betaFeedback

WorksMathematics

Nonradial Finite-Time Blow-up for the Mass-Critical Inhomogeneous Biharmonic NLS

Abstract

We consider the focusing mass-critical inhomogeneous biharmonic nonlinear Schrödinger equation i∂tu=Δ2u−∣x∣−b∣u∣8−2bNu,0<b<min⁡{4,N/2}. i\partial_tu=Δ^2u-|x|^{-b}|u|^{\frac{8-2b}{N}}u, \quad 0<b<\min\{4,N/2\}. We prove that every H2(RN)H^2(\mathbb R^N) solution with negative energy blows up in finite time, both forward and backward. No radial symmetry or finite-variance assumption is required. In particular, this rules out the global grow-up scenario in the previously known negative-energy blow-up alternative and establishes finite-time blow-up in the genuinely nonradial setting. The proof relies on a new localized virial argument exploiting the spatial inhomogeneity to control the localization errors.

Introduction and main result

We consider the focusing mass-critical inhomogeneous biharmonic nonlinear Schrödinger equation

\begin{cases} i\partial_tu=\Delta^2u-|x|^{-b}|u|^{\frac{8-2b}{N}}u,\\ u(0)=u_0\in H^2(\mathbb{R}^N), \end{cases} \qquad0<b<\min\{4,N/2\}. \tag*{(1)}\end{cases}

Equation (1) is L2L^2-critical (mass-critical) since it is invariant under the scaling

uλ(t,x)=λN/2u(λ4t,λx),u_\lambda(t,x)=\lambda^{N/2}u(\lambda^4t,\lambda x),

which also preserves the L2L^2-norm. Along the maximal H2H^2 solution, the mass

M(u)=∥u(t)∥L22M(u)=\|u(t)\|_{L^2}^2

and the energy

E(u)=12∥Δu(t)∥L22−N2N+8−2b∫RN∣x∣−b∣u(t,x)∣2N+8−2bN dxE(u)=\frac{1}{2}\|\Delta u(t)\|_{L^2}^2-\frac{N}{2N+8-2b}\int_{\mathbb{R}^N}|x|^{-b}|u(t,x)|^{\frac{2N+8-2b}{N}}\,dx

are conserved.

Fourth-order nonlinear Schrödinger equations were introduced by Karpman [14] and Karpman–Shagalov [15] as higher-order dispersive models arising in nonlinear optics. Since then, biharmonic Schrödinger equations have attracted considerable attention as natural higher-order counterparts of the classical nonlinear Schrödinger equation; see, among many others, [2, 4, 12, 20]. The inhomogeneous factor ∣x∣−b|x|^{-b} introduces a spatially decaying nonlinear interaction and leads to a substantially different dynamics from the translation-invariant problem. Local and global well-posedness, stability, and scattering for inhomogeneous biharmonic Schrödinger equations have been studied in [5, 7, 11, 13, 17, 21] and the references therein.

Our concern here is finite-time blow-up at the mass-critical power. For the homogeneous biharmonic NLS, a fundamental contribution is due to Boulenger and Lenzmann [4], who developed localized virial and Riesz-bivariance methods adapted to the fourth-order flow and established a general blow-up theory in the mass-critical and mass-supercritical regimes. In contrast with the classical NLS, however, the biharmonic equation does not possess a variance identity whose concavity directly forces finite-time blow-up. In particular, for the pure mass-critical biharmonic flow, their argument yields for radial negative-energy solutions only the alternative between finite-time blow-up and global grow-up.

Finite-time blow-up for inhomogeneous fourth-order Schrödinger equations was first investigated by Cho, Ozawa and Wang [8], who considered the special mass-critical nonlinearity ∣x∣−2∣u∣4Nu|x|^{-2}|u|^{\frac{4}{N}}u. Under suitable dimensional restrictions, they proved finite-time blow-up for negative-energy solutions assuming either radial symmetry or finite variance. For the same b=2b=2 mass-critical model, the second author [10] subsequently exploited the spatial decay of the inhomogeneous nonlinearity to remove these assumptions and proved finite-time blow-up for nonradial negative-energy solutions in the presence of a nontrivial lower-order dispersion term. In the intercritical regime, Saanouni and Ghanmi [22] established blow-up criteria and strong instability of standing waves, while, more recently, Scarpelli and Hespanha [23] obtained radial and nonradial blow-up results for the energy-critical inhomogeneous biharmonic equation.

The result most closely related to the present work is due to Bai, Majdoub and Saanouni [1], who considered the general mass-critical equation

i∂tu−Δ2u+νΔu=−∣x∣−b∣u∣8−2bNu,0<b<min⁡{4,N/2},ν≥0.i\partial_t u-\Delta^2u+\nu\Delta u=-|x|^{-b}|u|^{\frac{8-2b}{N}}u,\qquad0<b<\min\{4,N/2\},\qquad\nu\ge0.

When ν>0\nu>0, they proved finite-time blow-up for arbitrary negative-energy H2H^2 solutions, without radial symmetry. For the pure biharmonic case ν=0\nu=0, however, their argument yields only the alternative

T+<∞or∥Δu(t)∥L2≳t2for t≫1.T_+<\infty\quad\text{or}\quad\|\Delta u(t)\|_{L^2}\gtrsim t^2\quad\text{for }t\gg1.

Thus, even in the presence of the spatial inhomogeneity, the possibility of a global negative-energy solution undergoing unbounded grow-up remained open.

This phenomenon has no analogue in the corresponding mass-critical inhomogeneous second-order NLS. Cardoso and Farah [6] proved that negative energy forces finite-time blow-up for arbitrary H1H^1 solutions, without radial symmetry or finite variance. It is therefore natural to ask whether the inhomogeneity is likewise strong enough to eliminate the grow-up branch for the pure biharmonic flow. The main result of this paper answers this question affirmatively: we rule out the global grow-up scenario in [1] and prove that every negative-energy H2H^2 solution of the mass-critical inhomogeneous pure biharmonic NLS blows up in finite time, without any radial or finite-variance assumption.

We recall the local theory needed to formulate the result. By Liu and Zhang [17], for every u0∈H2(RN)u_0\in H^2(\mathbb{R}^N) and

0<b<min⁡{4,N/2},0<b<\min\{4,N/2\},

equation (1) admits a unique maximal solution

u∈C((−T−,T+);H2(RN)),T−,T+∈(0,∞].u\in C\bigl((-T_-,T_+);H^2(\mathbb{R}^N)\bigr),\qquad T_-,T_+\in(0,\infty].

The mass and energy are conserved on the maximal lifespan, and the usual blow-up alternative holds:

T+<∞⟹lim⁡t↑T+∥u(t)∥H2=∞,T_+<\infty\quad\Longrightarrow\quad\lim_{t\uparrow T_+}\|u(t)\|_{H^2}=\infty,

with the analogous statement backward in time.

We are now ready to state our main result.

Theorem 1.1 (Nonradial finite-time blow-up). Let N≥1N \ge1 and 0<b<min⁡{4,N/2}0 < b < \min\{4,N/2\}. Let u0∈H2(RN)u_0 \in H^2(\mathbb{R}^N), and let

u∈C((−T−,T+);H2(RN))u \in C((-T_-,T_+);H^2(\mathbb{R}^N))

be the maximal solution of (1). If E(u0)<0E(u_0) < 0, then

T+<∞andT−<∞.T_+ < \infty\quad\text{and} \quad T_- < \infty.

In particular,

lim⁡t↑T+∥u(t)∥H2=∞,lim⁡t↓−T−∥u(t)∥H2=∞.\lim_{t\uparrow T_+}\|u(t)\|_{H^2}=\infty,\qquad\lim_{t\downarrow-T_-}\|u(t)\|_{H^2}=\infty.

Remark 1.2 (On the range of bb). The restriction

0<b<min⁡{4,N/2}0 < b < \min\{4,N/2\}

has two different origins. The assumption b>0b>0 gives the strict inequality 1−b4<11-\frac{b}{4}<1, while b<4b<4 ensures 1−b4>01-\frac{b}{4}>0. The exponent θ=1−b4∈(0,1)\theta=1-\frac{b}{4}\in(0,1) will play an important role in the weighted interpolation argument (see Proposition 3.1). On the other hand, the additional condition b<N/2b<N/2 is inherited from the available H2H^2 local well-posedness theory for the inhomogeneous biharmonic equation; see Liu–Zhang [17]. In particular, the restriction b<N/2b<N/2 is not intrinsic to the blow-up mechanism developed here.

Remark 1.3 (The homogeneous limit). The argument developed in this paper is genuinely inhomogeneous. Indeed, when b=0b=0, equation (1) reduces to the homogeneous mass-critical biharmonic NLS, while the crucial gain

1−b4<11-\frac{b}{4}<1

degenerates to equality. Consequently, the absorption mechanism used in our proof no longer closes, and Theorem 1.1 does not extend to the homogeneous problem by simply letting b↓0b\downarrow0.

