WorksMathematics
Uniform local smoothing for the derivative NLS and the soliton mass threshold
Abstract
Harrop-Griffiths, Killip, Ntekoume and Vişan proved that the derivative nonlinear Schrödinger equation (DNLS) is globally well-posed in . A key ingredient of their proof is a local smoothing estimate. It holds on -equicontinuous sets of initial data, but it cannot hold on -balls, because rescaled stationary solitons, all of mass , violate it. This raises a natural question: for which masses does local smoothing hold uniformly on the ball ? We prove three results. First, we give a quantitative version of the local smoothing estimate in which the data enter only through the mass and an explicit frequency scale . On time intervals of length the bound grows like , and the stationary soliton shows that the power of cannot be lowered below one. Second, we prove uniform local smoothing on balls of small mass, together with a scale-invariant Morrey-type form and a localized bound. Third, we analyse the soliton family from both sides. The uniform constant must blow up at least like as , while every soliton of mass satisfies local smoothing with a constant . These facts single out the stationary-soliton mass , which is also the Hayashi–Ozawa coercivity threshold, as the natural candidate for the sharp threshold. We formulate this as a conjecture, give a spectral reformulation in terms of the Kaup–Newell scattering data, and discuss the obstacles to a proof.
Introduction and main results
We consider the derivative nonlinear Schrödinger equation on the line,
It was derived as a model for circularly polarized Alfvén waves propagating along a magnetic field in a plasma [53, 54], and it is completely integrable through the Kaup–Newell Lax pair [34]. The equation conserves the mass and is invariant under the scaling
which leaves unchanged. Hence is the scaling-critical space.
In a recent landmark paper, Harrop-Griffiths, Killip, Ntekoume and Višan [20] proved that (1) is globally well-posed in . More precisely, the data-to-solution map, defined on Schwartz data, extends to a jointly continuous map
This result is the end point of a long line of work, which we recall in § 1.2. Its proof relies on the second-generation method of commuting flows. The analytic heart of the method consists of two families of estimates. The first are local smoothing estimates, obtained from microscopic conservation laws for the logarithm of the Kaup–Newell perturbation determinant. The second express the tightness of orbits.
Let us recall the local smoothing statement of [20]. We use the localizing weights
By [20 Theorem 1.5], if is bounded and equicontinuous in , then for every
The authors stress that the estimate is claimed for equicontinuous sets and not for balls, and that this restriction is necessary. Indeed, [20 Proposition 1.11] exhibits solutions , obtained by rescaling the stationary soliton, such that while a localized local smoothing norm of on tends to infinity. In (5), the implicit constant comes from a rescaling that makes the relevant set of Schwartz functions “-good” in the sense of [20 Definition 1.8]. As the authors note in the proof of [20 Theorem 1.5], this constant depends on only through the rescaling parameter, but that dependence is not made explicit.
The counterexample of [20] lives exactly on the sphere , and is a distinguished number for DNLS. It is the mass of the stationary soliton [20 (1.25), (1.28)]. It is also the threshold below which the energy is coercive, a fact used by Hayashi and Ozawa [24, 26] to obtain global solutions. Moreover, Bahouri and Perelman [2] point out that and are the masses of the static bright soliton and of the algebraic soliton, whose profiles are extremals of the relevant Gagliardo–Nirenberg inequalities. This suggests the following problem.
Problem. Determine the largest such that local smoothing holds uniformly on every ball with , with a constant depending only on . Describe the size of the optimal constant, and its dependence on the data when equicontinuity replaces smallness.
To make this precise, we introduce, for a solution of (1) and , the local smoothing functional
where as in [20]. We also introduce the uniform local smoothing constant
For the integrand in is a lower semicontinuous, hence measurable, function of , because is continuous in and the norm is lower semicontinuous with respect to convergence (see the proof of Corollary 1.3). The function is nondecreasing, and we set . By [20 Proposition 1.11], or by Theorem 1.4(i) below, .
Main results
Throughout the paper we fix (for definiteness ; this parameter is denoted by in [20]). We let be the smallness parameter for which [20 Proposition 4.1] holds; see Proposition 2.2 below. For we use the equicontinuity quantity of [20],
which is the norm of [20 (2.2)] with ; for its square is the left-hand side of [20 (1.19)].
Definition 1.1 (Frequency scale of an orbit). For with solution we set
Lemma 2.1 below shows that for every Schwartz datum, that the infimum is attained, and that whenever . It also shows that, apart from the normalization , the frequency scale transforms linearly under the scaling (2). Our first result makes (5) quantitative.
Theorem 1.2 (Quantitative local smoothing). There is a constant , depending only on the fixed parameter , such that for every and every the solution satisfies
Conversely, the power of in (9) cannot be lowered below one. More precisely, there are constants , and , and Schwartz data , , such that , , and
Theorem 1.2 recovers [20 Theorem 1.5]. Indeed, if is bounded and equicontinuous, then [20 Corollary 1.9] gives ; see Remark 2.3. The new information is twofold. All dependence on the data beyond the mass is carried by the single number , and this dependence is polynomial. When the mass is small, and we obtain uniform estimates.
Corollary 1.3 (Uniform local smoothing for small mass). Let . For every with and every , the solution of [20] satisfies
In particular and .
The estimate (12) is scale invariant. It is the nonlinear counterpart, on parabolic cylinders, of the Kato smoothing estimate for the free Schrödinger group. The point of Corollary 1.3 is not its difficulty, since it follows from the proof of [20 Proposition 4.1].
Its point is that it locates the obstruction of [20 Proposition 1.11] at positive mass. Our second result analyses this obstruction.
To state it, recall from [20 Section 1.3] that for each DNLS has a soliton of mass travelling with velocity . The soliton with is stationary. We denote by the family of all solutions obtained from it by translations in space and time, phase rotations and the scaling (1.2); see (2.12).
Theorem 1.4 (The stationary-soliton threshold). There are absolute constants , and such that the following hold.
(i) for every .
(ii) for every . In particular as .
(iii) Every soliton with satisfies
Consequently . Moreover, whatever the value of , the constant is at least of order near , and the soliton family alone produces exactly this rate.
Part (i) is a reformulation of [20 Proposition 1.11]. We include a proof because the same lower bound, applied to slowly moving solitons, gives (ii). Part (iii) shows that solitons alone cannot push below . Their contribution to local smoothing is controlled by the time they need to cross a window, which is of order , against a local smoothing density of order ; see Figure 2. These considerations lead to the following conjecture, which is the main open problem of this paper.
