Introduction
Most physical systems can only be observed partially: measurements are taken at a few specific times, and/or only at a discrete set of points. Schrödinger-type equations provide several examples of this:
Optical fibers. The envelope of a light pulse obeys the cubic nonlinear Schrödinger equation, in which the distance along the fiber plays the role of time and time plays the role of the spatial variable [22 , 2 ] . The pulse is accessible essentially only at the two ends of the fiber, where it is recorded by detectors sampling at discrete instants.
Bose–Einstein condensates. The macroscopic wave function of Bose–Einstein condensates obeys the Gross–Pitaevskii equation, a cubic nonlinear Schrödinger equation with an external trapping potential, which reduces to a one-dimensional equation for elongated condensates [13 , 33 ] . In this setting, standard absorption imaging destroys the condensate, so one obtains only a few snapshots of the object, each recorded on a pixel grid.
Transmission electron microscopy. The electron wave crossing a specimen is governed by a linear Schrödinger equation, in which the direction of propagation plays the role of time and the potential is the electrostatic potential of the specimen [28 ] . The transmitted wave is recorded on a pixelated detector.
In each of these situations the detectors record the intensity ∣ u ∣ 2 |u|^2 ∣ u ∣ 2 : in particular, this allows us to detect when the wave vanishes at a given point, if the sampled quantity also vanishes. A natural question is therefore how much of the solution is encoded in such partial information. In particular, a more direct version would be: must a solution that vanishes on a discrete set of points at two different times vanish identically?
With this question in mind, and influenced by the examples above, we study the initial value problem (IVP) associated to the Schrödinger equation with potential
SP (1.1)
{ ∂ t u = i ( ∂ x 2 u + V ( x , t ) u ) in R × [ 0 , 1 ] , u ( x , 0 ) = u 0 (1) \begin{cases}
\partial_t u = i(\partial_x^2 u + V(x,t)u) & \text{in } \mathbb{R} \times[0,1],\\
u(x,0)=u_0 \tag*{(1)}
\end{cases} { ∂ t u = i ( ∂ x 2 u + V ( x , t ) u ) u ( x , 0 ) = u 0 in R × [ 0 , 1 ] , (1)
where V V V is a smooth, real-valued potential satisfying some additional conditions that we specify below. We will also consider the cubic nonlinear Schrödinger equation
NLS (1.2)
{ ∂ t u = i ( ∂ x 2 u + λ ∣ u ∣ 2 u ) in R × [ 0 , 1 ] , u ( x , 0 ) = u 0 , (2) \begin{cases}
\partial_t u = i(\partial_x^2 u + \lambda|u|^2u) & \text{in } \mathbb{R} \times[0,1],\\
u(x,0)=u_0, \tag*{(2)}
\end{cases} { ∂ t u = i ( ∂ x 2 u + λ ∣ u ∣ 2 u ) u ( x , 0 ) = u 0 , in R × [ 0 , 1 ] , (2)
for some λ ∈ R \lambda\in\mathbb{R} λ ∈ R . We denote u 1 = u ( x , 1 ) u_1=u(x,1) u 1 = u ( x , 1 ) , and assume that
u 0 ( ± n α ) = u 1 ( ± n β ) = 0 ∀ n ∈ N (3) u_0(\pm n^\alpha)=u_1(\pm n^\beta)=0 \quad\forall n\in\mathbb{N} \tag*{(3)} u 0 ( ± n α ) = u 1 ( ± n β ) = 0 ∀ n ∈ N (3)
for 0 < α , β < 1 0<\alpha,\beta<1 0 < α , β < 1 . We want to understand under which conditions on α \alpha α , β \beta β and V V V we can deduce that u ≡ 0 u\equiv0 u ≡ 0 .
More generally, assume that, for a pair of sets ( A , B ) (A,B) ( A , B ) we have u 0 ∣ A = u 1 ∣ B = 0 u_0|_A=u_1|_B=0 u 0 ∣ A = u 1 ∣ B = 0 . Our goal is to understand which conditions these sets must satisfy in order for the whole solution to be the zero solution. We refer to such a pair of sets as a Schrödinger uniqueness pair .
The question above is directly motivated by the analogous case of the Fourier transform . Indeed, in that context, we mention first the breakthrough result by Radchenko and Viazovska [36 ] , where it was shown that there are smooth functions a n ∈ S e v e n ( R ) a_n\in S_{\mathrm{even}}(\mathbb{R}) a n ∈ S even ( R ) such that, for a given f ∈ S e v e n ( R ) f\in S_{\mathrm{even}}(\mathbb{R}) f ∈ S even ( R ) , we have
f ( x ) = ∑ n ≥ 0 ( f ( n ) a n ( x ) + f ^ ( n ) a n ^ ( x ) ) . f(x)=\sum_{n\geq0}\left(f(\sqrt{n})a_n(x)+\widehat{f}(\sqrt{n})\widehat{a_n}(x)\right). f ( x ) = n ≥ 0 ∑ ( f ( n ) a n ( x ) + f ( n ) a n ( x ) ) .
Here, the Fourier transform is taken to have the normalization
f ^ ( ξ ) = ∫ R f ( x ) e − 2 π i ξ x d x . \widehat{f}(\xi)=\int_{\mathbb{R}}f(x)e^{-2\pi i\xi x}\,dx. f ( ξ ) = ∫ R f ( x ) e − 2 π i ξ x d x .
The first instance in which the general question of whether a function f ∈ S ( R ) f\in S(\mathbb{R}) f ∈ S ( R ) such that
f ( ± n α ) = f ^ ( ± n β ) = 0 , ∀ n ∈ N , f(\pm n^\alpha)=\widehat{f}(\pm n^\beta)=0,\quad\forall n\in\mathbb{N}, f ( ± n α ) = f ( ± n β ) = 0 , ∀ n ∈ N ,
is automatically f ≡ 0 f\equiv0 f ≡ 0 was posed in a paper by the third named author with M. Sousa [37 ] . They proved that the conclusion holds for a certain subset of the lower triangle of exponents
{ ( α , β ) ∈ ( 0 , 1 ) 2 : 0 < α + β < 1 } . \{(\alpha,\beta)\in(0,1)^2:0<\alpha+\beta<1\}. {( α , β ) ∈ ( 0 , 1 ) 2 : 0 < α + β < 1 } .
Subsequently, in [30 ] , A. Kulikov, F. Nazarov and M. Sodin showed optimal thresholds for the question above. Effectively, they answered this question by introducing the notion of supercritical and subcritical pairs of sets. Indeed, they define a pair of sets ( λ n ) n ∈ Z (\lambda_n)_{n\in\mathbb{Z}} ( λ n ) n ∈ Z and ( μ n ) n ∈ Z (\mu_n)_{n\in\mathbb{Z}} ( μ n ) n ∈ Z to be supercritical if
lim ∣ n ∣ → + ∞ sup ∣ λ n ∣ p − 1 ( λ n + 1 − λ n ) < 1 2 , lim ∣ n ∣ → + ∞ sup ∣ μ n ∣ q − 1 ( μ n + 1 − μ n ) < 1 2 , 1 p + 1 q = 1 , \lim_{|n|\to+\infty}\sup|\lambda_n|^{p-1}(\lambda_{n+1}-\lambda_n)<\frac{1}{2},\qquad
\lim_{|n|\to+\infty}\sup|\mu_n|^{q-1}(\mu_{n+1}-\mu_n)<\frac{1}{2},\qquad
\frac{1}{p}+\frac{1}{q}=1, ∣ n ∣ → + ∞ lim sup ∣ λ n ∣ p − 1 ( λ n + 1 − λ n ) < 2 1 , ∣ n ∣ → + ∞ lim sup ∣ μ n ∣ q − 1 ( μ n + 1 − μ n ) < 2 1 , p 1 + q 1 = 1 ,
and subcritical if
lim ∣ n ∣ → + ∞ inf ∣ λ n ∣ p − 1 ( λ n + 1 − λ n ) > 1 2 , lim ∣ n ∣ → + ∞ inf ∣ μ n ∣ q − 1 ( μ n + 1 − μ n ) > 1 2 , 1 p + 1 q = 1. \lim_{|n|\to+\infty}\inf|\lambda_n|^{p-1}(\lambda_{n+1}-\lambda_n)>\frac{1}{2},\qquad
\lim_{|n|\to+\infty}\inf|\mu_n|^{q-1}(\mu_{n+1}-\mu_n)>\frac{1}{2},\qquad
\frac{1}{p}+\frac{1}{q}=1. ∣ n ∣ → + ∞ lim inf ∣ λ n ∣ p − 1 ( λ n + 1 − λ n ) > 2 1 , ∣ n ∣ → + ∞ lim inf ∣ μ n ∣ q − 1 ( μ n + 1 − μ n ) > 2 1 , p 1 + q 1 = 1.
The main result of [30 ] is then that any supercritical pair is a uniqueness pair, and that any subcritical pair is a non-uniqueness pair. This result completed the program started in [37 ] when it comes to showing that the Fourier uniqueness property holds in the whole lower triangle, since the pair ( ± n α , ± n β ) (\pm n^\alpha,\pm n^\beta) ( ± n α , ± n β ) is supercritical exactly when 0 < α + β < 1 0<\alpha+\beta<1 0 < α + β < 1 . We also refer the reader to [3 , 42 , 40 , 11 , 1 ] for Fourier uniqueness and interpolation results in higher dimensions; to [29 , 6 , 31 ] for necessary density conditions and for uniqueness pairs in the critical regime, such as those arising from the zeros of the Riemann zeta function and other L L L -functions; to [38 , 9 ] for perturbations of interpolation formulas and the regularity of their basis functions; to [35 , 4 , 7 ] for constructions of non-uniqueness sets; to [39 , 32 ] for phaseless (Pauli-type) analogues of these questions; and to [34 ] for the closely related theory of Heisenberg uniqueness pairs.
With our goal of understanding the analogous question for Schrödinger-like equations, we first start by noticing that a solution to the free Schrödinger equation,
{ ∂ t u = i Δ u u ( x , 0 ) = u 0 , \left\{
\begin{aligned}
\partial_tu&=i\Delta u\\
u(x,0)&=u_0,
\end{aligned}
\right. { ∂ t u u ( x , 0 ) = i Δ u = u 0 ,
can be written as u ( x , t ) = c t ( e i ∣ ⋅ ∣ 2 / 4 t u 0 ) x ^ ( x / 2 t ) u(x,t)=c_t\left(e^{i|\cdot|^2/4t}u_0\right)^{\widehat{\phantom{x}}}(x/2t) u ( x , t ) = c t ( e i ∣ ⋅ ∣ 2 /4 t u 0 ) x ( x /2 t ) , so that at any fixed time t t t , we may relate the solution to the Fourier transform of (a suitable modulation of) the initial data. This promptly implies that if ( A , B ) (A,B) ( A , B ) is a Fourier uniqueness pair , in the sense that any reasonable f f f such that
f ∣ A = f ^ ∣ B ≡ 0 f|_A=\widehat{f}|_B\equiv0 f ∣ A = f ∣ B ≡ 0
is automatically f = 0 f=0 f = 0 , then any reasonable solution to the free Schrödinger equation above such that u 0 ∣ A = u 1 ∣ B = 0 u_0|_A=u_1|_B=0 u 0 ∣ A = u 1 ∣ B = 0 satisfies that u = 0 u=0 u = 0 . Since this is true for the free Schrödinger equation, it is natural to ask the question if this also holds for the solution to (1.1) and (1.2). In those cases, however, it is not immediately clear how to proceed: the relationship between the solutions to such equations and an integral transform are not always as clear as the one between the free Schrödinger equation and the Fourier transform, and new ideas and techniques seem to be needed.
Indeed, progress in our main question has been restricted to [27 ] , where the authors proved that the sets ( ± c 1 log ( 1 + n ) α , ± c 2 log ( 1 + n ) α ) (\pm c_1\log(1+n)^\alpha,\pm c_2\log(1+n)^\alpha) ( ± c 1 log ( 1 + n ) α , ± c 2 log ( 1 + n ) α ) , for α ∈ ( 0 , 1 / 2 ) \alpha\in(0,1/2) α ∈ ( 0 , 1/2 ) , form a Schrödinger uniqueness pair for both (1.1), under suitable conditions on V V V , and (1.2). In that setting, the techniques rely heavily on complex analysis.
As previously mentioned, our main task is to find a suitable subset of the lower triangle of exponents where one has that ( ± n α , ± n β ) (\pm n^\alpha,\pm n^\beta) ( ± n α , ± n β ) are Schrödinger uniqueness pairs for either (1.1) or (1.2). Our first result is, hence, an improvement when one has certain (almost) analyticity conditions on the potential V V V .
Theorem 1.1. Let V ∈ L ∞ ( [ 0 , 1 ] ; C b ∞ ( R ) ) V\in L^\infty([0,1];C^\infty_b(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] ; C b ∞ ( R )) and assume that there exists σ ≥ 0 \sigma\ge0 σ ≥ 0 such that, for all l ∈ N l\in\mathbb{N} l ∈ N ,
∥ ∂ x l V ∥ L t , x ∞ < A l + 1 ( l ! ) σ , \|\partial_x^l V\|_{L^\infty_{t,x}}<A^{l+1}(l!)^\sigma, ∥ ∂ x l V ∥ L t , x ∞ < A l + 1 ( l ! ) σ , together with
lim R → + ∞ ∫ 0 1 sup ∣ x ∣ ≥ R ∣ V ( x , t ) ∣ d t = 0. \lim_{R\to+\infty}\int_0^1\sup_{|x|\ge R}|V(x,t)|\,dt=0. R → + ∞ lim ∫ 0 1 ∣ x ∣ ≥ R sup ∣ V ( x , t ) ∣ d t = 0. Let u 0 ∈ H 1 : = H 1 ( R ) ∩ L 2 ( x 2 d x ) u_0\in\mathcal{H}^1:=H^1(\mathbb{R})\cap L^2(x^2dx) u 0 ∈ H 1 := H 1 ( R ) ∩ L 2 ( x 2 d x ) and u u u be the solution to equation (1.1). Moreover assume that ( α , β ) ∈ D σ ⊂ { ( α , β ) ∈ ( 0 , 1 2 ) 2 : α + β < 1 } (\alpha,\beta)\in D_\sigma\subset\{(\alpha,\beta)\in(0,\frac12)^2:\alpha+\beta<1\} ( α , β ) ∈ D σ ⊂ {( α , β ) ∈ ( 0 , 2 1 ) 2 : α + β < 1 } ,
u 0 ( ± n α ) = u 1 ( ± n β ) = 0 ∀ n ∈ N , u_0(\pm n^\alpha)=u_1(\pm n^\beta)=0\quad\forall n\in\mathbb{N}, u 0 ( ± n α ) = u 1 ( ± n β ) = 0 ∀ n ∈ N , where, in the case 0 ≤ σ ≤ 1 0\le\sigma\le1 0 ≤ σ ≤ 1 ,
D σ = { ( α , β ) ∈ ( 0 , 1 2 ) 2 : α 1 − α + β 1 − β < 1 } , D_\sigma=\left\{(\alpha,\beta)\in\left(0,\frac12\right)^2:\frac{\alpha}{1-\alpha}+\frac{\beta}{1-\beta}<1\right\}, D σ = { ( α , β ) ∈ ( 0 , 2 1 ) 2 : 1 − α α + 1 − β β < 1 } , and, in the case σ > 1 \sigma>1 σ > 1 ,
D σ = { ( α , β ) ∈ ( 0 , 1 2 ) 2 : α 1 − α + β 1 − β < 1 σ } . D_\sigma=\left\{(\alpha,\beta)\in\left(0,\frac12\right)^2:\frac{\alpha}{1-\alpha}+\frac{\beta}{1-\beta}<\frac{1}{\sigma}\right\}. D σ = { ( α , β ) ∈ ( 0 , 2 1 ) 2 : 1 − α α + 1 − β β < σ 1 } . Then u ≡ 0 u\equiv0 u ≡ 0 .
Remark 1.1. Condition (1.4) means that V V V is uniformly of Gevrey class G σ G^\sigma G σ in the sense of Section 2.2 (see [26 ] page 281 and [41 ] Section 1.4 for the classical, local Gevrey classes). More precisely, we can say the following:
in the case σ = 0 \sigma=0 σ = 0 , condition (1.4) is satisfied by band-limited potentials: if supp V ^ ⊂ [ − Ω , Ω ] \operatorname{supp}\widehat V\subset[-\Omega,\Omega] supp V ⊂ [ − Ω , Ω ] and V ^ ∈ L 1 ( R ) \widehat V\in L^1(\mathbb{R}) V ∈ L 1 ( R ) , then ∥ ∂ x l V ∥ L ∞ ≲ ( 2 π Ω ) l ∥ V ^ ∥ L 1 \|\partial_x^lV\|_{L^\infty}\lesssim(2\pi\Omega)^l\|\widehat V\|_{L^1} ∥ ∂ x l V ∥ L ∞ ≲ ( 2 π Ω ) l ∥ V ∥ L 1 ; more generally, if V ∈ L ∞ ( R ) V\in L^\infty(\mathbb{R}) V ∈ L ∞ ( R ) and supp V ^ ⊂ [ − Ω , Ω ] \operatorname{supp}\widehat V\subset[-\Omega,\Omega] supp V ⊂ [ − Ω , Ω ] , then ∥ ∂ x l V ∥ L ∞ ≲ ( 2 π Ω ) l ∥ V ∥ L ∞ \|\partial_x^lV\|_{L^\infty}\lesssim(2\pi\Omega)^l\|V\|_{L^\infty} ∥ ∂ x l V ∥ L ∞ ≲ ( 2 π Ω ) l ∥ V ∥ L ∞ by the Paley–Wiener–Schwartz theorem Theorem 7.3.1 and Bernstein’s inequality [5 ] ;
in the case σ = 1 2 \sigma=\frac12 σ = 2 1 , condition (1.4) is satisfied by potentials whose Fourier transform has Gaussian decay, ∣ V ^ ( ξ ) ∣ ≤ c e − τ ∣ ξ ∣ 2 |\widehat V(\xi)|\le ce^{-\tau|\xi|^2} ∣ V ( ξ ) ∣ ≤ c e − τ ∣ ξ ∣ 2 (see Lemma 2.1 ), such as V ( x ) = e − τ x 2 V(x)=e^{-\tau x^2} V ( x ) = e − τ x 2 ;
in the case σ = 1 \sigma=1 σ = 1 , condition (1.4) holds for some A > 0 A>0 A > 0 if and only if V V V extends to a bounded holomorphic function on some strip { z : ∣ I m z ∣ ≤ δ } \{z:|{\rm Im}\,z|\le\delta\} { z : ∣ Im z ∣ ≤ δ } , δ > 0 \delta>0 δ > 0 ; an example is V ( x ) = ( 1 + x 2 ) − 1 V(x)=(1+x^2)^{-1} V ( x ) = ( 1 + x 2 ) − 1 ;
for any σ > 1 \sigma>1 σ > 1 , there are smooth compactly supported functions which belong to G σ G^\sigma G σ (see Proposition 1.4.10 in [41 ] and Example (iv) in Subsection 2.2 for a concrete example in our setting).
Our next result deals with the case of the cubic nonlinear Schrödinger equation (1.2):
Theorem 1.2. Let λ ∈ R \lambda\in\mathbb{R} λ ∈ R and u 0 ∈ H 1 : = H 1 ( R ) ∩ L 2 ( x 2 d x ) u_0\in\mathcal{H}^1:=H^1(\mathbb{R})\cap L^2(x^2dx) u 0 ∈ H 1 := H 1 ( R ) ∩ L 2 ( x 2 d x ) . Let u u u be the solution to (1.2). Assume that u u u satisfies the zero-condition
u 0 ( ± n α ) = u 1 ( ± n β ) = 0 ∀ n ∈ N , u_0(\pm n^\alpha)=u_1(\pm n^\beta)=0\quad\forall n\in\mathbb{N}, u 0 ( ± n α ) = u 1 ( ± n β ) = 0 ∀ n ∈ N , for ( α , β ) ∈ D 1 (\alpha,\beta)\in D_1 ( α , β ) ∈ D 1 , where D 1 D_1 D 1 is defined in (1.6). Then it follows that u ≡ 0 u\equiv0 u ≡ 0 .
Figure 1. The set D σ D_{\sigma} D σ , in the σ = 1 \sigma=1 σ = 1 case.
Strategy of the proof
Our first step in the proof is showing, from the fact that u 0 ∈ H 1 u_0 \in\mathcal{H}^1 u 0 ∈ H 1 and the vanishing conditions (1.3), that the solution u u u enjoys more regularity; more precisely, we show that u ∈ C ( [ 0 , 1 ] : S ( R ) ) u \in C([0,1] : \mathcal{S}(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) . The proof of this result builds on the results from [30 ] together with the fact that if a solution to (1.2) decays at two different times u 0 , u 1 ∈ L 2 ( x 2 k d x ) u_0,u_1 \in L^2(x^{2k}dx) u 0 , u 1 ∈ L 2 ( x 2 k d x ) , then one has both regularity and decay for all intermediate times. This latter fact relies on arguments developed in [23 , 24 , 25 , 27 ] .
As a matter of fact, this result may be proved in the whole lower triangle exponent range, which is synthesized by the following results:
Theorem 1.3. Assume that V ∈ L ∞ ( [ 0 , 1 ] : C b ∞ ( R ) ) V \in L^{\infty}([0,1] : C^\infty_b(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b ∞ ( R )) . Let u 0 ∈ H 1 : = H 1 ( R ) ∩ L 2 ( x 2 d x ) u_0 \in\mathcal{H}^1 := H^1(\mathbb{R}) \cap L^2(x^2dx) u 0 ∈ H 1 := H 1 ( R ) ∩ L 2 ( x 2 d x ) and let u u u be the solution to the equation (1.1). Moreover, assume that, for α , β ∈ ( 0 , 1 ) \alpha,\beta\in(0,1) α , β ∈ ( 0 , 1 ) with α + β < 1 \alpha+\beta<1 α + β < 1 ,
u 0 ( ± n α ) = u 1 ( ∓ n β ) = 0. (4) u_0(\pm n^\alpha)=u_1(\mp n^\beta)=0. \tag*{(4)} u 0 ( ± n α ) = u 1 ( ∓ n β ) = 0. (4) Then u ∈ C ( [ 0 , 1 ] : S ( R ) ) u \in C([0,1] : \mathcal{S}(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) .
Remark 1.2. Note that in this result we only require C ∞ C^\infty C ∞ smoothness and boundedness of the potential: V ∈ L ∞ ( [ 0 , 1 ] : C b ∞ ( R ) ) V \in L^\infty([0,1] : C^\infty_b(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b ∞ ( R )) . We note furthermore, as already mentioned above, that the result actually holds in the entirety of the lower triangle.
Theorem 1.4. Let λ ∈ R \lambda\in\mathbb{R} λ ∈ R and u 0 ∈ H 1 : = H 1 ( R ) ∩ L 2 ( x 2 d x ) u_0 \in\mathcal{H}^1 := H^1(\mathbb{R}) \cap L^2(x^2dx) u 0 ∈ H 1 := H 1 ( R ) ∩ L 2 ( x 2 d x ) and let u u u be the solution to (1.2). Moreover, assume that, for α , β ∈ ( 0 , 1 ) \alpha,\beta\in(0,1) α , β ∈ ( 0 , 1 ) with α + β < 1 \alpha+\beta<1 α + β < 1 , (1.7) holds. Then u ∈ C ( [ 0 , 1 ] : S ( R ) ) u \in C([0,1] : \mathcal{S}(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) .
The next step in both proofs is then showing that the zeroes actually give more information about the decay at two times. We will prove that both
∫ R e c ∣ x ∣ p ∣ u 0 ∣ 2 d x and ∫ R e c ∣ x ∣ q ∣ u 1 ∣ 2 d x , \int_{\mathbb{R}} e^{c|x|^{p}}|u_0|^2dx \quad\text{and}\quad\int_{\mathbb{R}} e^{c|x|^{q}}|u_1|^2dx, ∫ R e c ∣ x ∣ p ∣ u 0 ∣ 2 d x and ∫ R e c ∣ x ∣ q ∣ u 1 ∣ 2 d x ,
are finite for 1 p + 1 q = 1 \frac{1}{p}+\frac{1}{q}=1 p 1 + q 1 = 1 , and any c > 0 c>0 c > 0 . In order to do so, we must actually control the order of growth of norms of the form
∥ x k u 0 ∥ L 2 , k = 1 , 2 , … . \|x^k u_0\|_{L^2},\ k=1,2,\ldots. ∥ x k u 0 ∥ L 2 , k = 1 , 2 , … .
This follows essentially the same basic structure as the main theorems from [37 , 38 , 30 ] : one seeks to prove a bound on such quantities in terms of lower-order quantities of the same kind. In order to do so, one wants to relate this quantity to some certain dual quantity, which in our case corresponds to bounding norms of
Γ t j u ( t ) , where Γ t = ( x + 2 i t ∂ x ) . \Gamma_t^j u(t),\ \text{where }\Gamma_t=(x+2it\partial_x). Γ t j u ( t ) , where Γ t = ( x + 2 i t ∂ x ) .
This is where the main techniques here differ from those in [27 ] : indeed, in spite of also using the operator Γ t \Gamma_t Γ t , our main idea in the case of both (1.1) and (1.2) is separating the main contribution of the action of $\Gamma_t^k on the Duhammel expansion of u u u , from the lower-order ones. In the potential case, this is compensated by the Gevrey control on the potential. In the cubic nonlinear case, the special structure of the nonlinearity allows for us to leave that contribution isolated, and induct on the order.
This refined comparison estimate (see (4.4) and (5.13)) is what enables us to achieve reasonable energy estimates, in contrast with what one obtains by crudely using Grönwall in order to estimate the energy of Γ t j \Gamma_t^j Γ t j , as done in the proof of Theorem 1.3. This improvement effectively means enabling the usage of the techniques from [37 , 30 ] .
Finally, the conclusion follows by applying the Morgan principle for the Schrödinger equation with potential. We remark that the precise version below is simply one of the many statements present in the otherwise vast theory of Hardy-like uncertainty principles for Schrödinger equations, developed by L. Escauriaza, C. Kenig, G. Ponce and L. Vega in a series of papers [14 , 15 , 16 , 17 , 12 , 18 , 19 ] ; see also their survey [20 ] .
Theorem 1.5. [18 ] Given p ∈ ( 0 , 1 ) p \in(0,1) p ∈ ( 0 , 1 ) , there exists N p > 0 N_p > 0 N p > 0 such that if u ∈ C ( [ 0 , 1 ] : L 2 ( R n ) ) u \in C([0,1] : L^2(\mathbb{R}^n)) u ∈ C ([ 0 , 1 ] : L 2 ( R n )) is a solution of
∂ t u = i ( Δ u + V ( x , t ) u ) , \partial_t u = i(\Delta u + V(x,t)u), ∂ t u = i ( Δ u + V ( x , t ) u ) , with V = V ( x , t ) V = V(x,t) V = V ( x , t ) complexed valued, ∥ V ∥ L ∞ ( R n × [ 0 , 1 ] ) < ∞ \|V\|_{L^\infty(\mathbb{R}^n \times[0,1])} < \infty ∥ V ∥ L ∞ ( R n × [ 0 , 1 ]) < ∞ and
lim R → + ∞ ∫ 0 1 sup ∣ x ∣ > R ∣ V ( x , t ) ∣ d t = 0 , \lim_{R \to+\infty} \int_0^1 \sup_{|x|>R} |V(x,t)|\,dt = 0, R → + ∞ lim ∫ 0 1 ∣ x ∣ > R sup ∣ V ( x , t ) ∣ d t = 0 , and there exists α , β > 0 \alpha,\beta> 0 α , β > 0 such that
∫ R n ∣ u ( x , 0 ) ∣ 2 e 2 α p ∣ x ∣ p d x + ∫ R n ∣ u ( x , 1 ) ∣ 2 e 2 β q ∣ x ∣ q d x < ∞ , \int_{\mathbb{R}^n} |u(x,0)|^2 e^{2\alpha p |x|^p}\,dx + \int_{\mathbb{R}^n} |u(x,1)|^2 e^{2\beta q |x|^q}\,dx < \infty, ∫ R n ∣ u ( x , 0 ) ∣ 2 e 2 α p ∣ x ∣ p d x + ∫ R n ∣ u ( x , 1 ) ∣ 2 e 2 β q ∣ x ∣ q d x < ∞ , 1 / p + 1 / q = 1 1/p + 1/q = 1 1/ p + 1/ q = 1 , with α β > N p \alpha\beta> N_p α β > N p , then u = 0 u = 0 u = 0 .
Remark 1.3. We note that the choice of nonlinearity in Theorem 1.2 is due mostly to the fact that 2 is the first even integer. Indeed, by following the exact same proof steps, executing the necessary modifications along the way, one is able to obtain the following:
Corollary 1.1. Let u 0 ∈ H 1 u_0 \in\mathcal{H}^1 u 0 ∈ H 1 , and let u u u be a solution to
{ ∂ t u = i ( ∂ x 2 u + λ ∣ u ∣ 2 k u ) in R × [ 0 , 1 ] , u ( x , 0 ) = u 0 . (5) \begin{cases}
\partial_t u = i(\partial_x^2 u + \lambda|u|^{2k}u) & \text{in } \mathbb{R} \times[0,1],\\
u(x,0) = u_0. \tag*{(5)}
\end{cases} { ∂ t u = i ( ∂ x 2 u + λ ∣ u ∣ 2 k u ) u ( x , 0 ) = u 0 . in R × [ 0 , 1 ] , (5) Assume that u u u satisfies the zero-condition (1.3) for ( α , β ) ∈ D 1 (\alpha,\beta) \in D_1 ( α , β ) ∈ D 1 . Then it follows that u = 0 u = 0 u = 0 .