During the preparation of this manuscript, we learned from Leonard Dinh that he has independently obtained finite-time blow-up for radial negative-energy solutions of the homogeneous mass-critical biharmonic NLS (see [16]). This resolves, in the radial setting, the finite-time blow-up problem left open by Boulenger and Lenzmann [4], whose result allowed the alternative of global grow-up. The result in [16] and the present theorem are therefore complementary: his result concerns the homogeneous equation under radial symmetry, whereas our argument exploits the spatial inhomogeneity b>0b>0 to obtain finite-time blow-up without radial symmetry or a finite-variance assumption.

Remark 1.4 (Negative energy). The condition E(u0)<0E(u_0)<0 is a sufficient blow-up criterion rather than a characterization of the blow-up region. Solutions with nonnegative energy may also blow up, and a finer description of the threshold dynamics would require additional variational information, typically involving the ground state and the associated Pohozaev functional. The present result isolates the negative-energy regime, in which finite-time blow-up holds for arbitrary H2H^2 data.

We next briefly describe the main idea behind the proof of Theorem 1.1. The key issue is to strengthen the localized virial mechanism sufficiently to rule out the global grow-up scenario that remains possible in the pure biharmonic case. In the absence of lower-order dispersion, the standard localized virial estimates alone do not force finite-time blow-up. Our approach is inspired by the exponential-weight virial method recently introduced by the authors in [18] for the mass-critical quadratic Schrödinger system without mass resonance. In that setting, a bounded exponential localization retains a weighted kinetic defect and ultimately leads to a superlinear Riccati-type inequality. The present fourth-order, nonradial problem requires additional ideas, however, because of the higher-order dispersion and the localization errors generated by the singular inhomogeneous nonlinearity.

To overcome these difficulties, we introduce a localization which is never exactly quadratic away from the origin. More precisely, we choose a radial weight satisfying

∇ϕR(x)=xexp⁡[−(∣x∣/R)4].\nabla\phi_R(x) = x \exp\left[-\left(|x|/R\right)^4\right].

Its Hessian retains a positive defect at every nonzero radius, which both controls the nonlinear localization error and provides the coercivity needed to estimate the localized virial quantity. The spatial inhomogeneity ∣x∣−b|x|^{-b} enters through a weighted dyadic Gagliardo–Nirenberg estimate with exponent

1−b4<1,1-\frac{b}{4}<1,

allowing the defect term to absorb the nonlinear error. A further weighted interpolation estimate then yields a superlinear Riccati-type differential inequality, which rules out global existence.

Although the exponential localization and Riccati mechanism are motivated by [18], the present argument requires a new weighted localization analysis adapted simultaneously to fourth-order dispersion, the singular inhomogeneity, and the absence of radial symmetry.

As an application of Theorem 1.1, we also obtain strong instability by finite-time blow-up of standing waves. Strong instability for fourth-order Schrödinger equations has been studied in the homogeneous setting by Bonheure, Castéras, Gou and Jeanjean [3], and for inhomogeneous fourth-order equations in the intercritical regime by Saanouni and Ghanmi [22]. At the mass-critical exponent considered here, the conclusion takes a particularly simple form. Indeed, every nontrivial standing-wave profile has zero energy as a consequence of the Pohozaev identities. An arbitrarily small increase of its amplitude therefore produces negative-energy initial data, to which Theorem 1.1 applies.

Corollary 1.5 (Strong instability of standing waves). Let N≥1N \ge1 and 0<b<min⁡{4,N/2}0 < b < \min\{4,N/2\}. Let ω∈R\omega\in\mathbb{R}, and let 0≠Q∈H2(RN)0 \ne Q \in H^2(\mathbb{R}^N) solve

Δ2Q+ωQ−∣x∣−b∣Q∣8−2bNQ=0.(2)\Delta^2 Q + \omega Q - |x|^{-b}|Q|^{\frac{8-2b}{N}}Q = 0. \tag*{(2)}

Then the standing wave

u(t,x)=eiωtQ(x)u(t,x)=e^{i\omega t}Q(x)

is strongly unstable by finite-time blow-up. More precisely, for every ε>0\varepsilon> 0 there exists u0∈H2(RN)u_0 \in H^2(\mathbb{R}^N) such that

∥u0−Q∥H2<ε,\|u_0-Q\|_{H^2}<\varepsilon,

and the corresponding solution of (1.1) blows up in finite time.

Remark 1.6. The conclusion of Corollary 1.5 holds for every nontrivial standing wave. In particular, no ground-state assumption, radial symmetry, or variational minimizing property is required.

Remark 1.7. Any nontrivial solution QQ of (1.2) necessarily satisfies ω>0\omega> 0. This is an immediate consequence of the Pohozaev identities for the stationary equation; see Lemma 7.1.

Throughout the remainder of the paper, we set

p:=8−2bN,q:=p+2=2N+8−2bN.(3)p := \frac{8-2b}{N}, \qquad q := p+2 = \frac{2N+8-2b}{N}. \tag*{(3)}

The conserved energy takes the form

E(u)=12∥Δu∥L22−1q∫RN∣x∣−b∣u∣q dx.E(u)=\frac{1}{2}\|\Delta u\|_{L^2}^{2}-\frac{1}{q}\int_{\mathbb{R}^N}|x|^{-b}|u|^q\,dx.

We also denote

m:=M(u0)=∥u(t)∥L22.m:=M(u_0)=\|u(t)\|_{L^2}^{2}.

The paper is organized as follows. In Section 2, we introduce the exponential multiplier and the associated defect quantities. Section 3 is devoted to the weighted interpolation estimates that control the inhomogeneous nonlinear and lower-order error terms. In Section 4, we derive the coercive localized virial inequality, while Section 5 establishes the virial–defect estimate needed to close the argument. Section 6 contains the proof of the main finite-time blow-up theorem. Finally, in Section 7, we prove the strong instability of standing waves.

The exponential multiplier and its defect

For R>0R>0, set r=∣x∣r=|x|, s=r/Rs=r/R, and

h(s):=e−s4.h(s):=e^{-s^4}.

We define the radial function ϕR\phi_R by

ϕR′(r)=rh(r/R),ϕR(0)=0.\phi_R'(r)=rh(r/R), \qquad\phi_R(0)=0.

Equivalently,

∇ϕR(x)=xh(∣x∣/R).\nabla\phi_R(x)=xh(|x|/R).

Associated with this multiplier is the defect

ωR(x):=1−h(∣x∣/R)=1−e−∣x∣4/R4.\omega_R(x):=1-h(|x|/R)=1-e^{-|x|^4/R^4}.

The elementary estimates

cmin⁡{∣x∣4R4,1}≤ωR(x)≤min⁡{∣x∣4R4,1}c\min\left\{\frac{|x|^4}{R^4},1\right\}\leq\omega_R(x)\leq\min\left\{\frac{|x|^4}{R^4},1\right\}

hold uniformly in R>0R>0. We shall repeatedly use the positive quantity

DR(u):=∫RNωR(x)∣D2u(x)∣F2 dx.D_R(u):=\int_{\mathbb{R}^N}\omega_R(x)|D^2u(x)|_F^2\,dx.

where

∣D2u∣F2:=∑j,k=1N∣∂jku∣2.|D^2u|_F^2:=\sum_{j,k=1}^{N}|\partial_{jk}u|^2.

By Plancherel’s identity,

∫RN∣D2u∣F2dx=∥Δu∥L22.(4)\int_{\mathbb{R}^N}|D^2u|_F^2dx=\|\Delta u\|_{L^2}^2. \tag*{(4)}

We first record the geometry of the multiplier.

Lemma 2.1 (Geometry of the exponential multiplier). The tangential and radial eigenvalues of D2ϕRD^{2}\phi_{R} are respectively

\lambda_{\tan}(r)=h(s),\qquad\lambda_{\rad}(r)=h(s)(1-4s^{4}).

Consequently,

I−D2ϕR≥ωRII-D^{2}\phi_{R}\geq\omega_{R}I

as quadratic forms. Moreover,

ΔϕR=h(s)(N−4s4).\Delta\phi_{R}=h(s)(N-4s^{4}).

Proof. For a radial function,

λtan⁡(r)=ϕR′(r)r=h(s).\lambda_{\tan}(r)=\frac{\phi_{R}'(r)}{r}=h(s).

On the other hand,

ϕR′′(r)=h(s)+sh′(s)=h(s)(1−4s4),\phi_{R}''(r)=h(s)+sh'(s)=h(s)(1-4s^{4}),

which gives the radial eigenvalue. Both eigenvalues are bounded above by h(s)=1−ωRh(s)=1-\omega_{R}, and hence

D2ϕR≤(1−ωR)I.D^{2}\phi_{R}\leq(1-\omega_{R})I.