Figure 2. Trajectories , , of rescaled solitons in the space-time window (schematic). Each soliton carries a local smoothing density of order , so its contribution to is of order . This gives Proposition 4.1. The slow solitons with and attain this order, which gives Theorem 1.4(ii).
Conjecture 1.5 (Mass threshold for uniform local smoothing). . More precisely, for every , and
In § 5 we explain why we believe this conjecture, using the trace formula for the Kaup–Newell scattering data. For Schwartz data of mass , every eigenvalue of the Kaup–Newell problem satisfies . Consequently, no component of the discrete spectrum can be stationary; see Remark 5.1. We also formulate a sharp form of Theorem 1.2 (Conjecture 5.3), in which is replaced by , and a global-in-time question for small mass (Question 5.4).
Figure 1 summarizes what is proved and what is conjectured about , and Table 1 places our results among the known local smoothing statements.

Figure 1. The uniform local smoothing constant along the mass axis (not to scale). Green: proved finite. Orange: proved lower bound, finiteness conjectured. Red: proved infinite. The soliton bound of Theorem 1.4(iii) matches the lower bound of Theorem 1.4(ii).
| Setting | Statement | Constant | Source |
| Free Schrödinger group | exact | [33, 36] | |
| DNLS, bounded and equicontinuous | implicit in | [20 Thm. 1.5] | |
| DNLS, rescaled stationary solitons, | local smoothing fails on balls | no bound | [20 Prop. 1.11] |
| DNLS, | ; power necessary | explicit in | Thm. 1.2 (new) |
| DNLS, | uniform bound, Morrey form, localized | uniform | Cor. 1.3 (implicit in [20]) |
| DNLS, | blows up | Thm. 1.4(i),(ii) (new) | |
| Solitons of mass | sharp rate | Thm. 1.4(iii) (new) | |
| DNLS, | , | conjectured | Conj. 1.5, 5.2 |
Table 1. Local smoothing statements on and the position of the present results.
Related literature
Local and global well-posedness. Local well-posedness of (1.1) in was proved for by Tsutsumi and Fukuda [65], and in by Hayashi and Ozawa [23, 24, 26] by means of a gauge transformation; see also Ozawa [57]. Takaoka [61] reached , , by combining the gauge transformation with the Fourier restriction method. Below the data-to-solution map fails to be [61, 62], and it is not even uniformly continuous on bounded sets [7]. Hence contraction arguments in cannot go further. Local results in Fourier–Lebesgue and modulation spaces, closer to the critical scaling, are due to Grünrock [16] and to Guo, Ren and Wang [18].
Global results first required a mass restriction. Using the gauged energy and the sharp Gagliardo–Nirenberg inequality, Hayashi and Ozawa [26] and Ozawa [57] obtained global solutions for . Colliander, Keel, Staffilani, Takaoka and Tao [9, 10] carried the condition to , , by the -method, and Miao, Wu and Xu [52] reached . Wu exploited the momentum to go beyond , first to [69] and then to [70] in . Guo and Wu [19] obtained the same threshold in ; see also Fukaya, Hayashi and Inui [14]. The value is the mass of the algebraic soliton [20 (1.26)], [2], for which all other polynomial conservation laws vanish; see the discussion following [20 (1.26)].
The inverse scattering approach, initiated by Lee [48], removed the mass restriction for data in weighted spaces [49, 30, 32, 58, 59]. It also produced long-time asymptotics for soliton-free data [50] and soliton resolution for generic weighted data [29]; see the survey [31]. Bahouri and Perelman [2] then proved global well-posedness in without restriction, with uniform-in-time bounds, by combining the integrable structure with profile decompositions. Bahouri, Leslie and Perelman [1] established uniform-in-time bounds for .
At the level of low-regularity conservation laws, Klaus and Schippa [45] obtained a priori estimates in Besov spaces of positive regularity for small mass, following the approach of Killip, Vișan and Zhang [43]. Tang and Xu [63] derived the corresponding microscopic conservation laws for small mass and pointed out that their coercive currents should yield local smoothing in . They did not state an -level estimate. Killip, Ntekoume and Vișan [41] proved equicontinuity of orbits for and global well-posedness in under the same restriction. Harrop-Griffiths, Killip and Vișan [21] then removed the mass restriction from the equicontinuity statement. The theorem [20] builds on all of these works, on the method of commuting flows of Killip and Vișan [42], and on its second-generation form developed for the cubic NLS and mKdV [22]. Table 2 summarizes this development.
| Space | Mass condition | Main tool | References |
| , (local) | none | energy method | [65] |
| (local) | none | gauge transformation | [23, 24, 26] |
| (global) | gauged energy, sharp Gagliardo–Nirenberg | [26, 57] | |
| , (local) | none | gauge, Fourier restriction | [61] |
| , (global) | -method | [9, 10] | |
| (global) | -method, resonant decomposition | [52] | |
| (global) | , then | momentum, variational arguments | [69, 70] |
| (global) | -method, momentum | [19] | |
| weighted spaces (global) | none | inverse scattering | [49, 30, 32, 59, 58] |
| (global) | none | integrability, profile decompositions | [2] |
| (global) | conservation laws, equicontinuity | [41] | |
| (global) | none | commuting flows, local smoothing | [20] |
Table 2. Well-posedness of DNLS on the line (selection).
The periodic problem. On the circle, local well-posedness in is due to Herr [27]. Grünrock and Herr [17] treated Fourier–Lebesgue data, and Deng, Nahmod and Yue [12] obtained the optimal Fourier–Lebesgue local theory. Global results under the restriction were obtained by Mosincat and Oh [56] in and by Mosincat [55] in . Bahouri and Perelman [3] first relaxed the restriction to , and very recently they proved global well-posedness in with no mass restriction [4]. Well-posedness in remains open. Local smoothing, which is the tool studied here, is not available on the circle.
Solitons. The two-parameter soliton family and its orbital stability were studied by Colin and Ohta [8]. Kwon and Wu [46] treated the algebraic soliton, and Le Coz and Wu [47] multi-solitons. The coincidence of and with the masses of the static and algebraic solitons is noted in [2].
