Abstract
We study low regularity local well-posedness of the nonlinear Schrödinger equation (NLS) with the quadratic nonlinearity \(\overline{u}^2\), posed on one-dimensional and two-dimensional tori. While the relevant bilinear estimate with respect to the \(X^{s, b}\)-space is known to fail when the regularity s is below some threshold value, we establish local well-posedness for such low regularity by introducing modifications on the \(X^{s, b}\)-space.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
1.1 Quadratic Nonlinear Schrödinger Equations
In this paper, we consider the following Cauchy problem for the quadratic nonlinear Schrödinger equation (NLS) on periodic domains:
where \(\mathcal {M} = \mathbb {T}\) or \(\mathbb {T}^2\) with \(\mathbb {T}= \mathbb {R}/ 2 \pi \mathbb {Z}\).
Our main goal is to establish low regularity local well-posedness of the quadratic NLS (1.1) on periodic domains \(\mathbb {T}\) or \(\mathbb {T}^2\). For instructive purposes, we first provide some background on the quadratic NLS
where \(\mathcal {N} (u, u)\) can be \(u^2\), \(\overline{u}^2\), or \(|u|^2\). Note that on \(\mathbb {R}^d\), if u is a solution to (1.2), then \(u_\lambda (x, t):= \lambda ^2 u (\lambda x, \lambda ^2 t)\) is also a solution to (1.2) for any \(\lambda > 0\). This scaling symmetry induces the following scaling critical Sobolev regularity:
When \(d \le 3\), the scaling critical regularity is negative, which often fails to predict well-posedness and ill-posedness issues. In this paper, we mainly focus on the cases when \(d = 1\) and \(d = 2\).
Let us now review some previous results on the quadratic NLS (1.2), starting with the real line case. In [16], Kenig-Ponce-Vega used the Bourgain space \(X^{s, b}\) (see Sect. 2.2) to prove local well-posedness of (1.2) on \(\mathbb {R}\) for all types of nonlinearities \(u^2\), \(\overline{u}^2\), and \(|u|^2\). Specifically, they established the following bilinear estimates:
for \(s > -\frac{3}{4}\) and \(b = \frac{1}{2} +\) andFootnote 1
for \(s > -\frac{1}{4}\) and \(b = \frac{1}{2} +\). In addition, in the same paper, they showed that (1.3) and (1.4) fail for \(s < -\frac{3}{4}\) and (1.5) fails for \(s < -\frac{1}{4}\). The failure of these bilinear estimates at the endpoint regularities were established in [25]. Despite the failure of the bilinear estimate (1.3), Bejenaru-Tao [2] showed local well-posedness of (1.2) on \(\mathbb {R}\) with nonlinearity \(\mathcal {N} (u, u) = u^2\) for \(s \ge -1\) by introducing weighted function spaces. Moreover, they proved ill-posedness of the same equation for \(s < -1\). Later, Kishimoto [17] proved local well-posedness of (1.2) on \(\mathbb {R}\) with \(\mathcal {N} (u, u) = \overline{u}^2\) for \(s \ge -1\) using different weighted function spaces. He also proved ill-posedness of the same equation for \(s < -1\). Regarding (1.2) on \(\mathbb {R}\) with \(\mathcal {N} (u, u) = |u|^2\), Kishimoto [18] showed local well-posedness for \(s \ge -\frac{1}{4}\) and ill-posedness for \(s < -\frac{1}{4}\) (see also [22]). See also [13, 14, 21] for stronger ill-posedness results in the same ranges of s. For convenience, we summarize these results in Table 1. Note that for all these nonlinearities \(u^2\), \(\overline{u}^2\), and \(|u|^2\), well-posedness and ill-posedness results are sharp. Also, for all these nonlinearities, ill-posedness occurs before s reaches the scaling critical regularity of (1.2) on \(\mathbb {R}\): \(s_{\text {crit}} = -\frac{3}{2}\).
Let us also mention well-posedness and ill-posedness results of (1.2) on \(\mathbb {R}^2\), which are again summarized in Table 1. The \(X^{s, b}\)-bilinear estimates (1.3), (1.4), and (1.5) were established in [5, 30]. The failure of these \(X^{s, b}\)-bilinear estimates for lower values of s was shown in [5, 25]. For local well-posedness of (1.2) on \(\mathbb {R}^2\), see [1, 18, 21]. For ill-posedness of (1.2) on \(\mathbb {R}^2\), see [13, 14, 21]. From Table 1, we note that the ill-posedness on \(\mathbb {R}^2\) for \(\mathcal {N} (u, u) = |u|^2\) occurs before s reaches the scaling critical regularity \(s_{\text {crit}} = -1\). Also, we can see that all well-posedness and ill-posedness results are sharp on \(\mathbb {R}^2\).
We now turn our attention to well-posedness and ill-posedness results of (1.2) on periodic domains \(\mathbb {T}\) and \(\mathbb {T}^2\). The results are summarized in Table 2. On \(\mathbb {T}\), for all nonlinearities \(u^2\), \(\overline{u}^2\), and \(|u|^2\), the \(X^{s, b}\)-bilinear estimates (1.3), (1.4), and (1.5) for \(s \ge 0\) follows immediately from the \(L^3\)-Strichartz estimate, which is obtained by interpolating the \(L^4\)-Strichartz estimate on \(\mathbb {T}\) (see [3, 32]) and the trivial \(L^2\)-bound. In [16], Kenig-Ponce-Vega established bilinear estimates (1.3) (for \(u^2\)) and (1.4) (for \(\overline{u}^2\)) on \(\mathbb {T}\) for \(s > -\frac{1}{2}\) and \(b = \frac{1}{2} +\) and showed corresponding local well-posedness results. They also showed that (1.3) and (1.4) fail on \(\mathbb {T}\) when \(s < -\frac{1}{2}\) and (1.5) (for \(|u|^2\)) fails on \(\mathbb {T}\) when \(s < 0\). Later, Kishimoto [21] showed ill-posedness of (1.2) on \(\mathbb {T}\) with all types of nonlinearities for regularity ranges shown in Table 2. Here, we note that there are gaps between local well-posedness and ill-posedness results for nonlinearities \(u^2\) and \(\overline{u}^2\). Also, the quadratic NLS (1.2) with nonlinearity \(|u|^2\) behaves worse on \(\mathbb {T}\) than on \(\mathbb {R}\), since ill-posedness on \(\mathbb {T}\) occurs for a wider range of s than on \(\mathbb {R}\).
For (1.2) on \(\mathbb {T}^2\) with all nonlinearities \(u^2\), \(\overline{u}^2\), and \(|u|^2\), the \(X^{s, b}\)-bilinear estimates (1.3), (1.4), and (1.5) for \(s > 0\) follows from the \(L^3\)-Strichartz estimate with an \(\varepsilon \) derivative loss, which is obtained by interpolating the \(L^4\)-Strichartz estimate on \(\mathbb {T}^2\) (see Lemma 2.4) and the trivial \(L^2\)-bound. In [9], Grünrock showed the bilinear estimate (1.4) (for \(\overline{u}^2\)) for \(s > -\frac{1}{2}\) and proved the corresponding local well-posedness result. In the same paper, he showed the failure of (1.3) (for \(u^2\)) on \(\mathbb {T}^2\) when \(s < 0\) and the failure of (1.4) (for \(\overline{u}^2\)) on \(\mathbb {T}^2\) when \(s < -\frac{1}{2}\). In [21], Kishimoto showed ill-posedness of (1.2) on \(\mathbb {T}^2\) with all types of nonlinearities for regularity ranges shown in Table 2. In a recent work, Oh and the author [23] proved local well-posedness of (1.2) with nonlinearities \(u^2\) and \(|u|^2\) for \(s = 0\) by establishing correponding \(X^{s, b}\)-bilinear estimates.
The long-time behaviors of the quadratic NLS (1.2) have also been studied. For global existence and scattering results, see [7, 8, 11, 15, 24, 28]. For nonexistence of non-trivial scattering solutions, see [27, 29]. For finite-time blowup results, see [12, 26].