Therefore

I−D2ϕR≥ωRI.I-D^{2}\phi_{R}\geq\omega_{R}I.

Finally,

ΔϕR=ϕR′′+N−1rϕR′=h(s)(N−4s4).\Delta\phi_{R}=\phi_{R}''+\frac{N-1}{r}\phi_{R}'=h(s)(N-4s^{4}).

The derivatives of ϕR\phi_{R} enjoy bounds adapted to the same defect.

Lemma 2.2 (Derivative bounds). There exists CN>0C_{N}>0 such that

∣∇2ΔϕR∣+∣Δ2ϕR∣≤CNR−2ωR1/2,∣Δ3ϕR∣≤CNR−4.\lvert\nabla^{2}\Delta\phi_{R}\rvert+\lvert\Delta^{2}\phi_{R}\rvert\leq C_{N}R^{-2}\omega_{R}^{1/2},\qquad\lvert\Delta^{3}\phi_{R}\rvert\leq C_{N}R^{-4}.

Proof. By scaling,

ϕR(x)=R2ϕ1(x/R).\phi_{R}(x)=R^{2}\phi_{1}(x/R).

Near the origin,

ϕ1′(r)=re−r4=r−r5+O(r9),\phi_{1}'(r)=re^{-r^{4}}=r-r^{5}+O(r^{9}),

and hence

ϕ1(r)=r22−r66+O(r10).\phi_{1}(r)=\frac{r^{2}}{2}-\frac{r^{6}}{6}+O(r^{10}).

Thus every fourth derivative of ϕ1−∣x∣2/2\phi_{1}-|x|^{2}/2 is O(∣x∣2)O(|x|^{2}) near the origin, while all sixth derivatives remain bounded. Since

ω11/2(x)≍min⁡{∣x∣2,1},\omega_{1}^{1/2}(x)\asymp\min\{|x|^{2},1\},

and all derivatives of the exponential profile decay rapidly at infinity, the corresponding estimates hold globally for R=1R=1. Scaling gives the result.

Weighted interpolation estimates

This section is devoted to the following weighted Gagliardo–Nirenberg estimate.

Proposition 3.1 (Weighted Gagliardo–Nirenberg estimate). Let N≥1N \ge1, 0<b<min⁡{4,N/2}0 < b < \min\{4,N/2\} and p,qp,q be as in (3). Then, for every ε>0\varepsilon> 0, there exists CN,b,ε>0C_{N,b,\varepsilon} > 0 such that, for every u∈H2(RN)u \in H^2(\mathbb{R}^N) and every R>0R > 0,

∫RNωR(x)∣x∣−b∣u(x)∣q dx≤εDR(u)+CN,b,εR−4(∥u∥L22+4p/b+∥u∥L2q).(5)\int_{\mathbb{R}^N} \omega_R(x)|x|^{-b}|u(x)|^q\,dx \le\varepsilon\mathcal{D}_R(u) + C_{N,b,\varepsilon}R^{-4}\left(\|u\|_{L^2}^{2+4p/b}+\|u\|_{L^2}^q\right). \tag*{(5)}

Proof. Set

θ:=1−b4∈(0,1),γ:=q2−θ=p2+b4.(6)\theta:= 1-\frac{b}{4}\in(0,1),\qquad\gamma:= \frac{q}{2}-\theta=\frac{p}{2}+\frac{b}{4}. \tag*{(6)}

Since q=p+2q=p+2 and Np=8−2bNp=8-2b, we have

θ=N(q−2)8=Np8.\theta=\frac{N(q-2)}{8}=\frac{Np}{8}.

Step 1. Annular localization and control of the localized H2H^2 norm.

For j∈Zj\in\mathbb{Z}, set

ρj:=2jR,Aj:={x∈RN:ρj<∣x∣<2ρj}.\rho_j:=2^jR,\qquad A_j:=\{x\in\mathbb{R}^N:\rho_j<|x|<2\rho_j\}.

Choose enlarged annuli

Aj⋐Aj∗⋐Aj∗∗A_j\Subset A_j^*\Subset A_j^{**}

with uniformly finite overlap, and choose ηj∈Cc∞(Aj∗)\eta_j\in C_c^\infty(A_j^*) such that

ηj≡1 on Aj,∣∇kηj∣≲ρj−k,k=1,2.\eta_j\equiv1\ \text{on }A_j,\qquad|\nabla^k\eta_j|\lesssim\rho_j^{-k},\qquad k=1,2.

Define

dj:=min⁡{(ρjR)4,1},d_j:=\min\left\{\left(\frac{\rho_j}{R}\right)^4,1\right\},

and

mj:=∫Aj∗∣u∣2 dx,Dj:=∫Aj∗ωR(x)∣D2u∣F2 dx.m_j:=\int_{A_j^*}|u|^2\,dx,\qquad D_j:=\int_{A_j^*}\omega_R(x)|D^2u|_F^2\,dx.

Since ∣x∣≃ρj|x|\simeq\rho_j on Aj∗∗A_j^{**} and

ωR(x)≃min⁡{∣x∣4R4,1},\omega_R(x)\simeq\min\left\{\frac{|x|^4}{R^4},1\right\},

we have

ωR(x)≃dj,x∈Aj∗,\omega_R(x)\simeq d_j,\qquad x\in A_j^*,

with constants independent of jj and RR.

We claim that

∥D2(ηju)∥L22≲dj−1Dj+ρj−4mj.(7)\|D^2(\eta_ju)\|_{L^2}^2\lesssim d_j^{-1}D_j+\rho_j^{-4}m_j. \tag*{(7)}

Indeed, the product rule and the bounds on ηj\eta_j give

∥D2(ηju)∥L22≲∥D2u∥L2(Aj∗)2+ρj−2∥∇u∥L2(Aj∗)2+ρj−4mj.\|D^2(\eta_ju)\|_{L^2}^2\lesssim\|D^2u\|_{L^2(A_j^*)}^2+\rho_j^{-2}\|\nabla u\|_{L^2(A_j^*)}^2+\rho_j^{-4}m_j.

To estimate the middle term, choose χj∈Cc∞(Aj∗∗)\chi_j\in C_c^\infty(A_j^{**}) satisfying

χj≡1 on Aj∗,∣∇χj∣≲ρj−1.\chi_j\equiv1\ \text{on }A_j^*,\qquad|\nabla\chi_j|\lesssim\rho_j^{-1}.

Integration by parts gives

∫χj2∣∇u∣2 dx=−Re⁡∫χj2u‾ Δu dx−2Re⁡∫χju‾∇χj⋅∇u dx.\int\chi_j^2|\nabla u|^2\,dx=-\operatorname{Re}\int\chi_j^2\overline{u}\,\Delta u\,dx-2\operatorname{Re}\int\chi_j\overline{u}\nabla\chi_j\cdot\nabla u\,dx.

By Cauchy–Schwarz,

∣∫χj2uΔu dx∣≲mj1/2∥D2u∥L2(Aj∗∗),\left|\int\chi_j^2u\Delta u\,dx\right|\lesssim m_j^{1/2}\lVert D^2u\rVert_{L^2(A_j^{**})},

while Young’s inequality yields

∣2∫χj∇χj⋅∇u dx∣≤12∫χj2∣∇u∣2 dx+Cρj−2mj.\left|2\int\chi_j\nabla\chi_j\cdot\nabla u\,dx\right|\leq\frac{1}{2}\int\chi_j^2|\nabla u|^2\,dx+C\rho_j^{-2}m_j.

Hence

∥∇u∥L2(Aj∗)2≲mj1/2∥D2u∥L2(Aj∗∗)+ρj−2mj.(8)\lVert\nabla u\rVert_{L^2(A_j^*)}^2\lesssim m_j^{1/2}\lVert D^2u\rVert_{L^2(A_j^{**})}+\rho_j^{-2}m_j. \tag*{(8)}

Multiplying by ρj−2\rho_j^{-2} and applying Young’s inequality once more,

ρj−2∥∇u∥L2(Aj∗)2≲∥D2u∥L2(Aj∗∗)2+ρj−4mj.\rho_j^{-2}\lVert\nabla u\rVert_{L^2(A_j^*)}^2\lesssim\lVert D^2u\rVert_{L^2(A_j^{**})}^2+\rho_j^{-4}m_j.

Since ωR≃dj\omega_R\simeq d_j on Aj∗∗A_j^{**},

∥D2u∥L2(Aj∗∗)2≲dj−1Dj.\lVert D^2u\rVert_{L^2(A_j^{**})}^2\lesssim d_j^{-1}D_j.

This proves (7).

Step 2. Local Gagliardo–Nirenberg estimate on each annulus.