A parallel model. The continuum Calogero–Moser model, also called the Calogero–Moser derivative NLS, is another integrable, mass-critical derivative Schrödinger equation. There the stationary soliton also has mass . Gérard and Lenzmann [15] proved global well-posedness of smooth solutions with mass at most . Killip, Laurens and Vişan [40] proved well-posedness in for focusing data of mass below . Above , Hogan and Kowalski [28] constructed solutions with unbounded Sobolev norms and mass arbitrarily close to , and Kim, Kim and Kwon [44] constructed smooth finite-time blow-up solutions with mass arbitrarily close to that of the soliton. Small-mass dispersive decay is proved in [39]. In that model, therefore, is a well-posedness threshold. For DNLS, well-posedness in holds for every mass [20], and Conjecture 1.5 proposes that is instead the threshold for uniform smoothing. A threshold given by a ground-state mass is also familiar from the focusing mass-critical NLS [68, 13] and from the scattering theory below the ground state developed by Kenig and Merle [35]. For recent one-dimensional and inhomogeneous instances of this philosophy, we refer to [6, 51].
Local smoothing. For the free Schrödinger equation, Kato-type smoothing goes back to Kato [33], Constantin and Saut [11], Sjölin [60] and Vega [66]. In one dimension, Kenig, Ponce and Vega [36] proved the exact identity
The same authors used local smoothing to treat Schrödinger equations with derivative nonlinearities [37, 38]. The finite-scale behaviour of the smoothing quantity was studied by Vega and Visciglia [67]. An exact identity and sharp constants at finite scale were obtained by Bai and the author [5]. The monotonicity approach to local smoothing used in [20] is discussed in [64].
To the best of our knowledge, the literature contains no uniform local smoothing statement for DNLS on -balls of positive mass beyond what is implicit in [20 Proposition 4.1]. Likewise, the only negative statement we are aware of is [20 Proposition 1.11], which shows failure at mass ; the remarks in [63] concern small mass in , . The question whether is sharp, and how the constant depends on the data, therefore appears to be open. We do not claim novelty for Corollary 1.3, which is implicit in the proof of [20 Proposition 4.1]. Our contributions are the explicit dependence in Theorem 1.2, the two-sided soliton analysis of Theorem 1.4, which pins down the expected rate , and the formulation of Conjectures 1.5 and 5.3 together with a spectral reformulation.
The paper is organized as follows. § 2 collects notation, the results of [20] that we use, and two elementary tools, namely a window-summation lemma for and bounds on the soliton profiles. § 3 proves Theorem 1.2 and Corollary 1.3, and § 4 proves Theorem 1.4. § 5 discusses the method, its limitations and the evidence for the conjectures. § 6 contains concluding remarks and further directions.
Preliminaries and technical tools
Notation
We use the unitary Fourier transform , so that . Following [20], we write
Since , we have
We shall use the Gagliardo representation
It follows from Plancherel’s theorem, applied for fixed , and from the identity
We also use the elementary interpolation bound
which follows from the Cauchy–Schwarz inequality in and from .
The weight from (4) satisfies and
The first inequality follows from . The second follows from and . The value of is recorded in [20 (2.1)]. Moreover, for .
Results from [20] used below
We use [20] as a black box through three statements. The first is the global well-posedness theorem [20 Theorem 1.1]. It provides (3) and the group property . Moreover, Schwartz data lead to global Schwartz solutions, which conserve the mass and the perturbation determinant .
The second is the equicontinuity theorem of [21], quoted as [20 Theorem 1.7]. It states that if is bounded and equicontinuous in , then so is the set
of [20 (1.11)]. We combine it with [20 Lemma 2.1], which states that as for every bounded equicontinuous set . (This last fact is elementary: splitting the integral (8) at a frequency , we get , and equicontinuity makes the last term uniformly small.) For we also set
Lemma 2.1. Let and .
(i) , and the infimum in Definition 1.1 is attained. More precisely, for every .
(ii) If , then .
(iii) For every and one has , where .
Consequently, and .
Proof. For set . The weight in (8) is at most one and decreases in . Hence, by dominated convergence, each map is continuous and nonincreasing. We infer that is nonincreasing and lower semicontinuous, being a supremum of continuous functions. Therefore the set is closed in and is a half-line: either , in which case , or .
(i) The DNLS flow conserves . Hence the whole orbit is contained in with . A single Schwartz function forms a bounded equicontinuous set, so is bounded and equicontinuous by [20 Theorem 1.7]. By [20 Lemma 2.1], as , so is nonempty and . Since is a closed half-line, belongs to , and . (ii) If , the weight in (8) being at most one, we get for all by mass conservation. Thus and .
(iii) We have , and the substitution gives
The identity follows from the half-line structure of . Finally, by the scaling invariance (2) and uniqueness of Schwartz solutions, . Taking the supremum in and using the identity just proved, we obtain , which yields . □
The third statement is the local smoothing estimate [20 Proposition 4.1, (4.1)]. It reads as follows. Let be bounded and equicontinuous in , and assume that the set of DNLS orbits
is -good for a sufficiently small . (In [20 Proposition 4.1] this set is written , where is the DNLS Hamiltonian, and it is also denoted by ; it should not be confused with the isospectral set (19) above.) Then uniformly for . Here -good means bounded, equicontinuous and subject to the smallness condition [20 (1.19)], namely ; see [20 Definition 1.8]. The local smoothing norm is defined in [20 (2.24), (2.25)] over the time interval , and by [20 (2.32)] with it satisfies
with absolute implicit constants, because and . Finally, a compact subset of is automatically bounded and equicontinuous, and the orbit of a Schwartz datum is compact. Applying [20 Proposition 4.1] with , we obtain the result in the following form.
Proposition 2.2 ([20 Proposition 4.1]). There are constants and , depending only on , such that the following holds. Let and , and suppose that for all . Then
Decreasing only strengthens the hypothesis, so the normalization costs nothing. The time translation from to uses only the group property and mass conservation: one applies the estimate to the Schwartz datum .
Remark 2.3. The essential point in Proposition 2.2 is that and depend on nothing but . We have checked this against the proof of [20 Proposition 4.1] on pp. 895–896 of the published version. Combining the identity [20 (4.4)] with [20 Lemmas 4.2 and 4.9], [20 (2.4)] and [20 (2.13)], the authors arrive at [20 (4.33)], which for reads
If with small compared with the implicit constant, the first term is absorbed into the left-hand side; this is legitimate since for Schwartz solutions. Using and mass conservation, we deduce that .