As can be seen from Table 2, local well-posedness for the quadratic NLS with nonlinearity \(|u|^2\) is complete, whereas for nonlinearities \(u^2\) and \(\overline{u}^2\), there are gaps between local well-posedness and ill-posedness results. The difference of well-posedness behaviors of these three nonlinearities is closed related to their distinct phase functions. By letting \(n_1, n_2\) be the frequencies of the nonlinearity and n be the frequency of the duality term, we can write out the frequency interactions and phase functions for these three nonlinearities as in Table 3.
When the phase function is large, we expect some gain of regularities. For example, for nonlinearity \(\overline{u}^2\) on \(\mathbb {T}^2\), the phase function \(|n|^2 + |n_1|^2 + |n_2|^2\) provides gain of derivatives, so that one can establish local well-posedness for nonlinearity \(\overline{u}^2\) with very rough initial data. On the other hand, for nonlinearity \(u^2\) on \(\mathbb {T}^2\), the phase function \(|n|^2 - |n_1|^2 - |n_2|^2 = 2 n_1 \cdot n_2\) can be very small if \(n_1\) and \(n_2\) are almost perpendicular to each other, so that local well-posedness with rough initial data is much harder. In this paper, we focus on shrinking the well-posedness gap for nonlinearity \(\overline{u}^2\) by establishing local well-posedness with lower regularity. We also discuss some well-posedness issues for nonlinearity \(u^2\) in Remark 1.6 below.
We now look back on low regularity local well-posedness of the quadratic NLS (1.1) on \(\mathbb {T}\) and \(\mathbb {T}^2\). In this paper, we prove the following theorem.
Theorem 1.1
Let \(\mathcal {M} = \mathbb {T}\) or \(\mathbb {T}^2\). Then, the quadratic NLS (1.1) is locally well-posed in \(H^s (\mathcal {M})\) for \(s > -\frac{2}{3}\). More precisely, given any \(u_0 \in H^s (\mathcal {M})\), there exists \(T = T(\Vert u_0 \Vert _{H^s}) > 0\) and a unique solution \(u \in C([-T, T]; H^s (\mathcal {M}))\) to (1.1) with \(u|_{t = 0} = u_0\), and the solution u depends continuously on the initial data \(u_0\).
Since local well-posedness of (1.1) for \(s > -\frac{1}{2}\) was already shown in [16] on \(\mathbb {T}\) and in [9] on \(\mathbb {T}^2\), we mainly focus on the situation when \(-\frac{2}{3} < s \le -\frac{1}{2}\). Our proof of Theorem 1.1 relies on modified \(X^{s,b}\)-spaces for the solutions, and so the uniqueness in the above statement holds only in the relevant function space (see the \(Z^{s, b}\)-norm in (2.4) and its local-in-time version in (2.6)). For the proof of Theorem 1.1, we will mainly focus on the case \(\mathcal {M} = \mathbb {T}^2\) (see Remark 1.2). The idea of the proof of Theorem 1.1 is to introduce modifications on the \(X^{s,b}\)-space which enable us to prove the corresponding bilinear estimate. See Sect. 1.2 for more discussion on it.
Theorem 1.1 improves the previous local well-posedness results in [9, 16]. In addition, to the best of the author’s knowledge, these are the first local well-posedness results for the quadratic NLS on periodic domains below the regularity thresholds where the usual \(X^{s, b}\)-bilinear estimates fail. We also remark that the bound \(s > -\frac{2}{3}\) is sharp (up to the endpoint regularity \(s = -\frac{2}{3}\)) in our approach. See Sect. 1.2 for more details.
Remark 1.2
In Theorem 1.1, the proof for \(\mathcal {M} = \mathbb {T}\) follows from the proof for \(\mathcal {M} = \mathbb {T}^2\) with minor modifications. Thus, in proving Theorem 1.1, we mainly restrict our attention on the case \(\mathcal {M} = \mathbb {T}^2\).
1.2 Modified Function Spaces
In this subsection, we briefly explain our strategy for proving Theorem 1.1.
In [2], Bejenaru-Tao reduced the well-posedness problem of the quadratic NLS (1.2) in \(H^s (\mathbb {R}^d)\) or \(H^s (\mathbb {T}^d)\) to finding a space-time norm \(\Vert \cdot \Vert _{W^s}\) that satisfy the following propertiesFootnote 2:
(i) (Monotonicity) If \(|\widehat{f}| \le |\widehat{g}|\) pointwise, then
Here, \(\widehat{f}\) is the space-time Fourier transform of f.
(ii) (\(H^s\)-energy estimate) The following inequality holds:
where \(\langle \cdot \rangle = (1 + |\cdot |^2)^{\frac{1}{2}}\).
(iii) (Homogeneous linear estimate) There exists \(b \in \mathbb {R}\) such that
where the \(X^{s, b}\)-norm is as defined in (2.1).
(iv) (Bilinear estimate) The following inequality holds:
where \(\widehat{W}^s\) is the same norm \(W^s\) on the Fourier side and \(\mathcal {B} (f, g)\) is equal to \(f * g\) (if \(\mathcal {N} (u, u) = u^2\)), \(\overline{\widetilde{f}} * \overline{\widetilde{g}}\) (if \(\mathcal {N} (u, u) = \overline{u}^2\)), or \(f * \overline{\widetilde{g}}\) (if \(\mathcal {N} (u, u) = |u|^2\)). Here, \(\widetilde{f} (\xi , \tau ) = f (- \xi , -\tau )\).
Now the task is to find suitable function spaces that satisfy the properties listed above. From now on, we restrict our attention to the nonlinearity \(\mathcal {N} (u, u) = \overline{u}^2\) and the domain \(\mathbb {T}^2\). As we have seen in the previous subsection, the usual \(X^{s, b}\)-bilinear estimate fails when the regularity is very low. This failure is caused by certain “dangerous” interactions. Thus, we need to introduce modifications on the \(X^{s, b}\)-space in order to reduce the effect by those “dangerous” interactions. In the following, we discuss several examples of such interactions and our strategy to deal with them.
Example 1
For a large number \(N \in {\mathbb {N}}\), let
where \(e_1 = (1, 0)\). Note that \(\Vert u_N \Vert _{X^{s, b}} \sim N^s\) and \(\Vert v_N \Vert _{X^{s, b}} \sim N^s\). A direct computation yields
and so \(\Vert \overline{u_N} \overline{v_N} \Vert _{X^{s, b - 1}} \ge N^{2b - 2}\). Thus, the bilinear estimate (1.4) holds only if \(2b - 2 \le 2\,s\) or \(s \ge b - 1\). Since we need \(b > \tfrac{1}{2}\), we require that \(s > -\tfrac{1}{2}\).
In the above example, the frequency interaction is “high-high to low” and the modulation interaction is “low-low to high”. However, the modulation for \(\widehat{\overline{u_N} \overline{v_N}}\) is not high enough for the desired \(X^{s,b}\)-bilinear estimate when \(s \le -\tfrac{1}{2}\). To control the above interaction when \(s \le - \tfrac{1}{2}\), we consider the following \(Y^{s, b}\)-norm introduced by Kishimoto [19]:
and we define the space \(Z^{s, b} = X^{s, b} + Y^{s, b}\) via the norm
The \(\ell _n^2 L_\tau ^1\)-term in the \(Y^{s, b}\)-norm is needed to ensure that the \(Z^{s, b}\)-norm satisfies the \(H^s\)-energy estimate (1.7). It is not hard to check that the \(Z^{s, b}\)-norm satisfies the monotonicity property (1.6), the \(H^s\)-energy estimate (1.7), and the homogeneous linear estimate (1.8). Note that for \(s \le 0\) and \(b > \frac{1}{2}\), if \(\mathop {\textrm{supp}}\limits \widehat{u} \subset \{ |\tau + |n|^2| \le |n|^2 \}\), then we have
if \(\mathop {\textrm{supp}}\limits \widehat{u} \subset \{ |\tau + |n|^2| \ge |n|^2 \}\), then we have
In Sect. 2.3, we will revisit this \(Z^{s, b}\)-norm, which will be defined in a more precise manner for practical purposes.