We recall the following H2H^2 Gagliardo–Nirenberg inequality (see e.g. [19]):

∥u∥Lqq≲∥D2u∥L2N(q−2)4∥u∥L2q−N(q−2)4,\lVert u\rVert_{L^q}^q\lesssim\lVert D^2u\rVert_{L^2}^{\frac{N(q-2)}{4}}\lVert u\rVert_{L^2}^{q-\frac{N(q-2)}{4}},

which, by (3) and (6), gives

∥u∥Lqq≲∥D2u∥L22θ∥u∥L22γ.\lVert u\rVert_{L^q}^q\lesssim\lVert D^2u\rVert_{L^2}^{2\theta}\lVert u\rVert_{L^2}^{2\gamma}.

Since ηj≡1\eta_j\equiv1 on AjA_j, the above inequality gives

∫Aj∣u∣q dx≤∥ηju∥Lqq≲∥D2(ηju)∥L22θ∥ηju∥L22γ.\int_{A_j}|u|^q\,dx\leq\lVert\eta_ju\rVert_{L^q}^q\lesssim\lVert D^2(\eta_ju)\rVert_{L^2}^{2\theta}\lVert\eta_ju\rVert_{L^2}^{2\gamma}.

Using (7), together with (a+b)θ≲aθ+bθ(a+b)^\theta\lesssim a^\theta+b^\theta for 0<θ<10<\theta<1, we obtain

∫Aj∣u∣q dx≲dj−θDjθmjγ+ρj−4θmjq/2.\int_{A_j}|u|^q\,dx\lesssim d_j^{-\theta}D_j^\theta m_j^\gamma+\rho_j^{-4\theta}m_j^{q/2}.

On AjA_j we also have

ωR(x)∣x∣−b≲djρj−b.\omega_R(x)|x|^{-b}\lesssim d_j\rho_j^{-b}.

Therefore

∫AjωR∣x∣−b∣u∣q dx≲ρj−bdj1−θDjθmjγ+djρj−b−4θmjq/2.\int_{A_j}\omega_R|x|^{-b}|u|^q\,dx\lesssim\rho_j^{-b}d_j^{1-\theta}D_j^\theta m_j^\gamma+d_j\rho_j^{-b-4\theta}m_j^{q/2}.

The two coefficients have the uniform bounds

ρj−bdj1−θ≲R−b,djρj−b−4θ≲R−4.\rho_j^{-b}d_j^{1-\theta}\lesssim R^{-b},\qquad d_j\rho_j^{-b-4\theta}\lesssim R^{-4}.

Indeed, since

1−θ=b4,b+4θ=4,1-\theta=\frac{b}{4},\qquad b+4\theta=4,

if ρj≤R\rho_j\leq R, then dj=(ρj/R)4d_j=(\rho_j/R)^4, and hence

ρj−bdj1−θ=R−b,djρj−b−4θ=R−4.\rho_j^{-b}d_j^{1-\theta}=R^{-b},\qquad d_j\rho_j^{-b-4\theta}=R^{-4}.

If ρj≥R\rho_j\geq R, then dj=1d_j=1, so

ρj−b≤R−b,ρj−b−4θ=ρj−4≤R−4.\rho_j^{-b}\leq R^{-b},\qquad\rho_j^{-b-4\theta}=\rho_j^{-4}\leq R^{-4}.

Consequently,

∫AjωR∣x∣−b∣u∣q dx≲R−bDjθmjγ+R−4mjq/2.(9)\int_{A_j}\omega_R|x|^{-b}|u|^q\,dx \lesssim R^{-b}D_j^\theta m_j^\gamma+R^{-4}m_j^{q/2}. \tag*{(9)}

Step 3. Summation over the annuli.

Summing (9) over jj gives

∫RNωR∣x∣−b∣u∣q dx≲R−b∑jDjθmjγ+R−4∑jmjq/2.\int_{\mathbb{R}^N}\omega_R|x|^{-b}|u|^q\,dx\lesssim R^{-b}\sum_jD_j^\theta m_j^\gamma+R^{-4}\sum_jm_j^{q/2}.

Because the enlarged annuli have uniformly finite overlap,

∑jDj≲DR(u),∑jmj≲∥u∥L22.\sum_jD_j\lesssim D_R(u),\qquad\sum_jm_j\lesssim\|u\|_{L^2}^2.

By Hölder’s inequality,

∑jDjθmjγ≤(∑jDj)θ(∑jmjγ/(1−θ))1−θ.\sum_jD_j^\theta m_j^\gamma\leq\left(\sum_jD_j\right)^\theta\left(\sum_jm_j^{\gamma/(1-\theta)}\right)^{1-\theta}.

Since

γ1−θ=1+2pb>1,\frac{\gamma}{1-\theta}=1+\frac{2p}{b}>1,

we have

∑jmjγ/(1−θ)≤(∑jmj)γ/(1−θ),\sum_jm_j^{\gamma/(1-\theta)}\leq\left(\sum_jm_j\right)^{\gamma/(1-\theta)},

and hence

(∑jmjγ/(1−θ))1−θ≲∥u∥L22γ.\left(\sum_jm_j^{\gamma/(1-\theta)}\right)^{1-\theta}\lesssim\|u\|_{L^2}^{2\gamma}.

Similarly, since q/2>1q/2>1,

∑jmjq/2≤(∑jmj)q/2≲∥u∥L2q.\sum_jm_j^{q/2}\leq\left(\sum_jm_j\right)^{q/2}\lesssim\|u\|_{L^2}^q.

We therefore obtain

∫RNωR∣x∣−b∣u∣q dx≲R−b∥u∥L22γDR(u)θ+R−4∥u∥L2q.(10)\int_{\mathbb{R}^N}\omega_R|x|^{-b}|u|^q\,dx\lesssim R^{-b}\|u\|_{L^2}^{2\gamma}D_R(u)^\theta+R^{-4}\|u\|_{L^2}^q. \tag*{(10)}

Step 4. Absorption of the defect term.

By Young’s inequality, for every ε>0\varepsilon>0,

R−b∥u∥L22γDR(u)θ≤εDR(u)+CN,b,εR−b/(1−θ)∥u∥L22γ/(1−θ).R^{-b}\|u\|_{L^2}^{2\gamma}D_R(u)^\theta\leq\varepsilon D_R(u)+C_{N,b,\varepsilon}R^{-b/(1-\theta)}\|u\|_{L^2}^{2\gamma/(1-\theta)}.

Finally,

b1−θ=4,2γ1−θ=2+4pb.\frac{b}{1-\theta}=4,\qquad\frac{2\gamma}{1-\theta}=2+\frac{4p}{b}.

Inserting these identities into (10) yields

∫RNωR(x)∣x∣−b∣u(x)∣q dx≤εDR(u)+CN,b,εR−4(∥u∥L22+4p/b+∥u∥L2q),\int_{\mathbb{R}^N}\omega_R(x)|x|^{-b}|u(x)|^q\,dx\leq\varepsilon D_R(u)+C_{N,b,\varepsilon}R^{-4}\left(\|u\|_{L^2}^{2+4p/b}+\|u\|_{L^2}^q\right),

which proves (5). □

The lower-order linear commutators require a second interpolation estimate.

Lemma 3.2 (Weighted gradient estimate). For every ε>0\varepsilon> 0, there exists CN,ε>0C_{N,\varepsilon} > 0 such that, for every u∈H2(RN)u \in H^2(\mathbb{R}^N) and every R>0R > 0,

R−2∫RNωR1/2∣∇u∣2 dx≤εDR(u)+CN,εR−4∥u∥L22.R^{-2}\int_{\mathbb{R}^N}\omega_R^{1/2}|\nabla u|^2\,dx \le\varepsilon D_R(u)+C_{N,\varepsilon}R^{-4}\|u\|_{L^2}^2.

Proof. We use the notation introduced in the proof of Proposition 3.1. Since Aj⊂Aj∗A_j \subset A_j^*, the localized estimate (8) gives

∥∇u∥L2(Aj)2≲mj1/2∥D2u∥L2(Aj∗∗)+ρj−2mj.\|\nabla u\|_{L^2(A_j)}^2 \lesssim m_j^{1/2}\|D^2u\|_{L^2(A_j^{**})}+\rho_j^{-2}m_j.

Moreover, since ωR≃dj\omega_R\simeq d_j on Aj∗∗A_j^{**},

∥D2u∥L2(Aj∗∗)≲dj−1/2Dj1/2.\|D^2u\|_{L^2(A_j^{**})}\lesssim d_j^{-1/2}D_j^{1/2}.

Hence

dj1/2∥∇u∥L2(Aj)2≲mj1/2Dj1/2+dj1/2ρj−2mj.d_j^{1/2}\|\nabla u\|_{L^2(A_j)}^2\lesssim m_j^{1/2}D_j^{1/2}+d_j^{1/2}\rho_j^{-2}m_j.

