betaFeedback

WorksMathematics

Power-Type Nonlinear Schrödinger Equations with the Logarithmic Laplacian

Abstract

We prove global existence of the Cauchy problem of a Schrodinger equation with the logarithmic laplacian and a pure power linearity. We address the focusing and the defocusing cases, and we obtained a sharp blow-up in time result.

Introduction and functional setting

We consider the Cauchy problem

\begin{cases} i\partial_t u = \mathcal{L}u+\mu\lvert u\rvert^{p-1}u, & (t,x)\in\mathbb{R}\times\mathbb{R}^{N},\\ u(0,x)=u_0(x), & x\in\mathbb{R}^{N}, \end{cases} \qquad\mathcal{L}=\log(-\Delta),\quad p>1, \tag*{(1)}\end{cases}

where uu is complex-valued and μ∈R\mu\in\mathbb{R}. The sign μ>0\mu>0 is defocusing; the focusing equation is written with μ=−κ\mu=-\kappa, κ>0\kappa>0. With the unitary Fourier transform,

Lf^(ξ)=log⁡(∣ξ∣2)f^(ξ).(2)\widehat{\mathcal{L}f}(\xi)=\log(\lvert\xi\rvert^{2})\widehat{f}(\xi). \tag*{(2)}

The logarithmic Laplacian is connected to the fractional family through the formal relation

log⁡(−Δ)=∂s(−Δ)s∣s=0.\log(-\Delta)=\left.\partial_s(-\Delta)^s\right|_{s=0}.

Chen and Weth [1] developed its integral representation and a functional framework for bounded-domain Dirichlet problems. Their work provides background for the operator; the evolution results proved here concern the whole-space nonlinear Cauchy problem (1). Throughout this paper, the logarithm occurs in the dispersion operator, rather than in a nonlinearity of the form ulog⁡∣u∣2u\log\lvert u\rvert^{2}.

The logarithmic symbol is unbounded both above and below. At high frequencies it grows more slowly than every positive power, while near zero it has a negative singularity. Accordingly, the Sobolev, operator, and logarithmic form domains must be distinguished. In particular, high-frequency Sobolev regularity does not by itself remove the low-frequency domain restriction for log⁡(−Δ)\log(-\Delta). This issue persists even though its Schrödinger group is unitary on every Sobolev space.

For sufficiently regular nonlinearities, the semigroup contraction method reduces local solvability to Sobolev product and composition estimates. We follow that framework, as presented for dispersive equations in [4], while keeping the low-frequency operator domain separate from the space of mild solutions. At the global level, unitarity alone is insufficient for the contraction argument. The additional observation used here is that the logarithmic phase has a nondegenerate Hessian on every fixed frequency annulus. This gives annular time decay. The abstract theorem of Keel and Tao [2], followed by spatial scaling and dyadic summation, then supplies a global space-time bound with Sobolev regularity m>N/2m > N/2. That bound closes a small-data argument for both signs of the nonlinearity.

Large-data energy methods have a different structure. For the defocusing equation, the signed logarithmic kinetic form prevents an immediate positive-energy estimate. We control its negative part directly through the Fourier Duhamel formula. For 2≤p≤32 \le p \le3, interpolation with the conserved mass reduces this control to a linear Gronwall inequality. Smooth Hamiltonian approximations and the compactness framework of Simon [3] then yield global weak solutions. This construction does not establish uniqueness in the full energy class or global persistence of Sobolev regularity.

The focusing equation presents an additional difficulty: mass-preserving concentration makes the energy unbounded below for every pure power p>1p > 1. Fractional Schrödinger blow-up results, such as those of Boulenger, Himmelsbach, and Lenzmann [5] and Dinh [6], use positive-order dispersion in the range 1/2<s<11/2 < s < 1. They cannot be applied directly to (1). We therefore distinguish the derivation of necessary growth rates for an assumed finite lifespan from a proof that such a lifespan actually occurs.

The results proved below are as follows.

  • (i) Local well-posedness in HmH^m, m>N/2m > N/2, for smooth pure powers, with mass conservation and an amplitude-based continuation criterion.

  • (ii) Frequency-localized dispersive estimates and global small-data well-posedness and scattering, for either sign, in dimensions N≥3N \ge3.

  • (iii) Global defocusing weak solutions in the logarithmic form space for 2≤p≤32 \le p \le3, without a Sobolev smoothness assumption on the power.

  • (iv) Focusing low-frequency estimates, failure of energy coercivity, and dilation and variance identities under the requisite moment regularity.

  • (v) Lower bounds on Sobolev growth conditional on a finite maximal lifespan, including an energy-refined logarithmic correction.

The local, global small-data, and weak-existence results are independent statements with different assumptions; their exact hypotheses are given in the corresponding theorems. The final two sections examine focusing a priori estimates and conditional singularity rates. No theorem asserting the occurrence of finite-time blow-up is proved. In particular, conditional blow-up bounds are not constructions of blow-up solutions. The arguments make no claim of optimality or priority.

Domains and solution concepts

All function spaces are over RN\mathbb{R}^{N} unless otherwise stated. Write ∣D∣s|D|^{s} for the Fourier multiplier ∣ξ∣s|\xi|^{s}. The self-adjoint realization of L\mathcal{L} on L2L^{2} has domain

D(L)={f∈L2:∫∣log⁡(∣ξ∣2)∣2∣f^(ξ)∣2 dξ<∞}.(3)D(\mathcal{L})=\left\{f\in L^{2}:\int\left|\log(|\xi|^{2})\right|^{2}|\widehat{f}(\xi)|^{2}\,d\xi<\infty\right\}. \tag*{(3)}

The logarithmic form space is

Hlog⁡=D(∣L∣1/2),∥f∥Hlog⁡2=∥f∥22+∫∣log⁡(∣ξ∣2)∣∣f^(ξ)∣2 dξ.(4)\mathcal{H}_{\log}=D(|\mathcal{L}|^{1/2}),\qquad\|f\|_{\mathcal{H}_{\log}}^{2}=\|f\|_{2}^{2}+\int|\log(|\xi|^{2})||\widehat{f}(\xi)|^{2}\,d\xi. \tag*{(4)}

Neither inclusion Hm⊂D(L)H^{m}\subset D(\mathcal{L}) nor Hm⊂Hlog⁡H^{m}\subset\mathcal{H}_{\log} follows from high-frequency regularity alone; the issue is at ξ=0\xi=0. For example, the Fourier profile

f^(ξ)=1{0<∣ξ∣<e−1}∣ξ∣−N/2(log⁡1∣ξ∣)−1\widehat{f}(\xi)=\mathbf{1}_{\{0<|\xi|<e^{-1}\}}|\xi|^{-N/2}\left(\log\frac{1}{|\xi|}\right)^{-1}

defines an element of every HmH^{m}, but K−(f)=∞K_{-}(f)=\infty: the corresponding radial integrals are respectively ∫1∞s−2 ds\int_{1}^{\infty}s^{-2}\,ds and a constant multiple of ∫1∞s−1 ds\int_{1}^{\infty}s^{-1}\,ds.