In Example 1, because of the high modulation of \(\widehat{\overline{u_N} \overline{v_N}}\), the \(\widehat{Z}^{s, b}\)-norm (i.e. the \(Z^{s, b}\)-norm on the Fourier side) of \(\langle \tau + |n|^2 \rangle ^{-1} \widehat{\overline{u_N} \overline{v_N}}\) is small enough to obtain the desired bilinear estimate (1.9) for \(s \le - \frac{1}{2}\). One can easily check that using the \(Z^{s, b}\)-norm, the bilinear estimate for the above example holds for \(s \ge 2b - 2\). This is better than \(s > -\tfrac{1}{2}\) as long as \(\tfrac{1}{2} < b \le \tfrac{3}{4}\).
Let us take a look at another example using the \(Z^{s, b}\)-norm assuming that \(s \le 0\).
Example 2
For a large number \(N \in {\mathbb {N}}\), let
A direct computation yields
Note that in this example, the frequency interaction is “high-high to low” and the modulation interaction is “low-high to low”. We can compute their corresponding \(Z^{s,b}\)-norms as follows:
Thus, the bilinear estimate (1.9) with \(W = Z^{s, b}\) holds only if \(0 \le 2\,s + 2b\) or \(s \ge -b\).
Combining Example 1 and Example 2, we notice that the regularity s needs to satisfy \(s \ge 2b - 2\) and \(s \ge -b\). These two lower bounds become optimal when \(b = \frac{2}{3}\), so that \(s = - \frac{2}{3}\) seems to be the threshold of the bilinear estimate (1.9) with \(W^s = Z^{s, b}\). In fact, we will show in Sect. 3 that the bilinear estimate (1.9) with \(W^s = Z^{s, \frac{2}{3}}\) holds when \(s > - \frac{2}{3}\) (see Remark 1.4 for a discussion on the slight loss of regularity). See Sect. 3 for more details.
We conclude this introduction by stating several remarks.
Remark 1.3
On \(\mathbb {T}^d\), it is possible to use a scaling argument to prove local well-posedness for large initial data given that one can first obtain small data local well-posedness. See [6]. However, we do not pursue the scaling argument in this paper and instead rely on the time localization (Lemma 2.3) to prove local well-posedness for large initial data.
Remark 1.4
In [2, 17], a Besov refinement was considered in constructing desired function spaces so that the endpoint regularity (i.e. \(s = -1\) for the quadratic NLS (1.2) on \(\mathbb {R}\) with \(\mathcal {N}(u, u) = u^2\) or \(\overline{u}^2\)) can be handled. Similar Besov refinements were used by [10, 20] in the context of the Korteweg-de Vris equation.
For the quadratic NLS (1.1) on \(\mathbb {T}^2\), however, such Besov modification does not seem to be enough to cover the case when \(s = - \frac{2}{3}\). This is mainly due to the fact that our approach relies heavily on the \(L^4\)-Strichartz estimate on \(\mathbb {T}^2\) (see Lemma 2.4), which has an \(\varepsilon \) loss of derivative.
For the quadratic NLS (1.1) on \(\mathbb {T}\), since the \(L^4\)-Strichartz estimate on \(\mathbb {T}\) (see [3]) does not have any derivative loss, it seems possible to adapt the Besov modification to our estimate so that the endpoint case can be included.
Remark 1.5
For the quadratic NLS (1.1) on \(\mathbb {T}\) and \(\mathbb {T}^2\), there are still gaps between local well-posedness and ill-posedness results (see Table 2). Specifically, on \(\mathbb {T}\), well-posedness issues of (1.1) for \(-1 \le s \le - \frac{2}{3}\) remain open; on \(\mathbb {T}^2\), well-posedness issues of (1.1) for \(-1 < s \le - \frac{2}{3}\) remain open. One possible strategy for improving our local well-posedness arguments is to introduce weighted spaces as in [1, 2, 17, 19] in the context of Euclidean spaces.
Remark 1.6
Let us consider the quadratic NLS (1.2) with \(\mathcal {N} (u, u) = u^2\). On \(\mathbb {T}\), local well-posedness is known to hold for \(s > -\frac{1}{2}\) and ill-posedness holds for \(s < -1\). We believe that the method of using modified function spaces should be able to produce better local well-posedness results, but one may need to use the weighted spaces as in [1, 2] to handle the corresponding bilinear estimate.
For the quadratic NLS (1.2) with \(\mathcal {N} (u, u) = u^2\) on \(\mathbb {T}^2\), local well-posedness is known to hold for \(s \ge 0\) and ill-posedness holds for \(s \le -1\). However, it seems unlikely that the method of finding modified function spaces as illustrated at the beginning of Sect. 1.2 works in the range \(s < 0\). This is due to the following example in [9]. For a large number \(N \in \mathbb {N}\), let
where \(e_1 = (1, 0)\) and \(e_2 = (0, 1)\). A direct computation yields
In this example, the frequency interaction is “high-high to high” and the modulation interaction is “low-low to low”, which means that there seems to be no way to utilize the modulation to improve the bilinear estimate. Note that this “low-low to low” interaction does not occur for the nonlinearity \(\mathcal {N} (u, u) = \overline{u}^2\), which can be seen from the computations at the beginning of Subcase 2.3 of Lemma 3.2 below and Case 3 of Lemma 3.3 below.
For any \(s \in \mathbb {R}\) and \(b \in \mathbb {R}\), we have
where the \(\widehat{X}^{s, b}\)-norm is the \(X^{s, b}\)-norm on the Fourier side. Thus, we observe that due to the homogeneous linear estimate (1.8) and the similar structures of \(\widehat{u_N}\), \(\widehat{v_N}\), and \(\widehat{u_N v_N}\), any qualified modified norm \(\Vert \cdot \Vert _{W^s}\) should decrease the corresponding norms of \(\widehat{u_N}\), \(\widehat{v_N}\), and \(\langle \tau + |n|^2 \rangle ^{-1} \widehat{u_N v_N}\) with the same rate (with respect to N). Suppose that there exists \(a \ge 0\) such that
where the \(\widehat{W}^s\)-norm is the \(W^s\)-norm on the Fourier side. Then, for the bilinear estimate (1.9) to hold, we must have
so that \(s - a \ge 0\) or \(s \ge a \ge 0\). Therefore, we do not expect that the method of finding the \(W^s\)-norm for proving local well-posedness works for the quadratic NLS (1.2) with \(\mathcal {N} (u, u) = u^2\) on \(\mathbb {T}^2\) for \(s < 0\), and it is possible that some ill-posedness results may hold in this range.
2 Notations and Function Spaces
In this section, we introduce some notations and function spaces that enable us to prove local well-posedness of (1.1) in low regularity settings.
2.1 Notations
Throughout this paper, we drop the inessential factor of \(2\pi \). For a space-time distribution u, we write \(\widehat{u}\) or \(\mathcal {F}_{x, t} u\) to denote the space-time Fourier transform of u. If a function \(\phi \) only has a space (or time) variable, then we use \(\widehat{\phi }\) to denote the Fourier transform of \(\phi \) with respect to the space (or time, respectively) variable. For any function f, the function \(\widetilde{f}\) is the reflection of f, i.e. \(\widetilde{f}(x) = f(-x)\). We also set \(\langle \,\cdot \, \rangle = (1 + |\cdot |^2)^\frac{1}{2}\).
We use \(A \le B\) to denote \(A \le CB\) for some constant \(C > 0\). We write \(A \sim B\) if we have \(A \le B\) and \(B \le A\). We may use subscripts to denote dependence on external parameters. We also use \(a +\) and \(a -\) to denote \(a + \varepsilon \) and \(a - \varepsilon \), respectively, for sufficiently small \(\varepsilon > 0\).
Given a dyadic number \(N \in 2^{\mathbb {N}\cup \{0\}}\), if \(N \ge 2\), we let \(P_N\) be the spatial frequency projector onto the frequencies
If \(N = 1\), we let \(P_1\) be the spatial frequency projector onto the frequencies
For a space-time distribution u, we also write \(u_N:= P_N u\) for simplicity.