In the proof of (4.1), the set enters only through the smallness [20 (1.19)] along the trajectory and through [20 Lemmas 4.2 and 4.9], whose estimates are stated uniformly for in any set satisfying the hypotheses of [20 Proposition 4.1]. Equicontinuity is used quantitatively in that section only through [20 (2.3)], and only for the refined estimate [20, (4.2)}, which we do not need. This is consistent with the proof of [20 Theorem 1.5] on p. 921, where the authors note that the resulting constant depends on solely through the rescaling parameter. The paper [20] does not fix a normalization of the Fourier transform. A different normalization multiplies every norm in (8), in [20 (1.19)] and in by the same absolute factor, so only the values of and change.
For a bounded equicontinuous set , [20 Corollary 1.9] provides such that the rescaled isospectral set is -good. Since contains every orbit issued from , Lemma 2.1(iii) gives for all , which is the bound used after Theorem 1.2. In this sense Theorem 1.2 is a quantitative form of the proof of [20 Theorem 1.5], in which a uniform rescaling makes all orbits -good and the estimate is then summed over unit time intervals.
Summing small windows
The following lemma converts local smoothing on windows of size into local smoothing on unit windows. For and we set
Lemma 2.4 (Window summation). There is an absolute constant such that for every , every and every ,
Proof. By translation invariance we may take . We first treat the term. Fix and let . Then , and (18) gives . We average the identity over these , insert the two bounds, and integrate in . We obtain
which controls the second term on the left of (22).
For the homogeneous term, we split the double integral in (16) into the far region and the near region . In the far region we use and . It follows that the far region contributes at most , which is controlled by (23).
In the near region we again average over the with . For such , both and lie in , where . On we write with and . From (18) and we get on . Moreover, and , so that
Writing and using the mean value theorem on the interval , we deduce that
We square this inequality, divide by and integrate over the near region, using for the second term. We conclude that the near region contributes at most
to the double integral. Dividing by and collecting the three contributions gives (22). □
The soliton family
We now record what we use about the solitons. Following [20 Section 1.3], for we put
Note that , so that (24) is exactly the formula of [20]. (We have also checked numerically that (25) below solves (1) with this sign of the phase, and not with the opposite one, when . Only enters our estimates.) By [20 (1.24)],
is a Schwartz solution of (1) of mass [20 (1.25)]. For it reduces to the stationary soliton [20 (1.28)]. We let be the set of all solutions
Every has mass , and by (25)
with centre . The second identity holds because the phase factor in (25) depends on only.
Lemma 2.5 (Profile bounds). Let and . Then for all
Moreover, is continuous from into .
Proof. Since and , we have
As , this gives the identity in (28). For we have . Hence , and gives the first bound.
Next, . Because and , we have , and therefore . Furthermore, is bounded by in absolute value. We infer that , and the product rule applied to (24) gives the second bound in (28). For the continuity statement, note that does not vanish on . Hence and are continuous in on this set. By (28) (with ), they are dominated by uniformly in . Dominated convergence yields continuity in . □
Quantitative local smoothing
Local smoothing at the natural scale
The first step is to rescale Proposition 2.2 to the scale on which the orbit is -good. In the original variables, this gives local smoothing on windows of size over times of length .
Proposition 3.1 (Scale-adapted local smoothing). Let and , . Then for every ,
where and .
Proof. Let . This is the rescaling (2) with . Hence is the Schwartz solution with data , and . By Lemma 2.1(iii) with , and then by Lemma 2.1(i), we have for all . Proposition 2.2 applied at time , together with (15), gives
Now with . A change of variables in the Fourier integral shows that . Substituting , we obtain
which is (29). □
Proof of Theorem 1.2
Let , write and , and fix . Consider a time interval . We apply Lemma 2.4 with to for each , and we integrate over . On the right-hand side, Proposition 3.1 bounds the time integral of by for every . Since , mass conservation gives . Hence, by Fubini’s theorem and (18),
Moreover, . Combining these two bounds with (15), we get with . The interval is covered by at most intervals of the form . Summing their contributions and taking the supremum over , we obtain (9) with .
For the lower bound, let be the stationary soliton, that is (25) with , and set . By (25), , so that by Lemma 2.1(iii). Set . It is finite by Lemma 2.1(i). It is positive, because as , by monotone convergence. Lemma 2.1(iii) then gives for , where . In particular grows linearly in . On the other hand, with , and its centre does not move. Proposition 4.2 below therefore gives for all and all . Integrating over , we find
Since , this is (10) with . Finally, if (9) held with replaced by for some , then for all , which is impossible as . □
Remark 3.2. The exponent 2 in (9) comes from parabolic scaling. Proposition 3.1 controls windows of size only during times of order . Moreover, the smallness that drives the bootstrap of [20] is available only at frequencies above , so one cannot simply enlarge the windows. The stationary soliton shows that the true growth is at least linear in ; see Conjecture 5.3.
Proof of Corollary 1.3
Let first with . By Lemma 2.1(ii), , and (11) is (9).
For (12), let and let be the solution with data , which has the same mass. Applying (11) with to and using (15), we obtain
for every . Now with , and . Hence the substitution turns the last bound into
Since ranges over , this is (12).
For (13), we use the one-dimensional inequality , which follows from Sobolev embedding and interpolation. We apply it to . As , integration in time and (11) give (13) with .
Now let with . Choose Schwartz functions in , with , and set . Then , , and in . By the joint continuity of [20 Theorem 1.1] and the compactness of , we get in uniformly for . For fixed and , it follows that in .
The norms and (with value allowed) are lower semicontinuous with respect to convergence. Indeed, if in , choose a subsequence along which tends to , and a further subsequence along which almost everywhere. Fatou’s lemma in then gives , and the same argument applies to . Fatou’s lemma in therefore transfers (11) and (12) from to . Finally, in . Hence a subsequence converges almost everywhere, and Fatou’s lemma also transfers (13). The bound is (11) with . □
Solitons and the threshold
The two propositions of this section quantify a simple geometric mechanism, illustrated in Figure 2. A soliton of scale carries a local smoothing density of order , and it contributes only while it stays in the window. A soliton of mass moves with speed , so it crosses a unit window in a time of order . Its total contribution is therefore of order , uniformly in . By contrast, the stationary soliton never leaves the window.
An upper bound for moving solitons.
