
+I
HHL
, with the following correspondence:
LHH
0

1

2

HLH
1

0

2

HHL
2

0

1
.
The letters M, N, L and their indexed counterparts will denote dyadic numbers of the
form 2
j
, j 0, 1, 2, . . . , and we will use the following notation for Fourier restrictions.
F
N
(X) =
N
F(X),
F
N,L
(X) =
||||L
F
N
(X),
F
N,L,
(X) =
0
F
N,L
(X).
One immediately has the following dyadic summation estimates:

N
|F
N
|
p
L
p
|F|
p
L
p
,

L
|F
N,L
|
p
L
p
|F
N
|
p
L
p
,

L
|F
N,L,
|
p
L
p
|F
N
|
p
L
p
.
6 GRIGORYAN AND TANGUAY
Using the trilinear convolution form
(11) J(F
0
, F
1
, F
2
) =
___
F
0
(X
0
)F
1
(X
1
)F
2
(X
2
)(X
0
+X
1
+X
2
) dX
0
dX
1
dX
2
,
we have
I

N,L
J(F
N
0
0
, F
N
1
,L
1
1
, F
N
2
,L
2
2
)
N

0
N

1
N

2
L
b
1
L
b
2
,
where N = (N
0
, N
1
, N
2
), L = (L
1
, L
2
).
The proof of estimate (8) relies on bilinear Fourier restriction estimates on R
1+2
of the
form
(12) |P
A
0
(P
A
1
u
1
P
A
2
u
2
)|

L
r
C|P
A
1
u
1
|

L
r
|P
A
2
u
2
|

L
r
,
where A
0
, A
1
, A
2
R
1+2
are measurable sets, and the projection P
A
is dened by

P
A
u =

A
u. We are interested in these restriction estimates with A
0
, A
1
and A
2
being thickened
subsets of the null cone K = (, ) R
1+2
: [[ = [[. Following Selberg [20, 21], we
introduce the following notation for thickened upper ( 0) and lower ( 0) cones,
truncated in the spatial frequency by balls, annuli and angular sectors:
K

N,L
= (, ) R
1+2
: [[ N, = [[ +O(L),

N,L
= (, ) R
1+2
: [[ N, = [[ +O(L),

N,L,
() = (, )

K

N,L
: (, ) ,
where N, L, > 0; S
1
; and (a, b) denotes the angle between nonzero a, b R
2
.
By duality, (12) is equivalent to
(13) J(F
A
0
0
, F
A
1
1
, F
A
2
2
) C|F
A
0
0
|
L
r |F
A
1
1
|
L
p|F
A
2
2
|
L
p,
where F
A
(X) =
A
(X)F(X); F
j
= u
j
, j = 1, 2; and J is given by (11).
We start with the following elementary lemma, which is a direct extension of [21, Lemma
1.1] to

L
r
(see also [20, Lemma 3.1]). We use [E[ for the volume of the set E R
1+2
.
Lemma 3.1. The estimate (12) holds with C equal to an absolute constant times
min
_
sup
XA
0
[A
1
(X A
2
)[
1
r
, sup
XA
1
[A
0
(X +A
2
)[
1
p
[A
2
[
1
r

1
p
,
sup
XA
2
[A
0
(X +A
1
)[
1
p
[A
1
[
1
r

1
p
_
,
provided this quantity is nite.
Proof. For the rst bound, we use the dual formulation, and rewrite
J(F
A
0
0
, F
A
1
1
, F
A
2
2
)
=
_
X
0
_
X
1

X
1
A
1
(X
0
A
2
)

X
0
A
0
F
A
1
1
(X
1
)F
A
2
2
(X
0
X
1
)F
A
0
0
(X
0
) dX
1
dX
0

_
X
0
_
sup
X
0
A
0
[A
1
(X
0
A
2
)[
1
r
_ __
X
1
_
F
A
1
1
(X
1
)F
A
2
2
(X
0
X
2
)
_
p
dX
1
_1
p
F
A
0
0
(X
0
) dX
0
,
IMPROVED LWP FOR QUADRATIC DERIVATIVE NLW IN 2D 7
where we used Holders inequality in the X
1
variable. Now applying Holders inequality in
the X
0
variable and using Fubinis theorem, we bound the above by
sup
X
0
A
0
[A
1
(X
0
A
2
)[
1
r