2.2 Fourier Restriction Norm Method
In this subsection, we recall the definition and estimates of \(X^{s, b}\)-spaces for the Schrödinger equations, which were first introduced by Bourgain [3]. Given \(s, b \in \mathbb {R}\), we define the space \(X^{s, b} = X^{s, b} (\mathbb {T}^2 \times \mathbb {R})\) to be the completion of functions that are smooth in space and Schwartz in time with respect to the following norm:
We now present and recall some estimates related to \(X^{s, b}\)-norms, starting with the following stronger version of the usual homogeneous linear estimate of the \(X^{s, b}\)-norm as in [3, 31].
Lemma 2.1
Let \(\varphi \) be a smooth function supported on \([-2, 2]\). Let \(s \in \mathbb {R}\), \(b \le 1\), and \(k \in \mathbb {N}\cup \{0\}\). Then, we have
Proof
Note that by the fact that \(b \le 1\), we have
as desired. \(\square \)
Remark 2.2
In fact, the estimate (2.2) holds for all \(b \in \mathbb {R}\). For the proof of our local well-posedness result, however, we will only need the estimate (2.2) for \(b \le 1\).
Next, we recall the following time localization estimate. For a proof, see [3, 31].
Lemma 2.3
Let \(s \in \mathbb {R}\), \(-\frac{1}{2}< b_1 \le b_2 < \frac{1}{2}\), and \(0 < T \le 1\). Let \(\varphi \) be a Schwartz function and let \(\varphi _T (t):= \varphi (t / T)\). Then, we have
We also record the following \(L^4\)-Strichartz estimate on \(\mathbb {T}^2\). For a proof, see [3, 4].
Lemma 2.4
Let N be a dyadic number. Then, we have
where \(0< s < \frac{1}{2}\) and \(b > \frac{1 - s}{2}\).
2.3 Modified Function Spaces
In this subsection, we define our solution space for the quadratic NLS (1.1) in the low regularity setting and establish corresponding linear estimates.
Given \(s, b \in \mathbb {R}\), we define the space \(Y^{s, b} = Y^{s, b} (\mathbb {T}^2 \times \mathbb {R})\) to be the completion of functions that are smooth in space and Schwartz in time with respect to the norm
The idea of this modification comes from Kishimoto [19].
We now define the space \(Z^{s, b}\) via the norm
where \(P_{\text {lo}}\) is the space-time frequency projector onto the frequencies \(\{ |\tau + |n|^2| < 2^{-10} |n|^2 \}\) and \(P_{\text {hi}}\) is the space-time frequency projector onto the frequencies \(\{ |\tau + |n|^2| \ge 2^{-10} |n|^2 \}\). From the definition, we observe that the \(Z^{s, b}\)-norm has the monotonicity property: if \(|\widehat{u_1}| \le |\widehat{u_2}|\) pointwise, then
For \(T > 0\), we define the space \(Z_T^{s, b}\) as the restriction of the \(Z^{s, b}\)-space onto the time interval \([-T, T]\) via the norm:
Note that the \(Z_T^{s, b}\)-space is complete.
For convenience and conciseness, later on we may use the notations \(\widehat{X}^{s, b}\), \(\widehat{Y}^{s, b}\), and \(\widehat{Z}^{s, b}\) to denote the corresponding norms on the Fourier side. In other words, for a complex-valued function f defined on \(\mathbb {Z}^2 \times \mathbb {R}\), we write
where \(\mathcal {F}^{-1}\) is the inverse Fourier transform.
We now establish some linear estimates of the \(Z^{s, b}\)-norm. We start with the following \(H^s\)-energy estimate.
Lemma 2.5
Let \(s \in \mathbb {R}\) and \(b > \frac{1}{2}\). Then, we have
Proof
By the definition of the \(Z^{s, b}\)-norm in (2.4), we know that it suffices to show the following two estimates:
Since \(b > \frac{1}{2}\), we use the Cauchy-Schwarz inequality in \(\tau \) to obtain
so that we obtain (2.7). Also, note that (2.8) is easily obtained from the definition of the \(Y^{s, b}\)-norm in (2.3). \(\square \)
The above lemma implies the following embedding result.
Lemma 2.6
Let \(s \in \mathbb {R}\), \(b > \frac{1}{2}\), and \(T > 0\). Then, we have
Consequently, the embedding
holds.
Proof
Let \(\varepsilon > 0\) and let v be an extension of u outside of \([-T, T]\) such that
Note that we have the following embedding
Thus, by (2.10), Lemma 2.5, and (2.9), we obtain
and so the desired estimate follows since \(\varepsilon > 0\) can be arbitrarily small. \(\square \)
Lastly, we show the following lemma, which shows that the \(X^{s, b}\)-space is embedded in the \(Z^{s, b}\)-space.
Lemma 2.7
Let \(s \le 0\) and \(b > \frac{1}{2}\). Then, we have
Proof
We recall from (2.4) that
where \(P_{lo }\) projects the space-time frequencies onto \(\{ |\tau + |n|^2| < 2^{-10} |n|^2 \}\) and \(P_{hi }\) projects the space-time frequencies onto \(\{ |\tau + |n|^2| \ge 2^{-10} |n|^2 \}\). Note that we have
For the \(\Vert P_{\text {hi}} u \Vert _{Y^{s, b}}\) term, note that by the Cauchy-Schwarz inequality, we have
since \(b > \frac{1}{2}\). Also, we have
Thus, we obtain that \(\Vert P_{\text {hi}} u \Vert _{Y^{s, b}} \le \Vert u \Vert _{X^{s, b}}\), so that we achieve the desired inequality. \(\square \)
3 Bilinear Estimate
In this section, we establish the crucial bilinear estimate with respect to the \(Z^{s, b}\)-norm introduced in the previous section. Specifically, we show the following proposition.
Proposition 3.1
Let \(- \frac{2}{3} < s \le - \frac{1}{2}\) and \(0 < T \le \frac{1}{2}\). Let \(\varphi : \mathbb {R}\rightarrow [0, 1]\) be a smooth function such that \(\varphi \equiv 1\) on \([-1, 1]\) and \(\varphi \equiv 0\) outside of \([-2, 2]\), and let \(\varphi _T (t):= \varphi (t / T)\). Then, we have
for some \(\theta > 0\).
Let us first consider two particular cases of Proposition 3.1. We start with the following “high-low interaction” estimate.
Lemma 3.2
Let \(-\frac{2}{3} < s \le -\frac{1}{2}\) and \(0 < T \le \frac{1}{2}\). Let N, \(N_1\), and \(N_2\) be dyadic numbers. Let \(\varphi : \mathbb {R}\rightarrow [0, 1]\) be a smooth function such that \(\varphi \equiv 1\) on \([-1, 1]\) and \(\varphi \equiv 0\) outside of \([-2, 2]\), and let \(\varphi _T (t):= \varphi (t / T)\).
(i) If \(2^{-5} N \le N_1 \le 2^5 N\) and \(N_2 \le 2^6 N\), we have
for some \(\delta > 0\) and \(\theta > 0\).
(ii) If \(2^{-5} N \le N_2 \le 2^5 N\) and \(N_1 \le 2^6 N\), we have
for some \(\delta > 0\) and \(\theta > 0\).
Proof
By the symmetry of u and v, it suffices to prove (i). Below we use \((n_1, \tau _1)\) as the variables of \(\widehat{ \varphi _T u_{N_1} }\) or \(\widehat{u_{N_1}}\) and \((n_2, \tau _2)\) as the variables of \(\widehat{ \varphi _T v_{N_2} }\) or \(\widehat{v_{N_2}}\). Note that we have the relations \(\tau + \tau _1 + \tau _2 = 0\) and \(n + n_1 + n_2 = 0\). We also recall the notation \(\widetilde{f} (x) = f(-x)\).
We divide the argument into two main cases depending on the relationship between the modulation function \(\tau + |n|^2\) and \(|n|^2\).
Case 1 \(| \tau + |n|^2 | \ge 2^{-10} |n|^2\).