Proposition 4.1. Let , and . Then
Proof. Write , fix , and let and be as in (27). Define
Since , we have . By (27) and (28),
Using and the second bound in (28), we also get
By (17), it follows that .
We now integrate in time. The centre moves with speed . Denote by the time at which , so that . Fubini’s theorem gives
Combining this with , we obtain
where we used the bound for the first summand and for the second. Taking the supremum over gives (4.1).
The bound (4.1) is uniform in the scaling parameter and in the translation and phase parameters. It degenerates only as , that is, as . This is the regime of slowly moving solitons with mass close to . For small mass the bound is uniform as well, because small solitons travel fast.
A lower bound for slow solitons.
Proposition 4.2. There are absolute constants and such that the following holds. Let and , and let have parameters in (26). Then for every with ,
Proof. By (26) and (25), for a unimodular constant , where . By (15), translation invariance and the scaling of ,
Set . This minimum is attained, by Lemma 2.5 and the continuity of the norm on . It is positive, because a function with vanishing norm vanishes, while . Since , we have , and hence .
It remains to estimate the difference . By (18), . By Lemma 2.5 with and , we also have . Moreover, . We deduce that
with an absolute constant . We choose . Then , and (32) holds with .
Proof of Theorem 1.4.
(i) Let . For , let be the rescaled stationary soliton with . It has mass and centre . By Proposition 4.2, . Since , we get for every , and therefore .
(ii) Let be as in Proposition 4.2 and set . Given , let , so that . Take with and . Then for , and Proposition 4.2 gives . Since and on by convexity, we have . As , we conclude that
(iii) For with we have and . Hence , and Proposition 4.1 gives (14) with .
The final assertions follow from Corollary 1.3 and parts (i)–(iii), with . □
Remark 4.3. Proposition 4.1 also covers solitons of very small mass. As , the amplitude of is of order , while its frequency is of order . These solitons are thus high-frequency wave packets of small mass, travelling at a speed comparable to their frequency. They are invisible to the threshold question. They do show, however, that small-mass solutions need not scatter. This is why the global-in-time question below is phrased in terms of smoothing and not of scattering; compare with the weighted-space asymptotics of [25, 50].
Discussion: methods, obstacles and conjectures
What the method gives and where it stops
The proof of Theorem 1.2 uses [20] in an essentially optimal way. Smallness of the norm at frequency scale is the only source of smallness in their bootstrap, and rescaling converts it into local smoothing at the natural parabolic scale. All further losses come from geometry, namely from summing small windows into a unit window and short time intervals into . Lemma 2.4 shows that the first step costs nothing. The second step is where the factor appears.
For data of mass below the frequency scale is irrelevant, and Corollary 1.3 follows. Between and the method gives no uniform information. The reason is that the norm of a datum of moderate mass need not be small, and making it small requires a rescaling that depends on the frequency distribution of the datum. This is not a technical defect. By Theorem 1.4(i), the same mechanism must fail at mass . Any proof of Conjecture 1.5 must therefore use a property of the region that the stationary soliton violates.
Heuristic evidence for Conjecture 1.5
A unit window can accumulate a large amount of local smoothing only if a solution concentrated at a high frequency remains near the window for a time much longer than . Linear wave packets at frequency move with speed of order and do not do this, but coherent structures could. Proposition 4.1 shows that single solitons of mass cannot. Indeed, their velocity is proportional to their frequency, with a coefficient bounded below in terms of . Theorem 1.4(ii) shows that this coefficient is the only source of growth among solitons. Figure 3 illustrates the resulting two-sided picture near .

Figure 3. Schematic behaviour near (constants not to scale). Both proved bounds blow up at the same rate . Conjecture 1.5 asserts that no other mechanism produces a faster rate below .
The scattering theory of the Kaup–Newell problem gives a more structural version of this observation. For , let be the transmission coefficient in the variable , where is the Kaup–Newell spectral parameter, and let be its zeros in the upper half-plane, with . Bahouri and Perelman record the trace formula [4 Lemma 2.3(ii)]; see also [2 (2.15)]:
where is a nonnegative measure accounting for the radiation.
Remark 5.1. [Spectral picture below ] By (33) and the nonnegativity of , every eigenvalue of a Schwartz function with satisfies . For a single soliton the picture is explicit. In the normalization of [2], the bright soliton with parameters , , travels with velocity . Its transmission coefficient has the single zero , and its mass is given in [2 (1.12)]. A direct computation shows that , hence . Consequently, there is no radiation in (33), the velocity is proportional to , and the soliton is stationary if and only if its eigenvalue lies on the imaginary axis, that is, if and only if its mass equals . In the parametrization (24) this corresponds to . The equation of [2] is related to (1) by the reflection , which reverses the direction of motion but does not affect these statements. For several solitons, breathers and general data, the analogous statement, namely that each coherent component attached to moves with a velocity proportional to , is suggested by the Lax-pair time dependence and by soliton resolution [29].
Under soliton resolution [29], a solution decomposes for large times into solitons, attached to the eigenvalues, and dispersive radiation. If all eigenvalues lie in the sector , every soliton moves with a speed bounded below by a multiple of its frequency, and radiation disperses. This motivates the following spectral strengthening of Conjecture 1.5.
Conjecture 5.2 (Spectral form). For every and there is with the following property. Every with , whose transmission coefficient has no zeros in the sector , satisfies . Moreover, one can take .
By Remark 5.1, Schwartz data of mass satisfy the hypothesis with . Therefore Conjecture 5.2, combined with the density argument in the proof of Corollary 1.3, implies Conjecture 1.5, including the rate , which Theorem 1.4 shows to be optimal. The spectral formulation also explains why breathers or multi-soliton bound states should not produce stationary structures below . Indeed, a stationary bound state would need all its eigenvalues on the imaginary axis, and each of them contributes to (33). This last assertion relies on the heuristic velocity–eigenvalue relation of Remark 5.1 for multi-soliton states.
The sharp dependence on the frequency scale
The lower bound (10) and the soliton heuristics suggest that the factor in (9) is an artefact of the method.
Conjecture 5.3 (Sharp quantitative local smoothing). There is a nondecreasing function such that for all and ,
The stationary soliton shows that the exponent one cannot be improved. A proof would require local smoothing on windows of size much larger than during times of order one. This seems to need a substitute for the parabolic rescaling that preserves the smallness used in [20].
