Introduction
Setting of the problem
In this paper, we consider the focusing energy supercritical nonlinear Schrödinger equation
{ i ψ t = − Δ ψ − ∣ ψ ∣ 2 p ψ , x ∈ R d , ψ ∣ t = 0 = ψ 0 . (1) \begin{cases}
i\psi_t=-\Delta\psi-|\psi|^{2p}\psi,\qquad x\in\mathbb{R}^d,\\
\left.\psi\right|_{t=0}=\psi_0. \tag*{(1)}
\end{cases} { i ψ t = − Δ ψ − ∣ ψ ∣ 2 p ψ , x ∈ R d , ψ ∣ t = 0 = ψ 0 . (1)
Sufficiently regular solutions conserve both mass and energy throughout their lifespan:
E ( ψ ( t ) ) = 1 2 ∫ ∣ ∇ ψ ( x , t ) ∣ 2 d x − 1 2 ( p + 1 ) ∫ ∣ ψ ( x , t ) ∣ 2 p + 2 d x = E ( ψ 0 ) , E(\psi(t))=\frac{1}{2}\int|\nabla\psi(x,t)|^2\,\mathrm{d}x-\frac{1}{2(p+1)}\int|\psi(x,t)|^{2p+2}\,\mathrm{d}x=E(\psi_0), E ( ψ ( t )) = 2 1 ∫ ∣∇ ψ ( x , t ) ∣ 2 d x − 2 ( p + 1 ) 1 ∫ ∣ ψ ( x , t ) ∣ 2 p + 2 d x = E ( ψ 0 ) ,
M ( ψ ( t ) ) = ∫ ∣ ψ ( x , t ) ∣ 2 d x = M ( ψ 0 ) . M(\psi(t))=\int|\psi(x,t)|^2\,\mathrm{d}x=M(\psi_0). M ( ψ ( t )) = ∫ ∣ ψ ( x , t ) ∣ 2 d x = M ( ψ 0 ) .
The scaling symmetry ψ λ ( t , x ) = λ 1 p ψ ( λ 2 t , λ x ) \psi_\lambda(t,x)=\lambda^{\frac{1}{p}}\psi(\lambda^2t,\lambda x) ψ λ ( t , x ) = λ p 1 ψ ( λ 2 t , λ x ) for λ > 0 \lambda>0 λ > 0 leaves the homogeneous Sobolev norm at the critical regularity invariant
∥ ψ λ ( t , ⋅ ) ∥ H ˙ s c = ∥ ψ ( λ 2 t , ⋅ ) ∥ H ˙ s c \|\psi_\lambda(t,\cdot)\|_{\dot{H}^{s_c}}=\|\psi(\lambda^2t,\cdot)\|_{\dot{H}^{s_c}} ∥ ψ λ ( t , ⋅ ) ∥ H ˙ s c = ∥ ψ ( λ 2 t , ⋅ ) ∥ H ˙ s c
for s c = − 1 p + d 2 s_c=-\frac{1}{p}+\frac{d}{2} s c = − p 1 + 2 d .
We focus on the energy supercritical regime:
s c > 1 i.e. p > 2 d − 2 , d ≥ 3. s_c>1\quad\text{i.e.}\quad p>\frac{2}{d-2},\quad d\ge3. s c > 1 i.e. p > d − 2 2 , d ≥ 3.
The Cauchy problem (1) is locally well-posed in H s H^s H s , s ≥ s c s\ge s_c s ≥ s c [3 ] . For s > s c s>s_c s > s c , the standard blowup alternative holds:
T < + ∞ ⟹ lim t ↑ T ∥ ψ ( t ) ∥ H s = + ∞ . T<+\infty\quad\Longrightarrow\quad\lim_{t\uparrow T}\|\psi(t)\|_{H^s}=+\infty. T < + ∞ ⟹ t ↑ T lim ∥ ψ ( t ) ∥ H s = + ∞.
The blowup dynamics of nonlinear Schrödinger equations exhibit different features depending on the criticality of the equation. For the mass critical case, Perelman [47 ] constructed finite time blowup solutions in one dimension with ground state concentration at the log-log rate. Merle and Raphaël [38 , 39 , 40 , 41 ] established this blowup regime and its stability in one and higher dimensions under suitable spectral assumptions.
In the slightly mass supercritical regime, Merle, Raphaël, and Szeftel [44 ] established stable self-similar finite time blowup in H 1 ( R d ) H^1(\mathbb{R}^d) H 1 ( R d ) for 1 ≤ d ≤ 5 1 \le d \le5 1 ≤ d ≤ 5 . See [1 ] for the construction of radial self-similar profiles. Beyond this near critical range, finite energy self-similar blowup solutions were constructed in three dimensions for the cubic NLS [11 ] and for powers close to three [10 ] .
For the energy critical NLS, Type II blowup solutions concentrating on the ground state have been constructed with different blowup rates and regularity properties. In three dimensions, Ortoleva and Perelman [46 ] constructed radial finite energy solutions that, as t → + ∞ t \to+\infty t → + ∞ , decompose into a single bubble e i α ( t ) λ 1 / 2 ( t ) Q λ ( t ) e^{i\alpha(t)}\lambda^{1/2}(t)Q_{\lambda(t)} e i α ( t ) λ 1/2 ( t ) Q λ ( t ) plus a radiation term, with concentration rates λ ( t ) = t − ν \lambda(t)=t^{-\nu} λ ( t ) = t − ν for sufficiently small ν > 0 \nu>0 ν > 0 . The construction combines matched asymptotic expansions with energy estimates for the linearized evolution. For finite time blowup, Schmid [50 ] constructed Type II blowup solutions with continuous rates λ ( t ) = ( T − t ) − 1 / 2 − ν \lambda(t)=(T-t)^{-1/2-\nu} λ ( t ) = ( T − t ) − 1/2 − ν , ν > 1 \nu>1 ν > 1 , and a radiation term converging in H ˙ 1 ∩ H ˙ 2 \dot H^1\cap\dot H^2 H ˙ 1 ∩ H ˙ 2 . Using a different approach based on modulation analysis and the energy method, Kim [28 ] constructed finite time Type II blowup solutions in three dimensions from smooth radial initial data, with quantized rates λ ( t ) ∼ c ( T − t ) − 2 L \lambda(t)\sim c(T-t)^{-2L} λ ( t ) ∼ c ( T − t ) − 2 L , c > 0 c>0 c > 0 , for every integer L ≥ 1 L\ge1 L ≥ 1 . Blowup through concentration of stationary profiles also occurs in other energy critical models, such as the Schrödinger map flow [24 , 42 , 48 ] .
In the energy supercritical setting, the slow algebraic decay of the ground state enters both the profile construction and the control of the radiation. Merle, Raphaël, and Rodnianski [43 ] used modulation analysis and high order energy estimates to construct radial Type II blowup solutions with quantized rates λ ( t ) ∼ c ( T − t ) L / λ + \lambda(t)\sim c(T-t)^{L/\lambda_+} λ ( t ) ∼ c ( T − t ) L / λ + , c > 0 c>0 c > 0 . Recently, Li [32 ] constructed a continuous family with λ ( t ) = t − 1 / 2 − ν \lambda(t)=t^{-1/2-\nu} λ ( t ) = t − 1/2 − ν for every ν > ν 0 ( d , p ) \nu>\nu_0(d,p) ν > ν 0 ( d , p ) , under suitable dimension and nonresonance assumptions. The proof combines matched asymptotic expansions with modified energy estimates to control the singular nonlinear background and the oscillations generated by the ground state tail.
Beyond the single bubble regime, blowup may occur through multi-bubble concentration. The study of such solutions raises questions concerning the concentration points, the blowup rates, and the nonlinear interactions between the bubbles. In one class of multi-bubble solutions, the bubbles are centered at the same point, with asymptotically separated scales. For the energy critical NLS, pure two bubble solutions of this type, global in one time direction, were constructed by Jendrej [14 ] in dimensions d ≥ 7 d\ge7 d ≥ 7 and by Jendrej, Li, and Xu [22 ] in dimension six. In both constructions, one bubble remains at a fixed scale, while the other concentrates in infinite time. Similar multi-bubble configurations also arise in other Schrödinger-type models. In particular, Yin [52 ] constructed pure infinite time bubble towers with an arbitrary finite number of bubbles for the energy critical k k k -equivariant Schrödinger map flow, for every integer k ≥ 3 k\ge3 k ≥ 3 .
Nonlinear interactions can also affect the concentration rates and the motion of the bubbles. Martel and Raphaël [36 ] constructed strongly interacting multi-bubble blowup solutions for the two dimensional mass critical NLS. The effect of interactions on spatial motion is illustrated by Nguyễn [45 ] , who constructed global two solitary wave solutions for energy subcritical NLS outside the mass critical case, with logarithmically growing separation.
Multi-bubble blowup solutions have also been constructed with concentration at prescribed distinct points. For the mass critical NLS, Merle [37 ] constructed finite time blowup solutions asymptotic to a sum of pseudo-conformal bubbles. Röckner, Su, and Zhang [49 ] constructed multi-bubble Bourgain–Wang solutions in dimensions one and two, in which bubbles concentrating at distinct points are accompanied by a regular background.
In the energy supercritical setting, the construction of multi-bubble blowup solutions has remained open. In this paper, we construct finite time Type II multi-bubble solutions in sufficiently high dimensions, concentrating at prescribed distinct points with the same quantized blowup rate. We also establish strong convergence of the radiation and determine the asymptotic behavior of the critical Sobolev norm.
Throughout this paper, [ a ] [a] [ a ] denotes the integer part of a a a , so that [ a ] ∈ Z [a] \in\mathbb{Z} [ a ] ∈ Z and [ a ] ≤ a < [ a ] + 1 [a] \le a < [a] + 1 [ a ] ≤ a < [ a ] + 1 . Our main result is the following theorem.
Theorem 1.1. Let K ≥ 2 K \ge2 K ≥ 2 and L ≥ 2 L \ge2 L ≥ 2 be integers, and let p p p be a positive integer. Define
d ∗ ( p , K , L ) = max { d 0 ( p , K ) , 2 L L − 1 ( 24 K + 11 ) } . d_*(p,K,L) = \max\left\{d_0(p,K), \frac{2L}{L-1}(24K+11)\right\}. d ∗ ( p , K , L ) = max { d 0 ( p , K ) , L − 1 2 L ( 24 K + 11 ) } . where
d 0 ( p , K ) = 2 + 2 K ( 1 p + 2 ) + 2 p Δ K , Δ K = K 2 ( 1 p + 2 ) 2 − K p ( 1 p + 2 ) . (2) d_0(p,K) = 2 + 2K\left(\frac{1}{p}+2\right) + 2p\sqrt{\Delta_K}, \qquad\Delta_K = K^2\left(\frac{1}{p}+2\right)^2 - \frac{K}{p}\left(\frac{1}{p}+2\right). \tag*{(2)} d 0 ( p , K ) = 2 + 2 K ( p 1 + 2 ) + 2 p Δ K , Δ K = K 2 ( p 1 + 2 ) 2 − p K ( p 1 + 2 ) . (2) For every integer d > d ∗ ( p , K , L ) d > d_*(p,K,L) d > d ∗ ( p , K , L ) satisfying Discr ∉ Q \sqrt{\operatorname{Discr}} \notin\mathbb{Q} Discr ∈ / Q , let { x j } j = 1 K ⊂ R d \{x_j\}_{j=1}^{K} \subset\mathbb{R}^d { x j } j = 1 K ⊂ R d be distinct points. Fix δ > 0 \delta> 0 δ > 0 sufficiently small such that
10 δ < min i ≠ j ∣ x i − x j ∣ . 10\delta< \min_{i\ne j} |x_i-x_j|. 10 δ < i = j min ∣ x i − x j ∣. Then there exists a time T > 0 T > 0 T > 0 and a solution ψ \psi ψ to (1) on ( 0 , T ] (0,T] ( 0 , T ] such that
ψ ∈ C ( ( 0 , T ] , H ˙ 1 ( R d ) ∩ H ˙ M ( R d ) ) with M ∈ ( d 2 , d ] , \psi\in C((0,T],\dot{H}^{1}(\mathbb{R}^d)\cap\dot{H}^{M}(\mathbb{R}^d)) \quad\text{with } M \in\left(\frac{d}{2},d\right], ψ ∈ C (( 0 , T ] , H ˙ 1 ( R d ) ∩ H ˙ M ( R d )) with M ∈ ( 2 d , d ] , where M M M is the largest admissible odd integer defined in (136) . The solution exhibits Type II blowup at t = 0 t=0 t = 0 . Specifically, it concentrates as a K K K -bubble profile of the form:
ψ ( t , x ) = λ 1 p ( t ) ∑ j = 1 K Q ( λ ( t ) ( x − x j ) ) + ζ ( t , x ) , (3) \psi(t,x)=\lambda^{\frac{1}{p}}(t)\sum_{j=1}^{K}Q\bigl(\lambda(t)(x-x_j)\bigr)+\zeta(t,x), \tag*{(3)} ψ ( t , x ) = λ p 1 ( t ) j = 1 ∑ K Q ( λ ( t ) ( x − x j ) ) + ζ ( t , x ) , (3) where
λ ( t ) = t − 1 / 2 − ν and ν : = ν L = − 1 2 − L λ + . (4) \lambda(t)=t^{-1/2-\nu} \qquad\text{and} \qquad\nu:=\nu_L=-\frac{1}{2}-\frac{L}{\lambda_+}. \tag*{(4)} λ ( t ) = t − 1/2 − ν and ν := ν L = − 2 1 − λ + L . (4) Moreover, it satisfies the following estimates:
∥ ψ ( t ) ∥ H ˙ s ( R d ) ≤ C , 0 < t ≤ T . \|\psi(t)\|_{\dot{H}^{s}(\mathbb{R}^d)} \le C, \qquad0<t\le T. ∥ ψ ( t ) ∥ H ˙ s ( R d ) ≤ C , 0 < t ≤ T . ∥ ψ ( t ) ∥ H ˙ s c = c ∞ 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) K ( 1 2 + ν ) ∣ S d − 1 ∣ ∣ log t ∣ + O δ ( 1 ) . (5) \|\psi(t)\|_{\dot{H}^{s_c}}=c_\infty2^{s_c}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}\sqrt{K\left(\frac{1}{2}+\nu\right)|S^{d-1}|}\sqrt{|\log t|}+O_\delta(1). \tag*{(5)} ∥ ψ ( t ) ∥ H ˙ s c = c ∞ 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) K ( 2 1 + ν ) ∣ S d − 1 ∣ ∣ log t ∣ + O δ ( 1 ) . (5) ∥ ζ ( t , ⋅ ) ∥ H ˙ s ( R d ) ≤ C δ s c − s + C t 2 ν − ( 1 2 + ν ) ( s − s c ) . (6) \|\zeta(t,\cdot)\|_{\dot{H}^{s}(\mathbb{R}^d)}\le C\delta^{s_c-s}+Ct^{2\nu-(\frac{1}{2}+\nu)(s-s_c)}. \tag*{(6)} ∥ ζ ( t , ⋅ ) ∥ H ˙ s ( R d ) ≤ C δ s c − s + C t 2 ν − ( 2 1 + ν ) ( s − s c ) . (6) Furthermore, there exists
ζ ∗ ∈ H ˙ s c + 4 ν 1 + 2 ν ( R d ) \zeta^*\in\dot{H}^{s_c+\frac{4\nu}{1+2\nu}}(\mathbb{R}^d) ζ ∗ ∈ H ˙ s c + 1 + 2 ν 4 ν ( R d ) such that as t → 0 t\to0 t → 0
ζ ( t ) → ζ ∗ in H ˙ σ ( R d ) for every s c < σ < s c + 4 ν 1 + 2 ν . \zeta(t)\to\zeta^* \quad\text{in }\dot{H}^{\sigma}(\mathbb{R}^d)\quad\text{for every }s_c<\sigma<s_c+\frac{4\nu}{1+2\nu}. ζ ( t ) → ζ ∗ in H ˙ σ ( R d ) for every s c < σ < s c + 1 + 2 ν 4 ν .
Comments on Theorem 1.1.
Distinct roles of the tail exponents for the ground state. Let Q = Q ( r ) Q=Q(r) Q = Q ( r ) be the radial positive stationary solution normalized at the origin, namely
Δ Q + Q 2 p + 1 = 0 , Q ( 0 ) = 1 , Q ′ ( 0 ) = 0. \Delta Q+Q^{2p+1}=0,\qquad Q(0)=1,\qquad Q'(0)=0. Δ Q + Q 2 p + 1 = 0 , Q ( 0 ) = 1 , Q ′ ( 0 ) = 0.
Its behavior at infinity is governed by the singular stationary profile c ∞ r − 1 p c_\infty r^{-\frac{1}{p}} c ∞ r − p 1 , where
c ∞ = ( 1 p ( − 1 p + d − 2 ) ) 1 2 p . c_\infty=\left(\frac{1}{p}\left(-\frac{1}{p}+d-2\right)\right)^{\frac{1}{2p}}. c ∞ = ( p 1 ( − p 1 + d − 2 ) ) 2 p 1 .
Linearizing the radial equation about this profile gives the characteristic exponents
λ ± = 1 p − d − 2 ∓ D i s c r 2 , D i s c r = ( d − 2 ) 2 − 4 ( 2 p + 1 ) c ∞ 2 p . (7) \lambda_{\pm}=\frac{1}{p}-\frac{d-2\mp\sqrt{\mathit{Discr}}}{2},\qquad\mathit{Discr}=(d-2)^2-4(2p+1)c_\infty^{2p}. \tag*{(7)} λ ± = p 1 − 2 d − 2 ∓ Discr , Discr = ( d − 2 ) 2 − 4 ( 2 p + 1 ) c ∞ 2 p . (7)
The condition that these exponents are real is precisely
f 1 ( d ) : = ( d − 2 ) 2 − 4 ( 1 p + 2 ) ( d − 2 ) + 4 p ( 1 p + 2 ) ≥ 0. (8) f_1(d):=(d-2)^2-4\left(\frac{1}{p}+2\right)(d-2)+\frac{4}{p}\left(\frac{1}{p}+2\right)\geq0. \tag*{(8)} f 1 ( d ) := ( d − 2 ) 2 − 4 ( p 1 + 2 ) ( d − 2 ) + p 4 ( p 1 + 2 ) ≥ 0. (8)
In this regime, the radial solution has the full asymptotic expansion
Q ( r ) ∼ r − 1 p ( c ∞ + ∑ m + n ≥ 1 m , n ≥ 0 c m , n r n λ + + m λ − ) , r → ∞ . (9) Q(r)\sim r^{-\frac{1}{p}}\left(c_\infty+\sum_{\substack{m+n\geq1\\m,n\geq0}}c_{m,n}r^{n\lambda_++m\lambda_-}\right),\qquad r\to\infty. \tag*{(9)} Q ( r ) ∼ r − p 1 c ∞ + m + n ≥ 1 m , n ≥ 0 ∑ c m , n r n λ + + m λ − , r → ∞. (9)
Here the coefficients c m , n c_{m,n} c m , n are real constants determined by the elliptic equation and the normalization at the origin. The expansion is classical and can be differentiated any number of times with respect to r r r . The truncated expansion used in the construction is established in Lemma 2.1.
The two exponents λ + \lambda_+ λ + and λ − \lambda_- λ − in (9) play different roles in the construction. The exponent λ + \lambda_+ λ + determines the powers appearing in the matching expansions and enters the quantized blowup rate. The λ − \lambda_- λ − remainder provides the spatial decay used in the matching and error estimates. The assumption D i s c r ∉ Q \sqrt{\mathit{Discr}}\notin\mathbb{Q} Discr ∈ / Q prevents resonances in the coefficient recursions for the approximate profiles.
Sobolev norms and the limiting radiation. The algebraic tail of Q Q Q explains the different Sobolev behaviors of the multi-bubble solution at various regularities and the form of the limiting radiation. For 1 ≤ s < s c 1\leq s<s_c 1 ≤ s < s c , the singular profile c ∞ r − 1 p c_\infty r^{-\frac{1}{p}} c ∞ r − p 1 has finite local H ˙ s \dot H^s H ˙ s norm near each concentration point. The outer truncation of the approximate solution removes the slow decay at infinity. The profile estimates and the control of ψ − ψ a p \psi-\psi^{ap} ψ − ψ a p therefore give uniform H ˙ s \dot H^s H ˙ s bounds for the solution.
At s = s c s=s_c s = s c , the stationary tail contributes to ∥ ψ a p ( t ) ∥ H ˙ s c 2 \|\psi^{ap}(t)\|_{\dot H^{s_c}}^2 ∥ ψ a p ( t ) ∥ H ˙ s c 2 a term proportional to
∫ λ − 1 ( t ) δ d r r = ( 1 2 + ν ) ∣ log t ∣ + log δ . \int_{\lambda^{-1}(t)}^{\delta}\frac{dr}{r}=\left(\frac{1}{2}+\nu\right)|\log t|+\log\delta. ∫ λ − 1 ( t ) δ r d r = ( 2 1 + ν ) ∣ log t ∣ + log δ .
Each bubble contributes the same logarithmic term, while the cross terms remain bounded. Together with the vanishing approximation error in H ˙ s c \dot H^{s_c} H ˙ s c , this gives the growth ∣ log t ∣ \sqrt{|\log t|} ∣ log t ∣ and the factor K \sqrt{K} K in (5) .
For s c < s ≤ M s_c<s\leq M s c < s ≤ M , Q ∈ H ˙ s Q\in\dot H^s Q ∈ H ˙ s , and the concentrating ground states contribute at the scale λ s − s c ( t ) \lambda^{s-s_c}(t) λ s − s c ( t ) , which increases to infinity as t → 0 t\to0 t → 0 . After rescaling and translation, (97) and the control of ψ − ψ a p \psi-\psi^{ap} ψ − ψ a p show that ∥ ζ ( t ) ∥ H ˙ s \|\zeta(t)\|_{\dot H^s} ∥ ζ ( t ) ∥ H ˙ s is of smaller order. Hence the ground states determine the leading growth of the solution norm.
The limiting radiation is determined by the outer expansion. As t → 0 t\to0 t → 0 , the time dependent corrections vanish, leaving the background profile f 0 f_0 f 0 , while the rescaled ground state tends to $c_\infty r^{-\frac{1}{p}}. for each fixed r > 0 r>0 r > 0 . Their difference therefore identifies the radial radiation limit. The profile estimates, interpolation, and the control of ψ − ψ a p \psi-\psi^{ap} ψ − ψ a p yield strong convergence of the radiation in H ˙ σ \dot{H}^{\sigma} H ˙ σ for
s c < σ < s c + 4 ν 1 + 2 ν . s_c<\sigma<s_c+\frac{4\nu}{1+2\nu}. s c < σ < s c + 1 + 2 ν 4 ν .
After translation to the prescribed centers, the limit is
ζ ∗ ( x ) = ∑ j = 1 K ( f 0 ( ∣ x − x j ∣ ) − c ∞ ∣ x − x j ∣ − 1 p ) . \zeta^*(x)=\sum_{j=1}^{K}\left(f_0(|x-x_j|)-c_\infty|x-x_j|^{-\frac{1}{p}}\right). ζ ∗ ( x ) = j = 1 ∑ K ( f 0 ( ∣ x − x j ∣ ) − c ∞ ∣ x − x j ∣ − p 1 ) .
Assumption on the dimensions
The dimensional restriction arises from two requirements: coercivity of the multi-bubble linearized operator and closure of the energy estimates.
First, although the approximate bubbles have separated supports, the translated ground state potentials retain their algebraic tails. The coercivity and energy estimates must therefore control their combined contribution. Hardy’s inequality and the pointwise bound on Q Q Q give the coercivity estimate in Lemma Lemma 2.4 provided that
f K ( d ) = ( d − 2 ) 2 − 4 K ( 1 p + 2 ) ( d − 2 ) + 4 K p ( 1 p + 2 ) > 0. f_K(d)=(d-2)^2-4K\left(\frac{1}{p}+2\right)(d-2)+\frac{4K}{p}\left(\frac{1}{p}+2\right)>0. f K ( d ) = ( d − 2 ) 2 − 4 K ( p 1 + 2 ) ( d − 2 ) + p 4 K ( p 1 + 2 ) > 0.
which holds for d > d 0 ( p , K ) d>d_0(p,K) d > d 0 ( p , K ) . In addition, since d > d 0 ( p , K ) ≥ d 0 ( p , 1 ) d>d_0(p,K)\ge d_0(p,1) d > d 0 ( p , K ) ≥ d 0 ( p , 1 ) , (8) gives f 1 ( d ) > 0 f_1(d)>0 f 1 ( d ) > 0 . Together with (7) , this yields λ − < λ + < − 2 \lambda_-<\lambda_+<-2 λ − < λ + < − 2 . In (9) , all terms containing r m λ − r^{m\lambda_-} r m λ − with m ≥ 1 m\ge1 m ≥ 1 , and all terms r n λ + r^{n\lambda_+} r n λ + with n > [ λ − λ + ] n>\left[\frac{\lambda_-}{\lambda_+}\right] n > [ λ + λ − ] , can be included in the remainder O ( r λ − ) O(r^{\lambda_-}) O ( r λ − ) . Consequently, the expansion of the ground state (9) can be truncated in the form
Q ( r ) = r − 1 p ( c ∞ + ∑ 1 ≤ n ≤ [ λ − λ + ] c n r λ + + n + O ( r λ − ) ) , r → ∞ . Q(r)=r^{-\frac{1}{p}}\left(c_\infty+\sum_{1\le n\le\left[\frac{\lambda_-}{\lambda_+}\right]}c_nr^{\lambda_++n}+O(r^{\lambda_-})\right),\qquad r\to\infty. Q ( r ) = r − p 1 c ∞ + 1 ≤ n ≤ [ λ + λ − ] ∑ c n r λ + + n + O ( r λ − ) , r → ∞.
Second, the construction of an exact solution from the approximate multi-bubble profile imposes additional dimension restrictions in the bootstrap argument. The t − 1 t^{-1} t − 1 growth terms in the energy inequalities require sufficient decay of the approximation error to obtain bounds uniform in the initial time. To this end, explicit formulas are needed for C 0 ( K , L ) C_0(K,L) C 0 ( K , L ) and C k ( K , L ) C_k(K,L) C k ( K , L ) , the growth coefficients in the H ˙ 1 \dot{H}^1 H ˙ 1 and H ˙ 2 k + 1 \dot{H}^{2k+1} H ˙ 2 k + 1 energy estimates, respectively. Their expressions in (170) and (195) show how the energy growth depends on the dimension and the Sobolev order. Uniform control of the combined energy estimates requires the following coefficient conditions:
C 0 ( K , L ) < 0 , C M − 1 2 ( K , L ) < − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) . (10) C_0(K,L)<0,\qquad C_{\frac{M-1}{2}}(K,L)<-4\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3). \tag*{(10)} C 0 ( K , L ) < 0 , C 2 M − 1 ( K , L ) < − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) . (10)
The first condition allows the contribution of the H ˙ 1 \dot{H}^1 H ˙ 1 energy to be absorbed in the H ˙ 2 k + 1 \dot{H}^{2k+1} H ˙ 2 k + 1 energy estimates. The second compares the energy growth coefficient with the decay exponent generated by the approximation error. This strict inequality yields energy bounds uniform in the initial time, as detailed in (208) .
Since C k ( K , L ) C_k(K,L) C k ( K , L ) increases with k k k , (10) imposes an upper bound on the Sobolev order M M M . The nonlinear estimates require M > d 2 M>\frac{d}{2} M > 2 d , so that Sobolev interpolation between the H ˙ 1 \dot{H}^1 H ˙ 1 and H ˙ M \dot{H}^M H ˙ M bounds gives the L ∞ L^\infty L ∞ control of f f f . The approximation error estimate (125) further requires M ≤ d M\le d M ≤ d . The additional restriction in d ∗ ( p , K , L ) d_*(p,K,L) d ∗ ( p , K , L ) ensures that an odd integer d 2 < M ≤ d \frac{d}{2}<M\le d 2 d < M ≤ d satisfying (10) exists.
Eliminating oscillations selects quantized blowup rates. The quantization arises from requiring the oscillatory branch to vanish in the self-similar profiles matched with the ground state tail. The absence of this oscillatory component simplifies the estimates of interactions between distinct bubbles in the multi-bubble construction. The selected rates satisfy
( 1 2 + ν ) ∣ λ + ∣ = L , ν = − 1 2 − L λ + , \left(\frac{1}{2}+\nu\right)|\lambda_{+}|=L,\qquad\nu=-\frac{1}{2}-\frac{L}{\lambda_{+}}, ( 2 1 + ν ) ∣ λ + ∣ = L , ν = − 2 1 − λ + L ,
where L L L is a positive integer. The requirement ν > 0 \nu>0 ν > 0 is equivalent to L > ∣ λ + ∣ 2 L>\frac{|\lambda_{+}|}{2} L > 2 ∣ λ + ∣ . Under the dimension assumption, the admissible integers are L ≥ 2 L\ge2 L ≥ 2 . For these rates, the self-similar profiles take the form
v n ( y ) = y n λ + ∑ k = 0 n L v n , k y 2 k , y = t − 1 2 ∣ x ∣ . v_n(y)=y^{n\lambda_{+}}\sum_{k=0}^{nL}v_{n,k}y^{2k},\qquad y=t^{-\frac{1}{2}}|x|. v n ( y ) = y n λ + k = 0 ∑ n L v n , k y 2 k , y = t − 2 1 ∣ x ∣.
Each profile is therefore y n λ + y^{n\lambda_{+}} y n λ + times a polynomial in y 2 y^2 y 2 , with no oscillatory component.
Comparison with single-bubble results. The estimates for the approximate solution and the difference ψ − ψ a p \psi-\psi^{ap} ψ − ψ a p give the blowup rate
∥ ψ ( t ) ∥ L ∞ ∼ λ 1 / p ( t ) = t L p λ + , t → 0. \|\psi(t)\|_{L^\infty}\sim\lambda^{1/p}(t)=t^{\frac{L}{p\lambda_{+}}},\qquad t\to0. ∥ ψ ( t ) ∥ L ∞ ∼ λ 1/ p ( t ) = t p λ + L , t → 0.
This is the same quantized family of rates as for the Type II blowup solutions established by Merle, Raphaël, and Rodnianski in [43 ] . The stronger dimension restriction comes from controlling the combined algebraic tails of the translated ground state potentials in the multi-bubble coercivity and energy estimates.
However, these two works take different approaches. In [43 ] , the blowup dynamics are constructed through modulation analysis and high order energy estimates. In this paper, the rate is selected by requiring the oscillatory branch of the self-similar profiles to vanish. The additional difficulty in constructing bubbles at prescribed distinct centers is to control the combined algebraic tails of the translated ground state potentials in the energy estimates.
Multi-bubble solutions for other models. In the parabolic setting, multi-bubble solutions have been constructed for the energy critical heat equation in finite time [8 ] and in infinite time [6 , 9 ] . Solutions concentrating at distinct points were also obtained for the two dimensional harmonic map heat flow [7 ] and the Keller–Segel system [2 ] . For the Keller–Segel system, blowup through the collision of two collapsing solitons was constructed in [4 ] .
For the energy critical wave equation, multi-soliton solutions were constructed in [33 , 34 , 53 ] . These constructions require control of interactions generated by the algebraic tails of the stationary profiles. Beyond existence, a classification of multi-solitons with distinct collinear velocities was obtained in [35 ] under a quantitative convergence assumption. Infinite time multi-bubble solutions concentrating at distinct points were constructed in [12 , 23 ] , and rigidity of this regime in dimension five was established in [26 ] . In the three dimensional radial setting, nonexistence of multi-bubble solutions was proved in [51 ] .
For the energy critical wave maps equation, concentric multi-bubble solutions were constructed in finite time [16 , 29 ] . For equivariance classes k ≥ 2 k\ge2 k ≥ 2 , nonexistence of finite time blowup solutions with smooth radiation was proved in [25 ] . Infinite time constructions were obtained in [13 , 18 ] . Jendrej and Lawrie also studied threshold two-bubble dynamics [17 ] and proved uniqueness of two-bubble solutions in high equivariance classes [19 ] . For general finite energy equivariant solutions, soliton resolution was established in [21 ] .
Multi-soliton dynamics have also been studied for generalized KdV and scalar field equations. For generalized KdV equations, multi-bubble blowup solutions were constructed in the mass critical case [5 , 30 ] , and strongly interacting multi-solitons were studied in [15 , 31 ] . For scalar field equations, a classification of kink clusters was established in [20 ] .
Strategy of the proof
We now outline the overall strategy, which is implemented in detail throughout Section 3–§ 8 . The proof of Theorem 1.1 proceeds in two steps. The first step constructs a sufficiently good multi-bubble approximate solution to (1) . To this end, we start by constructing a radial approximate solution that exhibits no oscillatory behavior in the self-similar region. This approximate solution is viewed as a perturbation of the concentrating soliton profile t − 1 2 ( 1 / 2 + ν ) Q ( x t 1 / 2 + ν ) t^{-\frac{1}{2}(1/2+\nu)}Q\left(\frac{x}{t^{1/2+\nu}}\right) t − 2 1 ( 1/2 + ν ) Q ( t 1/2 + ν x ) , where ν = − 1 2 − L λ + \nu=-\frac{1}{2}-\frac{L}{\lambda_{+}} ν = − 2 1 − λ + L . The core of this step is to prove the following theorem:
Theorem 1.2. Fix integers p ≥ 1 p \ge1 p ≥ 1 and L ≥ 2 L \ge2 L ≥ 2 . Let d > d 0 ( p , 1 ) d > d_0(p,1) d > d 0 ( p , 1 ) be a sufficiently large integer satisfying D i s c r ∉ Q \sqrt{\mathrm{Discr}} \notin\mathbb{Q} Discr ∈ / Q . For every sufficiently small δ > 0 \delta> 0 δ > 0 , there exist T > 0 T > 0 T > 0 and a smooth radial approximate solution ψ r a d a p \psi^{ap}_{rad} ψ r a d a p to (1) on ( 0 , T ] (0,T] ( 0 , T ] of the form:
ψ r a d a p ( t , x ) = λ 1 p ( t ) ( Q ( ρ ) + χ r a d a p ( t , ρ ) ) , ρ = λ ( t ) x , \psi^{ap}_{rad}(t,x)=\lambda^{\frac{1}{p}}(t)\left(Q(\rho)+\chi^{ap}_{rad}(t,\rho)\right),\qquad\rho=\lambda(t)x, ψ r a d a p ( t , x ) = λ p 1 ( t ) ( Q ( ρ ) + χ r a d a p ( t , ρ ) ) , ρ = λ ( t ) x , where
λ ( t ) = t − 1 2 − ν , ν = − 1 2 − L λ + . \lambda(t)=t^{-\frac{1}{2}-\nu},\qquad\nu=-\frac{1}{2}-\frac{L}{\lambda_{+}}. λ ( t ) = t − 2 1 − ν , ν = − 2 1 − λ + L . The function χ r a d a p \chi^{ap}_{rad} χ r a d a p satisfies the following properties for all 0 < t ≤ T 0<t\le T 0 < t ≤ T :
(i) Estimate on χ r a d a p \chi^{ap}_{rad} χ r a d a p . For every s c < s ≤ d s_c<s\le d s c < s ≤ d ,
∥ χ r a d a p ( t , ⋅ ) ∥ H ˙ s ( R d ) ≤ C t 2 ν + C δ s c − s t ( 1 2 + ν ) ( s − s c ) . \left\|\chi^{ap}_{rad}(t,\cdot)\right\|_{\dot H^s(\mathbb{R}^d)}\le Ct^{2\nu}+C\delta^{s_c-s}t^{(\frac{1}{2}+\nu)(s-s_c)}. ∥ χ r a d a p ( t , ⋅ ) ∥ H ˙ s ( R d ) ≤ C t 2 ν + C δ s c − s t ( 2 1 + ν ) ( s − s c ) . Moreover,
∥ ⟨ ρ ⟩ − k ∇ ρ l χ r a d a p ( t , ⋅ ) ∥ L ∞ ( R d ) ≤ C t 2 ν + C δ t 1 + 2 ν 2 p , 0 ≤ k ≤ 1 , l ≥ 0. \left\|\langle\rho\rangle^{-k}\nabla_\rho^l\chi^{ap}_{rad}(t,\cdot)\right\|_{L^\infty(\mathbb{R}^d)}\le Ct^{2\nu}+C_\delta t^{\frac{1+2\nu}{2p}},\qquad0\le k\le1,\ l\ge0. ⟨ ρ ⟩ − k ∇ ρ l χ r a d a p ( t , ⋅ ) L ∞ ( R d ) ≤ C t 2 ν + C δ t 2 p 1 + 2 ν , 0 ≤ k ≤ 1 , l ≥ 0. (ii) Estimate on the remainder. Define
R r a d a p = − i ∂ t ψ r a d a p − Δ ψ r a d a p − ∣ ψ r a d a p ∣ 2 p ψ r a d a p . R^{ap}_{rad}=-i\partial_t\psi^{ap}_{rad}-\Delta\psi^{ap}_{rad}-\left|\psi^{ap}_{rad}\right|^{2p}\psi^{ap}_{rad}. R r a d a p = − i ∂ t ψ r a d a p − Δ ψ r a d a p − ∣ ψ r a d a p ∣ 2 p ψ r a d a p . Then,
∥ R r a d a p ∥ H s ( R d ) ≤ C t ν ( 1 − 1 d ) ∣ λ − ∣ − ( 1 2 + ν d ) ( s − s c + 2 ) , for all 1 ≤ s ≤ d . \left\|R^{ap}_{rad}\right\|_{H^s(\mathbb{R}^d)}\le Ct^{\nu(1-\frac{1}{d})}\left|\lambda_{-}\right|^{-(\frac{1}{2}+\frac{\nu}{d})(s-s_c+2)},\qquad\text{for all }1\le s\le d. ∥ R r a d a p ∥ H s ( R d ) ≤ C t ν ( 1 − d 1 ) ∣ λ − ∣ − ( 2 1 + d ν ) ( s − s c + 2 ) , for all 1 ≤ s ≤ d . (iii) Support property. The approximate solution ψ r a d a p \psi^{ap}_{rad} ψ r a d a p is supported in { ∣ x ∣ ≤ 2 δ } \{|x|\le2\delta\} { ∣ x ∣ ≤ 2 δ } .
Given K K K distinct points { x j } j = 1 K \{x_j\}_{j=1}^K { x j } j = 1 K satisfying ∣ x i − x j ∣ > 10 δ |x_i-x_j|>10\delta ∣ x i − x j ∣ > 10 δ for i ≠ j i\ne j i = j , we define the K K K -bubble approximate solution by
ψ a p ( t , x ) = ∑ j = 1 K ψ r a d a p ( t , x − x j ) . \psi^{ap}(t,x)=\sum_{j=1}^{K}\psi^{ap}_{rad}(t,x-x_j). ψ a p ( t , x ) = j = 1 ∑ K ψ r a d a p ( t , x − x j ) .
The second step, carried out in § 8 , is to construct an exact solution ψ \psi ψ from this multi-bubble approximation using energy estimates and a compactness argument.
As we shall see, the blowup results established in this article rely heavily on the asymptotic behavior of the ground state Q Q Q and the analysis of the linearized operator associated with (1) around translated solitons. Consequently, § 3 is dedicated to a detailed study of these linear properties.
To construct a good approximate solution, we employ a finite term expansion across three regions, each corresponding to a different spatial scale: the inner region corresponding to ∣ x ∣ ≤ t 1 / 2 + ϵ 1 |x|\le t^{1/2+\epsilon_1} ∣ x ∣ ≤ t 1/2 + ϵ 1 , the self-similar region defined by 1 10 t 1 / 2 + ϵ 1 ≤ ∣ x ∣ ≤ 10 t 1 / 2 − ϵ 2 \frac{1}{10}t^{1/2+\epsilon_1}\leq|x|\leq10t^{1/2-\epsilon_2} 10 1 t 1/2 + ϵ 1 ≤ ∣ x ∣ ≤ 10 t 1/2 − ϵ 2 , and the remote region where ∣ x ∣ ≥ t 1 / 2 − ϵ 2 |x|\geq t^{1/2-\epsilon_2} ∣ x ∣ ≥ t 1/2 − ϵ 2 . Here 0 < ϵ 1 < ν 0<\epsilon_1<\nu 0 < ϵ 1 < ν and 0 < ϵ 2 < 1 2 0<\epsilon_2<\frac{1}{2} 0 < ϵ 2 < 2 1 are fixed in (89) by
ϵ 1 = ν d , ϵ 2 = 1 8 d . \epsilon_1=\frac{\nu}{d},\qquad\epsilon_2=\frac{1}{8d}. ϵ 1 = d ν , ϵ 2 = 8 d 1 .
The inner region is characterized as the zone of blowup concentration, where the solution is treated as a perturbation of the profile t − 1 p ( 1 / 2 + ν ) Q ( x t 1 / 2 + ν ) t^{-\frac{1}{p}(1/2+\nu)}Q\left(\frac{x}{t^{1/2+\nu}}\right) t − p 1 ( 1/2 + ν ) Q ( t 1/2 + ν x ) . In the self-similar region, the profile is uniquely determined by the matching conditions coming from the inner region. In the remote region, the profile depends on the cutoff parameter δ \delta δ .
In Section 3, we investigate the equation in the inner region ∣ x ∣ ≤ t 1 / 2 + ϵ 1 |x|\leq t^{1/2+\epsilon_1} ∣ x ∣ ≤ t 1/2 + ϵ 1 . In this regime, we seek an approximate solution in the form of a power series expansion in t 2 ν t^{2\nu} t 2 ν :
ψ i n a p ( t , x ) = t − 1 p ( 1 / 2 + ν ) ∑ k = 0 N i n t 2 ν k χ k ( ρ ) , ρ = t − ( 1 / 2 + ν ) ∣ x ∣ , \psi_{in}^{ap}(t,x)=t^{-\frac{1}{p}(1/2+\nu)}\sum_{k=0}^{N_{in}}t^{2\nu k}\chi_k(\rho),\qquad\rho=t^{-(1/2+\nu)}|x|, ψ in a p ( t , x ) = t − p 1 ( 1/2 + ν ) k = 0 ∑ N in t 2 ν k χ k ( ρ ) , ρ = t − ( 1/2 + ν ) ∣ x ∣ ,
where χ 0 \chi_0 χ 0 coincides with the ground state Q Q Q , and the functions χ k \chi_k χ k for 1 ≤ k ≤ N i n 1\leq k\leq N_{in} 1 ≤ k ≤ N in are determined recursively by solving a system of the form:
{ H χ ⃗ k = D k ( χ ⃗ 1 , … , χ ⃗ k − 1 ) , χ k ( 0 ) = i k C k , χ k ′ ( 0 ) = 0. \begin{cases}
H\vec{\chi}_k=D_k(\vec{\chi}_1,\ldots,\vec{\chi}_{k-1}),\\
\chi_k(0)=i^kC_k,\qquad\chi_k'(0)=0.
\end{cases} { H χ k = D k ( χ 1 , … , χ k − 1 ) , χ k ( 0 ) = i k C k , χ k ′ ( 0 ) = 0.
The real constants C k C_k C k are chosen recursively to impose the required behavior at infinity. The linearized operator is
H = − Δ σ 3 − ( p + 1 ) Q 2 p σ 3 − p Q 2 p σ 3 σ 1 , H=-\Delta\sigma_3-(p+1)Q^{2p}\sigma_3-pQ^{2p}\sigma_3\sigma_1, H = − Δ σ 3 − ( p + 1 ) Q 2 p σ 3 − p Q 2 p σ 3 σ 1 ,
where
σ 1 = ( 0 1 1 0 ) , σ 3 = ( 1 0 0 − 1 ) . \sigma_1=\begin{pmatrix}0&1\\1&0\end{pmatrix},\qquad\sigma_3=\begin{pmatrix}1&0\\0&-1\end{pmatrix}. σ 1 = ( 0 1 1 0 ) , σ 3 = ( 1 0 0 − 1 ) .
As established in Subsection 3.2, the functions χ k \chi_k χ k have the following asymptotic behavior at infinity:
χ k ( ρ ) = ρ − 1 p + 2 k ( ∑ max { 1 , k / L } ≤ n ≤ ⌊ λ − λ + ⌋ c n , k ρ λ + n + O ( ρ λ − ) ) , ρ → ∞ . (11) \chi_k(\rho)=\rho^{-\frac{1}{p}+2k}\left(\sum_{\max\{1,k/L\}\leq n\leq\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}c_{n,k}\rho^{\lambda_+n}+O(\rho^{\lambda_-})\right),\qquad\rho\to\infty. \tag*{(11)} χ k ( ρ ) = ρ − p 1 + 2 k m a x { 1 , k / L } ≤ n ≤ ⌊ λ + λ − ⌋ ∑ c n , k ρ λ + n + O ( ρ λ − ) , ρ → ∞. (11)
The coefficients c n , k c_{n,k} c n , k are purely imaginary for odd k k k and real for even k k k . To obtain a good approximate solution, we are then constrained to restrict the construction to the region ρ ≤ t − ν + ϵ 1 \rho\leq t^{-\nu+\epsilon_1} ρ ≤ t − ν + ϵ 1 .
The aim of Section 4 is to extend the approximate solution built in Section 3 to the self-similar region 1 10 t 1 / 2 + ϵ 1 ≤ ∣ x ∣ ≤ 10 t 1 / 2 − ϵ 2 \frac{1}{10}t^{1/2+\epsilon_1}\leq|x|\leq10t^{1/2-\epsilon_2} 10 1 t 1/2 + ϵ 1 ≤ ∣ x ∣ ≤ 10 t 1/2 − ϵ 2 . Taking into account the matching conditions from the inner region, we seek this extension in the form:
ψ s s a p ( t , x ) = t − 1 2 y − 1 p ( c ∞ + ∑ n = 1 N s s t ν ∣ λ + ∣ n v n ( y ) ) , y = t − 1 2 ∣ x ∣ , (12) \psi_{ss}^{ap}(t,x)=t^{-\frac{1}{2} }y^{-\frac{1}{p}}\left(c_\infty+\sum_{n=1}^{N_{ss}}t^{\nu|\lambda_+|n}v_n(y)\right),\qquad y=t^{-\frac{1}{2}}|x|, \tag*{(12)} ψ ss a p ( t , x ) = t − 2 1 y − p 1 ( c ∞ + n = 1 ∑ N ss t ν ∣ λ + ∣ n v n ( y ) ) , y = t − 2 1 ∣ x ∣ , (12)
The functions v n v_n v n are determined successively from (1) . The choice ν = − 1 2 − L λ + \nu=-\frac{1}{2}-\frac{L}{\lambda_+} ν = − 2 1 − λ + L makes the coefficient recursions terminate, giving
v n ( y ) = ∑ k = 0 n L v n , k y λ + n + 2 k . (13) v_n(y)=\sum_{k=0}^{nL}v_{n,k}y^{\lambda_+n+2k}. \tag*{(13)} v n ( y ) = k = 0 ∑ n L v n , k y λ + n + 2 k . (13)
Choosing v 1 , 0 = c 1 v_{1,0}=c_1 v 1 , 0 = c 1 gives v n , 0 = c n v_{n,0}=c_n v n , 0 = c n through the stationary coefficient equations. The remaining coefficients are determined recursively, and matching with (11) yields v n , k = c n , k v_{n,k}=c_{n,k} v n , k = c n , k , with c n , 0 = c n c_{n,0}=c_n c n , 0 = c n . Backward recursion from v 1 , L v_{1,L} v 1 , L gives an equivalent characterization of the same profiles. The relation ( 1 2 + ν ) λ + = − L \left(\frac{1}{2}+\nu\right)\lambda_+=-L ( 2 1 + ν ) λ + = − L also produces the cancellation in the inner recursion responsible for the lower bound n ≥ max { 1 , k L } n\geq\max\{1,\frac{k}{L}\} n ≥ max { 1 , L k } in (11) .
In § 6 , we construct an approximate solution ψ o u t N o u t \psi_{\mathrm{out}}^{N_{\mathrm{out}}} ψ out N out that extends ψ s s N s s \psi_{\mathrm{ss}}^{N_{\mathrm{ss}}} ψ ss N ss to the whole space by solving the nonlinear Schrödinger equation (1) with adapted Cauchy data in the remote region ∣ x ∣ ≥ t 1 / 2 − ϵ 2 |x| \ge t^{1/2-\epsilon_2} ∣ x ∣ ≥ t 1/2 − ϵ 2 . Let Θ 0 ∈ C ∞ ( R d ) \Theta_0 \in C^\infty(\mathbb{R}^d) Θ 0 ∈ C ∞ ( R d ) be radial, equal to 1 1 1 for r ≤ 1 r \le1 r ≤ 1 and to 0 0 0 for r ≥ 2 r \ge2 r ≥ 2 , and set Θ δ ( r ) = Θ 0 ( r / δ ) \Theta_\delta(r) = \Theta_0(r/\delta) Θ δ ( r ) = Θ 0 ( r / δ ) . Based on (12) and the expansion (13) from the self-similar region, we seek an approximate solution in the following form:
ψ o u t a p ( t , x ) = f 0 ( r ) + ∑ k = 1 N o u t t k g k ( r ) , r = ∣ x ∣ , \psi_{\mathrm{out}}^{\mathrm{ap}}(t,x)=f_0(r)+\sum_{k=1}^{N_{\mathrm{out}}}t^k g_k(r),\qquad r=|x|, ψ out ap ( t , x ) = f 0 ( r ) + k = 1 ∑ N out t k g k ( r ) , r = ∣ x ∣ ,
where the Cauchy datum
f 0 ( r ) = Θ δ ( r ) ∑ 0 ≤ n ≤ N s s a n r − 1 p − 2 ν λ + + n , a 0 = c ∞ , a n = v n , n L , f_0(r)=\Theta_\delta(r)\sum_{0\le n\le N_{\mathrm{ss}}}a_n r^{-\frac{1}{p}-2\nu\lambda_++n},\qquad a_0=c_\infty,\quad a_n=v_{n,nL}, f 0 ( r ) = Θ δ ( r ) 0 ≤ n ≤ N ss ∑ a n r − p 1 − 2 ν λ + + n , a 0 = c ∞ , a n = v n , n L ,
is determined by the matching conditions from the self-similar region. For k ≥ 0 k \ge0 k ≥ 0 , the functions g k g_k g k are determined successively by a recurrence relation of the form
g k + 1 = − i k + 1 H k + 1 ( r ; g j , 0 ≤ j ≤ k ) , g 0 = f 0 . g_{k+1}=-\frac{i}{k+1}\mathcal{H}_{k+1}(r;g_j,0\le j\le k),\qquad g_0=f_0. g k + 1 = − k + 1 i H k + 1 ( r ; g j , 0 ≤ j ≤ k ) , g 0 = f 0 .
As shown in Subsection 5.2, they are supported in r ≤ 2 δ r \le2\delta r ≤ 2 δ and admit differentiated expansions in powers of r 2 ν ∣ λ + ∣ r^{2\nu|\lambda_+|} r 2 ν ∣ λ + ∣ multiplying r − 1 p − 2 k r^{-\frac{1}{p}-2k} r − p 1 − 2 k . For k ≥ 1 k \ge1 k ≥ 1 , the coefficient b 0 k b_0^k b 0 k vanishes because c ∞ r − 1 p c_\infty r^{-\frac{1}{p}} c ∞ r − p 1 is stationary. This cancellation provides the additional decay used in the remote error estimate.
In § 7 we complete the proof of Theorem 1.2 by defining
ψ r a d a p ( t , x ) = Θ ( t − 1 / 2 − ϵ 1 ∣ x ∣ ) ψ i n a p ( t , x ) + ( 1 − Θ ( t − 1 / 2 − ϵ 1 ∣ x ∣ ) ) Θ ( t − 1 / 2 + ϵ 2 ∣ x ∣ ) ψ s s a p ( t , x ) + ( 1 − Θ ( t − 1 / 2 + ϵ 2 ∣ x ∣ ) ) ψ o u t a p ( t , x ) . (14) \begin{aligned}
\psi_{\mathrm{rad}}^{\mathrm{ap}}(t,x)=\Theta(t^{-1/2-\epsilon_1}|x|)\psi_{\mathrm{in}}^{\mathrm{ap}}(t,x)+\left(1-\Theta(t^{-1/2-\epsilon_1}|x|)\right)\Theta(t^{-1/2+\epsilon_2}|x|)\psi_{\mathrm{ss}}^{\mathrm{ap}}(t,x) \\
&+\left(1-\Theta(t^{-1/2+\epsilon_2}|x|)\right)\psi_{\mathrm{out}}^{\mathrm{ap}}(t,x). \tag*{(14)}
\end{aligned} ψ rad ap ( t , x ) = Θ ( t − 1/2 − ϵ 1 ∣ x ∣ ) ψ in ap ( t , x ) + ( 1 − Θ ( t − 1/2 − ϵ 1 ∣ x ∣ ) ) Θ ( t − 1/2 + ϵ 2 ∣ x ∣ ) ψ ss ap ( t , x ) + ( 1 − Θ ( t − 1/2 + ϵ 2 ∣ x ∣ ) ) ψ out ap ( t , x ) . (14)
with
N i n = ⌊ λ − λ + ⌋ , N s s = ⌊ λ − λ + ⌋ , and N o u t = ⌊ 4 ν d 2 ⌋ . N_{\mathrm{in}}=\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor,\qquad N_{\mathrm{ss}}=\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor,\qquad\text{and}\qquad N_{\mathrm{out}}=\left\lfloor4\nu d^2\right\rfloor. N in = ⌊ λ + λ − ⌋ , N ss = ⌊ λ + λ − ⌋ , and N out = ⌊ 4 ν d 2 ⌋ .
With these choices, combining the error estimates in the three regions with the matching estimates in the overlap regions gives the global estimate (125) , completing the construction of the radial approximate solution.
Finally, in § 8 , we complete the proof of Theorem 1.1 by introducing the multi-bubble approximation
ψ a p ( t , x ) = ∑ j = 1 K ψ r a d a p ( t , x − x j ) , \psi^{\mathrm{ap}}(t,x)=\sum_{j=1}^{K}\psi_{\mathrm{rad}}^{\mathrm{ap}}(t,x-x_j), ψ ap ( t , x ) = j = 1 ∑ K ψ rad ap ( t , x − x j ) ,
and correcting it to an exact solution through energy estimates and a compactness argument. To this end, we decompose the K K K -bubble blowup solution
ψ ( t , x ) = ∑ j = 1 K ψ r a d a p ( t , x − x j ) + λ 1 p ( t ) f ( t , y ) , y = λ ( t ) x . (15) \psi(t,x)=\sum_{j=1}^{K}\psi_{\mathrm{rad}}^{\mathrm{ap}}(t,x-x_j)+\lambda^{\frac{1}{p}}(t)f(t,y),\qquad y=\lambda(t)x. \tag*{(15)} ψ ( t , x ) = j = 1 ∑ K ψ rad ap ( t , x − x j ) + λ p 1 ( t ) f ( t , y ) , y = λ ( t ) x . (15)
Substituting (14) and (15) into (1) yields an equation for f f f , which we study by a continuity argument and suitable energy estimates. These estimates are established by combining the positivity of the multi-potential operators
L + , K = − Δ − ( 2 p + 1 ) ∑ j = 1 K Q 2 p ( y − y j ) , L − , K = − Δ − ∑ j = 1 K Q 2 p ( y − y j ) , L_{+,K}=-\Delta-(2p+1)\sum_{j=1}^{K}Q^{2p}(y-y_j),\qquad L_{-,K}=-\Delta-\sum_{j=1}^{K}Q^{2p}(y-y_j), L + , K = − Δ − ( 2 p + 1 ) j = 1 ∑ K Q 2 p ( y − y j ) , L − , K = − Δ − j = 1 ∑ K Q 2 p ( y − y j ) ,
together with the estimates of the approximate solution ψ r a d a p \psi_{\mathrm{rad}}^{\mathrm{ap}} ψ rad ap .
To simplify the notation, we shall omit in this text the dependence of all the functions on the parameter ν \nu ν . Throughout this article, T T T and C C C denote a positive time and a constant, respectively, both of which may depend on various parameters and vary from line to line. We also adopt the notation A ≲ B A \lesssim B A ≲ B to denote an estimate of the form A ≤ C B A \le CB A ≤ C B for some positive constant C C C .
Acknowledgments
This project began in 2023 while I was a visiting PhD student at Université Paris-Est Créteil (UPEC). I would like to thank my advisor Galina Perelman for many enriching and enlightening discussions and for her constant encouragement. I am also grateful to my advisor Lifeng Zhao for providing many invaluable suggestions. X. Li is supported by National Natural Science Foundation of China under contract No. 12271497.
Analysis of the stationary solution and the linearized operators
Expansion of the ground state
Our analysis is closely tied to the behavior at infinity of the stationary solution to the nonlinear Schrödinger equation (1) . In this subsection, we derive the expansion of Q Q Q , which is used throughout the construction of the approximate solution.
Lemma 2.1 ([27 ] ). Assume that p p p , d d d are positive integers satisfying d > d 0 ( p , 1 ) d > d_0(p,1) d > d 0 ( p , 1 ) and D i s c r ∉ Q \sqrt{\mathit{Discr}} \notin\mathbb{Q} Discr ∈ / Q . Then, the radial initial value problem
Δ Q + Q 2 p + 1 = 0 , with Q ( 0 ) = 1 , Q r ( 0 ) = 0 , (16) \Delta Q + Q^{2p+1} = 0,\quad\text{with } Q(0)=1,\quad Q_r(0)=0, \tag*{(16)} Δ Q + Q 2 p + 1 = 0 , with Q ( 0 ) = 1 , Q r ( 0 ) = 0 , (16) admits a smooth radial solution Q ∈ C ∞ ( R d ) Q \in C^\infty(\mathbb{R}^d) Q ∈ C ∞ ( R d ) with the asymptotic expansion
Q ( x ) = ∣ x ∣ − 1 p ( c ∞ + ∑ 1 ≤ n ≤ [ λ − λ + ] c n ∣ x ∣ λ + n + O ( ∣ x ∣ λ − ) ) , as r → + ∞ , (17) Q(x)=|x|^{-\frac{1}{p}}\left(c_\infty+\sum_{1\le n\le\left[\frac{\lambda_-}{\lambda_+}\right]}c_n|x|^{\lambda_+ n}+O(|x|^{\lambda_-})\right),\quad\text{as }r\to+\infty, \tag*{(17)} Q ( x ) = ∣ x ∣ − p 1 c ∞ + 1 ≤ n ≤ [ λ + λ − ] ∑ c n ∣ x ∣ λ + n + O ( ∣ x ∣ λ − ) , as r → + ∞ , (17) where λ ± \lambda_\pm λ ± are given by (7) and c 1 ≠ 0 c_1\ne0 c 1 = 0 . Furthermore, for all x ∈ R d ∖ { 0 } x\in\mathbb{R}^d\setminus\{0\} x ∈ R d ∖ { 0 } ,
Q ( x ) ≤ c ∞ ∣ x ∣ 1 p , (18) Q(x)\le\frac{c_\infty}{|x|^{\frac{1}{p}}}, \tag*{(18)} Q ( x ) ≤ ∣ x ∣ p 1 c ∞ , (18) and
Λ Q ( x ) ≤ 1 p Q 2 p + 1 , Λ Q : = ( 1 p + x ⋅ ∇ ) Q . (19) \Lambda Q(x)\le\frac{1}{p}Q^{2p+1},\quad\Lambda Q:=\left(\frac{1}{p}+x\cdot\nabla\right)Q. \tag*{(19)} Λ Q ( x ) ≤ p 1 Q 2 p + 1 , Λ Q := ( p 1 + x ⋅ ∇ ) Q . (19)
Proof. Truncating the full expansion (9) and using the fact λ − < λ + \lambda_-<\lambda_+ λ − < λ + , we obtain the truncated expansion (17) . The estimate (18) was established in [27 ] . It remains to prove (19) . By (18) , we have
0 < Q ( r ) < c ∞ r − 1 / p for r > 0. (20) 0<Q(r)<c_\infty r^{-1/p}\quad\text{for }r>0. \tag*{(20)} 0 < Q ( r ) < c ∞ r − 1/ p for r > 0. (20)
Indeed, suppose that Q ( r ∗ ) = c ∞ ( r ∗ ) − 1 / p Q(r^*)=c_\infty(r^*)^{-1/p} Q ( r ∗ ) = c ∞ ( r ∗ ) − 1/ p for some r ∗ > 0 r^*>0 r ∗ > 0 . By (18) , the derivative of c ∞ r − 1 / p − Q ( r ) c_\infty r^{-1/p}-Q(r) c ∞ r − 1/ p − Q ( r ) vanishes at r = r ∗ r=r^* r = r ∗ . Thus
Q ′ ( r ∗ ) = − c ∞ p ( r ∗ ) − 1 / p − 1 . Q'(r^*)=-\frac{c_\infty}{p}(r^*)^{-1/p-1}. Q ′ ( r ∗ ) = − p c ∞ ( r ∗ ) − 1/ p − 1 .
Uniqueness for the radial equation then implies Q ( ρ ) = c ∞ ρ − 1 / p Q(\rho)=c_\infty\rho^{-1/p} Q ( ρ ) = c ∞ ρ − 1/ p for all ρ > 0 \rho>0 ρ > 0 , contradicting Q ( 0 ) = 1 Q(0)=1 Q ( 0 ) = 1 .
Assume that the equation Λ Q = 0 \Lambda Q=0 Λ Q = 0 has a positive root. By continuity and Λ Q ( 0 ) = 1 p > 0 \Lambda Q(0)=\frac{1}{p}>0 Λ Q ( 0 ) = p 1 > 0 , we define
r ∗ = min { r > 0 : Λ Q ( r ) = 0 } . r_*=\min\{r>0:\Lambda Q(r)=0\}. r ∗ = min { r > 0 : Λ Q ( r ) = 0 } .
Then Λ Q ( r ) > 0 \Lambda Q(r)>0 Λ Q ( r ) > 0 for 0 ≤ r < r ∗ 0\le r<r_* 0 ≤ r < r ∗ . Consequently,
Q ′ ( r ∗ ) = − Q ( r ∗ ) p r ∗ , ( Λ Q ) ′ ( r ∗ ) ≤ 0. (21) Q'(r_*)=-\frac{Q(r_*)}{pr_*},\quad(\Lambda Q)'(r_*)\le0. \tag*{(21)} Q ′ ( r ∗ ) = − p r ∗ Q ( r ∗ ) , ( Λ Q ) ′ ( r ∗ ) ≤ 0. (21)
On the other hand, using the equation
Q ′ ′ + d − 1 r Q ′ + Q 2 p + 1 = 0 , Q''+\frac{d-1}{r}Q'+Q^{2p+1}=0, Q ′′ + r d − 1 Q ′ + Q 2 p + 1 = 0 ,
and ( Λ Q ) ( r ) = 1 p Q ( r ) + r Q ′ ( r ) (\Lambda Q)(r)=\frac{1}{p}Q(r)+rQ'(r) ( Λ Q ) ( r ) = p 1 Q ( r ) + r Q ′ ( r ) , we obtain
( Λ Q ) ′ ( r ) = ( 1 p + 2 − d ) Q ′ ( r ) − r Q 2 p + 1 ( r ) . (\Lambda Q)'(r)=\left(\frac{1}{p}+2-d\right)Q'(r)-rQ^{2p+1}(r). ( Λ Q ) ′ ( r ) = ( p 1 + 2 − d ) Q ′ ( r ) − r Q 2 p + 1 ( r ) .
Thus, by (2.6), we find at r ∗ r_\ast r ∗
( Λ Q ) ′ ( r ∗ ) = Q ( r ∗ ) r ∗ ( 1 p ( 1 p + 2 − d ) + r ∗ 2 Q 2 p ( r ∗ ) ) . (\Lambda Q)'(r_\ast)=\frac{Q(r_\ast)}{r_\ast}\left(\frac{1}{p}\left(\frac{1}{p}+2-d\right)+r_\ast^2Q^{2p}(r_\ast)\right). ( Λ Q ) ′ ( r ∗ ) = r ∗ Q ( r ∗ ) ( p 1 ( p 1 + 2 − d ) + r ∗ 2 Q 2 p ( r ∗ ) ) .
Since c ∞ 2 p = 1 p ( − 1 p + d − 2 ) c_\infty^{2p}=\frac{1}{p}\left(-\frac{1}{p}+d-2\right) c ∞ 2 p = p 1 ( − p 1 + d − 2 ) , combining this identity with (2.5), we conclude
( Λ Q ) ′ ( r ∗ ) = Q ( r ∗ ) r ∗ ( c ∞ 2 p − r ∗ 2 Q 2 p ( r ∗ ) ) > 0 , (\Lambda Q)'(r_\ast)=\frac{Q(r_\ast)}{r_\ast}\left(c_\infty^{2p}-r_\ast^2Q^{2p}(r_\ast)\right)>0, ( Λ Q ) ′ ( r ∗ ) = r ∗ Q ( r ∗ ) ( c ∞ 2 p − r ∗ 2 Q 2 p ( r ∗ ) ) > 0 ,
this contradicts (2.6). Hence Λ Q ( r ) > 0 \Lambda Q(r)>0 Λ Q ( r ) > 0 for all r > 0 r>0 r > 0 .
We next prove Λ Q ≤ 1 p Q 2 p + 1 \Lambda Q\leq\frac{1}{p}Q^{2p+1} Λ Q ≤ p 1 Q 2 p + 1 . Since Q 2 p + 1 ( 0 ) Λ Q ( 0 ) = p \frac{Q^{2p+1}(0)}{\Lambda Q(0)}=p Λ Q ( 0 ) Q 2 p + 1 ( 0 ) = p , it suffices to prove that Q 2 p + 1 Λ Q \frac{Q^{2p+1}}{\Lambda Q} Λ Q Q 2 p + 1 is nondecreasing. According to
( Q 2 p + 1 Λ Q ) ′ = ( Λ Q ) − 2 ( ( Q 2 p + 1 ) ′ Λ Q − Q 2 p + 1 ( Λ Q ) ′ ) , \left(\frac{Q^{2p+1}}{\Lambda Q}\right)'=(\Lambda Q)^{-2}\left((Q^{2p+1})'\Lambda Q-Q^{2p+1}(\Lambda Q)'\right), ( Λ Q Q 2 p + 1 ) ′ = ( Λ Q ) − 2 ( ( Q 2 p + 1 ) ′ Λ Q − Q 2 p + 1 ( Λ Q ) ′ ) ,
we compute
( r d − 1 ( Λ Q ) 2 ( Q 2 p + 1 Λ Q ) ′ ) ′ = r d − 1 ( Λ Q Δ ( Q 2 p + 1 ) − Q 2 p + 1 Δ ( Λ Q ) ) . (22) \left(r^{d-1}(\Lambda Q)^2\left(\frac{Q^{2p+1}}{\Lambda Q}\right)'\right)'=r^{d-1}\left(\Lambda Q\Delta(Q^{2p+1})-Q^{2p+1}\Delta(\Lambda Q)\right). \tag*{(22)} ( r d − 1 ( Λ Q ) 2 ( Λ Q Q 2 p + 1 ) ′ ) ′ = r d − 1 ( Λ Q Δ ( Q 2 p + 1 ) − Q 2 p + 1 Δ ( Λ Q ) ) . (22)
Using the facts
L + ( Λ Q ) = 0 , L + ( Q 2 p + 1 ) = − 2 p ( 2 p + 1 ) Q 2 p − 1 ( ∇ Q ) 2 , (23) L_+(\Lambda Q)=0,\qquad L_+(Q^{2p+1})=-2p(2p+1)Q^{2p-1}(\nabla Q)^2, \tag*{(23)} L + ( Λ Q ) = 0 , L + ( Q 2 p + 1 ) = − 2 p ( 2 p + 1 ) Q 2 p − 1 ( ∇ Q ) 2 , (23)
where the operator L + = − Δ − ( 2 p + 1 ) Q 2 p L_+=-\Delta-(2p+1)Q^{2p} L + = − Δ − ( 2 p + 1 ) Q 2 p . Combining (2.7) with (2.8) and using Λ Q > 0 \Lambda Q>0 Λ Q > 0 , we obtain
( r d − 1 ( Λ Q ) 2 ( Q 2 p + 1 Λ Q ) ′ ) ′ = 2 p ( 2 p + 1 ) r d − 1 Q 2 p − 1 ( Q ′ ) 2 Λ Q ≥ 0. \left(r^{d-1}(\Lambda Q)^2\left(\frac{Q^{2p+1}}{\Lambda Q}\right)'\right)'=2p(2p+1)r^{d-1}Q^{2p-1}(Q')^2\Lambda Q\geq0. ( r d − 1 ( Λ Q ) 2 ( Λ Q Q 2 p + 1 ) ′ ) ′ = 2 p ( 2 p + 1 ) r d − 1 Q 2 p − 1 ( Q ′ ) 2 Λ Q ≥ 0.
Integrating this inequality and using Λ Q > 0 \Lambda Q>0 Λ Q > 0 , we obtain ( Q 2 p + 1 Λ Q ) ′ ≥ 0 \left(\frac{Q^{2p+1}}{\Lambda Q}\right)'\geq0 ( Λ Q Q 2 p + 1 ) ′ ≥ 0 . Hence
Q 2 p + 1 ( r ) Λ Q ( r ) ≥ Q 2 p + 1 ( 0 ) Λ Q ( 0 ) = p , \frac{Q^{2p+1}(r)}{\Lambda Q(r)}\geq\frac{Q^{2p+1}(0)}{\Lambda Q(0)}=p, Λ Q ( r ) Q 2 p + 1 ( r ) ≥ Λ Q ( 0 ) Q 2 p + 1 ( 0 ) = p ,
which proves (2.4). □
Remark 2.2. For any collection of centers { y j } j = 1 K ⊂ R d \{y_j\}_{j=1}^K\subset\mathbb{R}^d { y j } j = 1 K ⊂ R d , the choice of d 0 ( p , K ) d_0(p,K) d 0 ( p , K ) ensures the positivity of the operator L + , K = − Δ − ( 2 p + 1 ) ∑ j = 1 K Q 2 p ( y − y j ) L_{+,K}=-\Delta-(2p+1)\sum_{j=1}^K Q^{2p}(y-y_j) L + , K = − Δ − ( 2 p + 1 ) ∑ j = 1 K Q 2 p ( y − y j ) which arises from the linearization around the K K K -bubble profile. Since
L − , K = − Δ − ∑ j = 1 K Q 2 p ( y − y j ) > L + , L_{-,K}=-\Delta-\sum_{j=1}^K Q^{2p}(y-y_j)>L_+, L − , K = − Δ − j = 1 ∑ K Q 2 p ( y − y j ) > L + ,
the corresponding multi-potential operator
H K = − Δ σ 3 − ( p + 1 ) ∑ j = 1 K Q 2 p ( y − y j ) σ 3 − p ∑ j = 1 K Q 2 p ( y − y j ) σ 3 σ 1 H_K=-\Delta\sigma_3-(p+1)\sum_{j=1}^K Q^{2p}(y-y_j)\sigma_3-p\sum_{j=1}^K Q^{2p}(y-y_j)\sigma_3\sigma_1 H K = − Δ σ 3 − ( p + 1 ) j = 1 ∑ K Q 2 p ( y − y j ) σ 3 − p j = 1 ∑ K Q 2 p ( y − y j ) σ 3 σ 1
is positive in the sense of Lemma 2.4.
Properties of the linearized operators around the ground state
Let { x j } 1 ≤ j ≤ K ⊂ R d \{x_j\}_{1\le j\le K}\subset\mathbb{R}^d { x j } 1 ≤ j ≤ K ⊂ R d be a collection of distinct points. The multi-bubble blowup solution constructed in this paper is a small perturbation of the profile
λ 1 p ( t ) ∑ j = 1 K Q ( λ ( t ) ( x − x j ) ) . \lambda^{\frac{1}{p}}(t)\sum_{j=1}^{K}Q\bigl(\lambda(t)(x-x_j)\bigr). λ p 1 ( t ) j = 1 ∑ K Q ( λ ( t ) ( x − x j ) ) .
In this subsection, we collect the spectral properties of the linearized operators around the ground state Q Q Q , as well as those of the corresponding multi-bubble linearized operator, which are used throughout the paper.
We first recall the linearization of (1) around Q Q Q , which plays a crucial role in the analysis. Writing
f = f + + i f − , f=f_++if_-, f = f + + i f − ,
the linearized equation is
− i ( ∂ t f + + i ∂ t f − ) + L + f + + i L − f − = 0. -i(\partial_tf_++i\partial_tf_-)+L_+f_++iL_-f_-=0. − i ( ∂ t f + + i ∂ t f − ) + L + f + + i L − f − = 0.
where
L + = − Δ − ( 2 p + 1 ) Q 2 p , L − = − Δ − Q 2 p . L_+=-\Delta-(2p+1)Q^{2p},\qquad L_-=-\Delta-Q^{2p}. L + = − Δ − ( 2 p + 1 ) Q 2 p , L − = − Δ − Q 2 p .
The symmetries of the equation give rise to natural elements in the kernels of these operators. Since Q Q Q solves (16) , we have
Moreover, by the scaling invariance of (1) , the rescaled family
Q λ ( r ) : = λ 1 p Q ( λ r ) , λ > 0 Q_\lambda(r):=\lambda^{\frac{1}{p}}Q(\lambda r),\qquad\lambda>0 Q λ ( r ) := λ p 1 Q ( λ r ) , λ > 0
also solves the stationary equation. Differentiating with respect to λ \lambda λ at λ = 1 \lambda=1 λ = 1 yields
Λ Q : = d d λ Q λ ∣ λ = 1 = ( 1 p + r ∂ r ) Q , \Lambda Q:=\left.\frac{d}{d\lambda}Q_\lambda\right|_{\lambda=1}=\left(\frac{1}{p}+r\partial_r\right)Q, Λ Q := d λ d Q λ λ = 1 = ( p 1 + r ∂ r ) Q ,
and hence
L + Λ Q = 0. L_+\Lambda Q=0. L + Λ Q = 0.
The spectral properties of L ± L_\pm L ± discussed below rely on the asymptotic behavior of Q Q Q at infinity, as provided by (17) .
Lemma 2.3. Assume that p p p is a positive integer and d > d 0 ( p , K ) d>d_0(p,K) d > d 0 ( p , K ) . Then the following functions form a fundamental system of solutions to L ± f = 0 L_\pm f=0 L ± f = 0 :
Θ − ( r ) = Q ( r ) , Φ − ( r ) = r − 1 p + λ − + λ + ( C − + ∑ 1 ≤ n ≤ [ λ − λ + ] c ~ n r λ + n + O ( r λ − ) ) , Θ + ( r ) = Λ Q ( r ) , Φ + ( r ) = r − 1 p + λ − ( C + + ∑ 1 ≤ n ≤ [ λ − λ + ] d n r λ + n + O ( r λ − ) ) . (24) \begin{aligned}
\Theta_-(r)&=Q(r),\qquad\Phi_-(r)=r^{-\frac{1}{p}+\lambda_-+\lambda_+}\left(C_-+\sum_{1\le n\le\left[\frac{\lambda_-}{\lambda_+}\right]}\widetilde c_n r^{\lambda_+n}+O(r^{\lambda_-})\right),\\
\Theta_+(r)&=\Lambda Q(r),\qquad\Phi_+(r)=r^{-\frac{1}{p}+\lambda_-}\left(C_++\sum_{1\le n\le\left[\frac{\lambda_-}{\lambda_+}\right]}d_n r^{\lambda_+n}+O(r^{\lambda_-})\right). \tag*{(24)}
\end{aligned} Θ − ( r ) Θ + ( r ) = Q ( r ) , Φ − ( r ) = r − p 1 + λ − + λ + C − + 1 ≤ n ≤ [ λ + λ − ] ∑ c n r λ + n + O ( r λ − ) , = Λ Q ( r ) , Φ + ( r ) = r − p 1 + λ − C + + 1 ≤ n ≤ [ λ + λ − ] ∑ d n r λ + n + O ( r λ − ) . (24) where C ± ≠ 0 C_\pm\ne0 C ± = 0 , and the expansions may be differentiated any number of times. Moreover,
W [ Θ ± , Φ ± ] ( r ) : = Θ ± ′ ( r ) Φ ± ( r ) − Θ ± ( r ) Φ ± ′ ( r ) = − r 1 − d . W[\Theta_\pm,\Phi_\pm](r):=\Theta_\pm'(r)\Phi_\pm(r)-\Theta_\pm(r)\Phi_\pm'(r)=-r^{1-d}. W [ Θ ± , Φ ± ] ( r ) := Θ ± ′ ( r ) Φ ± ( r ) − Θ ± ( r ) Φ ± ′ ( r ) = − r 1 − d .
Next, we consider the linearization of (1) around the multi-bubble profile ∑ j = 1 K Q ( λ ( t ) ( x − x j ) ) \sum_{j=1}^{K} Q(\lambda(t)(x-x_j)) ∑ j = 1 K Q ( λ ( t ) ( x − x j )) .
Set y = λ ( t ) x y=\lambda(t)x y = λ ( t ) x , y j = λ ( t ) x j y_j=\lambda(t)x_j y j = λ ( t ) x j , with 1 ≤ j ≤ K 1\le j\le K 1 ≤ j ≤ K .
As in the single-bubble case, the corresponding multi-potential operators are
L + , K = − Δ − ( 2 p + 1 ) ∑ j = 1 K Q 2 p ( y − y j ) , L − , K = − Δ − ∑ j = 1 K Q 2 p ( y − y j ) . L_{+,K}=-\Delta-(2p+1)\sum_{j=1}^{K}Q^{2p}(y-y_j),\qquad L_{-,K}=-\Delta-\sum_{j=1}^{K}Q^{2p}(y-y_j). L + , K = − Δ − ( 2 p + 1 ) j = 1 ∑ K Q 2 p ( y − y j ) , L − , K = − Δ − j = 1 ∑ K Q 2 p ( y − y j ) .
The global estimates for Q Q Q and Λ Q \Lambda Q Λ Q in Lemma 2.1 control the potentials in L ± , K L_{\pm,K} L ± , K and will be used in the estimates for the approximate and exact solutions. These estimates provide the basic control on the potentials appearing in L ± , K L_{\pm,K} L ± , K and are used repeatedly in the L ∞ L^\infty L ∞ -bounds for the approximate solution and in the energy estimates for the exact solution.
The positivity of the single-bubble operators L ± L_\pm L ± under the condition d > d 0 ( p , 1 ) d>d_0(p,1) d > d 0 ( p , 1 ) was established in [27 ] . Using the global estimates for Q Q Q from Lemma 2.1, we extend this positivity property to the multi-bubble setting:
Lemma 2.4. Let K ≥ 1 K\ge1 K ≥ 1 , p > 0 p>0 p > 0 be integers. Define
d 0 ( p , K ) = 2 + 2 K ( 1 p + 2 ) + 2 K 2 ( 1 p + 2 ) 2 − K p ( 1 p + 2 ) . d_0(p,K)=2+2K\left(\frac{1}{p}+2\right)+2\sqrt{K^2\left(\frac{1}{p}+2\right)^2-\frac{K}{p}\left(\frac{1}{p}+2\right)}. d 0 ( p , K ) = 2 + 2 K ( p 1 + 2 ) + 2 K 2 ( p 1 + 2 ) 2 − p K ( p 1 + 2 ) . Then, for every d > d 0 ( p , K ) d>d_0(p,K) d > d 0 ( p , K ) , there exists a positive constant C ( K , p , d ) C(K,p,d) C ( K , p , d ) such that
⟨ L ± , K f , f ⟩ ≥ C ( K , p , d ) ∥ ∇ f ∥ L 2 2 , C ( K , p , d ) = 1 − ( 2 p + 1 ) c ∞ 2 p 4 K ( d − 2 ) 2 , (25) \langle L_{\pm,K}f,f\rangle\ge C(K,p,d)\lVert\nabla f\rVert_{L^2}^2,\qquad C(K,p,d)=1-(2p+1)c_\infty^{2p}\frac{4K}{(d-2)^2}, \tag*{(25)} ⟨ L ± , K f , f ⟩ ≥ C ( K , p , d ) ∥ ∇ f ∥ L 2 2 , C ( K , p , d ) = 1 − ( 2 p + 1 ) c ∞ 2 p ( d − 2 ) 2 4 K , (25) for all f ∈ H ˙ 1 ( R d ) f\in\dot H^1(\mathbb{R}^d) f ∈ H ˙ 1 ( R d ) .
Proof. Since L − , K > L + , K L_{-,K}>L_{+,K} L − , K > L + , K , it suffices to prove the coercivity of L + , K L_{+,K} L + , K . By the definition of L + , K L_{+,K} L + , K and the estimate (18) , for any f ∈ H ˙ 1 ( R d ) f\in\dot H^1(\mathbb{R}^d) f ∈ H ˙ 1 ( R d ) ,
⟨ L + , K f , f ⟩ ≥ ∥ ∇ f ∥ L 2 2 − ( 2 p + 1 ) c ∞ 2 p ∑ j = 1 K ∫ R d ∣ f ( y + y j ) ∣ 2 ∣ y ∣ 2 d y . \langle L_{+,K}f,f\rangle\ge\lVert\nabla f\rVert_{L^2}^2-(2p+1)c_\infty^{2p}\sum_{j=1}^{K}\int_{\mathbb{R}^d}\frac{|f(y+y_j)|^2}{|y|^2}\,dy. ⟨ L + , K f , f ⟩ ≥ ∥ ∇ f ∥ L 2 2 − ( 2 p + 1 ) c ∞ 2 p j = 1 ∑ K ∫ R d ∣ y ∣ 2 ∣ f ( y + y j ) ∣ 2 d y .
Applying Hardy’s inequality yields
⟨ L + , K f , f ⟩ ≥ ( 1 − ( 2 p + 1 ) c ∞ 2 p 4 K ( d − 2 ) 2 ) ∥ ∇ f ∥ L 2 2 . \langle L_{+,K}f,f\rangle\ge\left(1-(2p+1)c_\infty^{2p}\frac{4K}{(d-2)^2}\right)\lVert\nabla f\rVert_{L^2}^2. ⟨ L + , K f , f ⟩ ≥ ( 1 − ( 2 p + 1 ) c ∞ 2 p ( d − 2 ) 2 4 K ) ∥ ∇ f ∥ L 2 2 .
For p > 0 p>0 p > 0 and K ≥ 1 K\ge1 K ≥ 1 , the following identity holds:
( d 0 ( p , K ) − 2 ) 2 4 K = ( 1 p + 2 ) ( 1 p + d 0 ( p , K ) − 2 ) = ( 2 p + 1 ) c ∞ 2 p ( d 0 ( p , K ) ) . \frac{(d_0(p,K)-2)^2}{4K}=\left(\frac{1}{p}+2\right)\left(\frac{1}{p}+d_0(p,K)-2\right)=(2p+1)c_\infty^{2p}(d_0(p,K)). 4 K ( d 0 ( p , K ) − 2 ) 2 = ( p 1 + 2 ) ( p 1 + d 0 ( p , K ) − 2 ) = ( 2 p + 1 ) c ∞ 2 p ( d 0 ( p , K )) .
Thus, for any d > d 0 ( p , K ) d>d_0(p,K) d > d 0 ( p , K ) , the coefficient C ( K , p , d ) = 1 − ( 2 p + 1 ) c ∞ 2 p 4 K ( d − 2 ) 2 C(K,p,d)=1-(2p+1)c_\infty^{2p}\frac{4K}{(d-2)^2} C ( K , p , d ) = 1 − ( 2 p + 1 ) c ∞ 2 p ( d − 2 ) 2 4 K is strictly positive, which completes the proof of (25) . □
Remark 2.5. Using (17) , (18) , and (25) , one easily obtains that, for any integer m ≥ 1 m\ge1 m ≥ 1 , there exists a positive constant c m = c m ( p , K , d ) c_m=c_m(p,K,d) c m = c m ( p , K , d ) such that
⟨ H K 2 m + 1 f , σ 3 f ⟩ + ⟨ H K f , σ 3 f ⟩ ≥ c m ∥ ∇ f ∥ H 2 m 2 , ∀ f ∈ H ˙ 1 ( R d ) ∩ H ˙ 2 m + 1 ( R d ) . (26) \langle H_K^{2m+1}f,\sigma_3f\rangle+\langle H_Kf,\sigma_3f\rangle\ge c_m\lVert\nabla f\rVert_{H^{2m}}^2,\qquad\forall f\in\dot H^1(\mathbb{R}^d)\cap\dot H^{2m+1}(\mathbb{R}^d). \tag*{(26)} ⟨ H K 2 m + 1 f , σ 3 f ⟩ + ⟨ H K f , σ 3 f ⟩ ≥ c m ∥ ∇ f ∥ H 2 m 2 , ∀ f ∈ H ˙ 1 ( R d ) ∩ H ˙ 2 m + 1 ( R d ) . (26)
Approximate solution in the inner region
General scheme for constructing the approximate solution in the inner region
In this section, we work in the inner region
S i n = { ( t , ρ ) ∣ 0 ≤ ρ ≤ 10 t − ( ν − ϵ 1 ) } , ρ = λ ( t ) ∣ x ∣ , S_{in}=\{(t,\rho)\mid0\leq\rho\leq10t^{-(\nu-\epsilon_1)}\},\qquad\rho=\lambda(t)|x|, S in = {( t , ρ ) ∣ 0 ≤ ρ ≤ 10 t − ( ν − ϵ 1 ) } , ρ = λ ( t ) ∣ x ∣ ,
where 0 < ϵ 1 < ν 0<\epsilon_1<\nu 0 < ϵ 1 < ν is a positive real number to be fixed later. In this region, we construct an approximate solution ψ i n N \psi_{in}^{N} ψ in N to (1) as a perturbation of the profile λ 1 p ( t ) Q ( λ ( t ) x ) \lambda^{\frac{1}{p}}(t)Q(\lambda(t)x) λ p 1 ( t ) Q ( λ ( t ) x ) .
Writing
ψ ( t , x ) = λ 1 p ( t ) U ( t , ρ ) , ρ = λ ( t ) ∣ x ∣ . (27) \psi(t,x)=\lambda^{\frac{1}{p}}(t)U(t,\rho),\qquad\rho=\lambda(t)|x|. \tag*{(27)} ψ ( t , x ) = λ p 1 ( t ) U ( t , ρ ) , ρ = λ ( t ) ∣ x ∣. (27)
A straightforward computation gives
ψ t ( t , x ) = λ 1 p ( t ) U t ( t , ρ ) + λ 1 p − 1 ( t ) λ ′ ( t ) Λ U ( t , ρ ) , \psi_t(t,x)=\lambda^{\frac{1}{p}}(t)U_t(t,\rho)+\lambda^{\frac{1}{p}-1}(t)\lambda'(t)\Lambda U(t,\rho), ψ t ( t , x ) = λ p 1 ( t ) U t ( t , ρ ) + λ p 1 − 1 ( t ) λ ′ ( t ) Λ U ( t , ρ ) ,
Δ ψ ( t , x ) = λ 1 p + 2 ( t ) Δ ρ U ( t , ρ ) , ∣ ψ ( t , x ) ∣ 2 p ψ ( t , x ) = λ 1 p + 2 ( t ) ∣ U ( t , ρ ) ∣ 2 p U ( t , ρ ) . \Delta\psi(t,x)=\lambda^{\frac{1}{p}+2}(t)\Delta_\rho U(t,\rho),\qquad|\psi(t,x)|^{2p}\psi(t,x)=\lambda^{\frac{1}{p}+2}(t)|U(t,\rho)|^{2p}U(t,\rho). Δ ψ ( t , x ) = λ p 1 + 2 ( t ) Δ ρ U ( t , ρ ) , ∣ ψ ( t , x ) ∣ 2 p ψ ( t , x ) = λ p 1 + 2 ( t ) ∣ U ( t , ρ ) ∣ 2 p U ( t , ρ ) .
We set λ ( t ) = t − 1 2 − ν L \lambda(t)=t^{-\frac{1}{2}-\nu_L} λ ( t ) = t − 2 1 − ν L , where L ≥ 2 L\geq2 L ≥ 2 is an integer and ν L \nu_L ν L is defined by
ν L = − 1 2 − L λ + , (28) \nu_L=-\frac{1}{2}-\frac{L}{\lambda_+}, \tag*{(28)} ν L = − 2 1 − λ + L , (28)
In what follows, we write ν \nu ν instead of ν L \nu_L ν L to simplify the notation. Substituting (27) into (1) and multiplying by λ − 1 / p − 2 ( t ) \lambda^{-1/p-2}(t) λ − 1/ p − 2 ( t ) gives
− i t 1 + 2 ν U t + ( 1 2 + ν ) i t 2 ν Λ U − Δ U − ∣ U ∣ 2 p U = 0 (29) -it^{1+2\nu}U_t+\left(\frac{1}{2}+\nu\right)it^{2\nu}\Lambda U-\Delta U-|U|^{2p}U=0 \tag*{(29)} − i t 1 + 2 ν U t + ( 2 1 + ν ) i t 2 ν Λ U − Δ U − ∣ U ∣ 2 p U = 0 (29)
We seek a solution U U U to (29) of the form
U ( t , ρ ) = Q ( ρ ) + ∑ k ≥ 1 t 2 ν k χ k ( ρ ) , U(t,\rho)=Q(\rho)+\sum_{k\geq1}t^{2\nu k}\chi_k(\rho), U ( t , ρ ) = Q ( ρ ) + k ≥ 1 ∑ t 2 ν k χ k ( ρ ) ,
where Q Q Q is the stationary solution introduced in Lemma 2.1 . Inserting this ansatz into (29) , we obtain the following system for the vector χ ⃗ k = ( χ k χ k ‾ ) \vec{\chi}_k=\begin{pmatrix}\chi_k\\\overline{\chi_k}\end{pmatrix} χ k = ( χ k χ k ) with k ≥ 1 k\geq1 k ≥ 1 ,
0 = t 2 ν ( H χ ⃗ 1 + N 0 ) + ∑ k ≥ 2 t 2 ν k ( H χ ⃗ k + N 1 ) + N 2 ( χ ) (30) 0=t^{2\nu}(H\vec{\chi}_1+N_0)+\sum_{k\geq2}t^{2\nu k}(H\vec{\chi}_k+N_1)+N_2(\chi) \tag*{(30)} 0 = t 2 ν ( H χ 1 + N 0 ) + k ≥ 2 ∑ t 2 ν k ( H χ k + N 1 ) + N 2 ( χ ) (30)
where
H = − Δ σ 3 − ( p + 1 ) Q 2 p σ 3 − p Q 2 p σ 3 σ 1 , N i ( χ ) = ( N i ( χ ) − N i ( χ ) ) , i = 0 , 1 , 2 , H=-\Delta\sigma_3-(p+1)Q^{2p}\sigma_3-pQ^{2p}\sigma_3\sigma_1,\qquad N_i(\chi)=\begin{pmatrix}N_i(\chi)\\-N_i(\chi)\end{pmatrix},\quad i=0,1,2, H = − Δ σ 3 − ( p + 1 ) Q 2 p σ 3 − p Q 2 p σ 3 σ 1 , N i ( χ ) = ( N i ( χ ) − N i ( χ ) ) , i = 0 , 1 , 2 ,
N 0 = i ( 1 2 + ν ) Λ Q , N 1 ( χ ) = i ( 1 2 + ν ) Λ χ k − 1 − 2 ν i ( k − 1 ) χ k − 1 , N_0=i\left(\frac{1}{2}+\nu\right)\Lambda Q,\qquad N_1(\chi)=i\left(\frac{1}{2}+\nu\right)\Lambda\chi_{k-1}-2\nu i(k-1)\chi_{k-1}, N 0 = i ( 2 1 + ν ) Λ Q , N 1 ( χ ) = i ( 2 1 + ν ) Λ χ k − 1 − 2 ν i ( k − 1 ) χ k − 1 ,
and
N 2 ( χ ) = − ∣ U ∣ 2 p U + Q 2 p + 1 + ( p + 1 ) Q 2 p ( U − Q ) + p Q 2 p ( U ‾ − Q ) . N_2(\chi)=-|U|^{2p}U+Q^{2p+1}+(p+1)Q^{2p}(U-Q)+pQ^{2p}(\overline{U}-Q). N 2 ( χ ) = − ∣ U ∣ 2 p U + Q 2 p + 1 + ( p + 1 ) Q 2 p ( U − Q ) + p Q 2 p ( U − Q ) .
Thus, for k ≥ 1 k\geq1 k ≥ 1 , we solve the following recurrence relations:
H χ ⃗ k = D k ( ρ ; Q , χ ⃗ p , 1 ≤ p ≤ k − 1 ) , D k = ( D k − D k ) (31) H\vec{\chi}_k=D_k(\rho;Q,\vec{\chi}_p,1\leq p\leq k-1),\qquad D_k=\begin{pmatrix}D_k\\-D_k\end{pmatrix} \tag*{(31)} H χ k = D k ( ρ ; Q , χ p , 1 ≤ p ≤ k − 1 ) , D k = ( D k − D k ) (31)
subject to the initial conditions
χ k ( 0 ) = i k C k , and χ k ′ ( 0 ) = 0. (32) \chi_k(0)=i^kC_k,\qquad\text{and}\qquad\chi_k'(0)=0. \tag*{(32)} χ k ( 0 ) = i k C k , and χ k ′ ( 0 ) = 0. (32)
where the constants { C k } k ≥ 1 \{C_k\}_{k\geq1} { C k } k ≥ 1 are chosen recursively to impose the required asymptotic behavior at infinity. We further decompose the source term D k D_k D k into two parts:
D k = D k ( 1 ) + D k ( 2 ) . D_k=D_k^{(1)}+D_k^{(2)}. D k = D k ( 1 ) + D k ( 2 ) .
In view of (30) , we have D 1 ( 1 ) = − i ( 1 2 + ν ) Λ Q D_1^{(1)}=-i\left(\frac{1}{2}+\nu\right)\Lambda Q D 1 ( 1 ) = − i ( 2 1 + ν ) Λ Q , D 1 ( 2 ) = 0 D_1^{(2)}=0 D 1 ( 2 ) = 0 , while for k ≥ 2 k\geq2 k ≥ 2 , D k ( 1 ) D_k^{(1)} D k ( 1 ) takes the explicit form
D k ( 1 ) ( ρ ) = − i ( 1 2 + ν ) Λ χ k − 1 ( ρ ) + 2 ν i ( k − 1 ) χ k − 1 ( ρ ) . (33) D_k^{(1)}(\rho)=-i\left(\frac{1}{2}+\nu\right)\Lambda\chi_{k-1}(\rho)+2\nu i(k-1)\chi_{k-1}(\rho). \tag*{(33)} D k ( 1 ) ( ρ ) = − i ( 2 1 + ν ) Λ χ k − 1 ( ρ ) + 2 ν i ( k − 1 ) χ k − 1 ( ρ ) . (33)
where D k ( 2 ) D_k^{(2)} D k ( 2 ) is determined by
N 2 ( ρ ) = − ∑ k ≥ 2 τ 2 ν k D k ( 2 ) ( ρ ) . (34) N_2(\rho)=-\sum_{k\geq2}\tau^{2\nu k}D_k^{(2)}(\rho). \tag*{(34)} N 2 ( ρ ) = − k ≥ 2 ∑ τ 2 ν k D k ( 2 ) ( ρ ) . (34)
Analysis of the functions χ k \chi_k χ k
We now prove the following result.
Lemma 3.1. The system (31) admits a solution { χ k } k ≥ 1 \{\chi_k\}_{k\geq1} { χ k } k ≥ 1 with the following properties:
(1) For any k ≥ 1 k\geq1 k ≥ 1 , χ k \chi_k χ k is C ∞ C^\infty C ∞ on R + \mathbb{R}_+ R + and smooth at the origin.
(2) As ρ → + ∞ \rho\to+\infty ρ → + ∞ , for 1 ≤ k ≤ L ⌊ λ λ + 1 ⌋ 1\leq k\leq L\left\lfloor\frac{\lambda}{\lambda+1}\right\rfloor 1 ≤ k ≤ L ⌊ λ + 1 λ ⌋ , the function χ k \chi_k χ k has an asymptotic expansion of the form
χ k ( ρ ) = ρ − 1 p + 2 k ( ∑ max { 1 , k / L } ≤ n ≤ ⌊ λ λ + 1 ⌋ c n , k ρ λ + n + O ( ρ λ − ) ) . (35) \chi_k(\rho)=\rho^{-\frac{1}{p}+2k}\left(\sum_{\max\{1,k/L\}\leq n\leq\left\lfloor\frac{\lambda}{\lambda+1}\right\rfloor}c_{n,k}\rho^{\lambda+n}+O(\rho^{\lambda-})\right). \tag*{(35)} χ k ( ρ ) = ρ − p 1 + 2 k m a x { 1 , k / L } ≤ n ≤ ⌊ λ + 1 λ ⌋ ∑ c n , k ρ λ + n + O ( ρ λ − ) . (35) where χ k ( ρ ) = ρ − 1 p + 2 k O ( ρ λ − ) \chi_k(\rho)=\rho^{-\frac{1}{p}+2k}O(\rho^{\lambda-}) χ k ( ρ ) = ρ − p 1 + 2 k O ( ρ λ − ) , for k ≥ L ⌊ λ λ + 1 ⌋ + 1 k\geq L\left\lfloor\frac{\lambda}{\lambda+1}\right\rfloor+1 k ≥ L ⌊ λ + 1 λ ⌋ + 1 . Furthermore, this asymptotic expansion can be differentiated any number of times with respect to ρ \rho ρ .
Proof. For k ≥ 1 k\geq1 k ≥ 1 , write
v k + = Re χ k , v k − = Im χ k , g k + = Re D k , g k − = Im D k . v_k^+=\operatorname{Re}\chi_k,\qquad v_k^-=\operatorname{Im}\chi_k,\qquad g_k^+=\operatorname{Re}D_k,\qquad g_k^-=\operatorname{Im}D_k. v k + = Re χ k , v k − = Im χ k , g k + = Re D k , g k − = Im D k .
Then (31) and (32) become
L ± v k ± = g k ± , L_\pm v_k^\pm=g_k^\pm, L ± v k ± = g k ± ,
v k ± ( 0 ) + i v k ∓ ( 0 ) = i k C k , ∂ ρ v k ± ( 0 ) = 0. (36) v_k^\pm(0)+iv_k^\mp(0)=i^kC_k,\qquad\partial_\rho v_k^\pm(0)=0. \tag*{(36)} v k ± ( 0 ) + i v k ∓ ( 0 ) = i k C k , ∂ ρ v k ± ( 0 ) = 0. (36)
For k = 1 k=1 k = 1 , (36) gives
L + v 1 + = 0 , L − v 1 − = − ( 1 2 + ν ) Λ Q , (37) L_+v_1^+=0,\qquad L_-v_1^-=-\left(\frac{1}{2}+\nu\right)\Lambda Q, \tag*{(37)} L + v 1 + = 0 , L − v 1 − = − ( 2 1 + ν ) Λ Q , (37)
with initial conditions
v 1 + ( 0 ) = 0 , v 1 − ( 0 ) = C 1 , ∂ ρ v 1 − ( 0 ) = 0. v_1^+(0)=0,\qquad v_1^-(0)=C_1,\qquad\partial_\rho v_1^-(0)=0. v 1 + ( 0 ) = 0 , v 1 − ( 0 ) = C 1 , ∂ ρ v 1 − ( 0 ) = 0.
First, v 1 + ( ρ ) = 0 v_1^+(\rho)=0 v 1 + ( ρ ) = 0 solves L + v 1 + = 0 L_+v_1^+=0 L + v 1 + = 0 with zero initial data at the origin. It remains to compute v 1 − ( ρ ) v_1^-(\rho) v 1 − ( ρ ) . Since Q ( 0 ) = 1 Q(0)=1 Q ( 0 ) = 1 , we choose
C 1 = − ( 1 2 + ν ) ∫ 0 ∞ Φ − ( s ) Λ Q ( s ) W − 1 [ Θ − , Φ − ] ( s ) d s . C_1=-\left(\frac{1}{2}+\nu\right)\int_0^\infty\Phi_-(s)\Lambda Q(s)W^{-1}[\Theta_-,\Phi_-](s)\,ds. C 1 = − ( 2 1 + ν ) ∫ 0 ∞ Φ − ( s ) Λ Q ( s ) W − 1 [ Θ − , Φ − ] ( s ) d s .
The solution of (37) is given by
v 1 − ( ρ ) = − ( 1 2 + ν ) ( Φ − ( ρ ) ∫ 0 ρ Θ − ( s ) Λ Q ( s ) W − 1 [ Θ − , Φ − ] ( s ) d s + Θ − ( ρ ) ∫ ρ ∞ Φ − ( s ) Λ Q ( s ) W − 1 [ Θ − , Φ − ] ( s ) d s ) . v_1^-(\rho)=-\left(\frac{1}{2}+\nu\right)\left(\Phi_-(\rho)\int_0^\rho\Theta_-(s)\Lambda Q(s)W^{-1}[\Theta_-,\Phi_-](s)\,ds
+\Theta_-(\rho)\int_\rho^\infty\Phi_-(s)\Lambda Q(s)W^{-1}[\Theta_-,\Phi_-](s)\,ds\right). v 1 − ( ρ ) = − ( 2 1 + ν ) ( Φ − ( ρ ) ∫ 0 ρ Θ − ( s ) Λ Q ( s ) W − 1 [ Θ − , Φ − ] ( s ) d s + Θ − ( ρ ) ∫ ρ ∞ Φ − ( s ) Λ Q ( s ) W − 1 [ Θ − , Φ − ] ( s ) d s ) .
where the Wronskian W [ Θ − , Φ − ] ( s ) = − s 1 − d W[\Theta_{-},\Phi_{-}](s)=-s^{1-d} W [ Θ − , Φ − ] ( s ) = − s 1 − d is defined as in Lemma 2.3 . Applying Lemma 2.3 together with the relation λ − + λ + = 2 p − d + 2 \lambda_{-}+\lambda_{+}=\frac{2}{p}-d+2 λ − + λ + = p 2 − d + 2 , we find that v 1 − v_{1}^{-} v 1 − has the following asymptotic expansion as ρ → + ∞ \rho\to+\infty ρ → + ∞ :
v 1 − ( ρ ) = ρ − 1 p + 2 ( ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ c n , 1 − ρ λ + n + O ( ρ λ − ) ) . v_{1}^{-}(\rho)=\rho^{-\frac{1}{p}+2}\left(\sum_{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor}c_{n,1}^{-}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right). v 1 − ( ρ ) = ρ − p 1 + 2 1 ≤ n ≤ ⌊ λ + λ − ⌋ ∑ c n , 1 − ρ λ + n + O ( ρ λ − ) .
Therefore, as ρ → ∞ \rho\to\infty ρ → ∞ , χ 1 \chi_{1} χ 1 admits the expansion
χ 1 ( ρ ) = v 1 + ( ρ ) + i v 1 − ( ρ ) = ρ − 1 p + 2 ( ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ c n , 1 ρ λ + n + O ( ρ λ − ) ) . (38) \chi_{1}(\rho)=v_{1}^{+}(\rho)+iv_{1}^{-}(\rho)=\rho^{-\frac{1}{p}+2}\left(\sum_{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor}c_{n,1}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right). \tag*{(38)} χ 1 ( ρ ) = v 1 + ( ρ ) + i v 1 − ( ρ ) = ρ − p 1 + 2 1 ≤ n ≤ ⌊ λ + λ − ⌋ ∑ c n , 1 ρ λ + n + O ( ρ λ − ) . (38)
where c n , 1 = i c n , 1 − c_{n,1}=ic_{n,1}^{-} c n , 1 = i c n , 1 − is purely imaginary.
For k = 2 k=2 k = 2 , the expressions (33) and (34) yield
D 2 ( ρ ) = − i ( 1 2 + ν ) Λ χ 1 + 2 ν i χ 1 + Q 2 p − 1 ( p ( p + 1 ) 2 χ 1 2 + p ( p − 1 ) 2 χ ‾ 1 2 + p ( p + 1 ) χ 1 χ ‾ 1 ) . D_{2}(\rho)=-i\left(\frac{1}{2}+\nu\right)\Lambda\chi_{1}+2\nu i\chi_{1}+Q^{2p-1}\left(\frac{p(p+1)}{2}\chi_{1}^{2}+\frac{p(p-1)}{2}\overline{\chi}_{1}^{2}+p(p+1)\chi_{1}\overline{\chi}_{1}\right). D 2 ( ρ ) = − i ( 2 1 + ν ) Λ χ 1 + 2 ν i χ 1 + Q 2 p − 1 ( 2 p ( p + 1 ) χ 1 2 + 2 p ( p − 1 ) χ 1 2 + p ( p + 1 ) χ 1 χ 1 ) .
Substituting (38) and (17) into this expression, we obtain
g 2 − ( ρ ) = 0 , g 2 + ( ρ ) = ρ − 1 p + 2 ( ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ d n , 2 ρ λ + n + O ( ρ λ − ) ) . g_{2}^{-}(\rho)=0,\qquad g_{2}^{+}(\rho)=\rho^{-\frac{1}{p}+2}\left(\sum_{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor}d_{n,2}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right). g 2 − ( ρ ) = 0 , g 2 + ( ρ ) = ρ − p 1 + 2 1 ≤ n ≤ ⌊ λ + λ − ⌋ ∑ d n , 2 ρ λ + n + O ( ρ λ − ) .
Thus, v 2 − ( ρ ) = 0 v_{2}^{-}(\rho)=0 v 2 − ( ρ ) = 0 solves L − v 2 − = g 2 − L_{-}v_{2}^{-}=g_{2}^{-} L − v 2 − = g 2 − with zero initial data at the origin. In the following, we consider the behavior of v 2 + v_{2}^{+} v 2 + . Take η ∈ C ∞ ( [ 0 , ∞ ) ) \eta\in C^{\infty}([0,\infty)) η ∈ C ∞ ([ 0 , ∞ )) with η = 0 \eta=0 η = 0 on [ 0 , 1 ] [0,1] [ 0 , 1 ] and η = 1 \eta=1 η = 1 on [ 2 , ∞ ) [2,\infty) [ 2 , ∞ ) . Since Discr ∉ Q \sqrt{\operatorname{Discr}}\notin\mathbb{Q} Discr ∈ / Q , we have
4 + λ + n − λ ± ≠ 0 , 1 ≤ n ≤ ⌊ λ − λ + ⌋ . (39) 4+\lambda_{+}n-\lambda_{\pm}\ne0,\qquad1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor. \tag*{(39)} 4 + λ + n − λ ± = 0 , 1 ≤ n ≤ ⌊ λ + λ − ⌋ . (39)
Thus, we write
v 2 + ( ρ ) = η ( ρ ) ρ − 1 p + 4 ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ 4 + λ + n − λ + > 0 c n , 2 + ρ λ + n + v ~ 2 + ( ρ ) . (40) v_{2}^{+}(\rho)=\eta(\rho)\rho^{-\frac{1}{p}+4}\sum_{\substack{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor\\4+\lambda_{+}n-\lambda_{+}>0}}c_{n,2}^{+}\rho^{\lambda_{+}n}+\widetilde{v}_{2}^{+}(\rho). \tag*{(40)} v 2 + ( ρ ) = η ( ρ ) ρ − p 1 + 4 1 ≤ n ≤ ⌊ λ + λ − ⌋ 4 + λ + n − λ + > 0 ∑ c n , 2 + ρ λ + n + v 2 + ( ρ ) . (40)
where c 1 , 2 + = − d 1 , 2 4 ( 4 + λ + − λ − ) c_{1,2}^{+}=-\frac{d_{1,2}}{4(4+\lambda_{+}-\lambda_{-})} c 1 , 2 + = − 4 ( 4 + λ + − λ − ) d 1 , 2 . A direct computation gives
L + ( ρ − 1 p + a ) = − ( a − λ + ) ( a − λ − ) ρ − 1 p + a − 2 − ( 2 p + 1 ) ( ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ c ^ n + ρ λ + n + O ( ρ λ − ) ) ρ − 1 p + a − 2 . (41) \begin{aligned}
L_{+}\left(\rho^{-\frac{1}{p}+a}\right)
={}&-(a-\lambda_{+})(a-\lambda_{-})\rho^{-\frac{1}{p}+a-2}\\
&-(2p+1)\left(\sum_{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor}\widehat{c}_{n}^{+}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right)\rho^{-\frac{1}{p}+a-2}. \tag*{(41)}
\end{aligned} L + ( ρ − p 1 + a ) = − ( a − λ + ) ( a − λ − ) ρ − p 1 + a − 2 − ( 2 p + 1 ) 1 ≤ n ≤ ⌊ λ + λ − ⌋ ∑ c n + ρ λ + n + O ( ρ λ − ) ρ − p 1 + a − 2 . (41)
where c ^ n + \widehat{c}_{n}^{+} c n + depends only on c ∞ c_{\infty} c ∞ and { c m } 1 ≤ m ≤ n \{c_{m}\}_{1\le m\le n} { c m } 1 ≤ m ≤ n in (17) . The function v ~ 2 + \widetilde{v}_{2}^{+} v 2 + satisfies
L + v ~ 2 + = g ~ 2 + , L_{+}\widetilde{v}_{2}^{+}=\widetilde{g}_{2}^{+}, L + v 2 + = g 2 + ,
where
g ~ 2 + ( ρ ) = g 2 + ( ρ ) − L + ( η ( ρ ) ρ − 1 p + 4 ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ 4 + λ + n − λ + > 0 c n , 2 + ρ λ + n ) . \widetilde{g}_{2}^{+}(\rho)=g_{2}^{+}(\rho)-L_{+}\left(\eta(\rho)\rho^{-\frac{1}{p}+4}\sum_{\substack{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor\\4+\lambda_{+}n-\lambda_{+}>0}}c_{n,2}^{+}\rho^{\lambda_{+}n}\right). g 2 + ( ρ ) = g 2 + ( ρ ) − L + η ( ρ ) ρ − p 1 + 4 1 ≤ n ≤ ⌊ λ + λ − ⌋ 4 + λ + n − λ + > 0 ∑ c n , 2 + ρ λ + n .
According to (41) , we can choose the coefficients c n , 2 + c_{n,2}^{+} c n , 2 + in (40) depending on { d m , 2 } 1 ≤ m ≤ n \{d_{m,2}\}_{1\le m\le n} { d m , 2 } 1 ≤ m ≤ n and { c j + } 1 ≤ j ≤ n − 1 \{c_{j}^{+}\}_{1\le j\le n-1} { c j + } 1 ≤ j ≤ n − 1 , such that as ρ \rho ρ goes to infinity, g ~ 2 + \widetilde{g}_{2}^{+} g 2 + has the expansion
g ~ 2 + ( ρ ) = ρ − 1 p + 2 ( ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ 4 + λ + n − λ + < 0 d ~ n , 2 ρ λ + n + O ( ρ λ − ) ) . \widetilde{g}_{2}^{+}(\rho)=\rho^{-\frac{1}{p}+2}\left(\sum_{\substack{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor\\4+\lambda_{+}n-\lambda_{+}<0}}\widetilde{d}_{n,2}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right). g 2 + ( ρ ) = ρ − p 1 + 2 1 ≤ n ≤ ⌊ λ + λ − ⌋ 4 + λ + n − λ + < 0 ∑ d n , 2 ρ λ + n + O ( ρ λ − ) .
We choose the constant
C 2 = − Θ + ( 0 ) ∫ 0 ∞ Φ + ( s ) g ~ 2 + ( s ) W − 1 [ Θ + , Φ + ] ( s ) d s . C_2=-\Theta_+(0)\int_0^\infty\Phi_+(s)\widetilde{g}_2^+(s)W^{-1}[\Theta_+,\Phi_+](s)\,\mathrm{d}s. C 2 = − Θ + ( 0 ) ∫ 0 ∞ Φ + ( s ) g 2 + ( s ) W − 1 [ Θ + , Φ + ] ( s ) d s .
Then solving L + v ~ 2 + = g ~ 2 + L_+\widetilde{v}_2^+=\widetilde{g}_2^+ L + v 2 + = g 2 + with initial data v ~ 2 + ( 0 ) = − C 2 \widetilde{v}_2^+(0)=-C_2 v 2 + ( 0 ) = − C 2 , ∂ ρ v ~ 2 + ( 0 ) = 0 \partial_\rho\widetilde{v}_2^+(0)=0 ∂ ρ v 2 + ( 0 ) = 0 and using the fact (39) , we deduce that
v ~ 2 + ( ρ ) = Φ + ( ρ ) ∫ 0 ρ Θ + ( s ) g ~ 2 + ( s ) W − 1 [ Θ + , Φ + ] ( s ) d s + Θ + ( ρ ) ∫ ρ ∞ Φ + ( s ) g ~ 2 + ( s ) W − 1 [ Θ + , Φ + ] ( s ) d s . \widetilde{v}_2^+(\rho)=\Phi_+(\rho)\int_0^\rho\Theta_+(s)\widetilde{g}_2^+(s)W^{-1}[\Theta_+,\Phi_+](s)\,\mathrm{d}s
+\Theta_+(\rho)\int_\rho^\infty\Phi_+(s)\widetilde{g}_2^+(s)W^{-1}[\Theta_+,\Phi_+](s)\,\mathrm{d}s. v 2 + ( ρ ) = Φ + ( ρ ) ∫ 0 ρ Θ + ( s ) g 2 + ( s ) W − 1 [ Θ + , Φ + ] ( s ) d s + Θ + ( ρ ) ∫ ρ ∞ Φ + ( s ) g 2 + ( s ) W − 1 [ Θ + , Φ + ] ( s ) d s .
Using (24) and (39) , and combining this expansion with (40) , we obtain that as ρ \rho ρ goes to infinity,
χ 2 ( ρ ) = v 2 + ( ρ ) + i v 2 − ( ρ ) = ρ − 1 p + 4 ( ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ c n , 2 ρ λ + n + O ( ρ λ − ) ) . \chi_2(\rho)=v_2^+(\rho)+\mathrm{i}v_2^-(\rho)=\rho^{-\frac{1}{p}+4}\left(\sum_{1\le n\le\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}c_{n,2}\rho^{\lambda_+ n}+O(\rho^{\lambda_-})\right). χ 2 ( ρ ) = v 2 + ( ρ ) + i v 2 − ( ρ ) = ρ − p 1 + 4 1 ≤ n ≤ ⌊ λ + λ − ⌋ ∑ c n , 2 ρ λ + n + O ( ρ λ − ) .
We next consider the general case k ≥ 3 k\ge3 k ≥ 3 . Proceeding by induction, assume that solutions χ j \chi_j χ j , 1 ≤ j ≤ k − 1 1\le j\le k-1 1 ≤ j ≤ k − 1 , have been constructed with fixed constants C j C_j C j and satisfy (1) and (2). Using (33) together with the expansion (35) , we verify that
D k ( 1 ) ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k − 1 L } ≤ n ≤ ⌊ λ − λ + ⌋ d n , k ( 1 ) ρ λ + n + O ( ρ λ − ) ) , as ρ → ∞ . D_k^{(1)}(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\max\{1,\frac{k-1}{L}\}\le n\le\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}d_{n,k}^{(1)}\rho^{\lambda_+ n}+O(\rho^{\lambda_-})\right),\qquad\text{as }\rho\to\infty. D k ( 1 ) ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , L k − 1 } ≤ n ≤ ⌊ λ + λ − ⌋ ∑ d n , k ( 1 ) ρ λ + n + O ( ρ λ − ) , as ρ → ∞.
where d n , k ( 1 ) = − i ( k − 1 + ( 1 2 + ν ) λ + n ) c n , k − 1 d_{n,k}^{(1)}=-\mathrm{i}\left(k-1+\left(\frac{1}{2}+\nu\right)\lambda_+n\right)c_{n,k-1} d n , k ( 1 ) = − i ( k − 1 + ( 2 1 + ν ) λ + n ) c n , k − 1 . In particular, when k = m L + 1 k=mL+1 k = m L + 1 , the assumption (4) on ν \nu ν gives
d n , k ( 1 ) = − i ( k − 1 − n L ) c n , k − 1 = − i ( m − n ) L c n , m L = 0 , for n = m . d_{n,k}^{(1)}=-\mathrm{i}(k-1-nL)c_{n,k-1}=-\mathrm{i}(m-n)Lc_{n,mL}=0,\qquad\text{for }n=m. d n , k ( 1 ) = − i ( k − 1 − n L ) c n , k − 1 = − i ( m − n ) L c n , m L = 0 , for n = m .
Therefore, D k ( 1 ) D_k^{(1)} D k ( 1 ) admits the following expansion at infinity:
D k ( 1 ) ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k L } ≤ n ≤ ⌊ λ − λ + ⌋ d n , k ( 1 ) ρ λ + n + O ( ρ λ − ) ) . (42) D_k^{(1)}(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\max\{1,\frac{k}{L}\}\le n\le\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}d_{n,k}^{(1)}\rho^{\lambda_+n}+O(\rho^{\lambda_-})\right). \tag*{(42)} D k ( 1 ) ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , L k } ≤ n ≤ ⌊ λ + λ − ⌋ ∑ d n , k ( 1 ) ρ λ + n + O ( ρ λ − ) . (42)
Similarly, substituting (35) and (17) into the polynomial expression (34) gives
D k ( 2 ) ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k L } ≤ n ≤ ⌊ λ − λ + ⌋ d n , k ( 2 ) ρ λ + n + O ( ρ λ − ) ) . (43) D_k^{(2)}(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\max\{1,\frac{k}{L}\}\le n\le\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}d_{n,k}^{(2)}\rho^{\lambda_+n}+O(\rho^{\lambda_-})\right). \tag*{(43)} D k ( 2 ) ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , L k } ≤ n ≤ ⌊ λ + λ − ⌋ ∑ d n , k ( 2 ) ρ λ + n + O ( ρ λ − ) . (43)
Combining (42) and (43) , we obtain, for 3 ≤ k ≤ L ⌊ λ − λ + ⌋ 3\le k\le L\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor 3 ≤ k ≤ L ⌊ λ + λ − ⌋ ,
g k + ( ρ ) = 0 , g k − ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k L } ≤ n ≤ ⌊ λ − λ + ⌋ d n , k − ρ λ + n + O ( ρ λ − ) ) , (44) g_k^+(\rho)=0,\qquad g_k^-(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\max\{1,\frac{k}{L}\}\le n\le\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}d_{n,k}^-\rho^{\lambda_+n}+O(\rho^{\lambda_-})\right), \tag*{(44)} g k + ( ρ ) = 0 , g k − ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , L k } ≤ n ≤ ⌊ λ + λ − ⌋ ∑ d n , k − ρ λ + n + O ( ρ λ − ) , (44)
for k k k odd, and
g k + ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k L } ≤ n ≤ ⌊ λ − λ + ⌋ d n , k + ρ λ + n + O ( ρ λ − ) ) , g k − ( ρ ) = 0 , (45) g_k^+(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\max\{1,\frac{k}{L}\}\le n\le\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}d_{n,k}^+\rho^{\lambda_+n}+O(\rho^{\lambda_-})\right),\qquad g_k^-(\rho)=0, \tag*{(45)} g k + ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , L k } ≤ n ≤ ⌊ λ + λ − ⌋ ∑ d n , k + ρ λ + n + O ( ρ λ − ) , g k − ( ρ ) = 0 , (45)
for k k k even, where d n , k ± ∈ R d_{n,k}^{\pm}\in\mathbb{R} d n , k ± ∈ R . Furthermore, for k ≥ L ⌊ λ − λ + ⌋ + 1 k\ge L\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor+1 k ≥ L ⌊ λ + λ − ⌋ + 1 ,
g k ± ( ρ ) = ρ − 1 p + 2 ( k − 1 ) O ( ρ λ − ) , g_k^\pm(\rho)=\rho^{-\frac{1}{p}+2(k-1)}O(\rho^{\lambda_-}), g k ± ( ρ ) = ρ − p 1 + 2 ( k − 1 ) O ( ρ λ − ) ,
the same parity follows from (33) and (34) .
For odd k k k , g k + = 0 g_k^+=0 g k + = 0 and v k + ( 0 ) = ∂ ρ v k + ( 0 ) = 0 v_k^+(0)=\partial_\rho v_k^+(0)=0 v k + ( 0 ) = ∂ ρ v k + ( 0 ) = 0 . Uniqueness therefore gives v k + ≡ 0 v_k^+\equiv0 v k + ≡ 0 . Similarly, v k − ≡ 0 v_k^-\equiv0 v k − ≡ 0 for even k k k . It remains to consider v k − v_k^- v k − for odd k k k and v k + v_k^+ v k + for even $k. It follows from Discr ∉ Q \sqrt{\operatorname{Discr}}\notin\mathbb{Q} Discr ∈ / Q and λ + + λ − = 2 p − d + 2 ∈ Q \lambda_{+}+\lambda_{-}=\frac{2}{p}-d+2\in\mathbb{Q} λ + + λ − = p 2 − d + 2 ∈ Q that
2 k + λ + n ≠ 0 , 2 k + λ + n + 2 ( s c − 1 ) ≠ 0 , 2 k + λ + n − λ ± ≠ 0 , (46) 2k+\lambda_{+}n\ne0,\qquad2k+\lambda_{+}n+2(s_{c}-1)\ne0,\qquad2k+\lambda_{+}n-\lambda_{\pm}\ne0, \tag*{(46)} 2 k + λ + n = 0 , 2 k + λ + n + 2 ( s c − 1 ) = 0 , 2 k + λ + n − λ ± = 0 , (46)
for 1 ≤ n ≤ ⌊ λ − λ + ⌋ 1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor 1 ≤ n ≤ ⌊ λ + λ − ⌋ . Thus, in view of the expansion (24) , (44) and (45) , we write
v k − ( ρ ) = η ( ρ ) ρ − 1 p + 2 k ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ 2 k + λ + n > 0 c n , k − ρ λ + n + v ~ k − ( ρ ) , for odd k , v k + ( ρ ) = η ( ρ ) ρ − 1 p + 2 k ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ 2 k + λ + n − λ + > 0 c n , k + ρ λ + n + v ~ k + ( ρ ) , for even k . (47) \begin{aligned}
v_{k}^{-}(\rho)&=\eta(\rho)\rho^{-\frac{1}{p}+2k}\sum_{\substack{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor\\2k+\lambda_{+}n>0}}c_{n,k}^{-}\rho^{\lambda_{+}n}+\widetilde{v}_{k}^{-}(\rho),\qquad\text{for odd }k,\\
v_{k}^{+}(\rho)&=\eta(\rho)\rho^{-\frac{1}{p}+2k}\sum_{\substack{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor\\2k+\lambda_{+}n-\lambda_{+}>0}}c_{n,k}^{+}\rho^{\lambda_{+}n}+\widetilde{v}_{k}^{+}(\rho),\qquad\text{for even }k. \tag*{(47)}
\end{aligned} v k − ( ρ ) v k + ( ρ ) = η ( ρ ) ρ − p 1 + 2 k 1 ≤ n ≤ ⌊ λ + λ − ⌋ 2 k + λ + n > 0 ∑ c n , k − ρ λ + n + v k − ( ρ ) , for odd k , = η ( ρ ) ρ − p 1 + 2 k 1 ≤ n ≤ ⌊ λ + λ − ⌋ 2 k + λ + n − λ + > 0 ∑ c n , k + ρ λ + n + v k + ( ρ ) , for even k . (47)
where
c 1 , k − = − d 1 , k − ( 2 k + λ + ) ( 2 k − λ − ) , c 1 , k + = − d 1 , k + 2 k ( 2 k + λ + − λ − ) . c_{1,k}^{-}=-\frac{d_{1,k}^{-}}{(2k+\lambda_{+})(2k-\lambda_{-})},\qquad c_{1,k}^{+}=-\frac{d_{1,k}^{+}}{2k(2k+\lambda_{+}-\lambda_{-})}. c 1 , k − = − ( 2 k + λ + ) ( 2 k − λ − ) d 1 , k − , c 1 , k + = − 2 k ( 2 k + λ + − λ − ) d 1 , k + .
Similarly, we compute
L − ( ρ − 1 p + a ) = − ( a ( a − λ + − λ − ) + ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ c ^ n − ρ λ + n + O ( ρ λ − ) ) ρ − 1 p + a − 2 . (48) L_{-}\left(\rho^{-\frac{1}{p}+a}\right)=-\left(a(a-\lambda_{+}-\lambda_{-})+\sum_{1\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor}\widehat{c}_{n}^{-}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right)\rho^{-\frac{1}{p}+a-2}. \tag*{(48)} L − ( ρ − p 1 + a ) = − a ( a − λ + − λ − ) + 1 ≤ n ≤ ⌊ λ + λ − ⌋ ∑ c n − ρ λ + n + O ( ρ λ − ) ρ − p 1 + a − 2 . (48)
Using (41) and (48) , we choose c n , k − c_{n,k}^{-} c n , k − for odd k k k and c n , k + c_{n,k}^{+} c n , k + for even k k k , successively in increasing order of n n n , to cancel the terms in (44) and (45) corresponding to the explicit sums in (47) . The conditions (46) ensure that each coefficient is uniquely determined. More precisely, the coefficient c n , k − c_{n,k}^{-} c n , k − depends only on { d m , k − } max { 1 , k / L } ≤ m ≤ n \{d_{m,k}^{-}\}_{\max\{1,k/L\}\le m\le n} { d m , k − } m a x { 1 , k / L } ≤ m ≤ n , { c ^ j − } 1 ≤ j ≤ n − 1 \{\widehat{c}_{j}^{-}\}_{1\le j\le n-1} { c j − } 1 ≤ j ≤ n − 1 for odd k k k , and c n , k + c_{n,k}^{+} c n , k + depends on { d m , k + } max { 1 , k / L } ≤ m ≤ n \{d_{m,k}^{+}\}_{\max\{1,k/L\}\le m\le n} { d m , k + } m a x { 1 , k / L } ≤ m ≤ n , { c ^ j + } 1 ≤ j ≤ n − 1 \{\widehat{c}_{j}^{+}\}_{1\le j\le n-1} { c j + } 1 ≤ j ≤ n − 1 for even k k k .
Therefore, v ~ k ± \widetilde{v}_{k}^{\pm} v k ± solves
L ± v ~ k ± = g ~ k ± , L_{\pm}\widetilde{v}_{k}^{\pm}=\widetilde{g}_{k}^{\pm}, L ± v k ± = g k ± ,
where g ~ k ± \widetilde{g}_{k}^{\pm} g k ± admits the following asymptotic expansion as ρ → + ∞ \rho\to+\infty ρ → + ∞ :
g ~ k − ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k / L } ≤ n ≤ ⌊ λ − λ + ⌋ 2 k + λ + n < 0 d ~ n , k − ρ λ + n + O ( ρ λ − ) ) , for odd k . (49) \widetilde{g}_{k}^{-}(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\substack{\max\{1,k/L\}\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor\\2k+\lambda_{+}n<0}}\widetilde{d}_{n,k}^{-}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right),\qquad\text{for odd }k. \tag*{(49)} g k − ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , k / L } ≤ n ≤ ⌊ λ + λ − ⌋ 2 k + λ + n < 0 ∑ d n , k − ρ λ + n + O ( ρ λ − ) , for odd k . (49)
and
g ~ k + ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k / L } ≤ n ≤ ⌊ λ − λ + ⌋ 2 k + λ + n − λ + < 0 d ~ n , k + ρ λ + n + O ( ρ λ − ) ) , for even k . (50) \widetilde{g}_{k}^{+}(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\substack{\max\{1,k/L\}\le n\le\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor\\2k+\lambda_{+}n-\lambda_{+}<0}}\widetilde{d}_{n,k}^{+}\rho^{\lambda_{+}n}+O(\rho^{\lambda_{-}})\right),\qquad\text{for even }k. \tag*{(50)} g k + ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , k / L } ≤ n ≤ ⌊ λ + λ − ⌋ 2 k + λ + n − λ + < 0 ∑ d n , k + ρ λ + n + O ( ρ λ − ) , for even k . (50)
Finally, if 2 k + λ − < 0 2k+\lambda_{-}<0 2 k + λ − < 0 for odd k k k , or 2 k + λ − − λ + < 0 2k+\lambda_{-}-\lambda_{+}<0 2 k + λ − − λ + < 0 for even k k k , we choose
C k = ( − 1 ) ⌊ k 2 ⌋ Θ ± ( 0 ) ∫ 0 ∞ Φ ± ( s ) g ~ k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s . C_{k}=(-1)^{\left\lfloor\frac{k}{2}\right\rfloor}\Theta_{\pm}(0)\int_{0}^{\infty}\Phi_{\pm}(s)\widetilde{g}_{k}^{\pm}(s)W^{-1}[\Theta_{\pm},\Phi_{\pm}](s)\,ds. C k = ( − 1 ) ⌊ 2 k ⌋ Θ ± ( 0 ) ∫ 0 ∞ Φ ± ( s ) g k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s .
Then the asymptotic expansion at infinity follows directly from the representation
v ~ k ± ( ρ ) = Φ ± ( ρ ) ∫ 0 ρ Θ ± ( s ) g ~ k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s + Θ ± ( ρ ) ∫ ρ ∞ Φ ± ( s ) g ~ k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s . \widetilde{v}_{k}^{\pm}(\rho)=\Phi_{\pm}(\rho)\int_{0}^{\rho}\Theta_{\pm}(s)\widetilde{g}_{k}^{\pm}(s)W^{-1}[\Theta_{\pm},\Phi_{\pm}](s)\,ds+\Theta_{\pm}(\rho)\int_{\rho}^{\infty}\Phi_{\pm}(s)\widetilde{g}_{k}^{\pm}(s)W^{-1}[\Theta_{\pm},\Phi_{\pm}](s)\,ds. v k ± ( ρ ) = Φ ± ( ρ ) ∫ 0 ρ Θ ± ( s ) g k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s + Θ ± ( ρ ) ∫ ρ ∞ Φ ± ( s ) g k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s .
If 2 k + λ − > 0 2k+\lambda_{-}>0 2 k + λ − > 0 for odd k k k , or 2 k + λ − − λ + > 0 2k+\lambda_{-}-\lambda_{+}>0 2 k + λ − − λ + > 0 for even k k k , (24) gives
Θ ± ( ρ ) = O ( ρ − 1 p + 2 k + λ − ) . \Theta_{\pm}(\rho)=O\left(\rho^{-\frac{1}{p}+2k+\lambda_{-}}\right). Θ ± ( ρ ) = O ( ρ − p 1 + 2 k + λ − ) .
In this case, we fix C k = 0 C_k=0 C k = 0 as a normalization and obtain
v k ± ( ρ ) = Φ ± ( ρ ) ∫ 0 ρ Θ ± ( s ) g ~ k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s − Θ ± ( ρ ) ∫ 0 ρ Φ ± ( s ) g ~ k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s . v_k^{\pm}(\rho)=\Phi_{\pm}(\rho)\int_0^\rho\Theta_{\pm}(s)\widetilde{g}_k^{\pm}(s)W^{-1}[\Theta_{\pm},\Phi_{\pm}](s)\,\mathrm{d}s-\Theta_{\pm}(\rho)\int_0^\rho\Phi_{\pm}(s)\widetilde{g}_k^{\pm}(s)W^{-1}[\Theta_{\pm},\Phi_{\pm}](s)\,\mathrm{d}s. v k ± ( ρ ) = Φ ± ( ρ ) ∫ 0 ρ Θ ± ( s ) g k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s − Θ ± ( ρ ) ∫ 0 ρ Φ ± ( s ) g k ± ( s ) W − 1 [ Θ ± , Φ ± ] ( s ) d s .
The equality cases are excluded by D i s c r ∉ Q \sqrt{\mathrm{Discr}}\notin\mathbb{Q} Discr ∈ / Q . Using (49) , (50) , (24) , and (46) , and combining the resulting expansions with (47) , we obtain the claimed asymptotic expansion of χ k \chi_k χ k . The expansion can be differentiated any number of times with respect to ρ \rho ρ . This completes the proof of Lemma 3.1. □
Estimates for the approximate solution in the inner region
With the above notation, for any integer N i n ≥ 2 N_{\mathrm{in}}\ge2 N in ≥ 2 , set
ψ i n a p ( t , x ) = λ 1 p ( t ) U i n a p ( t , ρ ) , U i n a p ( t , ρ ) = Q ( ρ ) + ∑ k = 1 N i n t 2 ν k χ k ( ρ ) , \psi_{\mathrm{in}}^{\mathrm{ap}}(t,x)=\lambda^{\frac{1}{p}}(t)U_{\mathrm{in}}^{\mathrm{ap}}(t,\rho),\qquad U_{\mathrm{in}}^{\mathrm{ap}}(t,\rho)=Q(\rho)+\sum_{k=1}^{N_{\mathrm{in}}}t^{2\nu k}\chi_k(\rho), ψ in ap ( t , x ) = λ p 1 ( t ) U in ap ( t , ρ ) , U in ap ( t , ρ ) = Q ( ρ ) + k = 1 ∑ N in t 2 ν k χ k ( ρ ) ,
with ρ = λ ( t ) ∣ x ∣ \rho=\lambda(t)|x| ρ = λ ( t ) ∣ x ∣ . By Lemma 3.1, U i n a p U_{\mathrm{in}}^{\mathrm{ap}} U in ap satisfies the following L ∞ L^\infty L ∞ estimates in S i n S_{\mathrm{in}} S in .
Lemma 3.2. For 0 < t ≤ T 0<t\le T 0 < t ≤ T , the following estimates hold:
∥ U i n a p − Q ∥ L ∞ ( S i n ) ≤ C t ( ν − ϵ 1 ) ( 1 p + ∣ λ + ∣ ) + 2 ϵ 1 + C t 2 ν , \left\|U_{\mathrm{in}}^{\mathrm{ap}}-Q\right\|_{L^\infty(S_{\mathrm{in}})}
\le Ct^{(\nu-\epsilon_1)(\frac{1}{p}+|\lambda_+|)+2\epsilon_1}+Ct^{2\nu}, ∥ U in ap − Q ∥ L ∞ ( S in ) ≤ C t ( ν − ϵ 1 ) ( p 1 + ∣ λ + ∣ ) + 2 ϵ 1 + C t 2 ν ,
∥ ∣ ρ ∣ − k ∇ ρ l ( U i n a p − Q ) ∥ L ∞ ( S i n ) ≤ C t 2 ν , k ≥ 0 , l ≥ 1. \left\||\rho|^{-k}\nabla_\rho^l\left(U_{\mathrm{in}}^{\mathrm{ap}}-Q\right)\right\|_{L^\infty(S_{\mathrm{in}})}
\le Ct^{2\nu},\qquad k\ge0,\ l\ge1. ∣ ρ ∣ − k ∇ ρ l ( U in ap − Q ) L ∞ ( S in ) ≤ C t 2 ν , k ≥ 0 , l ≥ 1.
Along the same lines, using Lemma 3.1, we obtain the following L 2 L^2 L 2 estimates.
Lemma 3.3. With the above notation, for any integer l ≥ 1 l\ge1 l ≥ 1 , the following estimates hold for all 0 < t ≤ T 0<t\le T 0 < t ≤ T :
∥ ∇ l ( U i n a p − Q ) ∥ L 2 ( S i n ) ≤ C t − ( ν − ϵ 1 ) ( s c − ∣ λ + ∣ − l ) + 2 ϵ 1 , 0 ≤ l < s c − ∣ λ + ∣ + 2 \left\|\nabla^l\left(U_{\mathrm{in}}^{\mathrm{ap}}-Q\right)\right\|_{L^2(S_{\mathrm{in}})}
\le Ct^{-(\nu-\epsilon_1)(s_c-|\lambda_+|-l)+2\epsilon_1},\qquad0\le l<s_c-|\lambda_+|+2 ∇ l ( U in ap − Q ) L 2 ( S in ) ≤ C t − ( ν − ϵ 1 ) ( s c − ∣ λ + ∣ − l ) + 2 ϵ 1 , 0 ≤ l < s c − ∣ λ + ∣ + 2
∥ ∇ l ( U i n a p − Q ) ∥ L 2 ( S i n ) ≤ C t 2 ν , l ≥ s c − ∣ λ + ∣ + 2 , \left\|\nabla^l\left(U_{\mathrm{in}}^{\mathrm{ap}}-Q\right)\right\|_{L^2(S_{\mathrm{in}})}
\le Ct^{2\nu},\qquad l\ge s_c-|\lambda_+|+2, ∇ l ( U in ap − Q ) L 2 ( S in ) ≤ C t 2 ν , l ≥ s c − ∣ λ + ∣ + 2 ,
where s c = − 1 p + d 2 s_c=-\frac{1}{p}+\frac{d}{2} s c = − p 1 + 2 d . Furthermore, for sufficiently large d d d , one has s c − ∣ λ + ∣ ∼ d 2 s_c-|\lambda_+|\sim\frac{d}{2} s c − ∣ λ + ∣ ∼ 2 d .
We conclude this section by estimating the remainder term:
R i n a p ( t , ρ ) = − i t 1 + 2 ν ∂ t U i n a p + i ( 1 2 + ν ) t 2 ν Δ U i n a p − Δ U i n a p − ∣ U i n a p ∣ 2 p U i n a p . \mathcal{R}_{\mathrm{in}}^{\mathrm{ap}}(t,\rho)=-it^{1+2\nu}\partial_tU_{\mathrm{in}}^{\mathrm{ap}}+i\left(\frac{1}{2}+\nu\right)t^{2\nu}\Delta U_{\mathrm{in}}^{\mathrm{ap}}-\Delta U_{\mathrm{in}}^{\mathrm{ap}}-|U_{\mathrm{in}}^{\mathrm{ap}}|^{2p}U_{\mathrm{in}}^{\mathrm{ap}}. R in ap ( t , ρ ) = − i t 1 + 2 ν ∂ t U in ap + i ( 2 1 + ν ) t 2 ν Δ U in ap − Δ U in ap − ∣ U in ap ∣ 2 p U in ap .
We have the following estimate.
Lemma 3.4. For any integers N i n ≥ 2 N_{\mathrm{in}}\ge2 N in ≥ 2 and l ≥ 0 l\ge0 l ≥ 0 , there exist a positive constant C l , N i n C_{l,N_{\mathrm{in}}} C l , N in and a small positive time T = T ( l , N i n ) T=T(l,N_{\mathrm{in}}) T = T ( l , N in ) such that, for all 0 < t ≤ T 0<t\le T 0 < t ≤ T ,
∥ ∇ ρ l R i n a p ∥ L 2 ( S i n ) ≤ C l , N i n ( t 2 ν ( N i n + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ + ∣ + l − s c ) + 2 N i n ϵ 1 ) , \left\|\nabla_\rho^l\mathcal{R}_{\mathrm{in}}^{\mathrm{ap}}\right\|_{L^2(S_{\mathrm{in}})}
\le C_{l,N_{\mathrm{in}}}\left(t^{2\nu(N_{\mathrm{in}}+1)}+t^{(\nu-\epsilon_1)(|\lambda_+|+l-s_c)+2N_{\mathrm{in}}\epsilon_1}\right), ∇ ρ l R in ap L 2 ( S in ) ≤ C l , N in ( t 2 ν ( N in + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ + ∣ + l − s c ) + 2 N in ϵ 1 ) ,
where s c = − 1 p + d 2 s_c=-\frac{1}{p}+\frac{d}{2} s c = − p 1 + 2 d . In particular, for N i n = ⌊ ∣ λ − ∣ λ + ⌋ N_{\mathrm{in}}=\left\lfloor\frac{|\lambda_-|}{\lambda_+}\right\rfloor N in = ⌊ λ + ∣ λ − ∣ ⌋ ,
∥ ∇ ρ l R i n a p ∥ L 2 ( S i n ) ≤ C l , N i n ( t 2 ν ( N i n + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ − ∣ + 2 + l − s c ) + 2 ( N i n + 1 ) ϵ 1 ) . \left\|\nabla_\rho^l\mathcal{R}_{\mathrm{in}}^{\mathrm{ap}}\right\|_{L^2(S_{\mathrm{in}})}
\le C_{l,N_{\mathrm{in}}}\left(t^{2\nu(N_{\mathrm{in}}+1)}+t^{(\nu-\epsilon_1)(|\lambda_-|+2+l-s_c)+2(N_{\mathrm{in}}+1)\epsilon_1}\right). ∇ ρ l R in ap L 2 ( S in ) ≤ C l , N in ( t 2 ν ( N in + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ − ∣ + 2 + l − s c ) + 2 ( N in + 1 ) ϵ 1 ) .
Proof. Substituting the finite expansion of U i n a p U_{in}^{ap} U in a p into the definition of R i n a p R_{in}^{ap} R in a p and using (31) , the coefficients of t 2 ν k t^{2\nu k} t 2 ν k vanish for 1 ≤ k ≤ N i n 1\leq k\leq N_{in} 1 ≤ k ≤ N in . Since the nonlinearity is a polynomial of degree 2 p + 1 2p+1 2 p + 1 , we obtain
R i n a p = ∑ k = N i n + 1 ( 2 p + 1 ) N i n t 2 ν k G k , R_{in}^{ap}=\sum_{k=N_{in}+1}^{(2p+1)N_{in}}t^{2\nu k}G_k, R in a p = k = N in + 1 ∑ ( 2 p + 1 ) N in t 2 ν k G k ,
where G k G_k G k are the coefficients obtained from this finite expansion and depend only on Q Q Q and χ j \chi_j χ j , 1 ≤ j ≤ N i n 1\leq j\leq N_{in} 1 ≤ j ≤ N in .
By (30) and Lemma 3.1 , G k G_k G k is smooth and O ( 1 ) O(1) O ( 1 ) as ρ → 0 \rho\to0 ρ → 0 , and it has the following expansion:
G k ( ρ ) = ρ − 1 p + 2 ( k − 1 ) ( ∑ max { 1 , k L } ≤ n ≤ ⌊ λ − λ + ⌋ c ^ n , k ρ λ + n + O ( ρ λ − ) ) , as ρ → ∞ . (51) G_k(\rho)=\rho^{-\frac{1}{p}+2(k-1)}\left(\sum_{\max\left\{1,\frac{k}{L}\right\}\leq n\leq\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}\widehat{c}_{n,k}\rho^{\lambda_+n}+O(\rho^{\lambda_-})\right),\qquad\text{as }\rho\to\infty. \tag*{(51)} G k ( ρ ) = ρ − p 1 + 2 ( k − 1 ) m a x { 1 , L k } ≤ n ≤ ⌊ λ + λ − ⌋ ∑ c n , k ρ λ + n + O ( ρ λ − ) , as ρ → ∞. (51)
Therefore, for N i n ≥ 2 N_{in}\geq2 N in ≥ 2 and 0 < t ≤ T ≤ 1 0<t\leq T\leq1 0 < t ≤ T ≤ 1 ,
∥ ∇ ρ l R i n a p ∥ L 2 ( S i n ) ≤ C l , N i n ( t 2 ν ( N i n + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ + ∣ + l − s c ) + 2 N i n ϵ 1 ) . (52) \left\|\nabla_\rho^l R_{in}^{ap}\right\|_{L^2(S_{in})}\leq C_{l,N_{in}}\left(t^{2\nu(N_{in}+1)}+t^{(\nu-\epsilon_1)(|\lambda_+|+l-s_c)+2N_{in}\epsilon_1}\right). \tag*{(52)} ∇ ρ l R in a p L 2 ( S in ) ≤ C l , N in ( t 2 ν ( N in + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ + ∣ + l − s c ) + 2 N in ϵ 1 ) . (52)
In particular, for N i n = ⌊ λ − / λ + ⌋ N_{in}=\left\lfloor\lambda_-/\lambda_+\right\rfloor N in = ⌊ λ − / λ + ⌋ , the sum in (51) is empty for all k ≥ N i n + 1 k\geq N_{in}+1 k ≥ N in + 1 . From the differentiated expansion in Lemma 3.1 , we have
∇ ρ l G k ( ρ ) = O ( ρ − 1 p + 2 ( k − 1 ) + λ − − l ) , ρ → ∞ . \nabla_\rho^lG_k(\rho)=O\left(\rho^{-\frac{1}{p}+2(k-1)+\lambda_--l}\right),\qquad\rho\to\infty. ∇ ρ l G k ( ρ ) = O ( ρ − p 1 + 2 ( k − 1 ) + λ − − l ) , ρ → ∞.
Since D i s c r ∉ Q \sqrt{\mathit{Discr}}\notin\mathbb{Q} Discr ∈ / Q , we have 2 ( k − 1 ) − ∣ λ − ∣ − l + s c ≠ 0 2(k-1)-|\lambda_-|-l+s_c\neq0 2 ( k − 1 ) − ∣ λ − ∣ − l + s c = 0 . Hence
∥ ∇ ρ l R i n a p ∥ L 2 ( S i n ) ≤ C l , N i n ∑ k = N i n + 1 ( 2 p + 1 ) N i n t 2 ν k ( 1 + ( ∫ 1 10 t − ( ν − ϵ 1 ) ρ 4 ( k − 1 ) − 2 ∣ λ − ∣ − 2 l + 2 s c − 1 d ρ ) 1 / 2 ) . \left\|\nabla_\rho^lR_{in}^{ap}\right\|_{L^2(S_{in})}\leq C_{l,N_{in}}\sum_{k=N_{in}+1}^{(2p+1)N_{in}}t^{2\nu k}\left(1+\left(\int_1^{10t^{-(\nu-\epsilon_1)}}\rho^{4(k-1)-2|\lambda_-|-2l+2s_c-1}\,d\rho\right)^{1/2}\right). ∇ ρ l R in a p L 2 ( S in ) ≤ C l , N in k = N in + 1 ∑ ( 2 p + 1 ) N in t 2 ν k 1 + ( ∫ 1 10 t − ( ν − ϵ 1 ) ρ 4 ( k − 1 ) − 2∣ λ − ∣ − 2 l + 2 s c − 1 d ρ ) 1/2 .
Then the error estimate (52) has been improved as follows
∥ ∇ ρ l R i n a p ∥ L 2 ( S i n ) ≤ C l , N i n ( t 2 ν ( N i n + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + l + 2 ) + 2 ( N i n + 1 ) ϵ 1 ) . (53) \left\|\nabla_\rho^lR_{in}^{ap}\right\|_{L^2(S_{in})}\leq C_{l,N_{in}}\left(t^{2\nu(N_{in}+1)}+t^{(\nu-\epsilon_1)(|\lambda_-|-s_c+l+2)+2(N_{in}+1)\epsilon_1}\right). \tag*{(53)} ∇ ρ l R in a p L 2 ( S in ) ≤ C l , N in ( t 2 ν ( N in + 1 ) + t ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + l + 2 ) + 2 ( N in + 1 ) ϵ 1 ) . (53)
For d d d sufficiently large that 2 ( N i n + 1 ) ≥ ∣ λ − ∣ − s c + d + 2 2(N_{in}+1)\geq|\lambda_-|-s_c+d+2 2 ( N in + 1 ) ≥ ∣ λ − ∣ − s c + d + 2 , (53) yields
∥ ∇ ρ l R i n a p ∥ L 2 ( S i n ) ≤ C l , N i n t ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + l + 2 ) , 0 ≤ l ≤ d . (54) \left\|\nabla_\rho^lR_{in}^{ap}\right\|_{L^2(S_{in})}\leq C_{l,N_{in}}t^{(\nu-\epsilon_1)(|\lambda_-|-s_c+l+2)},\qquad0\leq l\leq d. \tag*{(54)} ∇ ρ l R in a p L 2 ( S in ) ≤ C l , N in t ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + l + 2 ) , 0 ≤ l ≤ d . (54)
This completes the proof. □
Approximate solution in the self-similar region
General scheme for constructing the approximate solution in the self-similar region
In this section, we construct an approximate solution ψ s s a p \psi_{ss}^{ap} ψ ss a p to (1) in the region
S s s = { ( t , x ) | 1 10 t − ( ν − ϵ 1 ) ≤ ρ ≤ 10 t − ( ν + ϵ 2 ) } , ρ = λ ( t ) ∣ x ∣ . S_{ss}=\left\{(t,x)\,\middle|\,\frac{1}{10}t^{-(\nu-\epsilon_1)}\leq\rho\leq10t^{-(\nu+\epsilon_2)}\right\},\qquad\rho=\lambda(t)|x|. S ss = { ( t , x ) 10 1 t − ( ν − ϵ 1 ) ≤ ρ ≤ 10 t − ( ν + ϵ 2 ) } , ρ = λ ( t ) ∣ x ∣.
This approximation extends the approximate solution ψ i n a p \psi_{in}^{ap} ψ in a p constructed in the inner region S i n S_{in} S in . Here 0 < ϵ 2 < 1 2 0<\epsilon_2<\frac{1}{2} 0 < ϵ 2 < 2 1 is fixed later.
We seek a radial solution of (1) in the following form:
ψ ( t , x ) = t − 1 2 p y − 1 p ( c ∞ + v ( t , y ) ) , with y = t − 1 2 ∣ x ∣ . (55) \psi(t,x)=t^{-\frac{1}{2p}}y^{-\frac{1}{p}}\left(c_\infty+v(t,y)\right),\qquad\text{with }y=t^{-\frac{1}{2}}|x|. \tag*{(55)} ψ ( t , x ) = t − 2 p 1 y − p 1 ( c ∞ + v ( t , y ) ) , with y = t − 2 1 ∣ x ∣. (55)
By straightforward computations, we find that
ψ t ( t , x ) = t − 1 2 p y − 1 p ( v t ( t , y ) − 1 2 t y ∂ y v ( t , y ) ) , \psi_t(t,x)=t^{-\frac{1}{2p}}y^{-\frac{1}{p}}\left(v_t(t,y)-\frac{1}{2t}y\partial_yv(t,y)\right), ψ t ( t , x ) = t − 2 p 1 y − p 1 ( v t ( t , y ) − 2 t 1 y ∂ y v ( t , y ) ) ,
Δ ψ ( t , x ) = t − 1 2 p − 1 y − 1 p ( − c ∞ 2 p + 1 y − 2 − c ∞ 2 p y − 2 v ( t , y ) + Δ v ( t , y ) − 2 p y − 1 ∂ y v ( t , y ) ) , \Delta\psi(t,x)=t^{-\frac{1}{2p}-1}y^{-\frac{1}{p}}\left(-c_\infty^{2p+1}y^{-2}-c_\infty^{2p}y^{-2}v(t,y)+\Delta v(t,y)-\frac{2}{p}y^{-1}\partial_yv(t,y)\right), Δ ψ ( t , x ) = t − 2 p 1 − 1 y − p 1 ( − c ∞ 2 p + 1 y − 2 − c ∞ 2 p y − 2 v ( t , y ) + Δ v ( t , y ) − p 2 y − 1 ∂ y v ( t , y ) ) ,
∣ ψ ( t , x ) ∣ 2 p ψ ( t , x ) = t − 1 2 p − 1 y − 1 p − 2 ( c ∞ 2 p + 1 + ( p + 1 ) c ∞ 2 p v ( t , y ) + p c ∞ 2 p v ‾ ( t , y ) + W ~ ( t , y ) ) . |\psi(t,x)|^{2p}\psi(t,x)=t^{-\frac{1}{2p}-1}y^{-\frac{1}{p}-2}\left(c_\infty^{2p+1}+(p+1)c_\infty^{2p}v(t,y)+pc_\infty^{2p}\overline{v}(t,y)+\widetilde{W}(t,y)\right). ∣ ψ ( t , x ) ∣ 2 p ψ ( t , x ) = t − 2 p 1 − 1 y − p 1 − 2 ( c ∞ 2 p + 1 + ( p + 1 ) c ∞ 2 p v ( t , y ) + p c ∞ 2 p v ( t , y ) + W ( t , y ) ) .
with
W ~ ( t , y ) = ∣ c ∞ + v ( t , y ) ∣ 2 p ( c ∞ + v ( t , y ) ) − c ∞ 2 p + 1 − ( p + 1 ) c ∞ 2 p v ( t , y ) − p c ∞ 2 p v ‾ ( t , y ) . \widetilde{W}(t,y)=|c_\infty+v(t,y)|^{2p}(c_\infty+v(t,y))-c_\infty^{2p+1}-(p+1)c_\infty^{2p}v(t,y)-pc_\infty^{2p}\overline{v}(t,y). W ( t , y ) = ∣ c ∞ + v ( t , y ) ∣ 2 p ( c ∞ + v ( t , y )) − c ∞ 2 p + 1 − ( p + 1 ) c ∞ 2 p v ( t , y ) − p c ∞ 2 p v ( t , y ) .
Substituting the ansatz (55) into (1) and multiplying the equation by t 1 2 p + 1 y 1 p t^{\frac{1}{2p}+1}y^{\frac{1}{p}} t 2 p 1 + 1 y p 1 , we find that v v v satisfies
− i t v t − Δ v + 2 p y − 1 ∂ y v + i 2 y ∂ y v − p c ∞ 2 p y − 2 ( v + v ‾ ) − y − 2 W ~ = 0. (56) -itv_t-\Delta v+\frac{2}{p}y^{-1}\partial_yv+\frac{i}{2}y\partial_yv-pc_\infty^{2p}y^{-2}(v+\overline{v})-y^{-2}\widetilde{W}=0. \tag*{(56)} − i t v t − Δ v + p 2 y − 1 ∂ y v + 2 i y ∂ y v − p c ∞ 2 p y − 2 ( v + v ) − y − 2 W = 0. (56)
Combining (17) with (35) , we take N i n = L [ λ − λ + ] N_{\mathrm{in}}=L\left[\frac{\lambda_-}{\lambda_+}\right] N in = L [ λ + λ − ] . In the limit ρ → + ∞ \rho\to+\infty ρ → + ∞ , y → 0 y\to0 y → 0 , with y = t ν ρ y=t^\nu\rho y = t ν ρ , the identity
t 2 ν k ρ n λ + + 2 k = t ν ∣ λ + ∣ n y n λ + + 2 k t^{2\nu k}\rho^{n\lambda_++2k}=t^{\nu|\lambda_+|n}y^{n\lambda_++2k} t 2 ν k ρ n λ + + 2 k = t ν ∣ λ + ∣ n y n λ + + 2 k
allows us to rewrite the inner expansion as
ψ i n a p ( t , x ) = t − 1 2 p y − 1 p ( c ∞ + ∑ 1 ≤ n ≤ ⌊ λ − λ + ⌋ t ν ∣ λ + ∣ n ∑ 0 ≤ k ≤ L n c n , k y λ + n + 2 k + O ( t − ν λ − y λ − ) ) . (57) \psi_{\mathrm{in}}^{\mathrm{ap}}(t,x)=t^{-\frac{1}{2p}}y^{-\frac{1}{p}}\left(c_\infty+\sum_{1\le n\le\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor}t^{\nu|\lambda_+|n}\sum_{0\le k\le Ln}c_{n,k}y^{\lambda_+n+2k}+O(t^{-\nu\lambda_-}y^{\lambda_-})\right). \tag*{(57)} ψ in ap ( t , x ) = t − 2 p 1 y − p 1 c ∞ + 1 ≤ n ≤ ⌊ λ + λ − ⌋ ∑ t ν ∣ λ + ∣ n 0 ≤ k ≤ L n ∑ c n , k y λ + n + 2 k + O ( t − ν λ − y λ − ) . (57)
where c n , 0 = c n c_{n,0}=c_n c n , 0 = c n and c n c_n c n are the coefficients in (17) .
Combining (57) with (55) suggests looking for the approximate solution v v v in S s s S_{ss} S ss in the form
v ( t , y ) = ∑ n ≥ 1 t ν ∣ λ + ∣ n v n ( y ) . (58) v(t,y)=\sum_{n\ge1}t^{\nu|\lambda_+|n}v_n(y). \tag*{(58)} v ( t , y ) = n ≥ 1 ∑ t ν ∣ λ + ∣ n v n ( y ) . (58)
Substituting the ansatz (58) into (56) , we deduce that the function v ⃗ n = ( v n v ‾ n ) \vec{v}_n=\begin{pmatrix}v_n\\ \overline{v}_n\end{pmatrix} v n = ( v n v n ) satisfies
L ~ n v ⃗ n = G n ( v 1 , … , v n − 1 ) , (59) \widetilde{L}_n\vec{v}_n=G_n(v_1,\ldots,v_{n-1}), \tag*{(59)} L n v n = G n ( v 1 , … , v n − 1 ) , (59)
where L ~ n \widetilde{L}_n L n denotes the operator
L ~ n = ( − Δ + 2 p y − 1 ∂ y − p c ∞ 2 p y − 2 ) σ 3 − p c ∞ 2 p y − 2 σ 3 σ 1 + i 2 y ∂ y + i ν λ + n , \widetilde{L}_n=\left(-\Delta+\frac{2}{p}y^{-1}\partial_y-pc_\infty^{2p}y^{-2}\right)\sigma_3-pc_\infty^{2p}y^{-2}\sigma_3\sigma_1+\frac{i}{2}y\partial_y+i\nu\lambda_+n, L n = ( − Δ + p 2 y − 1 ∂ y − p c ∞ 2 p y − 2 ) σ 3 − p c ∞ 2 p y − 2 σ 3 σ 1 + 2 i y ∂ y + i ν λ + n ,
and the source term G n = ( g n − g ‾ n ) G_n=\begin{pmatrix}g_n\\-\overline{g}_n\end{pmatrix} G n = ( g n − g n ) is 0 0 0 for n = 1 n=1 n = 1 . For n ≥ 2 n\ge2 n ≥ 2 , it is determined by the expression
y − 2 W ~ ( t , y ) = ∑ n ≥ 2 t ν ∣ λ + ∣ n g n ( y ) . (60) y^{-2}\widetilde{W}(t,y)=\sum_{n\ge2}t^{\nu|\lambda_+|n}g_n(y). \tag*{(60)} y − 2 W ( t , y ) = n ≥ 2 ∑ t ν ∣ λ + ∣ n g n ( y ) . (60)
where W ~ ( t , y ) \widetilde{W}(t,y) W ( t , y ) is defined in (55) .
Analysis of the vector functions v n v_n v n
To determine the solutions v n v_n v n of the recurrence system (59) , we first investigate the homogeneous equation
L ~ n f ⃗ = 0. (61) \widetilde{\mathcal{L}}_n\vec{f}=0. \tag*{(61)} L n f = 0. (61)
To treat this equation, we split f f f into its real and imaginary parts, f = f + + i f − f=f_++if_- f = f + + i f − , and rewrite (61) as
L ~ + f + − ( 1 2 y ∂ y + ν λ + ) f − = 0 , L ~ − f − + ( 1 2 y ∂ y + ν λ + ) f + = 0. (62) \begin{aligned}
\widetilde{L}_+f_+-\left(\frac{1}{2}y\partial_y+\nu\lambda_+\right)f_-&=0,\\
\widetilde{L}_-f_-+\left(\frac{1}{2}y\partial_y+\nu\lambda_+\right)f_+&=0. \tag*{(62)}
\end{aligned} L + f + − ( 2 1 y ∂ y + ν λ + ) f − L − f − + ( 2 1 y ∂ y + ν λ + ) f + = 0 , = 0. (62)
where
L ~ + = − Δ + 2 p y − 1 ∂ y − 2 p c 2 p y − 2 , and L ~ − = − Δ + 2 p y − 1 ∂ y . \widetilde{L}_+=-\Delta+\frac{2}{p}y^{-1}\partial_y-2pc_2^p y^{-2},\qquad\text{and}\qquad\widetilde{L}_-=-\Delta+\frac{2}{p}y^{-1}\partial_y. L + = − Δ + p 2 y − 1 ∂ y − 2 p c 2 p y − 2 , and L − = − Δ + p 2 y − 1 ∂ y .
The following lemma gives fundamental systems for L ~ ± \widetilde{L}_{\pm} L ± .
Lemma 4.1. A fundamental system { e 1 + , e 2 + } \{e_1^+,e_2^+\} { e 1 + , e 2 + } for L ~ + \widetilde{L}_+ L + is given by
e 1 + ( y ) = y λ + , e 2 + ( y ) = y λ − , e_1^+(y)=y^{\lambda_+},\qquad e_2^+(y)=y^{\lambda_-}, e 1 + ( y ) = y λ + , e 2 + ( y ) = y λ − , and
e 1 − ( y ) = 1 , e 2 − ( y ) = y λ − + λ + . e_1^-(y)=1,\qquad e_2^-(y)=y^{\lambda_-+\lambda_+}. e 1 − ( y ) = 1 , e 2 − ( y ) = y λ − + λ + . In addition, the Wronskians are
W [ e 1 + , e 2 + ] = ( λ + − λ − ) y λ + + λ − − 1 , W [ e 1 − , e 2 − ] = − ( λ − + λ + ) y λ + + λ − − 1 . W[e_1^+,e_2^+]=(\lambda_+-\lambda_-)y^{\lambda_++\lambda_--1},\qquad W[e_1^-,e_2^-]=-(\lambda_-+\lambda_+)y^{\lambda_++\lambda_--1}. W [ e 1 + , e 2 + ] = ( λ + − λ − ) y λ + + λ − − 1 , W [ e 1 − , e 2 − ] = − ( λ − + λ + ) y λ + + λ − − 1 .
We now proceed by induction to show that, under the prescribed blowup rate, the system (59) admits a finite term polynomial solution. The precise statement is the following.
Lemma 4.2. (i) For any real constant v 1 , 0 v_{1,0} v 1 , 0 , the system (59) admits a unique solution of the form
v n ( y ) = ∑ k = 0 n L v n , k y λ + n + 2 k , (63) v_n(y)=\sum_{k=0}^{nL}v_{n,k}y^{\lambda_+n+2k}, \tag*{(63)} v n ( y ) = k = 0 ∑ n L v n , k y λ + n + 2 k , (63) where for each n ≥ 2 n\ge2 n ≥ 2 , the coefficient v n , 0 v_{n,0} v n , 0 is uniquely determined by v m , 0 v_{m,0} v m , 0 , 1 ≤ m ≤ n − 1 1\le m\le n-1 1 ≤ m ≤ n − 1 .
(ii) Alternatively, the solution to (59) is uniquely characterized by the value v 1 , L v_{1,L} v 1 , L .
Proof. We first consider the case n = 1 n=1 n = 1 . Fix an arbitrary real number v 1 , 0 v_{1,0} v 1 , 0 . To construct a solution of the form (63) , we determine the coefficients in the formal expansion
v 1 ( y ) = ∑ k ≥ 0 v 1 , k y λ + + 2 k , v 1 , k = v 1 , k + + i v 1 , k − , v_1(y)=\sum_{k\ge0}v_{1,k}y^{\lambda_++2k},\qquad v_{1,k}=v_{1,k}^++iv_{1,k}^-, v 1 ( y ) = k ≥ 0 ∑ v 1 , k y λ + + 2 k , v 1 , k = v 1 , k + + i v 1 , k − ,
with v 1 , 0 + = v 1 , 0 v_{1,0}^+=v_{1,0} v 1 , 0 + = v 1 , 0 and v 1 , 0 − = 0 v_{1,0}^-=0 v 1 , 0 − = 0 . Thus,
v 1 + ( y ) = ∑ k ≥ 0 v 1 , k + y λ + + 2 k , v 1 − ( y ) = ∑ k ≥ 1 v 1 , k − y λ + + 2 k . v_1^+(y)=\sum_{k\ge0}v_{1,k}^+y^{\lambda_++2k},\qquad v_1^-(y)=\sum_{k\ge1}v_{1,k}^-y^{\lambda_++2k}. v 1 + ( y ) = k ≥ 0 ∑ v 1 , k + y λ + + 2 k , v 1 − ( y ) = k ≥ 1 ∑ v 1 , k − y λ + + 2 k .
By (62) , these functions satisfy
{ L ~ + v 1 + = ( 1 2 y ∂ y + ν λ + ) v 1 − , L ~ − v 1 − = − ( 1 2 y ∂ y + ν λ + ) v 1 + . \left\{
\begin{aligned}
\widetilde{L}_+v_1^+&=\left(\frac{1}{2}y\partial_y+\nu\lambda_+\right)v_1^-,\\
\widetilde{L}_-v_1^-&=-\left(\frac{1}{2}y\partial_y+\nu\lambda_+\right)v_1^+.
\end{aligned}
\right. ⎩ ⎨ ⎧ L + v 1 + L − v 1 − = ( 2 1 y ∂ y + ν λ + ) v 1 − , = − ( 2 1 y ∂ y + ν λ + ) v 1 + .
Substituting the ansatz into this system, using ( 1 2 + ν ) ∣ λ + ∣ = L \left(\frac{1}{2}+\nu\right)|\lambda_{+}|=L ( 2 1 + ν ) ∣ λ + ∣ = L , Lemma 4.1 and comparing the coefficients of y λ + + 2 k − 2 y^{\lambda_{+}+2k-2} y λ + + 2 k − 2 , we obtain for k ≥ 1 k\geq1 k ≥ 1
v 1 , k − = ( k − L − 1 ) v 1 , k − 1 + ( λ + + 2 k ) ( − λ − + 2 k ) , v 1 , k + = − ( k − L − 1 ) v 1 , k − 1 − 2 k ( λ + − λ − + 2 k ) (64) v_{1,k}^{-}=\frac{(k-L-1)v_{1,k-1}^{+}}{(\lambda_{+}+2k)(-\lambda_{-}+2k)},\qquad v_{1,k}^{+}=-\frac{(k-L-1)v_{1,k-1}^{-}}{2k(\lambda_{+}-\lambda_{-}+2k)} \tag*{(64)} v 1 , k − = ( λ + + 2 k ) ( − λ − + 2 k ) ( k − L − 1 ) v 1 , k − 1 + , v 1 , k + = − 2 k ( λ + − λ − + 2 k ) ( k − L − 1 ) v 1 , k − 1 − (64)
where v 1 , 0 + = v 1 , 0 v_{1,0}^{+}=v_{1,0} v 1 , 0 + = v 1 , 0 and these recurrence relations are well defined in view of D i s c r ∉ Q \sqrt{\mathit{Discr}}\notin\mathbb{Q} Discr ∈ / Q . It follows that v 1 , k − = 0 v_{1,k}^{-}=0 v 1 , k − = 0 for all k ≥ L + 1 k\geq L+1 k ≥ L + 1 . Hence, the system admits a unique solution of the form
v 1 ( y ) = ∑ k = 0 L v 1 , k y λ + + 2 k , v_{1}(y)=\sum_{k=0}^{L}v_{1,k}y^{\lambda_{+}+2k}, v 1 ( y ) = k = 0 ∑ L v 1 , k y λ + + 2 k ,
where v 1 , 0 ∈ R v_{1,0}\in\mathbb{R} v 1 , 0 ∈ R is arbitrary and for 1 ≤ k ≤ L 1\leq k\leq L 1 ≤ k ≤ L , v 1 , k = v 1 , k + + i v 1 , k − v_{1,k}=v_{1,k}^{+}+iv_{1,k}^{-} v 1 , k = v 1 , k + + i v 1 , k − is uniquely determined by the coefficient v 1 , 0 v_{1,0} v 1 , 0 .
We now turn to the general case n ≥ 2 n\geq2 n ≥ 2 . To proceed by induction, we assume that, for each 1 ≤ j ≤ n − 1 1\leq j\leq n-1 1 ≤ j ≤ n − 1 , the function v j ( y ) v_j(y) v j ( y ) satisfies the expansion (63) . Following the same strategy as in the case n = 1 n=1 n = 1 , we decompose v n v_n v n and g n g_n g n into real and imaginary parts. It follows from (59) that the functions v n ± v_n^{\pm} v n ± satisfy the following coupled system
{ L ~ + v n + − ( y 2 ∂ y + ν λ + n ) v n − = g n + , L ~ − v n − + ( y 2 ∂ y + ν λ + n ) v n + = g n − . \begin{cases}
\widetilde{\mathcal{L}}_{+}v_n^{+}-\left(\dfrac{y}{2}\partial_y+\nu\lambda_{+n}\right)v_n^{-}=g_n^{+},\\
\widetilde{\mathcal{L}}_{-}v_n^{-}+\left(\dfrac{y}{2}\partial_y+\nu\lambda_{+n}\right)v_n^{+}=g_n^{-}.
\end{cases} ⎩ ⎨ ⎧ L + v n + − ( 2 y ∂ y + ν λ + n ) v n − = g n + , L − v n − + ( 2 y ∂ y + ν λ + n ) v n + = g n − .
By (60) and the expansion of v j v_j v j , the source term g n = g n + + i g n − g_n=g_n^{+}+ig_n^{-} g n = g n + + i g n − can be written as
g n ± ( y ) = ∑ k = 0 n L g n , k ± y λ + n + 2 k − 2 = ∑ k = − 1 n L − 1 g n , k + 1 ± y λ + n + 2 k (65) g_n^{\pm}(y)=\sum_{k=0}^{nL}g_{n,k}^{\pm}y^{\lambda_{+}n+2k-2}=\sum_{k=-1}^{nL-1}g_{n,k+1}^{\pm}y^{\lambda_{+}n+2k} \tag*{(65)} g n ± ( y ) = k = 0 ∑ n L g n , k ± y λ + n + 2 k − 2 = k = − 1 ∑ n L − 1 g n , k + 1 ± y λ + n + 2 k (65)
where the parity of the lower order coefficients gives g n , k + = 0 g_{n,k}^{+}=0 g n , k + = 0 for odd k k k and g n , k − = 0 g_{n,k}^{-}=0 g n , k − = 0 for even k k k . We look for v n v_n v n in the form
v n ( y ) = ∑ k ≥ 0 v n , k y λ + n + 2 k , v n , k = v n , k + + i v n , k − . v_n(y)=\sum_{k\geq0}v_{n,k}y^{\lambda_{+}n+2k},\qquad v_{n,k}=v_{n,k}^{+}+iv_{n,k}^{-}. v n ( y ) = k ≥ 0 ∑ v n , k y λ + n + 2 k , v n , k = v n , k + + i v n , k − .
Comparing the coefficients of y n λ + − 2 y^{n\lambda_{+}-2} y n λ + − 2 in (65) gives
v n , 0 + = − g n , 0 + ( n − 1 ) λ + ( n λ + − λ − ) , v n , 0 − = 0. v_{n,0}^{+}=-\frac{g_{n,0}^{+}}{(n-1)\lambda_{+}(n\lambda_{+}-\lambda_{-})},\qquad v_{n,0}^{-}=0. v n , 0 + = − ( n − 1 ) λ + ( n λ + − λ − ) g n , 0 + , v n , 0 − = 0.
Then, the ansatz for v n ± v_n^{\pm} v n ± is given by
v n + ( y ) = ∑ k ≥ 0 v n , k + y λ + n + 2 k , v n − ( y ) = ∑ k ≥ 1 v n , k − y λ + n + 2 k . v_n^{+}(y)=\sum_{k\geq0}v_{n,k}^{+}y^{\lambda_{+}n+2k},\qquad v_n^{-}(y)=\sum_{k\geq1}v_{n,k}^{-}y^{\lambda_{+}n+2k}. v n + ( y ) = k ≥ 0 ∑ v n , k + y λ + n + 2 k , v n − ( y ) = k ≥ 1 ∑ v n , k − y λ + n + 2 k .
Substituting this ansatz into the system and using 1 2 + ν = L ∣ λ + ∣ \frac{1}{2}+\nu=\frac{L}{|\lambda_{+}|} 2 1 + ν = ∣ λ + ∣ L , we get
( y 2 ∂ y + ν λ + n ) v n + ( y ) = ∑ k ≥ 0 v n , k + ( k − n L ) y λ + n + 2 k , ( y 2 ∂ y + ν λ + n ) v n − ( y ) = ∑ k ≥ 1 v n , k − ( k − n L ) y λ + n + 2 k . (66) \begin{aligned}
\left(\frac{y}{2}\partial_y+\nu\lambda_{+}n\right)v_n^{+}(y)&=\sum_{k\geq0}v_{n,k}^{+}(k-nL)y^{\lambda_{+}n+2k},\\
\left(\frac{y}{2}\partial_y+\nu\lambda_{+}n\right)v_n^{-}(y)&=\sum_{k\geq1}v_{n,k}^{-}(k-nL)y^{\lambda_{+}n+2k}. \tag*{(66)}
\end{aligned} ( 2 y ∂ y + ν λ + n ) v n + ( y ) ( 2 y ∂ y + ν λ + n ) v n − ( y ) = k ≥ 0 ∑ v n , k + ( k − n L ) y λ + n + 2 k , = k ≥ 1 ∑ v n , k − ( k − n L ) y λ + n + 2 k . (66)
Using (65) , (66) , and Lemma 4.1 , comparison of the coefficients of y n λ + + 2 k − 2 y^{n\lambda_{+}+2k-2} y n λ + + 2 k − 2 yields
v n , k − = − g n , k − − ( k − n L − 1 ) v n , k − 1 + ( λ + n + 2 k ) ( λ + ( n − 1 ) − λ − + 2 k ) , k ≥ 1 , v n , k + = − g n , k + + ( k − n L − 1 ) v n , k − 1 − ( λ + ( n − 1 ) + 2 k ) ( λ + n − λ − + 2 k ) , k ≥ 1. (67) \begin{aligned}
v^{-}_{n,k}&=-\frac{g^{-}_{n,k}-(k-nL-1)v^{+}_{n,k-1}}{(\lambda_{+}n+2k)(\lambda_{+}(n-1)-\lambda_{-}+2k)},\qquad k\geq1,\\
v^{+}_{n,k}&=-\frac{g^{+}_{n,k}+(k-nL-1)v^{-}_{n,k-1}}{(\lambda_{+}(n-1)+2k)(\lambda_{+}n-\lambda_{-}+2k)},\qquad k\geq1. \tag*{(67)}
\end{aligned} v n , k − v n , k + = − ( λ + n + 2 k ) ( λ + ( n − 1 ) − λ − + 2 k ) g n , k − − ( k − n L − 1 ) v n , k − 1 + , k ≥ 1 , = − ( λ + ( n − 1 ) + 2 k ) ( λ + n − λ − + 2 k ) g n , k + + ( k − n L − 1 ) v n , k − 1 − , k ≥ 1. (67)
These relations are well defined by Discr ∉ Q \sqrt{\operatorname{Discr}}\notin\mathbb{Q} Discr ∈ / Q and determine all coefficients uniquely. Since g n , k ± = 0 g^{\pm}_{n,k}=0 g n , k ± = 0 for k ≥ n L + 1 k\geq nL+1 k ≥ n L + 1 , the vanishing of k − n L − 1 k-nL-1 k − n L − 1 at k = n L + 1 k=nL+1 k = n L + 1 gives v n , n L + 1 ± = 0 v^{\pm}_{n,nL+1}=0 v n , n L + 1 ± = 0 . The recurrence relations then imply v n , k ± = 0 v^{\pm}_{n,k}=0 v n , k ± = 0 for every k ≥ n L + 1 k\geq nL+1 k ≥ n L + 1 . Consequently,
v n ( y ) = ∑ k = 0 n L v n , k y λ + n + 2 k , v_n(y)=\sum_{k=0}^{nL}v_{n,k}y^{\lambda_{+}n+2k}, v n ( y ) = k = 0 ∑ n L v n , k y λ + n + 2 k ,
where the coefficients are determined uniquely by (67) . Finally, for n = 1 n=1 n = 1 , the factor k − L − 1 k-L-1 k − L − 1 is nonzero for 1 ≤ k ≤ L 1\leq k\leq L 1 ≤ k ≤ L . Using the recurrence relations (64) determines v 1 , 0 v_{1,0} v 1 , 0 uniquely from v 1 , L v_{1,L} v 1 , L . Thus the solution to (59) is also uniquely characterized by v 1 , L v_{1,L} v 1 , L . □
Remark 4.3. The matching condition (57) fixes the leading coefficients by v n , 0 = c n v_{n,0}=c_n v n , 0 = c n for all 1 ≤ n ≤ ⌊ λ − λ + ⌋ 1\leq n\leq\left\lfloor\frac{\lambda_{-}}{\lambda_{+}}\right\rfloor 1 ≤ n ≤ ⌊ λ + λ − ⌋ . Although only v 1 , 0 v_{1,0} v 1 , 0 is chosen freely in the preceding proof, the coefficient equations for k = 0 k=0 k = 0 coincide with those arising from the stationary equation satisfied by Q Q Q , so choosing v 1 , 0 = c 1 v_{1,0}=c_1 v 1 , 0 = c 1 yields v n , 0 = c n v_{n,0}=c_n v n , 0 = c n . Since both expansions satisfy the same coefficient recursions (64) and (67) under y = t ν ρ y=t^{\nu}\rho y = t ν ρ , choosing v 1 , 0 = c 1 v_{1,0}=c_1 v 1 , 0 = c 1 gives v n , k = c n , k v_{n,k}=c_{n,k} v n , k = c n , k by uniqueness. Thus the inner expansion and the self-similar expansion connect at the common overlap.
Estimates for the approximate solution in the self-similar region
With the notation introduced above, for any integer N s s ≥ 2 N_{\mathrm{ss}}\geq2 N ss ≥ 2 , set
ψ s s a p ( t , x ) = λ 1 p ( t ) U s s a p ( t , λ ( t ) ∣ x ∣ ) , U s s a p ( t , ρ ) = ρ − 1 p ( c ∞ + v s s a p ( t , t ν ρ ) ) . (68) \psi^{\mathrm{ap}}_{\mathrm{ss}}(t,x)=\lambda^{\frac{1}{p}}(t)U^{\mathrm{ap}}_{\mathrm{ss}}(t,\lambda(t)|x|),\qquad U^{\mathrm{ap}}_{\mathrm{ss}}(t,\rho)=\rho^{-\frac{1}{p}}\left(c_{\infty}+v^{\mathrm{ap}}_{\mathrm{ss}}(t,t^{\nu}\rho)\right). \tag*{(68)} ψ ss ap ( t , x ) = λ p 1 ( t ) U ss ap ( t , λ ( t ) ∣ x ∣ ) , U ss ap ( t , ρ ) = ρ − p 1 ( c ∞ + v ss ap ( t , t ν ρ ) ) . (68)
with
v s s a p ( t , y ) = ∑ n = 1 N s s t ν ∣ λ + ∣ n ∑ k = 0 n L v n , k y λ + n + 2 k . (69) v^{\mathrm{ap}}_{\mathrm{ss}}(t,y)=\sum_{n=1}^{N_{\mathrm{ss}}}t^{\nu|\lambda_{+}|n}\sum_{k=0}^{nL}v_{n,k}y^{\lambda_{+}n+2k}. \tag*{(69)} v ss ap ( t , y ) = n = 1 ∑ N ss t ν ∣ λ + ∣ n k = 0 ∑ n L v n , k y λ + n + 2 k . (69)
The aim of this subsection is to estimate the function U s s a p U^{\mathrm{ap}}_{\mathrm{ss}} U ss ap defined by (68) in the self-similar region S s s S_{\mathrm{ss}} S ss and to study the remainder term for N s s N_{\mathrm{ss}} N ss sufficiently large. Recalling that L ≥ 2 L\geq2 L ≥ 2 and λ + + 2 L = − 2 ν λ + \lambda_{+}+2L=-2\nu\lambda_{+} λ + + 2 L = − 2 ν λ + , we obtain the following lemma from (68) and (69) .
Lemma 4.4. There exist a positive constant C C C and a small positive time T = T ( p , d , L ) T=T(p,d,L) T = T ( p , d , L ) such that, for any k , l ≥ 0 k,l\geq0 k , l ≥ 0 and all 0 < t ≤ T 0<t\leq T 0 < t ≤ T , the following L ∞ L^{\infty} L ∞ estimates hold:
∥ ρ − k ∇ ρ l U s s a p ∥ L ∞ ( S s s ) ≤ C t ( ν − ϵ 1 ) 1 p , \left\|\rho^{-k}\nabla_{\rho}^{l}U^{\mathrm{ap}}_{\mathrm{ss}}\right\|_{L^{\infty}(S_{\mathrm{ss}})}\leq Ct^{(\nu-\epsilon_{1})\frac{1}{p}}, ρ − k ∇ ρ l U ss ap L ∞ ( S ss ) ≤ C t ( ν − ϵ 1 ) p 1 , and
∥ ρ − k ∇ ρ l ( U s s a p − Q ) ∥ L ∞ ( S s s ) ≤ C ( t 2 ν + ( ν − ϵ 1 ) ( 1 p + ∣ λ + ∣ − 2 ) + t ( ν + ϵ 2 ) 1 p + ν ∣ λ + ∣ ( 1 − 2 ϵ 2 ) + t ( ν − ϵ 1 ) ( 1 p + ∣ λ + ∣ ( N s s + 1 ) ) + t ( ν − ϵ 1 ) ( 1 p + ∣ λ − ∣ ) ) . \left\|\rho^{-k}\nabla_{\rho}^{l}\left(U^{\mathrm{ap}}_{\mathrm{ss}}-Q\right)\right\|_{L^{\infty}(S_{\mathrm{ss}})}
\leq C\left(t^{2\nu+(\nu-\epsilon_{1})(\frac{1}{p}+|\lambda_{+}|-2)}+t^{(\nu+\epsilon_{2})\frac{1}{p}+\nu|\lambda_{+}|(1-2\epsilon_{2})}+t^{(\nu-\epsilon_{1})(\frac{1}{p}+|\lambda_{+}|(N_{\mathrm{ss}}+1))}+t^{(\nu-\epsilon_{1})(\frac{1}{p}+|\lambda_{-}|)}\right). ρ − k ∇ ρ l ( U ss ap − Q ) L ∞ ( S ss ) ≤ C ( t 2 ν + ( ν − ϵ 1 ) ( p 1 + ∣ λ + ∣ − 2 ) + t ( ν + ϵ 2 ) p 1 + ν ∣ λ + ∣ ( 1 − 2 ϵ 2 ) + t ( ν − ϵ 1 ) ( p 1 + ∣ λ + ∣ ( N ss + 1 )) + t ( ν − ϵ 1 ) ( p 1 + ∣ λ − ∣ ) ) .
In particular, for N s s = ⌊ λ − λ + ⌋ N_{ss}=\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor N ss = ⌊ λ + λ − ⌋ ,
∥ ρ − k ∇ ρ l ( U s s a p − Q ) ∥ L ∞ ( S s s ) ≤ C ( t 2 ν + ( ν − ε 1 ) ( 1 p + ∣ λ + ∣ − 2 ) + t ( ν + ε 2 ) 1 p + ν ∣ λ + ∣ ( 1 − 2 ε 2 ) + t ( ν − ε 1 ) ( 1 p + ∣ λ − ∣ ) ) . \left\|\rho^{-k}\nabla_\rho^l\left(U_{ss}^{ap}-Q\right)\right\|_{L^\infty(S_{ss})}
\leq C\left(t^{2\nu+(\nu-\varepsilon_1)\left(\frac{1}{p}+|\lambda_+|-2\right)}
+t^{(\nu+\varepsilon_2)\frac{1}{p}+\nu|\lambda_+|(1-2\varepsilon_2)}
+t^{(\nu-\varepsilon_1)\left(\frac{1}{p}+|\lambda_-|\right)}\right). ρ − k ∇ ρ l ( U ss a p − Q ) L ∞ ( S ss ) ≤ C ( t 2 ν + ( ν − ε 1 ) ( p 1 + ∣ λ + ∣ − 2 ) + t ( ν + ε 2 ) p 1 + ν ∣ λ + ∣ ( 1 − 2 ε 2 ) + t ( ν − ε 1 ) ( p 1 + ∣ λ − ∣ ) ) .
In the spirit of Lemma 3.3, we have the following result.
Lemma 4.5. The following L 2 L^2 L 2 estimate holds for all 0 < t ≤ T 0<t\leq T 0 < t ≤ T :
∥ ρ − k ∇ ρ l ( U s s a p − Q ) ∥ L 2 ( S s s ) ≤ C t ν ∣ λ + ∣ ( 1 − 2 ε 2 ) − ( ν + ε 2 ) s c , 0 ≤ k + l < s c − ∣ λ + ∣ + 2 , \left\|\rho^{-k}\nabla_\rho^l\left(U_{ss}^{ap}-Q\right)\right\|_{L^2(S_{ss})}
\leq Ct^{\nu|\lambda_+|(1-2\varepsilon_2)-(\nu+\varepsilon_2)s_c},
\qquad0\leq k+l<s_c-|\lambda_+|+2, ρ − k ∇ ρ l ( U ss a p − Q ) L 2 ( S ss ) ≤ C t ν ∣ λ + ∣ ( 1 − 2 ε 2 ) − ( ν + ε 2 ) s c , 0 ≤ k + l < s c − ∣ λ + ∣ + 2 , ∥ ρ − k ∇ ρ l ( U s s a p − Q ) ∥ L 2 ( S s s ) ≤ C t 2 ν − 2 ε 2 ( L − 1 ) , s c − ∣ λ + ∣ + 2 ≤ k + l < s c + 2 ν ∣ λ + ∣ , \left\|\rho^{-k}\nabla_\rho^l\left(U_{ss}^{ap}-Q\right)\right\|_{L^2(S_{ss})}
\leq Ct^{2\nu-2\varepsilon_2(L-1)},
\qquad s_c-|\lambda_+|+2\leq k+l<s_c+2\nu|\lambda_+|, ρ − k ∇ ρ l ( U ss a p − Q ) L 2 ( S ss ) ≤ C t 2 ν − 2 ε 2 ( L − 1 ) , s c − ∣ λ + ∣ + 2 ≤ k + l < s c + 2 ν ∣ λ + ∣ , ∥ ρ − k ∇ ρ l ( U s s a p − Q ) ∥ L 2 ( S s s ) ≤ C t 2 ν , k + l ≥ s c + 2 ν ∣ λ + ∣ . \left\|\rho^{-k}\nabla_\rho^l\left(U_{ss}^{ap}-Q\right)\right\|_{L^2(S_{ss})}
\leq Ct^{2\nu},
\qquad k+l\geq s_c+2\nu|\lambda_+|. ρ − k ∇ ρ l ( U ss a p − Q ) L 2 ( S ss ) ≤ C t 2 ν , k + l ≥ s c + 2 ν ∣ λ + ∣. where s c = − 1 p + d 2 s_c=-\frac{1}{p}+\frac{d}{2} s c = − p 1 + 2 d and − ∣ λ + ∣ + 2 L = 2 ν ∣ λ + ∣ -|\lambda_+|+2L=2\nu|\lambda_+| − ∣ λ + ∣ + 2 L = 2 ν ∣ λ + ∣ .
Let us now consider the remainder function
R s s a p ( t , ρ ) = − i t 1 + 2 ν ∂ t U s s a p + ( 1 2 + ν ) i t 2 ν Λ U s s a p − Δ U s s a p − ∣ U s s a p ∣ 2 p U s s a p . \mathcal{R}_{ss}^{ap}(t,\rho)=-it^{1+2\nu}\partial_tU_{ss}^{ap}+\left(\frac{1}{2}+\nu\right)it^{2\nu}\Lambda U_{ss}^{ap}-\Delta U_{ss}^{ap}-|U_{ss}^{ap}|^{2p}U_{ss}^{ap}. R ss a p ( t , ρ ) = − i t 1 + 2 ν ∂ t U ss a p + ( 2 1 + ν ) i t 2 ν Λ U ss a p − Δ U ss a p − ∣ U ss a p ∣ 2 p U ss a p .
This gives
R s s a p ( t , ρ ) = 1 ρ 1 p ( t 1 2 λ ( t ) ) 2 R ~ s s a p ( t , ρ t 1 2 λ ( t ) ) , \mathcal{R}_{ss}^{ap}(t,\rho)=\frac{1}{\rho^{\frac{1}{p}}\left(t^{\frac{1}{2}}\lambda(t)\right)^2}\widetilde{\mathcal{R}}_{ss}^{ap}\left(t,\frac{\rho}{t^{\frac{1}{2}}\lambda(t)}\right), R ss a p ( t , ρ ) = ρ p 1 ( t 2 1 λ ( t ) ) 2 1 R ss a p ( t , t 2 1 λ ( t ) ρ ) ,
where
R ~ s s a p ( t , y ) = − i t ∂ t v s s a p − Δ v s s a p + 2 p y − 1 ∂ y v s s a p + i 2 y ∂ y v s s a p − p c 2 p y − 2 ( v s s a p + v s s a p ‾ ) − y − 2 W ~ ( v s s a p ) . \widetilde{\mathcal{R}}_{ss}^{ap}(t,y)=-it\partial_tv_{ss}^{ap}-\Delta v_{ss}^{ap}+\frac{2}{p}y^{-1}\partial_yv_{ss}^{ap}+\frac{i}{2}y\partial_yv_{ss}^{ap}-pc^{2p}y^{-2}\left(v_{ss}^{ap}+\overline{v_{ss}^{ap}}\right)-y^{-2}\widetilde{W}(v_{ss}^{ap}). R ss a p ( t , y ) = − i t ∂ t v ss a p − Δ v ss a p + p 2 y − 1 ∂ y v ss a p + 2 i y ∂ y v ss a p − p c 2 p y − 2 ( v ss a p + v ss a p ) − y − 2 W ( v ss a p ) .
By construction, using (59) and (60) , we deduce that
R ~ s s a p ( t , y ) = − ∑ n = N s s + 1 ( 2 p + 1 ) N s s t ν ∣ λ + ∣ n g n ( y ) . \widetilde{\mathcal{R}}_{ss}^{ap}(t,y)=-\sum_{n=N_{ss}+1}^{(2p+1)N_{ss}}t^{\nu|\lambda_+|n}g_n(y). R ss a p ( t , y ) = − n = N ss + 1 ∑ ( 2 p + 1 ) N ss t ν ∣ λ + ∣ n g n ( y ) .
According to the computations in Lemma 4.2 , g n ( y ) g_n(y) g n ( y ) satisfies
g n ( y ) = y − 2 ∑ m = 2 min { n , 2 p + 1 } c ∞ 2 p + 1 − m ∑ a + b = m 0 ≤ a ≤ p + 1 0 ≤ b ≤ p ( p + 1 a ) ( p b ) ∑ j 1 + ⋯ + j m = n 1 ≤ j 1 , … , j m ≤ N s s ( ∏ μ = 1 a v j μ ( y ) ) ( ∏ μ = a + 1 m v j μ ( y ) ‾ ) . g_n(y)=y^{-2}\sum_{m=2}^{\min\{n,2p+1\}}c_\infty^{2p+1-m}
\sum_{\substack{a+b=m\\0\leq a\leq p+1\\0\leq b\leq p}}
\binom{p+1}{a}\binom{p}{b}
\sum_{\substack{j_1+\cdots+j_m=n\\1\leq j_1,\ldots,j_m\leq N_{ss}}}
\left(\prod_{\mu=1}^{a}v_{j_\mu}(y)\right)
\left(\prod_{\mu=a+1}^{m}\overline{v_{j_\mu}(y)}\right). g n ( y ) = y − 2 m = 2 ∑ m i n { n , 2 p + 1 } c ∞ 2 p + 1 − m a + b = m 0 ≤ a ≤ p + 1 0 ≤ b ≤ p ∑ ( a p + 1 ) ( b p ) j 1 + ⋯ + j m = n 1 ≤ j 1 , … , j m ≤ N ss ∑ ( μ = 1 ∏ a v j μ ( y ) ) ( μ = a + 1 ∏ m v j μ ( y ) ) .
Using (63) , we obtain the expansion of g n g_n g n ,
g n ( y ) = ∑ k = 0 n L g n , k y λ + n + 2 k − 2 , g_n(y)=\sum_{k=0}^{nL}g_{n,k}y^{\lambda_+n+2k-2}, g n ( y ) = k = 0 ∑ n L g n , k y λ + n + 2 k − 2 ,
where the coefficients g n , k g_{n,k} g n , k are determined by (58) and (63) . As a direct consequence of these properties, we obtain the following lemma.
Lemma 4.6. There exists a small positive time T = T ( N s s ) T=T(N_{ss}) T = T ( N ss ) such that, for every integer l ≥ 0 l\geq0 l ≥ 0 and all 0 < t ≤ T 0<t\leq T 0 < t ≤ T ,
∥ ∇ ρ l R s s a p ∥ L 2 ( S s s ) ≤ C N s s ( t ( ν − ε 1 ) ( ∣ λ + ∣ ( N s s + 1 ) − s c + l + 2 ) + t ν ∣ λ + ∣ ( 1 − 2 ε 2 ) ( N s s + 1 ) + ( ν + ε 2 ) ( l − s c + 2 ) ) . (70) \left\|\nabla_\rho^l\mathcal{R}_{ss}^{ap}\right\|_{L^2(S_{ss})}
\leq C_{N_{ss}}\left(t^{(\nu-\varepsilon_1)(|\lambda_+|(N_{ss}+1)-s_c+l+2)}
+t^{\nu|\lambda_+|(1-2\varepsilon_2)(N_{ss}+1)+(\nu+\varepsilon_2)(l-s_c+2)}\right). \tag*{(70)} ∇ ρ l R ss a p L 2 ( S ss ) ≤ C N ss ( t ( ν − ε 1 ) ( ∣ λ + ∣ ( N ss + 1 ) − s c + l + 2 ) + t ν ∣ λ + ∣ ( 1 − 2 ε 2 ) ( N ss + 1 ) + ( ν + ε 2 ) ( l − s c + 2 ) ) . (70)
We end this section by investigating ψ s s a p − ψ i n a p \psi_{ss}^{ap}-\psi_{in}^{ap} ψ ss a p − ψ in a p in the intersection of the inner and self-similar regions, namely
S i n ∩ S s s = { ρ ∈ R + : 1 10 t − ( ν − ϵ 1 ) ≤ ρ ≤ 10 t − ( ν − ϵ 1 ) } . S_{in}\cap S_{ss}=\left\{\rho\in\mathbb{R}_{+}:\frac{1}{10}t^{-(\nu-\epsilon_{1})}\leq\rho\leq10t^{-(\nu-\epsilon_{1})}\right\}. S in ∩ S ss = { ρ ∈ R + : 10 1 t − ( ν − ϵ 1 ) ≤ ρ ≤ 10 t − ( ν − ϵ 1 ) } .
In view of (57) , (68) , and (69) , for N i n = L [ λ − λ + ] N_{in}=L\left[\frac{\lambda_{-}}{\lambda_{+}}\right] N in = L [ λ + λ − ] and N s s ≥ 2 N_{ss}\geq2 N ss ≥ 2 , we have
U i n a p ( t , ρ ) − U s s a p ( t , ρ ) = ρ − 1 p ( ∑ n = 1 [ λ − λ + ] ∑ 0 ≤ k ≤ n L c n , k t ν ∣ λ + ∣ n y λ + n + 2 k + O ( t ν ∣ λ − ∣ y λ − ) − ∑ n = 1 N s s ∑ 0 ≤ k ≤ n L v n , k t ν ∣ λ + ∣ n y λ + n + 2 k ) . (71) U_{in}^{ap}(t,\rho)-U_{ss}^{ap}(t,\rho)=\rho^{-\frac{1}{p}}\left(\sum_{n=1}^{\left[\frac{\lambda_{-}}{\lambda_{+}}\right]}\sum_{0\leq k\leq nL}c_{n,k}t^{\nu|\lambda_{+}|n}y^{\lambda_{+}n+2k}+O\left(t^{\nu|\lambda_{-}|}y^{\lambda_{-}}\right)-\sum_{n=1}^{N_{ss}}\sum_{0\leq k\leq nL}v_{n,k}t^{\nu|\lambda_{+}|n}y^{\lambda_{+}n+2k}\right). \tag*{(71)} U in a p ( t , ρ ) − U ss a p ( t , ρ ) = ρ − p 1 n = 1 ∑ [ λ + λ − ] 0 ≤ k ≤ n L ∑ c n , k t ν ∣ λ + ∣ n y λ + n + 2 k + O ( t ν ∣ λ − ∣ y λ − ) − n = 1 ∑ N ss 0 ≤ k ≤ n L ∑ v n , k t ν ∣ λ + ∣ n y λ + n + 2 k . (71)
where v n , 0 = c n , 0 = c n v_{n,0}=c_{n,0}=c_n v n , 0 = c n , 0 = c n and c n c_n c n are the coefficients of the stationary solution Q Q Q . By uniqueness in Lemma 4.2 , for t t t sufficiently small, (71) leads to the following result.
Lemma 4.7. For any integers l ≥ 0 l\geq0 l ≥ 0 and N s s ≥ 2 N_{ss}\geq2 N ss ≥ 2 , the following estimate holds on S i n ∩ S s s S_{in}\cap S_{ss} S in ∩ S ss for all 0 < t ≤ T 0<t\leq T 0 < t ≤ T :
∣ ∇ ρ l ( U s s a p ( t , ρ ) − U i n a p ( t , ρ ) ) ∣ ≤ C ρ − 1 p − l ( t ( ν − ϵ 1 ) ∣ λ + ∣ ( N s s + 1 ) + t ( ν − ϵ 1 ) ∣ λ − ∣ ) . \left|\nabla_{\rho}^{l}\left(U_{ss}^{ap}(t,\rho)-U_{in}^{ap}(t,\rho)\right)\right|\leq C\rho^{-\frac{1}{p}-l}\left(t^{(\nu-\epsilon_{1})|\lambda_{+}|(N_{ss}+1)}+t^{(\nu-\epsilon_{1})|\lambda_{-}|}\right). ∇ ρ l ( U ss a p ( t , ρ ) − U in a p ( t , ρ ) ) ≤ C ρ − p 1 − l ( t ( ν − ϵ 1 ) ∣ λ + ∣ ( N ss + 1 ) + t ( ν − ϵ 1 ) ∣ λ − ∣ ) . In particular, for N s s = [ λ − λ + ] N_{ss}=\left[\frac{\lambda_{-}}{\lambda_{+}}\right] N ss = [ λ + λ − ] , there holds
∣ ∇ ρ l ( U s s a p ( t , ρ ) − U i n a p ( t , ρ ) ) ∣ ≤ C ρ − 1 p − l t ( ν − ϵ 1 ) ∣ λ − ∣ . (72) \left|\nabla_{\rho}^{l}\left(U_{ss}^{ap}(t,\rho)-U_{in}^{ap}(t,\rho)\right)\right|\leq C\rho^{-\frac{1}{p}-l}t^{(\nu-\epsilon_{1})|\lambda_{-}|}. \tag*{(72)} ∇ ρ l ( U ss a p ( t , ρ ) − U in a p ( t , ρ ) ) ≤ C ρ − p 1 − l t ( ν − ϵ 1 ) ∣ λ − ∣ . (72)
Approximate solution in the remote region
General scheme for constructing the approximate solution in the remote region
In the previous section, we constructed an approximate solution ψ s s a p \psi_{ss}^{ap} ψ ss a p in the self-similar region, extending the inner approximation ψ i n a p \psi_{in}^{ap} ψ in a p developed in Section 3. We now define the remote region by
S o u t = { ( t , ρ ) : ρ ≥ 1 10 t − ( ν + ϵ 2 ) } , ρ = λ ( t ) ∣ x ∣ . S_{out}=\left\{(t,\rho):\rho\geq\frac{1}{10}t^{-(\nu+\epsilon_{2})}\right\},\qquad\rho=\lambda(t)|x|. S o u t = { ( t , ρ ) : ρ ≥ 10 1 t − ( ν + ϵ 2 ) } , ρ = λ ( t ) ∣ x ∣.
Our objective in this section is to construct an approximate solution in S o u t S_{out} S o u t with prescribed asymptotic data for r ∼ 1 r\sim1 r ∼ 1 . These data are determined by matching the solution with ψ s s a p \psi_{ss}^{ap} ψ ss a p in the overlap S s s ∩ S o u t S_{ss}\cap S_{out} S ss ∩ S o u t .
Recalling the approximate solution ψ s s a p \psi_{ss}^{ap} ψ ss a p constructed in § 5 and introducing the change of variables r = t 1 2 y r=t^{\frac{1}{2}}y r = t 2 1 y , we observe that, as y → + ∞ y\to+\infty y → + ∞ ,
ψ s s a p ( t , x ) = r − 1 p ( c ∞ + ∑ n = 1 N s s t − ν λ + n ∑ k = 0 n L v n , k ( t − 1 2 r ) λ + n + 2 k ) . \psi_{ss}^{ap}(t,x)=r^{-\frac{1}{p}}\left(c_{\infty}+\sum_{n=1}^{N_{ss}}t^{-\nu\lambda_{+}n}\sum_{k=0}^{nL}v_{n,k}(t^{-\frac{1}{2}}r)^{\lambda_{+}n+2k}\right). ψ ss a p ( t , x ) = r − p 1 ( c ∞ + n = 1 ∑ N ss t − ν λ + n k = 0 ∑ n L v n , k ( t − 2 1 r ) λ + n + 2 k ) .
Using the identity − 1 2 − ν = L λ + -\frac{1}{2}-\nu=\frac{L}{\lambda_{+}} − 2 1 − ν = λ + L and interchanging the summations,
ψ s s a p ( t , x ) = r − 1 p ∑ q ≥ 0 t q ∑ q / L ≤ n ≤ N s s v n , n L − q r − 2 ν λ + n − 2 q . (73) \psi_{ss}^{ap}(t,x)=r^{-\frac{1}{p}}\sum_{q\geq0}t^{q}\sum_{q/L\leq n\leq N_{ss}}v_{n,nL-q}r^{-2\nu\lambda_{+}n-2q}. \tag*{(73)} ψ ss a p ( t , x ) = r − p 1 q ≥ 0 ∑ t q q / L ≤ n ≤ N ss ∑ v n , n L − q r − 2 ν λ + n − 2 q . (73)
where v 0 , 0 = c ∞ v_{0,0}=c_{\infty} v 0 , 0 = c ∞ and v n , 0 = c n v_{n,0}=c_n v n , 0 = c n .
Let Θ 0 \Theta_0 Θ 0 be a radial smooth cutoff function on R d \mathbb{R}^d R d that equals 1 1 1 on the unit ball centered at the origin and vanishes outside the ball of radius 2 2 2 centered at the origin. Motivated by (73) , for a small parameter δ > 0 \delta> 0 δ > 0 , we define the compactly supported function
f 0 ( r ) = Θ δ ( r ) ∑ 0 ≤ n ≤ N s s a n r − 1 p − 2 ν λ + n . (74) f_0(r)=\Theta_\delta(r)\sum_{0\le n\le N_{ss}}a_n r^{-\frac{1}{p}-2\nu\lambda+n}. \tag*{(74)} f 0 ( r ) = Θ δ ( r ) 0 ≤ n ≤ N ss ∑ a n r − p 1 − 2 ν λ + n . (74)
where a 0 = c ∞ a_0=c_\infty a 0 = c ∞ , a n = v n , n L a_n=v_{n,nL} a n = v n , n L , and Θ δ ( r ) = Θ 0 ( r δ ) \Theta_\delta(r)=\Theta_0\left(\frac{r}{\delta}\right) Θ δ ( r ) = Θ 0 ( δ r ) .
We seek the solution in the remote region in the form
ψ o u t ( t , r ) = f 0 ( r ) + ∑ k ≥ 1 t k g k ( r ) . (75) \psi_{out}(t,r)=f_0(r)+\sum_{k\ge1}t^k\mathfrak{g}_k(r). \tag*{(75)} ψ o u t ( t , r ) = f 0 ( r ) + k ≥ 1 ∑ t k g k ( r ) . (75)
To this end, we follow the strategy of Sections 3 and 4 and determine the functions g k \mathfrak{g}_k g k , k ≥ 1 k\ge1 k ≥ 1 , by induction, using the fact that ψ o u t \psi_{out} ψ o u t is a formal solution to the Cauchy problem
{ i ∂ t ψ o u t + Δ ψ o u t + ∣ ψ o u t ∣ 2 p ψ o u t = 0 , ψ o u t ∣ t = 0 = f 0 . (76) \begin{cases}
i\partial_t\psi_{out}+\Delta\psi_{out}+|\psi_{out}|^{2p}\psi_{out}=0,\\
\left.\psi_{out}\right|_{t=0}=f_0. \tag*{(76)}
\end{cases} { i ∂ t ψ o u t + Δ ψ o u t + ∣ ψ o u t ∣ 2 p ψ o u t = 0 , ψ o u t ∣ t = 0 = f 0 . (76)
Substituting (75) into (76) , we obtain the following recurrence relations for g ⃗ k = ( g k g ‾ k ) \vec{\mathfrak{g}}_k=\left(\begin{matrix}\mathfrak{g}_k\\\overline{\mathfrak{g}}_k\end{matrix}\right) g k = ( g k g k ) ,
g 1 = i ( Δ f 0 + ∣ f 0 ∣ 2 p f 0 ) . (77) \mathfrak{g}_1=i\left(\Delta f_0+|f_0|^{2p}f_0\right). \tag*{(77)} g 1 = i ( Δ f 0 + ∣ f 0 ∣ 2 p f 0 ) . (77)
g k + 1 = − i k + 1 H k + 1 ( r ; f 0 , g j , 1 ≤ j ≤ k ) , k ≥ 1. (78) \mathfrak{g}_{k+1}=-\frac{i}{k+1}\mathcal{H}_{k+1}(r;f_0,\mathfrak{g}_j,1\le j\le k),\qquad k\ge1. \tag*{(78)} g k + 1 = − k + 1 i H k + 1 ( r ; f 0 , g j , 1 ≤ j ≤ k ) , k ≥ 1. (78)
We decompose the source term H k + 1 \mathcal{H}_{k+1} H k + 1 in the above identity into two parts:
H k + 1 = H k + 1 ( 1 ) + H k + 1 ( 2 ) . \mathcal{H}_{k+1}=\mathcal{H}_{k+1}^{(1)}+\mathcal{H}_{k+1}^{(2)}. H k + 1 = H k + 1 ( 1 ) + H k + 1 ( 2 ) .
where the linear part is given by
H k + 1 ( 1 ) = ( − Δ − ( p + 1 ) ∣ f 0 ∣ 2 p ) g k − p ∣ f 0 ∣ 2 p − 2 f 0 2 g ‾ k , k ≥ 1. (79) \mathcal{H}_{k+1}^{(1)}=\left(-\Delta-(p+1)|f_0|^{2p}\right)\mathfrak{g}_k-p|f_0|^{2p-2}f_0^2\overline{\mathfrak{g}}_k,\qquad k\ge1. \tag*{(79)} H k + 1 ( 1 ) = ( − Δ − ( p + 1 ) ∣ f 0 ∣ 2 p ) g k − p ∣ f 0 ∣ 2 p − 2 f 0 2 g k , k ≥ 1. (79)
For the nonlinear contribution H k + 1 ( 2 ) \mathcal{H}_{k+1}^{(2)} H k + 1 ( 2 ) , we have H 2 ( 2 ) = 0 \mathcal{H}_2^{(2)}=0 H 2 ( 2 ) = 0 , while for any k ≥ 2 k\ge2 k ≥ 2 it is defined by
H k + 1 ( 2 ) = ∑ 0 ≤ a ≤ p + 1 , 0 ≤ b ≤ p 2 ≤ a + b ≤ min { k , 2 p + 1 } f 0 p + 1 − a f ‾ 0 p − b ∑ j 1 + ⋯ + j a + ℓ 1 + ⋯ + ℓ b = k j i ≥ 1 , ℓ q ≥ 1 c j 1 , … , j a , ℓ 1 , … , ℓ b a , b g j 1 ⋯ g j a g ‾ ℓ 1 ⋯ g ‾ ℓ b . (80) \mathcal{H}_{k+1}^{(2)}
=
\sum_{\substack{0\le a\le p+1,\ 0\le b\le p\\2\le a+b\le\min\{k,2p+1\}}}
f_0^{p+1-a}\overline{f}_0^{p-b}
\sum_{\substack{j_1+\cdots+j_a+\ell_1+\cdots+\ell_b=k\\j_i\ge1,\ \ell_q\ge1}}
c_{j_1,\ldots,j_a,\ell_1,\ldots,\ell_b}^{a,b}
\mathfrak{g}_{j_1}\cdots\mathfrak{g}_{j_a}\overline{\mathfrak{g}}_{\ell_1}\cdots\overline{\mathfrak{g}}_{\ell_b}. \tag*{(80)} H k + 1 ( 2 ) = 0 ≤ a ≤ p + 1 , 0 ≤ b ≤ p 2 ≤ a + b ≤ m i n { k , 2 p + 1 } ∑ f 0 p + 1 − a f 0 p − b j 1 + ⋯ + j a + ℓ 1 + ⋯ + ℓ b = k j i ≥ 1 , ℓ q ≥ 1 ∑ c j 1 , … , j a , ℓ 1 , … , ℓ b a , b g j 1 ⋯ g j a g ℓ 1 ⋯ g ℓ b . (80)
where the coefficients c j 1 , … , j a , ℓ 1 , … , ℓ b a , b c_{j_1,\ldots,j_a,\ell_1,\ldots,\ell_b}^{a,b} c j 1 , … , j a , ℓ 1 , … , ℓ b a , b are determined through the expansion
− ( ∣ f 0 + ∑ k ≥ 1 t k g k ∣ 2 p ( f 0 + ∑ k ≥ 1 t k g k ) − ∣ f 0 ∣ 2 p f 0 ) -\left(\left|f_0+\sum_{k\ge1}t^k\mathfrak{g}_k\right|^{2p}\left(f_0+\sum_{k\ge1}t^k\mathfrak{g}_k\right)-|f_0|^{2p}f_0\right) − f 0 + k ≥ 1 ∑ t k g k 2 p ( f 0 + k ≥ 1 ∑ t k g k ) − ∣ f 0 ∣ 2 p f 0
− ( p + 1 ) ∣ f 0 ∣ 2 p ∑ k ≥ 1 t k g k − p ∣ f 0 ∣ 2 p − 2 f 0 2 ∑ k ≥ 1 t k g ‾ k = ∑ k ≥ 2 t k H k + 1 ( 2 ) ( r ) . -(p+1)|f_0|^{2p}\sum_{k\ge1}t^k\mathfrak{g}_k-p|f_0|^{2p-2}f_0^2\sum_{k\ge1}t^k\overline{\mathfrak{g}}_k
=\sum_{k\ge2}t^k\mathcal{H}_{k+1}^{(2)}(r). − ( p + 1 ) ∣ f 0 ∣ 2 p k ≥ 1 ∑ t k g k − p ∣ f 0 ∣ 2 p − 2 f 0 2 k ≥ 1 ∑ t k g k = k ≥ 2 ∑ t k H k + 1 ( 2 ) ( r ) .
Analysis of the functions gk
We now investigate the functions g k g_k g k defined by (78) . We begin with the following definition.
Definition 5.1. We denote by A \mathcal{A} A the set of functions a ∈ C ∞ ( R + , C ) a \in C^\infty(\mathbb{R}_{+},\mathbb{C}) a ∈ C ∞ ( R + , C ) supported in { 0 < r ≤ 2 δ } \{0<r\le2\delta\} { 0 < r ≤ 2 δ } , where δ \delta δ is the positive parameter introduced in (74) , and admitting, for r < δ r<\delta r < δ , an absolutely convergent expansion of the form
a ( r ) = ∑ n ≥ 0 a n r − 1 2 − 2 ν λ + n , a n ∈ C . a(r)=\sum_{n\ge0}a_n r^{-\frac{1}{2}-2\nu\lambda+n},\qquad a_n\in\mathbb{C}. a ( r ) = n ≥ 0 ∑ a n r − 2 1 − 2 ν λ + n , a n ∈ C .
Remark 5.2. For any function a ∈ A a\in\mathcal{A} a ∈ A and any integer l ≥ 0 l\ge0 l ≥ 0 , we have
∇ l a ∈ r − l A . (81) \nabla^l a\in r^{-l}\mathcal{A}. \tag*{(81)} ∇ l a ∈ r − l A . (81)
We next establish the following lemma, which describes the behavior of the functions g k g_k g k .
Lemma 5.3. There exists δ 0 ( N o u t ) > 0 \delta_0(N_{\mathrm{out}})>0 δ 0 ( N out ) > 0 such that, for all positive real δ ≤ δ 0 ( N o u t ) \delta\le\delta_0(N_{\mathrm{out}}) δ ≤ δ 0 ( N out ) and any k ∈ N ∗ k\in\mathbb{N}^* k ∈ N ∗ ,
g k ∈ r − 2 k A . g_k\in r^{-2k}\mathcal{A}. g k ∈ r − 2 k A . In particular, g 0 = f 0 g_0=f_0 g 0 = f 0 .
Proof. First, (74) implies that f 0 ∈ A f_0\in\mathcal{A} f 0 ∈ A for any δ > 0 \delta>0 δ > 0 . By (77) and (81) , there exists δ 0 ( N o u t ) > 0 \delta_0(N_{\mathrm{out}})>0 δ 0 ( N out ) > 0 such that
g 1 ∈ r − 2 A , g_1\in r^{-2}\mathcal{A}, g 1 ∈ r − 2 A ,
for any δ ≤ δ 0 ( N o u t ) \delta\le\delta_0(N_{\mathrm{out}}) δ ≤ δ 0 ( N out ) .
We now show that, for any δ ≤ δ 0 ( N o u t ) \delta\le\delta_0(N_{\mathrm{out}}) δ ≤ δ 0 ( N out ) , the functions satisfy g k ∈ r − 2 k A g_k\in r^{-2k}\mathcal{A} g k ∈ r − 2 k A for all k ≥ 2 k\ge2 k ≥ 2 . We proceed by induction, assuming that, for each integer j ≤ k − 1 j\le k-1 j ≤ k − 1 , the function g j g_j g j belongs to r − 2 j A r^{-2j}\mathcal{A} r − 2 j A .
By Remark 5.2, (81) implies that − Δ g k − 1 ∈ r − 2 k A -\Delta g_{k-1}\in r^{-2k}\mathcal{A} − Δ g k − 1 ∈ r − 2 k A . Since f 0 ∈ A f_0\in\mathcal{A} f 0 ∈ A and g k − 1 ∈ r − 2 ( k − 1 ) A g_{k-1}\in r^{-2(k-1)}\mathcal{A} g k − 1 ∈ r − 2 ( k − 1 ) A , the term H k ( 1 ) \mathcal{H}_k^{(1)} H k ( 1 ) defined in (79) belongs to r − 2 k A r^{-2k}\mathcal{A} r − 2 k A . For H k ( 2 ) \mathcal{H}_k^{(2)} H k ( 2 ) , using (80) and the induction hypothesis g j ∈ r − 2 j A g_j\in r^{-2j}\mathcal{A} g j ∈ r − 2 j A for all j ≤ k − 1 j\le k-1 j ≤ k − 1 , we deduce that H k ( 2 ) \mathcal{H}_k^{(2)} H k ( 2 ) also belongs to r − 2 k A r^{-2k}\mathcal{A} r − 2 k A . This completes the proof. □
Remark 5.4. Combining Definition 5.1 with Lemma 5.3, we infer that, for any integer k ≥ 1 k\ge1 k ≥ 1 , the function g k g_k g k appearing in the asymptotic formula (75) admits an absolutely convergent expansion of the form
g k ( r ) = ∑ n ≥ 0 b n k r − 1 2 − 2 ν λ + n − 2 k , (82) g_k(r)=\sum_{n\ge0}b_n^k r^{-\frac{1}{2}-2\nu\lambda+n-2k}, \tag*{(82)} g k ( r ) = n ≥ 0 ∑ b n k r − 2 1 − 2 ν λ + n − 2 k , (82)
for r < δ r<\delta r < δ , where the coefficients b n k b_n^k b n k satisfy
b 0 0 = c ∞ , b n 0 = v n , n L for 0 ≤ n ≤ N s s , b n 0 = 0 for n ≥ N s s + 1. (83) b_0^0=c_\infty,\qquad b_n^0=v_{n,n_L}\quad\text{for }0\le n\le N_{ss},\qquad b_n^0=0\quad\text{for }n\ge N_{ss}+1. \tag*{(83)} b 0 0 = c ∞ , b n 0 = v n , n L for 0 ≤ n ≤ N ss , b n 0 = 0 for n ≥ N ss + 1. (83)
Moreover, for k ≥ 1 k\ge1 k ≥ 1 , we have
b 0 k = 0 , b n k = 0 if n > ( 2 k p + 1 ) N s s . b_0^k=0,\qquad b_n^k=0\quad\text{if }n>(2kp+1)N_{ss}. b 0 k = 0 , b n k = 0 if n > ( 2 k p + 1 ) N ss .
Estimates for the approximate solution in the remote region
Let N o u t = [ 4 ν d 2 ] N_{\mathrm{out}}=\left[4\nu d^2\right] N out = [ 4 ν d 2 ] 1 and set
ψ o u t a p ( t , x ) = ∑ k = 0 N o u t t k g k ( r ) . \psi_{\mathrm{out}}^{\mathrm{ap}}(t,x)=\sum_{k=0}^{N_{\mathrm{out}}}t^k g_k(r). ψ out ap ( t , x ) = k = 0 ∑ N out t k g k ( r ) .
and
U o u t a p ( t , ρ ) = λ − 1 p ( t ) ψ o u t a p ( t , ρ λ ( t ) ) . (84) U_{\mathrm{out}}^{\mathrm{ap}}(t,\rho)=\lambda^{-\frac{1}{p}}(t)\psi_{\mathrm{out}}^{\mathrm{ap}}\left(t,\frac{\rho}{\lambda(t)}\right). \tag*{(84)} U out ap ( t , ρ ) = λ − p 1 ( t ) ψ out ap ( t , λ ( t ) ρ ) . (84)
Using Lemma 5.3 and recalling that, for any integer k k k , the function g k g_k g k is compactly supported in { 0 ≤ r ≤ 2 δ } \{0\le r\le2\delta\} { 0 ≤ r ≤ 2 δ } , we infer that U o u t a p U_{\mathrm{out}}^{\mathrm{ap}} U out ap defined by (84) satisfies the following L ∞ L^\infty L ∞ estimates in the remote region S o u t S_{\mathrm{out}} S out .
Lemma 5.5. For any integers k , l ≥ 0 k,l\ge0 k , l ≥ 0 , there exists δ 0 ( N o u t ) > 0 \delta_0(N_{\mathrm{out}})>0 δ 0 ( N out ) > 0 such that, for every positive real number δ ≤ δ 0 ( N o u t ) \delta\le\delta_0(N_{\mathrm{out}}) δ ≤ δ 0 ( N out ) , we have
∥ ρ − k ∇ ρ l U o u t a p ∥ L ∞ ( S o u t ) ≤ C k , l t ( ν + ε 2 ) ( 1 p + k + l ) , \left\|\rho^{-k}\nabla_\rho^l U_{\mathrm{out}}^{\mathrm{ap}}\right\|_{L^\infty(S_{\mathrm{out}})}
\le C_{k,l}t^{(\nu+\varepsilon_2)\left(\frac{1}{p}+k+l\right)}, ρ − k ∇ ρ l U out ap L ∞ ( S out ) ≤ C k , l t ( ν + ε 2 ) ( p 1 + k + l ) , ∥ ρ − k ∇ ρ l ( U o u t a p − Q ) ∥ L ∞ ( S o u t ) ≤ C δ t ( 1 2 + ν ) 1 p , \left\|\rho^{-k}\nabla_\rho^l\left(U_{\mathrm{out}}^{\mathrm{ap}}-Q\right)\right\|_{L^\infty(S_{\mathrm{out}})}
\le C_\delta t^{\left(\frac{1}{2}+\nu\right)\frac{1}{p}}, ρ − k ∇ ρ l ( U out ap − Q ) L ∞ ( S out ) ≤ C δ t ( 2 1 + ν ) p 1 , ∥ ρ − k ∇ ρ l ( U o u t a p − t ( 1 / 2 + ν ) / p f 0 ( ρ λ ( t ) ) ) ∥ L ∞ ( S o u t ) ≤ C δ t ( 1 2 + ν ) 1 p + 2 ε 2 . \left\|\rho^{-k}\nabla_\rho^l\left(U_{\mathrm{out}}^{\mathrm{ap}}-t^{(1/2+\nu)/p}f_0\left(\frac{\rho}{\lambda(t)}\right)\right)\right\|_{L^\infty(S_{\mathrm{out}})}
\le C_\delta t^{\left(\frac{1}{2}+\nu\right)\frac{1}{p}+2\varepsilon_2}. ρ − k ∇ ρ l ( U out ap − t ( 1/2 + ν ) / p f 0 ( λ ( t ) ρ ) ) L ∞ ( S out ) ≤ C δ t ( 2 1 + ν ) p 1 + 2 ε 2 . for all 0 < t ≤ T 0<t\le T 0 < t ≤ T , where T = T ( δ , N o u t ) T=T(\delta,N_{\mathrm{out}}) T = T ( δ , N out ) .
We also have the following L 2 L^2 L 2 estimate.
Lemma 5.6. With the above notation, the following estimate holds for any 0 < δ ≤ δ 0 ( N o u t ) 0<\delta\le\delta_0(N_{\mathrm{out}}) 0 < δ ≤ δ 0 ( N out ) and all 0 < t ≤ T = T ( l , δ , N o u t ) 0<t\le T=T(l,\delta,N_{\mathrm{out}}) 0 < t ≤ T = T ( l , δ , N out ) :
∥ ∇ ρ l ( U o u t a p − Q ) ∥ L 2 ( S o u t ) ≤ C l ( t ( ν + ε 2 ) ( ∣ λ + ∣ + l − s c ) + t ( 1 / 2 + ν ) ( l − s c ) δ s c − l ) , \left\|\nabla_\rho^l\left(U_{\mathrm{out}}^{\mathrm{ap}}-Q\right)\right\|_{L^2(S_{\mathrm{out}})}
\le C_l\left(t^{(\nu+\varepsilon_2)(|\lambda_+|+l-s_c)}+t^{(1/2+\nu)(l-s_c)}\delta^{s_c-l}\right), ∇ ρ l ( U out ap − Q ) L 2 ( S out ) ≤ C l ( t ( ν + ε 2 ) ( ∣ λ + ∣ + l − s c ) + t ( 1/2 + ν ) ( l − s c ) δ s c − l ) , s c < l < s c + 2 ν ∣ λ + ∣ s_c<l<s_c+2\nu|\lambda_+| s c < l < s c + 2 ν ∣ λ + ∣ ,
∥ ∇ ρ l ( U o u t a p − Q ) ∥ L 2 ( S o u t ) ≤ C l t ( ν + ε 2 ) ( l − s c ) + ν ∣ λ + ∣ ( 1 − 2 ε 2 ) , \left\|\nabla_\rho^l\left(U_{\mathrm{out}}^{\mathrm{ap}}-Q\right)\right\|_{L^2(S_{\mathrm{out}})}
\le C_lt^{(\nu+\varepsilon_2)(l-s_c)+\nu|\lambda_+|(1-2\varepsilon_2)}, ∇ ρ l ( U out ap − Q ) L 2 ( S out ) ≤ C l t ( ν + ε 2 ) ( l − s c ) + ν ∣ λ + ∣ ( 1 − 2 ε 2 ) , l > s c + 2 ν ∣ λ + ∣ l>s_c+2\nu|\lambda_+| l > s c + 2 ν ∣ λ + ∣ ,
with s c = − 1 p + d 2 s_c=-\frac{1}{p}+\frac{d}{2} s c = − p 1 + 2 d .
We now consider the remainder
R o u t a p ( t , ρ ) = − i t 1 + 2 ν ∂ t U o u t a p + ( 1 2 + ν ) i t 2 ν Λ U o u t a p − Δ U o u t a p − ∣ U o u t a p ∣ 2 p U o u t a p . \mathcal{R}_{\mathrm{out}}^{\mathrm{ap}}(t,\rho)=-it^{1+2\nu}\partial_tU_{\mathrm{out}}^{\mathrm{ap}}+\left(\frac{1}{2}+\nu\right)it^{2\nu}\Lambda U_{\mathrm{out}}^{\mathrm{ap}}-\Delta U_{\mathrm{out}}^{\mathrm{ap}}-\left|U_{\mathrm{out}}^{\mathrm{ap}}\right|^{2p}U_{\mathrm{out}}^{\mathrm{ap}}. R out ap ( t , ρ ) = − i t 1 + 2 ν ∂ t U out ap + ( 2 1 + ν ) i t 2 ν Λ U out ap − Δ U out ap − ∣ U out ap ∣ 2 p U out ap .
We have
R o u t a p ( t , ρ ) = λ − 1 p − 2 ( t ) R ~ o u t a p ( t , ρ λ ( t ) ) , \mathcal{R}_{\mathrm{out}}^{\mathrm{ap}}(t,\rho)=\lambda^{-\frac{1}{p}-2}(t)\widetilde{\mathcal{R}}_{\mathrm{out}}^{\mathrm{ap}}\left(t,\frac{\rho}{\lambda(t)}\right), R out ap ( t , ρ ) = λ − p 1 − 2 ( t ) R out ap ( t , λ ( t ) ρ ) ,
where
R ~ o u t a p ( t , r ) = − i ∂ t ψ o u t a p − Δ ψ o u t a p − ∣ ψ o u t a p ∣ 2 p ψ o u t a p . \widetilde{\mathcal{R}}_{\mathrm{out}}^{\mathrm{ap}}(t,r)=-i\partial_t\psi_{\mathrm{out}}^{\mathrm{ap}}-\Delta\psi_{\mathrm{out}}^{\mathrm{ap}}-\left|\psi_{\mathrm{out}}^{\mathrm{ap}}\right|^{2p}\psi_{\mathrm{out}}^{\mathrm{ap}}. R out ap ( t , r ) = − i ∂ t ψ out ap − Δ ψ out ap − ∣ ψ out ap ∣ 2 p ψ out ap .
It follows directly from the proof of Lemma 5.3 that
∥ ∇ r l R o u t a p ( t , r ) ∥ L 2 ( S o u t ) ≤ C l , N o u t ( t N o u t + t − 1 + 2 ( N o u t + 1 ) ϵ 2 + ( 1 2 − ϵ 2 ) ( s c − l + 2 ν ∣ λ + ∣ ) ) . \left\|\nabla_r^l \mathcal{R}^{ap}_{out}(t,r)\right\|_{L^2(S_{out})} \le C_{l,N_{out}}\left(t^{N_{out}}+t^{-1+2(N_{out}+1)\epsilon_2+\left(\frac{1}{2}-\epsilon_2\right)(s_c-l+2\nu|\lambda_+|)}\right). ∇ r l R o u t a p ( t , r ) L 2 ( S o u t ) ≤ C l , N o u t ( t N o u t + t − 1 + 2 ( N o u t + 1 ) ϵ 2 + ( 2 1 − ϵ 2 ) ( s c − l + 2 ν ∣ λ + ∣ ) ) .
for any l ≥ 0 l\ge0 l ≥ 0 , provided that N o u t ≥ ( s c − 2 ) / 2 N_{out}\ge(s_c-2)/2 N o u t ≥ ( s c − 2 ) /2 . This yields the following lemma.
Lemma 5.8. For any integer l ≥ 0 l\ge0 l ≥ 0 , the following estimate holds:
∥ ∇ ρ l R o u t a p ( t , ρ ) ∥ L 2 ( S o u t ) ≤ C N o u t ( t N o u t − ( 1 2 + ν ) ( s c − 2 − l ) + t 2 N o u t ϵ 2 − ( ν + ϵ 2 ) ( s c − 2 − l ) + ν ∣ λ + ∣ ( 1 − 2 ϵ 2 ) ) , (85) \left\|\nabla_\rho^l\mathcal{R}^{ap}_{out}(t,\rho)\right\|_{L^2(S_{out})}
\le C_{N_{out}}\left(t^{N_{out}-\left(\frac{1}{2}+\nu\right)(s_c-2-l)}
+t^{2N_{out}\epsilon_2-(\nu+\epsilon_2)(s_c-2-l)+\nu|\lambda_+|(1-2\epsilon_2)}\right), \tag*{(85)} ∇ ρ l R o u t a p ( t , ρ ) L 2 ( S o u t ) ≤ C N o u t ( t N o u t − ( 2 1 + ν ) ( s c − 2 − l ) + t 2 N o u t ϵ 2 − ( ν + ϵ 2 ) ( s c − 2 − l ) + ν ∣ λ + ∣ ( 1 − 2 ϵ 2 ) ) , (85)
for all 0 < t ≤ T = T ( l , N o u t ) 0<t\le T=T(l,N_{out}) 0 < t ≤ T = T ( l , N o u t ) .
Finally, assuming N o u t ≥ L N s s N_{out}\ge LN_{ss} N o u t ≥ L N ss , we estimate the difference U o u t a p − U s s a p U^{ap}_{out}-U^{ap}_{ss} U o u t a p − U ss a p in the overlap region S o u t ∩ S s s S_{out}\cap S_{ss} S o u t ∩ S ss . For r < δ r<\delta r < δ , substituting r = t 1 2 y r=t^{\frac{1}{2}}y r = t 2 1 y into (82) and using λ + + 2 L = − 2 ν λ + \lambda_++2L=-2\nu\lambda_+ λ + + 2 L = − 2 ν λ + , we obtain
ϕ o u t a p ( t , r ) = t − 1 2 p y − 1 p ∑ n ≥ 0 t ν ∣ λ + ∣ n ∑ n L − N o u t ≤ k ≤ n L b n , k o u t y λ + n + 2 k , (86) \phi^{ap}_{out}(t,r)=t^{-\frac{1}{2}p}y^{-\frac{1}{p}}\sum_{n\ge0}t^{\nu|\lambda_+|n}\sum_{nL-N_{out}\le k\le nL}b^{out}_{n,k}y^{\lambda_+n+2k}, \tag*{(86)} ϕ o u t a p ( t , r ) = t − 2 1 p y − p 1 n ≥ 0 ∑ t ν ∣ λ + ∣ n n L − N o u t ≤ k ≤ n L ∑ b n , k o u t y λ + n + 2 k , (86)
where b n , k o u t = b n n L − k b^{out}_{n,k}=b^{nL-k}_n b n , k o u t = b n n L − k .
Combining (86) with (83) , we infer that b 0 , 0 o u t = c ∞ b^{out}_{0,0}=c_\infty b 0 , 0 o u t = c ∞ and b n , n L o u t = v n , n L b^{out}_{n,nL}=v_{n,nL} b n , n L o u t = v n , n L for 1 ≤ n ≤ N s s 1\le n\le N_{ss} 1 ≤ n ≤ N ss . Since nonlinear products add nonnegative indices, (59) and (77) –(78) give the same time coefficient equations for n ≤ N s s n\le N_{ss} n ≤ N ss . Induction on the time order therefore identifies b n o u t b^{out}_n b n o u t with the corresponding coefficient in (73) , which vanishes for q > n L q>nL q > n L . Using b n , k o u t = b n n L − k b^{out}_{n,k}=b^{nL-k}_n b n , k o u t = b n n L − k , we obtain
b n , k o u t = v n , k , 1 ≤ n ≤ N s s , 0 ≤ k ≤ n L , b n , k o u t = 0 , 0 ≤ n ≤ N s s , n L − N o u t ≤ k < 0. (87) \begin{aligned}
b^{out}_{n,k}&=v_{n,k},\qquad1\le n\le N_{ss},\quad0\le k\le nL,\\
b^{out}_{n,k}&=0,\qquad0\le n\le N_{ss},\quad nL-N_{out}\le k<0. \tag*{(87)}
\end{aligned} b n , k o u t b n , k o u t = v n , k , 1 ≤ n ≤ N ss , 0 ≤ k ≤ n L , = 0 , 0 ≤ n ≤ N ss , n L − N o u t ≤ k < 0. (87)
where v n , k v_{n,k} v n , k are the coefficients appearing in (63) . As a direct consequence of (87) , we obtain the following lemma.
Lemma 5.9. For any integer l ≥ 0 l\ge0 l ≥ 0 , the following estimate holds on S s s ∩ S o u t S_{ss}\cap S_{out} S ss ∩ S o u t for all 0 < t ≤ T ( l , N s s ) 0<t\le T(l,N_{ss}) 0 < t ≤ T ( l , N ss ) :
∣ ∇ ρ l ( U o u t a p ( t , ρ ) − U s s a p ( t , ρ ) ) ∣ ≤ C ρ − 1 p − l t ν ∣ λ + ∣ ( N s s + 1 ) ( 1 − 2 ϵ 2 ) . \left|\nabla_\rho^l\left(U^{ap}_{out}(t,\rho)-U^{ap}_{ss}(t,\rho)\right)\right|
\le C\rho^{-\frac{1}{p}-l}t^{\nu|\lambda_+|(N_{ss}+1)(1-2\epsilon_2)}. ∇ ρ l ( U o u t a p ( t , ρ ) − U ss a p ( t , ρ ) ) ≤ C ρ − p 1 − l t ν ∣ λ + ∣ ( N ss + 1 ) ( 1 − 2 ϵ 2 ) .
In particular, for N s s = [ λ − λ + ] N_{ss}=\left[\frac{\lambda_-}{\lambda_+}\right] N ss = [ λ + λ − ] , we have
∣ ∇ ρ l ( U o u t a p ( t , ρ ) − U s s a p ( t , ρ ) ) ∣ ≤ C ρ − 1 p − l t ( 1 − 2 ϵ 2 ) ν ∣ λ − ∣ . (88) \left|\nabla_\rho^l\left(U^{ap}_{out}(t,\rho)-U^{ap}_{ss}(t,\rho)\right)\right|
\le C\rho^{-\frac{1}{p}-l}t^{(1-2\epsilon_2)\nu|\lambda_-|}. \tag*{(88)} ∇ ρ l ( U o u t a p ( t , ρ ) − U ss a p ( t , ρ ) ) ≤ C ρ − p 1 − l t ( 1 − 2 ϵ 2 ) ν ∣ λ − ∣ . (88)
Approximate solution in the whole space
Fix
ϵ 1 = ν d , ϵ 2 = 1 8 d . (89) \epsilon_1=\frac{\nu}{d},\qquad\epsilon_2=\frac{1}{8d}. \tag*{(89)} ϵ 1 = d ν , ϵ 2 = 8 d 1 . (89)
These choices satisfy 0 < ϵ 1 < ν 0<\epsilon_1<\nu 0 < ϵ 1 < ν and 0 < ϵ 2 < 1 2 0<\epsilon_2<\frac{1}{2} 0 < ϵ 2 < 2 1 . Let Θ ∈ C c ∞ ( R d ) \Theta\in C^\infty_c(\mathbb{R}^d) Θ ∈ C c ∞ ( R d ) be a radial cutoff function satisfying 0 ≤ Θ ≤ 1 0\le\Theta\le1 0 ≤ Θ ≤ 1 and
Θ ( ξ ) = { 1 , ∣ ξ ∣ ≤ 1 4 , 0 , ∣ ξ ∣ ≥ 1 2 . \Theta(\xi)=
\begin{cases}
1, & |\xi|\le\frac{1}{4},\\
0, & |\xi|\ge\frac{1}{2}.
\end{cases} Θ ( ξ ) = { 1 , 0 , ∣ ξ ∣ ≤ 4 1 , ∣ ξ ∣ ≥ 2 1 .
Based on the constructions in the three regions above, we define the radial profile ψ r a d a p ( t , x ) \psi^{ap}_{rad}(t,x) ψ r a d a p ( t , x ) by
U r a d a p ( t , ρ ) : = Θ ( t ν − ϵ 1 ρ ) U i n a p ( t , ρ ) + ( 1 − Θ ( t ν − ϵ 1 ρ ) ) Θ ( t ν + ϵ 2 ρ ) U s s a p ( t , ρ ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) U o u t a p ( t , ρ ) , U^{ap}_{rad}(t,\rho) := \Theta(t^{\nu-\epsilon_1}\rho)U^{ap}_{in}(t,\rho) + \left(1-\Theta(t^{\nu-\epsilon_1}\rho)\right)\Theta(t^{\nu+\epsilon_2}\rho)U^{ap}_{ss}(t,\rho) + \left(1-\Theta(t^{\nu+\epsilon_2}\rho)\right)U^{ap}_{out}(t,\rho), U r a d a p ( t , ρ ) := Θ ( t ν − ϵ 1 ρ ) U in a p ( t , ρ ) + ( 1 − Θ ( t ν − ϵ 1 ρ ) ) Θ ( t ν + ϵ 2 ρ ) U ss a p ( t , ρ ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) U o u t a p ( t , ρ ) ,
ψ r a d a p ( t , x ) : = λ 1 p ( t ) U r a d a p ( t , λ ( t ) ∣ x ∣ ) . (90) \psi^{ap}_{rad}(t,x) := \lambda^{\frac{1}{p}}(t)U^{ap}_{rad}(t,\lambda(t)|x|). \tag*{(90)} ψ r a d a p ( t , x ) := λ p 1 ( t ) U r a d a p ( t , λ ( t ) ∣ x ∣ ) . (90)
With the choice
N i n = ⌊ λ − λ + ⌋ , N s s = ⌊ λ − λ + ⌋ , and N o u t = ⌊ 4 ν d 2 ⌋ , (91) N_{in}=\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor,\quad N_{ss}=\left\lfloor\frac{\lambda_-}{\lambda_+}\right\rfloor,\quad\text{and}\quad N_{out}=\left\lfloor4\nu d^2\right\rfloor, \tag*{(91)} N in = ⌊ λ + λ − ⌋ , N ss = ⌊ λ + λ − ⌋ , and N o u t = ⌊ 4 ν d 2 ⌋ , (91)
we combine Lemmas 3.2, 4.4 , and 5.5 and apply the Leibniz rule to the cutoff functions in (90) . For sufficiently large d d d , this gives a positive parameter δ 0 ( N o u t ) \delta_0(N_{out}) δ 0 ( N o u t ) such that, for every 0 < δ ≤ δ 0 ( N o u t ) 0<\delta\leq\delta_0(N_{out}) 0 < δ ≤ δ 0 ( N o u t ) , there exists a positive time T = T ( δ ) T=T(\delta) T = T ( δ ) for which the approximate solution U r a d a p U^{ap}_{rad} U r a d a p satisfies the following L ∞ L^\infty L ∞ estimates.
Lemma 6.1. For any k , l ≥ 0 k,l\geq0 k , l ≥ 0 , the following estimate hold for U r a d a p U^{ap}_{rad} U r a d a p and all 0 < t ≤ T 0<t\leq T 0 < t ≤ T :
∥ ⟨ ρ ⟩ − k ∇ ρ l ( U r a d a p ( t , ρ ) − Q ( ρ ) ) ∥ L ∞ ( R d ) ≤ C t 2 ν + C δ t ( 1 2 + ν ) p , k ≤ 1. (92) \left\|\langle\rho\rangle^{-k}\nabla_\rho^l\left(U^{ap}_{rad}(t,\rho)-Q(\rho)\right)\right\|_{L^\infty(\mathbb{R}^d)}\leq Ct^{2\nu}+C\delta t^{(\frac{1}{2}+\nu)p},\quad k\leq1. \tag*{(92)} ⟨ ρ ⟩ − k ∇ ρ l ( U r a d a p ( t , ρ ) − Q ( ρ ) ) L ∞ ( R d ) ≤ C t 2 ν + C δ t ( 2 1 + ν ) p , k ≤ 1. (92)
For u ∈ S ( R d ) u\in\mathcal{S}(\mathbb{R}^d) u ∈ S ( R d ) , we define its Fourier transform u ^ : R d → C \widehat{u}:\mathbb{R}^d\to\mathbb{C} u : R d → C by
u ^ ( ξ ) = ( 2 π ) − d / 2 ∫ R d e − i y ⋅ ξ u ( y ) d y . \widehat{u}(\xi)=(2\pi)^{-d/2}\int_{\mathbb{R}^d}e^{-iy\cdot\xi}u(y)\,\mathrm{d}y. u ( ξ ) = ( 2 π ) − d /2 ∫ R d e − i y ⋅ ξ u ( y ) d y .
For s ≥ 0 s\geq0 s ≥ 0 , we use the homogeneous Sobolev norm
∥ u ∥ H ˙ s 2 = ∫ R d ∣ ξ ∣ 2 s ∣ u ^ ( ξ ) ∣ 2 d ξ . \|u\|_{\dot{H}^s}^2=\int_{\mathbb{R}^d}|\xi|^{2s}|\widehat{u}(\xi)|^2\,\mathrm{d}\xi. ∥ u ∥ H ˙ s 2 = ∫ R d ∣ ξ ∣ 2 s ∣ u ( ξ ) ∣ 2 d ξ .
Since
( − Δ ) s / 2 u ^ ( ξ ) = ∣ ξ ∣ s u ^ ( ξ ) , \widehat{(-\Delta)^{s/2}u}(\xi)=|\xi|^s\widehat{u}(\xi), ( − Δ ) s /2 u ( ξ ) = ∣ ξ ∣ s u ( ξ ) ,
Plancherel’s theorem gives
∥ u ∥ H ˙ s = ∥ ( − Δ ) s / 2 u ∥ L 2 . (93) \|u\|_{\dot{H}^s}=\|(-\Delta)^{s/2}u\|_{L^2}. \tag*{(93)} ∥ u ∥ H ˙ s = ∥ ( − Δ ) s /2 u ∥ L 2 . (93)
To estimate Sobolev norms of noninteger order from the differentiated expansions, let u = u ( ρ ) u=u(\rho) u = u ( ρ ) be a smooth radial function on R d \mathbb{R}^d R d , where ρ = ∣ y ∣ \rho=|y| ρ = ∣ y ∣ . Using the smooth radial cutoff Θ 0 \Theta_0 Θ 0 introduced in § 6 , which equals 1 for ρ ≤ 1 \rho\leq1 ρ ≤ 1 and 0 for ρ ≥ 2 \rho\geq2 ρ ≥ 2 , we write
u ( ρ ) = Θ 0 ( ρ ) u ( ρ ) + ∑ j ≥ 0 ( Θ 0 ( ρ 2 j + 1 ) − Θ 0 ( ρ 2 j ) ) u ( ρ ) . (94) u(\rho)=\Theta_0(\rho)u(\rho)+\sum_{j\geq0}\left(\Theta_0\left(\frac{\rho}{2^{j+1}}\right)-\Theta_0\left(\frac{\rho}{2^j}\right)\right)u(\rho). \tag*{(94)} u ( ρ ) = Θ 0 ( ρ ) u ( ρ ) + j ≥ 0 ∑ ( Θ 0 ( 2 j + 1 ρ ) − Θ 0 ( 2 j ρ ) ) u ( ρ ) . (94)
The j j j th term in the sum is supported in 2 j ≤ ρ ≤ 2 j + 2 2^j\leq\rho\leq2^{j+2} 2 j ≤ ρ ≤ 2 j + 2 . For 0 ≤ s ≤ d 0\leq s\leq d 0 ≤ s ≤ d , scaling, interpolation, and the Leibniz rule give
∥ ( Θ 0 ( ρ 2 j + 1 ) − Θ 0 ( ρ 2 j ) ) u ( ρ ) ∥ H ˙ s ≤ C s 2 j ( d 2 − s ) ∑ l = 0 d 2 j l ∥ ∇ ρ l u ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) . (95) \left\|\left(\Theta_0\left(\frac{\rho}{2^{j+1}}\right)-\Theta_0\left(\frac{\rho}{2^j}\right)\right)u(\rho)\right\|_{\dot{H}^s}\leq C_s2^{j(\frac{d}{2}-s)}\sum_{l=0}^{d}2^{jl}\|\nabla_\rho^l u\|_{L^\infty(2^j\leq\rho\leq2^{j+2})}. \tag*{(95)} ( Θ 0 ( 2 j + 1 ρ ) − Θ 0 ( 2 j ρ ) ) u ( ρ ) H ˙ s ≤ C s 2 j ( 2 d − s ) l = 0 ∑ d 2 j l ∥ ∇ ρ l u ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) . (95)
The differentiated expansions in the three regions yield the following estimates.
Lemma 6.2. For any integers k , l ≥ 0 k,l\geq0 k , l ≥ 0 with k + l > s c k+l>s_c k + l > s c , the following estimate holds for sufficiently small 0 < t ≤ T ( δ ) 0<t\leq T(\delta) 0 < t ≤ T ( δ ) :
∥ ⟨ ρ ⟩ − k ∇ ρ l ( U r a d a p ( t , ρ ) − Q ( ρ ) ) ∥ L 2 ≤ C k , l ( t 2 ν + δ s c − k − l t ( 1 2 + ν ) ( k + l − s c ) ) . (96) \left\|\langle\rho\rangle^{-k}\nabla_\rho^l\left(U^{ap}_{rad}(t,\rho)-Q(\rho)\right)\right\|_{L^2}\leq C_{k,l}\left(t^{2\nu}+\delta^{s_c-k-l}t^{(\frac{1}{2}+\nu)(k+l-s_c)}\right). \tag*{(96)} ⟨ ρ ⟩ − k ∇ ρ l ( U r a d a p ( t , ρ ) − Q ( ρ ) ) L 2 ≤ C k , l ( t 2 ν + δ s c − k − l t ( 2 1 + ν ) ( k + l − s c ) ) . (96) Moreover, for every s c < s ≤ d s_c<s\leq d s c < s ≤ d ,
∥ U r a d a p ( t , ρ ) − Q ( ρ ) ∥ H s ≤ C s ( t 2 ν + δ s c − s t ( 1 2 + ν ) ( s − s c ) ) . (97) \left\|U^{ap}_{rad}(t,\rho)-Q(\rho)\right\|_{H^s}\leq C_s\left(t^{2\nu}+\delta^{s_c-s}t^{(\frac{1}{2}+\nu)(s-s_c)}\right). \tag*{(97)} ∥ U r a d a p ( t , ρ ) − Q ( ρ ) ∥ H s ≤ C s ( t 2 ν + δ s c − s t ( 2 1 + ν ) ( s − s c ) ) . (97)
Applying the Leibniz rule to (90) , we estimate each term using the differentiated expansions in Lemma 3.1, (63) , and (82) . Integrating yields (96) . For (97) , fix s c < s ≤ d s_c<s\le d s c < s ≤ d and apply (95) . The inner and self-similar terms give
∑ j ≥ 0 2 j ≤ C t − ( ν + ϵ 2 ) t 2 ν k 2 j ( 2 k + n λ + + s c − s ) ≤ C s t 2 ν , 1 ≤ n ≤ N s s , 1 ≤ k ≤ n L . (98) \sum_{\substack{j\ge0\\2^{j}\le Ct^{-(\nu+\epsilon_2)}}} t^{2\nu k}2^{j(2k+n\lambda_++s_c-s)}\le C_s t^{2\nu},\qquad1\le n\le N_{ss},\qquad1\le k\le nL. \tag*{(98)} j ≥ 0 2 j ≤ C t − ( ν + ϵ 2 ) ∑ t 2 ν k 2 j ( 2 k + n λ + + s c − s ) ≤ C s t 2 ν , 1 ≤ n ≤ N ss , 1 ≤ k ≤ n L . (98)
Indeed, geometric summation is controlled by the two endpoints, using 2 L = ( 1 + 2 ν ) ∣ λ + ∣ 2L=(1+2\nu)|\lambda_+| 2 L = ( 1 + 2 ν ) ∣ λ + ∣ and
ν ∣ λ + ∣ ( 1 − 2 ϵ 2 ) > 2 ν , ( ν − ϵ 1 ) ∣ λ − ∣ > 2 ν , \nu|\lambda_+|(1-2\epsilon_2)>2\nu,\qquad(\nu-\epsilon_1)|\lambda_-|>2\nu, ν ∣ λ + ∣ ( 1 − 2 ϵ 2 ) > 2 ν , ( ν − ϵ 1 ) ∣ λ − ∣ > 2 ν ,
which follow from the dimension assumption and (89) . The λ − \lambda_- λ − remainders satisfy the same bound. Their terms with k = 0 k=0 k = 0 occur only for ρ ≥ c t − ( ν − ϵ 1 ) \rho\ge ct^{-(\nu-\epsilon_1)} ρ ≥ c t − ( ν − ϵ 1 ) . In the remote region, (82) and (84) similarly give, after decreasing T ( δ ) T(\delta) T ( δ ) ,
∑ j ≥ 0 c t − ( ν + ϵ 2 ) ≤ 2 j ≤ δ λ ( t ) t q λ s c − s ( t ) ( 2 j λ ( t ) ) 2 ν ∣ λ + ∣ n − 2 q + s c − s ≤ C s ( t 2 ν + δ s c − s λ s c − s ( t ) ) , (99) \sum_{\substack{j\ge0\\ct^{-(\nu+\epsilon_2)}\le2^j\le\delta\lambda(t)}} t^q\lambda^{s_c-s}(t)\left(\frac{2^j}{\lambda(t)}\right)^{2\nu|\lambda_+|n-2q+s_c-s}\le C_s\left(t^{2\nu}+\delta^{s_c-s}\lambda^{s_c-s}(t)\right), \tag*{(99)} j ≥ 0 c t − ( ν + ϵ 2 ) ≤ 2 j ≤ δ λ ( t ) ∑ t q λ s c − s ( t ) ( λ ( t ) 2 j ) 2 ν ∣ λ + ∣ n − 2 q + s c − s ≤ C s ( t 2 ν + δ s c − s λ s c − s ( t ) ) , (99)
where 0 ≤ q ≤ N o u t 0\le q\le N_{out} 0 ≤ q ≤ N o u t and 1 ≤ n ≤ ( 2 p q + 1 ) N s s 1\le n\le(2pq+1)N_{ss} 1 ≤ n ≤ ( 2 pq + 1 ) N ss . If the exponent of 2 j 2^j 2 j is zero, summation gives an additional factor 1 + ∣ log t ∣ 1+|\log t| 1 + ∣ log t ∣ . In this case, the time exponent is strictly larger than 2 ν 2\nu 2 ν , so the contribution is still bounded by C s t 2 ν C_s t^{2\nu} C s t 2 ν . The region near the origin contributes C s t 2 ν C_s t^{2\nu} C s t 2 ν . The differentiated decay of Q Q Q and the Leibniz rule control the outer cutoff region and the remaining tail by C s ( δ λ ( t ) ) s c − s C_s(\delta\lambda(t))^{s_c-s} C s ( δ λ ( t ) ) s c − s . Combining (98) and (99) through (94) and using λ ( t ) = t − 1 2 − ν \lambda(t)=t^{-\frac12-\nu} λ ( t ) = t − 2 1 − ν proves (97) .
We next study the critical Sobolev norm. The Gaussian identity
( e − τ ρ 2 ) ∧ ( ξ ) = ( 2 τ ) − d 2 e − ∣ ξ ∣ 2 4 τ , τ > 0 , \left(e^{-\tau\rho^2}\right)^{\wedge}(\xi)=(2\tau)^{-\frac d2}e^{-\frac{|\xi|^2}{4\tau}},\qquad\tau>0, ( e − τ ρ 2 ) ∧ ( ξ ) = ( 2 τ ) − 2 d e − 4 τ ∣ ξ ∣ 2 , τ > 0 ,
together with the Gamma function integral representations
ρ − 1 / p = 1 Γ ( 1 / ( 2 p ) ) ∫ 0 ∞ τ 1 2 p − 1 e − τ ρ 2 d τ , \rho^{-1/p}=\frac{1}{\Gamma(1/(2p))}\int_0^\infty\tau^{\frac{1}{2p}-1}e^{-\tau\rho^2}\,d\tau, ρ − 1/ p = Γ ( 1/ ( 2 p )) 1 ∫ 0 ∞ τ 2 p 1 − 1 e − τ ρ 2 d τ ,
ρ − d 2 = 1 Γ ( d / 4 ) ∫ 0 ∞ τ d 4 − 1 e − τ ∣ y ∣ 2 d τ \rho^{-\frac d2}=\frac{1}{\Gamma(d/4)}\int_0^\infty\tau^{\frac d4-1}e^{-\tau|\mathbf y|^2}\,d\tau ρ − 2 d = Γ ( d /4 ) 1 ∫ 0 ∞ τ 4 d − 1 e − τ ∣ y ∣ 2 d τ
gives, as tempered distributions,
( ρ − 1 p ) ∧ ( ξ ) = 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ∣ ξ ∣ 1 / p − d , \left(\rho^{-\frac1p}\right)^{\wedge}(\xi)=2^{s_c}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}|\xi|^{1/p-d}, ( ρ − p 1 ) ∧ ( ξ ) = 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ∣ ξ ∣ 1/ p − d ,
( ρ − d 2 ) ∧ ( ξ ) = ∣ ξ ∣ − d / 2 . \left(\rho^{-\frac d2}\right)^{\wedge}(\xi)=|\xi|^{-d/2}. ( ρ − 2 d ) ∧ ( ξ ) = ∣ ξ ∣ − d /2 .
Since s c = − 1 p + d 2 s_c=-\frac1p+\frac d2 s c = − p 1 + 2 d , multiplying the first identity by ∣ ξ ∣ s c |\xi|^{s_c} ∣ ξ ∣ s c and using the second one yield
( − Δ ) s c 2 ρ − 1 p = 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ρ − d 2 . (100) (-\Delta)^{\frac{s_c}{2}}\rho^{-\frac1p}=2^{s_c}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}\rho^{-\frac d2}. \tag*{(100)} ( − Δ ) 2 s c ρ − p 1 = 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ρ − 2 d . (100)
Then, we introduce the following lemma.
Lemma 6.3. There exist T > 0 T>0 T > 0 and C δ > 0 C_\delta>0 C δ > 0 such that, for all 0 < t ≤ T 0<t\le T 0 < t ≤ T ,
∥ U r a d a p ( t , ρ ) − Θ δ ( ρ λ ( t ) ) Q ( ρ ) ∥ H ˙ s c ( R d ) ≤ C δ . (101) \left\|U_{rad}^{ap}(t,\rho)-\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)Q(\rho)\right\|_{\dot H^{s_c}(\mathbb{R}^d)}\le C_\delta. \tag*{(101)} U r a d a p ( t , ρ ) − Θ δ ( λ ( t ) ρ ) Q ( ρ ) H ˙ s c ( R d ) ≤ C δ . (101) Moreover, as t → 0 t\to0 t → 0 ,
∥ U r a d a p ( t ) ∥ H ˙ s c ( R d ) = c ∞ 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ∣ S d − 1 ∣ ( 1 2 + ν ) ∣ log t ∣ + O δ ( 1 ) . (102) \left\|U_{rad}^{ap}(t)\right\|_{\dot H^{s_c}(\mathbb{R}^d)}
=c_\infty2^{s_c}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}
\sqrt{|S^{d-1}|\left(\frac12+\nu\right)}\sqrt{|\log t|}+O_\delta(1). \tag*{(102)} ∥ U r a d a p ( t ) ∥ H ˙ s c ( R d ) = c ∞ 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ∣ S d − 1 ∣ ( 2 1 + ν ) ∣ log t ∣ + O δ ( 1 ) . (102)
Proof. Step 1. We first prove (101) . Using the partition (94) and applying (95) with s = s c s=s_c s = s c , we estimate the critical Sobolev norm from the differentiated expansions in Sections 3–§ 6 . For 0 ≤ l ≤ d 0\leq l\leq d 0 ≤ l ≤ d , the derivatives of the cutoff functions satisfy
2 j l ( ∥ ∇ ρ l Θ ( t ν − ϵ 1 ρ ) ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) + ∥ ∇ ρ l Θ ( t ν + ϵ 2 ρ ) ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) + ∥ ∇ ρ l Θ δ ( ρ λ ( t ) ) ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ) ≤ C l . (103) 2^{jl}\left(\left\|\nabla_\rho^l\Theta(t^{\nu-\epsilon_1}\rho)\right\|_{L^\infty(2^j\leq\rho\leq2^{j+2})}+\left\|\nabla_\rho^l\Theta(t^{\nu+\epsilon_2}\rho)\right\|_{L^\infty(2^j\leq\rho\leq2^{j+2})}+\left\|\nabla_\rho^l\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)\right\|_{L^\infty(2^j\leq\rho\leq2^{j+2})}\right)\leq C_l. \tag*{(103)} 2 j l ( ∇ ρ l Θ ( t ν − ϵ 1 ρ ) L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) + ∇ ρ l Θ ( t ν + ϵ 2 ρ ) L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) + ∇ ρ l Θ δ ( λ ( t ) ρ ) L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ) ≤ C l . (103)
Choose T T T sufficiently small so that Θ ( t ν − ϵ 1 ρ ) = Θ δ ( ρ λ ( t ) ) = 1 \Theta(t^{\nu-\epsilon_1}\rho)=\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)=1 Θ ( t ν − ϵ 1 ρ ) = Θ δ ( λ ( t ) ρ ) = 1 for ρ ≤ 2 \rho\leq2 ρ ≤ 2 . Then the definition (90) of U r a d a p U_{\mathrm{rad}}^{ap} U rad a p and Lemma 3.1 yield
∥ ∇ ρ l ( U r a d a p ( t , ρ ) − Θ δ ( ρ λ ( t ) ) Q ( ρ ) ) ∥ L ∞ ( ρ ≤ 2 ) ≤ C t 2 ν , 0 ≤ l ≤ d . (104) \left\|\nabla_\rho^l\left(U_{\mathrm{rad}}^{ap}(t,\rho)-\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)Q(\rho)\right)\right\|_{L^\infty(\rho\leq2)}\leq Ct^{2\nu},\qquad0\leq l\leq d. \tag*{(104)} ∇ ρ l ( U rad a p ( t , ρ ) − Θ δ ( λ ( t ) ρ ) Q ( ρ ) ) L ∞ ( ρ ≤ 2 ) ≤ C t 2 ν , 0 ≤ l ≤ d . (104)
Thus, in the inner and self-similar regions, substituting the expansions in Lemma 3.1 , (63) , and (17) into (90) , and applying the Leibniz rule with (103) , we obtain, for 0 ≤ l ≤ d 0\leq l\leq d 0 ≤ l ≤ d and j ≥ 0 j\geq0 j ≥ 0 with 2 j + 2 ≤ 1 4 t − ( ν + ϵ 2 ) 2^{j+2}\leq\frac{1}{4}t^{-(\nu+\epsilon_2)} 2 j + 2 ≤ 4 1 t − ( ν + ϵ 2 )
∥ ∇ ρ l ( U r a d a p ( t , ρ ) − Θ δ ( ρ λ ( t ) ) Q ( ρ ) ) ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ≤ C 2 − j ( 1 / p + l ) ( ∑ n = 1 N s s ∑ k = 1 n L t 2 ν k 2 j ( 2 k + λ + n ) + ∑ k = 0 N i n t 2 ν k 2 j ( 2 k + λ − ) ) . (105) \left\|\nabla_\rho^l\left(U_{\mathrm{rad}}^{ap}(t,\rho)-\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)Q(\rho)\right)\right\|_{L^\infty(2^j\leq\rho\leq2^{j+2})}
\leq C2^{-j(1/p+l)}\left(\sum_{n=1}^{N_{ss}}\sum_{k=1}^{nL}t^{2\nu k}2^{j(2k+\lambda_+n)}+\sum_{k=0}^{N_{in}}t^{2\nu k}2^{j(2k+\lambda_-)}\right). \tag*{(105)} ∇ ρ l ( U rad a p ( t , ρ ) − Θ δ ( λ ( t ) ρ ) Q ( ρ ) ) L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ≤ C 2 − j ( 1/ p + l ) ( n = 1 ∑ N ss k = 1 ∑ n L t 2 ν k 2 j ( 2 k + λ + n ) + k = 0 ∑ N in t 2 ν k 2 j ( 2 k + λ − ) ) . (105)
where the terms with k = 0 k=0 k = 0 and 1 ≤ n ≤ N s s 1\leq n\leq N_{ss} 1 ≤ n ≤ N ss cancel by matching of the coefficients in (57) . For the first sum in (105) , D i s c r ∉ Q \sqrt{\mathit{Discr}}\notin\mathbb{Q} Discr ∈ / Q implies 2 k + λ + n ≠ 0 2k+\lambda_+n\neq0 2 k + λ + n = 0 and
ν ∣ λ + ∣ n − ϵ 2 ( 2 k − ∣ λ + ∣ n ) ≥ ν ∣ λ + ∣ n ( 1 − 2 ϵ 2 ) > 0 (106) \nu|\lambda_+|n-\epsilon_2(2k-|\lambda_+|n)\geq\nu|\lambda_+|n(1-2\epsilon_2)>0 \tag*{(106)} ν ∣ λ + ∣ n − ϵ 2 ( 2 k − ∣ λ + ∣ n ) ≥ ν ∣ λ + ∣ n ( 1 − 2 ϵ 2 ) > 0 (106)
by k ≤ n L k\leq nL k ≤ n L and 2 L = ( 1 + 2 ν ) ∣ λ + ∣ 2L=(1+2\nu)|\lambda_+| 2 L = ( 1 + 2 ν ) ∣ λ + ∣ . For the second sum, the same argument applies since 2 k + λ − ≠ 0 2k+\lambda_-\neq0 2 k + λ − = 0 and, for k ≤ N i n = L N s s k\leq N_{in}=LN_{ss} k ≤ N in = L N ss ,
ν ∣ λ − ∣ − ϵ 2 ( 2 k − ∣ λ − ∣ ) ≥ ν ∣ λ − ∣ ( 1 − 2 ϵ 2 ) > 0. (107) \nu|\lambda_-|-\epsilon_2(2k-|\lambda_-|)\geq\nu|\lambda_-|(1-2\epsilon_2)>0. \tag*{(107)} ν ∣ λ − ∣ − ϵ 2 ( 2 k − ∣ λ − ∣ ) ≥ ν ∣ λ − ∣ ( 1 − 2 ϵ 2 ) > 0. (107)
Applying (95) to (105) and summing over j ≥ 0 j\geq0 j ≥ 0 with 2 j + 2 ≤ 1 4 t − ( ν + ϵ 2 ) 2^{j+2}\leq\frac{1}{4}t^{-(\nu+\epsilon_2)} 2 j + 2 ≤ 4 1 t − ( ν + ϵ 2 ) , we obtain a uniform bound by (106) and (107) . Combining this bound with (104) , we obtain
∥ Θ 0 ( ρ ) ( U r a d a p ( t , ρ ) − Θ δ ( ρ λ ( t ) ) Q ( ρ ) ) ∥ H ˙ s c + ∑ j ≥ 0 2 j + 2 ≤ 1 4 t − ( ν + ϵ 2 ) ∥ ( Θ 0 ( ρ 2 j + 1 ) − Θ 0 ( ρ 2 j ) ) ( U r a d a p ( t , ρ ) − Θ δ ( ρ λ ( t ) ) Q ( ρ ) ) ∥ H ˙ s c ≤ C . (108) \left\|\Theta_0(\rho)\left(U_{\mathrm{rad}}^{ap}(t,\rho)-\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)Q(\rho)\right)\right\|_{\dot H^{s_c}}
+\sum_{\substack{j\geq0\\2^{j+2}\leq\frac14t^{-(\nu+\epsilon_2)}}}
\left\|\left(\Theta_0\left(\frac{\rho}{2^{j+1}}\right)-\Theta_0\left(\frac{\rho}{2^j}\right)\right)\left(U_{\mathrm{rad}}^{ap}(t,\rho)-\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)Q(\rho)\right)\right\|_{\dot H^{s_c}}\leq C. \tag*{(108)} Θ 0 ( ρ ) ( U rad a p ( t , ρ ) − Θ δ ( λ ( t ) ρ ) Q ( ρ ) ) H ˙ s c + j ≥ 0 2 j + 2 ≤ 4 1 t − ( ν + ϵ 2 ) ∑ ( Θ 0 ( 2 j + 1 ρ ) − Θ 0 ( 2 j ρ ) ) ( U rad a p ( t , ρ ) − Θ δ ( λ ( t ) ρ ) Q ( ρ ) ) H ˙ s c ≤ C . (108)
In the outer region with ρ < δ λ ( t ) \rho<\delta\lambda(t) ρ < δ λ ( t ) , (74) , (82) , and (84) , together with (17) , give
∥ ∇ ρ l ( U o u t a p ( t , ρ ) − Q ( ρ ) ) ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ≤ C δ 2 − j ( 1 / p + l ) ( 2 j λ + + ∑ q = 0 N o u t ∑ n = 1 ( 2 p q + 1 ) N s s t q ( 2 j λ ( t ) ) 2 ν ∣ λ + ∣ n − 2 q ) , (109) \left\|\nabla_\rho^l\left(U_{\mathrm{out}}^{ap}(t,\rho)-Q(\rho)\right)\right\|_{L^\infty(2^j\leq\rho\leq2^{j+2})}
\leq C\delta2^{-j(1/p+l)}\left(2^{j\lambda_+}+\sum_{q=0}^{N_{\mathrm{out}}}\sum_{n=1}^{(2pq+1)N_{ss}}t^q\left(\frac{2^j}{\lambda(t)}\right)^{2\nu|\lambda_+|n-2q}\right), \tag*{(109)} ∇ ρ l ( U out a p ( t , ρ ) − Q ( ρ ) ) L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ≤ C δ 2 − j ( 1/ p + l ) 2 j λ + + q = 0 ∑ N out n = 1 ∑ ( 2 pq + 1 ) N ss t q ( λ ( t ) 2 j ) 2 ν ∣ λ + ∣ n − 2 q , (109)
for 0 ≤ l ≤ d 0\leq l\leq d 0 ≤ l ≤ d and j ≥ 0 j\geq0 j ≥ 0 with 1 10 t − ( ν + ϵ 2 ) ≤ 2 j \frac{1}{10}t^{-(\nu+\epsilon_2)}\leq2^j 10 1 t − ( ν + ϵ 2 ) ≤ 2 j and 2 j + 2 ≤ δ λ ( t ) 2^{j+2}\leq\delta\lambda(t) 2 j + 2 ≤ δ λ ( t ) . Here b 0 0 = c ∞ b_0^0=c_\infty b 0 0 = c ∞ and b 0 q = 0 b_0^q=0 b 0 q = 0 for q ≥ 1 q\geq1 q ≥ 1 . Similarly, the assumption D i s c r ∉ Q \sqrt{\mathit{Discr}}\notin\mathbb{Q} Discr ∈ / Q gives 2 ν ∣ λ + ∣ n − 2 q ≠ 0 2\nu|\lambda_+|n-2q\neq0 2 ν ∣ λ + ∣ n − 2 q = 0 . Moreover,
2 q ϵ 2 + ν ∣ λ + ∣ n ( 1 − 2 ϵ 2 ) > 0 , λ + < 0. (110) 2q\epsilon_2+\nu|\lambda_+|n(1-2\epsilon_2)>0,\qquad\lambda_+<0. \tag*{(110)} 2 q ϵ 2 + ν ∣ λ + ∣ n ( 1 − 2 ϵ 2 ) > 0 , λ + < 0. (110)
Thus, applying (95) to (109) and summing over j j j , we obtain, by (110) ,
∑ j ≥ 0 1 10 t − ( ν + ϵ 2 ) ≤ 2 j , 2 j + 2 ≤ δ λ ( t ) ∥ ( Θ 0 ( ρ 2 j + 1 ) − Θ 0 ( ρ 2 j ) ) ( U o u t a p ( t , ρ ) − Q ( ρ ) ) ∥ H ˙ s c ≤ C δ . (111) \sum_{\substack{j\ge0\\ \frac{1}{10}t^{-(\nu+\epsilon_{2})}\le2^{j},\ 2^{j+2}\le\delta\lambda(t)}}\left\|\left(\Theta_{0}\left(\frac{\rho}{2^{j+1}}\right)-\Theta_{0}\left(\frac{\rho}{2^{j}}\right)\right)\left(U_{\mathrm{out}}^{\mathrm{ap}}(t,\rho)-Q(\rho)\right)\right\|_{\dot{H}^{s_{c}}}\le C_{\delta}. \tag*{(111)} j ≥ 0 10 1 t − ( ν + ϵ 2 ) ≤ 2 j , 2 j + 2 ≤ δ λ ( t ) ∑ ( Θ 0 ( 2 j + 1 ρ ) − Θ 0 ( 2 j ρ ) ) ( U out ap ( t , ρ ) − Q ( ρ ) ) H ˙ s c ≤ C δ . (111)
On the overlap between the self-similar and remote regions, (90) gives
U r a d a p − Q = Θ ( t ν + ϵ 2 ρ ) ( U s s a p − Q ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) ( U o u t a p − Q ) . U_{\mathrm{rad}}^{\mathrm{ap}}-Q=\Theta\left(t^{\nu+\epsilon_{2}}\rho\right)\left(U_{\mathrm{ss}}^{\mathrm{ap}}-Q\right)+\left(1-\Theta\left(t^{\nu+\epsilon_{2}}\rho\right)\right)\left(U_{\mathrm{out}}^{\mathrm{ap}}-Q\right). U rad ap − Q = Θ ( t ν + ϵ 2 ρ ) ( U ss ap − Q ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) ( U out ap − Q ) .
The preceding derivative estimates and (103) , together with the Leibniz rule and (95) , bound each localized term by C δ C_{\delta} C δ . Since the number of indices satisfying 1 10 t − ( ν + ϵ 2 ) ≤ 2 j ≤ 10 t − ( ν + ϵ 2 ) \frac{1}{10}t^{-(\nu+\epsilon_{2})}\le2^{j}\le10t^{-(\nu+\epsilon_{2})} 10 1 t − ( ν + ϵ 2 ) ≤ 2 j ≤ 10 t − ( ν + ϵ 2 ) is uniformly bounded, the total contribution is bounded by C δ C_{\delta} C δ .
It remains to estimate the contribution of the outer cutoff region δ λ ( t ) ≤ ρ ≤ 2 δ λ ( t ) \delta\lambda(t)\le\rho\le2\delta\lambda(t) δ λ ( t ) ≤ ρ ≤ 2 δ λ ( t ) . The corresponding localized terms have indices j ≥ 0 j\ge0 j ≥ 0 satisfying δ λ ( t ) / 4 ≤ 2 j ≤ 2 δ λ ( t ) \delta\lambda(t)/4\le2^{j}\le2\delta\lambda(t) δ λ ( t ) /4 ≤ 2 j ≤ 2 δ λ ( t ) . By (84) and the differentiated decay of Q Q Q , we obtain
∥ ∇ ρ l ( U r a d a p ( t , ρ ) − Θ δ ( ρ λ ( t ) ) Q ( ρ ) ) ∥ L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ≤ C δ λ − 1 / p − l ( t ) , 0 ≤ l ≤ d . (112) \left\|\nabla_{\rho}^{l}\left(U_{\mathrm{rad}}^{\mathrm{ap}}(t,\rho)-\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)Q(\rho)\right)\right\|_{L^{\infty}(2^{j}\le\rho\le2^{j+2})}\le C_{\delta}\lambda^{-1/p-l}(t),\qquad0\le l\le d. \tag*{(112)} ∇ ρ l ( U rad ap ( t , ρ ) − Θ δ ( λ ( t ) ρ ) Q ( ρ ) ) L ∞ ( 2 j ≤ ρ ≤ 2 j + 2 ) ≤ C δ λ − 1/ p − l ( t ) , 0 ≤ l ≤ d . (112)
Applying (95) with (112) , and summing over the finitely many indices satisfying δ λ ( t ) / 4 ≤ 2 j ≤ 2 δ λ ( t ) \delta\lambda(t)/4\le2^{j}\le2\delta\lambda(t) δ λ ( t ) /4 ≤ 2 j ≤ 2 δ λ ( t ) , we obtain a bound C δ C_{\delta} C δ for these localized terms. Since the difference vanishes for ρ ≥ 2 δ λ ( t ) \rho\ge2\delta\lambda(t) ρ ≥ 2 δ λ ( t ) , combining this bound with (108) and (111) , and applying the triangle inequality to (94) , gives (101) .
Step 2. It remains to prove (102) . According to (101) , it is sufficient to prove
∥ Θ δ ( ρ λ ( t ) ) Q ( ρ ) ∥ H ˙ s c = c ∞ 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ∣ S d − 1 ∣ ln ( δ λ ( t ) ) + O δ ( 1 ) . \left\|\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)Q(\rho)\right\|_{\dot{H}^{s_{c}}}=c_{\infty}2^{s_{c}}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}\sqrt{\lvert S^{d-1}\rvert\ln(\delta\lambda(t))}+O_{\delta}(1). Θ δ ( λ ( t ) ρ ) Q ( ρ ) H ˙ s c = c ∞ 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ∣ S d − 1 ∣ ln ( δ λ ( t )) + O δ ( 1 ) .
Since the leading term of Q Q Q is c ∞ ρ − 1 / p c_{\infty}\rho^{-1/p} c ∞ ρ − 1/ p , which is singular at the origin, we multiply it by 1 − Θ 0 ( ρ ) 1-\Theta_{0}(\rho) 1 − Θ 0 ( ρ ) . Then, the resulting remainder Q ( ρ ) − ( 1 − Θ 0 ( ρ ) ) c ∞ ρ − 1 / p Q(\rho)-(1-\Theta_{0}(\rho))c_{\infty}\rho^{-1/p} Q ( ρ ) − ( 1 − Θ 0 ( ρ )) c ∞ ρ − 1/ p is smooth near the origin. By the expansion (17) and the estimate (95) , summing over j ≥ 0 j\ge0 j ≥ 0 , we obtain
∥ Θ δ ( ρ λ ( t ) ) ( Q ( ρ ) − c ∞ ( 1 − Θ 0 ( ρ ) ) ρ − 1 / p ) ∥ H ˙ s c ≤ C δ . (113) \left\|\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)\left(Q(\rho)-c_{\infty}(1-\Theta_{0}(\rho))\rho^{-1/p}\right)\right\|_{\dot{H}^{s_{c}}}\le C_{\delta}. \tag*{(113)} Θ δ ( λ ( t ) ρ ) ( Q ( ρ ) − c ∞ ( 1 − Θ 0 ( ρ )) ρ − 1/ p ) H ˙ s c ≤ C δ . (113)
Choose T T T sufficiently small so that δ λ ( t ) ≥ 2 \delta\lambda(t)\ge2 δ λ ( t ) ≥ 2 for all 0 < t ≤ T 0<t\le T 0 < t ≤ T . Since Θ 0 \Theta_{0} Θ 0 is supported in ρ ≤ 2 \rho\le2 ρ ≤ 2 , we have
Θ δ ( ρ λ ( t ) ) Θ 0 ( ρ ) = Θ 0 ( ρ ) . \Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)\Theta_{0}(\rho)=\Theta_{0}(\rho). Θ δ ( λ ( t ) ρ ) Θ 0 ( ρ ) = Θ 0 ( ρ ) .
Thus (113) and the reverse triangle inequality imply
∣ ∥ Θ δ ( ρ λ ( t ) ) Q ∥ H ˙ s c − c ∞ ∥ ( Θ δ ( ρ λ ( t ) ) − Θ 0 ( ρ ) ) ρ − 1 / p ∥ H ˙ s c ∣ ≤ C δ . \left|\left\|\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)Q\right\|_{\dot{H}^{s_{c}}}-c_{\infty}\left\|\left(\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)-\Theta_{0}(\rho)\right)\rho^{-1/p}\right\|_{\dot{H}^{s_{c}}}\right|\le C_{\delta}. Θ δ ( λ ( t ) ρ ) Q H ˙ s c − c ∞ ( Θ δ ( λ ( t ) ρ ) − Θ 0 ( ρ ) ) ρ − 1/ p H ˙ s c ≤ C δ .
We next compute the second norm. According to (93) , (100) and
Θ δ ( ρ λ ( t ) ) − Θ 0 ( ρ ) = ( 1 − Θ 0 ( ρ ) ) − ( 1 − Θ δ ( ρ λ ( t ) ) ) , \Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)-\Theta_{0}(\rho)=(1-\Theta_{0}(\rho))-\left(1-\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)\right), Θ δ ( λ ( t ) ρ ) − Θ 0 ( ρ ) = ( 1 − Θ 0 ( ρ )) − ( 1 − Θ δ ( λ ( t ) ρ ) ) ,
it suffices to estimate
∥ ( − Δ ) s c / 2 ( ( 1 − Θ δ ( ρ λ ( t ) ) ) ρ − 1 / p ) − 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ( 1 − Θ δ ( ρ λ ( t ) ) ) ρ − d / 2 ∥ L 2 . (114) \left\|(-\Delta)^{s_{c}/2}\left((1-\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right))\rho^{-1/p}\right)-2^{s_{c}}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}\left(1-\Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)\right)\rho^{-d/2}\right\|_{L^{2}}. \tag*{(114)} ( − Δ ) s c /2 ( ( 1 − Θ δ ( λ ( t ) ρ ) ) ρ − 1/ p ) − 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ( 1 − Θ δ ( λ ( t ) ρ ) ) ρ − d /2 L 2 . (114)
and
∥ ( − Δ ) s c / 2 ( ( 1 − Θ 0 ( ρ ) ) ρ − 1 / p ) − 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ( 1 − Θ 0 ( ρ ) ) ρ − d / 2 ∥ L 2 . (115) \left\|(-\Delta)^{s_{c}/2}\left((1-\Theta_{0}(\rho))\rho^{-1/p}\right)-2^{s_{c}}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}(1-\Theta_{0}(\rho))\rho^{-d/2}\right\|_{L^{2}}. \tag*{(115)} ( − Δ ) s c /2 ( ( 1 − Θ 0 ( ρ )) ρ − 1/ p ) − 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ( 1 − Θ 0 ( ρ )) ρ − d /2 L 2 . (115)
Since Θ δ ( ρ λ ( t ) ) = Θ 0 ( ρ λ ( t ) δ ) \Theta_{\delta}\left(\frac{\rho}{\lambda(t)}\right)=\Theta_{0}\left(\frac{\rho}{\lambda(t)\delta}\right) Θ δ ( λ ( t ) ρ ) = Θ 0 ( λ ( t ) δ ρ ) and s c = − 1 p + d 2 s_{c}=-\frac{1}{p}+\frac{d}{2} s c = − p 1 + 2 d , spatial scaling shows that (114) and (115) are equal. Therefore, it suffices to estimate (115) .
We first derive an L ∞ L^\infty L ∞ bound for the fractional Laplacian. Since d > s c 2 d > \frac{s_c}{2} d > 2 s c , we write
( − Δ ) s c 2 u = 1 Γ ( d − s c / 2 ) ∫ 0 ∞ τ d − s c 2 − 1 ( − Δ ) d e τ Δ u d τ . (116) (-\Delta)^{\frac{s_c}{2}}u = \frac{1}{\Gamma(d-s_c/2)}\int_0^\infty\tau^{d-\frac{s_c}{2}-1}(-\Delta)^d e^{\tau\Delta}u\,\mathrm{d}\tau. \tag*{(116)} ( − Δ ) 2 s c u = Γ ( d − s c /2 ) 1 ∫ 0 ∞ τ d − 2 s c − 1 ( − Δ ) d e τ Δ u d τ . (116)
The heat semigroup has the Gaussian kernel representation
( e τ Δ u ) ( y ) = ∫ R d ( 4 π τ ) − d 2 e − ∣ y − z ∣ 2 4 τ u ( z ) d z . (e^{\tau\Delta}u)(y)=\int_{\mathbb{R}^d}(4\pi\tau)^{-\frac{d}{2}}e^{-\frac{|y-z|^2}{4\tau}}u(z)\,\mathrm{d}z. ( e τ Δ u ) ( y ) = ∫ R d ( 4 π τ ) − 2 d e − 4 τ ∣ y − z ∣ 2 u ( z ) d z .
A direct computation gives
− Δ y e − ∣ y − z ∣ 2 4 τ = 1 2 τ ( d − ∣ y − z ∣ 2 2 τ ) e − ∣ y − z ∣ 2 4 τ . -\Delta_y e^{-\frac{|y-z|^2}{4\tau}}=\frac{1}{2\tau}\left(d-\frac{|y-z|^2}{2\tau}\right)e^{-\frac{|y-z|^2}{4\tau}}. − Δ y e − 4 τ ∣ y − z ∣ 2 = 2 τ 1 ( d − 2 τ ∣ y − z ∣ 2 ) e − 4 τ ∣ y − z ∣ 2 .
Repeated differentiation then yields
∣ ( − Δ y ) d ( ( 4 π τ ) − d 2 e − ∣ y − z ∣ 2 4 τ ) ∣ ≤ C τ − 3 d 2 e − ∣ y − z ∣ 2 8 τ . (117) \left|(-\Delta_y)^d\left((4\pi\tau)^{-\frac{d}{2}}e^{-\frac{|y-z|^2}{4\tau}}\right)\right|\le C\tau^{-\frac{3d}{2}}e^{-\frac{|y-z|^2}{8\tau}}. \tag*{(117)} ( − Δ y ) d ( ( 4 π τ ) − 2 d e − 4 τ ∣ y − z ∣ 2 ) ≤ C τ − 2 3 d e − 8 τ ∣ y − z ∣ 2 . (117)
Integrating (117) in z z z yields
∥ ( − Δ ) d e τ Δ u ∥ L ∞ ≤ C τ − 3 d 2 ∥ u ∥ L ∞ ∫ R d e − ∣ z ∣ 2 8 τ d z ≤ C τ − d ∥ u ∥ L ∞ . (118) \|(-\Delta)^d e^{\tau\Delta}u\|_{L^\infty}\le C\tau^{-\frac{3d}{2}}\|u\|_{L^\infty}\int_{\mathbb{R}^d}e^{-\frac{|z|^2}{8\tau}}\,\mathrm{d}z\le C\tau^{-d}\|u\|_{L^\infty}. \tag*{(118)} ∥ ( − Δ ) d e τ Δ u ∥ L ∞ ≤ C τ − 2 3 d ∥ u ∥ L ∞ ∫ R d e − 8 τ ∣ z ∣ 2 d z ≤ C τ − d ∥ u ∥ L ∞ . (118)
Alternatively, since the Gaussian kernel is nonnegative and has integral one,
∥ e τ Δ u ∥ L ∞ ≤ ∥ u ∥ L ∞ . \|e^{\tau\Delta}u\|_{L^\infty}\le\|u\|_{L^\infty}. ∥ e τ Δ u ∥ L ∞ ≤ ∥ u ∥ L ∞ .
Commuting ( − Δ ) d (-\Delta)^d ( − Δ ) d with e τ Δ e^{\tau\Delta} e τ Δ then gives
∥ ( − Δ ) d e τ Δ u ∥ L ∞ = ∥ e τ Δ ( − Δ ) d u ∥ L ∞ ≤ ∥ ( − Δ ) d u ∥ L ∞ . (119) \|(-\Delta)^d e^{\tau\Delta}u\|_{L^\infty}=\|e^{\tau\Delta}(-\Delta)^d u\|_{L^\infty}\le\|(-\Delta)^d u\|_{L^\infty}. \tag*{(119)} ∥ ( − Δ ) d e τ Δ u ∥ L ∞ = ∥ e τ Δ ( − Δ ) d u ∥ L ∞ ≤ ∥ ( − Δ ) d u ∥ L ∞ . (119)
Substituting (119) for 0 < τ ≤ 1 0<\tau\le1 0 < τ ≤ 1 and (118) for τ ≥ 1 \tau\ge1 τ ≥ 1 into (116) , we deduce
∥ ( − Δ ) s c 2 u ∥ L ∞ ≤ 1 Γ ( d − s c / 2 ) ( ∥ ( − Δ ) d u ∥ L ∞ ∫ 0 1 τ d − s c / 2 − 1 d τ + C ∥ u ∥ L ∞ ∫ 1 ∞ τ − s c / 2 − 1 d τ ) ≤ C ( ∥ ( − Δ ) d u ∥ L ∞ + ∥ u ∥ L ∞ ) . (120) \begin{aligned}
\|(-\Delta)^{\frac{s_c}{2}}u\|_{L^\infty}
&\le\frac{1}{\Gamma(d-s_c/2)}\left(\|(-\Delta)^d u\|_{L^\infty}\int_0^1\tau^{d-s_c/2-1}\,\mathrm{d}\tau+C\|u\|_{L^\infty}\int_1^\infty\tau^{-s_c/2-1}\,\mathrm{d}\tau\right)\\
&\le C\left(\|(-\Delta)^d u\|_{L^\infty}+\|u\|_{L^\infty}\right). \tag*{(120)}
\end{aligned} ∥ ( − Δ ) 2 s c u ∥ L ∞ ≤ Γ ( d − s c /2 ) 1 ( ∥ ( − Δ ) d u ∥ L ∞ ∫ 0 1 τ d − s c /2 − 1 d τ + C ∥ u ∥ L ∞ ∫ 1 ∞ τ − s c /2 − 1 d τ ) ≤ C ( ∥ ( − Δ ) d u ∥ L ∞ + ∥ u ∥ L ∞ ) . (120)
Next, we estimate (115) by decomposing R d = I ∪ I I \mathbb{R}^d=I\cup II R d = I ∪ I I , where
I = { y ∈ R d : ∣ y ∣ ≤ 4 } , I I = { y ∈ R d : ∣ y ∣ > 4 } , I=\{y\in\mathbb{R}^d:|y|\le4\},\qquad II=\{y\in\mathbb{R}^d:|y|>4\}, I = { y ∈ R d : ∣ y ∣ ≤ 4 } , I I = { y ∈ R d : ∣ y ∣ > 4 } ,
and ∣ y ∣ = ρ |y|=\rho ∣ y ∣ = ρ . Since Θ 0 \Theta_0 Θ 0 is supported in ρ ≤ 2 \rho\le2 ρ ≤ 2 , the radius 4 ensures that, for y ∈ I I y\in II y ∈ I I and ∣ z ∣ ≤ 2 |z|\le2 ∣ z ∣ ≤ 2 ,
∣ y − z ∣ ≥ ρ − 2 ≥ ρ 2 . |y-z|\ge\rho-2\ge\frac{\rho}{2}. ∣ y − z ∣ ≥ ρ − 2 ≥ 2 ρ .
On the region I, the function ( 1 − Θ 0 ( ρ ) ) ρ − 1 / p (1-\Theta_0(\rho))\rho^{-1/p} ( 1 − Θ 0 ( ρ )) ρ − 1/ p is smooth with bounded derivatives. Applying (120) , we obtain
∥ ( − Δ ) s c 2 ( ( 1 − Θ 0 ( ρ ) ) ρ − 1 / p ) ∥ L ∞ ( I ) ≤ C . (121) \left\|(-\Delta)^{\frac{s_c}{2}}\left((1-\Theta_0(\rho))\rho^{-1/p}\right)\right\|_{L^\infty(I)}\le C. \tag*{(121)} ( − Δ ) 2 s c ( ( 1 − Θ 0 ( ρ )) ρ − 1/ p ) L ∞ ( I ) ≤ C . (121)
Moreover, ( 1 − Θ 0 ( ρ ) ) ρ − d 2 (1-\Theta_0(\rho))\rho^{-\frac d2} ( 1 − Θ 0 ( ρ )) ρ − 2 d in (115) is bounded on I.
On the region II, we have Θ 0 ( ρ ) = 0 \Theta_0(\rho)=0 Θ 0 ( ρ ) = 0 . Thus (100) gives
( − Δ ) s c 2 ( ( 1 − Θ 0 ( ρ ) ) ρ − 1 / p − 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ρ − d 2 ) = − ( − Δ ) s c 2 ( Θ 0 ( ρ ) ρ − 1 / p ) . (-\Delta)^{\frac{s_c}{2}}\left((1-\Theta_0(\rho))\rho^{-1/p}-2^{s_c}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}\rho^{-\frac d2}\right)=-(-\Delta)^{\frac{s_c}{2}}\left(\Theta_0(\rho)\rho^{-1/p}\right). ( − Δ ) 2 s c ( ( 1 − Θ 0 ( ρ )) ρ − 1/ p − 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ρ − 2 d ) = − ( − Δ ) 2 s c ( Θ 0 ( ρ ) ρ − 1/ p ) .
Using ∣ y − z ∣ ≥ ρ 2 |y-z|\geq\frac{\rho}{2} ∣ y − z ∣ ≥ 2 ρ in (117) and applying (116) , we obtain
∣ ( − Δ ) s c 2 ( Θ 0 ( ρ ) ρ − 1 p ) ∣ ≲ C ∫ 0 ∞ τ − d + s c 2 − 1 ∫ ∣ z ∣ ≤ 2 e − ∣ y − z ∣ 2 8 τ ∣ Θ 0 ( ∣ z ∣ ) ∣ ∣ z ∣ − 1 p d z d τ ≤ C ∫ 0 ∞ τ − d + s c 2 − 1 e − ρ 2 32 τ d τ ≤ C ρ − d − s c , (122) \begin{aligned}
\left|(-\Delta)^{\frac{s_c}{2}}\left(\Theta_0(\rho)\rho^{-\frac{1}{p}}\right)\right|
\lesssim C\int_0^\infty\tau^{-\frac{d+s_c}{2}-1}\int_{|z|\leq2}e^{-\frac{|y-z|^2}{8\tau}}\left|\Theta_0(|z|)\right||z|^{-\frac{1}{p}}\,dz\,d\tau\\
&\leq C\int_0^\infty\tau^{-\frac{d+s_c}{2}-1}e^{-\frac{\rho^2}{32\tau}}\,d\tau\leq C\rho^{-d-s_c}, \tag*{(122)}
\end{aligned} ( − Δ ) 2 s c ( Θ 0 ( ρ ) ρ − p 1 ) ≲ C ∫ 0 ∞ τ − 2 d + s c − 1 ∫ ∣ z ∣ ≤ 2 e − 8 τ ∣ y − z ∣ 2 ∣ Θ 0 ( ∣ z ∣ ) ∣ ∣ z ∣ − p 1 d z d τ ≤ C ∫ 0 ∞ τ − 2 d + s c − 1 e − 32 τ ρ 2 d τ ≤ C ρ − d − s c , (122)
where we used Θ 0 ( ρ ) ρ − 1 / p ∈ L 1 ( R d ) \Theta_0(\rho)\rho^{-1/p}\in L^1(\mathbb{R}^d) Θ 0 ( ρ ) ρ − 1/ p ∈ L 1 ( R d ) .
Combining (121) on I and (122) on II, we conclude that
∥ ( − Δ ) s c 2 ( ( 1 − Θ 0 ( ρ ) ) ρ − 1 p ) − 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ( 1 − Θ 0 ( ρ ) ) ρ − d 2 ∥ L 2 ( R d ) 2 ≤ C + C ∫ 4 ∞ ρ − d − 2 s c − 1 d ρ ≤ C . (123) \begin{aligned}
\left\|(-\Delta)^{\frac{s_c}{2}}\left((1-\Theta_0(\rho))\rho^{-\frac{1}{p}}\right)-2^{s_c}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}(1-\Theta_0(\rho))\rho^{-\frac{d}{2}}\right\|_{L^2(\mathbb{R}^d)}^2 \\
&\leq C+C\int_4^\infty\rho^{-d-2s_c-1}\,d\rho\leq C. \tag*{(123)}
\end{aligned} ( − Δ ) 2 s c ( ( 1 − Θ 0 ( ρ )) ρ − p 1 ) − 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ( 1 − Θ 0 ( ρ )) ρ − 2 d L 2 ( R d ) 2 ≤ C + C ∫ 4 ∞ ρ − d − 2 s c − 1 d ρ ≤ C . (123)
Taking (123) and spatial scaling into account, we conclude that both errors in (114) and (115) are uniformly bounded in L 2 L^2 L 2 . Using (93) and the reverse triangle inequality, we deduce
∥ ( Θ δ ( ρ λ ( t ) ) − Θ 0 ( ρ ) ) ρ − 1 p ∥ H ˙ s c = 2 s c Γ ( ( d − 1 / p ) / 2 ) Γ ( 1 / ( 2 p ) ) ∥ ( Θ δ ( ρ λ ( t ) ) − Θ 0 ( ρ ) ) ρ − d 2 ∥ L 2 + O δ ( 1 ) . \left\|\left(\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)-\Theta_0(\rho)\right)\rho^{-\frac{1}{p}}\right\|_{\dot H^{s_c}}
=2^{s_c}\frac{\Gamma((d-1/p)/2)}{\Gamma(1/(2p))}
\left\|\left(\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)-\Theta_0(\rho)\right)\rho^{-\frac{d}{2}}\right\|_{L^2}+O_\delta(1). ( Θ δ ( λ ( t ) ρ ) − Θ 0 ( ρ ) ) ρ − p 1 H ˙ s c = 2 s c Γ ( 1/ ( 2 p )) Γ (( d − 1/ p ) /2 ) ( Θ δ ( λ ( t ) ρ ) − Θ 0 ( ρ ) ) ρ − 2 d L 2 + O δ ( 1 ) .
Since δ λ ( t ) ≥ 2 \delta\lambda(t)\geq2 δ λ ( t ) ≥ 2 , direct integration gives
∥ ( Θ δ ( ρ λ ( t ) ) − Θ 0 ( ρ ) ) ρ − d / 2 ∥ L 2 2 = ∣ S d − 1 ∣ ln ( δ λ ( t ) ) + O δ ( 1 ) . \left\|\left(\Theta_\delta\left(\frac{\rho}{\lambda(t)}\right)-\Theta_0(\rho)\right)\rho^{-d/2}\right\|_{L^2}^2
=|S^{d-1}|\ln(\delta\lambda(t))+O_\delta(1). ( Θ δ ( λ ( t ) ρ ) − Θ 0 ( ρ ) ) ρ − d /2 L 2 2 = ∣ S d − 1 ∣ ln ( δ λ ( t )) + O δ ( 1 ) .
Combining these identities with (113) , (101) , and ln ( δ λ ( t ) ) = ( 1 2 + ν ) ∣ ln t ∣ + ln δ \ln(\delta\lambda(t))=(\frac{1}{2}+\nu)|\ln t|+\ln\delta ln ( δ λ ( t )) = ( 2 1 + ν ) ∣ ln t ∣ + ln δ proves (102) . □
Finally, denote
R r a d a p ( t , ρ ) = − i t 1 + 2 ν ∂ t U r a d a p + ( 1 2 + ν ) i t 2 ν Λ U r a d a p − Δ U r a d a p − ∣ U r a d a p ∣ 2 p U r a d a p , (124) \mathcal{R}^{ap}_{rad}(t,\rho)=-it^{1+2\nu}\partial_tU^{ap}_{rad}+\left(\frac{1}{2}+\nu\right)it^{2\nu}\Lambda U^{ap}_{rad}-\Delta U^{ap}_{rad}-|U^{ap}_{rad}|^{2p}U^{ap}_{rad}, \tag*{(124)} R r a d a p ( t , ρ ) = − i t 1 + 2 ν ∂ t U r a d a p + ( 2 1 + ν ) i t 2 ν Λ U r a d a p − Δ U r a d a p − ∣ U r a d a p ∣ 2 p U r a d a p , (124)
and
R r a d a p ( t , x ) = − i ∂ t ψ r a d a p − Δ ψ r a d a p − ∣ ψ r a d a p ∣ 2 p ψ r a d a p . R^{ap}_{rad}(t,x)=-i\partial_t\psi^{ap}_{rad}-\Delta\psi^{ap}_{rad}-|\psi^{ap}_{rad}|^{2p}\psi^{ap}_{rad}. R r a d a p ( t , x ) = − i ∂ t ψ r a d a p − Δ ψ r a d a p − ∣ ψ r a d a p ∣ 2 p ψ r a d a p .
Then
R r a d a p ( t , x ) = λ 1 p + 2 ( t ) R r a d a p ( t , λ ( t ) r ) . R^{ap}_{rad}(t,x)=\lambda^{\frac{1}{p}+2}(t)\mathcal{R}^{ap}_{rad}(t,\lambda(t)r). R r a d a p ( t , x ) = λ p 1 + 2 ( t ) R r a d a p ( t , λ ( t ) r ) .
This relation leads to the following lemma.
Lemma 6.4. For any 1 ≤ s ≤ d 1\leq s\leq d 1 ≤ s ≤ d , there exists a positive constant C C C such that
∥ R r a d a p ( t , ⋅ ) ∥ H s ( R d ) ≤ C t ν ( 1 − 1 d ) ∣ λ − ∣ − ( 1 2 + ν d ) ( s − s c + 2 ) , (125) \left\|\mathcal{R}^{ap}_{rad}(t,\cdot)\right\|_{H^s(\mathbb{R}^d)}
\leq Ct^{\nu(1-\frac{1}{d})}|\lambda_-|^{-(\frac{1}{2}+\frac{\nu}{d})(s-s_c+2)}, \tag*{(125)} ∥ R r a d a p ( t , ⋅ ) ∥ H s ( R d ) ≤ C t ν ( 1 − d 1 ) ∣ λ − ∣ − ( 2 1 + d ν ) ( s − s c + 2 ) , (125) for all 0 < t ≤ T ( δ , L ) 0<t\leq T(\delta,L) 0 < t ≤ T ( δ , L ) .
Proof. Let Θ ~ ( ξ ) = ξ ⋅ ∇ Θ ( ξ ) \widetilde{\Theta}(\xi)=\xi\cdot\nabla\Theta(\xi) Θ ( ξ ) = ξ ⋅ ∇Θ ( ξ ) . By the definition (124) , R r a d a p \mathcal{R}^{ap}_{rad} R r a d a p can be decomposed into the following four parts:
R r a d a p = E 1 + E 2 + E 3 + E 4 , \mathcal{R}^{ap}_{rad}=E_1+E_2+E_3+E_4, R r a d a p = E 1 + E 2 + E 3 + E 4 ,
where
E 1 = i ( 1 2 + ϵ 1 ) t 2 ν Θ ~ ( t ν − ϵ 1 ρ ) ( U i n a p ( t , ρ ) − U s s a p ( t , ρ ) ) − 2 t ν − ϵ 1 ∇ Θ ( t ν − ϵ 1 ρ ) ⋅ ∇ ( U i n a p ( t , ρ ) − U s s a p ( t , ρ ) ) − t 2 ( ν − ϵ 1 ) Δ Θ ( t ν − ϵ 1 ρ ) ( U i n a p ( t , ρ ) − U s s a p ( t , ρ ) ) , E_1=i\left(\frac{1}{2}+\epsilon_1\right)t^{2\nu}\widetilde{\Theta}(t^{\nu-\epsilon_1}\rho)\left(U^{ap}_{in}(t,\rho)-U^{ap}_{ss}(t,\rho)\right)-2t^{\nu-\epsilon_1}\nabla\Theta(t^{\nu-\epsilon_1}\rho)\cdot\nabla\left(U^{ap}_{in}(t,\rho)-U^{ap}_{ss}(t,\rho)\right)
-t^{2(\nu-\epsilon_1)}\Delta\Theta(t^{\nu-\epsilon_1}\rho)\left(U^{ap}_{in}(t,\rho)-U^{ap}_{ss}(t,\rho)\right), E 1 = i ( 2 1 + ϵ 1 ) t 2 ν Θ ( t ν − ϵ 1 ρ ) ( U in a p ( t , ρ ) − U ss a p ( t , ρ ) ) − 2 t ν − ϵ 1 ∇Θ ( t ν − ϵ 1 ρ ) ⋅ ∇ ( U in a p ( t , ρ ) − U ss a p ( t , ρ ) ) − t 2 ( ν − ϵ 1 ) ΔΘ ( t ν − ϵ 1 ρ ) ( U in a p ( t , ρ ) − U ss a p ( t , ρ ) ) ,
E 2 = i ( 1 2 − ϵ 2 ) t 2 ν Θ ~ ( t ν + ϵ 2 ρ ) ( U s s a p ( t , ρ ) − U o u t a p ( t , ρ ) ) − 2 t ν + ϵ 2 ∇ Θ ( t ν + ϵ 2 ρ ) ⋅ ∇ ( U s s a p ( t , ρ ) − U o u t a p ( t , ρ ) ) − t 2 ( ν + ϵ 2 ) Δ Θ ( t ν + ϵ 2 ρ ) ( U s s a p ( t , ρ ) − U o u t a p ( t , ρ ) ) . E_2=i\left(\frac{1}{2}-\epsilon_2\right)t^{2\nu}\widetilde{\Theta}(t^{\nu+\epsilon_2}\rho)\left(U^{ap}_{ss}(t,\rho)-U^{ap}_{out}(t,\rho)\right)-2t^{\nu+\epsilon_2}\nabla\Theta(t^{\nu+\epsilon_2}\rho)\cdot\nabla\left(U^{ap}_{ss}(t,\rho)-U^{ap}_{out}(t,\rho)\right)
-t^{2(\nu+\epsilon_2)}\Delta\Theta(t^{\nu+\epsilon_2}\rho)\left(U^{ap}_{ss}(t,\rho)-U^{ap}_{out}(t,\rho)\right). E 2 = i ( 2 1 − ϵ 2 ) t 2 ν Θ ( t ν + ϵ 2 ρ ) ( U ss a p ( t , ρ ) − U o u t a p ( t , ρ ) ) − 2 t ν + ϵ 2 ∇Θ ( t ν + ϵ 2 ρ ) ⋅ ∇ ( U ss a p ( t , ρ ) − U o u t a p ( t , ρ ) ) − t 2 ( ν + ϵ 2 ) ΔΘ ( t ν + ϵ 2 ρ ) ( U ss a p ( t , ρ ) − U o u t a p ( t , ρ ) ) .
and E 3 E_3 E 3 , E 4 E_4 E 4 are given by
E 3 = Θ ( t ν − ϵ 1 ρ ) R i n a p ( t , ρ ) + ( 1 − Θ ( t ν − ϵ 1 ρ ) ) Θ ( t ν + ϵ 2 ρ ) R s s a p ( t , ρ ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) R o u t a p ( t , ρ ) , E_3=\Theta(t^{\nu-\epsilon_1}\rho)R^{\mathrm{ap}}_{\mathrm{in}}(t,\rho)+\left(1-\Theta(t^{\nu-\epsilon_1}\rho)\right)\Theta(t^{\nu+\epsilon_2}\rho)R^{\mathrm{ap}}_{\mathrm{ss}}(t,\rho)+\left(1-\Theta(t^{\nu+\epsilon_2}\rho)\right)R^{\mathrm{ap}}_{\mathrm{out}}(t,\rho), E 3 = Θ ( t ν − ϵ 1 ρ ) R in ap ( t , ρ ) + ( 1 − Θ ( t ν − ϵ 1 ρ ) ) Θ ( t ν + ϵ 2 ρ ) R ss ap ( t , ρ ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) R out ap ( t , ρ ) ,
E 4 = Θ ( t ν − ϵ 1 ρ ) ( ∣ U i n a p ∣ 2 p U i n a p − ∣ U r a d a p ∣ 2 p U r a d a p ) + ( 1 − Θ ( t ν − ϵ 1 ρ ) ) Θ ( t ν + ϵ 2 ρ ) ( ∣ U s s a p ∣ 2 p U s s a p − ∣ U r a d a p ∣ 2 p U r a d a p ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) ( ∣ U o u t a p ∣ 2 p U o u t a p − ∣ U r a d a p ∣ 2 p U r a d a p ) . \begin{aligned}
E_4={}&\Theta(t^{\nu-\epsilon_1}\rho)\left(\lvert U^{\mathrm{ap}}_{\mathrm{in}}\rvert^{2p}U^{\mathrm{ap}}_{\mathrm{in}}-\lvert U^{\mathrm{ap}}_{\mathrm{rad}}\rvert^{2p}U^{\mathrm{ap}}_{\mathrm{rad}}\right)\\
&+\left(1-\Theta(t^{\nu-\epsilon_1}\rho)\right)\Theta(t^{\nu+\epsilon_2}\rho)\left(\lvert U^{\mathrm{ap}}_{\mathrm{ss}}\rvert^{2p}U^{\mathrm{ap}}_{\mathrm{ss}}-\lvert U^{\mathrm{ap}}_{\mathrm{rad}}\rvert^{2p}U^{\mathrm{ap}}_{\mathrm{rad}}\right)\\
&+\left(1-\Theta(t^{\nu+\epsilon_2}\rho)\right)\left(\lvert U^{\mathrm{ap}}_{\mathrm{out}}\rvert^{2p}U^{\mathrm{ap}}_{\mathrm{out}}-\lvert U^{\mathrm{ap}}_{\mathrm{rad}}\rvert^{2p}U^{\mathrm{ap}}_{\mathrm{rad}}\right).
\end{aligned} E 4 = Θ ( t ν − ϵ 1 ρ ) ( ∣ U in ap ∣ 2 p U in ap − ∣ U rad ap ∣ 2 p U rad ap ) + ( 1 − Θ ( t ν − ϵ 1 ρ ) ) Θ ( t ν + ϵ 2 ρ ) ( ∣ U ss ap ∣ 2 p U ss ap − ∣ U rad ap ∣ 2 p U rad ap ) + ( 1 − Θ ( t ν + ϵ 2 ρ ) ) ( ∣ U out ap ∣ 2 p U out ap − ∣ U rad ap ∣ 2 p U rad ap ) .
In view of the estimate (72) , we obtain
∥ E 1 ∥ H ˙ s ( R d ) ≤ C t ( ν − ϵ 1 ) ( ∣ λ − ∣ + s − s c + 2 ) . (126) \lVert E_1\rVert_{\dot H^s(\mathbb{R}^d)}\le Ct^{(\nu-\epsilon_1)(\lvert\lambda_-\rvert+s-s_c+2)}. \tag*{(126)} ∥ E 1 ∥ H ˙ s ( R d ) ≤ C t ( ν − ϵ 1 ) (∣ λ − ∣ + s − s c + 2 ) . (126)
Similarly, from (28) and (88) , for 1 ≤ s ≤ d 1\le s\le d 1 ≤ s ≤ d , we deduce that
∥ E 2 ∥ H ˙ s ( R d ) ≤ C t 2 ν + ν ∣ λ − ∣ ( 1 − 2 ϵ 2 ) + ( ν + ϵ 2 ) ( s − s c ) ≤ C t ( ν − ϵ 1 ) ( ∣ λ − ∣ + s − s c + 2 ) , (127) \lVert E_2\rVert_{\dot H^s(\mathbb{R}^d)}\le Ct^{2\nu+\nu\lvert\lambda_-\rvert(1-2\epsilon_2)+(\nu+\epsilon_2)(s-s_c)}\le Ct^{(\nu-\epsilon_1)(\lvert\lambda_-\rvert+s-s_c+2)}, \tag*{(127)} ∥ E 2 ∥ H ˙ s ( R d ) ≤ C t 2 ν + ν ∣ λ − ∣ ( 1 − 2 ϵ 2 ) + ( ν + ϵ 2 ) ( s − s c ) ≤ C t ( ν − ϵ 1 ) (∣ λ − ∣ + s − s c + 2 ) , (127)
where the last inequality follows from ϵ 1 = ν d \epsilon_1=\frac{\nu}{d} ϵ 1 = d ν and ϵ 2 = 1 8 d \epsilon_2=\frac{1}{8d} ϵ 2 = 8 d 1 .
Next, we consider E 3 E_3 E 3 . With the choices (89) and (91) , the estimate (85) can be rewritten as
∥ ∇ l R o u t a p ∥ L 2 ( S o u t ) ≤ C l , N o u t t ν ( 1 − 1 / d ) ( ∣ λ − ∣ − s c + l + 2 ) , 1 ≤ l ≤ d . (128) \lVert\nabla^l R^{\mathrm{ap}}_{\mathrm{out}}\rVert_{L^2(S_{\mathrm{out}})}\le Cl,N_{\mathrm{out}}t^{\nu(1-1/d)(\lvert\lambda_-\rvert-s_c+l+2)},\qquad1\le l\le d. \tag*{(128)} ∥ ∇ l R out ap ∥ L 2 ( S out ) ≤ C l , N out t ν ( 1 − 1/ d ) (∣ λ − ∣ − s c + l + 2 ) , 1 ≤ l ≤ d . (128)
Then (54) , (70) , (128) , and (91) give
∥ E 3 ∥ H ˙ s ( R d ) ≤ C t ( ν − ϵ 1 ) ( ∣ λ − ∣ + s − s c + 2 ) , 1 ≤ s ≤ d . (129) \lVert E_3\rVert_{\dot H^s(\mathbb{R}^d)}\le Ct^{(\nu-\epsilon_1)(\lvert\lambda_-\rvert+s-s_c+2)},\qquad1\le s\le d. \tag*{(129)} ∥ E 3 ∥ H ˙ s ( R d ) ≤ C t ( ν − ϵ 1 ) (∣ λ − ∣ + s − s c + 2 ) , 1 ≤ s ≤ d . (129)
Finally, applying Lemma 3.2, Lemma 4.4, Lemma 5.5, the matching estimates (72) and (88) , and the choices of ϵ j \epsilon_j ϵ j , j = 1 , 2 j=1,2 j = 1 , 2 yields
∥ E 4 ∥ H ˙ s ( R d ) ≤ C t ( ν − ϵ 1 ) ( ∣ λ − ∣ + s − s c + 2 ) , 1 ≤ s ≤ d . (130) \lVert E_4\rVert_{\dot H^s(\mathbb{R}^d)}\le Ct^{(\nu-\epsilon_1)(\lvert\lambda_-\rvert+s-s_c+2)},\qquad1\le s\le d. \tag*{(130)} ∥ E 4 ∥ H ˙ s ( R d ) ≤ C t ( ν − ϵ 1 ) (∣ λ − ∣ + s − s c + 2 ) , 1 ≤ s ≤ d . (130)
Combining (126) with (127) , (129) , and (130) , we obtain, for fixed ϵ 1 = ν d \epsilon_1=\frac{\nu}{d} ϵ 1 = d ν , ϵ 2 = 1 8 d \epsilon_2=\frac{1}{8d} ϵ 2 = 8 d 1 , and any 1 ≤ s ≤ d 1\le s\le d 1 ≤ s ≤ d ,
∥ R r a d a p ( t ) ∥ H ˙ s ( R d ) ≤ C t ν ( 1 − 1 d ) ( ∣ λ − ∣ + s − s c + 2 ) . \lVert R^{\mathrm{ap}}_{\mathrm{rad}}(t)\rVert_{\dot H^s(\mathbb{R}^d)}\le Ct^{\nu(1-\frac{1}{d})(\lvert\lambda_-\rvert+s-s_c+2)}. ∥ R rad ap ( t ) ∥ H ˙ s ( R d ) ≤ C t ν ( 1 − d 1 ) (∣ λ − ∣ + s − s c + 2 ) .
Rescaling gives
∥ R r a d a p ( t ) ∥ H ˙ s ≤ C t ν ( 1 − 1 d ) ∣ λ − ∣ − ( 1 2 + ν d ) ( s − s c + 2 ) . \lVert R^{\mathrm{ap}}_{\mathrm{rad}}(t)\rVert_{\dot H^s}\le Ct^{\nu(1-\frac{1}{d})\lvert\lambda_-\rvert-(\frac{1}{2}+\frac{\nu}{d})(s-s_c+2)}. ∥ R rad ap ( t ) ∥ H ˙ s ≤ C t ν ( 1 − d 1 ) ∣ λ − ∣ − ( 2 1 + d ν ) ( s − s c + 2 ) .
Proof of the blowup result
Construction of a multi-bubble blowup solution
The radial approximate solution ψ r a d a p \psi^{\mathrm{ap}}_{\mathrm{rad}} ψ rad ap has been constructed in the previous sections. We now construct a K K K -bubble blowup solution ψ \psi ψ of (1) . Fix positive integers K ≥ 1 K\ge1 K ≥ 1 and p p p . Here d 0 ( p , K ) d_0(p,K) d 0 ( p , K ) is defined in (2) . For any dimension d > d 0 d>d_0 d > d 0 and any sequence of points { x j } 1 ≤ j ≤ K \{x_j\}_{1\le j\le K} { x j } 1 ≤ j ≤ K in R d \mathbb{R}^d R d , we seek a K K K -bubble blowup solution ψ ∈ H ˙ 1 ( R d ) ∩ H ˙ s ( R d ) \psi\in\dot H^1(\mathbb{R}^d)\cap\dot H^s(\mathbb{R}^d) ψ ∈ H ˙ 1 ( R d ) ∩ H ˙ s ( R d ) to (1) of the form
ψ ( t , x ) = λ 1 p ( t ) U ( t , y ) , y = λ ( t ) x , (131) \psi(t,x)=\lambda^{\frac{1}{p}}(t)U(t,y),\qquad y=\lambda(t)x, \tag*{(131)} ψ ( t , x ) = λ p 1 ( t ) U ( t , y ) , y = λ ( t ) x , (131)
where λ ( t ) = t − 1 2 − ν L \lambda(t)=t^{-\frac{1}{2}-\nu_L} λ ( t ) = t − 2 1 − ν L is as defined previously. We further decompose U U U as
U ( t , y ) = U a p ( t , y ) + f ( t , y ) , U a p ( t , y ) = ∑ j = 1 K U r a d a p ( t , y − y j ) . (132) U(t,y)=U^{\mathrm{ap}}(t,y)+f(t,y),\qquad U^{\mathrm{ap}}(t,y)=\sum_{j=1}^{K}U^{\mathrm{ap}}_{\mathrm{rad}}(t,y-y_j). \tag*{(132)} U ( t , y ) = U ap ( t , y ) + f ( t , y ) , U ap ( t , y ) = j = 1 ∑ K U rad ap ( t , y − y j ) . (132)
where y j = λ ( t ) x j y_j=\lambda(t)x_j y j = λ ( t ) x j . The parameter δ > 0 \delta>0 δ > 0 is chosen sufficiently small for the radial construction and, when K ≥ 2 K\geq2 K ≥ 2 , satisfies
0 < δ < 1 10 min i ≠ j ∣ x i − x j ∣ . (133) 0<\delta<\frac{1}{10}\min_{i\neq j}\lvert x_i-x_j\rvert. \tag*{(133)} 0 < δ < 10 1 i = j min ∣ x i − x j ∣ . (133)
The profile U r a d a p U^{ap}_{rad} U r a d a p denotes the approximate solution constructed in Section 3–Section 6.
To simplify the notation, we write
Q j ( y ) : = Q ( y − y j ) , U j a p ( t , y ) : = U r a d a p ( t , y − y j ) , Q_j(y):=Q(y-y_j),\qquad U^{ap}_j(t,y):=U^{ap}_{rad}(t,y-y_j), Q j ( y ) := Q ( y − y j ) , U j a p ( t , y ) := U r a d a p ( t , y − y j ) ,
and F ( u ) = ∣ u ∣ 2 p u F(u)=\lvert u\rvert^{2p}u F ( u ) = ∣ u ∣ 2 p u .
Substituting (131) into (1) and rescaling by the factor λ 2 p + 2 ( t ) \lambda^{\frac{2}{p}+2}(t) λ p 2 + 2 ( t ) , we rewrite the equation in terms of f f f and U a p U^{ap} U a p
∑ j = 1 K ( − i λ − 2 ∂ t U j a p − i λ − 3 λ ′ Λ U j a p − Δ U j a p − F ( U j a p ) ) − i λ − 2 f t − i λ − 3 λ ′ Λ f − Δ f − ( F ( U a p + f ) − ∑ j = 1 K F ( U j a p ) ) = 0. \sum_{j=1}^{K}\left(-i\lambda^{-2}\partial_tU^{ap}_j-i\lambda^{-3}\lambda'\Lambda U^{ap}_j-\Delta U^{ap}_j-F(U^{ap}_j)\right)
-i\lambda^{-2}f_t-i\lambda^{-3}\lambda'\Lambda f-\Delta f-\left(F(U^{ap}+f)-\sum_{j=1}^{K}F(U^{ap}_j)\right)=0. j = 1 ∑ K ( − i λ − 2 ∂ t U j a p − i λ − 3 λ ′ Λ U j a p − Δ U j a p − F ( U j a p ) ) − i λ − 2 f t − i λ − 3 λ ′ Λ f − Δ f − ( F ( U a p + f ) − j = 1 ∑ K F ( U j a p ) ) = 0.
Consequently, since the approximate solution U j a p U^{ap}_j U j a p satisfies (29) up to the remainder term R j a p \mathcal{R}^{ap}_j R j a p defined in (124) , the remainder f ⃗ = ( f f ˉ ) \vec f=\begin{pmatrix}f\\ \bar f\end{pmatrix} f = ( f f ˉ ) satisfies
i t 2 L ∣ λ + ∣ f ⃗ t = H K ( t ) f ⃗ + F ( f ) + r , (134) it^{\frac{2L}{\lvert\lambda_+\rvert}}\vec f_t=\mathcal{H}_K(t)\vec f+\mathcal{F}(f)+r, \tag*{(134)} i t ∣ λ + ∣ 2 L f t = H K ( t ) f + F ( f ) + r , (134)
where the principal operator H K \mathcal{H}_K H K is given by
H K f ⃗ = ( H K + i γ t 2 L ∣ λ + ∣ − 1 Λ ) f ⃗ = ( − Δ σ 3 − ( p + 1 ) ∑ j = 1 K Q j 2 p σ 3 − p ∑ j = 1 K Q j 2 p σ 3 σ 1 + i γ t 2 L ∣ λ + ∣ − 1 Λ ) f ⃗ . \begin{aligned}
\mathcal{H}_K\vec f=(H_K+i\gamma t^{\frac{2L}{\lvert\lambda_+\rvert}-1}\Lambda)\vec f \\
&=\left(-\Delta\sigma_3-(p+1)\sum_{j=1}^{K}Q_j^{2p}\sigma_3-p\sum_{j=1}^{K}Q_j^{2p}\sigma_3\sigma_1+i\gamma t^{\frac{2L}{\lvert\lambda_+\rvert}-1}\Lambda\right)\vec f.
\end{aligned} H K f = ( H K + iγ t ∣ λ + ∣ 2 L − 1 Λ ) f = ( − Δ σ 3 − ( p + 1 ) j = 1 ∑ K Q j 2 p σ 3 − p j = 1 ∑ K Q j 2 p σ 3 σ 1 + iγ t ∣ λ + ∣ 2 L − 1 Λ ) f .
with γ = L ∣ λ + ∣ \gamma=\frac{L}{\lvert\lambda_+\rvert} γ = ∣ λ + ∣ L . The term F ( f ) = F 0 + F 1 ( f ) + F 2 ( f ) \mathcal{F}(f)=F_0+F_1(f)+F_2(f) F ( f ) = F 0 + F 1 ( f ) + F 2 ( f ) , with F i = ( F i − F i ‾ ) F_i=\begin{pmatrix}F_i\\- \overline{F_i}\end{pmatrix} F i = ( F i − F i ) for i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 , is given by
F 0 = − F ( U a p ) + ∑ j = 1 K F ( U j a p ) , F 1 ( f ) = Υ 1 ( t ) f + Υ 2 ( t ) f ˉ , F_0=-F(U^{ap})+\sum_{j=1}^{K}F(U^{ap}_j),\qquad F_1(f)=\Upsilon_1(t)f+\Upsilon_2(t)\bar f, F 0 = − F ( U a p ) + j = 1 ∑ K F ( U j a p ) , F 1 ( f ) = Υ 1 ( t ) f + Υ 2 ( t ) f ˉ ,
Υ 1 ( t ) = ( p + 1 ) ( ∑ j = 1 K Q j 2 p − ∣ U a p ∣ 2 p ) , Υ 2 ( t ) = p ( ∑ j = 1 K Q j 2 p − ∣ U a p ∣ 2 p − 2 ( U a p ) 2 ) . (135) \Upsilon_1(t)=(p+1)\left(\sum_{j=1}^{K}Q_j^{2p}-\lvert U^{ap}\rvert^{2p}\right),\qquad
\Upsilon_2(t)=p\left(\sum_{j=1}^{K}Q_j^{2p}-\lvert U^{ap}\rvert^{2p-2}(U^{ap})^2\right). \tag*{(135)} Υ 1 ( t ) = ( p + 1 ) ( j = 1 ∑ K Q j 2 p − ∣ U a p ∣ 2 p ) , Υ 2 ( t ) = p ( j = 1 ∑ K Q j 2 p − ∣ U a p ∣ 2 p − 2 ( U a p ) 2 ) . (135)
F 2 ( f ) = − F ( U a p + f ) + F ( U a p ) + ( p + 1 ) ∣ U a p ∣ 2 p f + p ∣ U a p ∣ 2 p − 2 ( U a p ) 2 f ˉ , F_2(f)=-F(U^{ap}+f)+F(U^{ap})+(p+1)\lvert U^{ap}\rvert^{2p}f+p\lvert U^{ap}\rvert^{2p-2}(U^{ap})^2\bar f, F 2 ( f ) = − F ( U a p + f ) + F ( U a p ) + ( p + 1 ) ∣ U a p ∣ 2 p f + p ∣ U a p ∣ 2 p − 2 ( U a p ) 2 f ˉ ,
and the error term r = ( R a p − R a p ‾ ) r=\begin{pmatrix}\mathcal{R}^{ap}\\-\overline{\mathcal{R}^{ap}}\end{pmatrix} r = ( R a p − R a p ) is given by
R a p ( t , y ) = ∑ j = 1 K R j a p ( t , y ) , R j a p ( t , y ) : = R r a d a p ( t , y − y j ) . \mathcal{R}^{ap}(t,y)=\sum_{j=1}^{K}\mathcal{R}^{ap}_j(t,y),\qquad
\mathcal{R}^{ap}_j(t,y):=\mathcal{R}^{ap}_{rad}(t,y-y_j). R a p ( t , y ) = j = 1 ∑ K R j a p ( t , y ) , R j a p ( t , y ) := R r a d a p ( t , y − y j ) .
Here F 1 F_1 F 1 accounts for the difference between H K H_K H K and the linearization around U a p U^{ap} U a p . Indeed,
− Δ f − ( p + 1 ) ∑ j = 1 K Q j 2 p f − p ∑ j = 1 K Q j 2 p f ˉ + F 1 ( f ) = − Δ f − ( p + 1 ) ∣ U a p ∣ 2 p f − p ∣ U a p ∣ 2 p − 2 ( U a p ) 2 f ˉ . -\Delta f-(p+1)\sum_{j=1}^{K}Q_j^{2p}f-p\sum_{j=1}^{K}Q_j^{2p}\bar f+F_1(f)
=-\Delta f-(p+1)\lvert U^{ap}\rvert^{2p}f-p\lvert U^{ap}\rvert^{2p-2}(U^{ap})^2\bar f. − Δ f − ( p + 1 ) j = 1 ∑ K Q j 2 p f − p j = 1 ∑ K Q j 2 p f ˉ + F 1 ( f ) = − Δ f − ( p + 1 ) ∣ U a p ∣ 2 p f − p ∣ U a p ∣ 2 p − 2 ( U a p ) 2 f ˉ .
Main proposition
The goal of this subsection is to establish estimates for the remainder f f f , showing that the actual solution ψ \psi ψ to (1) blows up at t = 0 t=0 t = 0 while remaining close to the approximate solution
ψ a p ( t , x ) = λ 1 / p ( t ) U a p ( t , y ) . \psi^{ap}(t,x)=\lambda^{1/p}(t)U^{ap}(t,y). ψ a p ( t , x ) = λ 1/ p ( t ) U a p ( t , y ) .
The parameters ϵ 1 , ϵ 2 \epsilon_1,\epsilon_2 ϵ 1 , ϵ 2 are fixed by (89) . We first verify that there is an odd integer M ∈ ( d 2 , d ] M\in\left(\frac{d}{2},d\right] M ∈ ( 2 d , d ] satisfying
2 γ ( M − s c ) + 8 γ K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p ( ( 2 p + 1 ) M + 4 p ⌊ M + 1 4 ⌋ ) < − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) . (136) 2\gamma(M-s_c)+\frac{8\gamma Kc_{\infty}^{2p}}{(d-2)^2-4K(2p+1)c_{\infty}^{2p}}\left((2p+1)M+4p\left\lfloor\frac{M+1}{4}\right\rfloor\right)
< -4\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3). \tag*{(136)} 2 γ ( M − s c ) + ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 γ K c ∞ 2 p ( ( 2 p + 1 ) M + 4 p ⌊ 4 M + 1 ⌋ ) < − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) . (136)
Using c ∞ 2 p = 1 p ( d − 2 − 1 p ) c_{\infty}^{2p}=\frac{1}{p}\left(d-2-\frac{1}{p}\right) c ∞ 2 p = p 1 ( d − 2 − p 1 ) and 4 ⌊ M + 1 4 ⌋ ≤ M + 1 4\left\lfloor\frac{M+1}{4}\right\rfloor\leq M+1 4 ⌊ 4 M + 1 ⌋ ≤ M + 1 , together with p ≥ 1 p\geq1 p ≥ 1 , M ≤ d M\leq d M ≤ d , and d > 2 L L − 1 ( 24 K + 11 ) > 44 K + 8 d>\frac{2L}{L-1}(24K+11)>44K+8 d > L − 1 2 L ( 24 K + 11 ) > 44 K + 8 , we obtain
4 K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p ( ( 2 p + 1 ) M + 4 p ⌊ M + 1 4 ⌋ ) < 4 K ( 4 d + 1 ) d − 2 − 12 K < 22 K . (137) \frac{4Kc_{\infty}^{2p}}{(d-2)^2-4K(2p+1)c_{\infty}^{2p}}\left((2p+1)M+4p\left\lfloor\frac{M+1}{4}\right\rfloor\right)
<\frac{4K(4d+1)}{d-2-12K}<22K. \tag*{(137)} ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 4 K c ∞ 2 p ( ( 2 p + 1 ) M + 4 p ⌊ 4 M + 1 ⌋ ) < d − 2 − 12 K 4 K ( 4 d + 1 ) < 22 K . (137)
Moreover, the identities ν γ = 1 − ∣ λ + ∣ 2 L \frac{\nu}{\gamma}=1-\frac{|\lambda_+|}{2L} γ ν = 1 − 2 L ∣ λ + ∣ and ∣ λ − ∣ = 2 s c − 2 − ∣ λ + ∣ |\lambda_-|=2s_c-2-|\lambda_+| ∣ λ − ∣ = 2 s c − 2 − ∣ λ + ∣ give
− 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) 2 γ > ( 1 − 1 2 L ) d − s c − 7. (138) \frac{-4\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)}{2\gamma}
>\left(1-\frac{1}{2L}\right)d-s_c-7. \tag*{(138)} 2 γ − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) > ( 1 − 2 L 1 ) d − s c − 7. (138)
Here we used 2 < ∣ λ + ∣ < 5 2 2<|\lambda_+|<\frac{5}{2} 2 < ∣ λ + ∣ < 2 5 , ( ∣ λ + ∣ − 2 ) s c < 4 (|\lambda_+|-2)s_c<4 ( ∣ λ + ∣ − 2 ) s c < 4 , 0 < ν γ < 1 0<\frac{\nu}{\gamma}<1 0 < γ ν < 1 , and ∣ λ − ∣ − s c + 3 < d 2 |\lambda_-|-s_c+3<\frac{d}{2} ∣ λ − ∣ − s c + 3 < 2 d , which hold under the dimension assumption.
Now take the odd integer
M = 2 ⌊ ( 2 L − 1 ) d 4 L ⌋ − 24 K − 9. M=2\left\lfloor\frac{(2L-1)d}{4L}\right\rfloor-24K-9. M = 2 ⌊ 4 L ( 2 L − 1 ) d ⌋ − 24 K − 9.
Since a − 1 < [ a ] ≤ a a-1<[a]\leq a a − 1 < [ a ] ≤ a and d > 2 L L − 1 ( 24 K + 11 ) d>\frac{2L}{L-1}(24K+11) d > L − 1 2 L ( 24 K + 11 ) , we have
d 2 < ( 1 − 1 2 L ) d − 24 K − 11 < M ≤ ( 1 − 1 2 L ) d − 24 K − 9 < d , (139) \frac{d}{2}<\left(1-\frac{1}{2L}\right)d-24K-11<M\leq\left(1-\frac{1}{2L}\right)d-24K-9<d, \tag*{(139)} 2 d < ( 1 − 2 L 1 ) d − 24 K − 11 < M ≤ ( 1 − 2 L 1 ) d − 24 K − 9 < d , (139)
and hence
M − ( 1 − 1 2 L ) d + 22 K + 7 ≤ − 2 K − 2. (140) M-\left(1-\frac{1}{2L}\right)d+22K+7\leq-2K-2. \tag*{(140)} M − ( 1 − 2 L 1 ) d + 22 K + 7 ≤ − 2 K − 2. (140)
Combining (137) –(140) , we obtain
2 γ ( M − s c ) + 8 γ K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p ( ( 2 p + 1 ) M + 4 p ⌊ M + 1 4 ⌋ ) < − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − 4 γ ( K + 1 ) < − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) . \begin{aligned}
2\gamma(M-s_c)+\frac{8\gamma Kc_{\infty}^{2p}}{(d-2)^2-4K(2p+1)c_{\infty}^{2p}}\left((2p+1)M+4p\left\lfloor\frac{M+1}{4}\right\rfloor\right)
< -4\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)-4\gamma(K+1) \\
&< -4\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3).
\end{aligned} 2 γ ( M − s c ) + ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 γ K c ∞ 2 p ( ( 2 p + 1 ) M + 4 p ⌊ 4 M + 1 ⌋ ) < − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 4 γ ( K + 1 ) < − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) .
Thus the admissible set in (136) is nonempty. Let M M M be the largest odd integer satisfying (136) . By the first inequality in (139) and (137) –(138) , every odd integer in ( d 2 , d 2 + 4 ] \left(\frac{d}{2},\frac{d}{2}+4\right] ( 2 d , 2 d + 4 ] is admissible. Hence M > d 2 + 2 M>\frac{d}{2}+2 M > 2 d + 2 , and therefore
s c + 4 ν 1 + 2 ν < d 2 + 2 < M . (141) s_c+\frac{4\nu}{1+2\nu}<\frac{d}{2}+2<M. \tag*{(141)} s c + 1 + 2 ν 4 ν < 2 d + 2 < M . (141)
We now establish the following proposition for sufficiently small η > 0 \eta>0 η > 0 , whose choice will be fixed in the proof.
Proposition 7.1. For any 1 ≤ s ≤ M 1 \le s \le M 1 ≤ s ≤ M , there exists a small positive time T ∗ < T ( δ , K , L ) T^* < T(\delta, K, L) T ∗ < T ( δ , K , L ) such that, for any t ∈ [ t 1 , T ∗ ] t \in[t_1,T^*] t ∈ [ t 1 , T ∗ ] , the Cauchy problem
{ i ψ t = − Δ ψ − ∣ ψ ∣ 2 p ψ , ψ ∣ t = t 1 = ψ a p ( t 1 ) , \begin{cases}
i\psi_t=-\Delta\psi-|\psi|^{2p}\psi,\\
\left.\psi\right|_{t=t_1}=\psi^{ap}(t_1),
\end{cases} { i ψ t = − Δ ψ − ∣ ψ ∣ 2 p ψ , ψ ∣ t = t 1 = ψ a p ( t 1 ) , admits a solution ψ \psi ψ such that ψ − ψ a p ∈ C ( [ t 1 , T ∗ ] , H ˙ 1 ∩ H ˙ s ) \psi-\psi^{ap}\in C([t_1,T^*],\dot H^1\cap\dot H^s) ψ − ψ a p ∈ C ([ t 1 , T ∗ ] , H ˙ 1 ∩ H ˙ s ) and, for any 1 < s ≤ s c 1<s\le s_c 1 < s ≤ s c ,
∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ s ≤ C t − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η 2 , \|\psi(t,x)-\psi^{ap}(t,x)\|_{\dot H^s}\le Ct^{-2\nu+\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\frac{\eta}{2}}, ∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ s ≤ C t − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η , and for any s c < s ≤ M s_c<s\le M s c < s ≤ M ,
∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ s ≤ C t − L ∣ λ + ∣ ( s − s c ) − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η 2 . \|\psi(t,x)-\psi^{ap}(t,x)\|_{\dot H^s}\le Ct^{-\frac{L}{|\lambda_+|}(s-s_c)-2\nu+\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\frac{\eta}{2}}. ∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ s ≤ C t − ∣ λ + ∣ L ( s − s c ) − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η . Furthermore, ∣ x ∣ ( ψ ( t ) − ψ a p ( t ) ) ∈ L 2 |x|(\psi(t)-\psi^{ap}(t))\in L^2 ∣ x ∣ ( ψ ( t ) − ψ a p ( t )) ∈ L 2 and
∥ ∣ x ∣ ( ψ ( t , x ) − ψ a p ( t , x ) ) ∥ L 2 ≤ C t ν ( 1 − 1 d ) ∣ λ − ∣ + ( 1 2 + ν d ) ( s c + 1 ) − 1 − 4 ν d . \||x|(\psi(t,x)-\psi^{ap}(t,x))\|_{L^2}\le Ct^{\nu(1-\frac{1}{d})|\lambda_-|+(\frac{1}{2}+\frac{\nu}{d})(s_c+1)-1-\frac{4\nu}{d}}. ∥∣ x ∣ ( ψ ( t , x ) − ψ a p ( t , x )) ∥ L 2 ≤ C t ν ( 1 − d 1 ) ∣ λ − ∣ + ( 2 1 + d ν ) ( s c + 1 ) − 1 − d 4 ν . where the constant C C C changes from line to line and is independent of t 1 t_1 t 1 .
Before proving Proposition Proposition 7.1 , we introduce several key operators used throughout this subsection. Taking into account the single bubble expression (132) , we decompose F 1 F_1 F 1 as
F 1 ( f ) = t 2 ν B ( f ) + F ^ 1 ( f ) . (142) F_1(f)=t^{2\nu}\mathcal{B}(f)+\widehat{F}_1(f). \tag*{(142)} F 1 ( f ) = t 2 ν B ( f ) + F 1 ( f ) . (142)
where
B ( f ) = ( B ( f ) − B ( f ) ) \mathcal{B}(f)=\begin{pmatrix}B(f)\\-B(f)\end{pmatrix} B ( f ) = ( B ( f ) − B ( f ) )
is determined by the first order corrections in the inner region as follows
B ( f ) = − ( p + 1 ) ∑ j = 1 K Q j 2 p − 1 ( p χ 1 , j + p χ 1 , j ‾ ) f − p ∑ j = 1 K Q j 2 p − 1 ( ( p + 1 ) χ 1 , j + ( p − 1 ) χ 1 , j ‾ ) f ‾ . B(f)=-(p+1)\sum_{j=1}^{K}Q_j^{2p-1}(p\chi_{1,j}+p\overline{\chi_{1,j}})f-p\sum_{j=1}^{K}Q_j^{2p-1}\bigl((p+1)\chi_{1,j}+(p-1)\overline{\chi_{1,j}}\bigr)\overline{f}. B ( f ) = − ( p + 1 ) j = 1 ∑ K Q j 2 p − 1 ( p χ 1 , j + p χ 1 , j ) f − p j = 1 ∑ K Q j 2 p − 1 ( ( p + 1 ) χ 1 , j + ( p − 1 ) χ 1 , j ) f .
Since χ 1 , j ( y ) = χ 1 ( y − y j ) \chi_{1,j}(y)=\chi_1(y-y_j) χ 1 , j ( y ) = χ 1 ( y − y j ) is purely imaginary, we have χ 1 , j ‾ = − χ 1 , j \overline{\chi_{1,j}}=-\chi_{1,j} χ 1 , j = − χ 1 , j . Then,
B ( f ) = − 2 p ∑ j = 1 K Q j 2 p − 1 χ 1 , j f ‾ , B ( f ) = − 2 p ∑ j = 1 K Q j 2 p − 1 χ 1 , j σ 1 f ⃗ . (143) B(f)=-2p\sum_{j=1}^{K}Q_j^{2p-1}\chi_{1,j}\overline{f},\qquad\mathcal{B}(f)=-2p\sum_{j=1}^{K}Q_j^{2p-1}\chi_{1,j}\sigma_1\vec{f}. \tag*{(143)} B ( f ) = − 2 p j = 1 ∑ K Q j 2 p − 1 χ 1 , j f , B ( f ) = − 2 p j = 1 ∑ K Q j 2 p − 1 χ 1 , j σ 1 f . (143)
In particular, σ 3 B \sigma_3\mathcal{B} σ 3 B is self-adjoint and σ 3 B = − B σ 3 \sigma_3\mathcal{B}=-\mathcal{B}\sigma_3 σ 3 B = − B σ 3 by using
σ 1 ∗ = σ 1 , σ 3 ∗ = σ 3 , σ 1 σ 3 = − σ 3 σ 1 . \sigma_1^*=\sigma_1,\qquad\sigma_3^*=\sigma_3,\qquad\sigma_1\sigma_3=-\sigma_3\sigma_1. σ 1 ∗ = σ 1 , σ 3 ∗ = σ 3 , σ 1 σ 3 = − σ 3 σ 1 .
The term F ^ 1 ( f ) \widehat{F}_1(f) F 1 ( f ) is the remainder after subtracting t 2 ν B ( f ) t^{2\nu}\mathcal{B}(f) t 2 ν B ( f ) from F 1 ( f ) F_1(f) F 1 ( f ) defined in (135) . We also introduce the higher order derivatives of f f f associated with the linearized operator H K H_K H K , defined by
f k = H K k f , for k ≥ 0. (144) f_k=H_K^k f,\qquad\text{for }k\ge0. \tag*{(144)} f k = H K k f , for k ≥ 0. (144)
In particular, f 0 = f f_0=f f 0 = f .
We now prove Proposition Proposition 7.1 by a bootstrap argument. With M M M fixed by (136) , choose η = η ( p , d , K , L ) > 0 \eta=\eta(p,d,K,L)>0 η = η ( p , d , K , L ) > 0 sufficiently small and T ∗ < min { 1 , T ( δ , K , L ) } T^*<\min\{1,T(\delta,K,L)\} T ∗ < min { 1 , T ( δ , K , L )} . Write
ψ ( t , x ) − ψ a p ( t , x ) = λ 1 2 ( t ) f ( t , y ) , y = λ ( t ) x , \psi(t,x)-\psi^{ap}(t,x)=\lambda^{\frac{1}{2}}(t)f(t,y),\qquad y=\lambda(t)x, ψ ( t , x ) − ψ a p ( t , x ) = λ 2 1 ( t ) f ( t , y ) , y = λ ( t ) x ,
and assume on [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , where 0 < t 1 < t 2 ≤ T ∗ 0<t_1<t_2\le T^* 0 < t 1 < t 2 ≤ T ∗ , that
∥ ∇ f ∥ H M − 1 ≤ C t − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η . (145) \|\nabla f\|_{H^{M-1}}\le Ct^{-2\nu+\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\eta}. \tag*{(145)} ∥∇ f ∥ H M − 1 ≤ C t − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η . (145)
Since M > d 2 M>\frac{d}{2} M > 2 d , Sobolev interpolation gives
∥ f ∥ L ∞ ≤ C t − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η ≤ C t 2 ν + 3 η . (146) \|f\|_{L^\infty}\le Ct^{-2\nu+\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\eta}\le Ct^{2\nu+3\eta}. \tag*{(146)} ∥ f ∥ L ∞ ≤ C t − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η ≤ C t 2 ν + 3 η . (146)
Under the bootstrap assumption (7.15), we first give the following estimates for the nonlinear term and the error of approximation. These follow from Lemmas 6.1 and 6.4 and will be used in both the H ˙ 1 \dot{H}^{1} H ˙ 1 and higher order energy estimates.
Lemma 7.2. There exists a positive constant C C C such that, under the bootstrap assumption (7.15), the following estimates hold for any t ∈ [ t 1 , t 2 ] t \in[t_{1},t_{2}] t ∈ [ t 1 , t 2 ]
∥ ∇ H K k F 2 ( f ) ∥ L 2 ≤ C t − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − 2 η , 1 ≤ 2 k + 1 ≤ M , (147) \left\lVert\nabla H_{K}^{k}\mathcal{F}_{2}(f)\right\rVert_{L^{2}} \leq Ct^{-4\nu+2\nu(1-\frac{1}{d})(|\lambda_{-}|-s_{c}+3)-2\eta}, \qquad1 \leq2k+1 \leq M, \tag*{(147)} ∇ H K k F 2 ( f ) L 2 ≤ C t − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η , 1 ≤ 2 k + 1 ≤ M , (147) and
∥ ∇ H K k r ∥ L 2 ≤ C t ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) , 1 ≤ 2 k + 1 ≤ M . (148) \left\lVert\nabla H_{K}^{k}r\right\rVert_{L^{2}} \leq Ct^{\nu(1-\frac{1}{d})(|\lambda_{-}|-s_{c}+3)}, \qquad1 \leq2k+1 \leq M. \tag*{(148)} ∇ H K k r L 2 ≤ C t ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) , 1 ≤ 2 k + 1 ≤ M . (148) Here 0 < t 1 ≤ t 2 ≤ T 0 < t_{1} \leq t_{2} \leq T 0 < t 1 ≤ t 2 ≤ T .
The nonlinear estimate follows from the standard product estimates and Hardy’s inequality, using the differentiated profile bounds and (7.15). The approximation error estimate follows from (6.37) by rescaling and expanding ∇ H K k \nabla H_{K}^{k} ∇ H K k , together with Hardy’s inequality.
We now establish the energy estimates and close the bootstrap argument in the following four steps.
Step 1. Energy control. To establish the H ˙ 1 ( R d ) \dot{H}^{1}(\mathbb{R}^{d}) H ˙ 1 ( R d ) estimate for f f f , we begin with the modified energy functional
J 0 ( t ) = ⟨ H K f , σ 3 f ⟩ + t 2 L ∣ λ + ∣ − 1 Re ⟨ f , σ 3 B ( f ) ⟩ . (149) J_{0}(t)=\langle H_{K}f,\sigma_{3}f\rangle+t^{\frac{2L}{|\lambda_{+}|}-1}\operatorname{Re}\langle f,\sigma_{3}B(f)\rangle. \tag*{(149)} J 0 ( t ) = ⟨ H K f , σ 3 f ⟩ + t ∣ λ + ∣ 2 L − 1 Re ⟨ f , σ 3 B ( f )⟩ . (149)
The expansion for χ 1 \chi_{1} χ 1 in Lemma 3.1 gives
∣ Q j 2 p − 1 χ 1 , j ∣ ≤ C ⟨ y − y j ⟩ − ∣ λ + ∣ . \left|Q_{j}^{2p-1}\chi_{1,j}\right|\leq C\langle y-y_{j}\rangle^{-|\lambda_{+}|}. Q j 2 p − 1 χ 1 , j ≤ C ⟨ y − y j ⟩ − ∣ λ + ∣ .
Since ∣ λ + ∣ > 2 |\lambda_{+}|>2 ∣ λ + ∣ > 2 , (7.13) and Hardy’s inequality imply
∣ Re ⟨ f , σ 3 B ( f ) ⟩ ∣ ≤ C ∥ ∇ f ∥ L 2 2 . \left|\operatorname{Re}\langle f,\sigma_{3}B(f)\rangle\right|\leq C\left\lVert\nabla f\right\rVert_{L^{2}}^{2}. ∣ Re ⟨ f , σ 3 B ( f )⟩ ∣ ≤ C ∥ ∇ f ∥ L 2 2 .
Combining this bound with Lemma 2.4 and choosing T ∗ T^{*} T ∗ sufficiently small, we obtain
∥ ∇ f ∥ L 2 2 ≤ 2 ( d − 2 ) 2 ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p J 0 ( t ) . (150) \left\lVert\nabla f\right\rVert_{L^{2}}^{2}\leq\frac{2(d-2)^{2}}{(d-2)^{2}-4K(2p+1)c_{\infty}^{2p}}J_{0}(t). \tag*{(150)} ∥ ∇ f ∥ L 2 2 ≤ ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 2 ( d − 2 ) 2 J 0 ( t ) . (150)
The derivative of J 0 J_{0} J 0 is given by
d d t J 0 ( t ) = ⟨ V t f , σ 3 f ⟩ + ⟨ H K f t , σ 3 f ⟩ + ⟨ H K f , σ 3 f t ⟩ + ( 2 L ∣ λ + ∣ − 1 ) t 2 L ∣ λ + ∣ − 2 Re ⟨ f , σ 3 B ( f ) ⟩ + t 2 L ∣ λ + ∣ − 1 ( Re ⟨ f t , σ 3 B ( f ) ⟩ + Re ⟨ f , σ 3 ( B ( f ) ) t ⟩ ) , \frac{\mathrm{d}}{\mathrm{d}t}J_{0}(t)=\langle V_{t}f,\sigma_{3}f\rangle+\langle H_{K}f_{t},\sigma_{3}f\rangle+\langle H_{K}f,\sigma_{3}f_{t}\rangle+\left(\frac{2L}{|\lambda_{+}|}-1\right)t^{\frac{2L}{|\lambda_{+}|}-2}\operatorname{Re}\langle f,\sigma_{3}B(f)\rangle
+t^{\frac{2L}{|\lambda_{+}|}-1}\left(\operatorname{Re}\langle f_{t},\sigma_{3}B(f)\rangle+\operatorname{Re}\langle f,\sigma_{3}(B(f))_{t}\rangle\right), d t d J 0 ( t ) = ⟨ V t f , σ 3 f ⟩ + ⟨ H K f t , σ 3 f ⟩ + ⟨ H K f , σ 3 f t ⟩ + ( ∣ λ + ∣ 2 L − 1 ) t ∣ λ + ∣ 2 L − 2 Re ⟨ f , σ 3 B ( f )⟩ + t ∣ λ + ∣ 2 L − 1 ( Re ⟨ f t , σ 3 B ( f )⟩ + Re ⟨ f , σ 3 ( B ( f ) ) t ⟩ ) ,
where V = − ( p + 1 ) ∑ j = 1 K Q j 2 p σ 3 − p ∑ j = 1 K Q j 2 p σ 3 σ 1 V=-(p+1)\sum_{j=1}^{K}Q_{j}^{2p}\sigma_{3}-p\sum_{j=1}^{K}Q_{j}^{2p}\sigma_{3}\sigma_{1} V = − ( p + 1 ) ∑ j = 1 K Q j 2 p σ 3 − p ∑ j = 1 K Q j 2 p σ 3 σ 1 . By the self-adjointness of σ 3 B \sigma_{3}B σ 3 B , we have
Re ⟨ f , σ 3 B ( f t ) ⟩ = Re ⟨ f t , σ 3 B ( f ) ⟩ . \operatorname{Re}\langle f,\sigma_{3}B(f_{t})\rangle=\operatorname{Re}\langle f_{t},\sigma_{3}B(f)\rangle. Re ⟨ f , σ 3 B ( f t )⟩ = Re ⟨ f t , σ 3 B ( f )⟩ .
Combining this identity with
∂ t y j = L λ + + t y j t = − γ y j t , (151) \partial_{t}y_{j}=\frac{L}{\lambda_{+}+t}\frac{y_{j}}{t}=-\gamma\frac{y_{j}}{t}, \tag*{(151)} ∂ t y j = λ + + t L t y j = − γ t y j , (151)
we obtain
Re ⟨ f t , σ 3 B ( f ) ⟩ + Re ⟨ f , σ 3 ( B ( f ) ) t ⟩ = 2 Re ⟨ f t , σ 3 B ( f ) ⟩ − 2 p γ t Re ⟨ f , σ 3 ∑ j = 1 K ( y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) ) σ 1 f ⟩ . (152) \operatorname{Re}\langle f_{t},\sigma_{3}B(f)\rangle+\operatorname{Re}\langle f,\sigma_{3}(B(f))_{t}\rangle
=2\operatorname{Re}\langle f_{t},\sigma_{3}B(f)\rangle-\frac{2p\gamma}{t}\operatorname{Re}\left\langle f,\sigma_{3}\sum_{j=1}^{K}\left(y_{j}\cdot\nabla(Q_{j}^{2p-1}\chi_{1,j})\right)\sigma_{1}f\right\rangle. \tag*{(152)} Re ⟨ f t , σ 3 B ( f )⟩ + Re ⟨ f , σ 3 ( B ( f ) ) t ⟩ = 2 Re ⟨ f t , σ 3 B ( f )⟩ − t 2 p γ Re ⟨ f , σ 3 j = 1 ∑ K ( y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) ) σ 1 f ⟩ . (152)
Substituting (134) into the derivative of J 0 J_0 J 0 and using (151) , (152) , we obtain
d d t J 0 ( t ) = γ t ( ∑ j = 1 K ⟨ y j ⋅ ∇ V j f , σ 3 f ⟩ + ⟨ H K Λ f , σ 3 f ⟩ + ⟨ H K f , σ 3 Λ f ⟩ ) + i t 2 L ∣ λ + ∣ ( ⟨ H K f , σ 3 ( F ( f ) + r ) ⟩ − ⟨ H K ( F ( f ) + r ) , σ 3 f ⟩ ) + 2 t Im ⟨ H K f + F ( f ) + r , σ 3 B ( f ) ⟩ + t 2 L ∣ λ + ∣ − 2 ( Re ⟨ ( 2 γ − 1 ) f + 2 γ Λ f , σ 3 B ( f ) ⟩ − 2 p γ Re ⟨ f , σ 3 ∑ j = 1 K y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) σ 1 f ⟩ ) . \begin{aligned}
\frac{\mathrm{d}}{\mathrm{d}t}J_0(t)={}&\frac{\gamma}{t}\left(\sum_{j=1}^{K}\langle y_j\cdot\nabla V_jf,\sigma_3f\rangle+\langle H_K\Lambda f,\sigma_3f\rangle+\langle H_Kf,\sigma_3\Lambda f\rangle\right)\\
&+it^{\frac{2L}{|\lambda_+|}}\left(\langle H_Kf,\sigma_3(\mathcal{F}(f)+r)\rangle-\langle H_K(\mathcal{F}(f)+r),\sigma_3f\rangle\right)\\
&+\frac{2}{t}\operatorname{Im}\langle H_Kf+\mathcal{F}(f)+r,\sigma_3\mathcal{B}(f)\rangle\\
&+t^{\frac{2L}{|\lambda_+|}-2}\left(\operatorname{Re}\langle(2\gamma-1)f+2\gamma\Lambda f,\sigma_3\mathcal{B}(f)\rangle-2p\gamma\operatorname{Re}\left\langle f,\sigma_3\sum_{j=1}^{K}y_j\cdot\nabla(Q_j^{2p-1}\chi_{1,j})\sigma_1f\right\rangle\right).
\end{aligned} d t d J 0 ( t ) = t γ ( j = 1 ∑ K ⟨ y j ⋅ ∇ V j f , σ 3 f ⟩ + ⟨ H K Λ f , σ 3 f ⟩ + ⟨ H K f , σ 3 Λ f ⟩ ) + i t ∣ λ + ∣ 2 L ( ⟨ H K f , σ 3 ( F ( f ) + r )⟩ − ⟨ H K ( F ( f ) + r ) , σ 3 f ⟩ ) + t 2 Im ⟨ H K f + F ( f ) + r , σ 3 B ( f )⟩ + t ∣ λ + ∣ 2 L − 2 ( Re ⟨( 2 γ − 1 ) f + 2 γ Λ f , σ 3 B ( f )⟩ − 2 p γ Re ⟨ f , σ 3 j = 1 ∑ K y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) σ 1 f ⟩ ) .
Integrating by parts gives
⟨ H K Λ f , σ 3 f ⟩ + ⟨ H K f , σ 3 Λ f ⟩ = 2 ( 1 − s c ) ⟨ H K f , σ 3 f ⟩ − ⟨ ( y ⋅ ∇ V + 2 V ) f , σ 3 f ⟩ . (153) \langle H_K\Lambda f,\sigma_3f\rangle+\langle H_Kf,\sigma_3\Lambda f\rangle=2(1-s_c)\langle H_Kf,\sigma_3f\rangle-\langle(y\cdot\nabla V+2V)f,\sigma_3f\rangle. \tag*{(153)} ⟨ H K Λ f , σ 3 f ⟩ + ⟨ H K f , σ 3 Λ f ⟩ = 2 ( 1 − s c ) ⟨ H K f , σ 3 f ⟩ − ⟨( y ⋅ ∇ V + 2 V ) f , σ 3 f ⟩ . (153)
Using (143) and (153) , we decompose the derivative of (149) into the following parts:
d d t J 0 ( t ) = E 1 + E 2 + E 3 + E 4 , (154) \frac{\mathrm{d}}{\mathrm{d}t}J_0(t)=\mathcal{E}_1+\mathcal{E}_2+\mathcal{E}_3+\mathcal{E}_4, \tag*{(154)} d t d J 0 ( t ) = E 1 + E 2 + E 3 + E 4 , (154)
with
E 1 = 2 γ ( 1 − s c ) t ⟨ H K f , σ 3 f ⟩ , \mathcal{E}_1=\frac{2\gamma(1-s_c)}{t}\langle H_Kf,\sigma_3f\rangle, E 1 = t 2 γ ( 1 − s c ) ⟨ H K f , σ 3 f ⟩ ,
E 2 = − γ t ∑ j = 1 K ⟨ ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f , σ 3 f ⟩ , \mathcal{E}_2=-\frac{\gamma}{t}\sum_{j=1}^{K}\left\langle\left((y-y_j)\cdot\nabla V_j+2V_j\right)f,\sigma_3f\right\rangle, E 2 = − t γ j = 1 ∑ K ⟨ ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f , σ 3 f ⟩ ,
E 3 = − 2 t − 2 L ∣ λ + ∣ Im ⟨ H K f , σ 3 ( F 1 ( f ) + F 2 ( f ) + r ) ⟩ + 2 t Im ⟨ F ( f ) + r , σ 3 B ( f ) ⟩ , \mathcal{E}_3=-2t^{-\frac{2L}{|\lambda_+|}}\operatorname{Im}\langle H_Kf,\sigma_3(\mathcal{F}_1(f)+\mathcal{F}_2(f)+r)\rangle+\frac{2}{t}\operatorname{Im}\langle\mathcal{F}(f)+r,\sigma_3\mathcal{B}(f)\rangle, E 3 = − 2 t − ∣ λ + ∣ 2 L Im ⟨ H K f , σ 3 ( F 1 ( f ) + F 2 ( f ) + r )⟩ + t 2 Im ⟨ F ( f ) + r , σ 3 B ( f )⟩ ,
and
E 4 = t 2 L ∣ λ + ∣ − 2 ( Re ⟨ ( 2 γ − 1 ) f + 2 γ Λ f , σ 3 B ( f ) ⟩ − 2 p γ Re ⟨ f , σ 3 ∑ j = 1 K y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) σ 1 f ⟩ ) . \mathcal{E}_4=t^{\frac{2L}{|\lambda_+|}-2}\left(\operatorname{Re}\langle(2\gamma-1)f+2\gamma\Lambda f,\sigma_3\mathcal{B}(f)\rangle-2p\gamma\operatorname{Re}\left\langle f,\sigma_3\sum_{j=1}^{K}y_j\cdot\nabla(Q_j^{2p-1}\chi_{1,j})\sigma_1f\right\rangle\right). E 4 = t ∣ λ + ∣ 2 L − 2 ( Re ⟨( 2 γ − 1 ) f + 2 γ Λ f , σ 3 B ( f )⟩ − 2 p γ Re ⟨ f , σ 3 j = 1 ∑ K y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) σ 1 f ⟩ ) .
First, we estimate the term E 2 \mathcal{E}_2 E 2 . By the definitions
V j = − ( p + 1 ) Q 2 p ( y − y j ) σ 3 − p Q 2 p ( y − y j ) σ 3 σ 1 , V_j=-(p+1)Q^{2p}(y-y_j)\sigma_3-pQ^{2p}(y-y_j)\sigma_3\sigma_1, V j = − ( p + 1 ) Q 2 p ( y − y j ) σ 3 − p Q 2 p ( y − y j ) σ 3 σ 1 ,
and
Λ Q j ( y ) : = ( Λ Q ) ( y − y j ) = 1 p Q j ( y ) + ( y − y j ) ⋅ ∇ y Q j ( y ) , \Lambda Q_j(y):=(\Lambda Q)(y-y_j)=\frac{1}{p}Q_j(y)+(y-y_j)\cdot\nabla_yQ_j(y), Λ Q j ( y ) := ( Λ Q ) ( y − y j ) = p 1 Q j ( y ) + ( y − y j ) ⋅ ∇ y Q j ( y ) ,
E 2 \mathcal{E}_2 E 2 can be rewritten as
E 2 = 2 p γ t ∑ j = 1 K ⟨ ( ( p + 1 ) Q j 2 p − 1 Λ Q j σ 3 + p Q j 2 p − 1 Λ Q j σ 3 σ 1 ) f , σ 3 f ⟩ . \mathcal{E}_2=\frac{2p\gamma}{t}\sum_{j=1}^{K}\left\langle\left((p+1)Q_j^{2p-1}\Lambda Q_j\sigma_3+pQ_j^{2p-1}\Lambda Q_j\sigma_3\sigma_1\right)f,\sigma_3f\right\rangle. E 2 = t 2 p γ j = 1 ∑ K ⟨ ( ( p + 1 ) Q j 2 p − 1 Λ Q j σ 3 + p Q j 2 p − 1 Λ Q j σ 3 σ 1 ) f , σ 3 f ⟩ .
Using the estimates (2.3), (2.4), and γ = L ∣ λ + ∣ > 0 \gamma=\frac{L}{|\lambda_+|}>0 γ = ∣ λ + ∣ L > 0 , we obtain
E 2 ≤ 2 ( 2 p + 1 ) γ c ∞ 2 p K t ( 2 d − 2 ) 2 ∥ ∇ f ∥ L 2 2 . \mathcal{E}_2\leq\frac{2(2p+1)\gamma c_\infty^{2p}K}{t}\left(\frac{2}{d-2}\right)^2\|\nabla f\|_{L^2}^2. E 2 ≤ t 2 ( 2 p + 1 ) γ c ∞ 2 p K ( d − 2 2 ) 2 ∥∇ f ∥ L 2 2 .
by the Hardy inequality. Combining this with (2.10), we get
E 2 ≤ A 1 γ t ⟨ H K f , σ 3 f ⟩ , where A 1 = 8 ( 2 p + 1 ) K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p . (155) \mathcal{E}_2\leq\frac{A_1\gamma}{t}\langle H_Kf,\sigma_3f\rangle,\qquad\text{where}\quad A_1=\frac{8(2p+1)Kc_\infty^{2p}}{(d-2)^2-4K(2p+1)c_\infty^{2p}}. \tag*{(155)} E 2 ≤ t A 1 γ ⟨ H K f , σ 3 f ⟩ , where A 1 = ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 ( 2 p + 1 ) K c ∞ 2 p . (155)
We next estimate E 3 \mathcal{E}_3 E 3 . Since
supp U j a p ( t , y ) ⊂ { y ∈ R d : ∣ y − y j ∣ ≤ 2 δ λ ( t ) } , \operatorname{supp} U_j^{ap}(t,y) \subset\{y \in\mathbb{R}^d : |y-y_j| \le2\delta\lambda(t)\}, supp U j a p ( t , y ) ⊂ { y ∈ R d : ∣ y − y j ∣ ≤ 2 δ λ ( t )} ,
condition (133) ensures that these supports are well separated. Hence,
F 0 = − F ( U a p ) + ∑ j = 1 K F ( U j a p ) = 0 , (156) F_0=-F(U^{ap})+\sum_{j=1}^{K}F(U_j^{ap})=0, \tag*{(156)} F 0 = − F ( U a p ) + j = 1 ∑ K F ( U j a p ) = 0 , (156)
and
⟨ H K f , σ 3 F 0 ( f ) ⟩ = 0. \langle H_Kf,\sigma_3F_0(f)\rangle=0. ⟨ H K f , σ 3 F 0 ( f )⟩ = 0.
Moreover, the self-adjointness of σ 3 \sigma_3 σ 3 implies that
Im ⟨ B ( f ) , σ 3 B ( f ) ⟩ = 0. \operatorname{Im}\langle\mathcal{B}(f),\sigma_3\mathcal{B}(f)\rangle=0. Im ⟨ B ( f ) , σ 3 B ( f )⟩ = 0.
Thus, E 3 \mathcal{E}_3 E 3 splits into two parts:
E 3 = − 2 t − 2 L ∣ λ + ∣ Im ⟨ H K f , σ 3 ( F ^ 1 ( f ) + F 2 ( f ) + r ) ⟩ + 2 t Im ⟨ F ^ 1 ( f ) + F 2 ( f ) + r , σ 3 B ( f ) ⟩ . (157) \begin{aligned}
\mathcal{E}_3={}&-2t^{-\frac{2L}{|\lambda_+|}}\operatorname{Im}\langle H_Kf,\sigma_3(\widehat{F}_1(f)+F_2(f)+r)\rangle\\
&+\frac{2}{t}\operatorname{Im}\langle\widehat{F}_1(f)+F_2(f)+r,\sigma_3\mathcal{B}(f)\rangle. \tag*{(157)}
\end{aligned} E 3 = − 2 t − ∣ λ + ∣ 2 L Im ⟨ H K f , σ 3 ( F 1 ( f ) + F 2 ( f ) + r )⟩ + t 2 Im ⟨ F 1 ( f ) + F 2 ( f ) + r , σ 3 B ( f )⟩ . (157)
Substituting (142) and (143) into (135) and using the fact that χ 1 , j \chi_{1,j} χ 1 , j is purely imaginary, we obtain
F ^ 1 ( f ) = − ( p + 1 ) ( ∣ U a p ∣ 2 p − ∑ j = 1 K Q j 2 p ) f − p ( ∣ U a p ∣ 2 p − 2 ( U a p ) 2 − ∑ j = 1 K Q j 2 p − 1 − 2 t 2 ν ∑ j = 1 K Q j 2 p − 1 χ 1 , j ) f ‾ . (158) \widehat{F}_1(f)=-(p+1)\left(\lvert U^{ap}\rvert^{2p}-\sum_{j=1}^{K}Q_j^{2p}\right)f
-p\left(\lvert U^{ap}\rvert^{2p-2}(U^{ap})^2-\sum_{j=1}^{K}Q_j^{2p-1}-2t^{2\nu}\sum_{j=1}^{K}Q_j^{2p-1}\chi_{1,j}\right)\overline{f}. \tag*{(158)} F 1 ( f ) = − ( p + 1 ) ( ∣ U a p ∣ 2 p − j = 1 ∑ K Q j 2 p ) f − p ( ∣ U a p ∣ 2 p − 2 ( U a p ) 2 − j = 1 ∑ K Q j 2 p − 1 − 2 t 2 ν j = 1 ∑ K Q j 2 p − 1 χ 1 , j ) f . (158)
Set
χ s s a p ( t , ρ ) = U s s a p ( t , ρ ) − Q ( ρ ) , χ o u t a p ( t , ρ ) = U o u t a p ( t , ρ ) − Q ( ρ ) , \chi_{ss}^{ap}(t,\rho)=U_{ss}^{ap}(t,\rho)-Q(\rho),\qquad\chi_{out}^{ap}(t,\rho)=U_{out}^{ap}(t,\rho)-Q(\rho), χ ss a p ( t , ρ ) = U ss a p ( t , ρ ) − Q ( ρ ) , χ o u t a p ( t , ρ ) = U o u t a p ( t , ρ ) − Q ( ρ ) ,
and denote their translates by χ s s , j a p ( t , y ) = χ s s a p ( t , y − y j ) \chi_{ss,j}^{ap}(t,y)=\chi_{ss}^{ap}(t,y-y_j) χ ss , j a p ( t , y ) = χ ss a p ( t , y − y j ) and χ o u t , j a p ( t , y ) = χ o u t a p ( t , y − y j ) \chi_{out,j}^{ap}(t,y)=\chi_{out}^{ap}(t,y-y_j) χ o u t , j a p ( t , y ) = χ o u t a p ( t , y − y j ) . Using χ k ( y − y j ) = ( − 1 ) k χ k ( y − y j ) \chi_k(y-y_j)=(-1)^k\chi_k(y-y_j) χ k ( y − y j ) = ( − 1 ) k χ k ( y − y j ) in Lemma 3.1, and taking (132) into account, we find
F ^ 1 ( f ) = − ( p + 1 ) [ ∑ j = 1 K p Q j 2 p − 1 ( y ) ( Θ ( t ν − ϵ 1 ( y − y j ) ) ∑ 2 ≤ k ≤ N i n k e v e n t 2 ν k ( χ k , j ( y ) + χ k , j ( y ) ‾ ) + ( 1 − Θ ( t ν − ϵ 1 ( y − y j ) ) ) Θ ( t ν + ϵ 2 ( y − y j ) ) ( χ s s , j a p ( t , y ) + χ s s , j a p ( t , y ) ‾ ) + ( 1 − Θ ( t ν + ϵ 2 ( y − y j ) ) ) ( χ o u t , j a p ( t , y ) + χ o u t , j a p ( t , y ) ‾ ) ] + N 1 ( χ 2 ) ] f − p [ ∑ j = 1 K Q j 2 p − 1 ( y ) ( − 2 ( 1 − Θ ( t ν − ϵ 1 ( y − y j ) ) ) t 2 ν χ 1 , j ( y ) + 2 Θ ( t ν − ϵ 1 ( y − y j ) ) ( p ∑ 2 ≤ k ≤ N i n k e v e n t 2 ν k χ k , j ( y ) + ∑ 2 ≤ k ≤ N i n k o d d t 2 ν k χ k , j ( y ) ) + ( 1 − Θ ( t ν − ϵ 1 ( y − y j ) ) ) Θ ( t ν + ϵ 2 ( y − y j ) ) ( ( p + 1 ) χ s s , j a p ( t , y ) + ( p − 1 ) χ s s , j a p ( t , y ) ‾ ) + ( 1 − Θ ( t ν + ϵ 2 ( y − y j ) ) ) ( ( p + 1 ) χ o u t , j a p ( t , y ) + ( p − 1 ) χ o u t , j a p ( t , y ) ‾ ) ] + N 2 ( χ 2 ) ] f ‾ . \begin{aligned}
\widehat{F}_1(f)={}&-(p+1)\Bigg[\sum_{j=1}^{K}pQ_j^{2p-1}(y)\Bigg(\Theta(t^{\nu-\epsilon_1}(y-y_j))\sum_{\substack{2\le k\le N_{\mathrm{in}}\\ k\ \mathrm{even}}}t^{2\nu k}\big(\chi_{k,j}(y)+\overline{\chi_{k,j}(y)}\big)\\
&+\big(1-\Theta(t^{\nu-\epsilon_1}(y-y_j))\big)\Theta(t^{\nu+\epsilon_2}(y-y_j))\big(\chi_{ss,j}^{ap}(t,y)+\overline{\chi_{ss,j}^{ap}(t,y)}\big)\\
&+\big(1-\Theta(t^{\nu+\epsilon_2}(y-y_j))\big)\big(\chi_{out,j}^{ap}(t,y)+\overline{\chi_{out,j}^{ap}(t,y)}\big)\Bigg]+N_1(\chi^2)\Bigg]f\\
&-p\Bigg[\sum_{j=1}^{K}Q_j^{2p-1}(y)\Bigg(-2\big(1-\Theta(t^{\nu-\epsilon_1}(y-y_j))\big)t^{2\nu}\chi_{1,j}(y)\\
&+2\Theta(t^{\nu-\epsilon_1}(y-y_j))\Bigg(p\sum_{\substack{2\le k\le N_{\mathrm{in}}\\ k\ \mathrm{even}}}t^{2\nu k}\chi_{k,j}(y)+\sum_{\substack{2\le k\le N_{\mathrm{in}}\\ k\ \mathrm{odd}}}t^{2\nu k}\chi_{k,j}(y)\Bigg)\\
&+\big(1-\Theta(t^{\nu-\epsilon_1}(y-y_j))\big)\Theta(t^{\nu+\epsilon_2}(y-y_j))\big((p+1)\chi_{ss,j}^{ap}(t,y)+(p-1)\overline{\chi_{ss,j}^{ap}(t,y)}\big)\\
&+\big(1-\Theta(t^{\nu+\epsilon_2}(y-y_j))\big)\big((p+1)\chi_{out,j}^{ap}(t,y)+(p-1)\overline{\chi_{out,j}^{ap}(t,y)}\big)\Bigg]+N_2(\chi^2)\Bigg]\overline{f}.
\end{aligned} F 1 ( f ) = − ( p + 1 ) [ j = 1 ∑ K p Q j 2 p − 1 ( y ) ( Θ ( t ν − ϵ 1 ( y − y j )) 2 ≤ k ≤ N in k even ∑ t 2 ν k ( χ k , j ( y ) + χ k , j ( y ) ) + ( 1 − Θ ( t ν − ϵ 1 ( y − y j )) ) Θ ( t ν + ϵ 2 ( y − y j )) ( χ ss , j a p ( t , y ) + χ ss , j a p ( t , y ) ) + ( 1 − Θ ( t ν + ϵ 2 ( y − y j )) ) ( χ o u t , j a p ( t , y ) + χ o u t , j a p ( t , y ) ) ] + N 1 ( χ 2 ) ] f − p [ j = 1 ∑ K Q j 2 p − 1 ( y ) ( − 2 ( 1 − Θ ( t ν − ϵ 1 ( y − y j )) ) t 2 ν χ 1 , j ( y ) + 2Θ ( t ν − ϵ 1 ( y − y j )) ( p 2 ≤ k ≤ N in k even ∑ t 2 ν k χ k , j ( y ) + 2 ≤ k ≤ N in k odd ∑ t 2 ν k χ k , j ( y ) ) + ( 1 − Θ ( t ν − ϵ 1 ( y − y j )) ) Θ ( t ν + ϵ 2 ( y − y j )) ( ( p + 1 ) χ ss , j a p ( t , y ) + ( p − 1 ) χ ss , j a p ( t , y ) ) + ( 1 − Θ ( t ν + ϵ 2 ( y − y j )) ) ( ( p + 1 ) χ o u t , j a p ( t , y ) + ( p − 1 ) χ o u t , j a p ( t , y ) ) ] + N 2 ( χ 2 ) ] f .
where N i ( χ 2 ) N_i(\chi^2) N i ( χ 2 ) , i = 1 , 2 i=1,2 i = 1 , 2 , are second-order terms with respect to U r a d a p − Q U_{rad}^{ap}-Q U r a d a p − Q . Since χ 1 , j \chi_{1,j} χ 1 , j is purely imaginary, its linear contribution to ∣ U j a p ∣ 2 p \lvert U_j^{ap}\rvert^{2p} ∣ U j a p ∣ 2 p vanishes, while the corresponding first order term in the second coefficient is explicitly subtracted in (158) . The differentiated expansions in Lemma 3.1, (63) , and (82) , together with the endpoint bounds in the self-similar and remote regions and the Leibniz rule, yield the following estimate for 0 ≤ l ≤ M 0 \le l \le M 0 ≤ l ≤ M :
∥ ⟨ y − y j ⟩ l ∇ l ( ∣ U j a p ∣ 2 p − Q j 2 p ) ∥ L ∞ + ∥ ⟨ y − y j ⟩ l ∇ l ( ( U j a p ) 2 p − 2 ( U j a p ) 2 − Q j 2 p − 2 t 2 ν Q j 2 p − 1 χ 1 , j ) ∥ L ∞ ≤ C t 2 ν + 8 η . (159) \begin{aligned}
&\left\|\langle y-y_j\rangle^l\nabla^l\left(\lvert U_j^{ap}\rvert^{2p}-Q_j^{2p}\right)\right\|_{L^\infty}\\
&\quad+\left\|\langle y-y_j\rangle^l\nabla^l\left((U_j^{ap})^{2p-2}(U_j^{ap})^2-Q_j^{2p}-2t^{2\nu}Q_j^{2p-1}\chi_{1,j}\right)\right\|_{L^\infty}
\le Ct^{2\nu+8\eta}. \tag*{(159)}
\end{aligned} ⟨ y − y j ⟩ l ∇ l ( ∣ U j a p ∣ 2 p − Q j 2 p ) L ∞ + ⟨ y − y j ⟩ l ∇ l ( ( U j a p ) 2 p − 2 ( U j a p ) 2 − Q j 2 p − 2 t 2 ν Q j 2 p − 1 χ 1 , j ) L ∞ ≤ C t 2 ν + 8 η . (159)
By the dimension assumption and (89) , the differentiated expansions give time orders strictly larger than 2 ν 2\nu 2 ν . Thus (159) holds for η > 0 \eta>0 η > 0 sufficiently small. Since the supports of U j a p U_j^{ap} U j a p are well separated, combining (158) , (159) with Hardy’s inequality yields
∥ F ^ 1 ( f ) ∥ H ˙ 1 ≤ C t 2 ν + 8 η ∥ ∇ f ∥ L 2 . (160) \|\widehat F_1(f)\|_{\dot H^1}\le Ct^{2\nu+8\eta}\|\nabla f\|_{L^2}. \tag*{(160)} ∥ F 1 ( f ) ∥ H ˙ 1 ≤ C t 2 ν + 8 η ∥∇ f ∥ L 2 . (160)
Moreover, the expansion of χ 1 \chi_1 χ 1 in Lemma 3.1 gives
∣ ∇ l ( Q j 2 p − 1 χ 1 , j ) ∣ ≤ C ⟨ y − y j ⟩ − ∣ λ + ∣ − l , 0 ≤ l ≤ M . (161) \left|\nabla^l\left(Q_j^{2p-1}\chi_{1,j}\right)\right|\le C\langle y-y_j\rangle^{-|\lambda_+|-l},\qquad0\le l\le M. \tag*{(161)} ∇ l ( Q j 2 p − 1 χ 1 , j ) ≤ C ⟨ y − y j ⟩ − ∣ λ + ∣ − l , 0 ≤ l ≤ M . (161)
Therefore, applying Hardy’s inequality to (143) and using (161) , we obtain
∥ ∇ B ( f ) ∥ L 2 ≤ C ∥ ∇ f ∥ L 2 . (162) \|\nabla\mathcal{B}(f)\|_{L^2}\le C\|\nabla f\|_{L^2}. \tag*{(162)} ∥∇ B ( f ) ∥ L 2 ≤ C ∥∇ f ∥ L 2 . (162)
To estimate the first term in (157) , we bound its three contributions separately. By (160) and the bootstrap assumption (145) ,
∣ ⟨ H K f , σ 3 F ^ 1 ( f ) ⟩ ∣ ≤ C ∥ ∇ f ∥ L 2 ∥ F ^ 1 ( f ) ∥ H ˙ 1 ≤ C t − 2 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) + 6 η . (163) \left|\langle H_Kf,\sigma_3\widehat F_1(f)\rangle\right|
\le C\|\nabla f\|_{L^2}\|\widehat F_1(f)\|_{\dot H^1}
\le Ct^{-2\nu+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)+6\eta}. \tag*{(163)} ⟨ H K f , σ 3 F 1 ( f )⟩ ≤ C ∥∇ f ∥ L 2 ∥ F 1 ( f ) ∥ H ˙ 1 ≤ C t − 2 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) + 6 η . (163)
Similarly, (145) and (147) with k = 0 k=0 k = 0 give
∣ ⟨ H K f , σ 3 F 2 ( f ) ⟩ ∣ ≤ C ∥ ∇ f ∥ L 2 ∥ F 2 ( f ) ∥ H ˙ 1 ≤ C t − 6 ν + 3 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − 3 η . (164) \left|\langle H_Kf,\sigma_3\mathcal F_2(f)\rangle\right|
\le C\|\nabla f\|_{L^2}\|\mathcal F_2(f)\|_{\dot H^1}
\le Ct^{-6\nu+3\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-3\eta}. \tag*{(164)} ∣ ⟨ H K f , σ 3 F 2 ( f )⟩ ∣ ≤ C ∥∇ f ∥ L 2 ∥ F 2 ( f ) ∥ H ˙ 1 ≤ C t − 6 ν + 3 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 3 η . (164)
and
∣ ⟨ H K f , σ 3 r ⟩ ∣ ≤ C t − 2 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η . (165) \left|\langle H_Kf,\sigma_3r\rangle\right|
\le Ct^{-2\nu+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\eta}. \tag*{(165)} ∣ ⟨ H K f , σ 3 r ⟩ ∣ ≤ C t − 2 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η . (165)
For the second term in (157) , the estimates (160) and (162) , together with Lemma 7.2 and Hardy’s inequality, imply
2 t ∣ Im ⟨ F ^ 1 ( f ) + F 2 ( f ) + r , σ 3 B ( f ) ⟩ ∣ ≤ C t ∥ ∇ f ∥ L 2 ( ∥ F ^ 1 ( f ) ∥ H ˙ 1 + ∥ F 2 ( f ) ∥ H ˙ 1 + ∥ r ∥ H ˙ 1 ) ≤ C t − 2 ν − 1 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η . (166) \begin{aligned}
\frac{2}{t}\left|\operatorname{Im}\left\langle\widehat F_1(f)+\mathcal F_2(f)+r,\sigma_3\mathcal B(f)\right\rangle\right|
&\le\frac{C}{t}\|\nabla f\|_{L^2}\left(\|\widehat F_1(f)\|_{\dot H^1}+\|\mathcal F_2(f)\|_{\dot H^1}+\|r\|_{\dot H^1}\right)\\
&\le Ct^{-2\nu-1+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\eta}. \tag*{(166)}
\end{aligned} t 2 Im ⟨ F 1 ( f ) + F 2 ( f ) + r , σ 3 B ( f ) ⟩ ≤ t C ∥∇ f ∥ L 2 ( ∥ F 1 ( f ) ∥ H ˙ 1 + ∥ F 2 ( f ) ∥ H ˙ 1 + ∥ r ∥ H ˙ 1 ) ≤ C t − 2 ν − 1 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η . (166)
Combining (163) –(166) and using 2 L ∣ λ + ∣ = 1 + 2 ν \frac{2L}{|\lambda_+|}=1+2\nu ∣ λ + ∣ 2 L = 1 + 2 ν , we obtain
∣ E 3 ( t ) ∣ ≤ C t − 4 ν − 1 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η . (167) |\mathcal E_3(t)|\le Ct^{-4\nu-1+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\eta}. \tag*{(167)} ∣ E 3 ( t ) ∣ ≤ C t − 4 ν − 1 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η . (167)
We finally estimate E 4 \mathcal E_4 E 4 . Integrating the term containing Λ f \Lambda f Λ f by parts and combining it with the terms containing y j y_j y j , we obtain
E 4 = 2 p t 2 L ∣ λ + ∣ − 2 Re ⟨ f , σ 3 ∑ j = 1 K ( γ ( y − y j ) ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) + ( 2 γ s c − 2 ν ) Q j 2 p − 1 χ 1 , j ) σ 1 f ⟩ . \mathcal E_4=2pt^{\frac{2L}{|\lambda_+|}-2}\operatorname{Re}\left\langle f,\sigma_3\sum_{j=1}^{K}\left(\gamma(y-y_j)\cdot\nabla\left(Q_j^{2p-1}\chi_{1,j}\right)+(2\gamma s_c-2\nu)Q_j^{2p-1}\chi_{1,j}\right)\sigma_1f\right\rangle. E 4 = 2 p t ∣ λ + ∣ 2 L − 2 Re ⟨ f , σ 3 j = 1 ∑ K ( γ ( y − y j ) ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) + ( 2 γ s c − 2 ν ) Q j 2 p − 1 χ 1 , j ) σ 1 f ⟩ .
Using (161) , ∣ λ + ∣ > 2 |\lambda_{+}|>2 ∣ λ + ∣ > 2 , Hardy’s inequality, and (145) , we obtain
∣ E 4 ( t ) ∣ ≤ C t 2 ν − 1 ∥ ∇ f ∥ L 2 2 ≤ C t − 2 ν − 1 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − 2 η . (168) |\mathcal{E}_{4}(t)|\le Ct^{2\nu-1}\|\nabla f\|_{L^{2}}^{2}\le Ct^{-2\nu-1+2\nu(1-\frac{1}{d})(|\lambda_{-}|-s_{c}+3)-2\eta}. \tag*{(168)} ∣ E 4 ( t ) ∣ ≤ C t 2 ν − 1 ∥∇ f ∥ L 2 2 ≤ C t − 2 ν − 1 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η . (168)
The definition (149) and Hardy’s inequality show that replacing ⟨ H K f , σ 3 f ⟩ \langle H_{K}f,\sigma_{3}f\rangle ⟨ H K f , σ 3 f ⟩ by J 0 J_{0} J 0 in the estimate for E 1 + E 2 \mathcal{E}_{1}+\mathcal{E}_{2} E 1 + E 2 introduces an error bounded by C t 2 ν − 1 ∥ ∇ f ∥ L 2 2 Ct^{2\nu-1}\|\nabla f\|_{L^{2}}^{2} C t 2 ν − 1 ∥∇ f ∥ L 2 2 . This error satisfies the same bound as E 4 \mathcal{E}_{4} E 4 in (168) .
Substituting the estimates (155) , (167) , and (168) into (154) , and using 0 < η < 2 ν 0<\eta<2\nu 0 < η < 2 ν and T > 0 T>0 T > 0 sufficiently small, we conclude that
d d t J 0 ( t ) ≤ C 0 ( K , L ) t J 0 ( t ) + C t − 4 ν − 1 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η , (169) \frac{\mathrm{d}}{\mathrm{d}t}J_{0}(t)\le\frac{C_{0}(K,L)}{t}J_{0}(t)+Ct^{-4\nu-1+2\nu(1-\frac{1}{d})(|\lambda_{-}|-s_{c}+3)-\eta}, \tag*{(169)} d t d J 0 ( t ) ≤ t C 0 ( K , L ) J 0 ( t ) + C t − 4 ν − 1 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η , (169)
where
C 0 ( K , L ) = 2 γ ( 1 − s c ) + 8 ( 2 p + 1 ) γ K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p . (170) C_{0}(K,L)=2\gamma(1-s_{c})+\frac{8(2p+1)\gamma Kc_{\infty}^{2p}}{(d-2)^{2}-4K(2p+1)c_{\infty}^{2p}}. \tag*{(170)} C 0 ( K , L ) = 2 γ ( 1 − s c ) + ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 ( 2 p + 1 ) γ K c ∞ 2 p . (170)
Since c ∞ 2 p = 1 p ( d − 2 − 1 p ) c_{\infty}^{2p}=\frac{1}{p}(d-2-\frac{1}{p}) c ∞ 2 p = p 1 ( d − 2 − p 1 ) , p ≥ 1 p\ge1 p ≥ 1 , and d > 2 L L − 1 ( 24 K + 11 ) > 36 K + 2 d>\frac{2L}{L-1}(24K+11)>36K+2 d > L − 1 2 L ( 24 K + 11 ) > 36 K + 2 , we have
8 ( 2 p + 1 ) K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p < 24 K d − 2 − 12 K < 1. \frac{8(2p+1)Kc_{\infty}^{2p}}{(d-2)^{2}-4K(2p+1)c_{\infty}^{2p}}<\frac{24K}{d-2-12K}<1. ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 ( 2 p + 1 ) K c ∞ 2 p < d − 2 − 12 K 24 K < 1.
Consequently, C 0 ( K , L ) < γ ( 3 − 2 s c ) < 0 C_{0}(K,L)<\gamma(3-2s_{c})<0 C 0 ( K , L ) < γ ( 3 − 2 s c ) < 0 .
Step 2. Control of the higher regularity. For 1 ≤ k ≤ M − 1 2 L 1\le k\le\frac{M-1}{2L} 1 ≤ k ≤ 2 L M − 1 , we define the higher order energy functional by
J k ( t ) = ⟨ H K f k , σ 3 f k ⟩ + ( − 1 ) k t 2 L ∣ λ + ∣ − 1 Re ⟨ f k , σ 3 B ( f k ) ⟩ , (171) J_{k}(t)=\langle H_{K}f_{k},\sigma_{3}f_{k}\rangle+(-1)^{k}t^{\frac{2L}{|\lambda_{+}|}-1}\operatorname{Re}\langle f_{k},\sigma_{3}\mathcal{B}(f_{k})\rangle, \tag*{(171)} J k ( t ) = ⟨ H K f k , σ 3 f k ⟩ + ( − 1 ) k t ∣ λ + ∣ 2 L − 1 Re ⟨ f k , σ 3 B ( f k )⟩ , (171)
where f k f_{k} f k is given in (144) and B ( f ) \mathcal{B}(f) B ( f ) is defined in (143) . We first show that the corrected energy J k J_{k} J k , together with the lower bound of J 0 J_{0} J 0 , controls the higher Sobolev norms of f f f . By (161) , Hardy’s inequality, and Lemma 2.4,
∣ Re ⟨ f k , σ 3 B ( f k ) ⟩ ∣ ≤ C ∥ ∇ f k ∥ L 2 2 ≤ C ⟨ H K f k , σ 3 f k ⟩ . |\operatorname{Re}\langle f_{k},\sigma_{3}\mathcal{B}(f_{k})\rangle|\le C\|\nabla f_{k}\|_{L^{2}}^{2}\le C\langle H_{K}f_{k},\sigma_{3}f_{k}\rangle. ∣ Re ⟨ f k , σ 3 B ( f k )⟩ ∣ ≤ C ∥∇ f k ∥ L 2 2 ≤ C ⟨ H K f k , σ 3 f k ⟩ .
Thus, for T > 0 T>0 T > 0 sufficiently small, there exists a constant C > 0 C>0 C > 0 such that
( 1 − C t 2 ν ) ⟨ H K f k , σ 3 f k ⟩ ≤ J k ( t ) ≤ ( 1 + C t 2 ν ) ⟨ H K f k , σ 3 f k ⟩ . (172) (1-Ct^{2\nu})\langle H_{K}f_{k},\sigma_{3}f_{k}\rangle\le J_{k}(t)\le(1+Ct^{2\nu})\langle H_{K}f_{k},\sigma_{3}f_{k}\rangle. \tag*{(172)} ( 1 − C t 2 ν ) ⟨ H K f k , σ 3 f k ⟩ ≤ J k ( t ) ≤ ( 1 + C t 2 ν ) ⟨ H K f k , σ 3 f k ⟩ . (172)
Combining this estimate with the coercivity estimate (26) , we obtain
∥ ∇ f k ∥ L 2 2 ≤ C J k . (173) \|\nabla f_{k}\|_{L^{2}}^{2}\le CJ_{k}. \tag*{(173)} ∥∇ f k ∥ L 2 2 ≤ C J k . (173)
Since f k = H K k f f_{k}=H_{K}^{k}f f k = H K k f , expanding H K k H_{K}^{k} H K k and applying Hardy’s inequality to terms containing no derivative of f f f , and interpolating the remaining lower order terms gives
∥ ∇ f ∥ H 2 k ≤ C ( ∥ ∇ f k ∥ L 2 + ∥ ∇ f ∥ L 2 ) . \|\nabla f\|_{H^{2k}}\le C(\|\nabla f_{k}\|_{L^{2}}+\|\nabla f\|_{L^{2}}). ∥∇ f ∥ H 2 k ≤ C ( ∥∇ f k ∥ L 2 + ∥∇ f ∥ L 2 ) .
Together with the estimate (150) and (173) , this yields
∥ ∇ f ∥ H 2 k 2 ≤ C ( J k + J 0 ) . (174) \|\nabla f\|_{H^{2k}}^{2}\le C(J_{k}+J_{0}). \tag*{(174)} ∥∇ f ∥ H 2 k 2 ≤ C ( J k + J 0 ) . (174)
To compute the derivative of J k J_{k} J k , we first derive the equation satisfied by f k f_{k} f k .
Claim 7.3. For any k ≥ 1 k\ge1 k ≥ 1 , the function f k f_{k} f k defined in (144) solves the equation
i t 2 L ∣ λ + ∣ f k t = H K f k + i γ t 2 L ∣ λ + ∣ − 1 Λ f k + 2 k i γ t 2 L ∣ λ + ∣ − 1 f k − i γ t 2 L ∣ λ + ∣ − 1 ∑ l = 0 k − 1 ∑ j = 1 K H K l ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) + H K k ( F + r ) . (175) \mathrm{i}t^{\frac{2L}{|\lambda_{+}|}}f_{kt}=H_{K}f_{k}+\mathrm{i}\gamma t^{\frac{2L}{|\lambda_{+}|}-1}\Lambda f_{k}+2k\mathrm{i}\gamma t^{\frac{2L}{|\lambda_{+}|}-1}f_{k}
-\mathrm{i}\gamma t^{\frac{2L}{|\lambda_{+}|}-1}\sum_{l=0}^{k-1}\sum_{j=1}^{K}H_{K}^{l}\left(((y-y_{j})\cdot\nabla V_{j}+2V_{j})f_{k-l-1}\right)+H_{K}^{k}(F+r). \tag*{(175)} i t ∣ λ + ∣ 2 L f k t = H K f k + i γ t ∣ λ + ∣ 2 L − 1 Λ f k + 2 k i γ t ∣ λ + ∣ 2 L − 1 f k − i γ t ∣ λ + ∣ 2 L − 1 l = 0 ∑ k − 1 j = 1 ∑ K H K l ( (( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) + H K k ( F + r ) . (175)
Proof of Claim 7.3. We first consider the case k = 1 k=1 k = 1 . Using (134) , we deduce that f 1 = H K f f_1=H_Kf f 1 = H K f satisfies
i t 2 L ∣ λ + ∣ ∂ t f 1 = H K f 1 + i γ t 2 L ∣ λ + ∣ − 1 Λ f 1 + i γ t 2 L ∣ λ + ∣ − 1 [ H K , y ⋅ ∇ ] f + i t 2 L ∣ λ + ∣ V t f + H K ( F + r ) , \mathrm{i}t^{\frac{2L}{|\lambda_+|}}\partial_t f_1=H_Kf_1+\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}\Lambda f_1+\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}[H_K,y\cdot\nabla]f+\mathrm{i}t^{\frac{2L}{|\lambda_+|}}V_tf+H_K(\mathcal{F}+r), i t ∣ λ + ∣ 2 L ∂ t f 1 = H K f 1 + i γ t ∣ λ + ∣ 2 L − 1 Λ f 1 + i γ t ∣ λ + ∣ 2 L − 1 [ H K , y ⋅ ∇ ] f + i t ∣ λ + ∣ 2 L V t f + H K ( F + r ) ,
where [ A , B ] = A B − B A [A,B]=AB-BA [ A , B ] = A B − B A denotes the commutator of A A A and B B B . Using the identity
[ H K , y ⋅ ∇ ] f = 2 H K f − ( y ⋅ ∇ V + 2 V ) f , (176) [H_K,y\cdot\nabla]f=2H_Kf-(y\cdot\nabla V+2V)f, \tag*{(176)} [ H K , y ⋅ ∇ ] f = 2 H K f − ( y ⋅ ∇ V + 2 V ) f , (176)
and (151) , we rewrite the equation for f 1 f_1 f 1 as
i t 2 L ∣ λ + ∣ ∂ t f 1 = H K f 1 + i γ t 2 L ∣ λ + ∣ − 1 Λ f 1 + 2 i γ t 2 L ∣ λ + ∣ − 1 f 1 − i γ t 2 L ∣ λ + ∣ − 1 ∑ j = 1 K ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f + H K ( F + r ) . \mathrm{i}t^{\frac{2L}{|\lambda_+|}}\partial_t f_1=H_Kf_1+\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}\Lambda f_1+2\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}f_1-\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}\sum_{j=1}^{K}\bigl((y-y_j)\cdot\nabla V_j+2V_j\bigr)f+H_K(\mathcal{F}+r). i t ∣ λ + ∣ 2 L ∂ t f 1 = H K f 1 + i γ t ∣ λ + ∣ 2 L − 1 Λ f 1 + 2 i γ t ∣ λ + ∣ 2 L − 1 f 1 − i γ t ∣ λ + ∣ 2 L − 1 j = 1 ∑ K ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f + H K ( F + r ) .
We next consider the case k ≥ 2 k\ge2 k ≥ 2 . We proceed by induction and assume that, for every integer 1 ≤ k ~ ≤ k − 1 1\le\widetilde{k}\le k-1 1 ≤ k ≤ k − 1 , the function f k ~ f_{\widetilde{k}} f k satisfies (175) . Using the equation for f k − 1 f_{k-1} f k − 1 together with
i t 2 L ∣ λ + ∣ ∂ t f k = i t 2 L ∣ λ + ∣ ( V t f k − 1 + H K ( ∂ t f k − 1 ) ) , \mathrm{i}t^{\frac{2L}{|\lambda_+|}}\partial_t f_k=\mathrm{i}t^{\frac{2L}{|\lambda_+|}}\bigl(V_tf_{k-1}+H_K(\partial_t f_{k-1})\bigr), i t ∣ λ + ∣ 2 L ∂ t f k = i t ∣ λ + ∣ 2 L ( V t f k − 1 + H K ( ∂ t f k − 1 ) ) ,
we derive from (176)
i t 2 L ∣ λ + ∣ ∂ t f k = H K f k + i γ t 2 L ∣ λ + ∣ − 1 ( Λ f k + 2 f k − ∑ j = 1 K ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − 1 ) + 2 ( k − 1 ) i γ t 2 L ∣ λ + ∣ − 1 f k − i γ t 2 L ∣ λ + ∣ − 1 ∑ l = 0 k − 2 ∑ j = 1 K H K l + 1 ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 2 ) + H K k ( F + r ) . \begin{aligned}
\mathrm{i}t^{\frac{2L}{|\lambda_+|}}\partial_t f_k={}&H_Kf_k+\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}\left(\Lambda f_k+2f_k-\sum_{j=1}^{K}\bigl((y-y_j)\cdot\nabla V_j+2V_j\bigr)f_{k-1}\right)+2(k-1)\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}f_k\\
&-\mathrm{i}\gamma t^{\frac{2L}{|\lambda_+|}-1}\sum_{l=0}^{k-2}\sum_{j=1}^{K}H_K^{l+1}\left(\bigl((y-y_j)\cdot\nabla V_j+2V_j\bigr)f_{k-l-2}\right)+H_K^k(\mathcal{F}+r).
\end{aligned} i t ∣ λ + ∣ 2 L ∂ t f k = H K f k + i γ t ∣ λ + ∣ 2 L − 1 ( Λ f k + 2 f k − j = 1 ∑ K ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − 1 ) + 2 ( k − 1 ) i γ t ∣ λ + ∣ 2 L − 1 f k − i γ t ∣ λ + ∣ 2 L − 1 l = 0 ∑ k − 2 j = 1 ∑ K H K l + 1 ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 2 ) + H K k ( F + r ) .
Renaming the index l + 1 l+1 l + 1 as l l l concludes the proof of the claim.
Next, differentiating the functional J k ( t ) J_k(t) J k ( t ) and using (175) , we obtain
d d t J k ( t ) = γ t ( ⟨ H K Λ f k , σ 3 f k ⟩ + ⟨ H K f k , σ 3 Λ f k ⟩ ) + 4 k γ t ⟨ H K f k , σ 3 f k ⟩ + γ t ∑ j = 1 K ⟨ y j ⋅ ∇ V j f k , σ 3 f k ⟩ − 2 t 2 L ∣ λ + ∣ Im ⟨ H K f k , σ 3 H K k ( F + r ) ⟩ − 2 γ t ∑ l = 0 k − 1 ∑ j = 1 K Re ⟨ H K l + 1 ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) , σ 3 f k ⟩ + ( − 1 ) k d d t ( t 2 ν Re ⟨ f k , σ 3 B ( f k ) ⟩ ) . (177) \begin{aligned}
\frac{\mathrm{d}}{\mathrm{d}t}J_k(t)={}&\frac{\gamma}{t}\left(\langle H_K\Lambda f_k,\sigma_3f_k\rangle+\langle H_Kf_k,\sigma_3\Lambda f_k\rangle\right)+\frac{4k\gamma}{t}\langle H_Kf_k,\sigma_3f_k\rangle\\
&+\frac{\gamma}{t}\sum_{j=1}^{K}\langle y_j\cdot\nabla V_jf_k,\sigma_3f_k\rangle-2t^{\frac{2L}{|\lambda_+|}}\operatorname{Im}\langle H_Kf_k,\sigma_3H_K^k(\mathcal{F}+r)\rangle\\
&-\frac{2\gamma}{t}\sum_{l=0}^{k-1}\sum_{j=1}^{K}\operatorname{Re}\left\langle H_K^{l+1}\left(\bigl((y-y_j)\cdot\nabla V_j+2V_j\bigr)f_{k-l-1}\right),\sigma_3f_k\right\rangle\\
&+(-1)^k\frac{\mathrm{d}}{\mathrm{d}t}\left(t^{2\nu}\operatorname{Re}\langle f_k,\sigma_3\mathcal{B}(f_k)\rangle\right). \tag*{(177)}
\end{aligned} d t d J k ( t ) = t γ ( ⟨ H K Λ f k , σ 3 f k ⟩ + ⟨ H K f k , σ 3 Λ f k ⟩ ) + t 4 k γ ⟨ H K f k , σ 3 f k ⟩ + t γ j = 1 ∑ K ⟨ y j ⋅ ∇ V j f k , σ 3 f k ⟩ − 2 t ∣ λ + ∣ 2 L Im ⟨ H K f k , σ 3 H K k ( F + r )⟩ − t 2 γ l = 0 ∑ k − 1 j = 1 ∑ K Re ⟨ H K l + 1 ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) , σ 3 f k ⟩ + ( − 1 ) k d t d ( t 2 ν Re ⟨ f k , σ 3 B ( f k )⟩ ) . (177)
To simplify the scaling contribution, integration by parts gives
⟨ H K Λ f k , σ 3 f k ⟩ + ⟨ H K f k , σ 3 Λ f k ⟩ = − 2 ( s c − 1 ) ⟨ H K f k , σ 3 f k ⟩ − ⟨ ( y ⋅ ∇ V + 2 V ) f k , σ 3 f k ⟩ , \langle H_K\Lambda f_k,\sigma_3f_k\rangle+\langle H_Kf_k,\sigma_3\Lambda f_k\rangle=-2(s_c-1)\langle H_Kf_k,\sigma_3f_k\rangle-\langle(y\cdot\nabla V+2V)f_k,\sigma_3f_k\rangle, ⟨ H K Λ f k , σ 3 f k ⟩ + ⟨ H K f k , σ 3 Λ f k ⟩ = − 2 ( s c − 1 ) ⟨ H K f k , σ 3 f k ⟩ − ⟨( y ⋅ ∇ V + 2 V ) f k , σ 3 f k ⟩ ,
where s c = − 1 p + d 2 s_c=-\frac{1}{p}+\frac{d}{2} s c = − p 1 + 2 d . We separate the commutator contribution using
H K l + 1 ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) = [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 + ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k . H_K^{l+1}\left(\bigl((y-y_j)\cdot\nabla V_j+2V_j\bigr)f_{k-l-1}\right)
=\left[H_K^{l+1},(y-y_j)\cdot\nabla V_j+2V_j\right]f_{k-l-1}+\bigl((y-y_j)\cdot\nabla V_j+2V_j\bigr)f_k. H K l + 1 ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) = [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 + ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k .
Finally, as in the H ˙ 1 \dot H^1 H ˙ 1 energy estimate, using the self-adjointness of σ 3 B \sigma_3\mathcal{B} σ 3 B together with (151) and the identity 2 L ∣ λ + ∣ − 1 = 2 ν \frac{2L}{|\lambda_+|}-1=2\nu ∣ λ + ∣ 2 L − 1 = 2 ν gives
d d t ( t 2 ν Re ⟨ f k , σ 3 B ( f k ) ⟩ ) = t 2 ν ( 2 ν t Re ⟨ f k , σ 3 B ( f k ) ⟩ + 2 Re ⟨ ∂ t f k , σ 3 B ( f k ) ⟩ − 2 p γ t Re ⟨ f k , σ 3 ∑ j = 1 K ( y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) ) σ 1 f k ⟩ ) . \frac{\mathrm{d}}{\mathrm{d}t}\left(t^{2\nu}\operatorname{Re}\langle f_k,\sigma_3\mathcal{B}(f_k)\rangle\right)
=t^{2\nu}\left(\frac{2\nu}{t}\operatorname{Re}\langle f_k,\sigma_3\mathcal{B}(f_k)\rangle+2\operatorname{Re}\langle\partial_t f_k,\sigma_3\mathcal{B}(f_k)\rangle-\frac{2p\gamma}{t}\operatorname{Re}\left\langle f_k,\sigma_3\sum_{j=1}^{K}(y_j\cdot\nabla(Q_j^{2p-1}\chi_{1,j}))\sigma_1f_k\right\rangle\right). d t d ( t 2 ν Re ⟨ f k , σ 3 B ( f k )⟩ ) = t 2 ν ( t 2 ν Re ⟨ f k , σ 3 B ( f k )⟩ + 2 Re ⟨ ∂ t f k , σ 3 B ( f k )⟩ − t 2 p γ Re ⟨ f k , σ 3 j = 1 ∑ K ( y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j )) σ 1 f k ⟩ ) .
Substituting these identities into (177) and using (175) , we obtain the following decomposition:
d d t J k ( t ) = E k , 1 + E k , 2 + E k , 3 + E k , 4 , (178) \frac{\mathrm{d}}{\mathrm{d}t}J_k(t)=\mathcal{E}_{k,1}+\mathcal{E}_{k,2}+\mathcal{E}_{k,3}+\mathcal{E}_{k,4}, \tag*{(178)} d t d J k ( t ) = E k , 1 + E k , 2 + E k , 3 + E k , 4 , (178)
where
E k , 1 = 2 γ ( 2 k + 1 − s c ) t ⟨ H K f k , σ 3 f k ⟩ , \mathcal{E}_{k,1}=\frac{2\gamma(2k+1-s_c)}{t}\langle H_Kf_k,\sigma_3f_k\rangle, E k , 1 = t 2 γ ( 2 k + 1 − s c ) ⟨ H K f k , σ 3 f k ⟩ ,
E k , 2 = − ( 2 k + 1 ) γ t ∑ j = 1 K ⟨ ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k , σ 3 f k ⟩ , \mathcal{E}_{k,2}=-\frac{(2k+1)\gamma}{t}\sum_{j=1}^{K}\left\langle((y-y_j)\cdot\nabla V_j+2V_j)f_k,\sigma_3f_k\right\rangle, E k , 2 = − t ( 2 k + 1 ) γ j = 1 ∑ K ⟨ (( y − y j ) ⋅ ∇ V j + 2 V j ) f k , σ 3 f k ⟩ ,
E k , 3 = − 2 γ t ∑ l = 0 k − 1 ∑ j = 1 K Re ⟨ [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 , σ 3 f k ⟩ − 2 t − 1 − 2 ν Im ⟨ H K f k , σ 3 ( H K k ( F + r ) − ( − 1 ) k t 2 ν B ( f k ) ) ⟩ − 2 ( − 1 ) k γ t 2 ν − 1 ∑ l = 0 k − 1 ∑ j = 1 K Re ⟨ H K l ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 , σ 3 B ( f k ) ⟩ + 2 ( − 1 ) k t Im ⟨ H K k ( F + r ) , σ 3 B ( f k ) ⟩ , \begin{aligned}
\mathcal{E}_{k,3}={}&-\frac{2\gamma}{t}\sum_{l=0}^{k-1}\sum_{j=1}^{K}\operatorname{Re}\left\langle[H_K^{l+1},(y-y_j)\cdot\nabla V_j+2V_j]f_{k-l-1},\sigma_3f_k\right\rangle\\
&-2t^{-1-2\nu}\operatorname{Im}\left\langle H_Kf_k,\sigma_3(H_K^k(\mathcal{F}+r)-(-1)^kt^{2\nu}\mathcal{B}(f_k))\right\rangle\\
&-2(-1)^k\gamma t^{2\nu-1}\sum_{l=0}^{k-1}\sum_{j=1}^{K}\operatorname{Re}\left\langle H_K^l\left((y-y_j)\cdot\nabla V_j+2V_j\right)f_{k-l-1},\sigma_3\mathcal{B}(f_k)\right\rangle\\
&+\frac{2(-1)^k}{t}\operatorname{Im}\left\langle H_K^k(\mathcal{F}+r),\sigma_3\mathcal{B}(f_k)\right\rangle,
\end{aligned} E k , 3 = − t 2 γ l = 0 ∑ k − 1 j = 1 ∑ K Re ⟨ [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 , σ 3 f k ⟩ − 2 t − 1 − 2 ν Im ⟨ H K f k , σ 3 ( H K k ( F + r ) − ( − 1 ) k t 2 ν B ( f k )) ⟩ − 2 ( − 1 ) k γ t 2 ν − 1 l = 0 ∑ k − 1 j = 1 ∑ K Re ⟨ H K l ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 , σ 3 B ( f k ) ⟩ + t 2 ( − 1 ) k Im ⟨ H K k ( F + r ) , σ 3 B ( f k ) ⟩ ,
and
E k , 4 = 2 γ ( − 1 ) k t 2 ν − 1 ( Re ⟨ ( 2 k + ν γ ) f k + Λ f k , σ 3 B ( f k ) ⟩ − p Re ⟨ f k , σ 3 ∑ j = 1 K ( y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) ) σ 1 f k ⟩ ) . \mathcal{E}_{k,4}=2\gamma(-1)^kt^{2\nu-1}\left(\operatorname{Re}\left\langle\left(2k+\frac{\nu}{\gamma}\right)f_k+\Lambda f_k,\sigma_3\mathcal{B}(f_k)\right\rangle-p\operatorname{Re}\left\langle f_k,\sigma_3\sum_{j=1}^{K}(y_j\cdot\nabla(Q_j^{2p-1}\chi_{1,j}))\sigma_1f_k\right\rangle\right). E k , 4 = 2 γ ( − 1 ) k t 2 ν − 1 ( Re ⟨ ( 2 k + γ ν ) f k + Λ f k , σ 3 B ( f k ) ⟩ − p Re ⟨ f k , σ 3 j = 1 ∑ K ( y j ⋅ ∇ ( Q j 2 p − 1 χ 1 , j )) σ 1 f k ⟩ ) .
We first treat E k , 2 \mathcal{E}_{k,2} E k , 2 . By the pointwise bounds on Q Q Q and Λ Q \Lambda Q Λ Q , Hardy’s inequality, and the multibubble coercivity estimate (26) , we obtain
E k , 2 ≤ 2 ( 2 p + 1 ) ( 2 k + 1 ) γ c ∞ 2 p K t ( 2 d − 2 ) 2 ∥ ∇ f k ∥ L 2 2 ≤ ( 2 k + 1 ) A 1 γ t ( 1 + C t 2 ν ) J k ( t ) , (179) \mathcal{E}_{k,2}\leq\frac{2(2p+1)(2k+1)\gamma c_{\infty}^{2p}K}{t}\left(\frac{2}{d-2}\right)^2\|\nabla f_k\|_{L^2}^2\leq\frac{(2k+1)A_1\gamma}{t}(1+Ct^{2\nu})J_k(t), \tag*{(179)} E k , 2 ≤ t 2 ( 2 p + 1 ) ( 2 k + 1 ) γ c ∞ 2 p K ( d − 2 2 ) 2 ∥∇ f k ∥ L 2 2 ≤ t ( 2 k + 1 ) A 1 γ ( 1 + C t 2 ν ) J k ( t ) , (179)
where A 1 A_1 A 1 is the constant defined in (155) .
We next estimate the commutator contribution in E k , 3 \mathcal{E}_{k,3} E k , 3 . Using the identity
[ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] = ∑ q = 0 l H K q [ H K , ( y − y j ) ⋅ ∇ V j + 2 V j ] H K l − q . [H_K^{l+1},(y-y_j)\cdot\nabla V_j+2V_j]=\sum_{q=0}^{l}H_K^q[H_K,(y-y_j)\cdot\nabla V_j+2V_j]H_K^{l-q}. [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] = q = 0 ∑ l H K q [ H K , ( y − y j ) ⋅ ∇ V j + 2 V j ] H K l − q .
For any test function g g g , since each V j V_j V j is Q j 2 p Q_j^{2p} Q j 2 p times a constant matrix independent of j j j and σ 1 σ 3 = − σ 3 σ 1 \sigma_1\sigma_3=-\sigma_3\sigma_1 σ 1 σ 3 = − σ 3 σ 1 , we have
[ H K , ( y − y j ) ⋅ ∇ V j + 2 V j ] g = − σ 3 Δ ( ( y − y j ) ⋅ ∇ V j + 2 V j ) g − 2 σ 3 ∇ ( ( y − y j ) ⋅ ∇ V j + 2 V j ) ⋅ ∇ g + 2 ( p ( y − y j ) ⋅ ∇ Q j 2 p + 2 p Q j 2 p ) σ 1 Δ g . \left[H_K,(y-y_j)\cdot\nabla V_j+2V_j\right]g
=-\sigma_3\Delta\left((y-y_j)\cdot\nabla V_j+2V_j\right)g-2\sigma_3\nabla\left((y-y_j)\cdot\nabla V_j+2V_j\right)\cdot\nabla g
+2\left(p(y-y_j)\cdot\nabla Q_j^{2p}+2pQ_j^{2p}\right)\sigma_1\Delta g. [ H K , ( y − y j ) ⋅ ∇ V j + 2 V j ] g = − σ 3 Δ ( ( y − y j ) ⋅ ∇ V j + 2 V j ) g − 2 σ 3 ∇ ( ( y − y j ) ⋅ ∇ V j + 2 V j ) ⋅ ∇ g + 2 ( p ( y − y j ) ⋅ ∇ Q j 2 p + 2 p Q j 2 p ) σ 1 Δ g .
Expanding the powers of H K H_K H K in the above identity, we obtain
[ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 = p ( 1 + ( − 1 ) l ) ( ( y − y j ) ⋅ ∇ Q j 2 p + 2 Q j 2 p ) σ 3 σ 1 f k + ∑ m = 0 2 k − 1 B l , j , m ( t , y ) ∇ m f + R l , j ( t , y ) . (180) \left[H_K^{l+1},(y-y_j)\cdot\nabla V_j+2V_j\right]f_{k-l-1}
=p\left(1+(-1)^l\right)\left((y-y_j)\cdot\nabla Q_j^{2p}+2Q_j^{2p}\right)\sigma_3\sigma_1f_k+\sum_{m=0}^{2k-1}B_{l,j,m}(t,y)\nabla^m f+R_{l,j}(t,y). \tag*{(180)} [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 = p ( 1 + ( − 1 ) l ) ( ( y − y j ) ⋅ ∇ Q j 2 p + 2 Q j 2 p ) σ 3 σ 1 f k + m = 0 ∑ 2 k − 1 B l , j , m ( t , y ) ∇ m f + R l , j ( t , y ) . (180)
The coefficients involving only potentials centered at y j y_j y j satisfy
∣ B l , j , m ( t , y ) ∣ ≤ C ⟨ y − y j ⟩ − ( 2 k + 2 − m ) . (181) \left|B_{l,j,m}(t,y)\right|\le C\langle y-y_j\rangle^{-(2k+2-m)}. \tag*{(181)} ∣ B l , j , m ( t , y ) ∣ ≤ C ⟨ y − y j ⟩ − ( 2 k + 2 − m ) . (181)
The remainder R l , j R_{l,j} R l , j denotes the sum of cross terms
( ∂ α 0 ( ( y − y j ) ⋅ ∇ V j + 2 V j ) ) ( ∂ α 1 V i 1 ) ⋯ ( ∂ α s V i s ) ∇ m f , (182) \left(\partial^{\alpha_0}\left((y-y_j)\cdot\nabla V_j+2V_j\right)\right)\left(\partial^{\alpha_1}V_{i_1}\right)\cdots\left(\partial^{\alpha_s}V_{i_s}\right)\nabla^m f, \tag*{(182)} ( ∂ α 0 ( ( y − y j ) ⋅ ∇ V j + 2 V j ) ) ( ∂ α 1 V i 1 ) ⋯ ( ∂ α s V i s ) ∇ m f , (182)
where 1 ≤ s ≤ k 1\le s\le k 1 ≤ s ≤ k , 1 ≤ i μ ≤ K 1\le i_\mu\le K 1 ≤ i μ ≤ K and 1 ≤ μ ≤ s 1\le\mu\le s 1 ≤ μ ≤ s , at least one i μ ≠ j i_\mu\ne j i μ = j , and
0 ≤ m ≤ 2 k − 2 , ∣ α 0 ∣ + ⋯ + ∣ α s ∣ + 2 s + m = 2 k . 0\le m\le2k-2,\qquad|\alpha_0|+\cdots+|\alpha_s|+2s+m=2k. 0 ≤ m ≤ 2 k − 2 , ∣ α 0 ∣ + ⋯ + ∣ α s ∣ + 2 s + m = 2 k .
We first estimate the terms involving B l , j , m B_{l,j,m} B l , j , m . Using (181) and the Cauchy–Schwarz inequality
∣ Re ⟨ B l , j , m ∇ m f , σ 3 f k ⟩ ∣ ≤ C k ( ∫ ∣ ∇ m f ∣ 2 ⟨ y − y j ⟩ 2 ( 2 k + 1 − m ) d y ) 1 / 2 ( ∫ ∣ f k ∣ 2 ⟨ y − y j ⟩ 2 d y ) 1 / 2 . (183) \left|\operatorname{Re}\langle B_{l,j,m}\nabla^m f,\sigma_3f_k\rangle\right|
\le C_k\left(\int\frac{|\nabla^m f|^2}{\langle y-y_j\rangle^{2(2k+1-m)}}\,dy\right)^{1/2}\left(\int\frac{|f_k|^2}{\langle y-y_j\rangle^2}\,dy\right)^{1/2}. \tag*{(183)} ∣ Re ⟨ B l , j , m ∇ m f , σ 3 f k ⟩ ∣ ≤ C k ( ∫ ⟨ y − y j ⟩ 2 ( 2 k + 1 − m ) ∣ ∇ m f ∣ 2 d y ) 1/2 ( ∫ ⟨ y − y j ⟩ 2 ∣ f k ∣ 2 d y ) 1/2 . (183)
Hardy’s inequality, Lemma 2.4 and (172) imply
∫ R d ∣ f k ∣ 2 ⟨ y − y j ⟩ 2 d y ≤ 4 ( 1 + C t 2 ν ) ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p J k ( t ) . (184) \int_{\mathbb{R}^d}\frac{|f_k|^2}{\langle y-y_j\rangle^2}\,dy\le\frac{4(1+Ct^{2\nu})}{(d-2)^2-4K(2p+1)c_\infty^{2p}}J_k(t). \tag*{(184)} ∫ R d ⟨ y − y j ⟩ 2 ∣ f k ∣ 2 d y ≤ ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 4 ( 1 + C t 2 ν ) J k ( t ) . (184)
For the other term in (183) , use the Hardy inequality for m = 0 m=0 m = 0 , and boundedness for m ≥ 1 m\ge1 m ≥ 1 . Hence
∑ m = 0 2 k − 1 ∥ ⟨ y − y j ⟩ − ( 2 k + 1 − m ) ∇ m f ∥ L 2 ≤ C ∥ ∇ f ∥ H 2 k − 2 . (185) \sum_{m=0}^{2k-1}\left\|\langle y-y_j\rangle^{-(2k+1-m)}\nabla^m f\right\|_{L^2}\le C\|\nabla f\|_{H^{2k-2}}. \tag*{(185)} m = 0 ∑ 2 k − 1 ⟨ y − y j ⟩ − ( 2 k + 1 − m ) ∇ m f L 2 ≤ C ∥∇ f ∥ H 2 k − 2 . (185)
By interpolation and (174) ,
∥ ∇ f ∥ H 2 k − 2 ≤ C ∥ ∇ f ∥ L 2 1 / k ∥ ∇ f ∥ H 2 k 1 − 1 / k ≤ C J 0 1 2 k ( J k + J 0 ) k − 1 2 k . \|\nabla f\|_{H^{2k-2}}\le C\|\nabla f\|_{L^2}^{1/k}\|\nabla f\|_{H^{2k}}^{1-1/k}\le CJ_0^{\frac{1}{2k}}(J_k+J_0)^{\frac{k-1}{2k}}. ∥∇ f ∥ H 2 k − 2 ≤ C ∥∇ f ∥ L 2 1/ k ∥∇ f ∥ H 2 k 1 − 1/ k ≤ C J 0 2 k 1 ( J k + J 0 ) 2 k k − 1 .
Combining this bound with (183) , (184) , and (185) , and applying Young’s inequality with exponents 2 k / ( 2 k − 1 ) 2k/(2k-1) 2 k / ( 2 k − 1 ) and 2 k 2k 2 k , we obtain
∣ ∑ j = 1 K ∑ m = 0 2 k − 1 Re ⟨ B l , j , m ∇ m f , σ 3 f k ⟩ ∣ ≤ C ( J k + J 0 ) 1 − 1 2 k J 0 1 2 k ≤ η 8 k ( 1 + γ ) J k + C J 0 . (186) \left|\sum_{j=1}^K\sum_{m=0}^{2k-1}\operatorname{Re}\langle B_{l,j,m}\nabla^m f,\sigma_3f_k\rangle\right|
\le C(J_k+J_0)^{1-\frac{1}{2k}}J_0^{\frac{1}{2k}}\le\frac{\eta}{8k(1+\gamma)}J_k+CJ_0. \tag*{(186)} j = 1 ∑ K m = 0 ∑ 2 k − 1 Re ⟨ B l , j , m ∇ m f , σ 3 f k ⟩ ≤ C ( J k + J 0 ) 1 − 2 k 1 J 0 2 k 1 ≤ 8 k ( 1 + γ ) η J k + C J 0 . (186)
We next estimate the remainder R l , j R_{l,j} R l , j . The separation condition (133) gives
min i ≠ j ∣ y i − y j ∣ = λ ( t ) min i ≠ j ∣ x i − x j ∣ > 10 δ λ ( t ) , λ ( t ) = t − 1 / 2 − ν . \min_{i\ne j}|y_i-y_j|=\lambda(t)\min_{i\ne j}|x_i-x_j|>10\delta\lambda(t),\qquad\lambda(t)=t^{-1/2-\nu}. i = j min ∣ y i − y j ∣ = λ ( t ) i = j min ∣ x i − x j ∣ > 10 δ λ ( t ) , λ ( t ) = t − 1/2 − ν .
For each y y y , choose a center y i y_i y i closest to y y y among y j , y i 1 , … , y i s y_j,y_{i_1},\ldots,y_{i_s} y j , y i 1 , … , y i s . Every other distinct center in this set is then at distance at least 5 δ λ ( t ) 5\delta\lambda(t) 5 δ λ ( t ) from y y y . Since at least two distinct centers occur, we extract λ − 2 ( t ) = t 1 + 2 ν \lambda^{-2}(t)=t^{1+2\nu} λ − 2 ( t ) = t 1 + 2 ν from a factor centered at one of these other points. The remaining decay is bounded by ⟨ y − y i ⟩ − ( 2 k − m ) \langle y-y_i\rangle^{-(2k-m)} ⟨ y − y i ⟩ − ( 2 k − m ) . Hence the coefficients in (182) satisfy
∣ ( ∂ α 0 ( ( y − y j ) ⋅ ∇ V j + 2 V j ) ) ( ∂ α 1 V i 1 ) ⋯ ( ∂ α s V i s ) ∣ ≤ C t 1 + 2 ν ∑ i = 1 K ⟨ y − y i ⟩ − ( 2 k − m ) . \left|\left(\partial^{\alpha^0}\left((y-y_j)\cdot\nabla V_j+2V_j\right)\right)\left(\partial^{\alpha^1}V_{i_1}\right)\cdots\left(\partial^{\alpha^s}V_{i_s}\right)\right|\leq Ct^{1+2\nu}\sum_{i=1}^{K}\langle y-y_i\rangle^{-(2k-m)}. ( ∂ α 0 ( ( y − y j ) ⋅ ∇ V j + 2 V j ) ) ( ∂ α 1 V i 1 ) ⋯ ( ∂ α s V i s ) ≤ C t 1 + 2 ν i = 1 ∑ K ⟨ y − y i ⟩ − ( 2 k − m ) .
The same argument as above gives
∣ ∑ j = 1 K Re ⟨ R l , j , σ 3 f k ⟩ ∣ ≤ C t 1 + 2 ν ∥ ∇ f k ∥ L 2 ∥ ∇ f ∥ H 2 k − 2 ≤ η 8 k ( 1 + γ ) J k + C J 0 . (187) \left|\sum_{j=1}^{K}\operatorname{Re}\langle\mathcal{R}_{l,j},\sigma_3f_k\rangle\right|\leq Ct^{1+2\nu}\|\nabla f_k\|_{L^2}\|\nabla f\|_{H^{2k-2}}\leq\frac{\eta}{8k(1+\gamma)}J_k+CJ_0. \tag*{(187)} j = 1 ∑ K Re ⟨ R l , j , σ 3 f k ⟩ ≤ C t 1 + 2 ν ∥∇ f k ∥ L 2 ∥∇ f ∥ H 2 k − 2 ≤ 8 k ( 1 + γ ) η J k + C J 0 . (187)
Finally, the leading term in (180) satisfies
∣ ∑ j = 1 K Re ⟨ ( ( y − y j ) ⋅ ∇ Q j 2 p + 2 Q j 2 p ) σ 3 σ 1 f k , σ 3 f k ⟩ ∣ ≤ 8 K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p ⟨ H K f k , σ 3 f k ⟩ . \left|\sum_{j=1}^{K}\operatorname{Re}\left\langle\left((y-y_j)\cdot\nabla Q_j^{2p}+2Q_j^{2p}\right)\sigma_3\sigma_1f_k,\sigma_3f_k\right\rangle\right|\leq\frac{8Kc_\infty^{2p}}{(d-2)^2-4K(2p+1)c_\infty^{2p}}\langle H_Kf_k,\sigma_3f_k\rangle. j = 1 ∑ K Re ⟨ ( ( y − y j ) ⋅ ∇ Q j 2 p + 2 Q j 2 p ) σ 3 σ 1 f k , σ 3 f k ⟩ ≤ ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 K c ∞ 2 p ⟨ H K f k , σ 3 f k ⟩ .
Combining this bound with (186) and (187) , we obtain
∣ ∑ j = 1 K Re ⟨ [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 , σ 3 f k ⟩ ∣ ≤ 8 p ( 1 + ( − 1 ) l ) K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p ⟨ H K f k , σ 3 f k ⟩ + η 4 k ( 1 + γ ) J k + C J 0 . (188) \begin{aligned}
\left|\sum_{j=1}^{K}\operatorname{Re}\left\langle\left[H_K^{l+1},(y-y_j)\cdot\nabla V_j+2V_j\right]f_{k-l-1},\sigma_3f_k\right\rangle\right|\\
\leq\frac{8p(1+(-1)^l)Kc_\infty^{2p}}{(d-2)^2-4K(2p+1)c_\infty^{2p}}\langle H_Kf_k,\sigma_3f_k\rangle+\frac{\eta}{4k(1+\gamma)}J_k+CJ_0. \tag*{(188)}
\end{aligned} j = 1 ∑ K Re ⟨ [ H K l + 1 , ( y − y j ) ⋅ ∇ V j + 2 V j ] f k − l − 1 , σ 3 f k ⟩ ≤ ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 p ( 1 + ( − 1 ) l ) K c ∞ 2 p ⟨ H K f k , σ 3 f k ⟩ + 4 k ( 1 + γ ) η J k + C J 0 . (188)
We next consider the second term of E k , 3 \mathcal{E}_{k,3} E k , 3 . Since σ 3 B = − B σ 3 \sigma_3\mathcal{B}=-\mathcal{B}\sigma_3 σ 3 B = − B σ 3 , this shows that H K k B ( f ) − ( − 1 ) k B ( f k ) H_K^k\mathcal{B}(f)-(-1)^k\mathcal{B}(f_k) H K k B ( f ) − ( − 1 ) k B ( f k ) contains derivatives of f f f of order at most 2 k − 1 2k-1 2 k − 1 . Using (161) , integrating by parts and interpolating as above, we obtain
∣ Im ⟨ H K f k , σ 3 ( H K k B ( f ) − ( − 1 ) k B ( f k ) ) ⟩ ∣ ≤ C ∥ ∇ f k ∥ L 2 ∥ ∇ f ∥ H 2 k − 1 ≤ η 16 J k + C J 0 . (189) \left|\operatorname{Im}\left\langle H_Kf_k,\sigma_3\left(H_K^k\mathcal{B}(f)-(-1)^k\mathcal{B}(f_k)\right)\right\rangle\right|\leq C\|\nabla f_k\|_{L^2}\|\nabla f\|_{H^{2k-1}}\leq\frac{\eta}{16}J_k+CJ_0. \tag*{(189)} Im ⟨ H K f k , σ 3 ( H K k B ( f ) − ( − 1 ) k B ( f k ) ) ⟩ ≤ C ∥∇ f k ∥ L 2 ∥∇ f ∥ H 2 k − 1 ≤ 16 η J k + C J 0 . (189)
As in the H ˙ 1 \dot H^1 H ˙ 1 energy estimate, the separation of the supports of U j α p U_j^{\alpha p} U j α p , together with (158) , (159) , and the Leibniz rule applied to ∇ H K k \nabla H_K^k ∇ H K k , gives
∥ ∇ H K k F ~ 1 ( f ) ∥ L 2 ≤ C t 2 ν + 8 n ( ∑ m = 1 2 k + 1 ∥ ∇ m f ∥ L 2 + ∑ j = 1 K ∥ ⟨ y − y j ⟩ − 1 f ∥ L 2 ) ≤ C t 2 ν + 8 n ∥ ∇ f ∥ H 2 k . (190) \begin{aligned}
\|\nabla H_K^k\widetilde{F}_1(f)\|_{L^2}&\leq Ct^{2\nu+8n}\left(\sum_{m=1}^{2k+1}\|\nabla^m f\|_{L^2}+\sum_{j=1}^{K}\|\langle y-y_j\rangle^{-1}f\|_{L^2}\right)\\
&\leq Ct^{2\nu+8n}\|\nabla f\|_{H^{2k}}. \tag*{(190)}
\end{aligned} ∥∇ H K k F 1 ( f ) ∥ L 2 ≤ C t 2 ν + 8 n ( m = 1 ∑ 2 k + 1 ∥ ∇ m f ∥ L 2 + j = 1 ∑ K ∥ ⟨ y − y j ⟩ − 1 f ∥ L 2 ) ≤ C t 2 ν + 8 n ∥∇ f ∥ H 2 k . (190)
where the last inequality follows from Hardy’s inequality. Moreover, Lemma 7.2 yields
∥ ∇ H K k F 2 ( f ) ∥ L 2 + ∥ ∇ H K k r ∥ L 2 ≤ C t ν ( 1 − 1 q ) ( ∣ λ ∣ − ∣ s c + 3 ) , (191) \|\nabla H_K^k\mathcal{F}_2(f)\|_{L^2}+\|\nabla H_K^k r\|_{L^2}\leq Ct^{\nu(1-\frac{1}{q})(|\lambda|-|s_c+3)}, \tag*{(191)} ∥∇ H K k F 2 ( f ) ∥ L 2 + ∥∇ H K k r ∥ L 2 ≤ C t ν ( 1 − q 1 ) ( ∣ λ ∣ − ∣ s c + 3 ) , (191)
where the bound follows from T ∗ ≤ 1 T^*\leq1 T ∗ ≤ 1 and ( ν − ϵ 1 ) ( ∣ λ ∣ − ∣ s c + 3 ) > 4 ν + 2 η (\nu-\epsilon_1)(|\lambda|-|s_c+3)>4\nu+2\eta ( ν − ϵ 1 ) ( ∣ λ ∣ − ∣ s c + 3 ) > 4 ν + 2 η .
Applying Young’s inequality to (189) –(191) and then decreasing T T T , we obtain
2 t − 2 λ ∣ λ ∣ + 1 ∣ Im ⟨ H K f k , σ 3 ( H K k ( F + r ) − ( − 1 ) k t 2 B ( f k ) ) ⟩ ∣ ≤ η 2 t J k + C t J 0 + C t 4 ν − 1 + 2 ν ( 1 − 1 q ) ( ∣ λ ∣ − ∣ s c + 3 ) . 2t^{-\frac{2\lambda}{|\lambda|+1}}\left|\operatorname{Im}\left\langle H_Kf_k,\sigma_3\left(H_K^k(\mathcal{F}+r)-(-1)^kt^2\mathcal{B}(f_k)\right)\right\rangle\right|
\leq\frac{\eta}{2t}J_k+\frac{C}{t}J_0+Ct^{4\nu-1+2\nu(1-\frac{1}{q})(|\lambda|-|s_c+3)}. 2 t − ∣ λ ∣ + 1 2 λ Im ⟨ H K f k , σ 3 ( H K k ( F + r ) − ( − 1 ) k t 2 B ( f k ) ) ⟩ ≤ 2 t η J k + t C J 0 + C t 4 ν − 1 + 2 ν ( 1 − q 1 ) ( ∣ λ ∣ − ∣ s c + 3 ) .
We next estimate the last two terms in E k , 3 \mathcal{E}_{k,3} E k , 3 . By (161) , Hardy’s inequality, and the bounded derivatives of the potentials, we have
∥ ∇ H K k B ( f ) ∥ L 2 ≤ C ∥ ∇ f ∥ H 2 k , \|\nabla H_K^k\mathcal{B}(f)\|_{L^2}\leq C\|\nabla f\|_{H^{2k}}, ∥∇ H K k B ( f ) ∥ L 2 ≤ C ∥∇ f ∥ H 2 k ,
and
∑ l = 0 k − 1 ∑ j = 1 K ∥ ∇ H K l ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) ∥ L 2 ≤ C ∥ ∇ f ∥ H 2 k − 2 . \sum_{l=0}^{k-1}\sum_{j=1}^{K}\left\|\nabla H_K^l\left(((y-y_j)\cdot\nabla V_j+2V_j)f_{k-l-1}\right)\right\|_{L^2}\leq C\|\nabla f\|_{H^{2k-2}}. l = 0 ∑ k − 1 j = 1 ∑ K ∇ H K l ( (( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) L 2 ≤ C ∥∇ f ∥ H 2 k − 2 .
Together with the estimates (174) , (190) and (191) , we conclude
2 γ t 2 ν − 1 ∣ ∑ l = 0 k − 1 ∑ j = 1 K Re ⟨ H K l ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) , σ 3 B ( f k ) ⟩ ∣ + 2 t ∣ Im ⟨ H K k ( F ( f ) + r ) , σ 3 B ( f k ) ⟩ ∣ ≤ C t 2 ν − 1 ( J k + J 0 ) + C t − 1 + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) J k 1 / 2 . \begin{aligned}
2\gamma t^{2\nu-1}\left|\sum_{l=0}^{k-1}\sum_{j=1}^{K}\operatorname{Re}\left\langle H_K^l\left(((y-y_j)\cdot\nabla V_j+2V_j)f_{k-l-1}\right),\sigma_3\mathcal{B}(f_k)\right\rangle\right| \\
&+\frac{2}{t}\left|\operatorname{Im}\left\langle H_K^k(\mathcal{F}(f)+r),\sigma_3\mathcal{B}(f_k)\right\rangle\right|
\leq Ct^{2\nu-1}(J_k+J_0)+Ct^{-1+\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)}J_k^{1/2}.
\end{aligned} 2 γ t 2 ν − 1 l = 0 ∑ k − 1 j = 1 ∑ K Re ⟨ H K l ( (( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) , σ 3 B ( f k ) ⟩ + t 2 Im ⟨ H K k ( F ( f ) + r ) , σ 3 B ( f k ) ⟩ ≤ C t 2 ν − 1 ( J k + J 0 ) + C t − 1 + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) J k 1/2 .
Applying Young’s inequality to the second term and choosing T ∗ T^* T ∗ such that C ( T ∗ ) 2 ν ≤ η 2 C(T^*)^{2\nu}\leq\frac{\eta}{2} C ( T ∗ ) 2 ν ≤ 2 η , we obtain
2 γ t 2 ν − 1 ∣ ∑ l = 0 k − 1 ∑ j = 1 K Re ⟨ H K l ( ( ( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) , σ 3 B ( f k ) ⟩ ∣ + 2 t ∣ Im ⟨ H K k ( F ( f ) + r ) , σ 3 B ( f k ) ⟩ ∣ ≤ η t J k + C t J 0 + C t − 4 ν − 1 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) . \begin{aligned}
2\gamma t^{2\nu-1}\left|\sum_{l=0}^{k-1}\sum_{j=1}^{K}\operatorname{Re}\left\langle H_K^l\left(((y-y_j)\cdot\nabla V_j+2V_j)f_{k-l-1}\right),\sigma_3\mathcal{B}(f_k)\right\rangle\right| \\
&+\frac{2}{t}\left|\operatorname{Im}\left\langle H_K^k(\mathcal{F}(f)+r),\sigma_3\mathcal{B}(f_k)\right\rangle\right|
\leq\frac{\eta}{t}J_k+\frac{C}{t}J_0+Ct^{-4\nu-1+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)}.
\end{aligned} 2 γ t 2 ν − 1 l = 0 ∑ k − 1 j = 1 ∑ K Re ⟨ H K l ( (( y − y j ) ⋅ ∇ V j + 2 V j ) f k − l − 1 ) , σ 3 B ( f k ) ⟩ + t 2 Im ⟨ H K k ( F ( f ) + r ) , σ 3 B ( f k ) ⟩ ≤ t η J k + t C J 0 + C t − 4 ν − 1 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) .
Summing (188) over l = 0 , … , k − 1 l=0,\ldots,k-1 l = 0 , … , k − 1 and using ∑ l = 0 k − 1 ( 1 + ( − 1 ) l ) = 2 [ k + 1 2 ] \sum_{l=0}^{k-1}(1+(-1)^l)=2\left[\frac{k+1}{2}\right] ∑ l = 0 k − 1 ( 1 + ( − 1 ) l ) = 2 [ 2 k + 1 ] , we conclude that
E k , 3 ≤ ( 32 p γ [ k + 1 2 ] K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p + 2 η + C t 2 ν ) J k t + C t J 0 + C t − 4 ν − 1 + 2 ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) . (192) \mathcal{E}_{k,3}\leq\left(\frac{32p\gamma\left[\frac{k+1}{2}\right]Kc_\infty^{2p}}{(d-2)^2-4K(2p+1)c_\infty^{2p}}+2\eta+Ct^{2\nu}\right)\frac{J_k}{t}+\frac{C}{t}J_0+Ct^{-4\nu-1+2(\nu-\epsilon_1)(|\lambda_-|-s_c+3)}. \tag*{(192)} E k , 3 ≤ ( ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 32 p γ [ 2 k + 1 ] K c ∞ 2 p + 2 η + C t 2 ν ) t J k + t C J 0 + C t − 4 ν − 1 + 2 ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) . (192)
Finally, integrating the scaling term in E k , 4 \mathcal{E}_{k,4} E k , 4 by parts gives
E k , 4 = 2 p ( − 1 ) k t 2 ν − 1 Re ⟨ f k , σ 3 ∑ j = 1 K ( γ ( y − y j ) ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) + ( 2 γ s c − 2 ν − 4 k γ ) Q j 2 p − 1 χ 1 , j ) σ 1 f k ⟩ . \mathcal{E}_{k,4}=2p(-1)^kt^{2\nu-1}\operatorname{Re}\left\langle f_k,\sigma_3\sum_{j=1}^{K}\left(\gamma(y-y_j)\cdot\nabla\left(Q_j^{2p-1}\chi_{1,j}\right)+(2\gamma s_c-2\nu-4k\gamma)Q_j^{2p-1}\chi_{1,j}\right)\sigma_1f_k\right\rangle. E k , 4 = 2 p ( − 1 ) k t 2 ν − 1 Re ⟨ f k , σ 3 j = 1 ∑ K ( γ ( y − y j ) ⋅ ∇ ( Q j 2 p − 1 χ 1 , j ) + ( 2 γ s c − 2 ν − 4 k γ ) Q j 2 p − 1 χ 1 , j ) σ 1 f k ⟩ .
By (161) , ∣ λ + ∣ > 2 |\lambda_+|>2 ∣ λ + ∣ > 2 , and Hardy’s inequality,
∣ E k , 4 ∣ ≤ C t 2 ν − 1 ∑ j = 1 K ∥ ⟨ y − y j ⟩ − 1 f k ∥ L 2 2 ≤ C t 2 ν − 1 J k . (193) |\mathcal{E}_{k,4}|\leq Ct^{2\nu-1}\sum_{j=1}^{K}\left\|\langle y-y_j\rangle^{-1}f_k\right\|_{L^2}^2\leq Ct^{2\nu-1}J_k. \tag*{(193)} ∣ E k , 4 ∣ ≤ C t 2 ν − 1 j = 1 ∑ K ⟨ y − y j ⟩ − 1 f k L 2 2 ≤ C t 2 ν − 1 J k . (193)
Substituting (179) , (192) and (193) into (178) , we obtain
d d t J k ( t ) ≤ C k ( K , L ) + C t 2 ν t J k + C t J 0 + C t − 4 ν − 1 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) , (194) \frac{\mathrm{d}}{\mathrm{d}t}J_k(t)\leq\frac{C_k(K,L)+Ct^{2\nu}}{t}J_k+\frac{C}{t}J_0+Ct^{-4\nu-1+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)}, \tag*{(194)} d t d J k ( t ) ≤ t C k ( K , L ) + C t 2 ν J k + t C J 0 + C t − 4 ν − 1 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) , (194)
where
C k ( K , L ) = 2 γ ( 2 k + 1 − s c ) + 8 γ K c ∞ 2 p ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p ( ( 2 p + 1 ) ( 2 k + 1 ) + 4 p [ k + 1 2 ] ) + 2 η . (195) C_k(K,L)=2\gamma(2k+1-s_c)+\frac{8\gamma Kc_\infty^{2p}}{(d-2)^2-4K(2p+1)c_\infty^{2p}}\left((2p+1)(2k+1)+4p\left[\frac{k+1}{2}\right]\right)+2\eta. \tag*{(195)} C k ( K , L ) = 2 γ ( 2 k + 1 − s c ) + ( d − 2 ) 2 − 4 K ( 2 p + 1 ) c ∞ 2 p 8 γ K c ∞ 2 p ( ( 2 p + 1 ) ( 2 k + 1 ) + 4 p [ 2 k + 1 ] ) + 2 η . (195)
Step 3. Control of the weighted L 2 L^2 L 2 norm. Applying the equation (134) to the derivative d d t ∥ y ∥ L 2 2 \frac{\mathrm{d}}{\mathrm{d}t}\|y\|_{L^2}^2 d t d ∥ y ∥ L 2 2 , we obtain
t 2 L λ + + 1 d d t ∥ y ∥ L 2 2 = − i ( ⟨ y ∣ H K f , ∣ y ∣ f ⟩ − ⟨ ∣ y ∣ f , ∣ y ∣ H K f ⟩ ) + γ t 2 L λ + − 1 ( ⟨ ∣ y ∣ Λ f , ∣ y ∣ f ⟩ + ⟨ ∣ y ∣ f , ∣ y ∣ Λ f ⟩ ) − i ( ⟨ ∣ y ∣ ( F ( f ) + r ) , ∣ y ∣ f ⟩ − ⟨ ∣ y ∣ f , ∣ y ∣ ( F ( f ) + r ) ⟩ ) . t^{\frac{2L}{\lambda_+}+1}\frac{\mathrm{d}}{\mathrm{d}t}\|y\|_{L^2}^2=-i\left(\langle y|H_{K}f,|y|f\rangle-\langle|y|f,|y|H_{K}f\rangle\right)
+\gamma t^{\frac{2L}{\lambda_+}-1}\left(\langle|y|\Lambda f,|y|f\rangle+\langle|y|f,|y|\Lambda f\rangle\right)
-i\left(\langle|y|(\mathcal{F}(f)+r),|y|f\rangle-\langle|y|f,|y|(\mathcal{F}(f)+r)\rangle\right). t λ + 2 L + 1 d t d ∥ y ∥ L 2 2 = − i ( ⟨ y ∣ H K f , ∣ y ∣ f ⟩ − ⟨ ∣ y ∣ f , ∣ y ∣ H K f ⟩ ) + γ t λ + 2 L − 1 ( ⟨ ∣ y ∣Λ f , ∣ y ∣ f ⟩ + ⟨ ∣ y ∣ f , ∣ y ∣Λ f ⟩ ) − i ( ⟨ ∣ y ∣ ( F ( f ) + r ) , ∣ y ∣ f ⟩ − ⟨ ∣ y ∣ f , ∣ y ∣ ( F ( f ) + r )⟩ ) .
Integrating by parts gives
⟨ ∣ y ∣ H K f , ∣ y ∣ f ⟩ − ⟨ ∣ y ∣ f , ∣ y ∣ H K f ⟩ = 4 i Im ⟨ y ⋅ ∇ f , σ 3 f ⟩ − 2 p ∑ j = 1 K ⟨ ∣ y ∣ Q j 2 p σ 3 σ 1 f , ∣ y ∣ f ⟩ . \langle|y|H_{K}f,|y|f\rangle-\langle|y|f,|y|H_{K}f\rangle=4i\operatorname{Im}\langle y\cdot\nabla f,\sigma_{3}f\rangle-2p\sum_{j=1}^{K}\langle|y|Q_{j}^{2p}\sigma_{3}\sigma_{1}f,|y|f\rangle. ⟨ ∣ y ∣ H K f , ∣ y ∣ f ⟩ − ⟨ ∣ y ∣ f , ∣ y ∣ H K f ⟩ = 4 i Im ⟨ y ⋅ ∇ f , σ 3 f ⟩ − 2 p j = 1 ∑ K ⟨ ∣ y ∣ Q j 2 p σ 3 σ 1 f , ∣ y ∣ f ⟩ .
and
⟨ ∣ y ∣ Λ f , ∣ y ∣ f ⟩ + ⟨ ∣ y ∣ f , ∣ y ∣ Λ f ⟩ = − 2 ( 1 + s c ) ∥ y ∥ L 2 2 . \langle|y|\Lambda f,|y|f\rangle+\langle|y|f,|y|\Lambda f\rangle=-2(1+s_{c})\|y\|_{L^2}^2. ⟨ ∣ y ∣Λ f , ∣ y ∣ f ⟩ + ⟨ ∣ y ∣ f , ∣ y ∣Λ f ⟩ = − 2 ( 1 + s c ) ∥ y ∥ L 2 2 .
Since F = F 0 + F 1 + F 2 \mathcal{F}=\mathcal{F}_{0}+\mathcal{F}_{1}+\mathcal{F}_{2} F = F 0 + F 1 + F 2 and F 0 = 0 \mathcal{F}_{0}=0 F 0 = 0 by (156) , using 2 L ∣ λ + ∣ = 1 + 2 ν \frac{2L}{|\lambda_+|}=1+2\nu ∣ λ + ∣ 2 L = 1 + 2 ν , we obtain
d d t ∥ y ∥ L 2 2 = − 2 γ ( s c + 1 ) t ∥ y ∥ L 2 2 + 4 t − 1 − 2 ν Im ⟨ y ⋅ ∇ f , σ 3 f ⟩ + 2 t − 1 − 2 ν Im ⟨ ∣ y ∣ ( F 1 ( f ) − p ∑ j = 1 K Q j 2 p σ 3 σ 1 f ) , ∣ y ∣ f ⟩ + 2 t − 1 − 2 ν Im ⟨ ∣ y ∣ ( F 2 ( f ) + r ) , ∣ y ∣ f ⟩ . (196) \frac{\mathrm{d}}{\mathrm{d}t}\|y\|_{L^2}^2=-\frac{2\gamma(s_{c}+1)}{t}\|y\|_{L^2}^2+4t^{-1-2\nu}\operatorname{Im}\langle y\cdot\nabla f,\sigma_{3}f\rangle
+2t^{-1-2\nu}\operatorname{Im}\langle|y|(\mathcal{F}_{1}(f)-p\sum_{j=1}^{K}Q_{j}^{2p}\sigma_{3}\sigma_{1}f),|y|f\rangle
+2t^{-1-2\nu}\operatorname{Im}\langle|y|(\mathcal{F}_{2}(f)+r),|y|f\rangle. \tag*{(196)} d t d ∥ y ∥ L 2 2 = − t 2 γ ( s c + 1 ) ∥ y ∥ L 2 2 + 4 t − 1 − 2 ν Im ⟨ y ⋅ ∇ f , σ 3 f ⟩ + 2 t − 1 − 2 ν Im ⟨ ∣ y ∣ ( F 1 ( f ) − p j = 1 ∑ K Q j 2 p σ 3 σ 1 f ) , ∣ y ∣ f ⟩ + 2 t − 1 − 2 ν Im ⟨ ∣ y ∣ ( F 2 ( f ) + r ) , ∣ y ∣ f ⟩ . (196)
For the second line of (196) , the definition of F 1 \mathcal{F}_{1} F 1 in (135) gives
∣ Im ⟨ ∣ y ∣ ( F 1 ( f ) − p ∑ j = 1 K Q j 2 p σ 3 σ 1 f ) , ∣ y ∣ f ⟩ ∣ ≤ C ∫ ∣ y ∣ 2 ∣ U a p ∣ 2 p ∣ f ∣ 2 d y . \left|\operatorname{Im}\langle|y|(\mathcal{F}_{1}(f)-p\sum_{j=1}^{K}Q_{j}^{2p}\sigma_{3}\sigma_{1}f),|y|f\rangle\right|\le C\int|y|^2|U^{ap}|^{2p}|f|^2\,\mathrm{d}y. Im ⟨ ∣ y ∣ ( F 1 ( f ) − p j = 1 ∑ K Q j 2 p σ 3 σ 1 f ) , ∣ y ∣ f ⟩ ≤ C ∫ ∣ y ∣ 2 ∣ U a p ∣ 2 p ∣ f ∣ 2 d y .
By (92) and (18) ,
∣ U a p ( t , y ) ∣ 2 p ≤ C ∑ j = 1 K ⟨ y − y j ⟩ − 2 + C ( t 4 p ν + t 1 + 2 ν ) . |U^{ap}(t,y)|^{2p}\le C\sum_{j=1}^{K}\langle y-y_{j}\rangle^{-2}+C(t^{4p\nu}+t^{1+2\nu}). ∣ U a p ( t , y ) ∣ 2 p ≤ C j = 1 ∑ K ⟨ y − y j ⟩ − 2 + C ( t 4 p ν + t 1 + 2 ν ) .
Since ∣ y j ∣ ≤ C λ ( t ) |y_{j}|\le C\lambda(t) ∣ y j ∣ ≤ C λ ( t ) , by Hardy’s inequality, we obtain
∫ ∣ y ∣ 2 ∣ U a p ∣ 2 p ∣ f ∣ 2 d y ≤ C ∑ j = 1 K ( ∣ y j ∣ 2 ∥ ∣ y − y j ∣ − 1 f ∥ L 2 2 + ∥ f ∥ L 2 2 ) + C ( t 4 p ν + t 1 + 2 ν ) ∥ y f ∥ L 2 2 ≤ C t − 1 − 2 ν ∥ ∇ f ∥ L 2 2 + C ∥ f ∥ L 2 2 + C ( t 4 p ν + t 1 + 2 ν ) ∥ y f ∥ L 2 2 . (197) \int|y|^2|U^{ap}|^{2p}|f|^2\,\mathrm{d}y\le C\sum_{j=1}^{K}(|y_{j}|^2\||y-y_{j}|^{-1}f\|_{L^2}^2+\|f\|_{L^2}^2)+C(t^{4p\nu}+t^{1+2\nu})\|y f\|_{L^2}^2
\le Ct^{-1-2\nu}\|\nabla f\|_{L^2}^2+C\|f\|_{L^2}^2+C(t^{4p\nu}+t^{1+2\nu})\|y f\|_{L^2}^2. \tag*{(197)} ∫ ∣ y ∣ 2 ∣ U a p ∣ 2 p ∣ f ∣ 2 d y ≤ C j = 1 ∑ K ( ∣ y j ∣ 2 ∥∣ y − y j ∣ − 1 f ∥ L 2 2 + ∥ f ∥ L 2 2 ) + C ( t 4 p ν + t 1 + 2 ν ) ∥ y f ∥ L 2 2 ≤ C t − 1 − 2 ν ∥∇ f ∥ L 2 2 + C ∥ f ∥ L 2 2 + C ( t 4 p ν + t 1 + 2 ν ) ∥ y f ∥ L 2 2 . (197)
and integration by parts gives
∥ f ∥ L 2 2 = − 2 d Re ⟨ y ⋅ ∇ f , f ⟩ ≤ 2 d ∥ y f ∥ L 2 ∥ ∇ f ∥ L 2 . (198) \|f\|_{L^2}^2=-\frac{2}{d}\operatorname{Re}\langle y\cdot\nabla f,f\rangle\le\frac{2}{d}\|y f\|_{L^2}\|\nabla f\|_{L^2}. \tag*{(198)} ∥ f ∥ L 2 2 = − d 2 Re ⟨ y ⋅ ∇ f , f ⟩ ≤ d 2 ∥ y f ∥ L 2 ∥∇ f ∥ L 2 . (198)
Combining (197) and (198) with the bootstrap assumption (145) and applying Young’s inequality, for sufficiently small T ∗ T^* T ∗ , we obtain the following estimate
t − 1 − 2 ν ∣ Im ⟨ ∣ y ∣ ( F 1 ( f ) − p ∑ j = 1 K Q j 2 p σ 3 σ 1 f ) , ∣ y ∣ f ⟩ ∣ ≤ 1 8 t ∥ y f ∥ L 2 2 + C t − 2 − 8 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − 2 η . (199) t^{-1-2\nu}\left|\operatorname{Im}\langle|y|(\mathcal{F}_{1}(f)-p\sum_{j=1}^{K}Q_{j}^{2p}\sigma_{3}\sigma_{1}f),|y|f\rangle\right|
\le\frac{1}{8t}\|y f\|_{L^2}^2+Ct^{-2-8\nu+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-2\eta}. \tag*{(199)} t − 1 − 2 ν Im ⟨ ∣ y ∣ ( F 1 ( f ) − p j = 1 ∑ K Q j 2 p σ 3 σ 1 f ) , ∣ y ∣ f ⟩ ≤ 8 t 1 ∥ y f ∥ L 2 2 + C t − 2 − 8 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η . (199)
We next estimate the contribution involving F 2 + r \mathcal{F}_2+r F 2 + r . By (135) and (146) ,
∥ ∣ y ∣ F 2 ( f ) ∥ L 2 ≤ C ( ∥ f ∥ L ∞ + ∥ f ∥ L ∞ 2 p ) ∥ ∣ y ∣ f ∥ L 2 ≤ C t 2 ν + 3 η ∥ ∣ y ∣ f ∥ L 2 , (200) \lVert|y|\mathcal{F}_2(f)\rVert_{L^2}\leq C\left(\lVert f\rVert_{L^\infty}+\lVert f\rVert_{L^\infty}^{2p}\right)\lVert|y|f\rVert_{L^2}\leq Ct^{2\nu+3\eta}\lVert|y|f\rVert_{L^2}, \tag*{(200)} ∥ ∣ y ∣ F 2 ( f ) ∥ L 2 ≤ C ( ∥ f ∥ L ∞ + ∥ f ∥ L ∞ 2 p ) ∥ ∣ y ∣ f ∥ L 2 ≤ C t 2 ν + 3 η ∥ ∣ y ∣ f ∥ L 2 , (200)
where the last inequality follows from ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η (\nu-\epsilon_1)(|\lambda_-|-s_c+3)>4\nu+4\eta ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η . Moreover, Hardy’s inequality and (148) for k = 0 k=0 k = 0 imply
∥ ∣ y ∣ r ∥ L 2 ≤ C λ 2 ( t ) ∥ ∇ r ∥ L 2 ≤ C t − 1 − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) . (201) \lVert|y|r\rVert_{L^2}\leq C\lambda^2(t)\lVert\nabla r\rVert_{L^2}\leq Ct^{-1-2\nu+\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)}. \tag*{(201)} ∥ ∣ y ∣ r ∥ L 2 ≤ C λ 2 ( t ) ∥ ∇ r ∥ L 2 ≤ C t − 1 − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) . (201)
Combining (200) and (201) with Hölder and Young estimates, we obtain that for sufficiently small T ∗ T^* T ∗ ,
t − 1 − 2 ν ∣ Im ⟨ ∣ y ∣ ( F 2 ( f ) + r ) , ∣ y ∣ f ⟩ ∣ ≤ 1 8 t ∥ ∣ y ∣ f ∥ L 2 2 + C t − 3 − 8 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) . (202) t^{-1-2\nu}\left|\operatorname{Im}\langle|y|(\mathcal{F}_2(f)+r),|y|f\rangle\right|
\leq\frac{1}{8t}\lVert|y|f\rVert_{L^2}^{2}+Ct^{-3-8\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)}. \tag*{(202)} t − 1 − 2 ν ∣ Im ⟨ ∣ y ∣ ( F 2 ( f ) + r ) , ∣ y ∣ f ⟩ ∣ ≤ 8 t 1 ∥ ∣ y ∣ f ∥ L 2 2 + C t − 3 − 8 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) . (202)
Similarly, by Hölder estimate, the bootstrap assumption (145) and Young’s inequality, we get
t − 1 − 2 ν ∣ Im ⟨ y ⋅ ∇ f , σ 3 f ⟩ ∣ ≤ 1 8 t ∥ ∣ y ∣ f ∥ L 2 2 + C t − 3 − 8 ν + 2 ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) . (203) t^{-1-2\nu}\left|\operatorname{Im}\langle y\cdot\nabla f,\sigma_3f\rangle\right|
\leq\frac{1}{8t}\lVert|y|f\rVert_{L^2}^{2}+Ct^{-3-8\nu+2(\nu-\epsilon_1)(|\lambda_-|-s_c+3)}. \tag*{(203)} t − 1 − 2 ν ∣ Im ⟨ y ⋅ ∇ f , σ 3 f ⟩ ∣ ≤ 8 t 1 ∥ ∣ y ∣ f ∥ L 2 2 + C t − 3 − 8 ν + 2 ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) . (203)
Substituting (199) , (202) and (203) into (196) , we conclude
d d t ∥ ∣ y ∣ f ∥ L 2 2 ≤ 1 − 2 γ ( s c + 1 ) t ∥ ∣ y ∣ f ∥ L 2 2 + C t − 3 − 8 ν + 2 ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) . (204) \frac{\mathrm{d}}{\mathrm{d}t}\lVert|y|f\rVert_{L^2}^{2}\leq\frac{1-2\gamma(s_c+1)}{t}\lVert|y|f\rVert_{L^2}^{2}+Ct^{-3-8\nu+2(\nu-\epsilon_1)(|\lambda_-|-s_c+3)}. \tag*{(204)} d t d ∥ ∣ y ∣ f ∥ L 2 2 ≤ t 1 − 2 γ ( s c + 1 ) ∥ ∣ y ∣ f ∥ L 2 2 + C t − 3 − 8 ν + 2 ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) . (204)
Since ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η (\nu-\epsilon_1)(|\lambda_-|-s_c+3)>4\nu+4\eta ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η , we have
1 − 2 γ ( s c + 1 ) < − 2 < − 8 ν − 2 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) . 1-2\gamma(s_c+1)<-2<-8\nu-2+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3). 1 − 2 γ ( s c + 1 ) < − 2 < − 8 ν − 2 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) .
Integrating (204) with zero initial condition at t 1 t_1 t 1 gives
∥ ∣ y ∣ f ∥ L 2 ≤ C t − 4 ν − 1 + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) , t ∈ [ t 1 , t 2 ] . (205) \lVert|y|f\rVert_{L^2}\leq Ct^{-4\nu-1+\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)},\qquad t\in[t_1,t_2]. \tag*{(205)} ∥ ∣ y ∣ f ∥ L 2 ≤ C t − 4 ν − 1 + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) , t ∈ [ t 1 , t 2 ] . (205)
Step 4. Proof of Proposition 7.1. By (174) and C 0 ( K , L ) < 0 C_0(K,L)<0 C 0 ( K , L ) < 0 , choose c 0 ≥ 1 c_0\geq1 c 0 ≥ 1 sufficiently large that C + c 0 C 0 ( K , L ) ≤ 0 C+c_0C_0(K,L)\leq0 C + c 0 C 0 ( K , L ) ≤ 0 and
∥ ∇ f ∥ H 2 k 2 ≤ C ( J k + c 0 J 0 ) . (206) \lVert\nabla f\rVert_{H^{2k}}^{2}\leq C(J_k+c_0J_0). \tag*{(206)} ∥ ∇ f ∥ H 2 k 2 ≤ C ( J k + c 0 J 0 ) . (206)
Therefore, we set J ( t ) = J k ( t ) + c 0 J 0 ( t ) J(t)=J_k(t)+c_0J_0(t) J ( t ) = J k ( t ) + c 0 J 0 ( t ) . By (195) , C k ( K , L ) C_k(K,L) C k ( K , L ) is increasing in k k k . Hence
C k ( K , L ) ≤ C ( M − 1 ) / 2 ( K , L ) , 1 ≤ k ≤ M − 1 2 . C_k(K,L)\leq C_{(M-1)/2}(K,L),\qquad1\leq k\leq\frac{M-1}{2}. C k ( K , L ) ≤ C ( M − 1 ) /2 ( K , L ) , 1 ≤ k ≤ 2 M − 1 .
Combining (169) and (194) , we obtain
d d t J ( t ) ≤ C ( M − 1 ) / 2 ( K , L ) + C t 2 ν t J ( t ) + C t − 4 ν − 1 + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η . (207) \frac{\mathrm{d}}{\mathrm{d}t}J(t)\leq\frac{C_{(M-1)/2}(K,L)+Ct^{2\nu}}{t}J(t)+Ct^{-4\nu-1+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)-\eta}. \tag*{(207)} d t d J ( t ) ≤ t C ( M − 1 ) /2 ( K , L ) + C t 2 ν J ( t ) + C t − 4 ν − 1 + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η . (207)
The choice of M M M in (136) and sufficiently small η \eta η imply, after decreasing T ∗ T^* T ∗ , that
C ( M − 1 ) / 2 ( K , L ) + C t 2 ν < − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η , 0 < t ≤ T ∗ . (208) C_{(M-1)/2}(K,L)+Ct^{2\nu}<-4\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)-\eta,\qquad0<t\leq T^*. \tag*{(208)} C ( M − 1 ) /2 ( K , L ) + C t 2 ν < − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η , 0 < t ≤ T ∗ . (208)
Since f ( t 1 ) = 0 f(t_1)=0 f ( t 1 ) = 0 , we have J ( t 1 ) = 0 J(t_1)=0 J ( t 1 ) = 0 . Integrating (207) over [ t 1 , t ] [t_1,t] [ t 1 , t ] and using (208) , we obtain
J k ( t ) + c 0 J 0 ( t ) ≤ C t − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η , t ∈ [ t 1 , t 2 ] , J_k(t)+c_0J_0(t)\leq Ct^{-4\nu+2\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)-\eta},\qquad t\in[t_1,t_2], J k ( t ) + c 0 J 0 ( t ) ≤ C t − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η , t ∈ [ t 1 , t 2 ] ,
where C C C is independent of t 1 t_1 t 1 and t 2 t_2 t 2 . Together with (206) , this gives
∥ f ( t ) ∥ H ˙ 2 k + 1 ≤ ∥ ∇ f ( t ) ∥ H 2 k ≤ C t − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η 2 , 1 ≤ 2 k + 1 ≤ M . (209) \lVert f(t)\rVert_{\dot H^{2k+1}}\leq\lVert\nabla f(t)\rVert_{H^{2k}}\leq Ct^{-2\nu+\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)-\frac{\eta}{2}},\qquad1\leq2k+1\leq M. \tag*{(209)} ∥ f ( t ) ∥ H ˙ 2 k + 1 ≤ ∥ ∇ f ( t ) ∥ H 2 k ≤ C t − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η , 1 ≤ 2 k + 1 ≤ M . (209)
Taking 2 k + 1 = M 2k+1=M 2 k + 1 = M , this gives
∥ ∇ f ( t ) ∥ H M − 1 ≤ C t − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η 2 , \lVert\nabla f(t)\rVert_{H^{M-1}}\leq Ct^{-2\nu+\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)-\frac{\eta}{2}}, ∥ ∇ f ( t ) ∥ H M − 1 ≤ C t − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η ,
which improves the bootstrap estimate (145) . Therefore, the bootstrap argument closes on the interval [ t 1 , T ∗ ] [t_1,T^*] [ t 1 , T ∗ ] by continuity and the standard continuation criterion in H M H^M H M .
Recall that
ψ ( t , x ) − ψ a p ( t , x ) = λ 1 p ( t ) f ( t , y ) , y = λ ( t ) x . \psi(t,x)-\psi^{ap}(t,x)=\lambda^{\frac{1}{p}}(t)f(t,y),\qquad y=\lambda(t)x. ψ ( t , x ) − ψ a p ( t , x ) = λ p 1 ( t ) f ( t , y ) , y = λ ( t ) x .
For any 1 ≤ s ≤ M 1\le s\le M 1 ≤ s ≤ M , rescaling gives
∥ ψ − ψ a p ∥ H ˙ x s 2 + ∥ ∣ x ∣ ( ψ − ψ a p ) ∥ L x 2 2 = λ 2 ( s − s c ) ( t ) ∥ f ∥ H ˙ y s 2 + λ − 2 ( s c + 1 ) ( t ) ∥ ∣ y ∣ f ( t ) ∥ L y 2 2 . \|\psi-\psi^{ap}\|_{\dot H_x^s}^2+\||x|(\psi-\psi^{ap})\|_{L_x^2}^2=\lambda^{2(s-s_c)}(t)\|f\|_{\dot H_y^s}^2+\lambda^{-2(s_c+1)}(t)\||y|f(t)\|_{L_y^2}^2. ∥ ψ − ψ a p ∥ H ˙ x s 2 + ∥∣ x ∣ ( ψ − ψ a p ) ∥ L x 2 2 = λ 2 ( s − s c ) ( t ) ∥ f ∥ H ˙ y s 2 + λ − 2 ( s c + 1 ) ( t ) ∥∣ y ∣ f ( t ) ∥ L y 2 2 .
Combining this with (209) , we obtain
∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ C t − L ∣ λ + ∣ ( s − s c ) − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η 2 , 1 ≤ s ≤ M . (210) \|\psi(t)-\psi^{ap}(t)\|_{\dot H^s}\le Ct^{-\frac{L}{|\lambda_+|}(s-s_c)-2\nu+\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\frac{\eta}{2}},\qquad1\le s\le M. \tag*{(210)} ∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ C t − ∣ λ + ∣ L ( s − s c ) − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η , 1 ≤ s ≤ M . (210)
In particular, since T ≤ 1 T\le1 T ≤ 1 , for 1 ≤ s ≤ s c 1\le s\le s_c 1 ≤ s ≤ s c ,
∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ C t − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η 2 . (211) \|\psi(t)-\psi^{ap}(t)\|_{\dot H^s}\le Ct^{-2\nu+\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\frac{\eta}{2}}. \tag*{(211)} ∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ C t − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η . (211)
The exponent in (210) is minimized at s = M s=M s = M and is positive by (208) and C ( M − 1 ) / 2 ( K , L ) > 2 γ ( M − s c ) C(M-1)/2(K,L)>2\gamma(M-s_c) C ( M − 1 ) /2 ( K , L ) > 2 γ ( M − s c ) . At the same time, (205) yields
∥ ∣ x ∣ ( ψ ( t ) − ψ a p ( t ) ) ∥ L 2 ≤ C t ν ( 1 − 1 d ) ∣ λ − ∣ + ( 1 2 + ν d ) ( s c + 1 ) − 1 − 4 ν d . (212) \||x|(\psi(t)-\psi^{ap}(t))\|_{L^2}\le Ct^{\nu(1-\frac{1}{d})|\lambda_-|+(\frac{1}{2}+\frac{\nu}{d})(s_c+1)-1-\frac{4\nu}{d}}. \tag*{(212)} ∥∣ x ∣ ( ψ ( t ) − ψ a p ( t )) ∥ L 2 ≤ C t ν ( 1 − d 1 ) ∣ λ − ∣ + ( 2 1 + d ν ) ( s c + 1 ) − 1 − d 4 ν . (212)
This exponent is positive since ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η \nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)>4\nu+4\eta ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η and s c > 3 s_c>3 s c > 3 . This completes the proof of Proposition 7.1.
Proof of Theorem 1.1
Let { t j } j ≥ 1 ⊂ ( 0 , T ∗ ) \{t^j\}_{j\ge1}\subset(0,T^*) { t j } j ≥ 1 ⊂ ( 0 , T ∗ ) be a sequence such that t j → 0 t^j\to0 t j → 0 as j → ∞ j\to\infty j → ∞ . For each j j j , let ψ j \psi_j ψ j be the solution of
{ i ∂ t ψ j = − Δ ψ j − ∣ ψ j ∣ 2 p ψ j , ψ j ∣ t = t j = ψ a p ( t j ) . \begin{cases}
i\partial_t\psi_j=-\Delta\psi_j-|\psi_j|^{2p}\psi_j,\\
\psi_j|_{t=t^j}=\psi^{ap}(t^j).
\end{cases} { i ∂ t ψ j = − Δ ψ j − ∣ ψ j ∣ 2 p ψ j , ψ j ∣ t = t j = ψ a p ( t j ) .
Take N = M = 2 k + 1 N=M=2k+1 N = M = 2 k + 1 . By Proposition 7.1,
ψ j − ψ a p ∈ C ( [ t j , T ∗ ] , H ˙ 1 ∩ H ˙ 2 k + 1 ) \psi_j-\psi^{ap}\in C([t^j,T^*],\dot H^1\cap\dot H^{2k+1}) ψ j − ψ a p ∈ C ([ t j , T ∗ ] , H ˙ 1 ∩ H ˙ 2 k + 1 )
and, uniformly in j j j ,
∥ ψ j ( t , x ) − ψ a p ( t , x ) ∥ H ˙ 2 k + 1 ≤ C t − L ∣ λ + ∣ ( N − s c ) − 2 ν + ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) − η 2 , \|\psi_j(t,x)-\psi^{ap}(t,x)\|_{\dot H^{2k+1}}\le Ct^{-\frac{L}{|\lambda_+|}(N-s_c)-2\nu+(\nu-\epsilon_1)(|\lambda_-|-s_c+3)-\frac{\eta}{2}}, ∥ ψ j ( t , x ) − ψ a p ( t , x ) ∥ H ˙ 2 k + 1 ≤ C t − ∣ λ + ∣ L ( N − s c ) − 2 ν + ( ν − ϵ 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 η ,
∥ ∣ x ∣ ( ψ j ( t , x ) − ψ a p ( t , x ) ) ∥ L 2 ≤ C t ( ν − ϵ 1 ) ∣ λ − ∣ + ( 1 2 + ϵ 1 ) ( s c + 1 ) − 1 − 4 ϵ 1 . (213) \||x|(\psi_j(t,x)-\psi^{ap}(t,x))\|_{L^2}\le Ct^{(\nu-\epsilon_1)|\lambda_-|+(\frac{1}{2}+\epsilon_1)(s_c+1)-1-4\epsilon_1}. \tag*{(213)} ∥∣ x ∣ ( ψ j ( t , x ) − ψ a p ( t , x )) ∥ L 2 ≤ C t ( ν − ϵ 1 ) ∣ λ − ∣ + ( 2 1 + ϵ 1 ) ( s c + 1 ) − 1 − 4 ϵ 1 . (213)
for all t j ≤ t ≤ T ∗ t^j\le t\le T^* t j ≤ t ≤ T ∗ . By (210) with s = 1 s=1 s = 1 and (212) , integration by parts and Hölder’s inequality give
∥ ψ j ( t , x ) − ψ a p ( t , x ) ∥ L 2 2 ≤ 2 d ∥ ∣ x ∣ ( ψ j ( t , x ) − ψ a p ( t , x ) ) ∥ L 2 ∥ ∇ x ( ψ j ( t , x ) − ψ a p ( t , x ) ) ∥ L 2 ≤ C t − 2 γ ( N − s c ) − 4 ν + 2 ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − η . \begin{aligned}
\|\psi_j(t,x)-\psi^{ap}(t,x)\|_{L^2}^2\le\frac{2}{d}\||x|(\psi_j(t,x)-\psi^{ap}(t,x))\|_{L^2}\|\nabla_x(\psi_j(t,x)-\psi^{ap}(t,x))\|_{L^2} \\
&\le Ct^{-2\gamma(N-s_c)-4\nu+2\nu(1-\frac{1}{d})(|\lambda_-|-s_c+3)-\eta}.
\end{aligned} ∥ ψ j ( t , x ) − ψ a p ( t , x ) ∥ L 2 2 ≤ d 2 ∥∣ x ∣ ( ψ j ( t , x ) − ψ a p ( t , x )) ∥ L 2 ∥ ∇ x ( ψ j ( t , x ) − ψ a p ( t , x )) ∥ L 2 ≤ C t − 2 γ ( N − s c ) − 4 ν + 2 ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − η .
Here the last comparison follows from T ∗ ≤ 1 T^*\le1 T ∗ ≤ 1 , 2 γ = 1 + 2 ν 2\gamma=1+2\nu 2 γ = 1 + 2 ν , and 2 γ ( N − 1 ) + η 2 > 0 2\gamma(N-1)+\frac{\eta}{2}>0 2 γ ( N − 1 ) + 2 η > 0 . Consequently, the first bound in (213) also holds with H N H^N H N in place of H ˙ N \dot H^N H ˙ N .
Let R > 0 R>0 R > 0 and set B R = { x ∈ R d : ∣ x ∣ ≤ R } B_R=\{x\in\mathbb{R}^d:|x|\le R\} B R = { x ∈ R d : ∣ x ∣ ≤ R } . By (213) ,
sup j ∥ ( ψ j − ψ a p ) ( T ∗ ) ∥ H N < ∞ , sup j ∥ ∣ x ∣ ( ψ j − ψ a p ) ( T ∗ ) ∥ L 2 < ∞ , \sup_j\|(\psi_j-\psi^{ap})(T^*)\|_{H^N}<\infty,\qquad\sup_j\||x|(\psi_j-\psi^{ap})(T^*)\|_{L^2}<\infty, j sup ∥ ( ψ j − ψ a p ) ( T ∗ ) ∥ H N < ∞ , j sup ∥∣ x ∣ ( ψ j − ψ a p ) ( T ∗ ) ∥ L 2 < ∞ ,
Thus the sequence ψ j ( T ∗ ) − ψ a p ( T ∗ ) \psi_j(T^*)-\psi^{ap}(T^*) ψ j ( T ∗ ) − ψ a p ( T ∗ ) is uniformly bounded in H B R N H^N_{B_R} H B R N . Hence, by weak compactness in H N H^N H N and Rellich’s theorem, after extracting a subsequence, there exists a function w ∈ H N ( R d ) w\in H^N(\mathbb{R}^d) w ∈ H N ( R d ) such that
ψ j ( T ∗ ) − ψ a p ( T ∗ ) → w in H B R l , 0 ≤ l < N , (214) \psi_j(T^*)-\psi^{ap}(T^*)\to w\quad\text{in }H^l_{B_R},\qquad0\le l<N, \tag*{(214)} ψ j ( T ∗ ) − ψ a p ( T ∗ ) → w in H B R l , 0 ≤ l < N , (214)
for every fixed R > 0 R>0 R > 0 .
We next control the exterior region. By (213) , we have
sup j ∥ ψ j ( T ∗ , x ) − ψ a p ( T ∗ , x ) ∥ L 2 ( ∣ x ∣ > R ) ≤ 1 R sup j ∥ ∣ x ∣ ( ψ j ( T ∗ , x ) − ψ a p ( T ∗ , x ) ) ∥ L 2 ≤ C R . \sup_j\left\|\psi_j(T^*,x)-\psi^{ap}(T^*,x)\right\|_{L^2(|x|>R)}
\le\frac{1}{R}\sup_j\left\||x|\left(\psi_j(T^*,x)-\psi^{ap}(T^*,x)\right)\right\|_{L^2}
\le\frac{C}{R}. j sup ∥ ψ j ( T ∗ , x ) − ψ a p ( T ∗ , x ) ∥ L 2 ( ∣ x ∣ > R ) ≤ R 1 j sup ∥ ∣ x ∣ ( ψ j ( T ∗ , x ) − ψ a p ( T ∗ , x ) ) ∥ L 2 ≤ R C .
Consequently,
lim R → ∞ sup j ∥ ( ψ j ( T ∗ , x ) − ψ a p ( T ∗ , x ) ) ∥ L 2 ( ∣ x ∣ > R ) = 0. (215) \lim_{R\to\infty}\sup_j\left\|\left(\psi_j(T^*,x)-\psi^{ap}(T^*,x)\right)\right\|_{L^2(|x|>R)}=0. \tag*{(215)} R → ∞ lim j sup ∥ ( ψ j ( T ∗ , x ) − ψ a p ( T ∗ , x ) ) ∥ L 2 ( ∣ x ∣ > R ) = 0. (215)
Combining (214) with (215) , we obtain
ψ j ( T ∗ ) − ψ a p ( T ∗ ) → w in L 2 ( R d ) . (216) \psi_j(T^*)-\psi^{ap}(T^*)\to w\quad\text{in }L^2(\mathbb{R}^d). \tag*{(216)} ψ j ( T ∗ ) − ψ a p ( T ∗ ) → w in L 2 ( R d ) . (216)
Since ψ j ( T ∗ ) − ψ a p ( T ∗ ) \psi_j(T^*)-\psi^{ap}(T^*) ψ j ( T ∗ ) − ψ a p ( T ∗ ) is uniformly bounded in H N H^N H N and converges strongly in L 2 L^2 L 2 , interpolation gives, for every 0 ≤ l < N 0\le l<N 0 ≤ l < N ,
∥ ψ j ( T ∗ ) − ψ a p ( T ∗ ) − w ∥ H l ≤ ∥ ψ j ( T ∗ ) − ψ a p ( T ∗ ) − w ∥ L 2 1 − l N × ( sup j ∥ ( ψ j − ψ a p ) ( T ∗ ) ∥ H N + ∥ w ∥ H N ) l N . \left\|\psi_j(T^*)-\psi^{ap}(T^*)-w\right\|_{H^l}
\le\left\|\psi_j(T^*)-\psi^{ap}(T^*)-w\right\|_{L^2}^{1-\frac{l}{N}}
\times\left(\sup_j\left\|\left(\psi_j-\psi^{ap}\right)(T^*)\right\|_{H^N}+\left\|w\right\|_{H^N}\right)^{\frac{l}{N}}. ∥ ψ j ( T ∗ ) − ψ a p ( T ∗ ) − w ∥ H l ≤ ∥ ψ j ( T ∗ ) − ψ a p ( T ∗ ) − w ∥ L 2 1 − N l × ( j sup ∥ ( ψ j − ψ a p ) ( T ∗ ) ∥ H N + ∥ w ∥ H N ) N l .
Thus,
ψ j ( T ∗ ) − ψ a p ( T ∗ ) → w in H l ( R d ) , 0 ≤ l < N . (217) \psi_j(T^*)-\psi^{ap}(T^*)\to w\quad\text{in }H^l(\mathbb{R}^d),\qquad0\le l<N. \tag*{(217)} ψ j ( T ∗ ) − ψ a p ( T ∗ ) → w in H l ( R d ) , 0 ≤ l < N . (217)
Moreover, by weak compactness in H N H^N H N and (216) ,
ψ j ( T ∗ ) − ψ a p ( T ∗ ) ⇀ w in H N ( R d ) . \psi_j(T^*)-\psi^{ap}(T^*)\rightharpoonup w\quad\text{in }H^N(\mathbb{R}^d). ψ j ( T ∗ ) − ψ a p ( T ∗ ) ⇀ w in H N ( R d ) .
We now consider the Cauchy problem
{ i ψ t = − Δ ψ − ∣ ψ ∣ 2 p ψ , ψ ∣ t = T ∗ = ψ a p ( T ∗ ) + w . \begin{cases}
i\psi_t=-\Delta\psi-|\psi|^{2p}\psi,\\
\left.\psi\right|_{t=T^*}=\psi^{ap}(T^*)+w.
\end{cases} { i ψ t = − Δ ψ − ∣ ψ ∣ 2 p ψ , ψ ∣ t = T ∗ = ψ a p ( T ∗ ) + w .
Since N > d / 2 N>d/2 N > d /2 , local well-posedness gives a unique solution ψ ∈ C ( ( t ∗ , T ∗ ] , H N ) \psi\in C((t_*,T^*],H^N) ψ ∈ C (( t ∗ , T ∗ ] , H N ) for some 0 ≤ t ∗ < T ∗ 0\le t_*<T^* 0 ≤ t ∗ < T ∗ . Since t j → 0 t_j\to0 t j → 0 , the solutions ψ j \psi_j ψ j are defined on every compact subinterval of ( t ∗ , T ∗ ] (t_*,T^*] ( t ∗ , T ∗ ] for all sufficiently large j j j . By (217) , local well-posedness in H l H^l H l for d 2 < l < N \frac{d}{2}<l<N 2 d < l < N , and interpolation with the uniform H N H^N H N bounds,
ψ j ⟶ ψ in C l o c ( ( t ∗ , T ∗ ] , H l ) , 0 ≤ l < N . \psi_j\longrightarrow\psi\quad\text{in }C_{\mathrm{loc}}((t_*,T^*],H^l),\qquad0\le l<N. ψ j ⟶ ψ in C loc (( t ∗ , T ∗ ] , H l ) , 0 ≤ l < N .
Applying (210) to ψ j \psi_j ψ j and using strong convergence for s < N s<N s < N and weak lower semicontinuity for s = N s=N s = N , we obtain, for each fixed t ∈ ( t ∗ , T ∗ ] t\in(t_*,T^*] t ∈ ( t ∗ , T ∗ ] ,
∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ lim inf j → ∞ ∥ ψ j ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ C t − γ ( s − s c ) − 2 ν + ν ( 1 − 1 d ) ( ∣ λ − ∣ − s c + 3 ) − n 2 , 1 ≤ s ≤ N . (218) \left\|\psi(t)-\psi^{ap}(t)\right\|_{\dot H^s}
\le\liminf_{j\to\infty}\left\|\psi_j(t)-\psi^{ap}(t)\right\|_{\dot H^s}
\le Ct^{-\gamma(s-s_c)-2\nu+\nu\left(1-\frac{1}{d}\right)(|\lambda_-|-s_c+3)-\frac{n}{2}},\qquad1\le s\le N. \tag*{(218)} ∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ j → ∞ lim inf ∥ ψ j ( t ) − ψ a p ( t ) ∥ H ˙ s ≤ C t − γ ( s − s c ) − 2 ν + ν ( 1 − d 1 ) ( ∣ λ − ∣ − s c + 3 ) − 2 n , 1 ≤ s ≤ N . (218)
If t ∗ > 0 t_*>0 t ∗ > 0 , (218) and the smoothness of ψ a p \psi^{ap} ψ a p give
sup t ∗ < t ≤ T ∗ ∥ ψ ( t ) ∥ H N < ∞ , \sup_{t_*<t\le T^*}\|\psi(t)\|_{H^N}<\infty, t ∗ < t ≤ T ∗ sup ∥ ψ ( t ) ∥ H N < ∞ ,
which contradicts the maximality of the existence interval by the continuation criterion. Hence t ∗ = 0 t_*=0 t ∗ = 0 . The radiation bound (6) follows by scaling (97) and applying (210) .
We next prove the strong convergence of the radiation. The time independent term f 0 f_0 f 0 in the outer expansion (75) determines the limiting radial profile of ψ r a d a p \psi_{\mathrm{rad}}^{ap} ψ rad a p for r > 0 r>0 r > 0 . Since (17) gives
λ 1 p ( t ) Q ( λ ( t ) r ) → c ∞ r − 1 p , (219) \lambda^{\frac{1}{p}}(t)Q(\lambda(t)r)\to c_{\infty}r^{-\frac{1}{p}}, \tag*{(219)} λ p 1 ( t ) Q ( λ ( t ) r ) → c ∞ r − p 1 , (219)
after translation to the prescribed centers { x j } j = 1 K \{x_j\}_{j=1}^{K} { x j } j = 1 K , we define
ζ ∗ ( x ) = ∑ j = 1 K ( f 0 ( ∣ x − x j ∣ ) − c ∞ ∣ x − x j ∣ − 1 p ) . \zeta^*(x)=\sum_{j=1}^{K}\left(f_0(|x-x_j|)-c_{\infty}|x-x_j|^{-\frac{1}{p}}\right). ζ ∗ ( x ) = j = 1 ∑ K ( f 0 ( ∣ x − x j ∣ ) − c ∞ ∣ x − x j ∣ − p 1 ) .
Fix
s c < σ < σ ′ < s c + 4 ν 1 + 2 ν < M , s_c<\sigma<\sigma'<s_c+\frac{4\nu}{1+2\nu}<M, s c < σ < σ ′ < s c + 1 + 2 ν 4 ν < M ,
the last inequality follows by (141) . To establish strong convergence by interpolation, we first consider the main part
ψ r a d a p ( t , x ) − Θ δ ( r ) λ 1 p ( t ) Q ( λ ( t ) r ) , r = ∣ x ∣ , (220) \psi_{\mathrm{rad}}^{ap}(t,x)-\Theta_\delta(r)\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r),\qquad r=|x|, \tag*{(220)} ψ rad a p ( t , x ) − Θ δ ( r ) λ p 1 ( t ) Q ( λ ( t ) r ) , r = ∣ x ∣ , (220)
which is supported in r ≤ 2 δ r\leq2\delta r ≤ 2 δ . The expansions in Lemma 3.1 , (63) , and (82) , together with (18) , give
∣ ∇ x l ψ r a d a p ( t , x ) ∣ + λ 1 p ( t ) ∣ ∇ x l Q ( λ ( t ) r ) ∣ ≤ C δ , l r − 1 p − l , 0 ≤ l ≤ d . (221) \left|\nabla_x^l\psi_{\mathrm{rad}}^{ap}(t,x)\right|+\lambda^{\frac{1}{p}}(t)\left|\nabla_x^lQ(\lambda(t)r)\right|\leq C_{\delta,l}r^{-\frac{1}{p}-l},\qquad0\leq l\leq d. \tag*{(221)} ∇ x l ψ rad a p ( t , x ) + λ p 1 ( t ) ∇ x l Q ( λ ( t ) r ) ≤ C δ , l r − p 1 − l , 0 ≤ l ≤ d . (221)
Since d > 2 p d>\frac{2}{p} d > p 2 , (219) and (221) imply that
∥ ( ψ r a d a p ( t , x ) − Θ δ ( r ) λ 1 p ( t ) Q ( λ ( t ) r ) ) − ( f 0 ( r ) − Θ δ ( r ) c ∞ r − 1 p ) ∥ L x 2 → 0 (222) \left\|\left(\psi_{\mathrm{rad}}^{ap}(t,x)-\Theta_\delta(r)\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r)\right)-\left(f_0(r)-\Theta_\delta(r)c_{\infty}r^{-\frac{1}{p}}\right)\right\|_{L_x^2}\to0 \tag*{(222)} ( ψ rad a p ( t , x ) − Θ δ ( r ) λ p 1 ( t ) Q ( λ ( t ) r ) ) − ( f 0 ( r ) − Θ δ ( r ) c ∞ r − p 1 ) L x 2 → 0 (222)
by dominated convergence. For the remaining tail ( 1 − Θ δ ( r ) ) λ 1 p ( t ) Q ( λ ( t ) r ) (1-\Theta_\delta(r))\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r) ( 1 − Θ δ ( r )) λ p 1 ( t ) Q ( λ ( t ) r ) arising from the truncation in (220) , we estimate its difference from the limit c ∞ ( 1 − Θ δ ( r ) ) r − 1 p c_{\infty}(1-\Theta_\delta(r))r^{-\frac{1}{p}} c ∞ ( 1 − Θ δ ( r )) r − p 1 . The expansion (17) and scaling give
∥ ( 1 − Θ δ ( r ) ) ( λ 1 p ( t ) Q ( λ ( t ) r ) − c ∞ r − 1 p ) ∥ H ˙ x s ≤ C δ , s t L , s c < s ≤ d . (223) \left\|(1-\Theta_\delta(r))\left(\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r)-c_{\infty}r^{-\frac{1}{p}}\right)\right\|_{\dot H_x^s}\leq C_{\delta,s}t^L,\qquad s_c<s\leq d. \tag*{(223)} ( 1 − Θ δ ( r )) ( λ p 1 ( t ) Q ( λ ( t ) r ) − c ∞ r − p 1 ) H ˙ x s ≤ C δ , s t L , s c < s ≤ d . (223)
On the other hand, (97) and scaling imply
∥ ψ r a d a p ( t , x ) − λ 1 p ( t ) Q ( λ ( t ) r ) ∥ H ˙ x σ ′ ≤ C δ , σ ′ . \left\|\psi_{\mathrm{rad}}^{ap}(t,x)-\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r)\right\|_{\dot H_x^{\sigma'}}\leq C_{\delta,\sigma'}. ψ rad a p ( t , x ) − λ p 1 ( t ) Q ( λ ( t ) r ) H ˙ x σ ′ ≤ C δ , σ ′ .
Since ( 1 − Θ δ ( ∣ x ∣ ) ) r − 1 p ∈ H ˙ x σ ′ (1-\Theta_\delta(|x|))r^{-\frac{1}{p}}\in\dot H_x^{\sigma'} ( 1 − Θ δ ( ∣ x ∣ )) r − p 1 ∈ H ˙ x σ ′ , combining this estimate with (221) and (223) we deduce
∥ ψ r a d a p ( t , x ) − Θ δ ( r ) λ 1 p ( t ) Q ( λ ( t ) r ) ∥ H ˙ x σ ′ ≤ C δ , σ ′ . \left\|\psi_{\mathrm{rad}}^{ap}(t,x)-\Theta_\delta(r)\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r)\right\|_{\dot H_x^{\sigma'}}\leq C_{\delta,\sigma'}. ψ rad a p ( t , x ) − Θ δ ( r ) λ p 1 ( t ) Q ( λ ( t ) r ) H ˙ x σ ′ ≤ C δ , σ ′ .
By weak compactness in H σ ′ H^{\sigma'} H σ ′ , this bound and (222) imply that
f 0 ( r ) − c ∞ Θ δ ( r ) r − 1 p ∈ H σ ′ ( R d ) . f_0(r)-c_{\infty}\Theta_\delta(r)r^{-\frac{1}{p}}\in H^{\sigma'}(\mathbb{R}^d). f 0 ( r ) − c ∞ Θ δ ( r ) r − p 1 ∈ H σ ′ ( R d ) .
Interpolating (222) with the uniform H σ ′ H^{\sigma'} H σ ′ bound, we obtain
∥ ( ψ r a d a p ( t , x ) − Θ δ ( r ) λ 1 p ( t ) Q ( λ ( t ) r ) ) − ( f 0 ( r ) − Θ δ ( r ) c ∞ r − 1 p ) ∥ H ˙ x σ → 0 (224) \left\|\left(\psi_{\mathrm{rad}}^{ap}(t,x)-\Theta_\delta(r)\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r)\right)-\left(f_0(r)-\Theta_\delta(r)c_{\infty}r^{-\frac{1}{p}}\right)\right\|_{\dot H_x^\sigma}\to0 \tag*{(224)} ( ψ rad a p ( t , x ) − Θ δ ( r ) λ p 1 ( t ) Q ( λ ( t ) r ) ) − ( f 0 ( r ) − Θ δ ( r ) c ∞ r − p 1 ) H ˙ x σ → 0 (224)
For the multi-bubble solution, (3) and (132) give
∥ ζ ( t , x ) − ζ ∗ ( x ) ∥ H ˙ x σ ≤ ∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ x σ + K ∥ ψ r a d a p ( t , x ) − λ 1 p ( t ) Q ( λ ( t ) r ) − f 0 ( r ) + c ∞ ∣ x ∣ − 1 p ∥ H ˙ x σ . \|\zeta(t,x)-\zeta^*(x)\|_{\dot H_x^\sigma}\leq\|\psi(t,x)-\psi^{ap}(t,x)\|_{\dot H_x^\sigma}
+K\left\|\psi_{\mathrm{rad}}^{ap}(t,x)-\lambda^{\frac{1}{p}}(t)Q(\lambda(t)r)-f_0(r)+c_{\infty}|x|^{-\frac{1}{p}}\right\|_{\dot H_x^\sigma}. ∥ ζ ( t , x ) − ζ ∗ ( x ) ∥ H ˙ x σ ≤ ∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ x σ + K ψ rad a p ( t , x ) − λ p 1 ( t ) Q ( λ ( t ) r ) − f 0 ( r ) + c ∞ ∣ x ∣ − p 1 H ˙ x σ .
Combining (224) with (223) , and then applying (210) and (208) , we conclude that
ζ ( t , x ) ⟶ ζ ∗ ( x ) in H ˙ x σ ( R d ) , s c < σ < s c + 4 ν 1 + 2 ν . \zeta(t,x)\longrightarrow\zeta^*(x)\quad\text{in }\dot H_x^\sigma(\mathbb{R}^d),\qquad s_c<\sigma<s_c+\frac{4\nu}{1+2\nu}. ζ ( t , x ) ⟶ ζ ∗ ( x ) in H ˙ x σ ( R d ) , s c < σ < s c + 1 + 2 ν 4 ν .
We also establish the boundedness below the critical regularity. For 1 ≤ s < s c 1 \le s < s_c 1 ≤ s < s c , applying (95) at scales 2 − j δ 2^{-j}\delta 2 − j δ and using (221) , we obtain
∥ ( Θ 0 ( 2 j r δ ) − Θ 0 ( 2 j + 1 r δ ) ) ψ r a d a p ( t , x ) ∥ H ˙ x s ≤ C δ , s ( 2 − j δ ) s c − s , j ≥ 0. \left\|\left(\Theta_0\left(\frac{2^j r}{\delta}\right)-\Theta_0\left(\frac{2^{j+1}r}{\delta}\right)\right)\psi^{ap}_{rad}(t,x)\right\|_{\dot{H}^s_x} \le C_{\delta,s}(2^{-j}\delta)^{s_c-s}, \qquad j \ge0. ( Θ 0 ( δ 2 j r ) − Θ 0 ( δ 2 j + 1 r ) ) ψ r a d a p ( t , x ) H ˙ x s ≤ C δ , s ( 2 − j δ ) s c − s , j ≥ 0.
The remaining part ( 1 − Θ δ ( r ) ) ψ r a d a p ( t , x ) (1-\Theta_\delta(r))\psi^{ap}_{rad}(t,x) ( 1 − Θ δ ( r )) ψ r a d a p ( t , x ) is supported in δ ≤ r ≤ 2 δ \delta\le r \le2\delta δ ≤ r ≤ 2 δ and satisfies the same bound at scale δ \delta δ . Summing over j ≥ 0 j \ge0 j ≥ 0 , we obtain
∥ ψ r a d a p ( t , x ) ∥ H ˙ x s ≤ C δ , s δ s c − s ( 1 + ∑ j ≥ 0 2 − j ( s c − s ) ) ≤ C δ , s . \|\psi^{ap}_{rad}(t,x)\|_{\dot{H}^s_x} \le C_{\delta,s}\delta^{s_c-s}\left(1+\sum_{j\ge0}2^{-j(s_c-s)}\right)\le C_{\delta,s}. ∥ ψ r a d a p ( t , x ) ∥ H ˙ x s ≤ C δ , s δ s c − s ( 1 + j ≥ 0 ∑ 2 − j ( s c − s ) ) ≤ C δ , s .
According to (132) , translation invariance, and (211) ,
∥ ψ ( t , x ) ∥ H ˙ x s ≤ ∑ j = 1 K ∥ ψ r a d a p ( t , x − x j ) ∥ H ˙ x s + ∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ x s ≤ C δ , s , 1 ≤ s < s c . \|\psi(t,x)\|_{\dot{H}^s_x}\le\sum_{j=1}^{K}\|\psi^{ap}_{rad}(t,x-x_j)\|_{\dot{H}^s_x}+\|\psi(t,x)-\psi^{ap}(t,x)\|_{\dot{H}^s_x}\le C_{\delta,s},\qquad1\le s<s_c. ∥ ψ ( t , x ) ∥ H ˙ x s ≤ j = 1 ∑ K ∥ ψ r a d a p ( t , x − x j ) ∥ H ˙ x s + ∥ ψ ( t , x ) − ψ a p ( t , x ) ∥ H ˙ x s ≤ C δ , s , 1 ≤ s < s c .
It remains to determine the critical Sobolev norm. We first establish the L 1 L^1 L 1 bound needed to estimate the interactions between distinct bubbles. By (221) and the support property in Theorem 1.2 ,
(7.95)
∥ ψ r a d a p ( t , x ) ∥ L x 1 ≤ C δ ∫ r ≤ 2 δ r − 1 p + d − 1 d r ≤ C δ . \|\psi^{ap}_{rad}(t,x)\|_{L^1_x}\le C_\delta\int_{r\le2\delta}r^{-\frac1p+d-1}\,dr\le C_\delta. ∥ ψ r a d a p ( t , x ) ∥ L x 1 ≤ C δ ∫ r ≤ 2 δ r − p 1 + d − 1 d r ≤ C δ .
According to the definition (132) ,
ψ a p ( t , x ) = ∑ j = 1 K ψ r a d a p ( t , x − x j ) , \psi^{ap}(t,x)=\sum_{j=1}^{K}\psi^{ap}_{rad}(t,x-x_j), ψ a p ( t , x ) = j = 1 ∑ K ψ r a d a p ( t , x − x j ) ,
where ψ r a d a p ( t , x − x j ) \psi^{ap}_{rad}(t,x-x_j) ψ r a d a p ( t , x − x j ) is supported in ∣ x − x j ∣ ≤ 2 δ |x-x_j|\le2\delta ∣ x − x j ∣ ≤ 2 δ . Thus (133) implies that, for i ≠ j i\ne j i = j , the corresponding supports have distance at least ∣ x i − x j ∣ − 4 δ > 6 δ |x_i-x_j|-4\delta>6\delta ∣ x i − x j ∣ − 4 δ > 6 δ .
Since d > s c d>s_c d > s c , we apply (116) with s c 2 \frac{s_c}{2} 2 s c replaced by s c s_c s c . Using (117) , the separation of the supports, and (7.95), we obtain
(7.96)
∣ ⟨ ψ r a d a p ( t , ⋅ − x i ) , ( − Δ ) s c ψ r a d a p ( t , ⋅ − x j ) ⟩ ∣ ≤ C ∥ ψ r a d a p ( t ) ∥ L 1 2 ∫ 0 ∞ τ − d / 2 − s c − 1 e − ( 6 δ ) 2 / ( 8 τ ) d τ ≤ C δ . \left|\left\langle\psi^{ap}_{rad}(t,\cdot-x_i),(-\Delta)^{s_c}\psi^{ap}_{rad}(t,\cdot-x_j)\right\rangle\right|
\le C\|\psi^{ap}_{rad}(t)\|_{L^1}^2\int_0^\infty\tau^{-d/2-s_c-1}e^{-(6\delta)^2/(8\tau)}\,d\tau\le C_\delta. ∣ ⟨ ψ r a d a p ( t , ⋅ − x i ) , ( − Δ ) s c ψ r a d a p ( t , ⋅ − x j ) ⟩ ∣ ≤ C ∥ ψ r a d a p ( t ) ∥ L 1 2 ∫ 0 ∞ τ − d /2 − s c − 1 e − ( 6 δ ) 2 / ( 8 τ ) d τ ≤ C δ .
By (93) and translation invariance, (132) yields
(7.97)
∥ ψ a p ( t ) ∥ H ˙ s c 2 = K ∥ ψ r a d a p ( t ) ∥ H ˙ s c 2 + 2 Re ∑ 1 ≤ i < j ≤ K ⟨ ψ r a d a p ( t , ⋅ − x i ) , ( − Δ ) s c ψ r a d a p ( t , ⋅ − x j ) ⟩ = K ∥ ψ r a d a p ( t ) ∥ H ˙ s c 2 + O δ ( 1 ) , \begin{aligned}
\|\psi^{ap}(t)\|_{\dot{H}^{s_c}}^2=K\|\psi^{ap}_{rad}(t)\|_{\dot{H}^{s_c}}^2+2\operatorname{Re}\sum_{1\le i<j\le K}\left\langle\psi^{ap}_{rad}(t,\cdot-x_i),(-\Delta)^{s_c}\psi^{ap}_{rad}(t,\cdot-x_j)\right\rangle\\
&=K\|\psi^{ap}_{rad}(t)\|_{\dot{H}^{s_c}}^2+O_\delta(1),
\end{aligned} ∥ ψ a p ( t ) ∥ H ˙ s c 2 = K ∥ ψ r a d a p ( t ) ∥ H ˙ s c 2 + 2 Re 1 ≤ i < j ≤ K ∑ ⟨ ψ r a d a p ( t , ⋅ − x i ) , ( − Δ ) s c ψ r a d a p ( t , ⋅ − x j ) ⟩ = K ∥ ψ r a d a p ( t ) ∥ H ˙ s c 2 + O δ ( 1 ) ,
where the last equality follows from (7.96), since K K K is fixed. Since 1 p + s c = d 2 \frac1p+s_c=\frac d2 p 1 + s c = 2 d , critical scaling yields
∥ ψ r a d a p ( t ) ∥ H ˙ s c = λ 1 / p + s c − d / 2 ( t ) ∥ U r a d a p ( t ) ∥ H ˙ s c = ∥ U r a d a p ( t ) ∥ H ˙ s c . \|\psi^{ap}_{rad}(t)\|_{\dot{H}^{s_c}}=\lambda^{1/p+s_c-d/2}(t)\|U^{ap}_{rad}(t)\|_{\dot{H}^{s_c}}=\|U^{ap}_{rad}(t)\|_{\dot{H}^{s_c}}. ∥ ψ r a d a p ( t ) ∥ H ˙ s c = λ 1/ p + s c − d /2 ( t ) ∥ U r a d a p ( t ) ∥ H ˙ s c = ∥ U r a d a p ( t ) ∥ H ˙ s c .
Thus (7.97) implies
∥ ψ a p ( t ) ∥ H ˙ s c = K ∥ U r a d a p ( t ) ∥ H ˙ s c + O δ ( 1 ) . \|\psi^{ap}(t)\|_{\dot{H}^{s_c}}=\sqrt{K}\|U^{ap}_{rad}(t)\|_{\dot{H}^{s_c}}+O_\delta(1). ∥ ψ a p ( t ) ∥ H ˙ s c = K ∥ U r a d a p ( t ) ∥ H ˙ s c + O δ ( 1 ) .
Finally, applying (210) with s = s c s=s_c s = s c and using the fact ν ( 1 − 1 / d ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η \nu(1-1/d)(|\lambda_-|-s_c+3)>4\nu+4\eta ν ( 1 − 1/ d ) ( ∣ λ − ∣ − s c + 3 ) > 4 ν + 4 η , we obtain
∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s c ≤ C t − 2 ν + ν ( 1 − 1 / d ) ( ∣ λ − ∣ − s c + 3 ) − η / 2 , \|\psi(t)-\psi^{ap}(t)\|_{\dot{H}^{s_c}}\le Ct^{-2\nu+\nu(1-1/d)(|\lambda_-|-s_c+3)-\eta/2}, ∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s c ≤ C t − 2 ν + ν ( 1 − 1/ d ) ( ∣ λ − ∣ − s c + 3 ) − η /2 ,
thus
∣ ∥ ψ ( t ) ∥ H ˙ s c − ∥ ψ a p ( t ) ∥ H ˙ s c ∣ ≤ ∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s c \left|\|\psi(t)\|_{\dot{H}^{s_c}}-\|\psi^{ap}(t)\|_{\dot{H}^{s_c}}\right|\le\|\psi(t)-\psi^{ap}(t)\|_{\dot{H}^{s_c}} ∣ ∥ ψ ( t ) ∥ H ˙ s c − ∥ ψ a p ( t ) ∥ H ˙ s c ∣ ≤ ∥ ψ ( t ) − ψ a p ( t ) ∥ H ˙ s c
where the right-hand side tends to zero as t → 0 t \to0 t → 0 .
Combining this inequality with (7.97) and (102) , and using ∥ ψ r a d a p ( t ) ∥ H ˙ s c = ∥ U r a d a p ( t ) ∥ H ˙ s c \|\psi_{rad}^{ap}(t)\|_{\dot{H}^{s_c}}=\|U_{rad}^{ap}(t)\|_{\dot{H}^{s_c}} ∥ ψ r a d a p ( t ) ∥ H ˙ s c = ∥ U r a d a p ( t ) ∥ H ˙ s c , we obtain (5) . This completes the proof of Theorem 1.1.
References
The Institute of Geometry and Physics, University of Science and Technology of China, Hefei 230026, Anhui, China
Email address: lxy0418@ustc.edu.cn