In this case, we need to evaluate the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term using the \(\widehat{Y}^{s, \frac{2}{3}}\)-norm, and we need to evaluate both the \(\ell _n^2 L_\tau ^1\) term and the \(\ell _n^2 L_\tau ^2\) term. We consider the following three subcases.
Subcase 1.1 \(|\tau _1 + |n_1|^2| \ge 2^{-10} |n_1|^2\).
In this subcase, we need to estimate \(u_{N_1}\) using the \(Y^{s, \frac{2}{3}}\)-norm. By Young’s convolution inequality, Lemma 2.3, the Cauchy-Schwarz inequality, and Lemma 2.5, we obtain
where \(\varepsilon > 0\) is arbitrarily small. Since \( - s + 1 > 0\) given \(s \le -\frac{1}{2}\), the above estimate is acceptable if \(-s - 1 + 2\varepsilon < 0\), which is valid given \(s > - \frac{2}{3}\) and \(\varepsilon > 0\) sufficiently small.
Also, by the Cauchy-Schwarz inequality, we get
which can be estimated similarly as in (3.1). Combining the above two estimates, we obtain the desired inequality.
Subcase 1.2 \(|\tau _2 + |n_2|^2| \ge 2^{-10} |n_2|^2\).
In this subcase, we need to estimate \(v_{N_2}\) using the \(Y^{s, \frac{2}{3}}\)-norm. By Young’s convolution inequality, the Cauchy-Schwarz inequality, Lemma 2.3, and Lemma 2.5, we obtain
where \(\varepsilon > 0\) is arbitrarily small. Since \(-s - \frac{1}{3} + 2\varepsilon > 0\) given \(s \le - \frac{1}{2}\), the above estimate is acceptable if \(-s - 1 + 2\varepsilon < 0\), which is valid given \(s > - \frac{2}{3}\) and \(\varepsilon > 0\) sufficiently small.
Also, by the Cauchy-Schwarz inequality, we get
which can be estimated similarly as in (3.2). Combining the above two estimates, we obtain the desired inequality.
Subcase 1.3 \(|\tau _1 + |n_1|^2| < 2^{-10} |n_1|^2\) and \(|\tau _2 + |n_2|^2| < 2^{-10} |n_2|^2\).
In this subcase, we need to estimate both \(u_{N_1}\) and \(v_{N_2}\) using the \(X^{s, \frac{2}{3}}\)-norm. Using the fact that \(\varphi _T\) is supported on \([-1, 1]\) given \(0 < T \le \frac{1}{2}\), by the Plancherel theorem, Hölder’s inequality, Lemma 2.4, and Lemma 2.3, we obtain
where \(\varepsilon > 0\) is arbitrarily small. Since \(s \le - \frac{1}{2} < 0\), the above estimate is acceptable if \(-s - \frac{2}{3} + 8\varepsilon < 0\), which is valid given \(s > - \frac{2}{3}\) and \(\varepsilon > 0\) small enough.
Also, by the Cauchy-Schwarz inequality, we get
which can be estimated similarly as in (3.3). Combining the above two estimates, we obtain the desired inequality.
Case 2 \(|\tau + |n|^2| < 2^{-10} |n|^2\).
In this case, we need to evaluate the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term using the \(\widehat{X}^{s, \frac{2}{3}}\)-norm.
We assume that \(n \ne 0\). Note that if \(n = 0\), we have \(N = 1\) which then implies that \(N_1 \le 2^5\) and \(N_2 \le 2^6\), and so the estimate will follow in a similar (and much easier) manner.
We consider the following three subcases.
Subcase 2.1 \(|\tau _1 + |n_1|^2| \ge 2^{-10} |n_1|^2\) and \(|\tau _2 + |n_2|^2| \ge 2^{-10} |n_2|^2\).
In this subcase, we need to estimate both \(u_{N_1}\) and \(v_{N_2}\) using the \(Y^{s, \frac{2}{3}}\)-norm. By Hölder’s inequality, Young’s convolution inequality, and Lemma 2.3, we have
where \(\varepsilon > 0\) is arbitrarily small. The above estimate is acceptable if \(-s - \frac{4}{3} + 2\varepsilon < 0\), which is valid given \(s > - \frac{2}{3}\) and \(\varepsilon > 0\) sufficiently small.
Subcase 2.2 \(| \tau _1 + |n_1|^2 | \ge 2^{-10} |n_1|^2\) and \(| \tau _2 + |n_2|^2 | < 2^{-10} |n_2|^2\).
In this subcase, we need to estimate \(u_{N_1}\) using the \(Y^{s, \frac{2}{3}}\)-norm and estimate \(v_{N_2}\) using the \(X^{s, \frac{2}{3}}\)-norm. By duality and the Cauchy-Schwarz inequality, we have
Let \(w_N\) be a space-time distribution that satisfy \(\widehat{w_N} (n, \tau ) = h(n, \tau ) / \langle \tau + |n|^2 \rangle ^{\frac{1}{3}}\). Then, using the fact that \(\varphi _T\) is supported on \([-1, 1]\) given \(0 < T \le \frac{1}{2}\), by the Plancherel theorem, Hölder’s inequality, Lemma 2.4, and Lemma 2.3, we have
where \(\varepsilon > 0\) is arbitrarily small. Thus, continuing with (3.4), we use Lemma 2.3 to obtain
Since \(s < 0\), the above estimate is acceptable if \(-s - 1 + 6 \varepsilon < 0\), which is valid given \(s > - \frac{2}{3}\) and \(\varepsilon > 0\) small enough.
Subcase 2.3 \(|\tau _1 + |n_1|^2| < 2^{-10} |n_1|^2\).
In this subcase, we first note that
Note that since we assumed \(n \ne 0\), we have
Thus, we have
and \(|\tau _2 + |n_2|^2| \ge |n_2|^2 > 2^{-10} |n_2|^2\).
We need to estimate \(u_{N_1}\) using the \(X^{s, \frac{2}{3}}\)-norm and estimate \(v_{N_2}\) using the \(Y^{s, \frac{2}{3}}\)-norm. By using similar steps as in Subcase 2.2 by switching the roles of \(u_{N_1}\) and \(v_{N_2}\) along with the additional condition (3.5), we obtain
where \(\varepsilon > 0\) is arbitrarily small. The above estimate is acceptable if \(-s - 1 + 6 \varepsilon < 0\), which is valid given \(s > - \frac{2}{3}\) and \(\varepsilon > 0\) small enough.
Thus, we have finished our proof. \(\square \)
We now show the following “high-high interaction” estimate.
Lemma 3.3
Let \(-\frac{2}{3} < s \le -\frac{1}{2}\) and \(0 < T \le \frac{1}{2}\). Let N, \(N_1\), and \(N_2\) be dyadic numbers such that \(\frac{1}{2} N_1 \le N_2 \le 2 N_1\) and \(N < 2^{-5} N_1\). Let \(\varphi : \mathbb {R}\rightarrow [0, 1]\) be a smooth function such that \(\varphi \equiv 1\) on \([-1, 1]\) and \(\varphi \equiv 0\) outside of \([-2, 2]\), and let \(\varphi _T (t):= \varphi (t / T)\). Then, we have
for some \(\delta > 0\) and \(\theta > 0\).
Proof
As in the proof of the previous lemma, we use \((n_1, \tau _1)\) as the variables of \(\overline{\varphi _T u_{N_1}}\) or \(\overline{u_{N_1}}\), and \((n_2, \tau _2)\) as the variables of \(\overline{\varphi _T v_{N_2}}\) or \(\overline{v_{N_2}}\). Note that we have the relations \(\tau + \tau _1 + \tau _2 = 0\) and \(n + n_1 + n_2 = 0\). Also, the assumptions on the sizes of N, \(N_1\), and \(N_2\) ensure that \(n_1 \ne 0\) and \(n_2 \ne 0\). We also recall the notation \(\widetilde{f} (x) = f(-x)\).
We consider the following four main cases.
Case 1 \(|\tau + |n|^2| \ge 2^{-10} |n_1|^2\).