Global-in-time smoothing for small mass
The estimate (12) holds on parabolic cylinders of every size, which suggests asking whether smoothing for small mass holds globally in time. Some care is needed in the formulation. For the free flow with low-frequency data, decays like . Hence can diverge logarithmically, because is comparable to for such slowly varying functions, which reflects the commutator between and . The natural global quantity is the Kato–Kenig–Ponce–Vega one.
Question 5.4. Is there such that every with satisfies
For the free Schrödinger group the answer is yes, by the identity of [36]. Small-mass solitons, which by Remark 4.3 do not scatter, appear to satisfy this bound, since their frequency and speed are both of order . A positive answer would be a nonlinear Kato smoothing theorem at the critical regularity, in a setting without scattering. A natural tool is the family of microscopic conservation laws of [63, 20] with the spectral parameter coupled to time, in the spirit of an interaction Morawetz estimate. We have not been able to close such an argument. The finite-scale analysis of [5] may help to understand the linear part of the problem.
Strichartz-type quantities
By [20 Proposition 1.11], the rescaled stationary solitons also satisfy . (This is also immediate from scaling, since for the rescaled stationary soliton .) Our methods yield the localized bound (13) for small mass, but not an bound. The latter would require control of in rather than in , since . We do not know whether holds even for small mass. Its threshold, if it exists, is presumably also .
Possible routes to Conjecture 1.5 and their obstacles
A natural approach is by contradiction and concentration compactness, in the spirit of Kenig and Merle [35]. Suppose that for some , and take a minimal such . One would extract a sequence with . One would then rescale it so that the bad behaviour occurs at unit frequency on time intervals of length , and try to pass to a limit which is almost periodic modulo symmetries and has mass at most . Three obstacles stand in the way. First, a profile decomposition adapted to the DNLS flow in is not available. Bahouri and Perelman [2] use profile decompositions at the and levels, and the absence of uniform continuity of the flow on bounded sets of , [7], prevents the usual nonlinear profile approximation. Second, minimality of and the Lipschitz stability theory that usually comes with it have no counterpart here. Third, the rigidity step would have to exclude non-dispersing almost periodic solutions of mass below . Remark 5.1 suggests doing so through the spectral data, which are conserved and somehow well behaved for data in the sense of [4 Corollary 2.1].
A second approach is to strengthen the bootstrap of [20 Section 4] directly. There, the quartic and higher terms of the current are controlled by two copies of the local smoothing norm times a small factor . Below , one would need to replace this smallness by coercivity. The Hayashi–Ozawa coercivity [24, 26] is an -level statement for the gauged energy, and it does not obviously localize to the frequency-dependent quantities . Finding a coercive -level substitute would be a significant step, for instance a lower bound on for that is uniform across frequency scales.
Concluding remarks and further directions
We have isolated a quantitative question left open by the well-posedness theory of [20]: on which balls of does local smoothing for DNLS hold uniformly? We proved that it holds for small mass, with a scale-invariant Morrey-type form, and that it fails from mass on. We also proved that the uniform constant must blow up at least like , which is exactly the rate exhibited by the soliton family. Finally, we made explicit the dependence on the data in the local smoothing estimate of [20]. It is carried by a single frequency scale , and it is polynomial with an exponent between one and two.
The results proved here rest on [20] as a black box together with elementary arguments, and their value lies mainly in the questions they make precise. Conjecture 1.5 asserts that , the mass of the stationary soliton and the Hayashi–Ozawa coercivity threshold, is sharp. Conjecture 5.2 gives a spectral form, and Conjecture 5.3 the sharp dependence on the frequency scale. Question 5.4 asks for global-in-time smoothing at small mass. A positive answer would be a nonlinear Kato smoothing theorem at critical regularity, in a setting where small solitons prevent scattering. Several further directions are natural. The commuting-flow method of [20] applies, at least formally, to the whole Kaup–Newell hierarchy, and the threshold question can be asked for each flow. For odd flows the relevant soliton velocities change, and the threshold may differ from . The models (1.33) and (1.36) of [20] are obtained from (1) by a Galilean-type modulation and by a gauge transformation, respectively, and they are well-posed in by [20 Corollary 1.13]. The threshold can be studied for them as well, keeping in mind that gauge transformations do not commute with spatial localization. For the continuum Calogero–Moser model, where is already a well-posedness threshold [40, 44], it would be interesting to know whether uniform local smoothing holds up to . Finally, the periodic problem in remains open. Since local smoothing is unavailable on the circle, a different mechanism will be needed there.
Declarations
Funding. This research received no funding.
Conflict of interest. The author declares no conflict of interest.
Data availability. Data sharing is not applicable to this article, as no datasets were generated or analysed.
References
Department of Mathematics, College of Science, Imam Abdulrahman Bin Faisal University, P.O. Box 1982, Dammam, Saudi Arabia
Basic and Applied Scientific Research Center, Imam Abdulrahman Bin Faisal University, P.O. Box 1982, 31441, Dammam, Saudi Arabia
Email address: [email protected]
Email address: [email protected]
Email address: [email protected]
URL: https://orcid.org/0000-0001-6038-1069
References
- [1]H. Bahouri, T. M. Leslie and G. Perelman, Hˢ bounds for the derivative nonlinear Schrödinger equation, Math. Res. Lett. 30 (2023), 1299–1333.
- [2]H. Bahouri and G. Perelman, Global well-posedness for the derivative nonlinear Schrödinger equation, Invent. Math. 229 (2022), 639–688.arxiv.org/abs/2012.01923
- [3]H. Bahouri and G. Perelman, Global well-posedness for the derivative nonlinear Schrödinger equation with periodic boundary condition, Int. Math. Res. Not. IMRN 2024, no. 24, 14479–14518.DOI
- [4]H. Bahouri and G. Perelman, Global well-posedness for the derivative nonlinear Schrödinger equation on the circle, preprint, arXiv:2609.10005 (2026).arxiv.org/abs/2609.10005
- [5]R. Bai and M. Majdoub, Local smoothing at finite scale: an exact identity and sharp constants, in preparation.
- [6]R. Bai, M. Majdoub and R. Jia, Scattering below the ground state for the inhomogeneous nonlinear Schrödinger equation in one dimension, in preparation.