Define

K+(f)=∫∣ξ∣≥1log⁡(∣ξ∣2)∣f^(ξ)∣2 dξ,(5)K_{+}(f)=\int_{|\xi|\geq1}\log(|\xi|^{2})|\widehat{f}(\xi)|^{2}\,d\xi, \tag*{(5)}
K−(f)=∫∣ξ∣<1−log⁡(∣ξ∣2)∣f^(ξ)∣2 dξ.(6)K_{-}(f)=\int_{|\xi|<1}-\log(|\xi|^{2})|\widehat{f}(\xi)|^{2}\,d\xi. \tag*{(6)}

For f∈Hlog⁡f\in\mathcal{H}_{\log}, the signed kinetic form is K(f)=K+(f)−K−(f)K(f)=K_{+}(f)-K_{-}(f). The formal conserved quantities are

M(f)=∥f∥22,Eμ(f)=12K(f)+μp+1∥f∥p+1p+1.(7)M(f)=\|f\|_{2}^{2},\qquad E_{\mu}(f)=\frac{1}{2}K(f)+\frac{\mu}{p+1}\|f\|_{p+1}^{p+1}. \tag*{(7)}

Set U(t)=e−itLU(t)=e^{-it\mathcal{L}}, so

U(t)f^(ξ)=∣ξ∣−2itf^(ξ).(8)\widehat{U(t)f}(\xi)=|\xi|^{-2it}\widehat{f}(\xi). \tag*{(8)}

The value of the multiplier at ξ=0\xi=0 is irrelevant. Since it has modulus one almost everywhere, U(t)U(t) is a strongly continuous unitary group on every HsH^{s}, s∈Rs\in\mathbb{R}. It also preserves D(L)D(\mathcal{L}) and Hlog⁡\mathcal{H}_{\log} with their graph and form norms.

Definition 1.1. A mild HmH^{m} solution on an interval II containing zero is a function u∈C(I;Hm)u\in C(I;H^{m}) satisfying

u(t)=U(t)u0−iμ∫0tU(t−s)∣u(s)∣p−1u(s) ds(9)u(t)=U(t)u_{0}-i\mu\int_{0}^{t}U(t-s)|u(s)|^{p-1}u(s)\,ds \tag*{(9)}

in HmH^{m}. A strong L2L^{2} solution additionally belongs to C(I;D(L))∩C1(I;L2)C(I;D(\mathcal{L}))\cap C^{1}(I;L^{2}) and satisfies the equation in L2L^{2}.

For weak solutions, put

Vp=Hlog⁡∩Lp+1,∥f∥Vp=∥f∥Hlog⁡+∥f∥p+1.(10)V_{p}=\mathcal{H}_{\log}\cap L^{p+1},\qquad\|f\|_{V_{p}}=\|f\|_{\mathcal{H}_{\log}}+\|f\|_{p+1}. \tag*{(10)}

This is a reflexive Banach space, continuously and densely embedded in L2L^{2}. The kinetic form

l(u,v)=∫log⁡(∣ξ∣2)u^(ξ)v^(ξ)‾ dξ(11)\mathfrak{l}(u,v)=\int\log(|\xi|^{2})\widehat{u}(\xi)\overline{\widehat{v}(\xi)}\,d\xi \tag*{(11)}

defines a continuous map L:Hlog⁡→Hlog⁡∗\mathcal{L}:\mathcal{H}_{\log}\to\mathcal{H}_{\log}^{*}. A weak solution satisfies (1.1) in Vp∗V_{p}^{*}, with this form interpretation of L\mathcal{L}, and has the initial trace u0u_{0} in L2L^{2}.

Local well-posedness and conservation laws

Throughout the Sobolev theory, use the following sufficient smoothness hypothesis:

m∈N,m>N/2,p∈{3,5,7,…} or p>m+1.(12)m \in\mathbb{N},\qquad m > N/2,\qquad p \in\{3,5,7,\ldots\}\ \text{or}\ p > m+1. \tag*{(12)}

The restriction on real noninteger powers is convenient and is not claimed to be sharp.

Lemma 2.1 (Nonlinear estimates). Under (12), F(u)=∣u∣p−1uF(u) = |u|^{p-1}u is locally Lipschitz from HmH^m to itself. More precisely,

∥F(u)∥Hm≤C∥u∥∞p−1∥u∥Hm,(13)\|F(u)\|_{H^m} \le C\|u\|_{\infty}^{p-1}\|u\|_{H^m}, \tag*{(13)}
∥∣D∣mF(u)∥2≤C∥u∥∞p−1∥∣D∣mu∥2,(14)\||D|^mF(u)\|_2 \le C\|u\|_{\infty}^{p-1}\||D|^m u\|_2, \tag*{(14)}

and

∥F(u)−F(v)∥Hm≤C[(∥u∥∞+∥v∥∞)p−1∥u−v∥Hm+(∥u∥∞+∥v∥∞)p−2(∥u∥Hm+∥v∥Hm)∥u−v∥∞].(15)\begin{aligned} \|F(u)-F(v)\|_{H^m} &\le C\left[(\|u\|_{\infty}+\|v\|_{\infty})^{p-1}\|u-v\|_{H^m}\right.\\ &\qquad\left.+(\|u\|_{\infty}+\|v\|_{\infty})^{p-2}(\|u\|_{H^m}+\|v\|_{H^m})\|u-v\|_{\infty}\right]. \tag*{(15)} \end{aligned}

Proof. For odd p=2k+1p=2k+1, the nonlinearity is the polynomial uk+1u‾ku^{k+1}\overline{u}^{k}. For real p>m+1p>m+1, it is a Cm+1C^{m+1} map on R2\mathbb{R}^2, and ∣DjF(z)∣≤Cj∣z∣p−j|D^jF(z)|\le C_j|z|^{p-j} for the required derivative orders. The chain rule, Hölder’s inequality, and the integer-order Gagliardo–Nirenberg estimates give (13) and (14). For example, a term with kk differentiated factors whose orders sum to mm is bounded by C∥u∥∞p−k−1∥u∥∞k−1∥∣D∣mu∥2C\|u\|_{\infty}^{p-k-1}\|u\|_{\infty}^{k-1}\||D|^m u\|_2. For the difference, write

F(u)−F(v)=∫01DF(v+θ(u−v))(u−v) dθF(u)-F(v)=\int_0^1 DF(v+\theta(u-v))(u-v)\,d\theta

and apply the tame product and composition estimates. These are the standard Sobolev estimates used in the semigroup method; see [4 Chapter 3 and Appendix A].

Theorem 2.2 (Local theory). Assume (12). For every u0∈Hmu_0\in H^m and μ∈R\mu\in\mathbb{R}, there is a unique maximal mild solution

u∈C((−T−,T+);Hm).u\in C((-T_-,T_+);H^m).

The solution map is locally Lipschitz on bounded sets over a common sufficiently short interval. If T+<∞T_+<\infty, then

lim⁡t↑T+∥u(t)∥Hm=∞.(16)\lim_{t\uparrow T_+}\|u(t)\|_{H^m}=\infty. \tag*{(16)}

For μ≠0\mu\ne0, finite T+T_+ also implies

∫0T+∥u(t)∥∞p−1 dt=∞.(17)\int_0^{T_+}\|u(t)\|_{\infty}^{p-1}\,dt=\infty. \tag*{(17)}

The corresponding statements hold backward in time.

Proof. Use C([−T,T];Hm)C([-T,T];H^m) with its supremum norm. Sobolev embedding and Lemma 2.1, together with unitarity of U(t)U(t), give for the Duhamel map Φ\Phi

∥Φ(u)∥CHm≤∥u0∥Hm+C∣μ∣TRp,∥Φ(u)−Φ(v)∥CHm≤C∣μ∣TRp−1∥u−v∥CHm\|\Phi(u)\|_{CH^m}\le\|u_0\|_{H^m}+C|\mu|TR^p,\qquad \|\Phi(u)-\Phi(v)\|_{CH^m}\le C|\mu|TR^{p-1}\|u-v\|_{CH^m}

on a ball of radius RR. For u0≠0u_0\ne0, choose R=2∥u0∥HmR=2\|u_0\|_{H^m} and then TT so that the map preserves the ball and is a contraction. Zero data give the zero solution.