In this case, we have \(|\tau + |n|^2| \ge 2^{-10} |n_1|^2 \ge 2^{-10} |n|^2\) given \(N < 2^{-5} N_1\), so that we need to evaluate the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term using the \(\widehat{Y}^{s, \frac{2}{3}}\)-norm, and we need to evaluate both the \(\ell _n^2 L_\tau ^1\) term and the \(\ell _n^2 L_\tau ^2\) term. We consider the following three subcases.
Subcase 1.1 \(|\tau _1 + |n_1|^2| \ge 2^{-10} |n_1|^2\).
In this subcase, we need to estimate \(u_{N_1}\) using the \(Y^{s, \frac{2}{3}}\)-norm. By Young’s convolution inequality, Lemma 2.3, the Cauchy-Schwarz inequality, and Lemma 2.5, we obtain
which is acceptable given \(s > - \frac{2}{3}\) and \(\varepsilon > 0\) sufficiently small.
Also, by Hölder’s inequality, Young’s convolution inequality, Lemma 2.3, and Lemma 2.5, we have
where \(\varepsilon > 0\) is arbitrarily small. Since \(s + 1 > 0\) given \(s > - \frac{2}{3}\), the above estimate is acceptable if \(-s - \frac{4}{3} + 4 \varepsilon < 0\), which is valid given \(s > -\frac{2}{3}\) and \(\varepsilon > 0\) small enough. Combining the above two estimates, we obtain the desired inequality.
Subcase 1.2 \(|\tau _2 + |n_2|^2| \ge 2^{-10} |n_2|^2\).
This subcase is similar to Subcase 1.1 by switching the roles of \(u_{N_1}\) and \(v_{N_2}\), and so we omit details.
Subcase 1.3 \(|\tau _1 + |n_1|^2| < 2^{-10} |n_1|^2\) and \(|\tau _2 + |n_2|^2| < 2^{-10} |n_2|^2\).
In this subcase, we need to estimate both \(u_{N_1}\) and \(v_{N_2}\) using the \(X^{s, \frac{2}{3}}\)-norm. Using the fact that \(\varphi _T\) is supported on \([-1, 1]\) given \(0 < T \le \frac{1}{2}\), by the Plancherel theorem, Hölder’s inequality, Lemma 2.4, and Lemma 2.3, we obtain
where \(\varepsilon > 0\) is arbitrarily small. The above estimate is acceptable if \(-s - \frac{2}{3} + 8 \varepsilon < 0\), which is valid given \(s > -\frac{2}{3}\) and \(\varepsilon > 0\) small enough.
Regarding the \(\ell _n^2 L_\tau ^1\) norm of the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term, we first let \(\varepsilon _1, \varepsilon _2 > 0\) satisfying
Note that both \(\varepsilon _1\) and \(\varepsilon _2\) can be arbitrarily small. By Hölder’s inequality, Young’s convolution inequality, Hölder’s inequalities twice, and Lemma 2.3, we have
for some \(\theta > 0\). Since \(s + 1 > 0\) given \(s > - \frac{2}{3}\), the above estimate is acceptable if \(-s - 1 + 2 \varepsilon _1 < 0\), which is valid given \(s > - \frac{2}{3}\) and \(\varepsilon _1 > 0\) close enough to 0. Combining the above two estimates, we obtain the desired inequality.
Case 2 \(|\tau + |n|^2| < 2^{-10} |n_1|^2\), \(|\tau _1 + |n_1|^2| \ge 2^{-10} |n_1|^2\), and \(|\tau _2 + |n_2|^2| \ge 2^{-10} |n_2|^2\).
In this case, we need to estimate both \(u_{N_1}\) and \(v_{N_2}\) using the \(Y^{s, \frac{2}{3}}\)-norm. We consider the following two subcases.
Subcase 2.1 \(|\tau + |n|^2| < 2^{-10} |n|^2\).
In this subcase, we need to evaluate the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term using the \(\widehat{X}^{s, \frac{2}{3}}\)-norm. By Hölder’s inequality, Young’s convolution inequality, and Lemma 2.3, we have
where \(\varepsilon > 0\) is arbitrarily small. Since \(s > -\frac{2}{3}\), we have \(s + \frac{4}{3} > 0\). Thus, the above estimate is acceptable if \(- s - \frac{4}{3} + 4 \varepsilon < 0\), which is valid given \(s > -\frac{2}{3}\).
Subcase 2.2 \(2^{-10} |n|^2 \le |\tau + |n|^2| < 2^{-10} |n_1|^2\).
In this subcase, we need to evaluate the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term using the \(\widehat{Y}^{s, \frac{2}{3}}\)-norm. By Hölder’s inequality, Young’s convolution inequality, and Lemma 2.3, we have
where \(\varepsilon > 0\) is arbitrarily small. Note that the second inequality is valid since \(s + \frac{1}{3} + 2 \varepsilon < 0\) given \(s \le - \frac{1}{2}\) and \(\varepsilon > 0\) small enough. Since \(s + \frac{4}{3} + 2 \varepsilon > 0\) given \(s > -\frac{2}{3}\), the above estimate is acceptable if \(-s - \frac{4}{3} + 6 \varepsilon < 0\), which is valid given \(s > -\frac{2}{3}\).
Also, by the Cauchy-Schwarz inequality, Hölder’s inequality, Young’s convolution inequality, and Lemma 2.3, we get
where \(\varepsilon > 0\) is arbitrarily small. Since \(s + 1 > 0\) given \(s > -\frac{2}{3}\), the above estimate is acceptable if \(-s - \frac{5}{3} + 8 \varepsilon < 0\), which is valid given \(s > -\frac{2}{3}\) and \(\varepsilon > 0\) small enough. Combining the above two estimates, we obtain the desired inequality.
Case 3 \(|\tau + |n|^2| < 2^{-10} |n_1|^2\) and \(|\tau _1 + |n_1|^2| < 2^{-10} |n_1|^2\).
In this case, we need to estimate \(u_{N_1}\) using the \(X^{s, \frac{2}{3}}\)-norm. Note that we have
and so \(|\tau _2 + |n_2|^2|> |n_2|^2 > 2^{-10} |n_2|^2\). Thus, we need to estimate \(v_{N_2}\) using the \(Y^{s, \frac{2}{3}}\)-norm. We consider the following two subcases.
Subcase 3.1 \(|\tau + |n|^2| < 2^{-10} |n|^2\).
In this subcase, we need to evaluate the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term using the \(\widehat{X}^{s, \frac{2}{3}}\)-norm. By duality and the Cauchy-Schwarz inequality, we have
Let \(w_N\) be a space-time distribution that satisfy \(\widehat{w_N} (n, \tau ) = h (n, \tau ) / \langle \tau + |n|^2 \rangle ^{\frac{1}{3}}\). Then, using the fact that \(\varphi _T\) is supported on \([-1, 1]\) given \(0 < T \le \frac{1}{2}\), by the Plancherel theorem, Hölder’s inequality, Lemma 2.4, and Lemma 2.3, we have
where \(\varepsilon > 0\) is arbitrarily small. Thus, continuing with (3.6), we use Lemma 2.3 to obtain
Since \(s \le -\frac{1}{2}\), we have \(s + \frac{1}{3} + \varepsilon < 0\) for \(\varepsilon > 0\) small enough. Thus, the above estimate is acceptable if \(-2\,s - \frac{4}{3} + 5 \varepsilon \le 0\), which is valid given \(s > -\frac{2}{3}\) and \(\varepsilon > 0\) sufficiently small.
Subcase 3.2 \(2^{-10}|n|^2 \le |\tau + |n|^2| < 2^{-10} |n_1|^2\).