- [7]H. A. Biagioni and F. Linares, Ill-posedness for the derivative Schrödinger and generalized Benjamin–Ono equations, Trans. Amer. Math. Soc. 353 (2001), 3649–3659.DOI
- [8]M. Colin and M. Ohta, Stability of solitary waves for derivative nonlinear Schrödinger equation, Ann. Inst. H. Poincaré Anal. Non Linéaire 23 (2006), 753–764.DOI
- [9]J. Colliander, M. Keel, G. Staffilani, H. Takaoka and T. Tao, Global well-posedness for Schrödinger equations with derivative, SIAM J. Math. Anal. 33 (2001), 649–669.DOI
- [10]J. Colliander, M. Keel, G. Staffilani, H. Takaoka and T. Tao, A refined global well-posedness result for Schrödinger equations with derivative, SIAM J. Math. Anal. 34 (2002), 64–86.
- [11]P. Constantin and J.-C. Saut, Local smoothing properties of dispersive equations, J. Amer. Math. Soc. 1 (1988), 413–439.DOI
- [12]Y. Deng, A. R. Nahmod and H. Yue, Optimal local well-posedness for the periodic derivative nonlinear Schrödinger equation, Comm. Math. Phys. 384 (2021), 1061–1107.DOI
- [13]B. Dodson, Global well-posedness and scattering for the mass critical nonlinear Schrödinger equation with mass below the mass of the ground state, Adv. Math. 285 (2015), 1589–1618.arxiv.org/abs/1104.1114
- [14]N. Fukaya, M. Hayashi and T. Inui, A sufficient condition for global existence of solutions to a generalized derivative nonlinear Schrödinger equation, Anal. PDE 10 (2017), 1149–1167.DOI
- [15]P. Gérard and E. Lenzmann, The Calogero–Moser derivative nonlinear Schrödinger equation, Comm. Pure Appl. Math. 77 (2024), 4008–4062.arxiv.org/abs/2208.04105
- [16]A. Grünrock, Bi- and trilinear Schrödinger estimates in one space dimension with applications to cubic NLS and DNLS, Math. Res. Not. 2005, 2525–2558.
- [17]A. Grünrock and S. Herr, Low regularity local well-posedness of the derivative nonlinear Schrödinger equation with periodic initial data, SIAM J. Math. Anal. 39 (2008), 1890–1920.DOI
- [18]S. Guo, X. Ren and B. Wang, Local well-posedness for the derivative nonlinear Schrödinger equation with L²-subcritical data, Discrete Contin. Dyn. Syst. 41 (2021), 4207–4253.DOI
- [19]Z. Guo and Y. Wu, Global well-posedness for the derivative nonlinear Schrödinger equation in H¹/²(ℝ), Discrete Contin. Dyn. Syst. 37 (2017), 257–264.
- [20]B. Harrop-Griffiths, R. Killip, M. Ntekoume and M. Vişan, Global well-posedness for the derivative nonlinear Schrödinger equation in L²(ℝ), J. Eur. Math. Soc. 28 (2026), 843–924.
- [21]B. Harrop-Griffiths, R. Killip and M. Vişan, Large-data equicontinuity for the derivative NLS, Int. Math. Res. Not. IMRN 2023, 4601–4642.
- [22]B. Harrop-Griffiths, R. Killip and M. Vişan, Sharp well-posedness for the cubic NLS and mKdV in H³(ℝ), Forum Math. Pi 12 (2024), e6.
- [23]N. Hayashi, The initial value problem for the derivative nonlinear Schrödinger equation in the energy space, Nonlinear Anal. 20 (1993), 823–833.DOI
- [24]N. Hayashi and T. Ozawa, On the derivative nonlinear Schrödinger equation, Phys. D 55 (1992), 14–36.DOI
- [25]N. Hayashi and T. Ozawa, Modified wave operators for the derivative nonlinear Schrödinger equation, Math. Ann. 298 (1994), 557–576.DOI
- [26]N. Hayashi and T. Ozawa, Finite energy solutions of nonlinear Schrödinger equations of derivative type, SIAM J. Math. Anal. 25 (1994), 1488–1503.DOI
- [27]S. Herr, On the Cauchy problem for the derivative nonlinear Schrödinger equation with periodic boundary condition, Int. Math. Res. Not. 2006, Art. ID 96763.DOI
- [28]J. Hogan and M. Kowalski, Turbulent threshold for continuum Calogero–Moser models, Pure Appl. Anal. 6 (2024), 941–954.DOI
- [29]R. Jenkins, J. Liu, P. Perry and C. Sulem, Soliton resolution for the derivative nonlinear Schrödinger equation, Comm. Math. Phys. 363 (2018), 1003–1049.DOI
- [30]R. Jenkins, J. Liu, P. A. Perry and C. Sulem, Global well-posedness for the derivative nonlinear Schrödinger equation, Comm. Partial Differential Equations 43 (2018), 1151–1195.
- [31]R. Jenkins, J. Liu, P. Perry and C. Sulem, The derivative nonlinear Schrödinger equation: global well-posedness and soliton resolution, Quart. Appl. Math. 78 (2020), 33–73.arxiv.org/abs/1905.02866
- [32]R. Jenkins, J. Liu, P. Perry and C. Sulem, Global existence for the derivative nonlinear Schrödinger equation with arbitrary spectral singularities, Anal. PDE 13 (2020), 1539–1578.DOI
- [33]T. Kato, On the Cauchy problem for the (generalized) Korteweg–de Vries equation, in: Studies in Applied Mathematics, Adv. Math. Suppl. Stud. 8, Academic Press, New York, 1983, 93–128.
- [34]D. J. Kaup and A. C. Newell, An exact solution for a derivative nonlinear Schrödinger equation, J. Math. Phys. 19 (1978), 798–801.DOI
- [35]C. E. Kenig and F. Merle, Global well-posedness, scattering and blow-up for the energy-critical, focusing, non-linear Schrödinger equation in the radial case, Invent. Math. 166 (2006), 645–675.
- [36]C. E. Kenig, G. Ponce and L. Vega, Oscillatory integrals and regularity of dispersive equations, Indiana Univ. Math. J. 40 (1991), 33–69.