In particular, the proof does not rely on the integrability of the cubic nonlinear Schrödinger equation in dimension one in any way. In the same spirit, we also note that, by carefully redoing the same proof, one may go much further than mere powers of integers, establishing thus the following result with a gap condition, in similar terms as to that of [30 ] :
Corollary 1.2. Let ( λ j ) j ∈ Z (\lambda_j)_{j\in\mathbb{Z}} ( λ j ) j ∈ Z and ( γ j ) j ∈ Z (\gamma_j)_{j\in\mathbb{Z}} ( γ j ) j ∈ Z be strictly increasing sequences of real numbers such that λ j → + ∞ \lambda_j \to+\infty λ j → + ∞ and γ j → + ∞ \gamma_j \to+\infty γ j → + ∞ as j → + ∞ j \to+\infty j → + ∞ . Assume that there exist p , q > 1 p,q > 1 p , q > 1 such that
lim sup ∣ j ∣ → ∞ ∣ λ j ∣ p − 1 ( λ j + 1 − λ j ) < ∞ and lim sup ∣ j ∣ → ∞ ∣ γ j ∣ q − 1 ( γ j + 1 − γ j ) < ∞ . (6) \limsup_{|j|\to\infty} |\lambda_j|^{p-1}(\lambda_{j+1}-\lambda_j) < \infty\quad\text{and} \quad\limsup_{|j|\to\infty} |\gamma_j|^{q-1}(\gamma_{j+1}-\gamma_j) < \infty. \tag*{(6)} ∣ j ∣ → ∞ lim sup ∣ λ j ∣ p − 1 ( λ j + 1 − λ j ) < ∞ and ∣ j ∣ → ∞ lim sup ∣ γ j ∣ q − 1 ( γ j + 1 − γ j ) < ∞. (6) Let u 0 ∈ H 1 u_0 \in\mathcal{H}^1 u 0 ∈ H 1 , let u u u be the corresponding solution of (1.1) or of (1.2). Assume that
u 0 ( λ j ) = u 1 ( γ j ) = 0 for all j ∈ Z . (7) u_0(\lambda_j)=u_1(\gamma_j)=0 \quad\text{for all } j\in\mathbb{Z}. \tag*{(7)} u 0 ( λ j ) = u 1 ( γ j ) = 0 for all j ∈ Z . (7) (i) Suppose that u u u solves (1.1), where V V V is as in the statement of Theorem 1.2. If
1 p − 1 + 1 q − 1 < 1 max { 1 , σ } , \frac{1}{p-1}+\frac{1}{q-1}<\frac{1}{\max\{1,\sigma\}}, p − 1 1 + q − 1 1 < max { 1 , σ } 1 , then $u \equiv 0.
(ii) Suppose that u u u solves (2) , for any real value of the coefficient of the nonlinearity. If
1 p − 1 + 1 q − 1 < 1 , \frac{1}{p-1}+\frac{1}{q-1}<1, p − 1 1 + q − 1 1 < 1 ,
then u ≡ 0 u\equiv0 u ≡ 0 .
We have, however, chosen to keep the power of integer case, as well as the cubic NLS case, for simplicity of exposition.
The paper is organized as follows. In Section 2, we recall Stirling’s formula, describe the uniform Gårding class G σ G^\sigma G σ , and recall some identities related to the vector field Γ t k \Gamma_t^k Γ t k . In Section 3, we explain how an initial condition in H 1 H^1 H 1 together with the zero condition at two different times imply that the solution lies in the Schwartz class for all time; in short, we give the proof of Theorems 1.3 and 1.4. Finally, Section 4 is dedicated to the proof of the unique continuation theorem in the case of the potential (Theorem 1.1), while Section 5 is devoted to the proof of the unique continuation theorem in the case of the nonlinear Schrödinger equation.
Notation
Throughout the text, c c c , c β c_{\beta} c β , c α c_{\alpha} c α will denote arbitrary positive constants, only depending on α \alpha α and β \beta β , and that may change from line to line. In Section 3, the constants c c c may also depend on the indexes k , j , l k,j,l k , j , l involved in the proof, since the growth of these do not matter at this point. However, in Sections 4 and 5, the constants c c c only depend on α \alpha α and β \beta β , and we keep track carefully on the growth of the indexes k , j , l k,j,l k , j , l .
Acknowledgements
J. P. G. R. was supported by the FCT through project SHADE (2023.17881.ICDT), DOI 10.54499/2023.17881.ICDT. He was also supported by FAPERJ through JCNE grant no. SEI-260003/020475/2025, by Instituto Serrapilheira through grant Serra-R-2510-59765, and by CNPq through Research Productivity Fellowship no. 303861/2026-6. T.S.J. and D.P. were partially supported by the Trond Mohn Foundation. They also would like to thank the University of Lisbon and the Instituto de Matemática Pura e Aplicada (IMPA) for the kind hospitality during different stages of this work.
LLM Usage.
All the ideas and arguments, as well as their technical details, were conceived and verified by the authors without using artificial intelligence tools. In particular, in Sections 3 and 4, LLM were only used to check computations and language. A preliminary working version of Section 5 was typeset by using several Large Language Models, but the present final version was completely written by the authors. Large Language Models were only used in the final stages for looking for typos and searching references.
Preliminaries
We recall the definition of the Gamma function1
Γ ( t ) = ∫ 0 + ∞ e − x x t − 1 d x , t > 0. \Gamma(t)=\int_{0}^{+\infty}e^{-x}x^{t-1}\,dx,\qquad t>0. Γ ( t ) = ∫ 0 + ∞ e − x x t − 1 d x , t > 0.
For integers k ≥ 0 k\ge0 k ≥ 0 ,
Γ ( k + 1 ) = k ! . \Gamma(k+1)=k!. Γ ( k + 1 ) = k ! .
Moreover, the Stirling formula states that
Γ ( t + 1 ) = t t e − t 2 π t ( 1 + O ( 1 t ) ) , as t → + ∞ . (8) \Gamma(t+1)=t^{t}e^{-t}\sqrt{2\pi t}\left(1+O\left(\frac{1}{t}\right)\right),\qquad\text{as }t\to+\infty. \tag*{(8)} Γ ( t + 1 ) = t t e − t 2 π t ( 1 + O ( t 1 ) ) , as t → + ∞. (8)
In particular,
k ! = k k e − k 2 π k ( 1 + O ( 1 k ) ) , as k → + ∞ . (9) k!=k^{k}e^{-k}\sqrt{2\pi k}\left(1+O\left(\frac{1}{k}\right)\right),\qquad\text{as }k\to+\infty. \tag*{(9)} k ! = k k e − k 2 π k ( 1 + O ( k 1 ) ) , as k → + ∞. (9)
As a consequence, we will often use the simpler identities k ! = e k log k + O ( k ) k!=e^{k\log k+O(k)} k ! = e k l o g k + O ( k ) and Γ ( t ) = e t log t + O ( t ) \Gamma(t)=e^{t\log t+O(t)} Γ ( t ) = e t l o g t + O ( t ) , which provide the useful bounds
k ! ≤ k k ≤ e k k ! . (10) k!\le k^{k}\le e^{k}k!. \tag*{(10)} k ! ≤ k k ≤ e k k ! . (10)
Gevrey Spaces
We recall the definition of Gevrey functions of order σ ≥ 0 \sigma\ge0 σ ≥ 0 , (see [26 ] and Chapter I in [41 ] ),
{ V ∈ C ∞ ( R ) : ∣ ∂ x j f ( x ) ∣ ≤ c K j + 1 ( j ! ) σ , ∀ x ∈ K , j ∈ N } \left\{ V \in C^{\infty}(\mathbb{R}) : |\partial_x^j f(x)| \le c_K^{j+1}(j!)^\sigma,\quad\forall x \in K,\ j \in\mathbb{N} \right\} { V ∈ C ∞ ( R ) : ∣ ∂ x j f ( x ) ∣ ≤ c K j + 1 ( j ! ) σ , ∀ x ∈ K , j ∈ N }
for any compact K ⊂ R K \subset\mathbb{R} K ⊂ R , and c K c_K c K is a positive constant depending on K K K .
In this paper, we work with a stronger assumption on the potential. We say that V V V is uniformly of Gevrey class σ ≥ 0 \sigma\ge0 σ ≥ 0 if there exists A > 0 A > 0 A > 0 such that
sup x ∈ R ∣ ∂ x j V ( x ) ∣ ≤ A j + 1 ( j ! ) σ , ∀ j ∈ N . \sup_{x \in\mathbb{R}} |\partial_x^j V(x)| \le A^{j+1}(j!)^\sigma,\ \forall j \in\mathbb{N}. x ∈ R sup ∣ ∂ x j V ( x ) ∣ ≤ A j + 1 ( j ! ) σ , ∀ j ∈ N .
If V V V is uniformly of Gevrey class σ \sigma σ , we denote V ∈ G σ = G σ ( R ) V \in G^\sigma= G^\sigma(\mathbb{R}) V ∈ G σ = G σ ( R ) . Note that the notation G σ G^\sigma G σ differs from [26 , 41 ] , since it means uniformly of Gevrey class on R \mathbb{R} R in our case.
In the following, we derive some useful properties of the spaces G σ ( R ) G^\sigma(\mathbb{R}) G σ ( R ) and give some important examples. The rest of this subsection is largely inspired by Chapter I in [41 ] .
Observe that G σ 1 ⊂ G σ 2 G^{\sigma_1} \subset G^{\sigma_2} G σ 1 ⊂ G σ 2 if σ 1 ≤ σ 2 \sigma_1 \le\sigma_2 σ 1 ≤ σ 2 .
We start with a characterization of the spaces G σ G^\sigma G σ by using the Fourier transform.
Lemma 2.1. Suppose that there exist s > 0 s > 0 s > 0 , τ > 0 \tau> 0 τ > 0 and c > 0 c > 0 c > 0 such that ∣ f ^ ( ξ ) ∣ ≤ c e − τ ∣ ξ ∣ s |\hat f(\xi)| \le c e^{-\tau|\xi|^s} ∣ f ^ ( ξ ) ∣ ≤ c e − τ ∣ ξ ∣ s , then f ∈ G σ f \in G^\sigma f ∈ G σ , for σ = 1 / s \sigma= 1/s σ = 1/ s . In particular, the Gaussian e − τ x 2 e^{-\tau x^2} e − τ x 2 is uniformly of Gevrey class of order σ = 1 / 2 \sigma= 1/2 σ = 1/2 .
Proof. Observe that
∥ ∂ x j f ∥ L ∞ ≤ ∫ R ∣ ξ ∣ j f ^ ( ξ ) d ξ ≤ ∫ R ∣ ξ ∣ j c e − τ ∣ ξ ∣ s d ξ . \|\partial_x^j f\|_{L^\infty} \le\int_{\mathbb{R}} |\xi|^j \hat f(\xi)d\xi\le\int_{\mathbb{R}} |\xi|^j c e^{-\tau|\xi|^s}d\xi. ∥ ∂ x j f ∥ L ∞ ≤ ∫ R ∣ ξ ∣ j f ^ ( ξ ) d ξ ≤ ∫ R ∣ ξ ∣ j c e − τ ∣ ξ ∣ s d ξ .
We consider the integral
I j , s = ∫ 0 + ∞ r j e − τ r s d r . I_{j,s} = \int_0^{+\infty} r^j e^{-\tau r^s}\,dr. I j , s = ∫ 0 + ∞ r j e − τ r s d r .
By doing the change of variable r = ( t τ ) 1 / β r = \left(\frac{t}{\tau}\right)^{1/\beta} r = ( τ t ) 1/ β , it follows that
I j , s = s − 1 τ − ( j + 1 ) / s ∫ 0 + ∞ t ( j + 1 ) / s − 1 e − t d t = s − 1 τ − ( j + 1 ) / s Γ ( j + 1 s ) ≤ c s , τ j e j s log j + O ( j ) ≤ c s , τ j ( j ! ) 1 / s . I_{j,s} = s^{-1}\tau^{-(j+1)/s}\int_0^{+\infty} t^{(j+1)/s-1}e^{-t}\,dt = s^{-1}\tau^{-(j+1)/s}\Gamma\left(\frac{j+1}{s}\right) \le c_{s,\tau}^j e^{\frac{j}{s}\log j+O(j)} \le c_{s,\tau}^j(j!)^{1/s}. I j , s = s − 1 τ − ( j + 1 ) / s ∫ 0 + ∞ t ( j + 1 ) / s − 1 e − t d t = s − 1 τ − ( j + 1 ) / s Γ ( s j + 1 ) ≤ c s , τ j e s j l o g j + O ( j ) ≤ c s , τ j ( j ! ) 1/ s .
by using the Stirling formula for the Γ \Gamma Γ function in (8) .
Thus, if ∣ f ^ ( ξ ) ∣ ≤ c e − τ ∣ ξ ∣ s |\hat f(\xi)| \le c e^{-\tau|\xi|^s} ∣ f ^ ( ξ ) ∣ ≤ c e − τ ∣ ξ ∣ s , then ∥ ∂ x j f ∥ L ∞ ≲ c s , τ j ( j ! ) 1 / s \|\partial_x^j f\|_{L^\infty} \lesssim c_{s,\tau}^j(j!)^{1/s} ∥ ∂ x j f ∥ L ∞ ≲ c s , τ j ( j ! ) 1/ s , which means that f ∈ G 1 / s f \in G^{1/s} f ∈ G 1/ s .
Examples of uniformly Gevrey potentials depending on σ ≥ 0 \sigma\ge0 σ ≥ 0 .
(i) σ = 0 \sigma= 0 σ = 0 . If V ∈ L ∞ ( R ) V \in L^\infty(\mathbb{R}) V ∈ L ∞ ( R ) is band-limited, i.e. supp V ^ ⊂ [ − A , A ] \operatorname{supp}\hat V \subset[-A,A] supp V ^ ⊂ [ − A , A ] , for some A > 0 A > 0 A > 0 , then V ∈ G 0 V \in G^0 V ∈ G 0 .
(ii) 1 / 2 ≤ σ < 1 1/2 \le\sigma< 1 1/2 ≤ σ < 1 . Let m ≥ 1 m \ge1 m ≥ 1 be an integer, let τ > 0 \tau> 0 τ > 0 , and let V ( x ) = e − τ x 2 m V(x) = e^{-\tau x^{2m}} V ( x ) = e − τ x 2 m . Then V V V extends to an entire function which satisfies
∣ V ( x + i y ) ∣ ≤ e − τ 2 x 2 m + C m τ ∣ y ∣ 2 m , x , y ∈ R , |V(x+iy)| \le e^{-\frac{\tau}{2}x^{2m}+C_m\tau|y|^{2m}},\qquad x,y \in\mathbb{R}, ∣ V ( x + i y ) ∣ ≤ e − 2 τ x 2 m + C m τ ∣ y ∣ 2 m , x , y ∈ R ,
for some constant C m > 0 C_m > 0 C m > 0 depending only on m m m . By the characterization of the Gelfand–Shilov spaces in terms of entire extensions [21 Chapter IV] , this means that V V V belongs to the Gelfand–Shilov space S 1 2 m 1 − 1 2 m S^{1-\frac{1}{2m}}_{\frac{1}{2m}} S 2 m 1 1 − 2 m 1 . Since the Fourier transform maps S α β S^\beta_\alpha S α β onto S β α S^\alpha_\beta S β α (see [21 Chapter IV] and [10 ] ), it follows that
∣ V ^ ( ξ ) ∣ ≤ c e − τ ~ ∣ ξ ∣ 2 m 2 m − 1 . |\hat V(\xi)| \le c e^{-\tilde{\tau}|\xi|^{\frac{2m}{2m-1}}}. ∣ V ^ ( ξ ) ∣ ≤ c e − τ ~ ∣ ξ ∣ 2 m − 1 2 m .
for some $c,\tilde{\tau} > 0 (see also [8 ] for the precise asymptotic behaviour of V ^ \hat V V ^ ). Lemma 2.1 then implies that V ∈ G 2 m − 1 2 m V \in G^{\frac{2m-1}{2m}} V ∈ G 2 m 2 m − 1 . In particular, for any τ > 0 \tau> 0 τ > 0 , e − τ x 2 ∈ G 1 / 2 e^{-\tau x^2} \in G^{1/2} e − τ x 2 ∈ G 1/2 and e − τ x 4 ∈ G 3 / 4 e^{-\tau x^4} \in G^{3/4} e − τ x 4 ∈ G 3/4 .
(iii) σ = 1 \sigma= 1 σ = 1 . The class G 1 G^1 G 1 corresponds to analytic functions on R \mathbb{R} R that can be extended holomorphically at least on the strip { z : ∣ Im ( z ) ∣ < 1 / A } \{z : |\operatorname{Im}(z)| < 1/A\} { z : ∣ Im ( z ) ∣ < 1/ A } . A good example is the Poisson kernel P ( x ) = 1 1 + x 2 P(x) = \frac{1}{1+x^2} P ( x ) = 1 + x 2 1 whose Fourier transform is P ^ ( ξ ) = e − 2 π ∣ ξ ∣ \hat P(\xi) = e^{-2\pi|\xi|} P ^ ( ξ ) = e − 2 π ∣ ξ ∣ . So, it follows from Lemma 2.1 that P ∈ G 1 P \in G^1 P ∈ G 1 .
(iv) σ > 1 \sigma> 1 σ > 1 . We will show that there exist compactly supported potentials belonging to G σ G^\sigma G σ , σ > 1 \sigma> 1 σ > 1 .
For any s > 0 s>0 s > 0 , let us define
V ( x ) = { e − 1 ( 1 − x 2 ) s ∣ x ∣ < 1 , 0 ∣ x ∣ ≥ 1. V(x)=
\begin{cases}
e^{-\frac{1}{(1-x^2)^s}} & |x|<1,\\
0 & |x|\ge1.
\end{cases} V ( x ) = { e − ( 1 − x 2 ) s 1 0 ∣ x ∣ < 1 , ∣ x ∣ ≥ 1.
Then, it is well-known that V ∈ C c ∞ ( R ) V\in C^\infty_c(\mathbb{R}) V ∈ C c ∞ ( R ) . Moreover, we claim that ∥ ∂ x j V ∥ L ∞ ≤ c ( j ! ) 1 + 1 s \|\partial_x^j V\|_{L^\infty}\le c(j!)^{1+\frac{1}{s}} ∥ ∂ x j V ∥ L ∞ ≤ c ( j ! ) 1 + s 1 , so that V ∈ G 1 + 1 s V\in G^{1+\frac{1}{s}} V ∈ G 1 + s 1 . Of course this does not work for G 1 G^1 G 1 , since there are no nonzero analytic compactly supported functions. To show the claim, we compute the derivatives: V ′ ( x ) = − e − 1 ( 1 − x 2 ) s 2 x s ( 1 − x 2 ) s + 1 V'(x)=-e^{-\frac{1}{(1-x^2)^s}}\frac{2xs}{(1-x^2)^{s+1}} V ′ ( x ) = − e − ( 1 − x 2 ) s 1 ( 1 − x 2 ) s + 1 2 x s . Since ∣ x ∣ < 1 |x|<1 ∣ x ∣ < 1 in supp V \operatorname{supp} V supp V , and by denoting t = 1 − x 2 ∈ [ 0 , 1 ] t=1-x^2\in[0,1] t = 1 − x 2 ∈ [ 0 , 1 ] , it follows that
∣ V ′ ( x ) ∣ ≤ e − 1 ( 1 − x 2 ) s 2 s ( 1 − x 2 ) s + 1 ≤ c e − 1 / t s t − ( s + 1 ) . |V'(x)|\le e^{-\frac{1}{(1-x^2)^s}}\frac{2s}{(1-x^2)^{s+1}}\le ce^{-1/t^s}t^{-(s+1)}. ∣ V ′ ( x ) ∣ ≤ e − ( 1 − x 2 ) s 1 ( 1 − x 2 ) s + 1 2 s ≤ c e − 1/ t s t − ( s + 1 ) .
Similarly, we can show that ∣ V ( j ) ( x ) ∣ ≤ c j e − 1 / t s t − ( s + 1 ) j ≤ c j e − 1 2 s ( s + 1 ) j log t |V^{(j)}(x)|\le c^j e^{-1/t^s}t^{-(s+1)j}\le c^j e^{-\frac{1}{2s}(s+1)j\log t} ∣ V ( j ) ( x ) ∣ ≤ c j e − 1/ t s t − ( s + 1 ) j ≤ c j e − 2 s 1 ( s + 1 ) j l o g t for any j ≥ 1 j\ge1 j ≥ 1 . Moreover,
d d t ( − 1 t s − ( s + 1 ) j log t ) = s t s + 1 − ( s + 1 ) j 1 t = 0 ⟺ t s = s ( s + 1 ) j , \frac{d}{dt}\left(-\frac{1}{t^s}-(s+1)j\log t\right)=\frac{s}{t^{s+1}}-(s+1)j\frac{1}{t}=0\Longleftrightarrow t^s=\frac{s}{(s+1)j}, d t d ( − t s 1 − ( s + 1 ) j log t ) = t s + 1 s − ( s + 1 ) j t 1 = 0 ⟺ t s = ( s + 1 ) j s ,
and the function f ( t ) = − 1 t s − ( s + 1 ) j log t f(t)=-\frac{1}{t^s}-(s+1)j\log t f ( t ) = − t s 1 − ( s + 1 ) j log t attains its maximum for t = ( s ( s + 1 ) j ) 1 / s t=\left(\frac{s}{(s+1)j}\right)^{1/s} t = ( ( s + 1 ) j s ) 1/ s . It follows that
∣ V ( j ) ( x ) ∣ ≤ c j e − ( s + 1 ) j s e − ( s + 1 ) j log ( s ( s + 1 ) j ) 1 / s ≤ c j e ( s + 1 ) s j log j ≤ c j ( j ! ) 1 + 1 s , |V^{(j)}(x)|\le c^j e^{-\frac{(s+1)j}{s}}e^{-(s+1)j\log\left(\frac{s}{(s+1)j}\right)^{1/s}}\le c^j e^{\frac{(s+1)}{s}j\log j}\le c^j(j!)^{1+\frac{1}{s}}, ∣ V ( j ) ( x ) ∣ ≤ c j e − s ( s + 1 ) j e − ( s + 1 ) j l o g ( ( s + 1 ) j s ) 1/ s ≤ c j e s ( s + 1 ) j l o g j ≤ c j ( j ! ) 1 + s 1 ,
which shows that V ∈ G 1 + 1 s V\in G^{1+\frac{1}{s}} V ∈ G 1 + s 1 .
The operator t
We recall from [23 , 24 , 25 ] the operator
Γ t j = ( x + 2 i t ∂ x ) j . (11) \Gamma_t^j=(x+2it\partial_x)^j. \tag*{(11)} Γ t j = ( x + 2 i t ∂ x ) j . (11)
It is known that for j ∈ N n j\in\mathbb{N}^n j ∈ N n and t ≠ 0 t\ne0 t = 0 ,
Γ t j u ( t ) = e i ∣ x ∣ 2 / 4 t ( 2 i t ) ∣ j ∣ ∂ x j ( e − i ∣ x ∣ 2 / 4 t u ( t ) ) , (12) \Gamma_t^j u(t)=e^{i|x|^2/4t}(2it)^{|j|}\partial_x^j\left(e^{-i|x|^2/4t}u(t)\right), \tag*{(12)} Γ t j u ( t ) = e i ∣ x ∣ 2 /4 t ( 2 i t ) ∣ j ∣ ∂ x j ( e − i ∣ x ∣ 2 /4 t u ( t ) ) , (12)
and that
Γ t j f = e i t ∂ x 2 ( x j e − i t ∂ x 2 f ) . (13) \Gamma_t^j f=e^{it\partial_x^2}\left(x^j e^{-it\partial_x^2}f\right). \tag*{(13)} Γ t j f = e i t ∂ x 2 ( x j e − i t ∂ x 2 f ) . (13)
We let, for t ≠ 0 t\ne0 t = 0 , u ~ ( x , t ) = e − i ∣ x ∣ 2 / 4 t u ( x , t ) \widetilde u(x,t)=e^{-i|x|^2/4t}u(x,t) u ( x , t ) = e − i ∣ x ∣ 2 /4 t u ( x , t ) , so that
Γ t j u = e i ∣ x ∣ 2 / 4 t ( 2 i t ) ∣ j ∣ ∂ x j u ~ and ∥ Γ t j u ∥ L 2 = ( 2 t ) ∣ j ∣ ∥ ∂ x j u ~ ∥ L 2 . (14) \Gamma_t^j u=e^{i|x|^2/4t}(2it)^{|j|}\partial_x^j\widetilde u\quad\text{and}\quad\|\Gamma_t^j u\|_{L^2}=(2t)^{|j|}\|\partial_x^j\widetilde u\|_{L^2}. \tag*{(14)} Γ t j u = e i ∣ x ∣ 2 /4 t ( 2 i t ) ∣ j ∣ ∂ x j u and ∥ Γ t j u ∥ L 2 = ( 2 t ) ∣ j ∣ ∥ ∂ x j u ∥ L 2 . (14)
Moreover Γ t j \Gamma_t^j Γ t j commutes with the free Schrödinger flow, that is
Γ t j ( i ∂ t + Δ ) = ( i ∂ t + Δ ) Γ t j , (15) \Gamma_t^j(i\partial_t+\Delta)=(i\partial_t+\Delta)\Gamma_t^j, \tag*{(15)} Γ t j ( i ∂ t + Δ ) = ( i ∂ t + Δ ) Γ t j , (15)
and if u u u is a solution to (1.1), then Γ t j u \Gamma_t^j u Γ t j u is a solution to the IVP
{ ∂ t Γ t j u = i ( ∂ x 2 Γ t j u + Γ t j ( V u ) ) , Γ 0 j u ( x , 0 ) = x j u 0 . (16) \begin{cases}
\partial_t\Gamma_t^j u=i(\partial_x^2\Gamma_t^j u+\Gamma_t^j(Vu)),\\
\Gamma_0^j u(x,0)=x^j u_0. \tag*{(16)}
\end{cases} { ∂ t Γ t j u = i ( ∂ x 2 Γ t j u + Γ t j ( V u )) , Γ 0 j u ( x , 0 ) = x j u 0 . (16)
Since the Schrödinger flow is reversible in time, the solution at time t = 0 t=0 t = 0 , u 0 u_0 u 0 , can be recovered from the data at time t = 1 t=1 t = 1 . We can therefore write u 0 u_0 u 0 as the solution obtained by evolving u 1 u_1 u 1 backwards through the same equation. More precisely, if we let v ( x , t ) = u ( x , 1 − t ) v(x,t)=u(x,1-t) v ( x , t ) = u ( x , 1 − t ) , then
{ ∂ t v = − i ( ∂ x 2 v + V ( x , 1 − t ) v ) , v ( x , 0 ) = u 1 , \begin{cases}
\partial_t v=-i(\partial_x^2v+V(x,1-t)v),\\
v(x,0)=u_1,
\end{cases} { ∂ t v = − i ( ∂ x 2 v + V ( x , 1 − t ) v ) , v ( x , 0 ) = u 1 ,
and u 0 = v ( x , 1 ) = v 1 u_0=v(x,1)=v_1 u 0 = v ( x , 1 ) = v 1 . Then we can work with the operators
Γ − t j = ( x − 2 i t ∂ x ) j , \Gamma_{-t}^j=(x-2it\partial_x)^j, Γ − t j = ( x − 2 i t ∂ x ) j ,
so that
Γ − t j v ( t ) = e − i ∣ x ∣ 2 / 4 t ( − 2 i t ) ∣ j ∣ ∂ x j ( e i ∣ x ∣ 2 / 4 t v ( t ) ) . (17) \Gamma_{-t}^jv(t)=e^{-i|x|^2/4t}(-2it)^{|j|}\partial_x^j\left(e^{i|x|^2/4t}v(t)\right). \tag*{(17)} Γ − t j v ( t ) = e − i ∣ x ∣ 2 /4 t ( − 2 i t ) ∣ j ∣ ∂ x j ( e i ∣ x ∣ 2 /4 t v ( t ) ) . (17)
and
Γ − t j f = e − i t ∂ x 2 ( x j e i t ∂ x 2 f ) . (18) \Gamma^j_{-t}f=e^{-it\partial_x^2}\left(x^j e^{it\partial_x^2}f\right). \tag*{(18)} Γ − t j f = e − i t ∂ x 2 ( x j e i t ∂ x 2 f ) . (18)
We define for t ≠ 0 t\ne0 t = 0 v ~ : = e + i ∣ x ∣ 2 / 4 t v ( t ) \widetilde v:=e^{+i|x|^2/4t}v(t) v := e + i ∣ x ∣ 2 /4 t v ( t ) , so that v ~ − 1 = e i x 2 / 4 v 1 = e i x 2 / 4 u 0 \widetilde v_{-1}=e^{ix^2/4}v_1=e^{ix^2/4}u_0 v − 1 = e i x 2 /4 v 1 = e i x 2 /4 u 0 . Then
Γ − t j v = e − i ∣ x ∣ 2 / 4 t ( − 2 i t ) ∣ j ∣ ∂ x j v ~ , ∥ Γ − t j v ∥ L 2 = ( 2 t ) ∣ j ∣ ∥ ∂ x j v ~ ∥ L 2 . \Gamma^j_{-t}v=e^{-i|x|^2/4t}(-2it)^{|j|}\partial_x^j\widetilde v,\qquad\|\Gamma^j_{-t}v\|_{L^2}=(2t)^{|j|}\| \partial_x^j\widetilde v\|_{L^2}. Γ − t j v = e − i ∣ x ∣ 2 /4 t ( − 2 i t ) ∣ j ∣ ∂ x j v , ∥ Γ − t j v ∥ L 2 = ( 2 t ) ∣ j ∣ ∥ ∂ x j v ∥ L 2 .
Moreover, Γ − t j \Gamma^j_{-t} Γ − t j commutes with the “backwards” Schrödinger flow, so that Γ − t j v ( t ) \Gamma^j_{-t}v(t) Γ − t j v ( t ) satisfies the IVP
{ ∂ t Γ − t j v = − i ( ∂ x 2 Γ − t j v + Γ − t j ( V ( ⋅ , 1 − t ) v ) ) Γ 0 j v ( x , 0 ) = x j v 0 = x j u 1 . (19) \begin{cases}
\partial_t\Gamma^j_{-t}v=-i(\partial_x^2\Gamma^j_{-t}v+\Gamma^j_{-t}(V(\cdot,1-t)v))\\
\Gamma^j_0v(x,0)=x^jv_0=x^ju_1. \tag*{(19)}
\end{cases} { ∂ t Γ − t j v = − i ( ∂ x 2 Γ − t j v + Γ − t j ( V ( ⋅ , 1 − t ) v )) Γ 0 j v ( x , 0 ) = x j v 0 = x j u 1 . (19)
From H 1 \mathcal{H}^1 H 1 to S ( R ) \mathcal{S}(\mathbb{R}) S ( R ) .