In this subcase, we need to evaluate the \(\mathcal {F}_{x, t} \big ( \varphi _T \overline{u_{N_1}} \cdot \varphi _T \overline{v_{N_2}} \big )\) term using the \(Y^{s, \frac{2}{3}}\)-norm. By duality and the Cauchy-Schwarz inequality, we have
Note that the first inequality is valid since \(s + \frac{1}{3} < 0\) given \(s \le - \frac{1}{2}\). Let \(w_N\) be a space-time distribution that satisfy \(\widehat{w_N} (n, \tau ) = h (n, \tau ) / \langle \tau + |n|^2 \rangle ^{\frac{1}{2}}\). Then, using the fact that \(\varphi _T\) is supported on \([-1, 1]\) given \(0 < T \le \frac{1}{2}\), by the Plancherel theorem, Hölder’s inequality, Lemma 2.4, and Lemma 2.3, we have
where \(\varepsilon > 0\) is arbitrarily small. Thus, continuing with (3.7), we use Lemma 2.3 to obtain
Since \(s \le -\frac{1}{2}\), we have \(s + \frac{1}{3} + \varepsilon < 0\) for \(\varepsilon > 0\) small enough. Thus, the above estimate is acceptable if \(-2\,s - \frac{4}{3} + 5 \varepsilon \le 0\), which is valid given \(s > -\frac{2}{3}\) and \(\varepsilon > 0\) sufficiently small.
Also, by the Cauchy-Schwarz inequality, we get
where \(\varepsilon > 0\) is arbitrarily small. The above term can be estimated similarly as above (along with \(\langle \tau + |n|^2 \rangle ^{\varepsilon } \le N_1^{2 \varepsilon }\)) for the \(\ell _n^2 L_\tau ^2\) term. Combining the above two estimates, we obtain the desired inequality.
Case 4 \(|\tau + |n|^2| < 2^{-10} |n_1|^2\) and \(|\tau _2 + |n_2|^2| < 2^{-10} |n_2|^2\).
This case follows similarly from Case 3 by switching the roles of \(u_{N_1}\) and \(v_{N_2}\). We thus omit details.
Thus, we have finished our proof. \(\square \)
Before moving on to the proof of our main bilinear estimate in Proposition 3.1, we first observe that by definition of the \(X^{s, b}\)-norm in (2.1) and the \(Y^{s, b}\)-norm in (2.3), we have the following decompositions:
Thus, it follows that we have the following decomposition regarding the \(Z^{s, b}\) norm:
Proof of Proposition 3.1
By (3.8), we have
For each nonzero summand on the right-hand side of (3.9), we know that N, \(N_1\), and \(N_2\) must satisfy one of the following:
-
1.
\(2^{-5} N \le N_1 \le 2^5 N\) and \(N_2 \le 2^6 N\),
-
2.
\(2^{-5} N \le N_2 \le 2^5 N\) and \(N_1 \le 2^6 N\),
-
3.
\(\frac{1}{2} N_1 \le N_2 \le 2N_1\) and \(N < 2^{-5} N_1\).
We now treat the above three cases separately.
Case 1 \(2^{-5} N \le N_1 \le 2^5 N\) and \(N_2 \le 2^6 N\).
In this case, by Lemma 3.2, the Cauchy-Schwarz inequality, and (3.8) twice, we have
where in the right-hand side of the first inequality we have \(\delta > 0\).
Case 2 \(2^{-5} N \le N_2 \le 2^5 N\) and \(N_1 \le 2^6 N\).
This case can be treated in the same way as Case 1, and so we omit details.
Case 3 \(\frac{1}{2} N_1 \le N_2 \le 2N_1\) and \(N < 2^{-5} N_1\).
In this case, by Lemma 3.3, the Cauchy-Schwarz inequality, and (3.8) twice, we have
where in the right-hand side of the first inequality we have \(\delta > 0\).
Combining the above three cases, we have thus finished our proof. \(\square \)
4 Local Well-Posedness of the Quadratic NLS
In this section, we present the proof of Theorem 1.1, local well-posedness of the quadratic NLS (1.1) in the low regularity setting. As mentioned in Sect. 1, we mainly focus our attention on local well-posedness of (1.1) on \(H^s (\mathbb {T}^2)\) for \(- \frac{2}{3} < s \le -\frac{1}{2}\), using the estimates of the \(Z^{s, b}\)-norm in Sects. 2 and 3.
By writing (1.1) in the Duhamel formulation, we have
Since we are only interested in local well-posedness, we can insert time cut-off functions. For \(0 < T \le \frac{1}{2}\), we let \(\eta : \mathbb {R}\rightarrow [0, 1]\) be a smooth function such that \(\eta \equiv 1\) on \([-1, 1]\) and \(\eta \equiv 0\) outside of \([-2, 2]\) and let \(\eta _{2T} (t):= \eta (t / 2T)\). We first replace the two \(\overline{u}\)’s on the right-hand side of (4.1) by \(\eta _{2T} \overline{u}\). Also, note that for any function F that is smooth in space and Schwartz in time, we have
where \(\psi : \mathbb {R}\rightarrow [0,1]\) is a smooth cut-off function such that \(\psi \equiv 1\) on \([-1, 1]\) and \(\psi \equiv 0\) outside of \([-2, 2]\). Let us define the following nonlinear terms.
We consider the following formulation of the quadratic NLS (1.1):
4.1 Relevant Estimates
In this subsection, we present some relevant estimates for proving our local well-posedness result. We first show the following homogeneous linear estimate.
Lemma 4.1
Let \(s \in \mathbb {R}\), \(b \in \mathbb {R}\), and \(0 < T \le 1\). Then, we have
Proof
By the definition of the \(Z_T^{s, b}\)-norm in (2.6), Lemma 2.7, and Lemma 2.1 with \(k = 0\), we have
as desired. \(\square \)
We now take \(b = \frac{2}{3}\) and show the following bilinear estimate.
Lemma 4.2
Let \(-\frac{2}{3} < s \le -\frac{1}{2}\) and \(0 < T \le \frac{1}{4}\). Then, we have
for some \(\theta > 0\), where \(\mathcal {N}_1\), \(\mathcal {N}_2\), and \(\mathcal {N}_3\) are as defined in (4.2).
Proof
The idea of the proof comes from [3]. As in the proof of Lemma 2.6, by working with the extensions of u and v outside \([-T, T]\), it suffices to show the following three estimates:
for some \(\theta > 0\).
To deal with the \(\mathcal {N}_1\) term, by Lemma 2.7, the Taylor expansion, Lemma 2.1, Lemma 2.5, and Proposition 3.1, we obtain
for some \(\theta > 0\).
For the \(\mathcal {N}_2\) term, using Lemma 2.7, Lemma 2.1 with \(k = 0\), the fact that \(1 - \psi \) is bounded by 1 and supported outside of \([-1, 1]\), Lemma 2.5, and Proposition 3.1, we have
for some \(\theta > 0\).
For the \(\mathcal {N}_3\) term, since \(1 - \psi \) is bounded by 1 and supported outside of \([-1, 1]\), by the monotonicity property (2.5) and Proposition 3.1, we have
for some \(\theta > 0\). Thus, we finish our proof. \(\square \)
4.2 Local Well-Posedness
We now use the formulation (4.3) and the estimates in Sect. 4.1 to prove our local well-posedness result. We let \(0 < T \le \frac{1}{4}\) and fix \(-\frac{2}{3} < s \le - \frac{1}{2}\).
For the setting of \(\mathbb {T}^2\), by (4.3), Lemma 4.1, and Lemma 4.2, we have
for some \(\theta > 0\). Similarly, we obtain the following difference estimate:
Thus, by choosing \(T = T(\Vert u_0 \Vert _{H^s (\mathbb {T}^2)}) > 0\) sufficiently small, we have that \(\Gamma _1\) is a contraction on the ball \(B_R \subset Z^{s, \frac{2}{3}}\) of radius \(R \sim \Vert u_0 \Vert _{H^s (\mathbb {T}^2)}\). This gives the existence part of Theorem 1.1 when \(\mathcal {M} = \mathbb {T}^2\) and the uniqueness in the ball \(B_R\). Also, the continuous dependence of solutions on the initial data follows easily from the formulation (4.3), Lemma 4.1, (4.4), and (4.5).