Since ωR1/2≃dj1/2\omega_R^{1/2}\simeq d_j^{1/2} on AjA_j, summing over jj yields

R−2∫RNωR1/2∣∇u∣2 dx≲R−2∑jmj1/2Dj1/2+R−2∑jdj1/2ρj−2mj.(11)R^{-2}\int_{\mathbb{R}^N}\omega_R^{1/2}|\nabla u|^2\,dx\lesssim R^{-2}\sum_jm_j^{1/2}D_j^{1/2}+R^{-2}\sum_jd_j^{1/2}\rho_j^{-2}m_j. \tag*{(11)}

For the first sum, Cauchy–Schwarz and the finite overlap of the enlarged annuli give

R−2∑jmj1/2Dj1/2≤R−2(∑jmj)1/2(∑jDj)1/2≲R−2∥u∥L2DR(u)1/2.(12)R^{-2}\sum_jm_j^{1/2}D_j^{1/2}\le R^{-2}\left(\sum_jm_j\right)^{1/2}\left(\sum_jD_j\right)^{1/2}\lesssim R^{-2}\|u\|_{L^2}D_R(u)^{1/2}. \tag*{(12)}

It remains to estimate the second sum in (11). We distinguish the inner and outer annuli. If ρj<R\rho_j<R, then

dj1/2=(ρjR)2,d_j^{1/2}=\left(\frac{\rho_j}{R}\right)^2,

and therefore

R−2dj1/2ρj−2=R−2(ρjR)2ρj−2=R−4.R^{-2}d_j^{1/2}\rho_j^{-2}=R^{-2}\left(\frac{\rho_j}{R}\right)^2\rho_j^{-2}=R^{-4}.

Thus

R−2∑ρj<Rdj1/2ρj−2mj≲R−4∑ρj<Rmj≲R−4∥u∥L22.(13)R^{-2}\sum_{\rho_j<R}d_j^{1/2}\rho_j^{-2}m_j\lesssim R^{-4}\sum_{\rho_j<R}m_j\lesssim R^{-4}\|u\|_{L^2}^2. \tag*{(13)}

If ρj≥R\rho_j\ge R, then dj=1d_j=1. Since ρj=2jR≥R\rho_j=2^jR\ge R for j≥0j\ge0,

R−2dj1/2ρj−2=R−2ρj−2=2−2jR−4.R^{-2}d_j^{1/2}\rho_j^{-2}=R^{-2}\rho_j^{-2}=2^{-2j}R^{-4}.

Consequently,

R−2∑ρj≥Rdj1/2ρj−2mj≤R−4∑j≥02−2jmj≲R−4∑j≥0mj≲R−4∥u∥L22.(14)R^{-2}\sum_{\rho_j\ge R}d_j^{1/2}\rho_j^{-2}m_j\le R^{-4}\sum_{j\ge0}2^{-2j}m_j\lesssim R^{-4}\sum_{j\ge0}m_j\lesssim R^{-4}\|u\|_{L^2}^2. \tag*{(14)}

Combining (11)–(14), we obtain

R−2∫RNωR1/2∣∇u∣2 dx≲R−2∥u∥L2DR(u)1/2+R−4∥u∥L22.R^{-2}\int_{\mathbb{R}^N}\omega_R^{1/2}|\nabla u|^2\,dx\lesssim R^{-2}\|u\|_{L^2}D_R(u)^{1/2}+R^{-4}\|u\|_{L^2}^2.

Finally, Young’s inequality gives

R−2∥u∥L2DR(u)1/2≤εDR(u)+CεR−4∥u∥L22.R^{-2}\|u\|_{L^2}D_R(u)^{1/2}\le\varepsilon D_R(u)+C_\varepsilon R^{-4}\|u\|_{L^2}^2.

This proves the desired estimate.

The coercive localized virial estimate

For a smooth real-valued weight ϕ\phi, define

Mϕ(u)=2Im⁡∫RNu‾∇ϕ⋅∇u dx.\mathcal{M}_{\phi}(u)=2\operatorname{Im}\int_{\mathbb{R}^{N}}\overline{u}\nabla\phi\cdot\nabla u\,dx.

For the equation

i∂tu=Δ2u−∣x∣−b∣u∣pu,i\partial_tu=\Delta^2u-|x|^{-b}|u|^{p}u,

the localized biharmonic virial identity is

ddtMϕ(u)=8Re⁡∑i,j,k=1N∫(∂jkϕ)(∂iku)∂iju‾ dx−4Re⁡∑j,k=1N∫(∂jkΔϕ)(∂ju)∂ku‾ dx−2∫(Δ2ϕ)∣∇u∣2 dx+∫(Δ3ϕ)∣u∣2 dx−2pq∫(Δϕ)∣x∣−b∣u∣q dx+4q∫∇ϕ⋅∇(∣x∣−b)∣u∣q dx,\begin{aligned} \frac{d}{dt}\mathcal{M}_{\phi}(u)={}&8\operatorname{Re}\sum_{i,j,k=1}^{N}\int(\partial_{jk}\phi)(\partial_{ik}u)\overline{\partial_{ij}u}\,dx\\ &-4\operatorname{Re}\sum_{j,k=1}^{N}\int(\partial_{jk}\Delta\phi)(\partial_ju)\overline{\partial_ku}\,dx\\ &-2\int(\Delta^2\phi)|\nabla u|^2\,dx+\int(\Delta^3\phi)|u|^2\,dx\\ &-\frac{2p}{q}\int(\Delta\phi)|x|^{-b}|u|^q\,dx+\frac{4}{q}\int\nabla\phi\cdot\nabla(|x|^{-b})|u|^q\,dx, \end{aligned}

where q=p+2q=p+2. This is the standard biharmonic commutator identity, see e.g. [4 Lemma 3.1] and [9 Lemma 4.3] for the homogeneous setting, and [21 Lemma 3.3] for the corresponding inhomogeneous identity. Although [21 Lemma 3.3] is stated for N≥3N\geq3, the same commutator computation is valid for N=1,2N=1,2.

For the exponential weight ϕR\phi_R introduced in § 2, we define the localized virial quantity

MR(t):=MϕR(u(t))=2Im⁡∫RNu(t,x)‾∇ϕR(x)⋅∇u(t,x) dx.\mathcal{M}_R(t):=\mathcal{M}_{\phi_R}(u(t))=2\operatorname{Im}\int_{\mathbb{R}^{N}}\overline{u(t,x)}\nabla\phi_R(x)\cdot\nabla u(t,x)\,dx.

Since ∥∇ϕR∥L∞≲R\|\nabla\phi_R\|_{L^\infty}\lesssim R, this is well defined for every u∈H2(RN)u\in H^2(\mathbb{R}^{N}).

Proposition 4.1 (Coercive virial estimate). Let uu be an H2H^2-solution to (1) defined on the maximal time interval (−T−,T+)(-T_-,T_+). There exist constants cN,b,CN,b>0c_{N,b},C_{N,b}>0 such that, for every R>0R>0 and every t∈(−T−,T+)t\in(-T_-,T_+),

MR′(t)≤16E(u0)−cN,bDR(u(t))+CN,bR−4Fb,p(m),\mathcal{M}'_R(t)\leq16E(u_0)-c_{N,b}D_R(u(t))+C_{N,b}R^{-4}F_{b,p}(m),

where

Fb,p(m):=m+mq/2+m1+2p/b.F_{b,p}(m):=m+m^{q/2}+m^{1+2p/b}.

Proof. For simplicity, write u=u(t)u=u(t) and

s=∣x∣R,h(s)=e−s4.s=\frac{|x|}{R},\qquad h(s)=e^{-s^4}.

Applying the localized biharmonic virial identity with ϕ=ϕR\phi=\phi_R, we estimate its linear and nonlinear contributions separately.

Step 1. The principal Hessian term. By Lemma 2.1,

D2ϕR(x)≤(1−ωR(x))ID^2\phi_R(x)\leq(1-\omega_R(x))I

as quadratic forms. Hence, for every fixed ii,

Re⁡∑j,k(∂jkϕR)(∂iku)∂iju‾≤(1−ωR)∑j∣∂iju∣2.\operatorname{Re}\sum_{j,k}(\partial_{jk}\phi_R)(\partial_{ik}u)\overline{\partial_{ij}u}\leq(1-\omega_R)\sum_j|\partial_{ij}u|^2.

Summing over ii and integrating, we obtain

8Re⁡∑i,j,k∫RN(∂jkϕR)(∂iku)∂iju‾ dx≤8∫RN∣D2u∣F2 dx−8DR(u)=8∥Δu∥L22−8DR(u).\begin{aligned} 8\operatorname{Re}\sum_{i,j,k}\int_{\mathbb{R}^N}(\partial_{jk}\phi_R)(\partial_{ik}u)\overline{\partial_{ij}u}\,dx\le8\int_{\mathbb{R}^N}|D^2u|_F^2\,dx-8D_R(u) \\ &=8\|\Delta u\|_{L^2}^2-8D_R(u). \end{aligned}

where the last equality follows from Plancherel’s identity (4).