__
X
0
_
X
1
_
F
A
1
1
(X
1
)
_
p
_
F
A
2
2
(X
0
X
1
)
_
p
dX
1
dX
0
_1
p
__
X
0
_
F
A
0
0
(X
0
)
_
r
dX
0
_1
r
sup
X
0
A
0
[A
1
(X
0
A
2
)[
1
r

__
X
1
_
F
A
1
1
(X
1
)
_
p
_
X
0
_
F
A
2
2
(X
0
X
1
)
_
p
dX
0
dX
1
_1
p
|F
A
0
0
|
L
r
sup
X
0
A
0
[A
1
(X
0
A
2
)[
1
r
|F
A
0
0
|
L
r |F
A
1
1
|
L
p|F
A
2
2
|
L
p.
The proofs of the second and third bounds are similar, so we prove only the second
bound. Using the dual formulation we now rewrite
J(F
A
0
0
, F
A
1
1
, F
A
2
2
)
=
_
X
1
_
X
0

X
0
A
0
(X
1
+A
2
)

X
1
A
1
F
A
1
1
(X
1
)F
A
2
2
(X
0
X
1
)F
A
0
0
(X
0
) dX
1
dX
0

_
X
1
_
sup
X
1
A
1
[A
0
(X
1
+A
2
)[
1
p
_ __
X
0
_
F
A
0
0
(X
0
)F
A
2
2
(X
0
X
1
)
_
r
dX
0
_1
r
F
A
1
1
(X
1
) dX
1
sup
X
1
A
1
[A
0
(X
1
+A
2
)[
1
p

__
X
1
_
X
0
_
F
A
0
0
(X
0
)
_
r
_
F
A
2
2
(X
0
X
1
)
_
r
dX
0
dX
1
_1
r
|F
A
1
1
|
L
p
sup
X
1
A
1
[A
0
(X
1
+A
2
)[
1
p
|F
A
0
0
|
L
r |F
A
2
2
|
L
r |F
A
1
1
|
L
p
sup
X
1
A
1
[A
0
(X
1
+A
2
)[
1
p
[A
2
[
1
r

1
p
|F
A
0
0
|
L
r |F
A
1
1
|
L
p|F
A
2
2
|
L
p.

Using the above lemma, we can now prove the necessary bilinear restriction estimates in

L
r
, analogous to those established by Selberg in [21, Theorem 2.1].
Proposition 3.1. Let
3
2
< r 2, then the estimate
(14)
_
_
_
_
P
K

0
N
0
_
P
K

1
N
1
,L
1
u
1
P
K

2
N
2
,L
2
u
2
__
_
_
_

L
r
C|u
1
|

L
r
|u
2
|

L
r
holds with
C (N
012
min
)
2
p
(N
12
min
)
2
r

2
p
(L
12
min
)
1
r
, (15)
C (N
012
min
)
1
p
(N
12
min
)
3
2r

1
p
(L
12
min
)
1
r
(L
12
max
)
1
2r
, (16)
regardless of the choice of signs
j
.
Remark 3.2. The restriction
3
2
< r is only needed for (16). Furthermore, the bound with
(15) can be trivially generalized to any dimension, since Lemma 3.1, which is the only
8 GRIGORYAN AND TANGUAY
ingredient in its proof, holds irrespective of the dimension of the space. In n dimensions
(15) looks like
(17) C (N
012
min
)
n
p
(N
12
min
)
n
r

n
p
(L
12
min
)
1
r
.
Proof of (15). The bound with (15) immediately follows from Lemma 3.1. To see this, we
split into the cases N
0
N
1
N
2
(LHH), N
1
N
0
N
2
(HLH), and N
2
N
0
N
1
(HHL), and by symmetry consider only LHH and HLH cases.
In the HLH case, N
1
N
2
N
0
, we estimate the volume of the set
E = K

1
N
1
,L
1
(X
0
K

2
N
2
,L
2
)
1
N
1
;
1
=
1
[
1
[ +O(L
1
),
0

1
=
2
[
0

1
[ +O(L
2
),
for any X
0
K

0
N
0
,L
0
. Integrating rst in
1
and then in
1
, we obtain [E[ N
2
1
L
12
min
. Raising
to the power 1/r and using the rst bound from Lemma 3.1 gives the desired result.
For the LHH regime, i.e. when N
0
N
1
N
2
, due to symmetry we can assume L
1
L
2
,
and estimate the volume of the set
E = K