For two solutions bounded by RR on a compact interval, Gronwall’s inequality yields

∥u(t)−v(t)∥Hm≤eCR∣μ∣∣t∣∥u0−v0∥Hm.\|u(t)-v(t)\|_{H^m} \le e^{CR|\mu||t|}\|u_0-v_0\|_{H^m}.

This gives uniqueness and dependence on the data. Restarting the local construction at any time proves (16), since the local lifespan is bounded below on bounded HmH^m sets.

Finally, (13) gives

∥u(t)∥Hm≤∥u0∥Hmexp⁡(C∣μ∣∫0t∥u(s)∥∞p−1 ds).(18)\|u(t)\|_{H^m} \le\|u_0\|_{H^m}\exp\left(C|\mu|\int_0^t\|u(s)\|_{\infty}^{p-1}\,ds\right). \tag*{(18)}

If the integral in (17) were finite, this estimate would contradict (16).

Proposition 2.3 (Mass and strong solutions). Every mild solution in Theorem 2.2 conserves mass. If also u0∈D(L)u_0 \in D(\mathcal{L}), then on its maximal interval

u∈C(Hm∩D(L))∩C1(L2),u \in C(H^m \cap D(\mathcal{L})) \cap C^1(L^2),

where D(L)D(\mathcal{L}) has its graph norm, and Eμ(u(t))=Eμ(u0)E_\mu(u(t))=E_\mu(u_0).

Proof. The interaction variable v(t)=U(−t)u(t)v(t)=U(-t)u(t) satisfies

v(t)=u0−iμ∫0tU(−s)F(u(s)) ds,v(t)=u_0-i\mu\int_0^t U(-s)F(u(s))\,ds,

and is C1C^1 in HmH^m. Unitarity gives

ddt∥u(t)∥22=2Re⁡(−iμ∫∣u(t)∣p+1)=0.\frac{d}{dt}\|u(t)\|_2^2=2\operatorname{Re}\left(-i\mu\int|u(t)|^{p+1}\right)=0.

The powers in (12) satisfy p≥2p \ge2. Thus F(u)∈Hm∩L1F(u)\in H^m\cap L^1 and

∥F(u)∥1=∥u∥pp≤∥u∥∞p−2∥u∥22.(19)\|F(u)\|_1=\|u\|_p^p\le\|u\|_\infty^{p-2}\|u\|_2^2. \tag*{(19)}

For f∈Hm∩L1f\in H^m\cap L^1,

∥Lf∥2≤Cm∥f∥Hm+CN∥f∥1.(20)\|\mathcal{L}f\|_2\le C_m\|f\|_{H^m}+C_N\|f\|_1. \tag*{(20)}

At high frequencies this follows from the logarithm’s subpower growth; at low frequencies use ∥f^∥∞≤C∥f∥1\|\widehat f\|_\infty\le C\|f\|_1 and ∫∣ξ∣<1∣log⁡(∣ξ∣2)∣2 dξ<∞\int_{|\xi|<1}|\log(|\xi|^2)|^2\,d\xi<\infty. The nonlinearity is continuous in Hm∩L1H^m\cap L^1 along the solution. The Duhamel formula in the graph norm therefore proves persistence of D(L)D(\mathcal{L}) and C1(L2)C^1(L^2) regularity.

For this regularity, the derivative of the kinetic form is 2Re⁡⟨Lu,ut⟩2\operatorname{Re}\langle\mathcal{L}u,u_t\rangle. The potential energy is differentiable along the L2L^2 path, since uu is locally bounded in L∞L^\infty. Consequently

ddtEμ(u)=Re⁡⟨Lu+μF(u),ut⟩=Re⁡⟨iut,ut⟩=0.\frac{d}{dt}E_\mu(u)=\operatorname{Re}\langle\mathcal{L}u+\mu F(u),u_t\rangle=\operatorname{Re}\langle iu_t,u_t\rangle=0.

Remark 2.4 (The Bessel logarithm). The local theorem, mass conservation, and amplitude criterion hold also for log⁡(I−Δ)\log(I-\Delta). In that case Hm⊂D(log⁡(I−Δ))H^m\subset D(\log(I-\Delta)) automatically, so the local solutions above are strong L2L^2 solutions without an extra low-frequency assumption. The global dispersive argument in the next section is specifically for log⁡(−Δ)\log(-\Delta); it is not being transferred to the Bessel logarithm.

Frequency-localized dispersion and small-data scattering

Lemma 3.1 (Annular dispersion). Let ψ∈Cc∞(RN∖{0})\psi\in C^\infty_c(\mathbb{R}^N \setminus\{0\}) be supported in a fixed annulus. Then

∥U(t)ψ(D)f∥∞≤Cψ(1+∣t∣)−N/2∥f∥1.(21)\|U(t)\psi(D)f\|_\infty\le C_\psi(1+|t|)^{-N/2}\|f\|_1 . \tag*{(21)}

Proof. The kernel is

Kψ(t,x)=(2π)−N∫ei(x⋅ξ−tlog⁡∣ξ∣2)ψ(ξ) dξ.K_\psi(t,x)=(2\pi)^{-N}\int e^{i(x\cdot\xi-t\log|\xi|^2)}\psi(\xi)\,d\xi.

For ∣t∣≤1|t|\le1 it is bounded by C∥ψ∥1C\|\psi\|_1. For larger times, write the phase as tΦy(ξ)t\Phi_y(\xi) with y=x/ty=x/t and Φy(ξ)=y⋅ξ−log⁡∣ξ∣2\Phi_y(\xi)=y\cdot\xi-\log|\xi|^2. Its Hessian is the negative of

D2log⁡∣ξ∣2=2∣ξ∣2I−4∣ξ∣4ξ⊗ξ.D^2\log|\xi|^2=\frac{2}{|\xi|^2}I-\frac{4}{|\xi|^4}\xi\otimes\xi.

The radial eigenvalue is −2/∣ξ∣2-2/|\xi|^2, while tangential eigenvalues are 2/∣ξ∣22/|\xi|^2. They are uniformly nonzero on the annulus. The map ξ↦2ξ/∣ξ∣2\xi\mapsto2\xi/|\xi|^2 is injective, so there is at most one stationary point. Parameter-uniform stationary phase near it and nonstationary integration by parts on its complement give sup⁡x∣Kψ(t,x)∣≤C∣t∣−N/2\sup_x|K_\psi(t,x)|\le C|t|^{-N/2}. The parameter yy need only be considered in a fixed compact set; outside it the phase gradient is uniformly separated from zero. This proves the estimate.

Let PjP_j be a homogeneous smooth dyadic partition, supported where ∣ξ∣∼2j|\xi|\sim2^j.