- [37]C. E. Kenig, G. Ponce and L. Vega, Small solutions to nonlinear Schrödinger equations, Ann. Inst. H. Poincaré Anal. Non Linéaire 10 (1993), 255–288.DOI
- [38]C. E. Kenig, G. Ponce and L. Vega, Smoothing effects and local existence theory for the generalized nonlinear Schrödinger equations, Invent. Math. 134 (1998), 489–545.DOI
- [39]R. Killip, T. Laurens, J. Murphy and M. Vişan, Dispersive decay and scattering for continuum Calogero–Moser models, preprint, arXiv:2609.12985 (2026).arxiv.org/abs/2609.12985
- [40]R. Killip, T. Laurens and M. Vişan, Scaling-critical well-posedness for continuum Calogero–Moser models on the line, Comm. Amer. Math. Soc. 5 (2025), 284–320.DOI
- [41]R. Killip, M. Ntekoume and M. Vişan, On the well-posedness problem for the derivative nonlinear Schrödinger equation, Anal. PDE 16 (2023), 1245–1270.DOI
- [42]R. Killip and M. Vişan, KdV is well-posed in H⁻¹, Ann. of Math. (2) 190 (2019), 249–305.DOI
- [43]R. Killip, M. Vişan and X. Zhang, Low regularity conservation laws for integrable PDE, Geom. Funct. Anal. 28 (2018), 1062–1090.arxiv.org/abs/1708.05362
- [44]K. Kim, T. Kim and S. Kwon, Construction of smooth chiral finite-time blow-up solutions to Calogero–Moser derivative nonlinear Schrödinger equation, preprint, arXiv:2404.09603 (2024).arxiv.org/abs/2404.09603
- [45]F. Klaus and R. Schippa, A priori estimates for the derivative nonlinear Schrödinger equation, Funkcial. Ekvac. 65 (2022), 329–346.
- [46]S. Kwon and Y. Wu, Orbital stability of solitary waves for derivative nonlinear Schrödinger equation, J. Anal. Math. 135 (2018), 473–486.arxiv.org/abs/1603.03745
- [47]S. Le Coz and Y. Wu, Stability of multisolitons for the derivative nonlinear Schrödinger equation, Int. Math. Res. Not. IMRN 2018, 4120–4170.DOI
- [48]J.-H. Lee, Global solvability of the derivative nonlinear Schrödinger equation, Trans. Amer. Math. Soc. 314 (1989), 107–118.
- [49]J. Liu, P. A. Perry and C. Sulem, Global existence for the derivative nonlinear Schrödinger equation by the method of inverse scattering, Comm. Partial Differential Equations 41 (2016), 1692–1760.arxiv.org/abs/1511.01173
- [50]J. Liu, P. A. Perry and C. Sulem, Long-time behavior of solutions to the derivative nonlinear Schrödinger equation for soliton-free initial data, Ann. Inst. H. Poincaré Anal. Non Linéaire 35 (2018), 217–265.DOI
- [51]M. Majdoub and T. Saanouni, Normalized standing waves for the focusing inhomogeneous Schrödinger equation with spatially growing nonlinearity, preprint, arXiv:2602.07505 (2026).arxiv.org/abs/2602.07505
- [52]C. Miao, Y. Wu and G. Xu, Global well-posedness for Schrödinger equation with derivative in H¹/²(ℝ), J. Differential Equations 251 (2011), 2164–2195.DOI
- [53]K. Mio, T. Ogino, K. Minami and S. Takeda, Modified nonlinear Schrödinger equation for Alfvén waves propagating along the magnetic field in cold plasmas, J. Phys. Soc. Japan 41 (1976), 265–271.
- [54]E. Mjølhus, On the modulational instability of hydromagnetic waves parallel to the magnetic field, J. Plasma Phys. 16 (1976), 321–334.DOI
- [55]R. Mosincat, Global well-posedness of the derivative nonlinear Schrödinger equation with periodic boundary condition in H¹/², J. Differential Equations 263 (2017), 4658–4722.DOI
- [56]R. Mosincat and T. Oh, A remark on global well-posedness of the derivative nonlinear Schrödinger equation on the circle, C. R. Math. Acad. Sci. Paris 353 (2015), 837–841.arxiv.org/abs/1502.02261
- [57]T. Ozawa, On the nonlinear Schrödinger equations of derivative type, Indiana Univ. Math. J. 45 (1996), 137–163.DOI
- [58]D. E. Pelinovsky and Y. Shimabukuro, Existence of global solutions to the derivative NLS equation with the inverse scattering transform method, Int. Math. Res. Not. IMRN 2018, 5663–5728.arxiv.org/abs/1602.02118
- [59]D. E. Pelinovsky, A. Saalmann and Y. Shimabukuro, The derivative NLS equation: global existence with solitons, Dyn. Partial Differ. Equ. 14 (2017), 271–294.arxiv.org/abs/1703.05277
- [60]P. Sjölin, Regularity of solutions to the Schrödinger equation, Duke Math. J. 55 (1987), 699–715.DOI
- [61]H. Takaoka, Well-posedness for the one-dimensional nonlinear Schrödinger equation with the derivative nonlinearity, Adv. Differential Equations 4 (1999), 561–580.
- [62]H. Takaoka, Global well-posedness for Schrödinger equations with derivative in a nonlinear term and data in low-order Sobolev spaces, Electron. J. Differential Equations 2001, No. 42.
- [63]X. Tang and G. Xu, Microscopic conservation laws for the derivative nonlinear Schrödinger equation, Lett. Math. Phys. 111 (2021), Paper No. 138.
- [64]T. Tao, Nonlinear Dispersive Equations: Local and Global Analysis, CBMS Regional Conference Series in Mathematics 106, Amer. Math. Soc., Providence, RI, 2006.
- [65]M. Tsutsumi and I. Fukuda, On solutions of the derivative nonlinear Schrödinger equation. Existence and uniqueness theorem, Funkcial. Ekvac. 23 (1980), 259–277.DOI
- [66]L. Vega, Schrödinger equations: pointwise convergence to the initial data, Proc. Amer. Math. Soc. 102 (1988), 874–878.
- [67]L. Vega and N. Visciglia, On the local smoothing for the Schrödinger equation, Proc. Amer. Math. Soc. 135 (2007), 119–128.DOI
- [68]M. I. Weinstein, Nonlinear Schrödinger equations and sharp interpolation estimates, Comm. Math. Phys. 87 (1983), 567–576.
- [69]Y. Wu, Global well-posedness for the nonlinear Schrödinger equation with derivative in energy space, Anal. PDE 6 (2013), 1989–2002.arxiv.org/abs/1310.7166
- [70]Y. Wu, Global well-posedness on the derivative nonlinear Schrödinger equation, Anal. PDE 8 (2015), 1101–1112.DOI