0
N
0
,L
0
(X
2
+K

1
N
1
,L
1
)
0
N
0
;
0

2
=
1
[
0

2
[ +O(L
1
)[,
for any X
2
K

2
N
2
,L
2
. Again, integrating rst in
0
and then in
0
, we have [E[ N
2
0
L
1
,
which coupled with [K

1
N
1
,L
1
[ N
2
1
L
1
and the third bound in Lemma 3.1 gives the desired
result.
Proof of (16). We may assume L
12
max
N
012
min
, since otherwise (15) is already better than
(16). Additionally, we may replace K

j
N
j
,L
j
by

K

j
N
j
,L
j
due to the presence of the factor
(N
012
min
)
1
p
(N
12
min
)
2
r

3
2p
in the bound. Indeed, using the dual formulation and decomposing the
balls [
j
[ N
j
into annuli [
j
[ M
j
for dyadic 0 < M
j
N
j
, we can sum over M
012
max
N
012
min
using the factor (M
012
min
)
1/p
and the fact that the two largest of the M
j
s are comparable.
For the rest of the sum we use the duality of l
p
and l
r
in the HLH and HHL cases, or the
factor (N
12
min
)
2
r

3
2p
in the LHH case. We now proceed with the proof separately in the HLH
and LHH cases. In what follows, we use the calculations and ideas of [21, Section 3] in a
substantial way.
The HLH case: N
1
N
0
N
2
. By Lemma 3.1 we reduce to estimating the volume
of the set
E =

K

1
N
1
,L
1
(X
0

2
N
2
,L
2
)
uniformly in X
0


K

0
N
0
,L
0
. But by [21, Subsection 3.1], [E[ N
3
2
1
L
12
min
(L
12
max
)
1
2
. We then
have
[E[
1
r

_
N
3
2
1
L
12
min
(L
12
max
)
1
2
_1
r
N
3
2r
1
(L
12
min
)
1
r
(L
12
max
)
1
2r
.
The LHH case: N
0
N
1
N
2
. We may assume L
1
L
2
by symmetry, and need
to prove that (3.1) holds with
(18) C N
1
p
0
N
3
2r

1
p
1
L
1
r
1
L
1
2r
2
.
Recall that (14) is equivalent to the dual estimate (13). By a simple change of variables
in the integral in the trilinear form on the left, we can also see that (13) is equivalent to
the product estimate
(19)
_
_
_
_
P
K

2
N
2
,L
2
_
P
K

1
N
1
,L
1
u
1
P
K

0
N
0
u
0
__
_
_
_

L
p
C|u
1
|

L
r
|u
0
|

L
p
,
where u
0
is the inverse Fourier transform of F
0
. So it suces to prove (19) for C given by
(18).
IMPROVED LWP FOR QUADRATIC DERIVATIVE NLW IN 2D 9
Denoting
01
= (
1

1
,
0

0
), we split into the cases
01

0
, and
12

0
, where

0
=
_
L
2
N
1
_1
2
.
Assuming without loss of generality that u
1
, u
0
0, we have by [21, Lemma 2.4, 2.5],
|P
N
2
(u
1
u
0
)|

L
p

0
,
1
(
0
)

(
0
,
1
)
0
|TP
N
2
(u

0
,
1
1
u

0
,
0
0
)|
L
r
+

0
<<1

0
,
1
()

3(
0
,
1
)12
|TP
N
2
(u
,
1
1
u
,
0
0
)|
L
r =
1
+
2
,
where T denotes the time-space Fourier transform; () is a maximal -separated subset
of S
1
, and

u
,
=
(,)<
u.
For
1
, we estimate the volume of the set E =

K

0
N
0
,
0
(
0
) (X
2
+

K

1
N
1
,L
1
,
0
(
1
)), where
X
2
is any point in the support of

u
N
2
2
. Notice that
E (
0
,
0
) : [
0
[ N
0
, (
0
,
0
) <
0
,
0
=
1
[
0

2
[ +
2
+O(L
1
),
and hence [E[ N
0
N
0

0
L
1
N
0
N
1

0
L
1
. Further observing that [A
1
[ N
1
N
1

0
L
1
, we
have by Lemma 3.1 that
|TP
N
2
(u

0
,
1
1
u

0
,
0
0
)|
L
r (N
0
N
1

0
L
1
)
1
p
(N
2
1

0
L
1
)
1
r

1
p
|

0
,
1
1
|
L
p|

0
,
0
0
|
L
r ,
and hence,

1
N
1
p
0
N
2
r

1
p
1

1
r
0
L
1
r
1

1
,
0
(
0
)