Proposition 3.2 (Global space-time bound). Let N≥3N\ge3, m>N/2m>N/2, and 2≤q<∞2\le q<\infty. Then

∥U(t)f∥Lq(R;L∞)≤C∥f∥Hm.(22)\|U(t)f\|_{L^q(\mathbb{R};L^\infty)}\le C\|f\|_{H^m}. \tag*{(22)}

Moreover, for G∈L1(R;Hm)G\in L^1(\mathbb{R};H^m),

∥∫0tU(t−s)G(s) ds∥Lt∞Hxm+∥∫0tU(t−s)G(s) ds∥LtqLx∞≤C∥G∥Lt1Hxm.(23)\left\|\int_0^t U(t-s)G(s)\,ds\right\|_{L^\infty_tH^m_x} + \left\|\int_0^t U(t-s)G(s)\,ds\right\|_{L^q_tL^\infty_x} \le C\|G\|_{L^1_tH^m_x}. \tag*{(23)}

Proof. Apply the abstract Strichartz theorem of Keel–Tao [2 Theorem 1.2] to U(t)ψ(D)U(t)\psi(D). Its TT∗TT^* kernel has cutoff ∣ψ∣2|\psi|^2 and satisfies Lemma 3.1; its L2L^2 energy bound is immediate. Hence the unit-annulus estimate holds for q,r≥2q,r\ge2 with 2/q+N/r=N/22/q+N/r=N/2, including the endpoint (q,r)=(2,2N/(N−2))(q,r)=(2,2N/(N-2)). For N≥3N\ge3 each finite q≥2q\ge2 has such a finite rr.

Rescaling only space, and using log⁡∣2jξ∣2=2jlog⁡2+log⁡∣ξ∣2\log|2^j\xi|^2=2j\log2+\log|\xi|^2, gives

∥U(t)Pjf∥LtqLxr≤C22j/q∥Pjf∥2.(24)\|U(t)P_jf\|_{L^q_tL^r_x}\le C2^{2j/q}\|P_jf\|_2. \tag*{(24)}

Bernstein’s inequality therefore implies

∥U(t)Pjf∥LtqLx∞≤C2jN/r+2j/q∥Pjf∥2=C2jN/2∥Pjf∥2.\|U(t)P_jf\|_{L^q_tL^\infty_x}\le C2^{jN/r+2j/q}\|P_jf\|_2=C2^{jN/2}\|P_jf\|_2.

The dyadic sum is controlled by

∑j≤02jN/2∥Pjf∥2≤C∥f∥2,\sum_{j\le0}2^{jN/2}\|P_jf\|_2\le C\|f\|_2,
∑j>02jN/2∥Pjf∥2≤(∑j>022j(N/2−m))1/2(∑j>022jm∥Pjf∥22)1/2≤Cm∥f∥Hm.\sum_{j>0}2^{jN/2}\|P_jf\|_2 \le\left(\sum_{j>0}2^{2j(N/2-m)}\right)^{1/2} \left(\sum_{j>0}2^{2jm}\|P_jf\|_2^2\right)^{1/2} \le C_m\|f\|_{H^m}.

This proves (22), first on a dense class and then by completion. Thus the derivative loss in (24) is absorbed by the stated Sobolev regularity.

For (23), the L∞HmL^\infty H^m part follows from unitarity. For the other part, Minkowski’s inequality and (22) give, on the positive half-line,

∥∫0tU(t−s)G(s) ds∥Lt≥0qLx∞≤∫0∞∥1{t≥s}U(t−s)G(s)∥Lt≥0qLx∞ ds≤C∫0∞∥G(s)∥Hm ds.\left\|\int_0^t U(t-s)G(s)\,\mathrm{d}s\right\|_{L^q_{t\geq0}L^\infty_x} \leq\int_0^\infty\left\|\mathbf{1}_{\{t\geq s\}}U(t-s)G(s)\right\|_{L^q_{t\geq0}L^\infty_x}\,\mathrm{d}s \leq C\int_0^\infty\|G(s)\|_{H^m}\,\mathrm{d}s.

The negative half-line is identical. No retarded endpoint estimate beyond this L1L^1 forcing bound is needed.

Theorem 3.3 (Small-data global well-posedness and scattering). Let N≥3N \geq3 and assume (12). For each fixed μ∈R\mu\in\mathbb{R} there is ϵ∗>0\epsilon_* > 0 such that ∥u0∥Hm≤ϵ∗\|u_0\|_{H^m} \leq\epsilon_* implies a unique global mild solution

u∈Cb(R;Hm)∩Lp−1(R;L∞),u \in C_b(\mathbb{R};H^m) \cap L^{p-1}(\mathbb{R};L^\infty),

with

sup⁡t∈R∥u(t)∥Hm+∥u∥Ltp−1Lx∞≤C∥u0∥Hm.(25)\sup_{t\in\mathbb{R}}\|u(t)\|_{H^m}+\|u\|_{L^{p-1}_tL^\infty_x}\leq C\|u_0\|_{H^m}. \tag*{(25)}

The small-data solution map is Lipschitz in this space, and there are u±∈Hmu_\pm\in H^m such that

∥u(t)−U(t)u±∥Hm⟶0as t→±∞.(26)\|u(t)-U(t)u_\pm\|_{H^m}\longrightarrow0\quad\text{as }t\to\pm\infty. \tag*{(26)}

The theorem applies to the focusing as well as the defocusing sign.

Proof. Set q=p−1≥2q=p-1\geq2 and X=Cb(R;Hm)∩Lq(R;L∞)X=C_b(\mathbb{R};H^m)\cap L^q(\mathbb{R};L^\infty), with the sum norm. Lemma 2.1 and Hölder’s inequality yield

∥F(u)∥L1Hm≤C∥u∥Xp.(27)\|F(u)\|_{L^1H^m}\leq C\|u\|_X^p. \tag*{(27)}
∥F(u)−F(v)∥L1Hm≤C(∥u∥X+∥v∥X)p−1∥u−v∥X.(28)\|F(u)-F(v)\|_{L^1H^m}\leq C(\|u\|_X+\|v\|_X)^{p-1}\|u-v\|_X. \tag*{(28)}

For the second term of (15), the time exponents are q/(p−2)q/(p-2) and qq; their reciprocals sum to one. Proposition 3.2 gives

∥Φ(u)∥X≤C0∥u0∥Hm+C1∣μ∣∥u∥Xp,\|\Phi(u)\|_X\leq C_0\|u_0\|_{H^m}+C_1|\mu|\|u\|_X^p,

and the analogous contraction estimate from (28). On the ball of radius 2C0∥u0∥Hm2C_0\|u_0\|_{H^m} these close when C∣μ∣∥u0∥Hmp−1C|\mu|\|u_0\|_{H^m}^{p-1} is sufficiently small. Banach’s fixed point theorem gives a solution on the entire time axis, its Lipschitz dependence, and (25). Local uniqueness from Theorem 2.2 gives uniqueness among all C(R;Hm)C(\mathbb{R};H^m) mild solutions with these data.

By (27), F(u)∈L1(R;Hm)F(u)\in L^1(\mathbb{R};H^m). Define

u+=u0−iμ∫0∞U(−s)F(u(s)) ds.u_+=u_0-i\mu\int_0^\infty U(-s)F(u(s))\,\mathrm{d}s.

This integral converges in HmH^m, and

∥u(t)−U(t)u+∥Hm≤∣μ∣∫t∞∥F(u(s))∥Hm ds⟶0.\|u(t)-U(t)u_+\|_{H^m}\leq|\mu|\int_t^\infty\|F(u(s))\|_{H^m}\,\mathrm{d}s\longrightarrow0.

The backward integral gives u−u_-.

Corollary 3.4 (The focusing cubic equation). For N≥3N\geq3, m>N/2m>N/2 an integer, and sufficiently small u0∈Hmu_0\in H^m, the focusing cubic equation has a unique global mild solution and scatters in HmH^m. In particular,

sup⁡t∥u(t)∥Hm+(∫R∥u(t)∥∞2 dt)1/2≤C∥u0∥Hm.\sup_t\|u(t)\|_{H^m}+\left(\int_{\mathbb{R}}\|u(t)\|_\infty^2\,\mathrm{d}t\right)^{1/2}\leq C\|u_0\|_{H^m}.