It remains to extend the uniqueness of solutions to (1.1) to the entire \(Z_T^{s, \frac{2}{3}}\)-space. We let u and v be two solutions of (1.1) in \(Z_T^{s, \frac{2}{3}}\). Note that u and v satisfy the formulation (4.3) for \(t \in [-T, T]\). For \(0 < T_0 \le T\), we use (4.5) to obtain
Thus, by choosing
sufficiently small, we can use Lemma 2.6 to obtain
so that \(u \equiv v\) on \([-T_0, T_0]\). Since \(T_0\) depends only on \(\Vert u \Vert _{Z_T^{s, \frac{2}{3}}}\) and \(\Vert v \Vert _{Z_T^{s, \frac{2}{3}}}\), we can iterate the above argument on \([-T, -T_0]\) and \([T_0, T]\). This shows that \(u \equiv v\) on \([-T, T]\) after a finite number of iterations, and so the uniqueness of (1.1) on the entire \(Z_T^{s, \frac{2}{3}}\)-space follows.
Availability of data and materials
Not applicable.
Notes
Here, \(\frac{1}{2} +\) means \(\frac{1}{2} + \varepsilon \) for some \(\varepsilon > 0\).
On \(\mathbb {T}^d\), this framework works only for local well-posedness for small initial data. See Remark 1.3 for a discussion of local well-posedness on periodic domains for large initial data.
References
Bejenaru, I., de Silva, D.: Low regularity solutions for a 2D quadratic nonlinear Schrödinger equation. Trans. Am. Math. Soc. 360(11), 5805–5830 (2008)
Bejenaru, I., Tao, T.: Sharp well-posedness and ill-posedness results for a quadratic non-linear Schrödinger equation. J. Funct. Anal. 233(1), 228–259 (2006)
Bourgain, J.: Fourier transform restriction phenomena for certain lattice subsets and applications to nonlinear evolution equations, I: Schrödinger equations. Geom. Funct. Anal. 3, 107–156 (1993)
Bourgain, J.: Nonlinear Schrödinger equations, Hyperbolic equations and frequency interactions (Park City, UT, 1995), pp. 3–157, IAS/Park City Math. Ser., 5, Amer. Math. Soc., Providence, RI (1999)
Colliander, J., Delort, J., Kenig, C., Staffilani, G.: Bilinear estimates and applications to 2D NLS. Trans. Am. Math. Soc. 353(8), 3307–3325 (2001)
Colliander, J., Keel, M., Staffilani, G., Takaoka, H., Tao, T.: Multilinear estimates for periodic KdV equations, and applications. J. Funct. Anal. 211(1), 173–218 (2004)
Fujiwara, K., Georgiev, V.: On global existence of \(L^2\) solutions for 1D periodic NLS with quadratic nonlinearity. J. Math. Phys. 62(9), 091504 (2021)
Germain, P., Masmoudi, N., Shatah, J.: Global solutions for 2D quadratic Schrödinger equations. J. Math. Pures Appl. 97(5), 505–543 (2012)
Grünrock, A.: Some local wellposedness results for nonlinear Schrödinger equations below \(L^2\). Preprint arXiv:math/0011157v2
Guo, Z.: Global well-posedness of Korteweg-de Vries equation in \(H^{-3/4}({\mathbb{R} })\). J. Math. Pures Appl. 91(6), 583–597 (2009)
Hayashi, N., Naumkin, P., Shimomura, A., Tonegawa, S.: Modified wave operators for nonlinear Schrödinger equations in one and two dimensions. Electron. J. Differ. Equ. 2004(62), 1–16 (2004)
Ikeda, M., Wakasugi, Y.: Small-data blow-up of \(L^2\)-solution for the nonlinear Schrödinger equation without gauge invariance. Differ. Integral Equ. 26(11–12), 1275–1285 (2013)
Iwabuchi, T., Ogawa, T.: Ill-posedness for the nonlinear Schrödinger equation with quadratic non-linearity in low dimensions. Trans. Am. Math. Soc. 367(4), 2613–2630 (2015)
Iwabuchi, T., Uriya, K.: Ill-posedness for the quadratic nonlinear Schrödinger equation with nonlinearity \(|u|^2\). Commun. Pure Appl. Anal. 14(4), 1395–1405 (2015)
Jaquette, J., Lessard, J., Takayasu, A.: Global dynamics in nonconservative nonlinear Schrödinger equations. Adv. Math. 398, 108234 (2022)
Kenig, C., Ponce, G., Vega, L.: Quadratic forms for the 1-D semilinear Schrödinger equation. Trans. Am. Math. Soc. 348(8), 3323–3353 (1996)
Kishimoto, N.: Local well-posedness for the Cauchy problem of the quadratic Schrödinger equation with nonlinearity \(\overline{u}^2\). Commun. Pure Appl. Anal. 7(5), 1123–1143 (2008)
Kishimoto, N.: Low-regularity local well-posedness for quadratic nonlinear Schrödinger equations, Kyoto University, Master’s Thesis (2008)
Kishimoto, N.: Low-regularity bilinear estimates for a quadratic nonlinear Schrödinger equation. J. Differ. Equ. 247(5), 1397–1439 (2009)
Kishimoto, N.: Well-posedness of the Cauchy problem for the Korteweg-de Vries equation at the critical regularity. Differ. Integral Equ. 22(5–6), 447–464 (2009)
Kishimoto, N.: A remark on norm inflation for nonlinear Schrödinger equations. Commun. Pure Appl. Anal. 18(3), 1375–1402 (2019)
Kishimoto, N., Tsugawa, K.: Local well-posedness for quadratic nonlinear Schrödinger equations and the “good’’ Boussinesq equation. Differ. Integral Equ. 23(5–6), 463–493 (2010)
Liu, R., Oh, T.: Sharp local well-posedness of the two-dimensional periodic nonlinear Schrödinger equation with a quadratic nonlinearity \(|u|^2\). (to appear in Math. Res. Lett)
Moriyama, K., Tonegawa, S., Tsutsumi, Y.: Wave operators for the nonlinear Schrödinger equation with a nonlinearity of low degree in one or two space dimensions. Commun. Contemp. Math. 5(6), 983–996 (2003)
Nakanishi, K., Takaoka, H., Tsutsumi, Y.: Counterexamples to bilinear estimates related with the KdV equation and the nonlinear Schrödinger equation. Methods Appl. Anal. 8(4), 569–578 (2001)
Oh, T.: A blowup result for the periodic NLS without gauge invariance. C. R. Math. Acad. Sci. Paris 350(7–8), 389–392 (2012)
Shimomura, A.: Nonexistence of asymptotically free solutions for quadratic nonlinear Schrödinger equations in two space dimensions. Differ. Integral Equ. 18(3), 325–335 (2005)
Shimomura, A., Tonegawa, S.: Long-range scattering for nonlinear Schrödinger equations in one and two space dimensions. Differ. Integral Equ. 17(1–2), 127–150 (2004)
Shimomura, A., Tsutsumi, Y.: Nonexistence of scattering states for some quadratic nonlinear Schrödinger equations in two space dimensions. Differ. Integral Equ. 19(9), 1047–1060 (2006)
Staffilani, G.: Quadratic forms for a 2-D semilinear Schrödinger equation. Duke Math. J. 86(1), 79–107 (1997)
Tao, T.: Nonlinear dispersive equations. Local and global analysis, CBMS Regional Conference Series in Mathematics, 106. Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI. pp. xvi+373 (2006)
Zygmund, A.: On Fourier coefficients and transforms of functions of two variables. Studia Math. 50, 189–201 (1974)
Acknowledgements
The author would like to thank his advisor, Tadahiro Oh, for suggesting this problem and for his support throughout the entire work. The author is also grateful to the anonymous reviewers for the helpful comments. R.L. was supported by the European Research Council (grant no. 864138 “SingStochDispDyn”).
Funding
The author acknowledges funding from the European Research Council (Grant no. 864138 “SingStochDispDyn”).
Author information
Authors and Affiliations
Contributions
Not applicable.
Corresponding author
Ethics declarations
Conflict of interest
The author has no conflict of interest to declare.
Ethical approval
Not applicable.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Liu, R. Local Well-Posedness of the Periodic Nonlinear Schrödinger Equation with a Quadratic Nonlinearity \(\overline{u}^2\) in Negative Sobolev Spaces. J Dyn Diff Equat (2023). https://doi.org/10.1007/s10884-023-10295-x
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10884-023-10295-x