For k ∈ N k\in\mathbb{N} k ∈ N we define
H k : = H k ( R ) ∩ L 2 ( R ; x 2 k d x ) , \mathcal{H}^k:=H^k(\mathbb{R})\cap L^2(\mathbb{R};x^{2k}dx), H k := H k ( R ) ∩ L 2 ( R ; x 2 k d x ) ,
which is a Banach space with the norm ∥ u ∥ H k 2 = ∑ l = 0 k ( ∥ ∂ x l u ∥ L 2 2 + ∥ x l u ∥ L 2 2 ) \|u\|_{\mathcal{H}^k}^2=\sum_{l=0}^k(\|\partial_x^l u\|_{L^2}^2+\|x^l u\|_{L^2}^2) ∥ u ∥ H k 2 = ∑ l = 0 k ( ∥ ∂ x l u ∥ L 2 2 + ∥ x l u ∥ L 2 2 ) . We recall the following estimate whose proof can be found in Lemma 4.3 in [27 ] .
∑ l + j ≤ k ∥ x l ∂ x j f ∥ L 2 2 ≲ ∑ l , j ≤ k ( ∥ x l f ∥ L 2 2 + ∥ ∂ x j f ∥ L 2 2 ) . (20) \sum_{l+j\leq k}\|x^l\partial_x^j f\|_{L^2}^2\lesssim\sum_{l,j\leq k}(\|x^l f\|_{L^2}^2+\|\partial_x^j f\|_{L^2}^2). \tag*{(20)} l + j ≤ k ∑ ∥ x l ∂ x j f ∥ L 2 2 ≲ l , j ≤ k ∑ ( ∥ x l f ∥ L 2 2 + ∥ ∂ x j f ∥ L 2 2 ) . (20)
The above estimate shows that one may use the equivalent norm ∑ l + j ≤ k ∥ x l ∂ x j u ∥ L 2 \sum_{l+j\leq k}\|x^l\partial_x^j u\|_{L^2} ∑ l + j ≤ k ∥ x l ∂ x j u ∥ L 2 on H k \mathcal{H}^k H k .
We also define
C b k ( R ) : = { V ∈ C k ( R : R ) : sup x ∈ R ∣ V ( j ) ( x ) ∣ < ∞ , ∀ j ≤ k } C_b^k(\mathbb{R}):=\left\{V\in C^k(\mathbb{R}:\mathbb{R}):\sup_{x\in\mathbb{R}}|V^{(j)}(x)|<\infty,\ \forall j\leq k\right\} C b k ( R ) := { V ∈ C k ( R : R ) : x ∈ R sup ∣ V ( j ) ( x ) ∣ < ∞ , ∀ j ≤ k }
and
C b ∞ ( R ) : = { f ∈ C ∞ ( R : R ) : sup x ∈ R ∣ V ( j ) ( x ) ∣ < ∞ , ∀ j ∈ N } . C_b^\infty(\mathbb{R}):=\left\{f\in C^\infty(\mathbb{R}:\mathbb{R}):\sup_{x\in\mathbb{R}}|V^{(j)}(x)|<\infty,\ \forall j\in\mathbb{N}\right\}. C b ∞ ( R ) := { f ∈ C ∞ ( R : R ) : x ∈ R sup ∣ V ( j ) ( x ) ∣ < ∞ , ∀ j ∈ N } .
We start by stating a well-posedness result for the initial value problem (1.1) in H k ( R ) \mathcal{H}^k(\mathbb{R}) H k ( R ) when V ∈ L ∞ ( [ 0 , 1 ] : C b k ( R ) ) V\in L^\infty([0,1]:C_b^k(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b k ( R )) .
Proposition 3.1. Assume that V ∈ L ∞ ( [ 0 , 1 ] : C b k ( R ) ) V\in L^\infty([0,1]:C_b^k(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b k ( R )) and let k ≥ 1 k\geq1 k ≥ 1 . Then, for any initial datum u 0 ∈ H k u_0\in\mathcal{H}^k u 0 ∈ H k , there exists a unique solution to (1.1) in C ( [ 0 , 1 ] : H k ) C([0,1]:\mathcal{H}^k) C ([ 0 , 1 ] : H k ) . Moreover, the flow map u 0 ∈ H k ↦ u ∈ C ( [ 0 , 1 ] : H k ) u_0\in\mathcal{H}^k\mapsto u\in C([0,1]:\mathcal{H}^k) u 0 ∈ H k ↦ u ∈ C ([ 0 , 1 ] : H k ) is Lipschitz.
Proof. The proof is a classical application of the fixed point theorem, and follows from the arguments in the proof Theorem 1 in [25 ] .
Since S ( R ) = ⋂ k = 1 + ∞ H k ( R ) \mathcal{S}(\mathbb{R})=\bigcap_{k=1}^{+\infty}\mathcal{H}^k(\mathbb{R}) S ( R ) = ⋂ k = 1 + ∞ H k ( R ) , we get the following corollary.
Corollary 3.1. Assume that V ∈ L ∞ ( [ 0 , 1 ] : C b ∞ ( R ) ) V\in L^\infty([0,1]:C_b^\infty(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b ∞ ( R )) . Then, for any initial data u 0 ∈ S ( R ) u_0\in\mathcal{S}(\mathbb{R}) u 0 ∈ S ( R ) , there exists a unique solution to (1.1) in C ( [ 0 , 1 ] : S ( R ) ) C([0,1]:\mathcal{S}(\mathbb{R})) C ([ 0 , 1 ] : S ( R )) . Moreover, the flow map u 0 ∈ S ( R ) ↦ u ∈ C ( [ 0 , 1 ] : S ( R ) ) u_0\in\mathcal{S}(\mathbb{R})\mapsto u\in C([0,1]:\mathcal{S}(\mathbb{R})) u 0 ∈ S ( R ) ↦ u ∈ C ([ 0 , 1 ] : S ( R )) is Lipschitz.
Now, we state a similar well-posedness result in H k \mathcal{H}^k H k for the nonlinear Schrödinger equation (1.2).
Proposition 3.2. Let k ≥ 1 k\geq1 k ≥ 1 . For any initial datum u 0 ∈ H k u_0\in\mathcal{H}^k u 0 ∈ H k , there exists a unique solution u u u to (1.2) in C ( [ 0 , 1 ] : H k ) C([0,1]:\mathcal{H}^k) C ([ 0 , 1 ] : H k ) satisfying
sup 0 ≤ t ≤ 1 ∥ u ( t ) ∥ H k ≤ c ∥ u 0 ∥ H k . (21) \sup_{0\leq t\leq1}\|u(t)\|_{\mathcal{H}^k}\leq c\|u_0\|_{\mathcal{H}^k}. \tag*{(21)} 0 ≤ t ≤ 1 sup ∥ u ( t ) ∥ H k ≤ c ∥ u 0 ∥ H k . (21) Moreover, the flow map u 0 ∈ H k ↦ u ∈ C ( [ 0 , 1 ] : H k ) u_0\in\mathcal{H}^k\mapsto u\in C([0,1]:\mathcal{H}^k) u 0 ∈ H k ↦ u ∈ C ([ 0 , 1 ] : H k ) is Lipschitz.
Proof. The proof is a classical application of the fixed point theorem on the Duhamel formulation associated to (1.2), following the arguments in the proof of Theorem 1 in [25 ] . In order to extend the existence time to [ 0 , 1 ] [0,1] [ 0 , 1 ] , we use tamed estimates on the Duhamel formulation of (1.2) to obtain a priori estimates on the ∥ u ( t ) ∥ H k \|u(t)\|_{\mathcal{H}^k} ∥ u ( t ) ∥ H k -norm on a time existence depending only on the ∥ u ( t ) ∥ H 1 \|u(t)\|_{\mathcal{H}^1} ∥ u ( t ) ∥ H 1 norm, and then we use the H 1 \mathcal{H}^1 H 1 control on the solutions obtained in Theorem 2.1 in [23 ] to conclude. □ \square □
As a corollary, we state the well-known well-posedness result for (1.2) in S ( R ) S(\mathbb{R}) S ( R ) ).
Corollary 3.2. For any initial data u 0 ∈ S ( R ) u_0 \in S(\mathbb{R}) u 0 ∈ S ( R ) , there exists a unique solution to (1.2) in C ( [ 0 , 1 ] : S ( R ) ) C([0,1] : S(\mathbb{R})) C ([ 0 , 1 ] : S ( R )) . Moreover, the flow map : u 0 ∈ S ( R ) ↦ u ∈ C ( [ 0 , 1 ] : S ( R ) ) u_0 \in S(\mathbb{R}) \mapsto u \in C([0,1] : S(\mathbb{R})) u 0 ∈ S ( R ) ↦ u ∈ C ([ 0 , 1 ] : S ( R )) is Lipschitz.
As a consequence of Proposition 3.1, we prove that the decay of a solution at two different times imply regularity and decay for all times in between.
Proposition 3.3. Assume that V ∈ L ∞ ( [ 0 , 1 ] : C b k ( R ) ) V \in L^\infty([0,1] : C_b^k(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b k ( R )) . Let u 0 ∈ H 1 u_0 \in\mathcal{H}^1 u 0 ∈ H 1 and let u u u be the solution u ∈ C ( [ 0 , 1 ] : H 1 ) u \in C([0,1] : \mathcal{H}^1) u ∈ C ([ 0 , 1 ] : H 1 ) obtained in Proposition 3.1. Moreover, assume that u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) u_0,u_1 = u(\cdot,1) \in L^2(x^{2k}dx) u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) for some k ∈ N k \in\mathbb{N} k ∈ N . Then u ∈ C ( [ 0 , 1 ] : H k ( R ) ) u \in C([0,1] : \mathcal{H}^k(\mathbb{R})) u ∈ C ([ 0 , 1 ] : H k ( R )) .
Proof. Let { u 0 m } m \{u_0^m\}_m { u 0 m } m be a sequence of Schwartz functions such that u 0 m → u 0 u_0^m \to u_0 u 0 m → u 0 in H 1 ∩ L 2 ( x 2 k , d x ) \mathcal{H}^1 \cap L^2(x^{2k},dx) H 1 ∩ L 2 ( x 2 k , d x ) . Let { u m } ⊂ C [ 0 , 1 ] : S ( R ) \{u^m\} \subset C[0,1] : S(\mathbb{R}) { u m } ⊂ C [ 0 , 1 ] : S ( R ) be the sequence of the corresponding solutions obtained in Corollary 3.1. By Proposition 3.1 it follows that u m → u u^m \to u u m → u in C ( [ 0 , 1 ] : H 1 ) ⊂ C ( [ 0 , 1 ] : L 2 ( R ) ) C([0,1] : \mathcal{H}^1) \subset C([0,1] : L^2(\mathbb{R})) C ([ 0 , 1 ] : H 1 ) ⊂ C ([ 0 , 1 ] : L 2 ( R )) .
Now Γ t k u m \Gamma_t^k u^m Γ t k u m is well defined, but we need to make sense of Γ t k u \Gamma_t^k u Γ t k u . The first step is to prove that Γ t k \Gamma_t^k Γ t k can be defined as
Γ t k u : = lim m → + ∞ Γ t k , m u m \Gamma_t^k u := \lim_{m \to+\infty} \Gamma_t^{k,m}u^m Γ t k u := m → + ∞ lim Γ t k , m u m
and that Γ t k u ∈ C ( [ 0 , 1 ] : L 2 ( R ) ) \Gamma_t^k u \in C([0,1] : L^2(\mathbb{R})) Γ t k u ∈ C ([ 0 , 1 ] : L 2 ( R )) .
By making an energy estimate, we have
1 2 d d t ∥ Γ t k u m ( t ) ∥ L 2 2 = Re ∫ R ∂ t Γ t k u m ( t ) Γ t k u m ( t ) ‾ d x = Re ( i ∫ R ∂ x 2 k u m ( t ) Γ t k u m ( t ) ‾ d x ) + Re ( i ∫ R Γ t k ( V u m ( t ) ) Γ t k u m ( t ) ‾ d x ) = − Im ( ∫ R Γ t k ( V u m ( t ) ) Γ t k u m ( t ) ‾ d x ) = − Im ( ( 2 t ) 2 k ∫ R ∂ x k ( V u ~ m ( t ) ) ∂ x k u ~ m ( t ) ‾ d x ) = − ∑ l = 0 k − 1 ( k l ) Im ( ( 2 t ) 2 k ∫ R ∂ x k − l V ∂ x l u ~ m ( t ) ∂ x k u ~ m ( t ) ‾ d x ) + Im ∫ R V ∣ ∂ x k u ~ m ( t ) ∣ 2 d x = − ∑ l = 0 k − 1 ( k l ) Im ( ( 2 t ) 2 k ∫ R ∂ x k − l V ∂ x l u ~ m ( t ) ∂ x k u ~ m ( t ) ‾ d x ) ≤ c ( k ) sup 0 ≤ l ≤ k ∥ ∂ x l V ∥ L ∞ ∑ l = 0 k − 1 ∥ ∂ x l u ~ m ∥ L 2 ∥ ∂ x k u ~ m ∥ L 2 , (22) \begin{aligned}
\frac{1}{2}\frac{d}{dt}\|\Gamma_t^k u^m(t)\|_{L^2}^2
&= \operatorname{Re}\int_{\mathbb{R}}\partial_t\Gamma_t^k u^m(t)\overline{\Gamma_t^k u^m(t)}\,dx\\
&= \operatorname{Re}\left(i\int_{\mathbb{R}}\partial_x^{2k}u^m(t)\overline{\Gamma_t^k u^m(t)}\,dx\right)
+\operatorname{Re}\left(i\int_{\mathbb{R}}\Gamma_t^k(Vu^m(t))\overline{\Gamma_t^k u^m(t)}\,dx\right)\\
&= -\operatorname{Im}\left(\int_{\mathbb{R}}\Gamma_t^k(Vu^m(t))\overline{\Gamma_t^k u^m(t)}\,dx\right)\\
&= -\operatorname{Im}\left((2t)^{2k}\int_{\mathbb{R}}\partial_x^k(V\widetilde{u}^m(t))\overline{\partial_x^k\widetilde{u}^m(t)}\,dx\right)\\
&= -\sum_{l=0}^{k-1}\binom{k}{l}\operatorname{Im}\left((2t)^{2k}\int_{\mathbb{R}}\partial_x^{k-l}V\partial_x^l\widetilde{u}^m(t)\overline{\partial_x^k\widetilde{u}^m(t)}\,dx\right)
+\operatorname{Im}\int_{\mathbb{R}}V|\partial_x^k\widetilde{u}^m(t)|^2\,dx\\
&= -\sum_{l=0}^{k-1}\binom{k}{l}\operatorname{Im}\left((2t)^{2k}\int_{\mathbb{R}}\partial_x^{k-l}V\partial_x^l\widetilde{u}^m(t)\overline{\partial_x^k\widetilde{u}^m(t)}\,dx\right)\\
&\le c(k)\sup_{0\le l\le k}\|\partial_x^lV\|_{L^\infty}\sum_{l=0}^{k-1}\|\partial_x^l\widetilde{u}^m\|_{L^2}\|\partial_x^k\widetilde{u}^m\|_{L^2}, \tag*{(22)}
\end{aligned} 2 1 d t d ∥ Γ t k u m ( t ) ∥ L 2 2 = Re ∫ R ∂ t Γ t k u m ( t ) Γ t k u m ( t ) d x = Re ( i ∫ R ∂ x 2 k u m ( t ) Γ t k u m ( t ) d x ) + Re ( i ∫ R Γ t k ( V u m ( t )) Γ t k u m ( t ) d x ) = − Im ( ∫ R Γ t k ( V u m ( t )) Γ t k u m ( t ) d x ) = − Im ( ( 2 t ) 2 k ∫ R ∂ x k ( V u m ( t )) ∂ x k u m ( t ) d x ) = − l = 0 ∑ k − 1 ( l k ) Im ( ( 2 t ) 2 k ∫ R ∂ x k − l V ∂ x l u m ( t ) ∂ x k u m ( t ) d x ) + Im ∫ R V ∣ ∂ x k u m ( t ) ∣ 2 d x = − l = 0 ∑ k − 1 ( l k ) Im ( ( 2 t ) 2 k ∫ R ∂ x k − l V ∂ x l u m ( t ) ∂ x k u m ( t ) d x ) ≤ c ( k ) 0 ≤ l ≤ k sup ∥ ∂ x l V ∥ L ∞ l = 0 ∑ k − 1 ∥ ∂ x l u m ∥ L 2 ∥ ∂ x k u m ∥ L 2 , (22)
so that, by using our assumptions on V V V ,
d d t ∥ Γ t k u m ( t ) ∥ L 2 2 ≤ c ( k , V ) ∑ l = 0 k − 1 ∥ ∂ x l u ~ m ( t ) ∥ L 2 2 ≤ c ( k , V ) ( ∥ u ~ m ∥ L 2 2 + ∥ ∂ x k u ~ m ( t ) ∥ L 2 2 ) ≤ c ( k , V ) ( ∥ u 0 m ∥ L 2 2 + ∥ Γ t k u m ( t ) ∥ L 2 2 ) . \frac{d}{dt}\|\Gamma_t^k u^m(t)\|_{L^2}^2
\le c(k,V)\sum_{l=0}^{k-1}\|\partial_x^l\widetilde{u}^m(t)\|_{L^2}^2
\le c(k,V)(\|\widetilde{u}^m\|_{L^2}^2+\|\partial_x^k\widetilde{u}^m(t)\|_{L^2}^2)
\le c(k,V)(\|u_0^m\|_{L^2}^2+\|\Gamma_t^k u^m(t)\|_{L^2}^2). d t d ∥ Γ t k u m ( t ) ∥ L 2 2 ≤ c ( k , V ) l = 0 ∑ k − 1 ∥ ∂ x l u m ( t ) ∥ L 2 2 ≤ c ( k , V ) ( ∥ u m ∥ L 2 2 + ∥ ∂ x k u m ( t ) ∥ L 2 2 ) ≤ c ( k , V ) ( ∥ u 0 m ∥ L 2 2 + ∥ Γ t k u m ( t ) ∥ L 2 2 ) .
Here we used that for each 0 ≤ l ≤ k − 1 0 \le l \le k-1 0 ≤ l ≤ k − 1 we can write
∥ ∂ x l u ~ m ∥ L 2 2 = ∫ ∣ ξ ∣ ≤ 1 ξ 2 l ∣ u ~ m ∣ 2 d ξ + ∫ ∣ ξ ∣ ≥ 1 ξ 2 l ∣ u ~ m ∣ 2 ≤ ∥ u 0 m ∥ L 2 2 + ∥ ∂ x k u ~ m ∥ L 2 2 . \|\partial_x^l\widetilde{u}^m\|_{L^2}^2
=\int_{|\xi|\le1}\xi^{2l}|\widetilde{u}^m|^2\,d\xi
+\int_{|\xi|\ge1}\xi^{2l}|\widetilde{u}^m|^2
\le\|u_0^m\|_{L^2}^2+\|\partial_x^k\widetilde{u}^m\|_{L^2}^2. ∥ ∂ x l u m ∥ L 2 2 = ∫ ∣ ξ ∣ ≤ 1 ξ 2 l ∣ u m ∣ 2 d ξ + ∫ ∣ ξ ∣ ≥ 1 ξ 2 l ∣ u m ∣ 2 ≤ ∥ u 0 m ∥ L 2 2 + ∥ ∂ x k u m ∥ L 2 2 .
Thus, by Grönwall's inequality, we get that, for all 0 ≤ t ≤ 1 0 \le t \le1 0 ≤ t ≤ 1 ,
∥ Γ t k u m ( t ) ∥ L 2 ≤ e c ( k , V ) t ( ∥ Γ 0 k u 0 m ∥ L 2 + ∥ u 0 m ∥ L 2 ) ≤ c ( k , V ) ( ∥ x k u 0 m ∥ L 2 + ∥ u 0 m ∥ L 2 ) , (23) \|\Gamma_t^k u^m(t)\|_{L^2}
\le e^{c(k,V)t}\left(\|\Gamma_0^k u_0^m\|_{L^2}+\|u_0^m\|_{L^2}\right)
\le c(k,V)\left(\|x^k u_0^m\|_{L^2}+\|u_0^m\|_{L^2}\right), \tag*{(23)} ∥ Γ t k u m ( t ) ∥ L 2 ≤ e c ( k , V ) t ( ∥ Γ 0 k u 0 m ∥ L 2 + ∥ u 0 m ∥ L 2 ) ≤ c ( k , V ) ( ∥ x k u 0 m ∥ L 2 + ∥ u 0 m ∥ L 2 ) , (23)
where the right hand side is uniformly bounded. Thus
sup m sup t ∈ [ 0 , 1 ] ∥ Γ t k u m ( t ) ∥ L 2 ≤ c ( k , V ) sup m ( ∥ x k u 0 m ∥ L 2 + ∥ u 0 m ∥ L 2 ) . (24) \sup_{m}\sup_{t\in[0,1]}\|\Gamma_t^k u^m(t)\|_{L^2}\leq c(k,V)\sup_m\left(\|x^k u_0^m\|_{L^2}+\|u_0^m\|_{L^2}\right). \tag*{(24)} m sup t ∈ [ 0 , 1 ] sup ∥ Γ t k u m ( t ) ∥ L 2 ≤ c ( k , V ) m sup ( ∥ x k u 0 m ∥ L 2 + ∥ u 0 m ∥ L 2 ) . (24)
Since u 0 m → u 0 u_0^m\to u_0 u 0 m → u 0 in H 1 ∩ L 2 ( x 2 k d x ) \mathcal{H}^1\cap L^2(x^{2k}\,dx) H 1 ∩ L 2 ( x 2 k d x ) , we know that the right hand side of (24) converges to c ( ∥ x k u 0 ∥ L 2 + ∥ u 0 ∥ L 2 ) c\left(\|x^k u_0\|_{L^2}+\|u_0\|_{L^2}\right) c ( ∥ x k u 0 ∥ L 2 + ∥ u 0 ∥ L 2 ) . Now we will prove that { Γ t k u m } m \{\Gamma_t^k u^m\}_m { Γ t k u m } m is a Cauchy sequence in C ( [ 0 , 1 ] : L 2 ( R ) ) C([0,1]:L^2(\mathbb{R})) C ([ 0 , 1 ] : L 2 ( R )) . By the same estimate as above, we have for all t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] ,
∥ Γ t k u m 1 ( t ) − Γ t k u m 2 ( t ) ∥ L 2 ≤ c ( k , V ) ( ∥ x k ( u 0 m 1 − u 0 m 2 ) ∥ L 2 + ∥ u 0 m 1 − u 0 m 2 ∥ L 2 ) ⟶ m 1 , m 2 → + ∞ 0. \|\Gamma_t^k u^{m_1}(t)-\Gamma_t^k u^{m_2}(t)\|_{L^2}\leq c(k,V)\left(\|x^k(u_0^{m_1}-u_0^{m_2})\|_{L^2}+\|u_0^{m_1}-u_0^{m_2}\|_{L^2}\right)\underset{m_1,m_2\to+\infty}{\longrightarrow}0. ∥ Γ t k u m 1 ( t ) − Γ t k u m 2 ( t ) ∥ L 2 ≤ c ( k , V ) ( ∥ x k ( u 0 m 1 − u 0 m 2 ) ∥ L 2 + ∥ u 0 m 1 − u 0 m 2 ∥ L 2 ) m 1 , m 2 → + ∞ ⟶ 0.
which shows that { Γ t k u m } m \{\Gamma_t^k u^m\}_m { Γ t k u m } m is a Cauchy sequence in C ( [ 0 , 1 ] : L 2 ( R ) ) C([0,1]:L^2(\mathbb{R})) C ([ 0 , 1 ] : L 2 ( R )) . Since C ( [ 0 , 1 ] : L 2 ( R ) ) C([0,1]:L^2(\mathbb{R})) C ([ 0 , 1 ] : L 2 ( R )) is complete, there exists a unique w w w such that Γ t k u m → w \Gamma_t^k u^m\to w Γ t k u m → w in C ( [ 0 , 1 ] : L 2 ( R ) ) C([0,1]:L^2(\mathbb{R})) C ([ 0 , 1 ] : L 2 ( R )) .
Moreover, for each m m m we have that Γ t k u m = e i x 2 / 4 t ( 2 i t ) k ∂ x k u ~ m \Gamma_t^k u^m=e^{ix^2/4t}(2it)^k\partial_x^k\widetilde{u}^m Γ t k u m = e i x 2 /4 t ( 2 i t ) k ∂ x k u m . Then, for all t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] ,
∂ x k u ~ m → m → + ∞ h = e − i x 2 / 4 t ( 2 i t ) − k w , in L 2 ( R ) \partial_x^k\widetilde{u}^m\xrightarrow[m\to+\infty]{}h=e^{-ix^2/4t}(2it)^{-k}w,\quad\text{in }L^2(\mathbb{R}) ∂ x k u m m → + ∞ h = e − i x 2 /4 t ( 2 i t ) − k w , in L 2 ( R )
and in particular h = ∂ x k u ~ h=\partial_x^k\widetilde{u} h = ∂ x k u in the weak sense, because { u m } \{u^m\} { u m } converges to u u u in C ( [ 0 , 1 ] : L 2 ( R ) ) C([0,1]:L^2(\mathbb{R})) C ([ 0 , 1 ] : L 2 ( R )) . Moreover
Γ t k u m = ( 2 i t ) k e i x 2 / 4 t ∂ x k u ~ m → ( 2 i t ) k e i x 2 / 4 t ∂ x k u ~ weakly in L 2 ( R ) . \Gamma_t^k u^m=(2it)^k e^{ix^2/4t}\partial_x^k\widetilde{u}^m\to(2it)^k e^{ix^2/4t}\partial_x^k\widetilde{u}\quad\text{weakly in }L^2(\mathbb{R}). Γ t k u m = ( 2 i t ) k e i x 2 /4 t ∂ x k u m → ( 2 i t ) k e i x 2 /4 t ∂ x k u weakly in L 2 ( R ) .
Since Γ t k u m → w \Gamma_t^k u^m\to w Γ t k u m → w strongly, we also have Γ t k u m → w \Gamma_t^k u^m\to w Γ t k u m → w , and by uniqueness of the weak limit, we must have Γ t k u : = w = ( 2 i t ) k e i x 2 / 4 t ∂ x k u ~ \Gamma_t^k u:=w=(2it)^k e^{ix^2/4t}\partial_x^k\widetilde{u} Γ t k u := w = ( 2 i t ) k e i x 2 /4 t ∂ x k u . In particular we get the bound
sup t ∈ [ 0 , 1 ] ∥ Γ t k u ∥ L 2 ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 + ∥ u 0 ∥ L 2 ) . (25) \sup_{t\in[0,1]}\|\Gamma_t^k u\|_{L^2}\leq c(k,V)\left(\|x^k u_0\|_{L^2}+\|u_0\|_{L^2}\right). \tag*{(25)} t ∈ [ 0 , 1 ] sup ∥ Γ t k u ∥ L 2 ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 + ∥ u 0 ∥ L 2 ) . (25)
The next step is to show that u 1 u_1 u 1 belongs to H k \mathcal{H}^k H k . Since ∥ Γ 1 j u ( 1 ) ∥ L 2 = ∥ ∂ x j u ~ 1 ∥ L 2 \|\Gamma_1^j u(1)\|_{L^2}=\|\partial_x^j\widetilde{u}_1\|_{L^2} ∥ Γ 1 j u ( 1 ) ∥ L 2 = ∥ ∂ x j u 1 ∥ L 2 and ∥ x j u ~ 1 ∥ L 2 = ∥ x j u 1 ∥ L 2 \|x^j\widetilde{u}_1\|_{L^2}=\|x^j u_1\|_{L^2} ∥ x j u 1 ∥ L 2 = ∥ x j u 1 ∥ L 2 , it follows that ∂ x j u ~ 1 ∈ L 2 \partial_x^j\widetilde{u}_1\in L^2 ∂ x j u 1 ∈ L 2 and x j u ~ 1 ∈ L 2 x^j\widetilde{u}_1\in L^2 x j u 1 ∈ L 2 for all j ≤ k j\leq k j ≤ k . This implies that u ~ 1 ∈ H k \widetilde{u}_1\in\mathcal{H}^k u 1 ∈ H k .
By using estimate (3.1), we deduce that
∑ l + j ≤ k ∥ x l Γ 1 j u 1 ∥ L 2 = ∑ l + j ≤ k ∥ x l ∂ x j u ~ 1 ∥ L 2 ≲ ∥ u ~ 1 ∥ H k . \sum_{l+j\leq k}\|x^l\Gamma_1^j u_1\|_{L^2}=\sum_{l+j\leq k}\|x^l\partial_x^j\widetilde{u}_1\|_{L^2}\lesssim\|\widetilde{u}_1\|_{\mathcal{H}^k}. l + j ≤ k ∑ ∥ x l Γ 1 j u 1 ∥ L 2 = l + j ≤ k ∑ ∥ x l ∂ x j u 1 ∥ L 2 ≲ ∥ u 1 ∥ H k .
Finally, since Γ 1 j = ( x + 2 i ∂ x ) j \Gamma_1^j=(x+2i\partial_x)^j Γ 1 j = ( x + 2 i ∂ x ) j , it follows by induction that
∂ x j = ∑ l + s ≤ j c ( l , s , j ) x l Γ 1 s , \partial_x^j=\sum_{l+s\leq j}c(l,s,j)x^l\Gamma_1^s, ∂ x j = l + s ≤ j ∑ c ( l , s , j ) x l Γ 1 s ,
for some positive constants c ( l , s , j ) c(l,s,j) c ( l , s , j ) . This implies that
∑ l + j ≤ k ∥ x l ∂ x j u 1 ∥ L 2 ≲ ∥ u ~ 1 ∥ H k , \sum_{l+j\leq k}\|x^l\partial_x^j u_1\|_{L^2}\lesssim\|\widetilde{u}_1\|_{\mathcal{H}^k}, l + j ≤ k ∑ ∥ x l ∂ x j u 1 ∥ L 2 ≲ ∥ u 1 ∥ H k ,
so that u 1 ∈ H k u_1\in\mathcal{H}^k u 1 ∈ H k .
We can prove a similar estimate to (3.6) for the function v ( x , t ) = u ( x , 1 − t ) v(x,t)=u(x,1-t) v ( x , t ) = u ( x , 1 − t ) ,
sup t ∈ [ 0 , 1 ] ∥ Γ − t k v ( t ) ∥ L 2 = sup t ∈ [ 0 , 1 ] ∥ Γ t − 1 k u ( t ) ∥ L 2 ≤ c ( k , V ) ( ∥ x k u 1 ∥ L 2 + ∥ u 0 ∥ L 2 ) . (26) \sup_{t\in[0,1]}\|\Gamma_{-t}^k v(t)\|_{L^2}=\sup_{t\in[0,1]}\|\Gamma_{t-1}^k u(t)\|_{L^2}\leq c(k,V)\left(\|x^k u_1\|_{L^2}+\|u_0\|_{L^2}\right). \tag*{(26)} t ∈ [ 0 , 1 ] sup ∥ Γ − t k v ( t ) ∥ L 2 = t ∈ [ 0 , 1 ] sup ∥ Γ t − 1 k u ( t ) ∥ L 2 ≤ c ( k , V ) ( ∥ x k u 1 ∥ L 2 + ∥ u 0 ∥ L 2 ) . (26)
which by arguing as for u 1 u_1 u 1 will imply that u 0 ∈ H k u_0\in\mathcal{H}^k u 0 ∈ H k . By Proposition 3.1, we finally conclude that u ∈ C ( [ 0 , 1 ] , H k ( R ) ) u\in C([0,1],\mathcal{H}^k(\mathbb{R})) u ∈ C ([ 0 , 1 ] , H k ( R )) .
Corollary 3.3. Assume that V ∈ L ∞ ( [ 0 , 1 ] : C b ∞ ( R ) ) V\in L^\infty([0,1]:C_b^\infty(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b ∞ ( R )) . Let u 0 ∈ H 1 u_0\in\mathcal{H}^1 u 0 ∈ H 1 and let u u u be the solution u ∈ C ( [ 0 , 1 ] : H 1 ) u\in C([0,1]:\mathcal{H}^1) u ∈ C ([ 0 , 1 ] : H 1 ) obtained in Proposition 3.1. Moreover, assume that u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) u_0,u_1=u(\cdot,1)\in L^2(x^{2k}\,dx) u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) for all k ∈ N k\in\mathbb{N} k ∈ N . Then u ∈ C ( [ 0 , 1 ] : S ( R ) ) u\in C([0,1]:\mathcal{S}(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) .
Similar results are known for solutions to (1.2). The proof is almost identical to the proof of Proposition 3.3, see [23 , 24 , 25 , 27 ] .
Proposition 3.4. Let u 0 ∈ H 1 u_0\in\mathcal{H}^1 u 0 ∈ H 1 and let u u u be the solution u ∈ C ( [ 0 , 1 ] : H 1 ) u\in C([0,1]:\mathcal{H}^1) u ∈ C ([ 0 , 1 ] : H 1 ) obtained in Proposition 3.2. Moreover, assume that u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) u_0,u_1=u(\cdot,1)\in L^2(x^{2k}\,dx) u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) for some k ∈ N k\in\mathbb{N} k ∈ N . Then u ∈ C ( [ 0 , 1 ] : H k ( R ) ∩ L 2 ( x 2 k d x ) ) : = C ( [ 0 , 1 ] : H k ( R ) ) u\in C([0,1]:\mathcal{H}^k(\mathbb{R})\cap L^2(x^{2k}\,dx)):=C([0,1]:\mathcal{H}^k(\mathbb{R})) u ∈ C ([ 0 , 1 ] : H k ( R ) ∩ L 2 ( x 2 k d x )) := C ([ 0 , 1 ] : H k ( R )) .
Corollary 3.4. Let u 0 ∈ H 1 u_0 \in\mathcal{H}^1 u 0 ∈ H 1 and let u u u be the solution u ∈ C ( [ 0 , 1 ] : H 1 ) u \in C([0,1] : \mathcal{H}^1) u ∈ C ([ 0 , 1 ] : H 1 ) obtained in Proposition 3.2. Moreover, assume that u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) u_0,u_1=u(\cdot,1)\in L^2(x^{2k}dx) u 0 , u 1 = u ( ⋅ , 1 ) ∈ L 2 ( x 2 k d x ) for all k ∈ N k\in\mathbb{N} k ∈ N . Then u ∈ C ( [ 0 , 1 ] : S ( R ) ) u\in C([0,1] : \mathcal{S}(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) .
Now, we are in position to prove Theorem 1.3 and Theorem 1.4.
Remark 3.1. Thanks to Proposition 3.1, the corresponding solution u u u belongs to C ( [ 0 , 1 ] : H 1 ) C([0,1] : \mathcal{H}^1) C ([ 0 , 1 ] : H 1 ) . As a consequence u 0 u_0 u 0 and u 1 u_1 u 1 are continuous on R \mathbb{R} R , so that the condition (1.7) makes sense.
Proof of Theorem 1.3. Let
I k ( f ) = ∫ R ∣ x ∣ 2 k ∣ f ( x ) ∣ 2 d x . (27) I_k(f)=\int_{\mathbb{R}} |x|^{2k}|f(x)|^2dx. \tag*{(27)} I k ( f ) = ∫ R ∣ x ∣ 2 k ∣ f ( x ) ∣ 2 d x . (27)
We will prove that
I k ( u 0 ) < ∞ and I k ( u 1 ) < ∞ for all k ≥ 1. I_k(u_0)<\infty\quad\text{and} \quad I_k(u_1)<\infty\quad\text{for all } k\geq1. I k ( u 0 ) < ∞ and I k ( u 1 ) < ∞ for all k ≥ 1.
Then, by Corollary 3.3, it follows that u ∈ C ( [ 0 , 1 ] : S ( R ) ) u\in C([0,1] : \mathcal{S}(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) .
To this end, we follow closely the argument in Section 5.1 in [30 ] . Let X 0 > 0 X_0>0 X 0 > 0 be a large number to be fixed later. For ∣ x ∣ ≥ X 0 |x|\geq X_0 ∣ x ∣ ≥ X 0 , the interval between two zeroes of u 1 u_1 u 1 containing x x x has length at most ( 2 λ 1 ∣ x ∣ b ) − 1 (2\lambda_1|x|^b)^{-1} ( 2 λ 1 ∣ x ∣ b ) − 1 , where 0 < b < 1 β − 1 0<b<\frac{1}{\beta}-1 0 < b < β 1 − 1 and λ 1 = λ 1 ( X 0 ) \lambda_1=\lambda_1(X_0) λ 1 = λ 1 ( X 0 ) can be chosen as large as we want by fixing X 0 X_0 X 0 large.
Indeed, assume without loss of generality that x > X 0 x>X_0 x > X 0 and x ∈ ( n β , ( n + 1 ) β ) x\in(n^\beta,(n+1)^\beta) x ∈ ( n β , ( n + 1 ) β ) . Then, by the mean value theorem we have
∣ ( n + 1 ) β − n β ∣ ≤ β n − ( 1 − β ) . |(n+1)^\beta-n^\beta|\leq\beta n^{-(1-\beta)}. ∣ ( n + 1 ) β − n β ∣ ≤ β n − ( 1 − β ) .
Since x ≤ ( n + 1 ) β ≤ ( 2 n ) β x\leq(n+1)^\beta\leq(2n)^\beta x ≤ ( n + 1 ) β ≤ ( 2 n ) β , we have that
n − ( 1 − β ) ≤ 2 1 − β x − 1 − β β ≤ 2 1 − β x − b x − δ ≤ 2 1 − β ∣ x ∣ − b X 0 − δ , n^{-(1-\beta)}\leq2^{1-\beta}x^{-\frac{1-\beta}{\beta}}\leq2^{1-\beta}x^{-b}x^{-\delta}\leq2^{1-\beta}|x|^{-b}X_0^{-\delta}, n − ( 1 − β ) ≤ 2 1 − β x − β 1 − β ≤ 2 1 − β x − b x − δ ≤ 2 1 − β ∣ x ∣ − b X 0 − δ ,
where δ = 1 β − 1 − b > 0 \delta=\frac{1}{\beta}-1-b>0 δ = β 1 − 1 − b > 0 . Thus,
∣ ( n + 1 ) β − n β ∣ ≤ β 2 1 − β X 0 − δ ∣ x ∣ − b = ( 2 λ 1 ∣ x ∣ b ) − 1 , |(n+1)^\beta-n^\beta|\leq\beta2^{1-\beta}X_0^{-\delta}|x|^{-b}=(2\lambda_1|x|^b)^{-1}, ∣ ( n + 1 ) β − n β ∣ ≤ β 2 1 − β X 0 − δ ∣ x ∣ − b = ( 2 λ 1 ∣ x ∣ b ) − 1 ,
for λ 1 = 2 β X 0 δ 4 β \lambda_1=\frac{2^\beta X_0^\delta}{4\beta} λ 1 = 4 β 2 β X 0 δ .
We argue exactly as in Section 5.1 in [30 ] for f = u ~ 1 f=\tilde{u}_1 f = u ~ 1 . Although our definitions of a a a and b b b differ, this does not affect the argument. Then, we get from estimate (5.6) in [30 ] that
∫ R ∣ u ~ 1 ( x ) ∣ 2 Φ R ( ∣ x ∣ 2 b ) d x ≤ 1 λ 1 ∫ R ∣ u ~ 1 ( ξ ) ∣ 2 Φ R ( ξ 2 ) d ξ + c ∥ u ~ 1 ∥ L 2 2 , (28) \int_{\mathbb{R}}|\tilde{u}_1(x)|^2\Phi_R(|x|^{2b})dx\leq\frac{1}{\sqrt{\lambda_1}}\int_{\mathbb{R}}|\tilde{u}_1(\xi)|^2\Phi_R(\xi^2)d\xi+c\|\tilde{u}_1\|_{L^2}^2, \tag*{(28)} ∫ R ∣ u ~ 1 ( x ) ∣ 2 Φ R ( ∣ x ∣ 2 b ) d x ≤ λ 1 1 ∫ R ∣ u ~ 1 ( ξ ) ∣ 2 Φ R ( ξ 2 ) d ξ + c ∥ u ~ 1 ∥ L 2 2 , (28)
where
Φ R ( t ) = { t k t ∈ [ 0 , R ] R k + k R k − 1 ( t − R ) t > R . \Phi_R(t)=
\begin{cases}
t^k & t\in[0,R]\\
R^k+kR^{k-1}(t-R) & t>R.
\end{cases} Φ R ( t ) = { t k R k + k R k − 1 ( t − R ) t ∈ [ 0 , R ] t > R .
for k ≥ 1 k\geq1 k ≥ 1 . Similarly, replacing the role of u 1 u_1 u 1 and u 0 u_0 u 0 , and working with the function f = v ~ − 1 = e i x 2 / 4 v 1 = e i x 2 / 4 u 0 f=\tilde{v}_{-1}=e^{ix^2/4}v_1=e^{ix^2/4}u_0 f = v ~ − 1 = e i x 2 /4 v 1 = e i x 2 /4 u 0 , instead of u ~ 1 \tilde{u}_1 u ~ 1 , and for 0 < a < 1 α − 1 0<a<\frac{1}{\alpha}-1 0 < a < α 1 − 1 , λ 0 = 2 α X 0 δ 4 α \lambda_0=\frac{2^\alpha X_0^\delta}{4\alpha} λ 0 = 4 α 2 α X 0 δ , for some δ > 0 \delta>0 δ > 0 , we obtain that
∫ R ∣ v ~ − 1 ( x ) ∣ 2 Φ R ( ∣ x ∣ 2 a ) d x ≤ 1 λ 0 ∫ R ∣ v ~ − 1 ( ξ ) ∣ 2 Φ R ( ξ 2 ) d ξ + c ∥ u 0 ∥ L 2 2 . (29) \int_{\mathbb{R}}|\tilde{v}_{-1}(x)|^2\Phi_R(|x|^{2a})dx\leq\frac{1}{\sqrt{\lambda_0}}\int_{\mathbb{R}}|\tilde{v}_{-1}(\xi)|^2\Phi_R(\xi^2)d\xi+c\|u_0\|_{L^2}^2. \tag*{(29)} ∫ R ∣ v ~ − 1 ( x ) ∣ 2 Φ R ( ∣ x ∣ 2 a ) d x ≤ λ 0 1 ∫ R ∣ v ~ − 1 ( ξ ) ∣ 2 Φ R ( ξ 2 ) d ξ + c ∥ u 0 ∥ L 2 2 . (29)
We will now prove by induction that the estimates (3.9) and (3.10) imply that u 0 , u 1 ∈ L 2 ( x 2 k d x ) u_0,u_1\in L^2(x^{2k}dx) u 0 , u 1 ∈ L 2 ( x 2 k d x ) for all k ≥ 1 k\geq1 k ≥ 1 . Indeed, since u 0 ∈ H 1 u_0\in\mathcal{H}^1 u 0 ∈ H 1 , the property holds for k = 1 k=1 k = 1 thanks to Proposition 3.1 with k = 1 k=1 k = 1 . Now, assume that x k u 0 ∈ L 2 x^ku_0\in L^2 x k u 0 ∈ L 2 for some k ∈ N k\in\mathbb{N} k ∈ N , k ≥ 1 k\geq1 k ≥ 1 . We observe from Proposition 3.3, in particular (3.6), that
∫ R ∣ u ~ 1 ( ξ ) ∣ 2 Φ R ( ξ 2 ) d ξ ≤ ∫ R ∣ u ~ 1 ( ξ ) ∣ 2 ξ 2 k d ξ ≤ ∫ R ∣ ∂ x k u ~ 1 ∣ 2 d x = ( 2 ) − k ∫ R ∣ Γ 1 k u 1 ∣ 2 d x ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) . \int_{\mathbb{R}}|\tilde{u}_1(\xi)|^2\Phi_R(\xi^2)d\xi\leq\int_{\mathbb{R}}|\tilde{u}_1(\xi)|^2\xi^{2k}d\xi\leq\int_{\mathbb{R}}|\partial_x^k\tilde{u}_1|^2dx=(2)^{-k}\int_{\mathbb{R}}|\Gamma_1^k u_1|^2dx\leq c(k,V)\left(\|x^ku_0\|_{L^2}^2+\|u_0\|_{L^2}^2\right). ∫ R ∣ u ~ 1 ( ξ ) ∣ 2 Φ R ( ξ 2 ) d ξ ≤ ∫ R ∣ u ~ 1 ( ξ ) ∣ 2 ξ 2 k d ξ ≤ ∫ R ∣ ∂ x k u ~ 1 ∣ 2 d x = ( 2 ) − k ∫ R ∣ Γ 1 k u 1 ∣ 2 d x ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) .
Arguing as in the proof of Proposition 3.3, the last estimate is rigorously justified assuming only that x k u 0 ∈ L 2 x^ku_0\in L^2 x k u 0 ∈ L 2 . This estimate combined with (3.9) yields that the bound
∫ R ∣ u ~ 1 ( x ) ∣ 2 Φ R ( ∣ x ∣ 2 b ) d x ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) (30) \int_{\mathbb{R}}|\tilde{u}_1(x)|^2\Phi_R(|x|^{2b})dx\leq c(k,V)\left(\|x^ku_0\|_{L^2}^2+\|u_0\|_{L^2}^2\right) \tag*{(30)} ∫ R ∣ u ~ 1 ( x ) ∣ 2 Φ R ( ∣ x ∣ 2 b ) d x ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) (30)
holds. By letting R → + ∞ R \to+\infty R → + ∞ and using the monotone convergence theorem, it follows that
∫ R ∣ x ∣ 2 b k ∣ u ~ 1 ∣ 2 d x ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) , (31) \int_{\mathbb{R}} |x|^{2bk} |\widetilde{u}_1|^2 \, dx \le c(k,V) \left( \|x^k u_0\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right), \tag*{(31)} ∫ R ∣ x ∣ 2 bk ∣ u 1 ∣ 2 d x ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) , (31)
which implies that x b u 1 ∈ L 2 x^b u_1 \in L^2 x b u 1 ∈ L 2 . Arguing similarly for v ~ − 1 \widetilde{v}_{-1} v − 1 , and using (3.10) instead (3.9), we deduce that
∫ R ∣ x ∣ 2 a j ∣ v ~ − 1 ∣ 2 d x ≤ c ( j , V ) ( ∥ x j u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) (32) \int_{\mathbb{R}} |x|^{2aj} |\widetilde{v}_{-1}|^2 \, dx \le c(j,V) \left( \|x^j u_1\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right) \tag*{(32)} ∫ R ∣ x ∣ 2 aj ∣ v − 1 ∣ 2 d x ≤ c ( j , V ) ( ∥ x j u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) (32)
for j = j ( k ) = ⌊ b k ⌋ j=j(k)=\lfloor bk\rfloor j = j ( k ) = ⌊ bk ⌋ . This estimate combined with (3.12), (recall also that v ~ − 1 = e i x 2 / 4 u 0 \widetilde{v}_{-1}=e^{ix^2/4}u_0 v − 1 = e i x 2 /4 u 0 ) implies that
∫ R ∣ x ∣ 2 a j ( k ) ∣ u 0 ∣ 2 d x ≤ c ( k , V ) ( ∥ x j ( k ) u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) . (33) \int_{\mathbb{R}} |x|^{2aj(k)} |u_0|^2 \, dx \le c(k,V) \left( \|x^{j(k)}u_1\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right) \le c(k,V) \left( \|x^k u_0\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right). \tag*{(33)} ∫ R ∣ x ∣ 2 aj ( k ) ∣ u 0 ∣ 2 d x ≤ c ( k , V ) ( ∥ x j ( k ) u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) . (33)
Next, we assume x k u 1 ∈ L 2 x^k u_1 \in L^2 x k u 1 ∈ L 2 . We deduce from a similar reasoning that
∫ R ∣ x ∣ 2 b j ^ ( k ) ∣ u ~ 1 ∣ 2 d x ≤ c ( k , V ) ( ∥ x j ^ ( k ) u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) ≤ c ( k , V ) ( ∥ x k u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) , (34) \int_{\mathbb{R}} |x|^{2b\hat{j}(k)} |\widetilde{u}_1|^2 \, dx \le c(k,V) \left( \|x^{\hat{j}(k)}u_0\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right) \le c(k,V) \left( \|x^k u_1\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right), \tag*{(34)} ∫ R ∣ x ∣ 2 b j ^ ( k ) ∣ u 1 ∣ 2 d x ≤ c ( k , V ) ( ∥ x j ^ ( k ) u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) ≤ c ( k , V ) ( ∥ x k u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) , (34)
for j ^ ( k ) = ⌊ a k ⌋ \hat{j}(k)=\lfloor ak\rfloor j ^ ( k ) = ⌊ ak ⌋ .
By hypothesis, we know that 0 < α , β < 1 0<\alpha,\beta<1 0 < α , β < 1 satisfy α + β < 1 \alpha+\beta<1 α + β < 1 , which is equivalent to the condition
( 1 − α α ) ( 1 − β β ) > 1. \left(\frac{1-\alpha}{\alpha}\right)\left(\frac{1-\beta}{\beta}\right)>1. ( α 1 − α ) ( β 1 − β ) > 1.
Thus we can choose a < 1 α − 1 a<\frac{1}{\alpha}-1 a < α 1 − 1 and b < 1 β − 1 b<\frac{1}{\beta}-1 b < β 1 − 1 such that a b > 1 ab>1 ab > 1 .
Since a j ( k ) = a ⌊ b k ⌋ ≥ a ( b k − 1 ) aj(k)=a\lfloor bk\rfloor\ge a(bk-1) aj ( k ) = a ⌊ bk ⌋ ≥ a ( bk − 1 ) , there exists an integer k 0 k_0 k 0 such that
a j ( k ) ≥ k + 1 , ∀ k ≥ k 0 . aj(k)\ge k+1,\qquad\forall\ k\ge k_0. aj ( k ) ≥ k + 1 , ∀ k ≥ k 0 .
Then, it follows from (3.14) that
∫ R ∣ x ∣ 2 ( k + 1 ) ∣ u 0 ∣ 2 ≤ ∫ R ∣ x ∣ 2 a j ( k ) ∣ u 0 ∣ 2 + ∥ u 0 ∥ L 2 2 ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) . \int_{\mathbb{R}} |x|^{2(k+1)} |u_0|^2 \le\int_{\mathbb{R}} |x|^{2aj(k)} |u_0|^2 + \|u_0\|_{L^2}^2 \le c(k,V) \left( \|x^k u_0\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right). ∫ R ∣ x ∣ 2 ( k + 1 ) ∣ u 0 ∣ 2 ≤ ∫ R ∣ x ∣ 2 aj ( k ) ∣ u 0 ∣ 2 + ∥ u 0 ∥ L 2 2 ≤ c ( k , V ) ( ∥ x k u 0 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) .
Arguing similarly, there exists k 1 ∈ N k_1\in\mathbb{N} k 1 ∈ N such that
b j ^ ( k ) ≥ k + 1 , ∀ k ≥ k 1 , b\hat{j}(k)\ge k+1,\qquad\forall\ k\ge k_1, b j ^ ( k ) ≥ k + 1 , ∀ k ≥ k 1 ,
so by (3.15)
∫ R ∣ x ∣ 2 ( k + 1 ) ∣ u 1 ∣ 2 ≤ c ( k , V ) ( ∥ x k u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) . \int_{\mathbb{R}} |x|^{2(k+1)} |u_1|^2 \le c(k,V) \left( \|x^k u_1\|_{L^2}^2 + \|u_0\|_{L^2}^2 \right). ∫ R ∣ x ∣ 2 ( k + 1 ) ∣ u 1 ∣ 2 ≤ c ( k , V ) ( ∥ x k u 1 ∥ L 2 2 + ∥ u 0 ∥ L 2 2 ) .
For k ≤ K 0 : = max { k 0 , k 1 } k\le K_0:=\max\{k_0,k_1\} k ≤ K 0 := max { k 0 , k 1 } , we have directly
∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 0 ∣ 2 d x + ∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 1 ∣ 2 d x ≤ ∥ u 0 ∥ L 2 2 + ∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 0 ∣ 2 d x + ∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 1 ∣ 2 d x < ∞ . \int_{\mathbb{R}} |x|^{2(K_0+1)} |u_0|^2 \, dx+\int_{\mathbb{R}} |x|^{2(K_0+1)} |u_1|^2 \, dx\le\|u_0\|_{L^2}^2+\int_{\mathbb{R}} |x|^{2(K_0+1)} |u_0|^2 \, dx+\int_{\mathbb{R}} |x|^{2(K_0+1)} |u_1|^2 \, dx<\infty. ∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 0 ∣ 2 d x + ∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 1 ∣ 2 d x ≤ ∥ u 0 ∥ L 2 2 + ∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 0 ∣ 2 d x + ∫ R ∣ x ∣ 2 ( K 0 + 1 ) ∣ u 1 ∣ 2 d x < ∞.
This concludes the induction process. Therefore, we deduce that u 1 , u 0 ∈ L 2 ( x 2 k d x ) u_1,u_0\in L^2(x^{2k}\,dx) u 1 , u 0 ∈ L 2 ( x 2 k d x ) for all k ≥ 1 k\ge1 k ≥ 1 . Finally we apply Corollary 3.3 to conclude that u ∈ C ( [ 0 , 1 ] ; S ( R ) ) u\in C([0,1];S(\mathbb{R})) u ∈ C ([ 0 , 1 ] ; S ( R )) .
Now, we prove Theorem 1.4. The proof for the nonlinear case follows the same argument as for the potential case, and by adapting estimate (3.3) on ∥ Γ t k u ∥ L 2 \|\Gamma_t^k u\|_{L^2} ∥ Γ t k u ∥ L 2 in the case of the nonlinear Schrödinger equation (1.2), (see [27 , 23 , 24 , 25 ] ).
Proof of Theorem 1.4. We follow the same strategy as in the potential case. The argument is identical until (3.10). By Proposition 3.2, we know that if u 0 ∈ H 1 u_0\in\mathcal{H}^1 u 0 ∈ H 1 , then u 1 ∈ H 1 u_1\in\mathcal{H}^1 u 1 ∈ H 1 , so the induction hypothesis holds for k = 1 k=1 k = 1 . Instead of the estimate (3.6), we use the bound
sup t ∈ [ 0 , 1 ] ∥ Γ t k u ∥ L 2 ≤ c ( k , ∥ u 0 ∥ H 1 ) ∥ x k u 0 ∥ . (35) \sup_{t\in[0,1]} \|\Gamma_t^k u\|_{L^2}\le c(k,\|u_0\|_{\mathcal{H}^1})\|x^k u_0\|. \tag*{(35)} t ∈ [ 0 , 1 ] sup ∥ Γ t k u ∥ L 2 ≤ c ( k , ∥ u 0 ∥ H 1 ) ∥ x k u 0 ∥. (35)
This estimate corresponds to (4.12) in the proof of Lemma 4.1 in [27 ] and is also rigorously justified by only assuming x k u 0 ∈ L 2 x^k u_0\in L^2 x k u 0 ∈ L 2 .
Then, the rest of the argument follows as for the potential case, and the result follows by Corollary 3.4.
Proof of the unique continuation result in the case of the potential
This section is dedicated to the proof Theorem 1.1. Below, we give a short overview of the proof.
We have seen in Section 3 that, for potential V ∈ L ∞ ( [ 0 , 1 ] : C b ∞ ( R ) ) V \in L^\infty([0,1] : C_b^\infty(\mathbb{R})) V ∈ L ∞ ([ 0 , 1 ] : C b ∞ ( R )) , the solutions u u u of (1.1) emanating from initial data in H 1 \mathcal{H}^1 H 1 are in Schwartz for all t ∈ [ 0 , 1 ] t \in[0,1] t ∈ [ 0 , 1 ] if one assumes that u ( ± n , 0 ) = u ( ± n β , 1 ) = 0 u(\pm n,0)=u(\pm n^\beta,1)=0 u ( ± n , 0 ) = u ( ± n β , 1 ) = 0 , for all n ∈ N n \in\mathbb{N} n ∈ N and α , β ∈ ( 0 , 1 ) \alpha,\beta\in(0,1) α , β ∈ ( 0 , 1 ) with 0 < α + β < 1 0<\alpha+\beta<1 0 < α + β < 1 . In this section, we impose stronger decay assumptions on V V V , which in each case will provide a subset of ( α , β ) (\alpha,\beta) ( α , β ) for which the unique continuation result from the zeroes holds. By using Theorem 1.3, we know that u ∈ C ( [ 0 , 1 ] : S ( R ) ) u \in C([0,1] : \mathcal{S}(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) , so that all the computations below are rigorous.
We set
I k ( f ) = ∫ R ∣ x ∣ 2 k ∣ f ( x ) ∣ 2 d x . I_k(f)=\int_{\mathbb{R}} |x|^{2k}|f(x)|^2dx. I k ( f ) = ∫ R ∣ x ∣ 2 k ∣ f ( x ) ∣ 2 d x .
Moreover, let
0 < a < 1 − α α , 0 < b < 1 − β β . 0<a<\frac{1-\alpha}{\alpha},\qquad0<b<\frac{1-\beta}{\beta}. 0 < a < α 1 − α , 0 < b < β 1 − β .
We will prove that
I p k 2 ( u 0 ) ≤ c k ( k ! ) p δ 0 , I q k 2 ( u 1 ) ≤ c k ( k ! ) q δ 1 , (36) I_{\frac{p k}{2}}(u_0)\leq c^k(k!)^{p\delta_0},\qquad I_{\frac{q k}{2}}(u_1)\leq c^k(k!)^{q\delta_1}, \tag*{(36)} I 2 p k ( u 0 ) ≤ c k ( k ! ) p δ 0 , I 2 q k ( u 1 ) ≤ c k ( k ! ) q δ 1 , (36)
for 1 p + 1 q = 1 \frac{1}{p}+\frac{1}{q}=1 p 1 + q 1 = 1 , where δ 0 \delta_0 δ 0 and δ 1 \delta_1 δ 1 depend on a a a and b b b .
With estimate (36) in hand, we deduce Gaussian decay for u 0 u_0 u 0 and u 1 u_1 u 1 as soon as p δ 0 < 1 p\delta_0<1 p δ 0 < 1 and q δ 1 < 1 q\delta_1<1 q δ 1 < 1 . Then, we will use Morgan’s uncertainty principle, Theorem 1.5, to conclude that u ≡ 0 u\equiv0 u ≡ 0 . The conditions p δ 0 < 1 p\delta_0<1 p δ 0 < 1 and q δ 1 < 1 q\delta_1<1 q δ 1 < 1 will give us the final restrictions on α \alpha α and β \beta β by optimizing the choice of p p p and q q q .
Decay estimates from the zeros.
Here we follow closely the argument in Section 5.2 in [30 ] . As in the proof of Theorem 1.3, we have that for ∣ x ∣ ≥ X 0 |x|\geq X_0 ∣ x ∣ ≥ X 0 , the interval interval between two zeroes of u 1 u_1 u 1 containing x x x has length at most ( 2 λ 1 ∣ x ∣ b ) − 1 (2\lambda_1|x|^b)^{-1} ( 2 λ 1 ∣ x ∣ b ) − 1 for λ 1 = 2 β X 0 δ 4 β \lambda_1=\frac{2^\beta X_0^\delta}{4\beta} λ 1 = 4 β 2 β X 0 δ , 0 < b < 1 β − 1 0<b<\frac{1}{\beta}-1 0 < b < β 1 − 1 and δ = 1 β − 1 − b \delta=\frac{1}{\beta}-1-b δ = β 1 − 1 − b .
We argue exactly as in Section 5.2 in [30 ] for f = u ~ 1 f=\widetilde{u}_1 f = u 1 . Although our definitions of a a a and b b b differ, this does not affect the argument. Then, we get from estimate (5.11) in [30 ] , by taking X 0 X_0 X 0 large enough, that for k ≥ k 0 k\geq k_0 k ≥ k 0 ,
I b k ( u 1 ) = I b k ( u ~ 1 ) = ∫ R ∣ x ∣ 2 b k ∣ u ~ 1 ( x ) ∣ 2 d x ≤ 1 2 ∫ R ∣ ξ ∣ 2 k ∣ u ~ 1 ( ξ ) ∣ 2 d ξ + c β k k 2 b b + 1 k I 0 ( u ~ 1 ) . I_{bk}(u_1)=I_{bk}(\widetilde{u}_1)=\int_{\mathbb{R}}|x|^{2bk}|\widetilde{u}_1(x)|^2dx\leq\frac{1}{2}\int_{\mathbb{R}}|\xi|^{2k}|\widetilde{u}_1(\xi)|^2d\xi+c_\beta^k k^{\frac{2b}{b+1}k}I_0(\widetilde{u}_1). I bk ( u 1 ) = I bk ( u 1 ) = ∫ R ∣ x ∣ 2 bk ∣ u 1 ( x ) ∣ 2 d x ≤ 2 1 ∫ R ∣ ξ ∣ 2 k ∣ u 1 ( ξ ) ∣ 2 d ξ + c β k k b + 1 2 b k I 0 ( u 1 ) .
By using Plancherel’s identity, it follows that
∫ R ∣ x ∣ 2 b k ∣ u ~ 1 ( x ) ∣ 2 d x ≤ 1 2 ( 2 π ) − 2 k ∫ R ∣ ∂ x k u ~ 1 ( x ) ∣ 2 d x + c β k k 2 b b + 1 k I 0 ( u 1 ) ≤ 1 2 2 − 2 k ∥ Γ 1 k u 1 ∥ L 2 2 + c β k k 2 b b + 1 k I 0 ( u 1 ) . (37) \begin{aligned}
\int_{\mathbb{R}}|x|^{2bk}|\widetilde{u}_1(x)|^2dx\leq\frac{1}{2}(2\pi)^{-2k}\int_{\mathbb{R}}|\partial_x^k\widetilde{u}_1(x)|^2dx+c_\beta^k k^{\frac{2b}{b+1}k}I_0(u_1) \\
&\leq\frac{1}{2}2^{-2k}\|\Gamma_1^k u_1\|_{L^2}^2+c_\beta^k k^{\frac{2b}{b+1}k}I_0(u_1). \tag*{(37)}
\end{aligned} ∫ R ∣ x ∣ 2 bk ∣ u 1 ( x ) ∣ 2 d x ≤ 2 1 ( 2 π ) − 2 k ∫ R ∣ ∂ x k u 1 ( x ) ∣ 2 d x + c β k k b + 1 2 b k I 0 ( u 1 ) ≤ 2 1 2 − 2 k ∥ Γ 1 k u 1 ∥ L 2 2 + c β k k b + 1 2 b k I 0 ( u 1 ) . (37)
Similarly, replacing the role of u 1 u_1 u 1 and u 0 u_0 u 0 and α \alpha α and β \beta β we deduce that,
I a k ( u 0 ) = ∫ R ∣ x ∣ 2 a k ∣ v ~ − 1 ( x ) ∣ 2 d x ≤ 1 2 2 − 2 k ∥ Γ − 1 k u 0 ∥ L 2 2 + c α k k 2 a a + 1 k I 0 ( u 0 ) , (38) I_{ak}(u_0)=\int_{\mathbb{R}}|x|^{2ak}|\widetilde{v}_{-1}(x)|^2dx\leq\frac{1}{2}2^{-2k}\|\Gamma_{-1}^k u_0\|_{L^2}^2+c_\alpha^k k^{\frac{2a}{a+1}k}I_0(u_0), \tag*{(38)} I ak ( u 0 ) = ∫ R ∣ x ∣ 2 ak ∣ v − 1 ( x ) ∣ 2 d x ≤ 2 1 2 − 2 k ∥ Γ − 1 k u 0 ∥ L 2 2 + c α k k a + 1 2 a k I 0 ( u 0 ) , (38)
for 0 < a < 1 α − 1 0<a<\frac{1}{\alpha}-1 0 < a < α 1 − 1 , and k ≥ k 0 k\geq k_0 k ≥ k 0 , by increasing k 0 k_0 k 0 if necessary.
Controlling the evolution of tk u in L2
The aim of this subsection is to control the L 2 L^2 L 2 -norms of Γ t k u \Gamma_t^k u Γ t k u and Γ − t v ~ \Gamma_{-t}\widetilde{v} Γ − t v that appear on the right-hand side of (37) -(38) .
Proposition 4.1. For any integer k ≥ 1 k\geq1 k ≥ 1
1 2 sup t ∈ [ 0 , 1 ] ∥ Γ t k u ( t ) ∥ L 2 ≤ ∥ x k u 0 ∥ L 2 + c k ( k ! ) τ ∥ u 0 ∥ L 2 , (39) \frac{1}{2}\sup_{t\in[0,1]}\|\Gamma_t^k u(t)\|_{L^2}\leq\|x^k u_0\|_{L^2}+c^k(k!)^\tau\|u_0\|_{L^2}, \tag*{(39)} 2 1 t ∈ [ 0 , 1 ] sup ∥ Γ t k u ( t ) ∥ L 2 ≤ ∥ x k u 0 ∥ L 2 + c k ( k ! ) τ ∥ u 0 ∥ L 2 , (39) 1 2 sup t ∈ [ 0 , 1 ] ∥ Γ − t k v ( t ) ∥ L 2 ≤ ∥ x k u 1 ∥ L 2 + c k ( k ! ) τ ∥ u 0 ∥ L 2 . (40) \frac{1}{2}\sup_{t\in[0,1]}\|\Gamma_{-t}^k v(t)\|_{L^2}\leq\|x^k u_1\|_{L^2}+c^k(k!)^\tau\|u_0\|_{L^2}. \tag*{(40)} 2 1 t ∈ [ 0 , 1 ] sup ∥ Γ − t k v ( t ) ∥ L 2 ≤ ∥ x k u 1 ∥ L 2 + c k ( k ! ) τ ∥ u 0 ∥ L 2 . (40)
where
τ = { 1 , when 0 ≤ σ < 1 , σ + , when σ ≥ 1 , \tau=
\begin{cases}
1, & \text{when } 0 \le\sigma< 1,\\
\sigma^{+}, & \text{when } \sigma\ge1,
\end{cases} τ = { 1 , σ + , when 0 ≤ σ < 1 , when σ ≥ 1 ,
where σ + \sigma^{+} σ + is any number slightly larger than σ \sigma σ .
Proof of Lemma 4.1. To prove (4.4), we follow the same strategy as in Section 3, but now we will also track the constant to get a more precise decay estimate. By using (3.3) and Hölder’s inequality, we get
1 2 d d t ∥ Γ t k u ( t ) ∥ L 2 2 ≤ ∑ l = 0 k − 1 ( k l ) ( 2 t ) 2 k ∥ ∂ x k − l V ∥ L ∞ ∥ ∂ x l u ~ ( t ) ∥ L 2 ∥ ∂ x k u ~ ( t ) ∥ L 2 = ∑ l = 0 k − 1 ( k l ) ( 2 t ) k ∥ ∂ x k − l V ∥ L ∞ ∥ ∂ x l u ~ ( t ) ∥ L 2 ∥ Γ t k u ( t ) ∥ L 2 . (41) \begin{aligned}
\frac{1}{2}\frac{d}{dt}\|\Gamma_t^k u(t)\|_{L^2}^2
\le\sum_{l=0}^{k-1} \binom{k}{l}(2t)^{2k}\|\partial_x^{k-l}V\|_{L^\infty}\|\partial_x^l\widetilde{u}(t)\|_{L^2}\|\partial_x^k\widetilde{u}(t)\|_{L^2} \\
&= \sum_{l=0}^{k-1} \binom{k}{l}(2t)^k\|\partial_x^{k-l}V\|_{L^\infty}\|\partial_x^l\widetilde{u}(t)\|_{L^2}\|\Gamma_t^k u(t)\|_{L^2}. \tag*{(41)}
\end{aligned} 2 1 d t d ∥ Γ t k u ( t ) ∥ L 2 2 ≤ l = 0 ∑ k − 1 ( l k ) ( 2 t ) 2 k ∥ ∂ x k − l V ∥ L ∞ ∥ ∂ x l u ( t ) ∥ L 2 ∥ ∂ x k u ( t ) ∥ L 2 = l = 0 ∑ k − 1 ( l k ) ( 2 t ) k ∥ ∂ x k − l V ∥ L ∞ ∥ ∂ x l u ( t ) ∥ L 2 ∥ Γ t k u ( t ) ∥ L 2 . (41)
This implies that
d d t ∥ Γ t k u ( t ) ∥ L 2 ≤ ∑ l = 0 k − 1 ( k l ) ( 2 t ) k ∥ ∂ x k − l V ∥ L ∞ ∥ ∂ x l u ~ ( t ) ∥ L 2 . \frac{d}{dt}\|\Gamma_t^k u(t)\|_{L^2}
\le\sum_{l=0}^{k-1}\binom{k}{l}(2t)^k\|\partial_x^{k-l}V\|_{L^\infty}\|\partial_x^l\widetilde{u}(t)\|_{L^2}. d t d ∥ Γ t k u ( t ) ∥ L 2 ≤ l = 0 ∑ k − 1 ( l k ) ( 2 t ) k ∥ ∂ x k − l V ∥ L ∞ ∥ ∂ x l u ( t ) ∥ L 2 .
For all 1 ≤ l ≤ k − 1 1 \le l \le k-1 1 ≤ l ≤ k − 1 , let θ = θ l > 0 \theta=\theta_l>0 θ = θ l > 0 to be fixed later. Then
∥ ∂ x l u ~ ∥ L 2 = ( 2 π ) l ∥ ξ l u ~ ^ ( t ) ∥ L 2 = ( 2 π ) l ∥ 1 ∣ ξ ∣ ≤ θ ξ l u ~ ^ ( t ) ∥ L 2 + ( 2 π ) l ∥ 1 ∣ ξ ∣ > θ u ~ ^ ( t ) ∥ L 2 ≤ ( 2 π ) l θ l ∥ u 0 ∥ L 2 + ( 2 π ) l − k θ − ( k − l ) ∥ ∂ x k u ~ ( t ) ∥ L 2 = ( 2 π θ ) l ∥ u 0 ∥ L 2 + ( 2 π θ ) − ( k − l ) ( 2 t ) − k ∥ Γ t k u ( t ) ∥ L 2 . \begin{aligned}
\|\partial_x^l\widetilde{u}\|_{L^2}
=(2\pi)^l\|\xi^l\widehat{\widetilde{u}}(t)\|_{L^2}
=(2\pi)^l\|\boldsymbol{1}_{|\xi|\le\theta}\xi^l\widehat{\widetilde{u}}(t)\|_{L^2}
+(2\pi)^l\|\boldsymbol{1}_{|\xi|>\theta}\widehat{\widetilde{u}}(t)\|_{L^2} \\
&\le(2\pi)^l\theta^l\|u_0\|_{L^2}
+(2\pi)^{l-k}\theta^{-(k-l)}\|\partial_x^k\widetilde{u}(t)\|_{L^2} \\
&=(2\pi\theta)^l\|u_0\|_{L^2}
+(2\pi\theta)^{-(k-l)}(2t)^{-k}\|\Gamma_t^k u(t)\|_{L^2}.
\end{aligned} ∥ ∂ x l u ∥ L 2 = ( 2 π ) l ∥ ξ l u ( t ) ∥ L 2 = ( 2 π ) l ∥ 1 ∣ ξ ∣ ≤ θ ξ l u ( t ) ∥ L 2 + ( 2 π ) l ∥ 1 ∣ ξ ∣ > θ u ( t ) ∥ L 2 ≤ ( 2 π ) l θ l ∥ u 0 ∥ L 2 + ( 2 π ) l − k θ − ( k − l ) ∥ ∂ x k u ( t ) ∥ L 2 = ( 2 π θ ) l ∥ u 0 ∥ L 2 + ( 2 π θ ) − ( k − l ) ( 2 t ) − k ∥ Γ t k u ( t ) ∥ L 2 .
By using that sup t ∈ [ 0 , 1 ] ∥ ∂ x l V ∥ L ∞ ≤ A l ( l ! ) σ \sup_{t\in[0,1]}\|\partial_x^lV\|_{L^\infty}\le A^l(l!)^\sigma sup t ∈ [ 0 , 1 ] ∥ ∂ x l V ∥ L ∞ ≤ A l ( l ! ) σ , it follows that
d d t ∥ Γ t k u ( t ) ∥ L 2 ≤ ∑ l = 0 k − 1 ( k l ) ( 2 t ) k A k − l ( ( k − l ) ! ) σ ( 2 π θ ) l ∥ u 0 ∥ L 2 + ∑ l = 1 k − 1 ( k l ) ( 2 π A − 1 θ ) − ( k − l ) ( ( k − l ) ! ) σ ∥ Γ t k u ( t ) ∥ L 2 . (42) \begin{aligned}
\frac{d}{dt}\|\Gamma_t^k u(t)\|_{L^2}
\le\sum_{l=0}^{k-1}\binom{k}{l}(2t)^kA^{k-l}((k-l)!)^\sigma(2\pi\theta)^l\|u_0\|_{L^2} \\
&+\sum_{l=1}^{k-1}\binom{k}{l}(2\pi A^{-1}\theta)^{-(k-l)}((k-l)!)^\sigma\|\Gamma_t^k u(t)\|_{L^2}. \tag*{(42)}
\end{aligned} d t d ∥ Γ t k u ( t ) ∥ L 2 ≤ l = 0 ∑ k − 1 ( l k ) ( 2 t ) k A k − l (( k − l )! ) σ ( 2 π θ ) l ∥ u 0 ∥ L 2 + l = 1 ∑ k − 1 ( l k ) ( 2 π A − 1 θ ) − ( k − l ) (( k − l )! ) σ ∥ Γ t k u ( t ) ∥ L 2 . (42)
and integrating (4.6) from 0 to t t t , we get that
sup t ∈ [ 0 , 1 ] ∥ Γ t k u ( t ) ∥ L 2 ≤ ∥ x k u 0 ∥ L 2 + ∑ l = 0 k − 1 ( k l ) 2 k A k − l ( ( k − l ) ! ) σ ( 2 π θ ) l ∥ u 0 ∥ L 2 + ∑ l = 1 k − 1 ( k l ) ( 2 π A − 1 θ ) − ( k − l ) ( ( k − l ) ! ) σ sup t ∈ [ 0 , 1 ] ∥ Γ t k u ( t ) ∥ L 2 . (43) \begin{aligned}
\sup_{t\in[0,1]}\|\Gamma_t^k u(t)\|_{L^2}
\le\|x^k u_0\|_{L^2}
+\sum_{l=0}^{k-1}\binom{k}{l}2^kA^{k-l}((k-l)!)^\sigma(2\pi\theta)^l\|u_0\|_{L^2} \\
&+\sum_{l=1}^{k-1}\binom{k}{l}(2\pi A^{-1}\theta)^{-(k-l)}((k-l)!)^\sigma
\sup_{t\in[0,1]}\|\Gamma_t^k u(t)\|_{L^2}. \tag*{(43)}
\end{aligned} t ∈ [ 0 , 1 ] sup ∥ Γ t k u ( t ) ∥ L 2 ≤ ∥ x k u 0 ∥ L 2 + l = 0 ∑ k − 1 ( l k ) 2 k A k − l (( k − l )! ) σ ( 2 π θ ) l ∥ u 0 ∥ L 2 + l = 1 ∑ k − 1 ( l k ) ( 2 π A − 1 θ ) − ( k − l ) (( k − l )! ) σ t ∈ [ 0 , 1 ] sup ∥ Γ t k u ( t ) ∥ L 2 . (43)
We let J 0 ∈ N J_0\in\mathbb{N} J 0 ∈ N , J 0 = J 0 ( σ ) ≥ 2 J_0=J_0(\sigma)\ge2 J 0 = J 0 ( σ ) ≥ 2 , and ρ > 0 \rho>0 ρ > 0 to be fixed later. Let
θ = { k σ 2 π ρ − 1 , 1 ≤ l ≤ ⌊ k J 0 ⌋ , l max { 1 , σ } A 2 π ρ − 1 , ⌊ k J 0 ⌋ + 1 ≤ l ≤ k − 1. \theta=
\begin{cases}
k^{{\frac{\sigma}{2\pi}}\rho^{-1}}, & 1\le l\le\left\lfloor\frac{k}{J_0}\right\rfloor,\\
l^{\max\{1,\sigma\}}{\frac{A}{2\pi}}\rho^{-1}, & \left\lfloor\frac{k}{J_0}\right\rfloor+1\le l\le k-1.
\end{cases} θ = ⎩ ⎨ ⎧ k 2 π σ ρ − 1 , l m a x { 1 , σ } 2 π A ρ − 1 , 1 ≤ l ≤ ⌊ J 0 k ⌋ , ⌊ J 0 k ⌋ + 1 ≤ l ≤ k − 1.
We will estimate the two sums separately. By Stirling’s formula it follows, in the case where 1 ≤ l ≤ ⌊ k J 0 ⌋ 1\le l\le\left\lfloor\frac{k}{J_0}\right\rfloor 1 ≤ l ≤ ⌊ J 0 k ⌋ ,
∑ l = 0 ⌊ k / J 0 ⌋ ( k l ) 2 k A k − l ( ( k − l ) ! ) σ ( 2 π θ ) l ≤ k ! ( 4 π A ) k ∑ l = 0 ⌊ k / J 0 ⌋ ( ( k − l ) ! ) σ − 1 ρ − l k σ k J 0 ≤ { c k ( k ! ) σ + 1 J 0 , σ ≤ 1 , c k ( k ! ) ( 1 + 1 J 0 ) σ , σ > 1. (44) \sum_{l=0}^{\left\lfloor k/J_0\right\rfloor}\binom{k}{l}2^kA^{k-l}((k-l)!)^\sigma(2\pi\theta)^l
\le k!(4\pi A)^k\sum_{l=0}^{\left\lfloor k/J_0\right\rfloor}((k-l)!)^{\sigma-1}\rho^{-l}k^{\frac{\sigma k}{J_0}}
\le
\begin{cases}
c^k(k!)^{\sigma+\frac{1}{J_0}}, & \sigma\le1,\\
c^k(k!)^{(1+\frac{1}{J_0})\sigma}, & \sigma>1.
\end{cases} \tag*{(44)} l = 0 ∑ ⌊ k / J 0 ⌋ ( l k ) 2 k A k − l (( k − l )! ) σ ( 2 π θ ) l ≤ k ! ( 4 π A ) k l = 0 ∑ ⌊ k / J 0 ⌋ (( k − l )! ) σ − 1 ρ − l k J 0 σ k ≤ { c k ( k ! ) σ + J 0 1 , c k ( k ! ) ( 1 + J 0 1 ) σ , σ ≤ 1 , σ > 1. (44)
In the case where l > ⌊ k J 0 ⌋ l > \left\lfloor\frac{k}{J_0} \right\rfloor l > ⌊ J 0 k ⌋ , we need to bound
∑ l = ⌊ k / J 0 ⌋ + 1 k − 1 ( k l ) 2 k A k − l ( ( k − l ) ! ) σ ( 2 π θ ) l ≤ c k ∑ l = ⌊ k / J 0 ⌋ + 1 k − 1 k ! ( ( k − l ) ! ) σ − 1 l max { 1 , σ } − 1 . \sum_{l=\left\lfloor k/J_0\right\rfloor+1}^{k-1} \binom{k}{l}2^k A^{k-l}((k-l)!)^\sigma(2\pi\theta)^l \le c^k \sum_{l=\left\lfloor k/J_0\right\rfloor+1}^{k-1} k!((k-l)!)^{\sigma-1}l^{\max\{1,\sigma\}-1}. l = ⌊ k / J 0 ⌋ + 1 ∑ k − 1 ( l k ) 2 k A k − l (( k − l )! ) σ ( 2 π θ ) l ≤ c k l = ⌊ k / J 0 ⌋ + 1 ∑ k − 1 k ! (( k − l )! ) σ − 1 l m a x { 1 , σ } − 1 .
When σ ≤ 1 \sigma\le1 σ ≤ 1 , we get the rough bound
∑ l = ⌊ k / J 0 ⌋ + 1 k − 1 ( k l ) 2 k A k − l ( ( k − l ) ! ) σ ( 2 π θ ) l ≤ C k k ! . (45) \sum_{l=\left\lfloor k/J_0\right\rfloor+1}^{k-1} \binom{k}{l}2^k A^{k-l}((k-l)!)^\sigma(2\pi\theta)^l \le C^k k!. \tag*{(45)} l = ⌊ k / J 0 ⌋ + 1 ∑ k − 1 ( l k ) 2 k A k − l (( k − l )! ) σ ( 2 π θ ) l ≤ C k k ! . (45)
When σ > 1 \sigma> 1 σ > 1 , we observe from the Stirling formula that
( k − l ) ! l l ≤ ( k − l ) ( k − l ) l l = k k k k ( k − l ) ( k − l ) l l = ( k − l k ) k − l ( l k ) l k k ≤ c k k ! . (k-l)!l^l \le(k-l)^{(k-l)}l^l = \frac{k^k}{k^k}(k-l)^{(k-l)}l^l = \left(\frac{k-l}{k}\right)^{k-l}\left(\frac{l}{k}\right)^l k^k \le c^k k!. ( k − l )! l l ≤ ( k − l ) ( k − l ) l l = k k k k ( k − l ) ( k − l ) l l = ( k k − l ) k − l ( k l ) l k k ≤ c k k ! .
Thus,
∑ l = ⌊ k / J 0 ⌋ + 1 k − 1 ( k l ) 2 k A k − l ( ( k − l ) ! ) σ ( 2 π θ ) l ≤ c k k ! ∑ l = ⌊ k / J 0 ⌋ + 1 k − 1 ( ( k − l ) ! l l ) σ − 1 ≤ c k ( k ! ) σ . (46) \sum_{l=\left\lfloor k/J_0\right\rfloor+1}^{k-1} \binom{k}{l}2^k A^{k-l}((k-l)!)^\sigma(2\pi\theta)^l \le c^k k! \sum_{l=\left\lfloor k/J_0\right\rfloor+1}^{k-1}((k-l)!l^l)^{\sigma-1} \le c^k(k!)^\sigma. \tag*{(46)} l = ⌊ k / J 0 ⌋ + 1 ∑ k − 1 ( l k ) 2 k A k − l (( k − l )! ) σ ( 2 π θ ) l ≤ c k k ! l = ⌊ k / J 0 ⌋ + 1 ∑ k − 1 (( k − l )! l l ) σ − 1 ≤ c k ( k ! ) σ . (46)
For the other sum in (43) , we will prove the following claim.
Claim 4.1. For l , k ∈ N l,k \in\mathbb{N} l , k ∈ N with 1 ≤ l ≤ k − 1 1 \le l \le k-1 1 ≤ l ≤ k − 1 , let
S l , k = ( k l ) ( ( k − l ) ! ) σ θ l − ( k − l ) A k − l = { ( k l ) ( ( k − l ) ! ) σ h − σ ( k − l ) ρ k − l , 1 ≤ l ≤ ⌊ k J 0 ⌋ , ( k l ) ( ( k − l ) ! ) σ l − max { 1 , σ } ( k − l ) ρ k − l , ⌊ k J 0 ⌋ + 1 ≤ l ≤ k − 1. S_{l,k}=\binom{k}{l}((k-l)!)^\sigma\theta_l^{-(k-l)}A^{k-l} =
\begin{cases}
\binom{k}{l}((k-l)!)^\sigma h^{-\sigma(k-l)}\rho^{k-l}, & 1 \le l \le\left\lfloor\frac{k}{J_0}\right\rfloor,\\
\binom{k}{l}((k-l)!)^\sigma l^{-\max\{1,\sigma\}(k-l)}\rho^{k-l}, & \left\lfloor\frac{k}{J_0}\right\rfloor+1 \le l \le k-1.
\end{cases} S l , k = ( l k ) (( k − l )! ) σ θ l − ( k − l ) A k − l = ⎩ ⎨ ⎧ ( l k ) (( k − l )! ) σ h − σ ( k − l ) ρ k − l , ( l k ) (( k − l )! ) σ l − m a x { 1 , σ } ( k − l ) ρ k − l , 1 ≤ l ≤ ⌊ J 0 k ⌋ , ⌊ J 0 k ⌋ + 1 ≤ l ≤ k − 1. Then, there exists ρ 0 = ρ 0 ( σ ) > 0 \rho_0=\rho_0(\sigma)>0 ρ 0 = ρ 0 ( σ ) > 0 such that, for all 0 < ρ < ρ 0 0<\rho<\rho_0 0 < ρ < ρ 0 , for all k ≥ 2 k\ge2 k ≥ 2 ,
∑ l = 1 k − 1 S l , k ≤ 1 2 . (47) \sum_{l=1}^{k-1}S_{l,k}\le\frac{1}{2}. \tag*{(47)} l = 1 ∑ k − 1 S l , k ≤ 2 1 . (47)
Assuming the claim, it follows that
∑ l = 1 k − 1 ( k l ) ( A θ − 1 ) k − l ( ( k − l ) ! ) σ sup t ∈ [ 0 , 1 ] ∥ Γ t k u ( t ) ∥ L 2 ≤ 1 2 sup t ∈ [ 0 , 1 ] ∥ Γ t k u ( t ) ∥ L 2 , \sum_{l=1}^{k-1}\binom{k}{l}(A\theta^{-1})^{k-l}((k-l)!)^\sigma\sup_{t\in[0,1]}\left\|\Gamma_t^k u(t)\right\|_{L^2}
\le\frac{1}{2}\sup_{t\in[0,1]}\left\|\Gamma_t^k u(t)\right\|_{L^2}, l = 1 ∑ k − 1 ( l k ) ( A θ − 1 ) k − l (( k − l )! ) σ t ∈ [ 0 , 1 ] sup Γ t k u ( t ) L 2 ≤ 2 1 t ∈ [ 0 , 1 ] sup Γ t k u ( t ) L 2 ,
and combining this with (4.8) and (4.9), we conclude the proof of (4.4). It remains to prove Claim 4.1.
Proof of Claim 4.1. First let M ∈ N M\in\mathbb{N} M ∈ N , M > J 0 M>J_0 M > J 0 . Let m = k − l m=k-l m = k − l . Observe that for k ≤ M k\le M k ≤ M , and 1 ≤ l ≤ k − 1 1\le l\le k-1 1 ≤ l ≤ k − 1 , since ( A θ − 1 ) m ≤ ρ m (A\theta^{-1})^m\le\rho^m ( A θ − 1 ) m ≤ ρ m as long as σ ≥ 0 \sigma\ge0 σ ≥ 0 , we always have
S l , k ≤ ( k m ) ( m ! ) σ ρ m ≤ 2 M ( M ! ) σ ρ m ≤ ( 2 M ( M ! ) σ ρ ) m ≤ ( 1 4 ) m S_{l,k}\le\binom{k}{m}(m!)^\sigma\rho^m\le2^M(M!)^\sigma\rho^m\le\left(2^M(M!)^\sigma\rho\right)^m\le\left(\frac{1}{4}\right)^m S l , k ≤ ( m k ) ( m ! ) σ ρ m ≤ 2 M ( M ! ) σ ρ m ≤ ( 2 M ( M ! ) σ ρ ) m ≤ ( 4 1 ) m
for ρ < 1 4 ( 2 M ( M ! ) σ ) − 1 \rho<\frac{1}{4}\left(2^M(M!)^\sigma\right)^{-1} ρ < 4 1 ( 2 M ( M ! ) σ ) − 1 . Thus we can assume k ≥ M > J 0 k\ge M>J_0 k ≥ M > J 0 .
We split the proof into two cases depending on whether l l l is larger or smaller than ⌊ k J 0 ⌋ \left\lfloor\frac{k}{J_0}\right\rfloor ⌊ J 0 k ⌋ .
Case ⌊ k J 0 ⌋ ≤ l ≤ k − 1 \left\lfloor\frac{k}{J_0}\right\rfloor\le l\le k-1 ⌊ J 0 k ⌋ ≤ l ≤ k − 1 . Then, l ≥ k J 0 − 1 ≥ k J 0 − k M = k ( M − J 0 M J 0 ) = : k N 0 l\ge\frac{k}{J_0}-1\ge\frac{k}{J_0}-\frac{k}{M}=k\left(\frac{M-J_0}{MJ_0}\right)=:kN_0 l ≥ J 0 k − 1 ≥ J 0 k − M k = k ( M J 0 M − J 0 ) =: k N 0 . (recall that k ≥ M k\ge M k ≥ M ). Let m = k − l m=k-l m = k − l . Since ( k l ) = ( k m ) ≤ k m m ! \binom{k}{l}=\binom{k}{m}\le\frac{k^m}{m!} ( l k ) = ( m k ) ≤ m ! k m , we have
S l , k = ( k m ) ( m ! ) σ l − max { 1 , σ } m ρ m ≤ k m m ! ( m ! ) σ ( ρ l max { 1 , σ } ) m . S_{l,k}=\binom{k}{m}(m!)^\sigma l^{-\max\{1,\sigma\}m}\rho^m\le\frac{k^m}{m!}(m!)^\sigma\left(\frac{\rho}{l^{\max\{1,\sigma\}}}\right)^m. S l , k = ( m k ) ( m ! ) σ l − m a x { 1 , σ } m ρ m ≤ m ! k m ( m ! ) σ ( l m a x { 1 , σ } ρ ) m .
Then, when σ ≤ 1 \sigma\le1 σ ≤ 1 , it follows that
S l , k ≤ k m ( ρ k N 0 ) m = ( N 0 − 1 ρ ) m < ( 1 4 ) m , S_{l,k}\le k^m\left(\frac{\rho}{kN_0}\right)^m=(N_0^{-1}\rho)^m<\left(\frac{1}{4}\right)^m, S l , k ≤ k m ( k N 0 ρ ) m = ( N 0 − 1 ρ ) m < ( 4 1 ) m ,
for ρ < N 0 4 \rho< \frac{N_0}{4} ρ < 4 N 0 . When σ > 1 \sigma> 1 σ > 1 ,
S l , k ≤ k m ( 1 − σ ) ( m ! ) σ − 1 ( N 0 − σ ρ ) m . S_{l,k} \leq k^{m(1-\sigma)}(m!)^{\sigma-1}(N_0^{-\sigma}\rho)^m. S l , k ≤ k m ( 1 − σ ) ( m ! ) σ − 1 ( N 0 − σ ρ ) m .
Since m ! ≤ m m m! \leq m^m m ! ≤ m m , it follows that
S l , k = k m ( 1 − σ ) m m ( σ − 1 ) ( N 0 − σ ρ ) m ≤ e m ( 1 − σ ) ( log k − log m ) ( N 0 − σ ρ ) m ≤ ( N 0 − σ ρ ) m < ( 1 4 ) m S_{l,k}=k^{m(1-\sigma)}m^{m(\sigma-1)}(N_0^{-\sigma}\rho)^m \leq e^{m(1-\sigma)(\log k-\log m)}(N_0^{-\sigma}\rho)^m \leq(N_0^{-\sigma}\rho)^m < \left(\frac{1}{4}\right)^m S l , k = k m ( 1 − σ ) m m ( σ − 1 ) ( N 0 − σ ρ ) m ≤ e m ( 1 − σ ) ( l o g k − l o g m ) ( N 0 − σ ρ ) m ≤ ( N 0 − σ ρ ) m < ( 4 1 ) m
for ρ < N 0 σ 4 \rho< \frac{N_0^\sigma}{4} ρ < 4 N 0 σ .
Thus, for ⌊ k J 0 ⌋ < l ≤ k − 1 \left\lfloor\frac{k}{J_0}\right\rfloor< l \leq k-1 ⌊ J 0 k ⌋ < l ≤ k − 1 , S l , k ≤ ( 1 4 ) m S_{l,k} \leq\left(\frac{1}{4}\right)^m S l , k ≤ ( 4 1 ) m for ρ < ρ 1 : = 1 4 min { N 0 , N 0 σ , ( 2 M ( M ! ) σ ) − 1 } \rho< \rho_1 := \frac{1}{4}\min\{N_0,N_0^\sigma,(2^M(M!)^\sigma)^{-1}\} ρ < ρ 1 := 4 1 min { N 0 , N 0 σ , ( 2 M ( M ! ) σ ) − 1 } .
Case 1 ≤ l ≤ ⌊ k J 0 ⌋ \leq l \leq\left\lfloor\frac{k}{J_0}\right\rfloor ≤ l ≤ ⌊ J 0 k ⌋ . Then k ≥ J 0 l k \geq J_0l k ≥ J 0 l . First, we observe that S l , k ≥ S l , k + 1 S_{l,k} \geq S_{l,k+1} S l , k ≥ S l , k + 1 . Indeed,
S l , k + 1 S l , k = ( k + 1 l ) ( ( k + 1 − l ) ! ) σ ( k + 1 ) − σ ( k + 1 − l ) ρ k + 1 − l ( k l ) ( ( k − l ) ! ) σ k − σ ( k − l ) ρ k − l = k + 1 k + 1 − l ( k + 1 − l ) σ ( k k + 1 ) σ ( k − l ) ( k + 1 ) − σ ρ . \frac{S_{l,k+1}}{S_{l,k}}=\frac{\binom{k+1}{l}((k+1-l)!)^\sigma(k+1)^{-\sigma(k+1-l)}\rho^{k+1-l}}{\binom{k}{l}((k-l)!)^\sigma k^{-\sigma(k-l)}\rho^{k-l}}=\frac{k+1}{k+1-l}(k+1-l)^\sigma\left(\frac{k}{k+1}\right)^{\sigma(k-l)}(k+1)^{-\sigma}\rho. S l , k S l , k + 1 = ( l k ) (( k − l )! ) σ k − σ ( k − l ) ρ k − l ( l k + 1 ) (( k + 1 − l )! ) σ ( k + 1 ) − σ ( k + 1 − l ) ρ k + 1 − l = k + 1 − l k + 1 ( k + 1 − l ) σ ( k + 1 k ) σ ( k − l ) ( k + 1 ) − σ ρ .
When we fix l ≤ k / J 0 l \leq k/J_0 l ≤ k / J 0 , the function k ↦ k + 1 k + 1 − l k\mapsto\frac{k+1}{k+1-l} k ↦ k + 1 − l k + 1 is decreasing, so
k + 1 k + 1 − l ≤ J 0 l + 1 J 0 l + 1 − l = J 0 l + 1 l ( J 0 − 1 ) + 1 = 1 + l ( J 0 − 1 ) l + 1 ≤ 1 + 1 J 0 − 1 ≤ 2 , \frac{k+1}{k+1-l}\leq\frac{J_0l+1}{J_0l+1-l}=\frac{J_0l+1}{l(J_0-1)+1}=1+\frac{l}{(J_0-1)l+1}\leq1+\frac{1}{J_0-1}\leq2, k + 1 − l k + 1 ≤ J 0 l + 1 − l J 0 l + 1 = l ( J 0 − 1 ) + 1 J 0 l + 1 = 1 + ( J 0 − 1 ) l + 1 l ≤ 1 + J 0 − 1 1 ≤ 2 ,
for J 0 ≥ 2 J_0\geq2 J 0 ≥ 2 . Thus,
S l , k + 1 S l , k ≤ 2 ( k + 1 − l ) σ ( k + 1 ) σ ( k k + 1 ) σ ( k − l ) ρ ≤ 2 ρ , \frac{S_{l,k+1}}{S_{l,k}}\leq2\frac{(k+1-l)^\sigma}{(k+1)^\sigma}\left(\frac{k}{k+1}\right)^{\sigma(k-l)}\rho\leq2\rho, S l , k S l , k + 1 ≤ 2 ( k + 1 ) σ ( k + 1 − l ) σ ( k + 1 k ) σ ( k − l ) ρ ≤ 2 ρ ,
since σ ≥ 0 \sigma\geq0 σ ≥ 0 . So, S l , k + 1 S l , k < 1 \frac{S_{l,k+1}}{S_{l,k}}<1 S l , k S l , k + 1 < 1 when ρ < 1 2 \rho<\frac{1}{2} ρ < 2 1 .
It follows that S l , k ≤ S l , J 0 l S_{l,k}\leq S_{l,J_0l} S l , k ≤ S l , J 0 l for k ≥ J 0 l k\geq J_0l k ≥ J 0 l . Moreover, we have by using the Stirling formula,
S l , J 0 l = ( J 0 l l ) ( ( ( J 0 − 1 ) l ) ! ) σ ( J 0 l ) − σ ( J 0 − 1 ) l κ ( J 0 − 1 ) l ≤ ( J 0 l ) l l ! ( ( ( J 0 − 1 ) l ) ! ) σ ( J 0 l ) − σ ( J 0 − 1 ) l κ ( J 0 − 1 ) l ≤ C l J 0 l ( ( ( J 0 − 1 ) l ) ) σ ( J 0 − 1 ) l ( J 0 l ) − σ ( J 0 − 1 ) l κ ( J 0 − 1 ) l ≤ C l J 0 l κ ( J 0 − 1 ) l ≤ ( 1 4 ) l \begin{aligned}
S_{l,J_0l}=\binom{J_0l}{l}(((J_0-1)l)!)^\sigma(J_0l)^{-\sigma(J_0-1)l}\kappa^{(J_0-1)l} \\
&\leq\frac{(J_0l)^l}{l!}(((J_0-1)l)!)^\sigma(J_0l)^{-\sigma(J_0-1)l}\kappa^{(J_0-1)l} \\
&\leq C^lJ_0^l(((J_0-1)l))^{\sigma(J_0-1)l}(J_0l)^{-\sigma(J_0-1)l}\kappa^{(J_0-1)l} \\
&\leq C^lJ_0^l\kappa^{(J_0-1)l}\leq\left(\frac{1}{4}\right)^l
\end{aligned} S l , J 0 l = ( l J 0 l ) ((( J 0 − 1 ) l )! ) σ ( J 0 l ) − σ ( J 0 − 1 ) l κ ( J 0 − 1 ) l ≤ l ! ( J 0 l ) l ((( J 0 − 1 ) l )! ) σ ( J 0 l ) − σ ( J 0 − 1 ) l κ ( J 0 − 1 ) l ≤ C l J 0 l ((( J 0 − 1 ) l ) ) σ ( J 0 − 1 ) l ( J 0 l ) − σ ( J 0 − 1 ) l κ ( J 0 − 1 ) l ≤ C l J 0 l κ ( J 0 − 1 ) l ≤ ( 4 1 ) l
for κ < κ 2 : = ( 4 C J 0 ) 1 − J 0 \kappa<\kappa_2:=(4CJ_0)^{1-J_0} κ < κ 2 := ( 4 C J 0 ) 1 − J 0 . This concludes the proof of the claim by choosing
0 < κ < κ 0 = min { κ 1 , κ 2 } . 0<\kappa<\kappa_0=\min\{\kappa_1,\kappa_2\}. 0 < κ < κ 0 = min { κ 1 , κ 2 } .
Therefore, we deduce gathering (4.8), (4.9), (4.10) and Claim 4.1 that
1 2 sup t ∈ [ 0 , 1 ] ∥ Γ t k u ( t ) ∥ L 2 ≤ ∥ x k u 0 ∥ L 2 + c k ( k ! ) τ ~ ∥ u ~ 0 ∥ L 2 (48) \frac{1}{2}\sup_{t\in[0,1]}\|\Gamma_t^k u(t)\|_{L^2}\leq\|x^ku_0\|_{L^2}+c^k(k!)^{\widetilde{\tau}}\|\widetilde{u}_0\|_{L^2} \tag*{(48)} 2 1 t ∈ [ 0 , 1 ] sup ∥ Γ t k u ( t ) ∥ L 2 ≤ ∥ x k u 0 ∥ L 2 + c k ( k ! ) τ ∥ u 0 ∥ L 2 (48)
where
τ ~ = { max { σ + 1 J 0 , 1 } , 0 ≤ σ < 1 , σ ( 1 + 1 J 0 ) , σ ≥ 1. \widetilde{\tau}=
\begin{cases}
\max\{\sigma+\frac{1}{J_0},1\}, & 0\leq\sigma<1,\\
\sigma(1+\frac{1}{J_0}), & \sigma\geq1.
\end{cases} τ = { max { σ + J 0 1 , 1 } , σ ( 1 + J 0 1 ) , 0 ≤ σ < 1 , σ ≥ 1.
When 0 ≤ σ < 1 0\leq\sigma<1 0 ≤ σ < 1 , we choose J 0 J_0 J 0 large enough so that σ + 1 J 0 < 1 \sigma+\frac{1}{J_0}<1 σ + J 0 1 < 1 . When σ ≥ 1 \sigma\geq1 σ ≥ 1 , we choose J 0 J_0 J 0 large enough so that σ ( 1 + 1 J 0 ) < σ + \sigma(1+\frac{1}{J_0})<\sigma^+ σ ( 1 + J 0 1 ) < σ + , for any number σ + > σ \sigma^+>\sigma σ + > σ . This concludes the proof of (4.4). Arguing similarly, for v ( x , t ) = u ( x , 1 − t ) v(x,t)=u(x,1-t) v ( x , t ) = u ( x , 1 − t ) and Γ − t k v ( t ) \Gamma_{-t}^kv(t) Γ − t k v ( t ) , we deduce (4.5). This concludes the proof of Lemma 4.1.
Conclusion of the proof
By inserting (39) and (40) into (38) and (37) we obtain, for k ≥ k 0 k \ge k_0 k ≥ k 0 ,
∫ R ∣ x ∣ 2 b k ∣ u 1 ( x ) ∣ 2 d x ≤ 1 2 ∥ x k u 0 ∥ L 2 2 + ( c k ( k ! ) 2 τ + c β k k 2 b b + 1 k ) I 0 ( u 0 ) . (49) \int_{\mathbb{R}} |x|^{2bk}|u_1(x)|^2\,dx \le\frac{1}{2}\|x^k u_0\|_{L^2}^2+\left(c^k(k!)^{2\tau}+c_\beta^k k^{\frac{2b}{b+1}k}\right)I_0(u_0). \tag*{(49)} ∫ R ∣ x ∣ 2 bk ∣ u 1 ( x ) ∣ 2 d x ≤ 2 1 ∥ x k u 0 ∥ L 2 2 + ( c k ( k ! ) 2 τ + c β k k b + 1 2 b k ) I 0 ( u 0 ) . (49)
and
∫ R ∣ x ∣ 2 a k ∣ u 0 ( x ) ∣ 2 d x ≤ 1 2 ∥ x k u 1 ∥ L 2 2 + ( c k ( k ! ) 2 τ + c α k k 2 a a + 1 k ) I 0 ( u 0 ) . (50) \int_{\mathbb{R}} |x|^{2ak}|u_0(x)|^2\,dx \le\frac{1}{2}\|x^k u_1\|_{L^2}^2+\left(c^k(k!)^{2\tau}+c_\alpha^k k^{\frac{2a}{a+1}k}\right)I_0(u_0). \tag*{(50)} ∫ R ∣ x ∣ 2 ak ∣ u 0 ( x ) ∣ 2 d x ≤ 2 1 ∥ x k u 1 ∥ L 2 2 + ( c k ( k ! ) 2 τ + c α k k a + 1 2 a k ) I 0 ( u 0 ) . (50)
Let 1 < p , q < ∞ 1<p,q<\infty 1 < p , q < ∞ with 1 p + 1 q = 1 \frac{1}{p}+\frac{1}{q}=1 p 1 + q 1 = 1 to be fixed later, and m ∈ N m\in\mathbb{N} m ∈ N . Applying (49) with k = ⌊ q m 2 b ⌋ k=\left\lfloor\frac{qm}{2b}\right\rfloor k = ⌊ 2 b q m ⌋ and using Stirling’s formula imply
∫ R ∣ x ∣ q m ∣ u 1 ( x ) ∣ 2 d x ≤ 1 2 ∫ R ∣ x ∣ q m b ∣ u 0 ( x ) ∣ 2 d x + ( c m ( m ! ) q τ b + c β m m q b + 1 m ) I 0 ( u 0 ) . (51) \int_{\mathbb{R}} |x|^{qm}|u_1(x)|^2\,dx \le\frac{1}{2}\int_{\mathbb{R}} |x|^{\frac{qm}{b}}|u_0(x)|^2\,dx+\left(c^m(m!)^{\frac{q\tau}{b}}+c_\beta^m m^{\frac{q}{b+1}m}\right)I_0(u_0). \tag*{(51)} ∫ R ∣ x ∣ q m ∣ u 1 ( x ) ∣ 2 d x ≤ 2 1 ∫ R ∣ x ∣ b q m ∣ u 0 ( x ) ∣ 2 d x + ( c m ( m ! ) b q τ + c β m m b + 1 q m ) I 0 ( u 0 ) . (51)
Applying (50) with k = ⌊ q m 2 a ⌋ k=\left\lfloor\frac{qm}{2a}\right\rfloor k = ⌊ 2 a q m ⌋ and using Stirling’s formula imply
∫ R ∣ x ∣ q m b ∣ u 0 ( x ) ∣ 2 d x ≤ 1 2 ∫ R ∣ x ∣ q m a b ∣ u 1 ( x ) ∣ 2 d x + ( c m ( m ! ) q τ a b + c α m m q ( a + 1 ) b m ) I 0 ( u 0 ) . (52) \int_{\mathbb{R}} |x|^{\frac{qm}{b}}|u_0(x)|^2\,dx \le\frac{1}{2}\int_{\mathbb{R}} |x|^{\frac{qm}{ab}}|u_1(x)|^2\,dx+\left(c^m(m!)^{\frac{q\tau}{ab}}+c_\alpha^m m^{\frac{q}{(a+1)b}m}\right)I_0(u_0). \tag*{(52)} ∫ R ∣ x ∣ b q m ∣ u 0 ( x ) ∣ 2 d x ≤ 2 1 ∫ R ∣ x ∣ ab q m ∣ u 1 ( x ) ∣ 2 d x + ( c m ( m ! ) ab q τ + c α m m ( a + 1 ) b q m ) I 0 ( u 0 ) . (52)
Inserting (52) into (51) yields
∫ R ∣ x ∣ q m ∣ u 1 ( x ) ∣ 2 d x ≤ 1 4 ∫ R ∣ x ∣ q m a b ∣ u 1 ( x ) ∣ 2 d x + c m ( ( m ! ) q τ a b + m q ( a + 1 ) b m + ( m ! ) q τ b + m q b + 1 m ) . (53) \int_{\mathbb{R}} |x|^{qm}|u_1(x)|^2\,dx \le\frac{1}{4}\int_{\mathbb{R}} |x|^{\frac{qm}{ab}}|u_1(x)|^2\,dx+c^m\left((m!)^{\frac{q\tau}{ab}}+m^{\frac{q}{(a+1)b}m}+(m!)^{\frac{q\tau}{b}}+m^{\frac{q}{b+1}m}\right). \tag*{(53)} ∫ R ∣ x ∣ q m ∣ u 1 ( x ) ∣ 2 d x ≤ 4 1 ∫ R ∣ x ∣ ab q m ∣ u 1 ( x ) ∣ 2 d x + c m ( ( m ! ) ab q τ + m ( a + 1 ) b q m + ( m ! ) b q τ + m b + 1 q m ) . (53)
Recall that condition α + β < 1 \alpha+\beta<1 α + β < 1 is equivalent to
( 1 − α α ) ( 1 − β β ) > 1. \left(\frac{1-\alpha}{\alpha}\right)\left(\frac{1-\beta}{\beta}\right)>1. ( α 1 − α ) ( β 1 − β ) > 1.
Therefore, it is possible to choose a < 1 − α α a<\frac{1-\alpha}{\alpha} a < α 1 − α and b < 1 − β β b<\frac{1-\beta}{\beta} b < β 1 − β such that a b > 1 ab>1 ab > 1 . Then
1 4 ∫ R ∣ x ∣ q a b ∣ u 1 ( x ) ∣ 2 d x ≤ 1 4 ∫ R ∣ x ∣ q ∣ u 1 ( x ) ∣ 2 d x + I 0 ( u 0 ) . \frac{1}{4}\int_{\mathbb{R}} |x|^{\frac{q}{ab}}|u_1(x)|^2\,dx\le\frac{1}{4}\int_{\mathbb{R}} |x|^{q}|u_1(x)|^2\,dx+I_0(u_0). 4 1 ∫ R ∣ x ∣ ab q ∣ u 1 ( x ) ∣ 2 d x ≤ 4 1 ∫ R ∣ x ∣ q ∣ u 1 ( x ) ∣ 2 d x + I 0 ( u 0 ) .
It follows from Stirling’s formula that, for k ≥ k 0 k\ge k_0 k ≥ k 0 ,
∫ R ∣ x ∣ q k ∣ u 1 ( x ) ∣ 2 d x ≤ e q δ 1 k log k + O ( k ) , (54) \int_{\mathbb{R}} |x|^{qk}|u_1(x)|^2\,dx\le e^{q\delta_1 k\log k+O(k)}, \tag*{(54)} ∫ R ∣ x ∣ q k ∣ u 1 ( x ) ∣ 2 d x ≤ e q δ 1 k l o g k + O ( k ) , (54)
where δ 1 = max { τ a b , 1 ( a + 1 ) b , τ b , 1 b + 1 } = max { τ a b , τ b } \delta_1=\max\left\{\frac{\tau}{ab},\frac{1}{(a+1)b},\frac{\tau}{b},\frac{1}{b+1}\right\}=\max\left\{\frac{\tau}{ab},\frac{\tau}{b}\right\} δ 1 = max { ab τ , ( a + 1 ) b 1 , b τ , b + 1 1 } = max { ab τ , b τ } . Here, we used that τ ≥ 1 \tau\ge1 τ ≥ 1 and that a , b > 0 a,b>0 a , b > 0 .
Applying (50) with k = ⌊ p m 2 a ⌋ k=\left\lfloor\frac{pm}{2a}\right\rfloor k = ⌊ 2 a p m ⌋ and (49) with k = ⌊ p m 2 a b ⌋ k=\left\lfloor\frac{pm}{2ab}\right\rfloor k = ⌊ 2 ab p m ⌋ and arguing similarly, for all k ≥ k 0 k\ge k_0 k ≥ k 0 ,
∫ R ∣ x ∣ p k ∣ u 0 ( x ) ∣ 2 d x ≤ e p δ 0 k log k + O ( k ) , (55) \int_{\mathbb{R}} |x|^{pk}|u_0(x)|^2\,dx\le e^{p\delta_0 k\log k+O(k)}, \tag*{(55)} ∫ R ∣ x ∣ p k ∣ u 0 ( x ) ∣ 2 d x ≤ e p δ 0 k l o g k + O ( k ) , (55)
where δ 0 = max { τ a b , 1 ( b + 1 ) a , τ a , 1 a + 1 } = max { τ a b , τ a } \delta_0=\max\left\{\frac{\tau}{ab},\frac{1}{(b+1)a},\frac{\tau}{a},\frac{1}{a+1}\right\}=\max\left\{\frac{\tau}{ab},\frac{\tau}{a}\right\} δ 0 = max { ab τ , ( b + 1 ) a 1 , a τ , a + 1 1 } = max { ab τ , a τ } .
Note that these estimates ((55) and (54) ) actually hold for all k ∈ N k\in\mathbb{N} k ∈ N . Indeed, in the case where k ≤ k 0 k\le k_0 k ≤ k 0 , we use the fact that u ∈ C ( [ 0 , 1 ] : S ( R ) ) u\in C([0,1]:S(\mathbb{R})) u ∈ C ([ 0 , 1 ] : S ( R )) (see Corollary 3.3), so that I k ( u 0 ) + I k ( u 1 ) ≤ c ( k 0 , ∥ u 0 ∥ H 0 ) I_k(u_0)+I_k(u_1)\le c(k_0,\|u_0\|_{\mathcal{H}^0}) I k ( u 0 ) + I k ( u 1 ) ≤ c ( k 0 , ∥ u 0 ∥ H 0 ) .
Then, we conclude that
∫ R e c ∣ x ∣ q ∣ u 1 ( x ) ∣ 2 d x = ∑ k = 0 ∞ c k k ! ∫ R ∣ x ∣ q k ∣ u 1 ( x ) ∣ 2 d x ≤ ∑ k = 0 ∞ e ( q δ 1 − 1 ) k log k + O ( k ) < + ∞ (56) \int_{\mathbb{R}} e^{c|x|^q}|u_1(x)|^2\,dx=\sum_{k=0}^{\infty}\frac{c^k}{k!}\int_{\mathbb{R}} |x|^{qk}|u_1(x)|^2\,dx\le\sum_{k=0}^{\infty}e^{(q\delta_1-1)k\log k+O(k)}<+\infty \tag*{(56)} ∫ R e c ∣ x ∣ q ∣ u 1 ( x ) ∣ 2 d x = k = 0 ∑ ∞ k ! c k ∫ R ∣ x ∣ q k ∣ u 1 ( x ) ∣ 2 d x ≤ k = 0 ∑ ∞ e ( q δ 1 − 1 ) k l o g k + O ( k ) < + ∞ (56)
and
∫ R e c ∣ x ∣ p ∣ u 0 ( x ) ∣ 2 d x = ∑ k = 0 ∞ c k k ! ∫ R ∣ x ∣ p k ∣ u 0 ( x ) ∣ 2 d x ≤ ∑ k = 0 ∞ e ( p δ 0 − 1 ) k log k + O ( k ) < + ∞ (57) \int_{\mathbb{R}} e^{c|x|^p}|u_0(x)|^2\,dx=\sum_{k=0}^{\infty}\frac{c^k}{k!}\int_{\mathbb{R}} |x|^{pk}|u_0(x)|^2\,dx\le\sum_{k=0}^{\infty}e^{(p\delta_0-1)k\log k+O(k)}<+\infty \tag*{(57)} ∫ R e c ∣ x ∣ p ∣ u 0 ( x ) ∣ 2 d x = k = 0 ∑ ∞ k ! c k ∫ R ∣ x ∣ p k ∣ u 0 ( x ) ∣ 2 d x ≤ k = 0 ∑ ∞ e ( p δ 0 − 1 ) k l o g k + O ( k ) < + ∞ (57)
if we require that q δ 1 < 1 q\delta_1<1 q δ 1 < 1 and p δ 0 < 1 p\delta_0<1 p δ 0 < 1 .
Now, we optimize the choice of p p p and q q q to get the best conditions for α , β \alpha,\beta α , β . We choose p p p and q q q such that
p δ 0 = q δ 1 ⟺ q q − 1 δ 0 = q δ 1 ⟺ q = δ 0 + δ 1 δ 1 . p\delta_0=q\delta_1 \Longleftrightarrow\frac{q}{q-1}\delta_0=q\delta_1 \Longleftrightarrow q=\frac{\delta_0+\delta_1}{\delta_1}. p δ 0 = q δ 1 ⟺ q − 1 q δ 0 = q δ 1 ⟺ q = δ 1 δ 0 + δ 1 .
Then, it follows that
q δ 1 < 1 and p δ 0 < 1 ⟺ δ 0 + δ 1 < 1. (58) q\delta_{1}<1\text{ and }p\delta_{0}<1\iff\delta_{0}+\delta_{1}<1. \tag*{(58)} q δ 1 < 1 and p δ 0 < 1 ⟺ δ 0 + δ 1 < 1. (58)
Below, we look at the meaning of condition (58) for α \alpha α and β \beta β .
If δ 0 = τ a b \delta_{0}=\frac{\tau}{ab} δ 0 = ab τ , then we must have
τ a b ≥ τ a ⟺ b ≤ 1. \frac{\tau}{ab}\geq\frac{\tau}{a}\iff b\leq1. ab τ ≥ a τ ⟺ b ≤ 1.
But b ≤ 1 b\leq1 b ≤ 1 implies δ 1 ≥ τ b ≥ 1 \delta_{1}\geq\frac{\tau}{b}\geq1 δ 1 ≥ b τ ≥ 1 , so that δ 1 + δ 0 ≥ 1 \delta_{1}+\delta_{0}\geq1 δ 1 + δ 0 ≥ 1 , which contradicts (58) .
Similarly, the case δ 1 = τ a b ⟺ a ≤ 1 \delta_{1}=\frac{\tau}{ab}\iff a\leq1 δ 1 = ab τ ⟺ a ≤ 1 is impossible.
Thus, in order for (58) to be satisfied, we must have δ 1 = τ b \delta_{1}=\frac{\tau}{b} δ 1 = b τ and δ 0 = τ a \delta_{0}=\frac{\tau}{a} δ 0 = a τ , with a > 1 a>1 a > 1 and b > 1 b>1 b > 1 . In this case,
δ 0 + δ 1 < 1 ⟺ τ ( 1 a + 1 b ) < 1. (59) \delta_{0}+\delta_{1}<1\iff\tau\left(\frac{1}{a}+\frac{1}{b}\right)<1. \tag*{(59)} δ 0 + δ 1 < 1 ⟺ τ ( a 1 + b 1 ) < 1. (59)
First, observe that
α 1 − α < 1 a < 1 and β 1 − β < 1 b < 1 , \frac{\alpha}{1-\alpha}<\frac{1}{a}<1\quad\text{and}\quad\frac{\beta}{1-\beta}<\frac{1}{b}<1, 1 − α α < a 1 < 1 and 1 − β β < b 1 < 1 ,
so we must have α , β ∈ ( 0 , 1 2 ) \alpha,\beta\in\left(0,\frac{1}{2}\right) α , β ∈ ( 0 , 2 1 ) . Now, assume that
τ ( α 1 − α + β 1 − β ) < 1. (60) \tau\left(\frac{\alpha}{1-\alpha}+\frac{\beta}{1-\beta}\right)<1. \tag*{(60)} τ ( 1 − α α + 1 − β β ) < 1. (60)
Since α < 1 2 \alpha<\frac{1}{2} α < 2 1 and β < 1 2 \beta<\frac{1}{2} β < 2 1 , then both 1 − α α > 1 \frac{1-\alpha}{\alpha}>1 α 1 − α > 1 and 1 − β β > 1 \frac{1-\beta}{\beta}>1 β 1 − β > 1 . Hence, we may choose 1 < a < 1 − α α 1<a<\frac{1-\alpha}{\alpha} 1 < a < α 1 − α and 1 < b < 1 − β β 1<b<\frac{1-\beta}{\beta} 1 < b < β 1 − β such that (59) is satisfied, which in turn implies (58) .
Finally, we conclude by using (56) -(57) and applying the Morgan principle proved by Escauriaza, Kening, Ponce and Vega (see Theorem 1.5) that u ≡ 0 u\equiv0 u ≡ 0 . This finishes the proof of Theorem 1.1.
Proof of the unique continuation result in the nonlinear case
We dedicate this section to the proof of Theorem 1.2. Without loss of generality, we assume that λ = 1 \lambda=1 λ = 1 in (2) . Below, we give a brief overview of the proof.
From Theorem 1.4 in Section 3, we know that solutions of (2) with u 0 ∈ H 1 u_{0}\in H^{1} u 0 ∈ H 1 that satisfies (3) for 0 < α + β < 1 0<\alpha+\beta<1 0 < α + β < 1 are in Schwartz for all time t ∈ [ 0 , 1 ] t\in[0,1] t ∈ [ 0 , 1 ] . The proof follows the same overall strategy as in the preceding section. However, instead of controlling the quantities I k ( u 0 ) I_k(u_0) I k ( u 0 ) and I k ( u 1 ) I_k(u_1) I k ( u 1 ) , we control the quantities
E k ( f ) : = I k ( f ) + I k ( ∂ x f ) = ∫ ∣ x ∣ 2 k ( ∣ f ( x ) ∣ 2 + ∣ ∂ x f ( x ) ∣ 2 ) d x . (61) E_k(f):=I_k(f)+I_k(\partial_xf)=\int|x|^{2k}\left(|f(x)|^2+|\partial_xf(x)|^2\right)\,dx. \tag*{(61)} E k ( f ) := I k ( f ) + I k ( ∂ x f ) = ∫ ∣ x ∣ 2 k ( ∣ f ( x ) ∣ 2 + ∣ ∂ x f ( x ) ∣ 2 ) d x . (61)
Note that, for all 0 ≤ k ≤ r 0\leq k\leq r 0 ≤ k ≤ r ,
E k ( f ) ≤ E r ( f ) + E 0 ( f ) . (62) E_k(f)\leq E_r(f)+E_0(f). \tag*{(62)} E k ( f ) ≤ E r ( f ) + E 0 ( f ) . (62)
It is worth noting that E 0 ( f ) = ∥ f ∥ H 1 2 ≤ ∥ f ∥ H 1 2 E_0(f)=\|f\|_{H^1}^{2}\leq\|f\|_{H^1}^{2} E 0 ( f ) = ∥ f ∥ H 1 2 ≤ ∥ f ∥ H 1 2 , and that
∥ x k f ∥ H 1 2 = ∥ x k f ∥ L 2 2 + ∥ ∂ x ( x k f ) ∥ L 2 2 ≤ 2 ( k 2 + 1 ) E k ( f ) + 2 k 2 ∥ f ∥ L 2 2 . (63) \|x^k f\|_{H^1}^{2}=\|x^k f\|_{L^2}^{2}+\|\partial_x(x^k f)\|_{L^2}^{2}\leq2(k^2+1)E_k(f)+2k^2\|f\|_{L^2}^{2}. \tag*{(63)} ∥ x k f ∥ H 1 2 = ∥ x k f ∥ L 2 2 + ∥ ∂ x ( x k f ) ∥ L 2 2 ≤ 2 ( k 2 + 1 ) E k ( f ) + 2 k 2 ∥ f ∥ L 2 2 . (63)
We will again use the functions u ~ 1 = e − i x 2 / 4 u 1 \tilde{u}_{1}=e^{-ix^2/4}u_1 u ~ 1 = e − i x 2 /4 u 1 and v ~ − 1 = e i x 2 / 4 u 0 \tilde{v}_{-1}=e^{ix^2/4}u_0 v ~ − 1 = e i x 2 /4 u 0 . Differentiating the exponential, we obtain that
E k ( u 1 ) ≤ c ( E k + 1 ( u ~ 1 ) + E 0 ( u ~ 1 ) ) ≤ c ( E k + 1 ( u ~ 1 ) + ∥ u 1 ∥ H 1 2 ) , (64) E_k(u_1)\leq c\left(E_{k+1}(\tilde{u}_1)+E_0(\tilde{u}_1)\right)\leq c\left(E_{k+1}(\tilde{u}_1)+\|u_1\|_{H^1}^{2}\right), \tag*{(64)} E k ( u 1 ) ≤ c ( E k + 1 ( u ~ 1 ) + E 0 ( u ~ 1 ) ) ≤ c ( E k + 1 ( u ~ 1 ) + ∥ u 1 ∥ H 1 2 ) , (64)
and
E k ( u 0 ) ≤ c ( E k + 1 ( v ~ − 1 ) + E 0 ( v ~ − 1 ) ) ≤ c ( E k + 1 ( v ~ − 1 ) + ∥ u 0 ∥ H 1 2 ) , (65) E_k(u_0)\leq c\left(E_{k+1}(\tilde{v}_{-1})+E_0(\tilde{v}_{-1})\right)\leq c\left(E_{k+1}(\tilde{v}_{-1})+\|u_0\|_{H^1}^{2}\right), \tag*{(65)} E k ( u 0 ) ≤ c ( E k + 1 ( v ~ − 1 ) + E 0 ( v ~ − 1 ) ) ≤ c ( E k + 1 ( v ~ − 1 ) + ∥ u 0 ∥ H 1 2 ) , (65)
Let α , β ∈ ( 0 , 1 2 ) \alpha,\beta\in\left(0,\frac{1}{2}\right) α , β ∈ ( 0 , 2 1 ) . We work with a , b a,b a , b satisfying
1 < a < 1 − α α , 1 < b < 1 − β β . 1<a<\frac{1-\alpha}{\alpha},\qquad1<b<\frac{1-\beta}{\beta}. 1 < a < α 1 − α , 1 < b < β 1 − β .