Global defocusing weak solutions

The theorem in this section does not require (12). The restriction 2≤p≤32 \le p \le3 arises from a low-frequency differential inequality and is not claimed to be an optimal range.

Theorem 4.1 (Global weak existence). Let N≥1N \ge1, μ>0\mu> 0, and 2≤p≤32 \le p \le3. For every u0∈Vpu_0 \in V_p, there is a global weak solution of (1) satisfying

u∈C(R;L2)∩Lloc∞(R;Vp),ut∈Lloc∞(R;Vp∗).(29)u \in C(\mathbb{R};L^2) \cap L^\infty_{\mathrm{loc}}(\mathbb{R};V_p), \qquad u_t \in L^\infty_{\mathrm{loc}}(\mathbb{R};V_p^*). \tag*{(29)}

It conserves mass and obeys, for constants depending on the initial mass, form norm, nonlinear energy, and the parameters,

K+(u(t))+K−(u(t))+∥u(t)∥p+1p+1≤C0eC1∣t∣(30)K_+(u(t)) + K_-(u(t)) + \|u(t)\|_{p+1}^{p+1} \le C_0 e^{C_1|t|} \tag*{(30)}

for almost every tt. No general uniqueness, persistence of HmH^m, or energy equality is asserted for this weak solution class.

The low-frequency estimate

For a regular solution, write B(t)=K−(u(t))1/2B(t) = K_-(u(t))^{1/2} and X(t)=∥u(t)∥p+1p+1X(t) = \|u(t)\|_{p+1}^{p+1}. Since the group preserves the low-frequency weighted L2L^2 norm,

B(t)≤B(0)+CN∣μ∣∫0t∥u(s)∥pp ds(t≥0).(31)B(t) \le B(0) + C_N|\mu| \int_0^t \|u(s)\|_p^p\,ds \qquad(t \ge0). \tag*{(31)}

Here we used

∥1{∣ξ∣<1}−log⁡∣ξ∣2 f^∥2≤CN∥f∥1,\left\|\mathbf{1}_{\{|\xi|<1\}}\sqrt{-\log|\xi|^2}\,\widehat{f}\right\|_2 \le C_N\|f\|_1,

which follows from the Fourier L1L^1 bound and integrability of −log⁡∣ξ∣2-\log|\xi|^2 on the unit ball.

For p≥2p \ge2, interpolation gives

∥u∥pp≤∥u∥22/(p−1)∥u∥p+1(p+1)(p−2)/(p−1).(32)\|u\|_p^p \le\|u\|_2^{2/(p-1)}\|u\|_{p+1}^{(p+1)(p-2)/(p-1)}. \tag*{(32)}

In the defocusing case, conservation of energy yields

12K+(u(t))+μp+1X(t)=Eμ(u0)+12B(t)2,(33)\frac{1}{2}K_+(u(t)) + \frac{\mu}{p+1}X(t) = E_\mu(u_0) + \frac{1}{2}B(t)^2, \tag*{(33)}

and hence X+K+≤C(1+B2)X + K_+ \le C(1+B^2). Set α=(p−2)/(p−1)\alpha= (p-2)/(p-1). For 2≤p≤32 \le p \le3, 0≤α≤1/20 \le\alpha\le1/2, so (31)–(33) reduce to

B(t)≤B(0)+C∫0t(1+B(s)2)α ds≤B(0)+C∫0t(1+B(s)) ds.B(t) \le B(0) + C\int_0^t (1+B(s)^2)^\alpha\,ds \le B(0) + C\int_0^t (1+B(s))\,ds.

Gronwall’s inequality gives the exponential bounds needed for global weak compactness. The same argument works backward in time.

Approximation, compactness, and the initial trace

Proof of Theorem 4.1. Choose a real even χ∈Cc∞\chi\in C^\infty_c with 0≤χ≤10 \le\chi\le1 and χ=1\chi= 1 on the unit ball. Let Jn=χ(D/n)J_n = \chi(D/n), with support contained in ∣ξ∣≤2n|\xi| \le2n. These operators are uniformly bounded on LrL^r, 1<r<∞1 < r < \infty, are L2L^2 contractions, commute with L\mathcal{L}, and map L2L^2 into Hk∩L∞H^k \cap L^\infty for every fixed kk.

Choose u0,n→u0u_{0,n} \to u_0 in VpV_p with u0,n∈D(L)∩⋂kHku_{0,n} \in D(\mathcal{L}) \cap\bigcap_k H^k. Such an approximation is obtained by smooth Fourier cutoffs away from zero and infinity. Convergence in Hlog⁡H_{\log} follows by dominated convergence. For the Lp+1L^{p+1} component, the high cutoff is an approximate identity; the low cutoff tends strongly to zero on every finite LrL^r, 1<r<∞1 < r < \infty. The latter follows first for L1∩LrL^1 \cap L^r functions by scaling of the convolution kernel, and then by density and uniform LrL^r bounds.

Consider

i∂tun=Lun+μJnF(Jnun),un(0)=u0,n.(34)i\partial_t u_n = \mathcal{L}u_n + \mu J_nF(J_nu_n), \qquad u_n(0) = u_{0,n}. \tag*{(34)}

For fixed nn, its nonlinearity is locally Lipschitz on L2L^2 and is bounded on each L2L^2 ball. The interaction-variable argument gives mass conservation because JnJ_n is self-adjoint:

⟨JnF(Jnun),un⟩=∫∣Jnun∣p+1∈R.\langle J_nF(J_nu_n),u_n\rangle= \int|J_nu_n|^{p+1} \in\mathbb{R}.

Thus the approximate solution is global in L2L^2. The forcing is band-limited, belongs to L1L^1 for p≥2p \ge2, and hence belongs to D(L)D(\mathcal{L}) by the low-frequency bound in (20). This gives graph-domain persistence for each fixed nn. Its conserved regularized energy is

En=12K+(un)−12K−(un)+μp+1∥Jnun∥p+1p+1.(35)E_n = \frac{1}{2}K_+(u_n) - \frac{1}{2}K_-(u_n) + \frac{\mu}{p+1}\|J_nu_n\|_{p+1}^{p+1}. \tag*{(35)}

The initial EnE_n, masses, and K−(u0,n)K_-(u_{0,n}) are uniformly bounded.

Put Bn=K−(un)1/2B_n = K_-(u_n)^{1/2} and Xn=∥Jnun∥p+1p+1X_n = \|J_nu_n\|_{p+1}^{p+1}. The weighted Duhamel estimate uses ∥JnF(Jnun)∥∞≤C∥F(Jnun)∥1\|J_nF(J_nu_n)\|_\infty\le C\|F(J_nu_n)\|_1. Interpolation (32), with ∥Jnun∥2≤∥un∥2\|J_nu_n\|_2 \le\|u_n\|_2, gives

Bn(t)≤Bn(0)+C∫0t(1+Xn(s))α ds,K+(un)+Xn≤C(1+Bn2).B_n(t) \le B_n(0) + C\int_0^t (1+X_n(s))^\alpha\,ds,\qquad K_+(u_n) + X_n \le C(1+B_n^2).