Step 2. The remaining linear terms. The lower-order linear contribution in the localized virial identity is

LR(u):=−4Re⁡∑j,k∫RN(∂jkΔϕR)(∂ju)∂ku‾ dxL_R(u):=-4\operatorname{Re}\sum_{j,k}\int_{\mathbb{R}^N}(\partial_{jk}\Delta\phi_R)(\partial_j u)\overline{\partial_k u}\,dx
−2∫RN(Δ2ϕR)∣∇u∣2 dx+∫RN(Δ3ϕR)∣u∣2 dx.-2\int_{\mathbb{R}^N}(\Delta^2\phi_R)|\nabla u|^2\,dx+\int_{\mathbb{R}^N}(\Delta^3\phi_R)|u|^2\,dx.

Notice in particular that the factor ∂jkΔϕR\partial_{jk}\Delta\phi_R is paired with first derivatives of uu, not with second derivatives.

By Lemma 2.22.2,

∣∇2ΔϕR∣+∣Δ2ϕR∣≲R−2ωR1/2,∣Δ3ϕR∣≲R−4.|\nabla^2\Delta\phi_R|+|\Delta^2\phi_R|\lesssim R^{-2}\omega_R^{1/2},\qquad|\Delta^3\phi_R|\lesssim R^{-4}.

Therefore

∣LR(u)∣≲R−2∫RNωR1/2∣∇u∣2 dx+R−4∥u∥L22.|L_R(u)|\lesssim R^{-2}\int_{\mathbb{R}^N}\omega_R^{1/2}|\nabla u|^2\,dx+R^{-4}\|u\|_{L^2}^2.

Using Lemma 3.23.2, for every ε>0\varepsilon>0,

∣LR(u)∣≤εDR(u)+CN,εR−4m.|L_R(u)|\le\varepsilon D_R(u)+C_{N,\varepsilon}R^{-4}m.

Step 3. The nonlinear contribution. The nonlinear part of the localized virial identity is

−2pq∫RN(ΔϕR)∣x∣−b∣u∣q dx+4q∫RN∇ϕR⋅∇(∣x∣−b)∣u∣q dx.-\frac{2p}{q}\int_{\mathbb{R}^N}(\Delta\phi_R)|x|^{-b}|u|^q\,dx+\frac{4}{q}\int_{\mathbb{R}^N}\nabla\phi_R\cdot\nabla(|x|^{-b})|u|^q\,dx.

Since

ΔϕR=h(s)(N−4s4)\Delta\phi_R=h(s)(N-4s^4)

and

∇ϕR⋅∇(∣x∣−b)=−bh(s)∣x∣−b,\nabla\phi_R\cdot\nabla(|x|^{-b})=-bh(s)|x|^{-b},

this becomes

−2q∫RNh(s)(Np+2b−4ps4)∣x∣−b∣u∣q dx.-\frac{2}{q}\int_{\mathbb{R}^N}h(s)(Np+2b-4ps^4)|x|^{-b}|u|^q\,dx.

Using the mass-critical relation

Np+2b=8,Np+2b=8,

we rewrite it as

−16q∫RN∣x∣−b∣u∣q dx+2q∫RNΘR(x)∣x∣−b∣u∣q dx,-\frac{16}{q}\int_{\mathbb{R}^N}|x|^{-b}|u|^q\,dx+\frac{2}{q}\int_{\mathbb{R}^N}\Theta_R(x)|x|^{-b}|u|^q\,dx,

where

ΘR(x):=8ωR(x)+4ps4h(s).\Theta_R(x):=8\omega_R(x)+4ps^4h(s).

Since

s4e−s4≤1−e−s4=ωR,s^4e^{-s^4}\le1-e^{-s^4}=\omega_R,

we have

0≤ΘR(x)≤(8+4p)ωR(x).0\le\Theta_R(x)\le(8+4p)\omega_R(x).

Consequently, Proposition 3.1 yields, for every ε>0\varepsilon> 0,

2q∫RNΘR(x)∣x∣−b∣u∣q dx≤εDR(u)+CN,b,εR−4(m1+2p/b+mq/2).\frac{2}{q}\int_{\mathbb{R}^{N}}\Theta_{R}(x)|x|^{-b}|u|^{q}\,dx \le\varepsilon\mathcal{D}_{R}(u)+C_{N,b,\varepsilon}R^{-4}\left(m^{1+2p/b}+m^{q/2}\right).

Step 4. Recovery of the conserved energy. Combining the principal kinetic contribution with the main nonlinear term gives

8∥Δu∥L22−16q∫RN∣x∣−b∣u∣q dx=16E(u0).8\|\Delta u\|_{L^{2}}^{2}-\frac{16}{q}\int_{\mathbb{R}^{N}}|x|^{-b}|u|^{q}\,dx=16E(u_{0}).

Hence Steps 1–3 give

MR′(t)≤16E(u0)−(8−2ε)DR(u(t))+CN,b,εR−4(m+mq/2+m1+2p/b).\mathcal{M}_{R}'(t)\le16E(u_{0})-(8-2\varepsilon)\mathcal{D}_{R}(u(t)) +C_{N,b,\varepsilon}R^{-4}\left(m+m^{q/2}+m^{1+2p/b}\right).

Choosing ε>0\varepsilon> 0 sufficiently small and absorbing it into the defect term, we obtain

MR′(t)≤16E(u0)−cN,bDR(u(t))+CN,bR−4(m+mq/2+m1+2p/b),\mathcal{M}_{R}'(t)\le16E(u_{0})-c_{N,b}\mathcal{D}_{R}(u(t)) +C_{N,b}R^{-4}\left(m+m^{q/2}+m^{1+2p/b}\right),

which proves the proposition. □

Remark 4.2. The point of Proposition 4.1 is that the defect DR\mathcal{D}_{R} is retained with a strictly negative sign. If this term were discarded, one would recover only a monotonicity estimate for MR\mathcal{M}_{R}, which by itself still allows a global grow-up scenario. The remainder of the argument couples the same defect back to the localized virial quantity.

Control of the virial by the defect

We next show that the same defect appearing in the coercive virial estimate also controls the localized virial itself.

Proposition 5.1 (Virial–defect estimate). Let N≥1N \ge1, $0 < b < \min{4,N/2}$, and uu be an H2H^{2}-solution to (1.1) defined on the maximal time interval (−T−,T+)(-T_{-},T_{+}). There exists $C_{N} > 0 * such that, for every R>0R > 0 and every t∈(−T−,T+)t \in(-T_{-},T_{+}),

∣MR(t)∣≤CNm+CNRm3/4DR(u(t))1/4.|\mathcal{M}_{R}(t)|\le C_{N}m+C_{N}Rm^{3/4}\mathcal{D}_{R}(u(t))^{1/4}.

Proof. Set

a(x):=∇φR(x)=xh(∣x∣/R)a(x):=\nabla\varphi_{R}(x)=x h(|x|/R)

and

X:=∥a⋅∇u∥L2.X:=\|a\cdot\nabla u\|_{L^{2}}.

By Cauchy–Schwarz,

∣MR(u)∣≤2m1/2X.|\mathcal{M}_{R}(u)|\le2m^{1/2}X.

Using the summation convention,

X2=Re⁡∫RNajak(∂ju)∂ku‾ dx.X^{2}=\operatorname{Re}\int_{\mathbb{R}^{N}}a_{j}a_{k}(\partial_{j}u)\overline{\partial_{k}u}\,dx.

Integrating by parts in the jj variable,

X2=−Re⁡∫RNu∂j(ajak)∂ku‾ dx−Re⁡∫RNuajak∂jku‾ dx.X^{2}=-\operatorname{Re}\int_{\mathbb{R}^{N}}u\partial_{j}(a_{j}a_{k})\overline{\partial_{k}u}\,dx -\operatorname{Re}\int_{\mathbb{R}^{N}}ua_{j}a_{k}\overline{\partial_{jk}u}\,dx.

A direct computation gives

∂j(ajak)=h(s)(N+1−8s4)ak.\partial_j(a_j a_k) = h(s)(N+1-8s^4)a_k.

Since the scalar factor h(s)(N+1−8s4)h(s)(N+1-8s^4) is uniformly bounded, the first term is bounded by

Cm1/2X.Cm^{1/2}X.

For the second term, observe that

∣a(x)∣4=∣x∣4e−4∣x∣4/R4.|a(x)|^4=|x|^4e^{-4|x|^4/R^4}.

Since

s4e−4s4≲1−e−s4,s^4e^{-4s^4}\lesssim1-e^{-s^4},

we obtain

∣a(x)∣4≲R4ωR(x).|a(x)|^4\lesssim R^4\omega_R(x).

Therefore

∣∫RNuajak∂jku‾ dx∣≤m1/2(∫RN∣a∣4∣D2u∣F2 dx)1/2≲R2m1/2DR(u)1/2.\left|\int_{\mathbb{R}^N}ua_ja_k\overline{\partial_jku}\,dx\right| \leq m^{1/2}\left(\int_{\mathbb{R}^N}|a|^4|D^2u|_F^2\,dx\right)^{1/2} \lesssim R^2m^{1/2}D_R(u)^{1/2}.

Thus

X2≤Cm1/2X+CR2m1/2DR(u)1/2.X^2\leq Cm^{1/2}X+CR^2m^{1/2}D_R(u)^{1/2}.

Solving this quadratic inequality gives

X≤Cm1/2+CRm1/4DR(u)1/4.X\leq Cm^{1/2}+CRm^{1/4}D_R(u)^{1/4}.

Hence

∣MR(u)∣≤CNm+CNRm3/4DR(u)1/4.|\mathcal{M}_R(u)|\leq C_Nm+C_NR m^{3/4}D_R(u)^{1/4}.

Proof of the main theorem

Proof of Theorem 1.1. Suppose now that

E(u0)=−e<0.E(u_0)=-e<0.

Choose RR sufficiently large, depending only on ee, mm, NN, and bb, so that

CN,bR−4Fb,p(m)≤8e.C_{N,b}R^{-4}F_{b,p}(m)\leq8e.

Then Proposition 4.1 yields

MR′(t)≤−8e−cN,bDR(u(t)).\mathcal{M}_R'(t)\leq-8e-c_{N,b}D_R(u(t)).

We first prove finite-time blow-up forward in time. Suppose, toward a contradiction, that

T+=∞.T_+=\infty.

Integrating the constant negative term gives

MR(t)≤MR(0)−8et.\mathcal{M}_R(t)\leq\mathcal{M}_R(0)-8et.

Hence

MR(t)⟶−∞as t⟶∞.\mathcal{M}_R(t)\longrightarrow-\infty\quad\text{as }t\longrightarrow\infty.

Set

y(t):=−MR(t).y(t):=-\mathcal{M}_R(t).

There exists t0<∞t_0<\infty such that

y(t)≥2CNm>0for all t≥t0,y(t)\geq2C_Nm>0\quad\text{for all }t\geq t_0,

where CNC_N is at least as large as the first constant in Proposition 5.1. For t≥t0t \ge t_0, Proposition 5.1 gives

12y(t)≤CNR3/4m3/4DR(u(t))1/4.\frac{1}{2}y(t) \le C_N R^{3/4}m^{3/4}D_R(u(t))^{1/4}.

Consequently,

DR(u(t))≥cNR−4m−3y(t)4,t≥t0.D_R(u(t)) \ge c_N R^{-4}m^{-3}y(t)^4,\qquad t \ge t_0.

On the other hand,

y′(t)=−MR′(t)≥cN,bDR(u(t)).y'(t)=-\mathcal{M}'_R(t) \ge c_{N,b}D_R(u(t)).

Therefore

y′(t)≥cN,bR−4m−3y(t)4,t≥t0.y'(t) \ge c_{N,b}R^{-4}m^{-3}y(t)^4,\qquad t \ge t_0.

Since y(t)>0y(t)>0,

ddty(t)−3=−3y(t)−4y′(t)≤−cN,bR−4m−3.\frac{d}{dt}y(t)^{-3}=-3y(t)^{-4}y'(t)\le-c_{N,b}R^{-4}m^{-3}.

Integrating from t0t_0 to tt yields

y(t)−3≤y(t0)−3−cN,bR−4m−3(t−t0).y(t)^{-3}\le y(t_0)^{-3}-c_{N,b}R^{-4}m^{-3}(t-t_0).

Set

t∗:=t0+y(t0)−3cN,bR−4m−3>t0.t_*:=t_0+\frac{y(t_0)^{-3}}{c_{N,b}R^{-4}m^{-3}}>t_0.

Since T+=∞T_+=\infty and u∈C([0,t∗];H2(RN))u\in C([0,t_*];H^2(\mathbb{R}^N)), y(t∗)y(t_*) is finite and positive. However, the preceding estimate gives y(t∗)−3≤0y(t_*)^{-3}\le0, a contradiction. Therefore, T+<∞T_+<\infty. The blow-up alternative then gives

lim⁡t↑T+∥u(t)∥H2=∞.\lim_{t\uparrow T_+}\|u(t)\|_{H^2}=\infty.

For the backward statement, define

u~(t,x):=u(−t,x)‾.\widetilde{u}(t,x):=\overline{u(-t,x)}.

Then u~\widetilde{u} solves the same equation and satisfies

M(u~)=M(u),E(u~)=E(u)<0.M(\widetilde{u})=M(u),\qquad E(\widetilde{u})=E(u)<0.

Applying the forward argument to u~\widetilde{u} gives T−<∞T_-<\infty and

lim⁡t↓−T−∥u(t)∥H2=∞.\lim_{t\downarrow-T_-}\|u(t)\|_{H^2}=\infty.

This completes the proof. □

Strong instability of standing waves

In this section, we prove Corollary 1.5. Let ω∈R\omega\in\mathbb{R}, and let 0≠Q∈H2(RN)0\ne Q\in H^2(\mathbb{R}^N) be a standing-wave profile, that is, a solution of

Δ2Q+ωQ−∣x∣−b∣Q∣pQ=0.\Delta^2Q+\omega Q-|x|^{-b}|Q|^pQ=0.

Then

u(t,x)=eiωtQ(x)u(t,x)=e^{i\omega t}Q(x)

is a standing-wave solution of (1).

We first record the Pohozaev identities associated with the stationary equation.

Lemma 7.1 (Pohozaev identities). Let N≥1N \ge1, 0<b<min⁡{4,N/2}0 < b < \min\{4,N/2\}, and p,qp,q be as in (3). Let 0≠Q∈H2(RN)0 \ne Q \in H^2(\mathbb{R}^N) solve

Δ2Q+ωQ−∣x∣−b∣Q∣pQ=0.\Delta^2 Q+\omega Q-|x|^{-b}|Q|^pQ=0.

Then

∥ΔQ∥L22+ω∥Q∥L22−∫RN∣x∣−b∣Q∣q dx=0.\|\Delta Q\|_{L^2}^2+\omega\|Q\|_{L^2}^2-\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx=0.

and

N−42∥ΔQ∥L22+N2ω∥Q∥L22−N−bq∫RN∣x∣−b∣Q∣q dx=0.\frac{N-4}{2}\|\Delta Q\|_{L^2}^2+\frac{N}{2}\omega\|Q\|_{L^2}^2-\frac{N-b}{q}\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx=0.

In particular,

E(Q)=0andω>0.E(Q)=0 \quad\text{and}\quad\omega>0.

Proof. Multiplying the stationary equation by Q‾\overline{Q}, integrating over RN\mathbb{R}^N, and taking real parts gives

∥ΔQ∥L22+ω∥Q∥L22−∫RN∣x∣−b∣Q∣q dx=0.\|\Delta Q\|_{L^2}^2+\omega\|Q\|_{L^2}^2-\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx=0.

Similarly, multiplying by x⋅∇Q‾x\cdot\nabla\overline{Q}, taking real parts, and integrating by parts yields

N−42∥ΔQ∥L22+N2ω∥Q∥L22−N−bq∫RN∣x∣−b∣Q∣q dx=0.\frac{N-4}{2}\|\Delta Q\|_{L^2}^2+\frac{N}{2}\omega\|Q\|_{L^2}^2-\frac{N-b}{q}\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx=0.

The latter identity is justified for H2H^2 solutions by the standard cutoff argument, see e.g. [3 Lemma 2.1] for the corresponding biharmonic Pohozaev computation.

Set

A:=∥ΔQ∥L22,B:=∥Q∥L22,C:=∫RN∣x∣−b∣Q∣q dx.A:=\|\Delta Q\|_{L^2}^2,\qquad B:=\|Q\|_{L^2}^2,\qquad C:=\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx.

Subtracting N/2N/2 times the first identity from the second gives

−2A+(N2−N−bq)C=0.-2A+\left(\frac{N}{2}-\frac{N-b}{q}\right)C=0.

Since

q=2N+8−2bN,q=\frac{2N+8-2b}{N},

we have

N2−N−bq=4q.\frac{N}{2}-\frac{N-b}{q}=\frac{4}{q}.

Therefore

12A=1qC.\frac{1}{2}A=\frac{1}{q}C.

This is precisely

E(Q)=12∥ΔQ∥L22−1q∫RN∣x∣−b∣Q∣q dx=0.E(Q)=\frac{1}{2}\|\Delta Q\|_{L^2}^2-\frac{1}{q}\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx=0.

Returning to the first Pohozaev identity and using C=qA/2C=qA/2, we obtain

ωB=(q2−1)A=p2A.\omega B=\left(\frac{q}{2}-1\right)A=\frac{p}{2}A.

Equivalently,

ω∥Q∥L22=p2∥ΔQ∥L22=4−bN∥ΔQ∥L22.\omega\|Q\|_{L^2}^2=\frac{p}{2}\|\Delta Q\|_{L^2}^2=\frac{4-b}{N}\|\Delta Q\|_{L^2}^2.

Since Q≠0Q\ne0 and 0<b<40<b<4, the right-hand side is strictly positive. Hence

ω>0.\omega>0.

We are now in a position to prove the strong instability result.

Proof of Corollary 1.5. Let ε>0\varepsilon> 0. For c>1c > 1, define

Qc:=cQ.Q_c := cQ.

By Lemma 7.1,

E(Q)=0,E(Q) = 0,

and hence

12∥ΔQ∥L22=1q∫RN∣x∣−b∣Q∣q dx.\frac{1}{2}\lVert\Delta Q\rVert_{L^2}^2 = \frac{1}{q}\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx.

It follows that

E(Qc)=c22∥ΔQ∥L22−cqq∫RN∣x∣−b∣Q∣q dx=12∥ΔQ∥L22(c2−cq).E(Q_c) = \frac{c^2}{2}\lVert\Delta Q\rVert_{L^2}^2 - \frac{c^q}{q}\int_{\mathbb{R}^N}|x|^{-b}|Q|^q\,dx = \frac{1}{2}\lVert\Delta Q\rVert_{L^2}^2\left(c^2-c^q\right).

Since p>0p > 0, we have q=p+2>2q = p + 2 > 2. Therefore, for every c>1c > 1,

cq>c2,c^q > c^2,

and consequently

E(Qc)<0.E(Q_c) < 0.

On the other hand,

∥Qc−Q∥H2=(c−1)∥Q∥H2⟶0as c↓1.\lVert Q_c-Q\rVert_{H^2} = (c-1)\lVert Q\rVert_{H^2} \longrightarrow0 \quad\text{as } c \downarrow1.

Thus we may choose c>1c > 1 sufficiently close to 11 so that

∥Qc−Q∥H2<ε.\lVert Q_c-Q\rVert_{H^2} < \varepsilon.

Applying Theorem 1.1 with initial datum u0=Qcu_0 = Q_c, we conclude that the corresponding solution blows up in finite positive time. Hence the standing wave eiωtQe^{i\omega t}Q is strongly unstable by finite-time blow-up. □\square

Remark 7.2. The argument uses only the fact that QQ is a nontrivial solution of the stationary equation. In particular, no ground-state characterization, minimizing property, positivity, or radial symmetry is required. Thus Corollary 1.5 applies to every nontrivial standing wave.

(N. U. C. Nguyen) CLAIRE AND LEO LLC, HOUSTON, TEXAS, USA Email address: [email protected]

(V. D. Dinh) CLAIRE AND LEO LLC, HOUSTON, TEXAS, USA Email address: [email protected]

References

  1. [1]Bai, R. Bai, M. Majdoub, and T. Saanouni, Non-radial blow-up for a mass-critical fourth-order inhomogeneous nonlinear Schrödinger equation, Dyn. Partial Differ. Equ. 23 (2026), no. 2, 195–208.DOI
  2. [2]M. Ben-Artzi, H. Koch, and J.-C. Saut, Dispersion estimates for fourth order Schrödinger equations, C. R. Acad. Sci. Paris Sér. I Math. 330 (2000), no. 2, 87–92.
  3. [3]D. Bonheure, J.-B. Casterás, T. Gou, and L. Jeanjean, Strong instability of ground states to a fourth order Schrödinger equation, Int. Math. Res. Not. IMRN 2019 (2019), no. 17, 5299–5315.
  4. [4]T. Boulenger and E. Lenzmann, Blowup for biharmonic NLS, Ann. Sci. Éc. Norm. Supér. (4) 50 (2017), no. 3, 503–544.DOI
  5. [5]L. Campos and C. M. Guzmán, Scattering for the non-radial inhomogeneous biharmonic NLS equation, Calc. Var. Partial Differential Equations 61 (2022), no. 4, Paper No. 156, 15 pp.arxiv.org/abs/2107.12359
  6. [6]M. Cardoso and L. G. Farah, Blow-up of non-radial solutions for the L² critical inhomogeneous NLS equation, Nonlinearity 35 (2022), no. 8, 4426–4436.DOI
  7. [7]M. Cardoso, C. M. Guzmán, and A. Pastor, Global well-posedness and critical norm concentration for inhomogeneous biharmonic NLS, Monatsh. Math. 198 (2022), no. 1, 1–29.arxiv.org/abs/2011.04715
  8. [8]Y. Cho, T. Ozawa, and C. Wang, Finite time blowup for the fourth-order NLS, Bull. Korean Math. Soc. 53 (2016), no. 2, 615–640.DOI
  9. [9]V. D. Dinh, Dynamics of radial solutions for the focusing fourth-order nonlinear Schrödinger equations, Nonlinearity 34 (2021), no. 2, 776–821.arxiv.org/abs/2001.03022
  10. [10]V. D. Dinh, Non-radial finite time blow-up for the fourth-order nonlinear Schrödinger equations, Appl. Math. Lett. 132 (2022), Paper No. 108084.DOI
  11. [11]V. D. Dinh and S. Keraani, Energy scattering for a class of inhomogeneous biharmonic nonlinear Schrödinger equations in low dimensions, arXiv:2211.11824, 2022.arxiv.org/abs/2211.11824
  12. [12]G. Fibich, B. Ilan, and G. Papanicolaou, Self-focusing with fourth-order dispersion, SIAM J. Appl. Math. 62 (2002), no. 4, 1437–1462.
  13. [13]C. M. Guzmán and A. Pastor, On the inhomogeneous biharmonic nonlinear Schrödinger equation: Local, global and stability results, Nonlinear Anal. Real World Appl. 56 (2020), Paper No. 103174.DOI
  14. [14]V. I. Karpman, Stabilization of soliton instabilities by higher-order dispersion: Fourth-order nonlinear Schrödinger-type equations, Phys. Rev. E 53 (1996), no. 2, R1336–R1339.pubmed.ncbi.nlm.nih.gov/9964480
  15. [15]V. I. Karpman and A. G. Shagalov, Stability of solitons described by nonlinear Schrödinger-type equations with higher-order dispersion, Phys. D 144 (2000), no. 1–2, 194–210.DOI
  16. [16]L. Dinh Finite-time blow-up of radial negative-energy solutions for the mass-critical biharmonic nonlinear Schrödinger equation, https://ai.meta.com/research/publications/finite-time-blow-up-of-radial-negative-energy-solutions-for-the-mass-critical-biharmonic-nonlinear-schrodinger-equation/ (2026).
  17. [17]X. Liu and T. Zhang, Bilinear Strichartz’s type estimates in Besov spaces with application to inhomogeneous nonlinear biharmonic Schrödinger equation, J. Differential Equations 296 (2021), 335–368.arxiv.org/abs/2103.08154
  18. [18]N. U. C. Nguyen and V. D. Dinh, Finite-time blow-up for the four-dimensional mass-critical quadratic nonlinear Schrödinger system without mass resonance, arXiv:2608.18453, 2026.arxiv.org/abs/2608.18453
  19. [19]L. Nirenberg, On elliptic partial differential equations, Annali della Scuola Normale Superiore di Pisa. 13 (1959), no. 2, 115-162.
  20. [20]B. Pausader, Global well-posedness for energy critical fourth-order Schrödinger equations in the radial case, Dyn. Partial Differ. Equ. 4 (2007), no. 3, 197–225.DOI
  21. [21]T. Saanouni, Energy scattering for radial focusing inhomogeneous bi-harmonic Schrödinger equations, Calc. Var. Partial Differential Equations 60 (2021), no. 3, Paper No. 113.arxiv.org/abs/2010.13315
  22. [22]T. Saanouni and R. Ghanmi, A note on the inhomogeneous fourth-order Schrödinger equation, J. Pseudo-Differ. Oper. Appl. 13 (2022), no. 4, Paper No. 56.DOI
  23. [23]R. Scarpelli and M. Hespanha, Blow-up results for inhomogeneous fourth-order nonlinear Schrödinger equation, Z. Angew. Math. Phys. 77 (2026), Paper No. 176.DOI

Paper details

vlnc:W_jq585jxp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

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