SearcharxivSearch

arXiv subjects

Alexis Knezevitch

Publications and source records attributed to Alexis Knezevitch.

3 recordsLinked to original sources

Quantitative quasi-invariance of Gaussian measures below the energy level for the 1D generalized nonlinear Schr\"odinger equation and application to global well-posedness

We consider the Schr\"odinger equation on the one dimensional torus with a general odd-power nonlinearity $p \geq 5$, which is known to be globally well-posed in the Sobolev space $H^\sigma(\mathbb{T})$, for every $\sigma \geq 1$, thanks to the conservation and finiteness of the energy. For regularities $\sigma < 1$, where this energy is infinite, we explore a globalization argument adapted to random initial data distributed according to the Gaussian measures $\mu_s$, with covariance operator $(1-\Delta)^s$, for $s$ in a range $(s_p,\frac{3}{2}]$. We combine a deterministic local Cauchy theory with the quasi-invariance of Gaussian measures $\mu_s$, with additional $L^q$-bounds on the Radon-Nikodym derivatives, to prove that the Gaussian initial data generate almost surely global solutions. These $L^q$-bounds are obtained with respect to Gaussian measures accompanied by a cutoff on a renormalization of the energy; the main tools to prove them are the Bou\'e-Dupuis variational formula and a Poincar\'e-Dulac normal form reduction. This approach is similar in spirit to Bourgain's invariant argument and to a recent work by Forlano-Tolomeo.

math.AP

Qualitative quasi-invariance of low regularity Gaussian measures for the 1d quintic nonlinear Schr\"odinger equation

We consider the 1d quintic nonlinear Schr\"odinger equation (NLS) on the torus with initial data distributed according to the Gaussian measures with covariance operator $(1-\Delta)^{-s}$, and denoted $\mu_s$. For the full range $s>\frac{9}{10}$, we prove that these Gaussian measures are quasi-invariant along the flow of (NLS), meaning that the law of the solution at any time is absolutely continuous with respect to the initial Gaussian measure. Moreover, the condition $s>\frac{9}{10}$ corresponds to the threshold where the Sobolev space $H^{\frac{2}{5}+}(\mathbb{T})$ is of $\mu_s$-full measure (it is of zero $\mu_s$-measure otherwise). This is the lower regularity Sobolev space where we currently know that (NLS) is globally well-posed, thanks to a work by LI-WU-XU. The present work extends the known threshold $s>\frac{3}{2}$ for the quasi-invariance down to $s>\frac{9}{10}$, but we do not obtain here quantitative results on the Radon-Nikodym derivatives. Our approach is based on a work of Sun-Tzvetkov, combining a Poincar\'e-Dulac normal form reduction with energy estimates. However, our main tool to obtain these energy estimates differs: we use the Bou\'e-Dupuis variational formula instead of Wiener Chaos.

math.AP

Transport of low regularity Gaussian measures for the 1d quintic nonlinear Schr\"odinger equation

We consider the 1d nonlinear Schr\"odinger equation (NLS) on the torus with initial data distributed according to the Gaussian measure with covariance operator $(1 - \Delta)^{-s}$, where $\Delta$ is the Laplace operator. We prove that the Gaussian measures are quasi-invariant along the flow of (NLS) for the full range $s > \frac{3}{2}$. This improves a previous result obtained by Planchon, Tzvetkov and Visciglia (in 2019), where the quasi-invariance is proven for $s=2k$, for all integers $k\geq 1$. In our approach, to prove the quasi-invariance, we directly establish an explicit formula for the Radon-Nikodym derivative $G_s(t,.)$ of the transported measures, which is obtained as the limit of truncated Radon-Nikodym derivatives $G_{s,N}(t,.)$ for transported measures associated with a truncated system. We also prove that the Radon-Nikodym derivatives belong to $L^p$, $p>1$, with respect to $H^1(\mathbb{T})$-cutoff Gaussian measures, relying on the introduction of weighted Gaussian measures produced by a normal form reduction, following a recent work by Sun and Tzvetkov (in 2023). Additionally, we prove that the truncated densities $G_{s,N}(t,.)$ converges to $G_s(t,.)$ in $L^p$ (with respect to the $H^1(\mathbb{T})$-cutoff Gaussian measures).

math.AP