For a precise Gronwall argument, bound BnB_n by its integral envelope ZnZ_n, which satisfies Zn′≤C(1+Zn2)α≤C(1+Zn)Z_n' \le C(1+Z_n^2)^\alpha\le C(1+Z_n). Consequently, on every finite interval [−T,T][-T,T],

sup⁡nsup⁡∣t∣≤T(∥un(t)∥Hlog⁡2+Xn(t))≤CT.(36)\sup_n\sup_{|t|\le T}\left(\|u_n(t)\|_{H_{\log}}^2 + X_n(t)\right) \le C_T. \tag*{(36)}

with exponential dependence on TT.

The equations bound (un)t(u_n)_t in L∞(−T,T;Vp∗)L^\infty(-T,T;V_p^*): the kinetic term is bounded by the form norm and

∣⟨JnF(Jnun),ϕ⟩∣≤∥Jnun∥p+1p∥Jnϕ∥p+1≤CT∥ϕ∥Vp.|\langle J_nF(J_nu_n),\phi\rangle| \le\|J_nu_n\|_{p+1}^p\|J_n\phi\|_{p+1} \le C_T\|\phi\|_{V_p}.

To obtain local strong compactness, note that the high-frequency logarithmic space with norm

∫log⁡(2+∣ξ∣2)∣f^(ξ)∣2 dξ\int\log(2+|\xi|^2)|\widehat{f}(\xi)|^2\,d\xi

is compactly embedded in L2L^2 on sets of functions supported in a fixed bounded region. Indeed, its Fourier tails are bounded by C/log⁡(2+R2)C/\log(2+R^2), giving uniform translation continuity; bounded support gives spatial tightness. Multiplication by a fixed smooth compact cutoff is bounded in this high-frequency logarithmic space.

After such localization, the time derivatives are bounded in H−kH^{-k} for k>N/2k>N/2. For the low-frequency part of Lun\mathcal{L}u_n, use

∫∣ξ∣<1∣log⁡∣ξ∣2∣∣u^n∣ dξ≤K−(un)1/2(∫∣ξ∣<1∣log⁡∣ξ∣2∣ dξ)1/2;\int_{|\xi|<1} \left|\log|\xi|^2\right| |\widehat{u}_n|\,d\xi\le K_-(u_n)^{1/2}\left(\int_{|\xi|<1} \left|\log|\xi|^2\right|\,d\xi\right)^{1/2};

its inverse transform is bounded locally. The high-frequency part is bounded in H−kH^{-k} by (36). The nonlinear term has the same negative-Sobolev bound, since Hk↪Lp+1H^k\hookrightarrow L^{p+1}. The Aubin–Lions–Simon compactness theorem [3] therefore gives

un→uin L2((−T,T)×BR)(37)u_n\to u\quad\text{in }L^2((-T,T)\times B_R) \tag*{(37)}

after extraction, for every finite T,RT,R.

Also,

∥(I−Jn)un∥L∞(−T,T;L2)2≤CTlog⁡(n2)⟶0.(38)\|(I-J_n)u_n\|_{L^\infty(-T,T;L^2)}^2\le\frac{C_T}{\log(n^2)}\longrightarrow0. \tag*{(38)}

Thus Jnun→uJ_nu_n\to u locally in L2L^2 and almost everywhere. Its global Lp+1L^{p+1} bound identifies the weak limit

F(Jnun)⇀F(u)in L(p+1)/p((−T,T)×RN).F(J_nu_n)\rightharpoonup F(u)\quad\text{in }L^{(p+1)/p}((-T,T)\times\mathbb{R}^N).

Passing to the limit in (34), using Jnϕ→ϕJ_n\phi\to\phi in Lp+1L^{p+1} for test functions, gives the equation in Vp∗V_p^*. Weak lower semicontinuity gives u∈L∞(−T,T;Vp)u\in L^\infty(-T,T;V_p) and (30). Equicontinuity of pairings with VpV_p tests and convergence of u0,nu_{0,n} identify the initial trace as u0u_0.

Finally, the Gelfand triple Vp⊂L2⊂Vp∗V_p\subset L^2\subset V_p^* gives u∈C([−T,T];L2)u\in C([-T,T];L^2) and its norm chain rule. Taking the real part,

ddt∥u∥22=2Re⁡⟨−iLu−iμF(u),u⟩=0,\frac{d}{dt}\|u\|_2^2=2\operatorname{Re}\langle-i\mathcal{L}u-i\mu F(u),u\rangle=0,

since l(u,u)l(u,u) and ∫∣u∣p+1\int|u|^{p+1} are real. A diagonal extraction over increasing TT gives the global solution.

Proposition 4.2 (Conditional uniqueness in the energy class). Two solutions in (29) with the same initial data agree on any common interval on which both belong to Ltp−1Lx∞L_t^{p-1}L_x^\infty.

Proof. The pointwise estimate ∣F(z)−F(w)∣≤Cp(∣z∣p−1+∣w∣p−1)∣z−w∣|F(z)-F(w)|\le C_p(|z|^{p-1}+|w|^{p-1})|z-w| and cancellation of the real kinetic form give

ddt∥u−v∥22≤Cp∣μ∣(∥u∥∞p−1+∥v∥∞p−1)∥u−v∥22.\frac{d}{dt}\|u-v\|_2^2\le C_p|\mu|(\|u\|_\infty^{p-1}+\|v\|_\infty^{p-1})\|u-v\|_2^2.

The chain rule can be taken in Vp∗V_p^*; alternatively the assumed amplitude integrability gives a mild L2L^2 formulation. Gronwall’s inequality finishes the proof. The extra space-time bound is not a consequence of Theorem 4.1.

Focusing estimates and limitations of the energy method

In this section μ=−κ<0\mu=-\kappa<0. For strong solutions from Proposition 2.3, write E0=E−κ(u0)E_0=E_{-\kappa}(u_0) and M=∥u0∥22M=\|u_0\|_2^2.

Proposition 5.1 (Low-frequency and kinetic bounds). For p≥2p\ge2, whenever the indicated quantities are finite,

K−(u(t))1/2≤K−(u0)1/2+CNκM∫0t∥u(s)∥∞p−2 ds,(39)K_{-}(u(t))^{1/2}\le K_{-}(u_0)^{1/2}+C_N\kappa M\int_0^t\|u(s)\|_{\infty}^{p-2}\,ds, \tag*{(39)}
K+(u(t))≤2∣E0∣+K−(u(t))+2κMp+1∥u(t)∥∞p−1.(40)K_{+}(u(t))\le2|E_0|+K_{-}(u(t))+\frac{2\kappa M}{p+1}\|u(t)\|_{\infty}^{p-1}. \tag*{(40)}

Proof. Combine (31) with (19). Rearrangement of the focusing energy and ∥u∥p+1p+1≤M∥u∥∞p−1\|u\|_{p+1}^{p+1}\le M\|u\|_{\infty}^{p-1} give the second inequality. These estimates depend on amplitude information; they do not provide it independently.

Proposition 5.2 (Failure of energy coercivity at every mass). For every M>0M>0, p>1p>1, and κ>0\kappa>0, the focusing energy is unbounded below on smooth functions of mass MM, even among functions with K−=0K_-=0. No estimate

∥u∥p+1p+1≤C(M)(1+K+(u))\|u\|_{p+1}^{p+1}\le C(M)(1+K_+(u))

holds on that class.