(
1
,
0
)
0
|

0
,
1
1
|
L
p|

0
,
0
0
|
L
r
N
1
p
0
N
2
r

1
p
1

1
r
0
L
1
r
1
|u
1
|
L
p|u
0
|
L
r ,
where in the last step we used Holders inequality, relying on the almost orthogonality
condition (see [21, (10)])
(20) #

() : (

, ) k 2k + 1.
Finally, using
0
= (L
2
/N
1
)
1
2
gives the desired estimate.
For
2
we again estimate the volume of the set E =

K
N
0
(X
2
+A
1
), using now the fact
that 3 (
1
,
0
) 12. First observe that
E = (
0
,
0
) :
0
R,
0
=
2

1
[
0

2
[ +O(L
1
),
where
R =
0
R
2
: [
0
[ N
0
,
0

(
0
),
1
(
0

2
)

(
1
).
Then for a xed
0
,
f(
0
) :=
1
[
0

2
[ =
0

2
+O(L
1
),
and we dene
e = f(
0
) =
1

2
[
0

2
[
.
Choosing coordinates (
1
0
,
2
0
), so that

1

0
|
1

2
|
= (1, 0), we have for all
0
E,

1
f(
0
) = e

1

0
[
1

0
[
=
cos (e,
1
) cos (e,
0
)
[
1

0
[
.
10 GRIGORYAN AND TANGUAY
Noticing that (e,
1
) , the fact that (
1
,
0
) 3 implies that (e,
0
) 2. Then
from the monotonic decreasing property of the cosine function for small angles,

1
f(
0
)
cos cos 2
[
1

0
[


2
[
1

0
[
,
where we used the expansion of cos around = 0 and that [
1

0
[ . Then integrating
rst in the
1
0
direction, we obtain
[E[ [
0
R :
0
T, f(
0
) =
0

2
+O(L
1
)[[T[

L
1

[
2
0
: [
2
0

2
2
[ =
1
(
0

2
) +O(L
1
)[[T[
L
2

L
1
N
0
,
where T =
0
:
0
=
2

1
[
2
[ +O(N
0
) +O(L
1
).
Using this estimate for the volume of E, and that [A
1
[ N
2
1
L
1
, as before, we have from
Lemma 3.1,
|TP
N
0
(u
,
1
1
u
,
0
2
)|
L
r
_
N
0
L
1
L
2

_1
p
(N
2
1
L
1
)
1
r

1
p
|

u
,
1
1
|
L
p|

u
,
0
0
|
L
r .
Hence,

2
N
1
p
0
L
1
r
1
N
2
r

2
p
1
L
1
p
2

0
<<1

1
r

2
p

1
,
0
()

3(
1
,
0
)12
|

u
,
1
1
|
L
p|

u
,
0
0
|
L
r
N
1
p
0
L
1
r
1
N
2
r

2
p
1
L
1
p
2

1
r

2
p
0
|u
1
|
L
p|u
0
|
L
r ,
where we estimated the sum in
1
,
0
as in the case of
1
, and summed in using the
restriction r >
3
2
, which guarantees that the exponent is negative. But
0
= (L
2
/N
1
)
1
2
, so

1
r

2
p
0
= L
1
2r

1
p
2
N
1
p

1
2r
1
, and, thus,

2
N
1
p
0
N
3
2r

1
p
1
L
1
r
1
L
1
2r
2
|u
1
|
L
p|u
2
|
L
p.

4. The proof of estimate (8)

We start by establishing some simple summation lemmas.
Lemma 4.1. Let A, B R, then

N
0
N
1
N
A
0
N
B
1
,
provided (i) B A, (ii) B 0 and (iii) A = B = 0 is excluded.
Proof. The statement of the lemma follows immediately from the trivial dyadic summation
rule
(21)

N
0

N
0
N
1
N
A
0

_
_
_
N
A
1
if A > 0,
logN
1
if A = 0,
1 if A < 0.