The main goal is to prove that
E p k 2 ( u 0 ) ≤ c k ( k ! ) p δ 0 , E q k 2 ( u 1 ) ≤ c k ( k ! ) q δ 1 , (66) E_{\frac{pk}{2}}(u_0) \le c^k(k!)^{p\delta_0}, \qquad E_{\frac{qk}{2}}(u_1) \le c^k(k!)^{q\delta_1}, \tag*{(66)} E 2 p k ( u 0 ) ≤ c k ( k ! ) p δ 0 , E 2 q k ( u 1 ) ≤ c k ( k ! ) q δ 1 , (66)
for 1 p + 1 q = 1 \frac{1}{p}+\frac{1}{q}=1 p 1 + q 1 = 1 , and δ 0 = 1 a \delta_0=\frac{1}{a} δ 0 = a 1 , δ 1 = 1 b \delta_1=\frac{1}{b} δ 1 = b 1 . These estimates provide Gaussian decay for u 0 u_0 u 0 and u 1 u_1 u 1 as soon as p δ 0 < 1 p\delta_0<1 p δ 0 < 1 and q δ 1 < 1 q\delta_1<1 q δ 1 < 1 . We will then optimize p p p and q q q as in the proof of Theorem 1.1 and conclude by using the Morgan uncertainty principle (Theorem 1.5).
Decay estimates from the zeros
Proposition 5.1. Let 0 < α , β < 1 0<\alpha,\beta<1 0 < α , β < 1 and 1 < b < 1 β − 1 1<b<\frac{1}{\beta}-1 1 < b < β 1 − 1 , 1 < a < 1 α − 1 1<a<\frac{1}{\alpha}-1 1 < a < α 1 − 1 . Then, there exist c ≥ 1 c\ge1 c ≥ 1 independent of u u u and k 0 ∈ N k_0\in\mathbb{N} k 0 ∈ N with k 0 ≥ 2 k_0\ge2 k 0 ≥ 2 , such that for k ≥ k 0 k\ge k_0 k ≥ k 0 ,
E b k ( u ~ 1 ) ≤ 1 2 ∥ u ~ 1 ∥ H k + 1 2 + c k k 2 b b + 1 k ∥ u 1 ∥ H 1 2 . (67) E_{bk}(\widetilde{u}_1) \le\frac{1}{2}\|\widetilde{u}_1\|_{H^{k+1}}^2+c^k k^{\frac{2b}{b+1}k}\|u_1\|_{\mathcal{H}^1}^2. \tag*{(67)} E bk ( u 1 ) ≤ 2 1 ∥ u 1 ∥ H k + 1 2 + c k k b + 1 2 b k ∥ u 1 ∥ H 1 2 . (67) E a k ( v ~ − 1 ) ≤ 1 2 ∥ v ~ − 1 ∥ H k + 1 2 + c k k 2 a a + 1 k ∥ u 0 ∥ H 1 2 . (68) E_{ak}(\widetilde{v}_{-1}) \le\frac{1}{2}\|\widetilde{v}_{-1}\|_{H^{k+1}}^2+c^k k^{\frac{2a}{a+1}k}\|u_0\|_{\mathcal{H}^1}^2. \tag*{(68)} E ak ( v − 1 ) ≤ 2 1 ∥ v − 1 ∥ H k + 1 2 + c k k a + 1 2 a k ∥ u 0 ∥ H 1 2 . (68)
Proof. We begin with the estimate from [30 (5.11)] , used earlier. For ∣ x ∣ ≥ X 0 |x|\ge X_0 ∣ x ∣ ≥ X 0 and x ∈ ( n β , ( n + 2 ) β ) x\in(n^\beta,(n+2)^\beta) x ∈ ( n β , ( n + 2 ) β ) , we have by the mean value theorem that
( n + 2 ) β − n β ≤ 2 β n β − 1 ≤ 2 β 3 1 − β ∣ x ∣ − 1 / β ≤ ( 2 λ 1 ∣ x ∣ b ) − 1 , (n+2)^\beta-n^\beta\le2\beta n^{\beta-1}\le2\beta3^{1-\beta}|x|^{-1/\beta}\le(2\lambda_1|x|^b)^{-1}, ( n + 2 ) β − n β ≤ 2 β n β − 1 ≤ 2 β 3 1 − β ∣ x ∣ − 1/ β ≤ ( 2 λ 1 ∣ x ∣ b ) − 1 ,
for λ 1 > 1 \lambda_1>1 λ 1 > 1 . Hence, the consecutive zeros of u ~ 1 \widetilde{u}_1 u 1 satisfy the gap hypothesis of [30 (5.11)] . In particular, there exists k 0 ∈ N k_0\in\mathbb{N} k 0 ∈ N with k 0 ≥ 2 k_0\ge2 k 0 ≥ 2 such that, for all k ≥ k 0 k\ge k_0 k ≥ k 0 ,
I b k ( u ~ 1 ) ≤ 1 2 ∫ ∣ ∂ x k u ~ 1 ( x ) ∣ 2 d x + c β k k 2 b b + 1 k ∥ u 1 ∥ L 2 2 . (69) I_{bk}(\widetilde{u}_1)\le\frac{1}{2}\int|\partial_x^k\widetilde{u}_1(x)|^2dx+c_\beta^k k^{\frac{2b}{b+1}k}\|u_1\|_{L^2}^2. \tag*{(69)} I bk ( u 1 ) ≤ 2 1 ∫ ∣ ∂ x k u 1 ( x ) ∣ 2 d x + c β k k b + 1 2 b k ∥ u 1 ∥ L 2 2 . (69)
We now apply the same estimate to the derivative ∂ x u ~ 1 \partial_x\widetilde{u}_1 ∂ x u 1 . As in [37 Section 2.1] , Rolle’s theorem gives, for each n ≥ 1 n\ge1 n ≥ 1 ,
{ x ∈ ( n β , ( n + 1 ) β ) : ∂ x Re u ~ 1 ( x ) = 0 } ≠ ∅ , \{x\in(n^\beta,(n+1)^\beta):\partial_x\operatorname{Re}\widetilde{u}_1(x)=0\}\ne\varnothing, { x ∈ ( n β , ( n + 1 ) β ) : ∂ x Re u 1 ( x ) = 0 } = ∅ ,
{ x ∈ ( n β , ( n + 1 ) β ) : ∂ x Im u ~ 1 ( x ) = 0 } ≠ ∅ . \{x\in(n^\beta,(n+1)^\beta):\partial_x\operatorname{Im}\widetilde{u}_1(x)=0\}\ne\varnothing. { x ∈ ( n β , ( n + 1 ) β ) : ∂ x Im u 1 ( x ) = 0 } = ∅ .
The same argument applies between consecutive negative zeros. Consecutive zeros chosen in these intervals lie within two consecutive gaps of { ± n β : n ≥ 1 } \{\pm n^\beta:n\ge1\} { ± n β : n ≥ 1 } , so their distance is at most ( n + 2 ) β − n β (n+2)^\beta-n^\beta ( n + 2 ) β − n β on the positive half-line, with the corresponding bound on the negative half-line. The previous gap estimate then yields
I b k ( ∂ x Re f ) ≤ 1 2 ∫ ∣ ∂ x k + 1 Re u ~ 1 ( x ) ∣ 2 d x + c k k 2 b b + 1 k ∥ ∂ x Re u ~ 1 ∥ L 2 2 , (70) I_{bk}(\partial_x\operatorname{Re}f)\le\frac{1}{2}\int|\partial_x^{k+1}\operatorname{Re}\widetilde{u}_1(x)|^2dx+c^k k^{\frac{2b}{b+1}k}\|\partial_x\operatorname{Re}\widetilde{u}_1\|_{L^2}^2, \tag*{(70)} I bk ( ∂ x Re f ) ≤ 2 1 ∫ ∣ ∂ x k + 1 Re u 1 ( x ) ∣ 2 d x + c k k b + 1 2 b k ∥ ∂ x Re u 1 ∥ L 2 2 , (70)
I b k ( ∂ x Im u ~ 1 ) ≤ 1 2 ∫ ∣ ∂ x k + 1 Im u ~ 1 ( x ) ∣ 2 d x + c k k 2 b b + 1 k ∥ ∂ x Im f ∥ L 2 2 . (71) I_{bk}(\partial_x\operatorname{Im}\widetilde{u}_1)\le\frac{1}{2}\int|\partial_x^{k+1}\operatorname{Im}\widetilde{u}_1(x)|^2dx+c^k k^{\frac{2b}{b+1}k}\|\partial_x\operatorname{Im}f\|_{L^2}^2. \tag*{(71)} I bk ( ∂ x Im u 1 ) ≤ 2 1 ∫ ∣ ∂ x k + 1 Im u 1 ( x ) ∣ 2 d x + c k k b + 1 2 b k ∥ ∂ x Im f ∥ L 2 2 . (71)
Combining the estimates (69) , (70) and (71) , we deduce that
E b k ( u ~ 1 ) ≤ 1 2 ( ∥ ∂ x k u ~ 1 ∥ L 2 2 + ∥ ∂ x k + 1 u ~ 1 ∥ L 2 2 ) + c k k 2 b b + 1 k ∥ u 1 ∥ H 1 2 ≤ 1 2 ∥ u ~ 1 ∥ H k + 1 2 + c k k 2 b b + 1 k ∥ u 1 ∥ H 1 2 , \begin{aligned}
E_{bk}(\widetilde{u}_1)\le\frac{1}{2}\left(\|\partial_x^k\widetilde{u}_1\|_{L^2}^2+\|\partial_x^{k+1}\widetilde{u}_1\|_{L^2}^2\right)+c^k k^{\frac{2b}{b+1}k}\|u_1\|_{\mathcal{H}^1}^2 \\
&\le\frac{1}{2}\|\widetilde{u}_1\|_{H^{k+1}}^2+c^k k^{\frac{2b}{b+1}k}\|u_1\|_{\mathcal{H}^1}^2,
\end{aligned} E bk ( u 1 ) ≤ 2 1 ( ∥ ∂ x k u 1 ∥ L 2 2 + ∥ ∂ x k + 1 u 1 ∥ L 2 2 ) + c k k b + 1 2 b k ∥ u 1 ∥ H 1 2 ≤ 2 1 ∥ u 1 ∥ H k + 1 2 + c k k b + 1 2 b k ∥ u 1 ∥ H 1 2 ,
which concludes the proof of (67) . The proof for (68) follows by the same argument for u 0 u_0 u 0 instead of u 1 u_1 u 1 .
Control of the evolution of Γ t k \Gamma_t^k Γ t k in L 2 L^2 L 2
In order to control the quantities ∥ u ~ 1 ∥ H k + 1 \|\widetilde{u}_1\|_{H^{k+1}} ∥ u 1 ∥ H k + 1 and ∥ v ~ − 1 ∥ H k + 1 \|\widetilde{v}_{-1}\|_{H^{k+1}} ∥ v − 1 ∥ H k + 1 in (67) –(68) , we estimate the L 2 L^2 L 2 -norm of the vector field operators Γ t k \Gamma_t^k Γ t k in L 2 L^2 L 2 along the flow. Recall that Γ t \Gamma_t Γ t is defined in (2.4); in particular, Γ 0 = x \Gamma_0=x Γ 0 = x .
Proposition 5.2. There exists K = K ( ∥ u 0 ∥ H 2 ) ≥ 1 K=K(\|u_0\|_{\mathcal{H}^2})\ge1 K = K ( ∥ u 0 ∥ H 2 ) ≥ 1 such that the following is true.
(i) Assume that there exist B ≥ 1 B\ge1 B ≥ 1 and J ∈ N J\in\mathbb{N} J ∈ N with J ≥ 1 J\ge1 J ≥ 1 such that
E j ( u 0 ) ≤ ( B j j ! ) 2 , 1 ≤ j ≤ J . (72) E_j(u_0)\le(B^j j!)^2,\qquad1\le j\le J. \tag*{(72)} E j ( u 0 ) ≤ ( B j j ! ) 2 , 1 ≤ j ≤ J . (72)
Then,
sup 0 ≤ t ≤ 1 ∥ Γ t j u ( t ) ∥ H 1 ≤ ( K B ) j j ! , 1 ≤ j ≤ J . (73) \sup_{0\leq t\leq1}\left\|\Gamma_t^j u(t)\right\|_{H^1} \leq(KB)^j j!, \qquad1\leq j\leq J. \tag*{(73)} 0 ≤ t ≤ 1 sup Γ t j u ( t ) H 1 ≤ ( K B ) j j ! , 1 ≤ j ≤ J . (73)
(ii) Assume that there exist B ≥ 1 B\geq1 B ≥ 1 and J ∈ N J\in\mathbb{N} J ∈ N with J ≥ 1 J\geq1 J ≥ 1 such that
E j ( u 1 ) ≤ ( B j j ! ) 2 , 1 ≤ j ≤ J . (74) E_j(u_1)\leq(B^j j!)^2,\qquad1\leq j\leq J. \tag*{(74)} E j ( u 1 ) ≤ ( B j j ! ) 2 , 1 ≤ j ≤ J . (74)
Then,
sup 0 ≤ t ≤ 1 ∥ Γ t j u ( 1 − t ) ∥ H 1 ≤ ( K B ) j j ! , 1 ≤ j ≤ J . (75) \sup_{0\leq t\leq1}\left\|\Gamma_t^j u(1-t)\right\|_{H^1}\leq(KB)^j j!,\qquad1\leq j\leq J. \tag*{(75)} 0 ≤ t ≤ 1 sup Γ t j u ( 1 − t ) H 1 ≤ ( K B ) j j ! , 1 ≤ j ≤ J . (75)
In order to prove Proposition 5.2, we need the following technical lemma.
Lemma 5.1. There exists a universal constant c > 0 c>0 c > 0 such that for any j ∈ N j\in\mathbb{N} j ∈ N , j ≥ 2 j\geq2 j ≥ 2 , we have
∑ ℓ = 1 j − 1 1 ( ℓ + 1 ) 2 ( j − ℓ + 1 ) 2 ≤ c ( j + 1 ) 2 . (76) \sum_{\ell=1}^{j-1}\frac{1}{(\ell+1)^2(j-\ell+1)^2}\leq\frac{c}{(j+1)^2}. \tag*{(76)} ℓ = 1 ∑ j − 1 ( ℓ + 1 ) 2 ( j − ℓ + 1 ) 2 1 ≤ ( j + 1 ) 2 c . (76) ∑ j 1 + j 2 + j 3 = j j 1 , j 2 , j 3 ≥ 1 1 ( j 1 + 1 ) 2 ( j 2 + 1 ) 2 ( j 3 + 1 ) 2 ≤ c ( j + 1 ) 2 . (77) \sum_{\substack{j_1+j_2+j_3=j\\ j_1,j_2,j_3\geq1}}\frac{1}{(j_1+1)^2(j_2+1)^2(j_3+1)^2}\leq\frac{c}{(j+1)^2}. \tag*{(77)} j 1 + j 2 + j 3 = j j 1 , j 2 , j 3 ≥ 1 ∑ ( j 1 + 1 ) 2 ( j 2 + 1 ) 2 ( j 3 + 1 ) 2 1 ≤ ( j + 1 ) 2 c . (77)
Proof. The proofs of (5.16) and (5.17) are classical, but we redo them for the sake of completeness. To prove (5.16), consider first the case 1 ≤ ℓ ≤ j / 2 1\leq\ell\leq j/2 1 ≤ ℓ ≤ j /2 . Then j − ℓ + 1 ≥ ( j + 1 ) / 2 j-\ell+1\geq(j+1)/2 j − ℓ + 1 ≥ ( j + 1 ) /2 , so the left-hand side of (5.16) is bounded in this case by 4 ( j + 1 ) − 2 ( ℓ + 1 ) − 2 4(j+1)^{-2}(\ell+1)^{-2} 4 ( j + 1 ) − 2 ( ℓ + 1 ) − 2 . The case where j / 2 ≤ ℓ ≤ j − 1 j/2\leq\ell\leq j-1 j /2 ≤ ℓ ≤ j − 1 is handled similarly by using symmetry and the change of variable ℓ ~ = j − ℓ \tilde{\ell}=j-\ell ℓ ~ = j − ℓ . Hence
∑ ℓ = 1 j − 1 1 ( ℓ + 1 ) 2 ( j − ℓ + 1 ) 2 ≤ 8 ( j + 1 ) 2 ∑ ℓ = 1 ∞ 1 ( ℓ + 1 ) 2 , \sum_{\ell=1}^{j-1}\frac{1}{(\ell+1)^2(j-\ell+1)^2}\leq\frac{8}{(j+1)^2}\sum_{\ell=1}^{\infty}\frac{1}{(\ell+1)^2}, ℓ = 1 ∑ j − 1 ( ℓ + 1 ) 2 ( j − ℓ + 1 ) 2 1 ≤ ( j + 1 ) 2 8 ℓ = 1 ∑ ∞ ( ℓ + 1 ) 2 1 ,
and the series on the right converges, which concludes the proof. To prove (5.17), we first sum over j 1 j_1 j 1 and then over j 2 j_2 j 2 and use twice (5.16) to find
∑ j 1 + j 2 + j 3 = j j 1 , j 2 , j 3 ≥ 1 1 ( j 1 + 1 ) 2 ( j 2 + 1 ) 2 ( j 3 + 1 ) 2 = ∑ j 2 = 1 j ( ∑ j 1 = 1 j − j 2 − 1 1 ( j 1 + 1 ) 2 ( j − j 1 − j 2 + 1 ) 2 ) 1 ( j 2 + 1 ) 2 ≤ c ∑ j 2 = 1 j 1 ( j − j 2 + 1 ) 2 ( j 2 + 1 ) 2 ≤ c 1 ( j + 1 ) 2 . \begin{aligned}
\sum_{\substack{j_1+j_2+j_3=j\\ j_1,j_2,j_3\geq1}}\frac{1}{(j_1+1)^2(j_2+1)^2(j_3+1)^2}
=\sum_{j_2=1}^{j}\left(\sum_{j_1=1}^{j-j_2-1}\frac{1}{(j_1+1)^2(j-j_1-j_2+1)^2}\right)\frac{1}{(j_2+1)^2} \\
&\leq c\sum_{j_2=1}^{j}\frac{1}{(j-j_2+1)^2(j_2+1)^2} \\
&\leq c\frac{1}{(j+1)^2}.
\end{aligned} j 1 + j 2 + j 3 = j j 1 , j 2 , j 3 ≥ 1 ∑ ( j 1 + 1 ) 2 ( j 2 + 1 ) 2 ( j 3 + 1 ) 2 1 = j 2 = 1 ∑ j ( j 1 = 1 ∑ j − j 2 − 1 ( j 1 + 1 ) 2 ( j − j 1 − j 2 + 1 ) 2 1 ) ( j 2 + 1 ) 2 1 ≤ c j 2 = 1 ∑ j ( j − j 2 + 1 ) 2 ( j 2 + 1 ) 2 1 ≤ c ( j + 1 ) 2 1 .
:::
Proof of Proposition 5.2. To prove (5.13), we prove a slightly stronger property. There exists K ≥ 1 K\geq1 K ≥ 1 , 0 < δ ≤ 1 0<\delta\leq1 0 < δ ≤ 1 , such that under the hypothesis (5.12), for all 1 ≤ ℓ ≤ J 1\leq\ell\leq J 1 ≤ ℓ ≤ J
sup 0 ≤ t ≤ 1 ∥ Γ t ℓ u ( t ) ∥ H 1 ≤ δ ( ℓ + 1 ) 2 ( K B ) ℓ ℓ ! . (78) \sup_{0\leq t\leq1}\left\|\Gamma_t^\ell u(t)\right\|_{H^1}\leq\frac{\delta}{(\ell+1)^2}(KB)^\ell\ell!. \tag*{(78)} 0 ≤ t ≤ 1 sup Γ t ℓ u ( t ) H 1 ≤ ( ℓ + 1 ) 2 δ ( K B ) ℓ ℓ ! . (78)
We will prove (5.18) by induction. First, we verify it for ℓ = 1 \ell=1 ℓ = 1 . From the definition of Γ \Gamma Γ in (2.4) and then (3.1), it follows that, for all 0 ≤ t ≤ 1 0\leq t\leq1 0 ≤ t ≤ 1 ,
∥ Γ t 1 u ( t ) ∥ H 1 ≤ c ( ∥ x u ( t ) ∥ H 1 + ∥ ∂ x u ( t ) ∥ H 1 ) ≤ c ∥ u ( t ) ∥ H 2 . \left\|\Gamma_t^1u(t)\right\|_{H^1}\leq c\left(\|xu(t)\|_{H^1}+\|\partial_xu(t)\|_{H^1}\right)\leq c\|u(t)\|_{\mathcal{H}^2}. Γ t 1 u ( t ) H 1 ≤ c ( ∥ xu ( t ) ∥ H 1 + ∥ ∂ x u ( t ) ∥ H 1 ) ≤ c ∥ u ( t ) ∥ H 2 .
From the well-posedness of (1.2) in H 2 \mathcal{H}^2 H 2 in Proposition 3.2, we conclude that
∥ Γ t 1 u ( t ) ∥ H 1 ≤ c ∥ u 0 ∥ H 2 ≤ δ K B , \left\|\Gamma_t^1u(t)\right\|_{H^1}\leq c\|u_0\|_{\mathcal{H}^2}\leq\delta KB, Γ t 1 u ( t ) H 1 ≤ c ∥ u 0 ∥ H 2 ≤ δ K B ,
under the condition
δ K ≥ c ∥ u 0 ∥ H 2 . (79) \delta K\geq c\|u_0\|_{\mathcal{H}^2}. \tag*{(79)} δ K ≥ c ∥ u 0 ∥ H 2 . (79)
and recalling B ≥ 1 B \ge1 B ≥ 1 . This verifies that (78) holds in the case l = 1 l=1 l = 1 .
Now, let 0 ≤ j ≤ J − 1 0 \le j \le J-1 0 ≤ j ≤ J − 1 be fixed. Assume that (78) holds for l = 1 , … , j − 1 l=1,\ldots,j-1 l = 1 , … , j − 1 . We want to prove that (78) holds for l = j l=j l = j .
Let 0 < t ≤ 1 0 < t \le1 0 < t ≤ 1 . By using the Duhamel principle associated to (16) with V = ∣ u ∣ 2 V=|u|^2 V = ∣ u ∣ 2 and (13) , we have that
Γ t j u ( t ) = e i t ∂ x 2 ( x k u 0 ) + i ∫ 0 t e i ( t − s ) ∂ x 2 Γ s j ( ∣ u ∣ 2 u ) ( s ) d s . (80) \Gamma_t^j u(t)=e^{it\partial_x^2}(x^k u_0)+i\int_0^t e^{i(t-s)\partial_x^2}\Gamma_s^j(|u|^2u)(s)\,ds. \tag*{(80)} Γ t j u ( t ) = e i t ∂ x 2 ( x k u 0 ) + i ∫ 0 t e i ( t − s ) ∂ x 2 Γ s j ( ∣ u ∣ 2 u ) ( s ) d s . (80)
Moreover, formula (12) and the Leibniz rule yield
Γ s j ( ∣ u ∣ 2 u ) = e i ∣ x ∣ 2 4 s ( 2 i s ) ∣ j ∣ ∂ x j ( ∣ u ~ ∣ 2 u ~ ) = e i ∣ x ∣ 2 4 s ( 2 i s ) ∣ j ∣ ∑ j 1 + j 2 + j 3 = j j ! ( j 1 ) ! ( j 2 ) ! ( j 3 ) ! ( − 1 ) j 3 ( ∂ x j 1 u ~ ) ( ∂ x j 2 u ~ ) ( ∂ x j 3 u ~ ) = ∑ j 1 + j 2 + j 3 = j j ! ( j 1 ) ! ( j 2 ) ! ( j 3 ) ! ( − 1 ) j 3 ( Γ s j 1 u ) ( Γ s j 2 u ) ( Γ s j 3 u ‾ ) , \begin{aligned}
\Gamma_s^j(|u|^2u)=e^{\frac{i|x|^2}{4s}}(2is)^{|j|}\partial_x^j(|\tilde u|^2\tilde u) \\
&=e^{\frac{i|x|^2}{4s}}(2is)^{|j|}\sum_{j_1+j_2+j_3=j}\frac{j!}{(j_1)!(j_2)!(j_3)!}(-1)^{j_3}(\partial_x^{j_1}\tilde u)(\partial_x^{j_2}\tilde u)(\partial_x^{j_3}\tilde u) \\
&=\sum_{j_1+j_2+j_3=j}\frac{j!}{(j_1)!(j_2)!(j_3)!}(-1)^{j_3}(\Gamma_s^{j_1}u)(\Gamma_s^{j_2}u)(\overline{\Gamma_s^{j_3}u}),
\end{aligned} Γ s j ( ∣ u ∣ 2 u ) = e 4 s i ∣ x ∣ 2 ( 2 i s ) ∣ j ∣ ∂ x j ( ∣ u ~ ∣ 2 u ~ ) = e 4 s i ∣ x ∣ 2 ( 2 i s ) ∣ j ∣ j 1 + j 2 + j 3 = j ∑ ( j 1 )! ( j 2 )! ( j 3 )! j ! ( − 1 ) j 3 ( ∂ x j 1 u ~ ) ( ∂ x j 2 u ~ ) ( ∂ x j 3 u ~ ) = j 1 + j 2 + j 3 = j ∑ ( j 1 )! ( j 2 )! ( j 3 )! j ! ( − 1 ) j 3 ( Γ s j 1 u ) ( Γ s j 2 u ) ( Γ s j 3 u ) ,
where we recall the notation u ~ ( x , s ) = e i ∣ x ∣ 2 4 s u ( x , s ) \tilde u(x,s)=e^{\frac{i|x|^2}{4s}}u(x,s) u ~ ( x , s ) = e 4 s i ∣ x ∣ 2 u ( x , s ) . Moreover, we separate the three terms in the above sum for which one of the index j i j_i j i is equal to j j j , so that
Γ s j ( ∣ u ∣ 2 u ) = 2 ∣ u ∣ 2 Γ s j u + ( − 1 ) j u 2 Γ s j u ‾ + ∑ j 1 + j 2 + j 3 = j 0 ≤ j 1 , j 2 , j 3 ≤ j − 1 j ! ( j 1 ) ! ( j 2 ) ! ( j 3 ) ! ( − 1 ) j 3 ( Γ s j 1 u ) ( Γ s j 2 u ) ( Γ s j 3 u ‾ ) . \Gamma_s^j(|u|^2u)=2|u|^2\Gamma_s^j u+(-1)^j u^2\overline{\Gamma_s^j u}+
\sum_{\substack{j_1+j_2+j_3=j\\0\le j_1,j_2,j_3\le j-1}}\frac{j!}{(j_1)!(j_2)!(j_3)!}(-1)^{j_3}(\Gamma_s^{j_1}u)(\Gamma_s^{j_2}u)(\overline{\Gamma_s^{j_3}u}). Γ s j ( ∣ u ∣ 2 u ) = 2∣ u ∣ 2 Γ s j u + ( − 1 ) j u 2 Γ s j u + j 1 + j 2 + j 3 = j 0 ≤ j 1 , j 2 , j 3 ≤ j − 1 ∑ ( j 1 )! ( j 2 )! ( j 3 )! j ! ( − 1 ) j 3 ( Γ s j 1 u ) ( Γ s j 2 u ) ( Γ s j 3 u ) .