Proof. Choose a nonzero Schwartz function φ\varphi with Fourier support in 1<∣ξ∣<21<|\xi|<2 and mass MM. For λ≥1\lambda\ge1, put φλ(x)=λN/2φ(λx)\varphi_\lambda(x)=\lambda^{N/2}\varphi(\lambda x). Then

M(φλ)=M,K−(φλ)=0,M(\varphi_\lambda)=M,\qquad K_-(\varphi_\lambda)=0,
K+(φλ)=K+(φ)+2Mlog⁡λ,∥φλ∥p+1p+1=λN(p−1)/2∥φ∥p+1p+1.K_+(\varphi_\lambda)=K_+(\varphi)+2M\log\lambda,\qquad \|\varphi_\lambda\|_{p+1}^{p+1}=\lambda^{N(p-1)/2}\|\varphi\|_{p+1}^{p+1}.

Therefore

E−κ(φλ)=12K+(φ)+Mlog⁡λ−κp+1λN(p−1)/2∥φ∥p+1p+1⟶−∞.E_{-\kappa}(\varphi_\lambda)=\frac{1}{2}K_+(\varphi)+M\log\lambda-\frac{\kappa}{p+1}\lambda^{N(p-1)/2}\|\varphi\|_{p+1}^{p+1}\longrightarrow-\infty.

Polynomial concentration dominates logarithmic kinetic growth, also contradicting the proposed inequality.

Remark 5.3. Proposition 5.2 is not a blow-up theorem. It also does not contradict Theorem 3.3, which assumes smallness of an HmH^m norm, not smallness of mass alone.

Dilation and variance identities

The identities below are conditional on sufficient regularity and moment control to differentiate the displayed pairings and pass the commutator or integration-by-parts limits. They are not asserted for every energy-class weak solution. In particular, no propagation of these moment assumptions is used to claim a blow-up theorem.

Let

D(t)=Im⁡∫u‾x⋅∇u dx.(41)D(t)=\operatorname{Im}\int\overline{u}x\cdot\nabla u\,dx. \tag*{(41)}

For the focusing equation, the exact identity is

D′(t)=2M−κN(p−1)p+1∥u(t)∥p+1p+1.(42)D'(t)=2M-\frac{\kappa N(p-1)}{p+1}\|u(t)\|_{p+1}^{p+1}. \tag*{(42)}

Indeed, the linear commutator contributes

∫ξ⋅∇ξlog⁡∣ξ∣2∣u^∣2 dξ=2M.\int\xi\cdot\nabla_{\xi}\log|\xi|^{2} |\widehat{u}|^{2}\,d\xi= 2M.

Writing ρ=∣u∣2\rho= |u|^{2}, the nonlinear contribution is

κ∫ρx⋅∇(ρ(p−1)/2) dx=κ(p−1)p+1∫x⋅∇(∣u∣p+1) dx=−κN(p−1)p+1∥u∥p+1p+1.\kappa\int\rho x \cdot\nabla\left(\rho^{(p-1)/2}\right)\,dx = \frac{\kappa(p-1)}{p+1}\int x\cdot\nabla\left(|u|^{p+1}\right)\,dx = -\frac{\kappa N(p-1)}{p+1}\|u\|_{p+1}^{p+1}.

Integration gives the space-time relation

κN(p−1)p+1∫0T∥u(t)∥p+1p+1 dt=2MT+D(0)−D(T).(43)\frac{\kappa N(p-1)}{p+1}\int_{0}^{T}\|u(t)\|_{p+1}^{p+1}\,dt = 2MT+D(0)-D(T). \tag*{(43)}

For V(t)=∫∣x∣2∣u(t,x)∣2 dxV(t)=\int|x|^{2}|u(t,x)|^{2}\,dx, the variance identity, when justified, is

V′(t)=−4Im⁡∫u^(t,ξ)ξ∣ξ∣2⋅∇ξu^(t,ξ) dξ.(44)V'(t)=-4\operatorname{Im}\int\widehat{u}(t,\xi)\frac{\xi}{|\xi|^{2}}\cdot\nabla_{\xi}\widehat{u}(t,\xi)\,d\xi. \tag*{(44)}

This follows from Plancherel and ∇ξlog⁡∣ξ∣2=2ξ/∣ξ∣2\nabla_{\xi}\log|\xi|^{2}=2\xi/|\xi|^{2}; the local nonlinearity makes no contribution to V′V'. For comparison, D=−Im⁡∫u ξ⋅∇ξu^D=-\operatorname{Im}\int u\,\xi\cdot\nabla_{\xi}\widehat{u}. Thus V′=4DV'=4D, the relation used for the classical Laplacian, is absent. If xuxu and ∣D∣−1u|D|^{-1}u belong to L2L^{2}, the right side of (44) is absolutely controlled by

∣V′(t)∣≤4∥xu(t)∥2∥∣D∣−1u(t)∥2.|V'(t)|\leq4\|xu(t)\|_{2}\||D|^{-1}u(t)\|_{2}.

Conditional lower bounds on finite-time blow-up

Here a blow-up rate means a necessary lower bound on the Sobolev norm assuming the maximal lifespan is finite. No upper bound, asymptotic equivalence, or singularity construction is inferred.

Theorem 6.1 (Mass-based lower bound). Assume (12), μ≠0\mu\ne0, and u0∈Hm∖{0}u_{0}\in H^{m}\setminus\{0\}. If T+<∞T_{+}<\infty, then, for t<T+t<T_{+},

∥∣D∣mu(t)∥2≥cN,m,p∣μ∣−2m/[N(p−1)]∥u0∥2−(2m−N)/N(T+−t)−2m/[N(p−1)].(45)\||D|^{m}u(t)\|_{2}\geq c_{N,m,p}|\mu|^{-2m/[N(p-1)]}\|u_{0}\|_{2}^{-(2m-N)/N}(T_{+}-t)^{-2m/[N(p-1)]}. \tag*{(45)}

Proof. Put Y(t)=∥∣D∣mu(t)∥2Y(t)=\||D|^{m}u(t)\|_{2}, A=∥u0∥2>0A=\|u_{0}\|_{2}>0, and θ0=N/(2m)\theta_{0}=N/(2m). The interaction variable is C1(Hm)C^{1}(H^{m}), so YY is locally absolutely continuous. Lemma 2.1 gives

Y′≤C∣μ∣∥u∥∞p−1Y.Y'\leq C|\mu|\|u\|_{\infty}^{p-1}Y.

The Gagliardo–Nirenberg inequality and mass conservation imply

∥u∥∞≤CA1−θ0Yθ0,Y′≤KY1+q0,q0=θ0(p−1),K=C∣μ∣A(1−θ0)(p−1).\|u\|_{\infty}\leq CA^{1-\theta_{0}}Y^{\theta_{0}},\qquad Y'\leq KY^{1+q_{0}},\qquad q_{0}=\theta_{0}(p-1),\qquad K=C|\mu|A^{(1-\theta_{0})(p-1)}.

Since mass is fixed, (16) implies Y(t)→∞Y(t)\to\infty as t↑T+t\uparrow T_{+}. Integrating the inequality for Y−q0Y^{-q_{0}} gives Y(t)−q0≤q0K(T+−t)Y(t)^{-q_{0}}\leq q_{0}K(T_{+}-t), which is (6.1).

Theorem 6.2 (Energy-refined focusing lower bound). Assume (12), μ=−κ<0\mu=-\kappa<0, and 0≠u0∈Hm∩D(L)0\ne u_0\in H^m\cap D(L). If T+<∞T_+<\infty, then for all tt sufficiently close to T+T_+,

∥∣D∣mu(t)∥2≥c(T+−t)−γ[log⁡eT+−t]−δ,(46)\left\lVert|D|^m u(t)\right\rVert_2 \ge c(T_+-t)^{-\gamma}\left[\log\frac{e}{T_+-t}\right]^{-\delta}, \tag*{(46)}

where

γ=m(p+1)N(p−1)−12,δ=mN−12.(47)\gamma=\frac{m(p+1)}{N(p-1)}-\frac{1}{2},\qquad\delta=\frac{m}{N}-\frac{1}{2}. \tag*{(47)}

The constant and the closeness threshold depend on the parameters, mass, and energy. This theorem presupposes, rather than proves, T+<∞T_+<\infty.