The next two dyadic summation lemmas will be used for estimating HLH (and by sym-
metry, HHL) and LHH interactions respectively.
IMPROVED LWP FOR QUADRATIC DERIVATIVE NLW IN 2D 11
Lemma 4.2. Let A, B R. The estimate

N
1
N
0
N
2
N
A
1
N
B
0
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p |F
0
|
L
r |F
1
|
L
p|F
2
|
L
p
holds, provided that (i) B A, (ii) B 0 and (iii) A = B = 0 is excluded.
Proof. The proof of this lemma is contained in the proof of [1, Lemma 2.1], where, after
applying Lemma 4.1, one simply needs to use the duality of l
p
and l
r
(Holders inequality
for l
p
), instead of the Cauchy-Schwartz inequality.
Lemma 4.3. Let A, B R. The estimate

N
0
N
1
N
2
N
A
0
N
B
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p |F
0
|
L
r |F
1
|
L
p|F
2
|
L
p
holds, provided that (i) B > A, (ii) B > 0.
Proof. The proof is similar to the proof of the previous lemma, with the only dierence in
that one cannot sum in N
1
N
2
using Holders inequality, since l
p
is not self-dual. But
due to the strict inequalities in the hypothesis, we have by Lemma 4.1:

N
0
N
1
N
2
N
A
0
N
B
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p |F
0
|
L
r

N
1
N
2
N

1
|F
N
1
1
|
L
p|F
N
2
2
|
L
p
|F
0
|
L
r |F
1
|
L
p
_
N

1
_
l
r
_
|F
N
2
2
|
L
p
_
l
p
|F
0
|
L
r |F
1
|
L
p|F
2
|
L
p

We are now ready to prove the key estimate (8).

Proof of estimate (8). It is enough to prove the dual estimate (10). We will use dyadic
decompositions, and by symmetry can assume L
1
L
2
. Estimate (10) then reduces to
showing

N
S
N
N

0
N

1
N

2
|F
0
|
L
r |F
1
|
L
p|F
2
|
L
p,
where
S
N
=

L
1
L
2
J(F
N
0
0
, F
N
1
,L
1
1
, F
N
2
,L
2
2
)
L
b
1
L
b
2
,
with J again dened as in (11). By symmetry, we only need to consider the HLH and LHH
cases.
The HLH case: N
1
N
0
N
2
. From (15), (16) and the dual formulation we know
that the estimate
(22) J(F
N
0
0
, F
N
1
,L
1
1
, F
N
2
,L
2
2
) C|F
N
0
0
|
L
r |F
N
1
,L
1
1
|
L
p|F
N
2
,L
2
2
|
L
p
holds with
(23) C N
3
2r
1
L
1
r
1
[min(N
1
, L
2
)]
1
2r
.
Hence, we split into cases L
2
N
1
and L
2
> N
1
. In the rst case (16) will be used, while
the second case requires (15).
12 GRIGORYAN AND TANGUAY
The subcase L
2
N
1
. In this case using (23) gives
S
N
N
3
2r
1

L
L
1
r
b
1
L
1
2r
b
2
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
N
3
2r
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p,
as b >
1
r
. But then by Lemma 4.2,

N
S
N
N

0
N

1
N

N
1
N
0
N
2
N
3
2r

1
N
0
0
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
|F
0
|
L
r |F
1
|
L
p|F
2
|
L
p,
as >
3
2r
.
The subcase L
2
> N
1
. In this case (23) reads C N
2
r
1
L
1
r
1
, hence,
S
N
N
2
r
1

L
2
>N
1
L
1
r
b
1
L
b
2
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
N
2
r
b
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p.
Then by Lemma 4.2 gives,

N
S
N
N

0
N

1
N

N
1
N
0
N
2
N
2
r
b
1
N
0
0
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
|F
0
|
L
r |F
1
|
L
p|F
2
|
L
p,
as >
3
2r
and b >
1
r
.
Notice that this subcase shows that estimate (15) will only establish (8) for >
2
r
, which
is above the best known result, and, in fact, coincides with what one can prove with classical
energy estimates.
The LHH case: N
0
N
1
N
2
. From (15), (16) we know that (22) holds with
(24) C N
1
p
0
N
3
2r

1
p
1
L
1
r
1
min(N
1
p
0
N
1
2r

1
p
1
, L
1
2r
2
).
Hence, we split into cases L
2
N
2r
p
0
/N
2r
p
1
1
and L
2
> N
2r
p
0
/N
2r
p
1
1
. Again, we will use (16)
in the rst case, while the second case requires (15).
The subcase L
2
N
2r
p
0
/N
2r
p
1
1
. In this case (24) becomes C N
1
p
0
N
3
2r

1
p
1
L
1
r
1
L
1
2r
2
, and
we have
S
N
N
1
p
0
N
3
2r

1
p
1

L
L
1
r
b
1
L
1
2r
b
2
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
N
1
p
0
N
3
2r

1
p
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p.
Lemma 4.3 then gives

N
S
N
N

0
N

1
N

N
0
N
1
N
2
N
1
p
+
0
N
2
3
2r
+
1
p
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
|F
0
|
L
r |F
1
|
L
p|F
2
|
L
p,
since >
3
2r
implies that 2 >
3
2r

1
p
, and 2
3
2r
+
1
p
> +
1
p
.
IMPROVED LWP FOR QUADRATIC DERIVATIVE NLW IN 2D 13
The subcase L
2
> N
2r
p
0
/N
2r
p
1
1
. Now (24) becomes C N
2
p
0
N
2
r

2
p
1
L
1
r
1
, and
S
N
N
2
p
0
N
2
r

2
p
1

L
2
>N
2r
p
0
/N
2r
p
1
1
L
1
r
b
1
L
b
2
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
N
2
p

2r
p
b
0
N
2
r

2
p
+(
2r
p
1)b
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p.
Lemma 4.3 then gives

N
S
N
N

0
N

1
N

N
0
N
1
N
2
N
+
2
p

2r
p
b
0
N
2
2
r
+
2
p
(
2r
p
1)b
1
|F
N
0
0
|
L
r |F
N
1
1
|
L
p|F
N
2
2
|
L
p
|F
0
|
L
r |F
1
|
L
p|F
2
|
L
p,
as
2
2
r
+
2
p

_
2r
p
1
_
b = 2
2
r
+ 2
_
1
1
r
_
2r
_
1
1
r
_
b +b
= 2
4
r
+ 2 2rb + 3b

2
r
+ 2 2rb > 0,
since >
3
2r
, and we can chose b to be as close to
1
r
as we want, and the dierence of the
exponents of N
1
and N
0
is
_
2
2
r
+
2
p

_
2r
p
1
_
b
_

_
+
2
p

2r
p
b
_
=
2
r
+b > 0.

Appendix A. Local well-posedness for Sobolev data in H

7
4
+
For simplicity, we will only sketch the proof for the equation
(25) u = (Du)
2
,
where D stands for the spatial gradient. To treat the general equation (1), which may
contain time derivatives in the nonlinearity as well, one can apply the arguments below to
the equation (
t
u) =
t
(u)
2
with appropriate initial data, for which the estimates can
be closed at one derivative less regularity. This, along with the presented argument, will
allow one to close the estimates for u in H
3
4
+
.
We start by stating the Strichartz estimates for the wave equation. For proofs of these
estimates see [11] or [5].
A pair of indices (p, q) is called wave-admissible in dimension n, if it belongs to the set
/ =
_
(p, q) : 2 p , 2 q < ,
2
p
+
n 1
q

n 1
2
_
.
We will use the notation p

for the Lebesgue conjugate of the index p, that is,

1
p
+
1
p

= 1.
Theorem A.1 (Strichartz estimates). if (q, r) and ( q, r) are wave-admissible, then
|D
s
u|
L
q
t
L
r
x
|(u
0
, u
1
)|

H
1
+|D
s
u|
L
q

t
L
r

x
,
provided the following scaling condition holds
1
q
+
n
r
s =
n
2
=
1
q

+
n
r

2 s.
14 GRIGORYAN AND TANGUAY
The pair (p, q) = (4, ) barely fails to be wave-admissible in dimension n = 2. However,
to keep the notation simple, we will assume that it is wave-admissible, and proceed with the
estimates using this pair. For the proper proof one can take for example (p, q) =
_
12
34
,
3
2
_
for a small > 0, which will give s =
1
4
+, and the estimates will close in H
7
4
+
instead of
H
7
4
.
For the following choice of indeces in dimension n = 2,
p = 4, q = , s =
1
4
; = 1; p

= 1, q

= 2, s = 0,
for which the scaling condition is satised, one gets the estimate
|D
1
4
u|
L
4
t
L

x
|(u
0
, u
1
)|

H
1
L
2
+|u|
L
1
t
L
2
x
.
Applying this estimate to D
3
4
u leads to
(26) |Du|
L
4
t
L

x
|(u
0
, u
1
)|
H
7
4 H
3
4
+|u|
L
1
t
H
3
4
x
.
On the other hand, by the classical energy inequality for (25) in S
T
= [0, T) R
2
,
|Du(t, )|
H
3
4
|u(0, )|
H
7
4
+|
t
u(0, )|
H
3
4
+
_
t
0
|(DuDu)(t

, )|
H
3
4
dt

.
But, by Holders inequality, the fractional Leibniz rule, and estimate (26),
_
t
0
|(DuDu)(t

, )|
H
3
4
dt

|u|
L
1
t
H
3
4
x
T
3
4
|Du|
L
4
t
L

x
|Du|
L

t
H
3
4
x
T
3
4
(|(u
0
, u
1
)|
H
7
4 H
3
4
+|u|
L
1
t
H
3
4
x
)|Du|
L

t
H
3
4
.
Taking T small, the last estimate combined with the previous energy inequality is enough
to show the LWP for (25) with data in H
7
4
H
3
4
using the Pickards iteration method. For
the proper proof for a wave admissible pair (p, q), one would additionally need to use the
Sobolevs inequality for the spatial norm in the previous to last estimate in order to place
Du in L
q
x
.
Appendix B. The general well-posedness theorem
In this appendix we state the general local well-posedness theorem for nonlinear wave
equations with data in

H
r
s


H
r
s1
. We only include the main idea of the proof here, as,
with minor dierences, it is similar to the proof of the analogous theorem in the L
2
case,
for which we refer the reader to [19, Theorem 4.1] (see also [13, Theorem 5.3]).
Consider the Cauchy problem
u = F(u, u), (t, x) R
1+n
, (27)
(u,
t
u)[
t=0
= (f, g)

H
r
s


H
r
s1
, (28)
where u is the space-time gradient of u, and F is a smooth function with F(0) = 0.
Theorem B.1 (c.f Theorem 14 in [19]). Assume s R,
1
r
< b < 1, (0, 1 b). If
|F(u, u)|
X
r
1,b+1
A

(|u|
Z
r
s,b
)|u|
Z
r
,b
for all s, and
|F(u, u) F(v, v)|
X
r
1,b+1
A
s
(|u|
Z
r
s,b
+|v|
Z
r
s,b
)|u v|
Z
r
,b
,
IMPROVED LWP FOR QUADRATIC DERIVATIVE NLW IN 2D 15
for all u, v Z
r
s,b
, where A

: R
+
R
+
is increasing and locally Lipschitz for every s,
then
(existence) There exists u Z
r
s,b
, which solves (27)-(28) on [0, T] R
n
, where T =
T(|f|

H
r
s
+|g|

H
r
s
) > 0 depends continuously on |f|

H
r
s
+|g|

H
r
s
.
(uniqueness) The solution is unique in the class Z
r
s,b
, i.e., if u, v Z
r
s,b
both solve
(27)-(28) on [0, T] R
n
for some T > 0, then
u(t) = v(t) for t [0, T].
(Lipschitz) The solution map
(f, g) u,

H
r
s


H
r
s1
Z
r
s,b
is locally Lipschitz.
(higher regularity) If the data has higher regularity
f

H
r

, g

H
r
1
for > s,
then u C([0, T],

H
r

)C
1
([0, T],

H
r
1
) for any T > 0 for which u solves (27)-(28).
In particular, if (f, g) o, then u C

([0, T] R
n
).
The proof of the general LWP theorem relies on a xed point argument by showing that
the solution map , given by
u(t) = cos(tD) f +D
1
sin(tD)g +
_
t
0
D
1
sin((t t

)D) F(u, u) dt

,
is a contraction of some closed ball in Z
r
s,b;T
for appropriately chosen T. This is achieved
through the following estimate for the solution of the inhomogeneous equation u = F(t, x)
with initial data (f, g),
|u|
Z
r
s,b
C
_
|f|

H
r
s
+|g|

H
r
s1
+T
/2
|F|
X
r
s1,b+1
_
.
The above estimate exhibits a time factor in front of the inhomogeneity, if one is allowed to
place F in X
r
s1,b+1
, instead of the natural X
r
s1,b1
. Exploiting the time factor in this
estimate by choosing T small, and relying on the estimates in the theorem, one can show
that is a contraction of some closed ball in the space Z
r
s,b;T
, and hence has a xed point.
References
[1] P. DAncona, D. Foschi, S. Selberg Product estimates for wave-Sobolev spaces in 2 + 1 and 1 + 1
dimensions, Nonlinear partial dierential equations and hyperbolic wave phenomena, 125150, Contemp.
Math., 526, Amer. Math. Soc., Providence, RI, 2010.
[2] M. Christ, Power series solution of a nonlinear Schrodinger equation. Mathematical aspects of nonlinear
dispersive equations, 131155, Ann. of Math. Stud., 163, Princeton Univ. Press, Princeton, NJ, 2007.
[3] V. Grigoryan, A. R. Nahmod Almost critical well-posedness for nonlinear wave equation with Q null
forms in 2D , arXiv:1307.6194v1 [Math.AP].
[4] D. Foschi, S. Klainerman, Bilinear space-time estimates for homogeneous wave equations. Ann. Sci.

Ecole Norm. Sup. (4) 33 (2000), no. 2, 211274.

[5] J. Ginibre, G. Velo, Generalized Strichartz inequalities for the wave equation. J. Funct. Anal. 133 (1995),
no. 1, 5068.
[6] A. Gr unrock, An improved local well-posedness result for the modied KdV equation. Int. Math. Res.
Not. 2004, no. 61, 32873308.
[7] A. Gr unrock, Bi- and trilinear Schrodinger estimates in one space dimension with applications to cubic
NLS and DNLS, Int. Math. Res. Not. 2005, no. 41, 25252558.
[8] A. Gr unrock, S. Herr, Low regularity local well-posedness of the derivative nonlinear Schrodinger equa-
tion with periodic initial data. SIAM J. Math. Anal. 39 (2008), no. 6, 18901920.
[9] A. Gr unrock, On the wave equation with quadratic nonlinearities in three space dimensions. J. Hyper-
bolic Dier. Equ. 8 (2011), no. 1, 18
16 GRIGORYAN AND TANGUAY
[10] L. Hormander, The analysis of linear partial dierential operators. II. Dierential operators with con-
stant coecients. Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Math-
ematical Sciences], 257. Springer-Verlag, Berlin, 1983. viii+391 pp. ISBN: 3-540-12139-0
[11] M. Keel, T. Tao Endpoint Strichartz estimates. Amer. J. Math. 120 (1998), no. 5, 955980.
[12] S. Klainerman, M. Machedon, Estimates for null forms and the spaces Hs,. Internat. Math. Res. Notices,
1996, no. 17, 853865.
[13] S. Klainerman, S. Selberg, Bilinear estimates and applications to nonlinear wave equations. Commun.
Contemp. Math. 4 (2002), no. 2, 223295.
[14] S. Klainerman, S. Selberg, Remark on the optimal regularity for equations of wave maps type, C.P.D.E.,
22 (1997), 901918.
[15] H. Lindblad, A sharp counterexample to the local existence of low-regularity solutions to nonlinear wave
equations. Duke Math. J. 72 (1993), no. 2, 503539.
[16] H. Lindblad, Counterexamples to local existence for semi-linear wave equations. Amer. J. Math. 118
(1996), no. 1, 116.
[17] H. Lindblad, Counterexamples to local existence for quasilinear wave equations. Math. Res. Lett. 5
(1998), no. 5, 605622.
[18] G. Ponce, T. Sideris Local regularity of nonlinear wave equations in three space dimensions, Comm.
Partial Dierential Equations 18 (1993), no. 1-2, 169177.
[19] S. Selberg Multilinear space-time estimates and applications to local existence theory for nonlinear wave
equations. Ph.D. Thesis, Princeton University, 1999.
[20] S. Selberg, Anisotropic bilinear L
2
estimates related to the 3D wave equation. Int. Math. Res. Not.
IMRN 2008, Art. ID rnn 107, 63 pp
[21] S. Selberg Bilinear Fourier restriction estimates related to the 2D wave equation. Adv. Dierential
Equations 16 (2011), no. 7-8, 667690
[22] A. Tanguay, New bilinear estimates for quadratic-derivative nonlinear wave equations in 2+1 dimension,
Ph.D. Thesis, University of Massachusetts Amherst, 2012.
[23] A. Vargas, L. Vega Global wellposedness for 1D non-linear Schrodinger equation for data with an innite
L
2
norm, J. Math. Pures Appl. (9) 80 (2001), no. 10, 10291044.
[24] Y. Zhou, Local existence with minimal regularity for nonlinear wave equations, Amer. J. Math. 119
(1997), no. 3, 671703.
1
Department of Mathematics, Occidental College, 1600 Campus Road, Los Angeles, Cali-
fornia 90041
E-mail address: vgrigoryan@oxy.edu
2
Mathematisches Institut, Universit at T ubingen, Auf der Morgenstelle 1072076 T ubingen
E-mail address: ajtanguay@gmail.com