Thus, we deduce inserting this identity in (80) , taking the H 1 H^1 H 1 -norm, and then using that H 1 H^1 H 1 is a Banach algebra, that
∥ Γ t j u ( t ) ∥ H 1 ≤ ∥ x j u 0 ∥ H 1 + 3 ∫ 0 t ∥ u ( s ) ∥ H 1 2 ∥ Γ s j u ( s ) ∥ H 1 d s + ∑ j 1 + j 2 + j 3 = j 0 ≤ j 1 , j 2 , j 3 ≤ j − 1 j ! ( j 1 ) ! ( j 2 ) ! ( j 3 ) ! ∫ 0 t ∥ Γ s j 1 u ( s ) ∥ H 1 ∥ Γ s j 2 u ( s ) ∥ H 1 ∥ Γ s j 3 u ( s ) ∥ H 1 d s = : I + I I + I I I (81) \begin{aligned}
\|\Gamma_t^j u(t)\|_{H^1}&\le\|x^j u_0\|_{H^1}+3\int_0^t\|u(s)\|_{H^1}^2\|\Gamma_s^j u(s)\|_{H^1}\,ds\\
&\quad+\sum_{\substack{j_1+j_2+j_3=j\\0\le j_1,j_2,j_3\le j-1}}\frac{j!}{(j_1)!(j_2)!(j_3)!}\int_0^t\|\Gamma_s^{j_1}u(s)\|_{H^1}\|\Gamma_s^{j_2}u(s)\|_{H^1}\|\Gamma_s^{j_3}u(s)\|_{H^1}\,ds\\
&=:I+II+III \tag*{(81)}
\end{aligned} ∥ Γ t j u ( t ) ∥ H 1 ≤ ∥ x j u 0 ∥ H 1 + 3 ∫ 0 t ∥ u ( s ) ∥ H 1 2 ∥ Γ s j u ( s ) ∥ H 1 d s + j 1 + j 2 + j 3 = j 0 ≤ j 1 , j 2 , j 3 ≤ j − 1 ∑ ( j 1 )! ( j 2 )! ( j 3 )! j ! ∫ 0 t ∥ Γ s j 1 u ( s ) ∥ H 1 ∥ Γ s j 2 u ( s ) ∥ H 1 ∥ Γ s j 3 u ( s ) ∥ H 1 d s =: I + I I + I I I (81)
To estimate the first term on the right-hand side of (81) , we use (63) and the hypothesis on E j E_j E j (72) ,
I ≤ ( 2 ( j 2 + 1 ) E j ( u 0 ) + 2 j 2 ∥ u 0 ∥ L 2 2 ) 1 2 ≤ c j 2 + 1 B j j ! + c j ∥ u 0 ∥ L 2 . I\le\left(2(j^2+1)E_j(u_0)+2j^2\|u_0\|_{L^2}^2\right)^{\frac12}\le c\sqrt{j^2+1}B^j j!+c_j\|u_0\|_{L^2}. I ≤ ( 2 ( j 2 + 1 ) E j ( u 0 ) + 2 j 2 ∥ u 0 ∥ L 2 2 ) 2 1 ≤ c j 2 + 1 B j j ! + c j ∥ u 0 ∥ L 2 .
From the well-posedness theory in H 1 H^1 H 1 , (see Proposition 3.2 ),
∥ u ( s ) ∥ H 1 ≤ c ∥ u ( s ) ∥ H 1 ≤ c ∥ u 0 ∥ H 1 , (82) \|u(s)\|_{H^1}\le c\|u(s)\|_{\mathcal H^1}\le c\|u_0\|_{\mathcal H^1}, \tag*{(82)} ∥ u ( s ) ∥ H 1 ≤ c ∥ u ( s ) ∥ H 1 ≤ c ∥ u 0 ∥ H 1 , (82)
so that we bound the second term on the right-hand side of (81) by
I I ≤ c ∥ u 0 ∥ H 1 2 ∫ 0 t ∥ Γ s j u ( s ) ∥ H 1 d s . II\le c\|u_0\|_{\mathcal H^1}^2\int_0^t\|\Gamma_s^j u(s)\|_{H^1}\,ds. I I ≤ c ∥ u 0 ∥ H 1 2 ∫ 0 t ∥ Γ s j u ( s ) ∥ H 1 d s .
To deal with the last term on the right-hand side of (81) , observe that at least two indexes of j 1 , j 2 , j 3 j_1,j_2,j_3 j 1 , j 2 , j 3 are larger or equal to 1. We split the sum depending if the remaining of these indexes is 0 or larger than 1. Hence,
I I I = 3 ∑ l = 1 j − 1 j ! l ! ( j − l ) ! ∫ 0 t ∥ u ( s ) ∥ H 1 ∥ Γ s l u ( s ) ∥ H 1 ∥ Γ s j − l u ( s ) ∥ H 1 d s + ∑ j 1 + j 2 + j 3 = j 1 ≤ j 1 , j 2 , j 3 ≤ j − 1 j ! ( j 1 ) ! ( j 2 ) ! ( j 3 ) ! ∫ 0 t ∥ Γ s j 1 u ( s ) ∥ H 1 ∥ Γ s j 2 u ( s ) ∥ H 1 ∥ Γ s j 3 u ( s ) ∥ H 1 d s . \begin{aligned}
III={}&3\sum_{l=1}^{j-1}\frac{j!}{l!(j-l)!}\int_0^t\|u(s)\|_{H^1}\|\Gamma_s^l u(s)\|_{H^1}\|\Gamma_s^{j-l}u(s)\|_{H^1}\,ds\\
&+\sum_{\substack{j_1+j_2+j_3=j\\1\le j_1,j_2,j_3\le j-1}}\frac{j!}{(j_1)!(j_2)!(j_3)!}\int_0^t\|\Gamma_s^{j_1}u(s)\|_{H^1}\|\Gamma_s^{j_2}u(s)\|_{H^1}\|\Gamma_s^{j_3}u(s)\|_{H^1}\,ds.
\end{aligned} I I I = 3 l = 1 ∑ j − 1 l ! ( j − l )! j ! ∫ 0 t ∥ u ( s ) ∥ H 1 ∥ Γ s l u ( s ) ∥ H 1 ∥ Γ s j − l u ( s ) ∥ H 1 d s + j 1 + j 2 + j 3 = j 1 ≤ j 1 , j 2 , j 3 ≤ j − 1 ∑ ( j 1 )! ( j 2 )! ( j 3 )! j ! ∫ 0 t ∥ Γ s j 1 u ( s ) ∥ H 1 ∥ Γ s j 2 u ( s ) ∥ H 1 ∥ Γ s j 3 u ( s ) ∥ H 1 d s .
Thus, we deduce by using (5.22), the induction hypothesis (5.18) for all 1 ≤ l ≤ j − 1 1 \le l \le j-1 1 ≤ l ≤ j − 1 , and then Lemma 5.1 that
I I I ≤ c ∥ u 0 ∥ H 1 j ! ( ∑ l = 1 j − 1 1 ( l + 1 ) 2 ( j − l + 1 ) 2 ) δ 2 ( K B ) j + c j ! ( ∑ j 1 + j 2 + j 3 = j 1 ≤ j 1 , j 2 , j 3 ≤ j − 1 1 ( j 1 + 1 ) 2 ( j 2 + 1 ) 2 ( j 3 + 1 ) 2 ) δ 3 ( K B ) j ≤ c ( ∥ u 0 ∥ H 1 δ 2 + δ 3 ( j + 1 ) 2 ) ( K B ) j j ! . III \le c\lVert u_0\rVert_{\mathcal{H}_1}j!\left(\sum_{l=1}^{j-1}\frac{1}{(l+1)^2(j-l+1)^2}\right)\delta^2(KB)^j
+ cj!\left(\sum_{\substack{j_1+j_2+j_3=j\\1\le j_1,j_2,j_3\le j-1}}\frac{1}{(j_1+1)^2(j_2+1)^2(j_3+1)^2}\right)\delta^3(KB)^j
\le c\left(\frac{\lVert u_0\rVert_{\mathcal{H}_1}\delta^2+\delta^3}{(j+1)^2}\right)(KB)^j j!. I I I ≤ c ∥ u 0 ∥ H 1 j ! ( l = 1 ∑ j − 1 ( l + 1 ) 2 ( j − l + 1 ) 2 1 ) δ 2 ( K B ) j + c j ! j 1 + j 2 + j 3 = j 1 ≤ j 1 , j 2 , j 3 ≤ j − 1 ∑ ( j 1 + 1 ) 2 ( j 2 + 1 ) 2 ( j 3 + 1 ) 2 1 δ 3 ( K B ) j ≤ c ( ( j + 1 ) 2 ∥ u 0 ∥ H 1 δ 2 + δ 3 ) ( K B ) j j ! .
By inserting these estimates into (5.21), we find that
∥ Γ t j u ( t ) ∥ H 1 ≤ 1 + j 2 B j j ! + c j ∥ u 0 ∥ L 2 + c ( ∥ u 0 ∥ H 1 δ 2 + δ 3 ) ( j + 1 ) 2 ( K B ) j j ! + c ∥ u 0 ∥ H 1 2 ∫ 0 t ∥ Γ s j u ( s ) ∥ H 1 d s , \lVert\Gamma_t^j u(t)\rVert_{\mathcal{H}_1}\le\sqrt{1+j^2B^j}j!+cj\lVert u_0\rVert_{L^2}+c\frac{\left(\lVert u_0\rVert_{\mathcal{H}_1}\delta^2+\delta^3\right)}{(j+1)^2}(KB)^j j!+c\lVert u_0\rVert_{\mathcal{H}_1^2}\int_0^t\lVert\Gamma_s^j u(s)\rVert_{\mathcal{H}_1}\,ds, ∥ Γ t j u ( t ) ∥ H 1 ≤ 1 + j 2 B j j ! + c j ∥ u 0 ∥ L 2 + c ( j + 1 ) 2 ( ∥ u 0 ∥ H 1 δ 2 + δ 3 ) ( K B ) j j ! + c ∥ u 0 ∥ H 1 2 ∫ 0 t ∥ Γ s j u ( s ) ∥ H 1 d s ,
which implies by using Gronwall’s inequality that for all 0 ≤ t ≤ 1 0 \le t \le1 0 ≤ t ≤ 1 ,
∥ Γ t j u ( t ) ∥ H 1 ≤ e c ∥ u 0 ∥ H 1 2 ( 1 + j 2 B j j ! + j ∥ u 0 ∥ L 2 + c ( ∥ u 0 ∥ H 1 δ 2 + δ 3 ) ( j + 1 ) 2 ( K B ) j j ! ) . \lVert\Gamma_t^j u(t)\rVert_{\mathcal{H}_1}\le e^{c\lVert u_0\rVert_{\mathcal{H}_1^2}}\left(\sqrt{1+j^2B^j}j!+j\lVert u_0\rVert_{L^2}+c\frac{\left(\lVert u_0\rVert_{\mathcal{H}_1}\delta^2+\delta^3\right)}{(j+1)^2}(KB)^j j!\right). ∥ Γ t j u ( t ) ∥ H 1 ≤ e c ∥ u 0 ∥ H 1 2 ( 1 + j 2 B j j ! + j ∥ u 0 ∥ L 2 + c ( j + 1 ) 2 ( ∥ u 0 ∥ H 1 δ 2 + δ 3 ) ( K B ) j j ! ) .
We now fix
δ = min { 1 2 c e c ∥ u 0 ∥ H 1 2 ( 1 + ∥ u 0 ∥ H 1 ) , 1 } ∈ ( 0 , 1 ] (83) \delta=\min\left\{\frac{1}{2ce^{c\lVert u_0\rVert_{\mathcal{H}_1^2}}(1+\lVert u_0\rVert_{\mathcal{H}_1})},1\right\}\in(0,1] \tag*{(83)} δ = min { 2 c e c ∥ u 0 ∥ H 1 2 ( 1 + ∥ u 0 ∥ H 1 ) 1 , 1 } ∈ ( 0 , 1 ] (83)
so that
c e c ∥ u 0 ∥ H 1 2 ( ∥ u 0 ∥ H 1 δ 2 + δ 3 ) ≤ δ 2 . ce^{c\lVert u_0\rVert_{\mathcal{H}_1^2}}\left(\lVert u_0\rVert_{\mathcal{H}_1}\delta^2+\delta^3\right)\le\frac{\delta}{2}. c e c ∥ u 0 ∥ H 1 2 ( ∥ u 0 ∥ H 1 δ 2 + δ 3 ) ≤ 2 δ .
Finally, we choose K = K ( ∥ u 0 ∥ H 2 ) ≥ 1 K=K(\lVert u_0\rVert_{\mathcal{H}_2})\ge1 K = K (∥ u 0 ∥ H 2 ) ≥ 1 satisfying (5.19) and
c e c ∥ u 0 ∥ H 1 2 ( 1 + j 2 + j ∥ u 0 ∥ L 2 ) ≤ K j δ 2 ( j + 1 ) 2 . ce^{c\lVert u_0\rVert_{\mathcal{H}_1^2}}\left(\sqrt{1+j^2}+j\lVert u_0\rVert_{L^2}\right)\le\frac{K^j\delta}{2(j+1)^2}. c e c ∥ u 0 ∥ H 1 2 ( 1 + j 2 + j ∥ u 0 ∥ L 2 ) ≤ 2 ( j + 1 ) 2 K j δ .
Then, we conclude that
sup 0 ≤ t ≤ 1 ∥ Γ t j u ( t ) ∥ H 1 ≤ δ ( j + 1 ) 2 ( K B ) j j ! , \sup_{0\le t\le1}\lVert\Gamma_t^j u(t)\rVert_{\mathcal{H}_1}\le\frac{\delta}{(j+1)^2}(KB)^j j!, 0 ≤ t ≤ 1 sup ∥ Γ t j u ( t ) ∥ H 1 ≤ ( j + 1 ) 2 δ ( K B ) j j ! ,
which is (5.18) in the case l = j l=j l = j . This concludes the proof of (5.18) by induction, which implies directly (5.13).
The proof of (5.15) follows similarly by applying the same argument to the function v ( x , t ) = u ( x , 1 − t ) v(x,t)=u(x,1-t) v ( x , t ) = u ( x , 1 − t ) that also solves (1.2).
Conclusion of the proof of the unique continuation theorem
In order to prove (5.6) we will prove the following proposition.
Proposition 5.3. Let 0 < α , β ≤ 1 2 0<\alpha,\beta\le\frac{1}{2} 0 < α , β ≤ 2 1 . Let 1 < b < 1 β − 1 1<b<\frac{1}{\beta}-1 1 < b < β 1 − 1 and 1 < a < 1 α − 1 1<a<\frac{1}{\alpha}-1 1 < a < α 1 − 1 . There exists a constant K 1 ≥ 1 K_1\ge1 K 1 ≥ 1 such that, for all j ∈ N j\in\mathbb{N} j ∈ N , j ≥ 1 j\ge1 j ≥ 1 , we have
E j ( u 1 ) ≤ K 1 j ( j ! ) 2 b (84) E_j(u_1)\le K_1^j(j!)^{\frac{2}{b}} \tag*{(84)} E j ( u 1 ) ≤ K 1 j ( j ! ) b 2 (84)
and
E j ( u 0 ) ≤ K 1 j ( j ! ) 2 a (85) E_j(u_0)\le K_1^j(j!)^{\frac{2}{a}} \tag*{(85)} E j ( u 0 ) ≤ K 1 j ( j ! ) a 2 (85)
Proof. We will again proceed by induction. For j = 1 j=1 j = 1 , we have by using (3.1), and the well-posedness in H 2 \mathcal{H}^2 H 2 ,
E 1 ( u 1 ) = ∥ x u 1 ∥ L 2 2 + ∥ x ∂ x u 1 ∥ L 2 2 ≤ c ∥ u 1 ∥ H 2 2 ≤ c ∥ u 0 ∥ H 2 2 ≤ K 1 E_1(u_1)=\lVert xu_1\rVert_{L^2}^2+\lVert x\partial_xu_1\rVert_{L^2}^2\le c\lVert u_1\rVert_{\mathcal{H}_2}^2\le c\lVert u_0\rVert_{\mathcal{H}_2}^2\le K_1 E 1 ( u 1 ) = ∥ x u 1 ∥ L 2 2 + ∥ x ∂ x u 1 ∥ L 2 2 ≤ c ∥ u 1 ∥ H 2 2 ≤ c ∥ u 0 ∥ H 2 2 ≤ K 1
for $K_1>c\lVert u_0\rVert_{\mathcal{H}_2}^2.
Now, let k ≥ 2 k \ge2 k ≥ 2 be fixed with k ≥ k 0 k \ge k_0 k ≥ k 0 , where k 0 k_0 k 0 is given in Proposition 5.1. Assume that (5.24) holds for j = 1 , … , k − 1 j=1,\ldots,k-1 j = 1 , … , k − 1 . We want to prove that (5.24) holds for j = k j=k j = k . From (5.2), (5.4), and using that b > 1 b>1 b > 1 , it follows that, for any integer l ≥ 1 l \ge1 l ≥ 1 ,
E b l ( u 1 ) ≤ c ( E b ( l + 1 ) ( u ~ 1 ) + ∥ u 1 ∥ H 1 2 ) . E_{bl}(u_1) \le c\left(E_{b(l+1)}(\widetilde{u}_1)+\lVert u_1\rVert_{\mathcal{H}_1}^2\right). E b l ( u 1 ) ≤ c ( E b ( l + 1 ) ( u 1 ) + ∥ u 1 ∥ H 1 2 ) .
By Proposition 5.1, with k = l + 1 k=l+1 k = l + 1 it follows that
E b l ( u 1 ) ≤ c ( ∥ u ~ 1 ∥ H l + 2 2 + c 2 b b + 1 l ∥ u 1 ∥ H 1 2 ) ≤ c l ( max 0 ≤ j ≤ l + 1 ∥ Γ 1 j u 1 ∥ H 1 2 + b 2 b b + 1 l ∥ u 1 ∥ H 1 2 ) . \begin{aligned}
E_{bl}(u_1) \le c\left(\lVert\widetilde{u}_1\rVert_{\mathcal{H}_{l+2}}^2+c^{\frac{2b}{b+1}l}\lVert u_1\rVert_{\mathcal{H}_1}^2\right) \\
&\le c^l\left(\max_{0\le j\le l+1}\lVert\Gamma_1^j u_1\rVert_{\mathcal{H}_1}^2+b^{\frac{2b}{b+1}l}\lVert u_1\rVert_{\mathcal{H}_1}^2\right).
\end{aligned} E b l ( u 1 ) ≤ c ( ∥ u 1 ∥ H l + 2 2 + c b + 1 2 b l ∥ u 1 ∥ H 1 2 ) ≤ c l ( 0 ≤ j ≤ l + 1 max ∥ Γ 1 j u 1 ∥ H 1 2 + b b + 1 2 b l ∥ u 1 ∥ H 1 2 ) .
Fix b < b ~ < 1 β − 1 b<\widetilde{b}<\frac{1}{\beta}-1 b < b < β 1 − 1 . We choose l = l ( k ) = ⌊ k b ⌋ l=l(k)=\left\lfloor\frac{k}{b}\right\rfloor l = l ( k ) = ⌊ b k ⌋ , i.e. l ∈ N l\in\mathbb{N} l ∈ N and k b ≤ l < k b + 1 \frac{k}{b}\le l<\frac{k}{b}+1 b k ≤ l < b k + 1 , then k ≤ b ~ l k\le\widetilde{b}l k ≤ b l . Using (5.2) and the above estimate with b = b ~ b=\widetilde{b} b = b , it follows that
E k ( u 1 ) ≤ c l ( max 0 ≤ j ≤ l + 1 ∥ Γ 1 j u 1 ∥ H 1 2 + l 2 b ~ b ~ + 1 ∥ u 1 ∥ H 1 2 ) . (86) E_k(u_1)\le c^l\left(\max_{0\le j\le l+1}\lVert\Gamma_1^j u_1\rVert_{\mathcal{H}_1}^2+l^{\frac{2\widetilde{b}}{\widetilde{b}+1}}\lVert u_1\rVert_{\mathcal{H}_1}^2\right). \tag*{(86)} E k ( u 1 ) ≤ c l ( 0 ≤ j ≤ l + 1 max ∥ Γ 1 j u 1 ∥ H 1 2 + l b + 1 2 b ∥ u 1 ∥ H 1 2 ) . (86)
Let k 1 = 2 b ~ b ~ − 1 + 1 k_1=\frac{2\widetilde{b}}{\widetilde{b}-1}+1 k 1 = b − 1 2 b + 1 . If k ≥ k 1 k\ge k_1 k ≥ k 1 , we have l + 1 < k b + 2 < k l+1<\frac{k}{b}+2<k l + 1 < b k + 2 < k . The induction hypothesis then yields
E j ( u 1 ) ≤ K 1 j ( j ! ) 2 b < K 1 j ( j ! ) 2 , ∀ 1 ≤ j ≤ l + 1. E_j(u_1)\le K_1^j(j!)^{\frac{2}{b}}<K_1^j(j!)^2,\quad\forall\ 1\le j\le l+1. E j ( u 1 ) ≤ K 1 j ( j ! ) b 2 < K 1 j ( j ! ) 2 , ∀ 1 ≤ j ≤ l + 1.
Hence, we can apply Proposition 5.2, with J = l + 1 J=l+1 J = l + 1 and B = K 1 1 2 B=K_1^{\frac{1}{2}} B = K 1 2 1 . It follows that there exists K = K ( ∥ u 0 ∥ H 2 ) ≥ 1 K=K(\lVert u_0\rVert_{\mathcal{H}^2})\ge1 K = K (∥ u 0 ∥ H 2 ) ≥ 1 such that
E k ( u 1 ) ≤ c l ( ( K K 1 ) l + 1 ( ( l + 1 ) ! ) 2 + l 2 b ~ b ~ + 1 ∥ u 1 ∥ H 1 2 ) ≤ K 1 k ( c l K l + 1 ( l ! ) 2 + l 2 b ~ b ~ + 1 ∥ u 1 ∥ H 1 2 ) . \begin{aligned}
E_k(u_1)\le c^l\left((KK_1)^{l+1}((l+1)!)^2+l^{\frac{2\widetilde{b}}{\widetilde{b}+1}}\lVert u_1\rVert_{\mathcal{H}_1}^2\right) \\
&\le K_1^k\left(c^lK^{l+1}(l!)^2+l^{\frac{2\widetilde{b}}{\widetilde{b}+1}}\lVert u_1\rVert_{\mathcal{H}_1}^2\right).
\end{aligned} E k ( u 1 ) ≤ c l ( ( K K 1 ) l + 1 (( l + 1 )! ) 2 + l b + 1 2 b ∥ u 1 ∥ H 1 2 ) ≤ K 1 k ( c l K l + 1 ( l ! ) 2 + l b + 1 2 b ∥ u 1 ∥ H 1 2 ) .
Since l = k b + O ( 1 ) l=\frac{k}{b}+O(1) l = b k + O ( 1 ) , it follows by using Stirling’s formula that
E k ( u 1 ) ≤ K 1 k c k K k + 1 ( k ! ) 2 b ≤ 1 2 K 1 k ( k ! ) 2 b E_k(u_1)\le K_1^k c^k K^{k+1}(k!)^{\frac{2}{b}}\le\frac{1}{2}K_1^k(k!)^{\frac{2}{b}} E k ( u 1 ) ≤ K 1 k c k K k + 1 ( k ! ) b 2 ≤ 2 1 K 1 k ( k ! ) b 2
for k ≥ k 2 ( ∥ u 0 ∥ H 2 ) k\ge k_2(\lVert u_0\rVert_{\mathcal{H}^2}) k ≥ k 2 (∥ u 0 ∥ H 2 ) , where k 2 ≥ max { k 0 , k 1 } k_2\ge\max\{k_0,k_1\} k 2 ≥ max { k 0 , k 1 } is chosen such that
sup k ≥ k 1 c k K k + 1 e 2 k ( 1 b − 1 b ~ ) log k ≤ 1 2 , \sup_{k\ge k_1}c^kK^{k+1}e^{2k\left(\frac{1}{b}-\frac{1}{\widetilde{b}}\right)\log k}\le\frac{1}{2}, k ≥ k 1 sup c k K k + 1 e 2 k ( b 1 − b 1 ) l o g k ≤ 2 1 ,
since b < b ~ b<\widetilde{b} b < b . This proves (5.24) for j = k j=k j = k and k ≥ k 2 k\ge k_2 k ≥ k 2 .
For k < k 2 k<k_2 k < k 2 , we know, since u 1 ∈ S ( R ) u_1\in\mathcal{S}(\mathbb{R}) u 1 ∈ S ( R ) that E k ( u 1 ) < ∞ E_k(u_1)<\infty E k ( u 1 ) < ∞ for all 1 ≤ k ≤ k 2 1\le k\le k_2 1 ≤ k ≤ k 2 . Then, there exists a constant K 2 ≥ 1 K_2\ge1 K 2 ≥ 1 such that
E k ( u 1 ) ≤ K 2 , ∀ 1 ≤ k ≤ k 2 . E_k(u_1)\le K_2,\quad\forall\ 1\le k\le k_2. E k ( u 1 ) ≤ K 2 , ∀ 1 ≤ k ≤ k 2 .
By enlarging the constant K 1 K_1 K 1 if necessary, we conclude that (5.24) holds for all k ≥ 1 k\ge1 k ≥ 1 .
The proof of (5.25) follows in a similar way. □
Now, we conclude the proof of Theorem 1.2.
Proof of Theorem 1.2. Let 1 < p , q < ∞ 1<p,q<\infty 1 < p , q < ∞ be such that 1 p + 1 q = 1 \frac{1}{p}+\frac{1}{q}=1 p 1 + q 1 = 1 to be fixed later. By applying (5.24) for j = ⌊ q k 2 ⌋ j=\left\lfloor\frac{qk}{2}\right\rfloor j = ⌊ 2 q k ⌋ , and using Stirling’s formula, we deduce
E q k 2 ( u 1 ) ≤ C 1 k ( k ! ) q δ 1 , (87) E_{\frac{qk}{2}}(u_1)\le C_1^k(k!)^{q\delta_1}, \tag*{(87)} E 2 q k ( u 1 ) ≤ C 1 k ( k ! ) q δ 1 , (87)
for δ 1 = 1 b \delta_1=\frac{1}{b} δ 1 = b 1 . Similarly, (5.25) with j = ⌊ p k 2 ⌋ j=\left\lfloor\frac{pk}{2}\right\rfloor j = ⌊ 2 p k ⌋ yields
E p k 2 ( u 0 ) ≤ C 1 k ( k ! ) p δ 0 . (88) E_{\frac{pk}{2}}(u_0)\le C_1^k(k!)^{p\delta_0}. \tag*{(88)} E 2 p k ( u 0 ) ≤ C 1 k ( k ! ) p δ 0 . (88)
for δ 0 = 1 a \delta_0=\frac{1}{a} δ 0 = a 1 . Now, we optimize p p p and q q q as in the proof of Theorem 1.1. We fix p p p and q q q such that
p δ 0 = q δ 1 ⟺ q = δ 0 + δ 1 δ 1 . p\delta_0=q\delta_1 \Longleftrightarrow q=\frac{\delta_0+\delta_1}{\delta_1}. p δ 0 = q δ 1 ⟺ q = δ 1 δ 0 + δ 1 .
Then p δ 0 < 1 p\delta_0<1 p δ 0 < 1 and q δ 1 < 1 ⟺ δ 0 + δ 1 = 1 a + 1 b < 1 q\delta_1<1 \Longleftrightarrow\delta_0+\delta_1=\frac{1}{a}+\frac{1}{b}<1 q δ 1 < 1 ⟺ δ 0 + δ 1 = a 1 + b 1 < 1 .
Remind by hypothesis that
α 1 − α + β 1 − β < 1. \frac{\alpha}{1-\alpha}+\frac{\beta}{1-\beta}<1. 1 − α α + 1 − β β < 1.
Now choose 1 < a < 1 α − 1 1<a<\frac{1}{\alpha}-1 1 < a < α 1 − 1 and 1 < b < 1 β − 1 1<b<\frac{1}{\beta}-1 1 < b < β 1 − 1 such that
α 1 − α + β 1 − β < 1 a + 1 b < 1 , \frac{\alpha}{1-\alpha}+\frac{\beta}{1-\beta}<\frac{1}{a}+\frac{1}{b}<1, 1 − α α + 1 − β β < a 1 + b 1 < 1 ,
which guarantees then that p δ 0 < 1 p\delta_0<1 p δ 0 < 1 and q δ 1 < 1 q\delta_1<1 q δ 1 < 1 .
Therefore, we conclude that
∫ R e c ∣ x ∣ q ∣ u 1 ( x ) ∣ 2 d x = ∑ k = 0 ∞ c k k ! ∫ ∣ x ∣ q k ∣ u 1 ( x ) ∣ 2 d x ≤ ∑ k = 0 ∞ c k k ! E q k 2 ( u 1 ) ≤ ∑ k = 0 ∞ e ( q δ 1 − 1 ) k log k + O ( k ) < + ∞ \int_{\mathbb{R}} e^{c|x|^q}|u_1(x)|^2dx=\sum_{k=0}^{\infty}\frac{c^k}{k!}\int|x|^{qk}|u_1(x)|^2dx\leq\sum_{k=0}^{\infty}\frac{c^k}{k!}E_{\frac{qk}{2}}(u_1)\leq\sum_{k=0}^{\infty}e^{(q\delta_1-1)k\log k+O(k)}<+\infty ∫ R e c ∣ x ∣ q ∣ u 1 ( x ) ∣ 2 d x = k = 0 ∑ ∞ k ! c k ∫ ∣ x ∣ q k ∣ u 1 ( x ) ∣ 2 d x ≤ k = 0 ∑ ∞ k ! c k E 2 q k ( u 1 ) ≤ k = 0 ∑ ∞ e ( q δ 1 − 1 ) k l o g k + O ( k ) < + ∞
and
∫ R e c ∣ x ∣ p ∣ u 0 ( x ) ∣ 2 d x = ∑ k = 0 ∞ c k k ! ∫ ∣ x ∣ p k ∣ u 0 ( x ) ∣ 2 d x ≤ ∑ k = 0 ∞ c k k ! E p k 2 ( u 0 ) ≤ ∑ k = 0 ∞ e ( p δ 0 − 1 ) k log k + O ( k ) < + ∞ \int_{\mathbb{R}} e^{c|x|^p}|u_0(x)|^2dx=\sum_{k=0}^{\infty}\frac{c^k}{k!}\int|x|^{pk}|u_0(x)|^2dx\leq\sum_{k=0}^{\infty}\frac{c^k}{k!}E_{\frac{pk}{2}}(u_0)\leq\sum_{k=0}^{\infty}e^{(p\delta_0-1)k\log k+O(k)}<+\infty ∫ R e c ∣ x ∣ p ∣ u 0 ( x ) ∣ 2 d x = k = 0 ∑ ∞ k ! c k ∫ ∣ x ∣ p k ∣ u 0 ( x ) ∣ 2 d x ≤ k = 0 ∑ ∞ k ! c k E 2 p k ( u 0 ) ≤ k = 0 ∑ ∞ e ( p δ 0 − 1 ) k l o g k + O ( k ) < + ∞
and by the the Morgan uncertainty principle (Theorem 1.5) we deduce that u ≡ 0 u\equiv0 u ≡ 0 .
References
Torunn S. Jensen, Department of Mathematics, University of Bergen, PO Box 7803, 5020 Bergen, Norway Email address: Torunn.Jensen@uib.no
Didier Pilod, Department of Mathematics, University of Bergen, PO Box 7803, 5020 Bergen, Norway Email address: Didier.Pilod@uib.no
João P. G. Ramos, IMPA, Rio de Janeiro, Brazil Email address: joaopgramos95@gmail.com