Proof. Let Y=∥∣D∣mu∥2Y=\left\lVert|D|^m u\right\rVert_2, M=∥u0∥22>0M=\lVert u_0\rVert_2^2>0, and X=∥u∥p+1p+1X=\lVert u\rVert_{p+1}^{p+1}. Energy conservation and K−≥0K_-\ge0 imply

X≤p+12κ(K++2∣E0∣).(48)X\le\frac{p+1}{2\kappa}(K_++2|E_0|). \tag*{(48)}

For ∣ξ∣≥1|\xi|\ge1, log⁡∣ξ∣2≤m−1log⁡(1+∣ξ∣2m)\log|\xi|^2\le m^{-1}\log(1+|\xi|^{2m}). Jensen’s inequality with probability measure M−1∣u^∣2 dξM^{-1}|\widehat{u}|^2\,d\xi gives

K+(u)≤Mmlog⁡(1+Y2M).(49)K_+(u)\le\frac{M}{m}\log\left(1+\frac{Y^2}{M}\right). \tag*{(49)}

Thus X≤Clog⁡(e+Y)X\le C\log(e+Y).

Interpolation between Lp+1L^{p+1} and H˙m\dot H^m gives

∥u∥∞≤C∥u∥p+11−θYθ,θ=Nm(p+1)−N(p−1)/2∈(0,1).\lVert u\rVert_\infty\le C\lVert u\rVert_{p+1}^{1-\theta}Y^\theta,\qquad \theta=\frac{N}{m(p+1)-N(p-1)/2}\in(0,1).

Consequently

∥u∥∞p−1≤C[log⁡(e+Y)]aYb,a=(p−1)(1−θ)p+1,b=(p−1)θ>0.\lVert u\rVert_\infty^{p-1}\le C[\log(e+Y)]^aY^b,\qquad a=\frac{(p-1)(1-\theta)}{p+1},\qquad b=(p-1)\theta>0.

The same interaction-variable argument as in Theorem 6.1 gives, almost everywhere,

Y′≤CY1+b[log⁡(e+Y)]a.(50)Y'\le CY^{1+b}[\log(e+Y)]^a. \tag*{(50)}

Since Y(t)→∞Y(t)\to\infty, integration toward T+T_+ yields

C(T+−t)≥∫Y(t)∞dηη1+b[log⁡(e+η)]a≥cY(t)−b[log⁡(e+Y(t))]−a.C(T_+-t)\ge\int_{Y(t)}^\infty\frac{d\eta}{\eta^{1+b}[\log(e+\eta)]^a}\ge cY(t)^{-b}[\log(e+Y(t))]^{-a}.

The last inequality follows already by integrating over [Y,2Y][Y,2Y] for large YY. Inverting this increasing-power logarithmic relation gives

Y(t)≥c(T+−t)−1/b[log⁡eT+−t]−a/b.Y(t)\ge c(T_+-t)^{-1/b}\left[\log\frac{e}{T_+-t}\right]^{-a/b}.

Direct computation shows 1/b=m(p+1)/[N(p−1)]−1/21/b=m(p+1)/[N(p-1)]-1/2 and a/b=m/N−1/2a/b=m/N-1/2.

Corollary 6.3 (Three-dimensional cubic example). For N=3N=3, p=3p=3, m=2m=2, a focusing strong solution with a finite maximal H2H^2 lifespan necessarily satisfies, near that time,

∥Δu(t)∥2≥c(T+−t)−5/6[log⁡eT+−t]−1/6.\lVert\Delta u(t)\rVert_2\ge c(T_+-t)^{-5/6}\left[\log\frac{e}{T_+-t}\right]^{-1/6}.

Without the operator-domain/energy assumption, the mass-based bound still gives ∥Δu(t)∥2≥c(T+−t)−2/3\lVert\Delta u(t)\rVert_2\ge c(T_+-t)^{-2/3}.

Scope, unresolved questions, and comparison with fractional NLS

Theorems 2.2 and 3.3 are uniqueness statements in Sobolev classes. Theorem 4.1 is a global existence result in a larger energy class; its construction does not establish uniqueness there or global persistence of a local HmH^m solution. Proposition 4.2 states the additional space-time condition under which energy-class uniqueness does follow.

For focusing data of arbitrary size, the paper does not prove global Sobolev existence or finite-time blow-up. Negative energy alone is not shown to force blow-up. The rate theorems assume a finite maximal lifespan and cannot be used as evidence that such a lifespan occurs.

One tempting comparison is the family

Ls=(−Δ)s−Is,s>0,\mathcal{L}_s = \frac{(-\Delta)^s-I}{s}, \qquad s>0,

whose symbols converge pointwise to log⁡∣ξ∣2\log|\xi|^2 away from zero. Its linear dilation contribution is

∫ξ⋅∇ξ(∣ξ∣2s−1s)∣u^∣2 dξ=2∥∣D∣su∥22.\int\xi\cdot\nabla_{\xi}\left(\frac{|\xi|^{2s}-1}{s}\right)|\hat{u}|^2\,\mathrm{d}\xi= 2\left\lVert|D|^s u\right\rVert_2^2.

On suitable fixed test functions this tends to 2M2M, not a positive-order derivative norm. The fractional blow-up theorems in [5, 6] assume 1/2<s<11/2<s<1 and do not apply to logarithmic dispersion. Passing to a formal operator limit therefore does not transfer their blow-up conclusions.

The remaining occurrence problem is to find admissible data and a mechanism proving T+<∞T_+<\infty, for example a justified localized functional with a closed finite-time forcing inequality, or a constructed singularity profile with error control. Neither mechanism is supplied here. This is a statement about the scope of the present arguments, not a claim of nonexistence of blow-up or an exhaustive classification of the literature.

References

  1. [1]H. Chen and T. Weth, The Dirichlet problem for the logarithmic Laplacian, Commun. Partial Differential Equations 44 (2019), 1100–1139. doi:10.1080/03605302.2019.1611851.DOI
  2. [2]M. Keel and T. Tao, Endpoint Strichartz estimates, Amer. J. Math. 120 (1998), 955–980. doi:10.1353/ajm.1998.0039.DOI
  3. [3]J. Simon, Compact sets in the space L^p(0, T; B), Ann. Mat. Pura Appl. (4) 146 (1986), 65–96. doi:10.1007/BF01762360.
  4. [4]T. Tao, Nonlinear Dispersive Equations: Local and Global Analysis, CBMS Regional Conference Series in Mathematics, vol. 106, American Mathematical Society, 2006. doi:10.1090/cbms/106.DOI
  5. [5]T. Boulenger, D. Himmelsbach, and E. Lenzmann, Blowup for fractional NLS, J. Funct. Anal. 271 (2016), 2569–2603. doi:10.1016/j.jfa.2016.08.011.DOI
  6. [6]V. D. Dinh, Blow-up criteria for fractional nonlinear Schrödinger equation, arXiv:1808.07368. arxiv.org/abs/1808.07368.arxiv.org/abs/1808.07368

Paper details

vlnc:W_euq88pyj
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents