Quantitative periodic homogenization of parabolic equations with large drift and potential
Kshitij Sinha · 2025
Read Quantitative periodic homogenization of parabolic equations with large drift and potential by Kshitij Sinha (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-22003.txt.
Introduction
Kshitij Sinha, writing alone, studies how fast a smoothing process takes over in equations with a repeating microscopic pattern. The 2025 paper treats parabolic equations, which describe diffusion-like change over time, with two large extra terms: a drift and a potential. Homogenization replaces the rapidly varying equation by an averaged one, and the quantitative question is how fast the two solutions merge as the pattern shrinks.
Sinha shows the solution splits into a product of three parts. One part depends only on time. Another is the ground state of an exponential eigenvalue problem posed on a single pattern cell. The third solves an averaged equation whose effective drift is zero. For that last part, the paper derives rates of convergence in the energy norm as the pattern scale goes to zero. The result clarifies how strong drift and potential terms shape the averaged behavior of such equations.
Catalog record
| CREATOR | Kshitij Sinha |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-22003
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Quantitative periodic homogenization of parabolic equations with large drift and potential Kshitij Sinha Abstract This work aims to study the rates in the context of periodic homogenization of parabolic problems with large lower order terms (both drift and potential). We demonstrate that the solution is a product of three terms: (i) a function of time, (ii) the ground-state of an exponential cell eigenvalue problem and (iii) the solution to a parabolic equation with zero effective drift. For the latter, we derive L 2 \displaystyle\mathrm{L}^{2} rates in the homogenization limit. 1 Introduction In this paper, for the unknown \displaystyle u_{\varepsilon}:\overline{\Omega}\times[0,T)\to\mathbb{R} , we study the following parabolic initial boundary value problem: \left\{\begin{aligned} \frac{\partial u_{\varepsilon}}{\partial t}-\dv\left(A\left(\frac{x}{\varepsilon}\right)\nabla u_{\varepsilon}\right)+\frac{1}{\varepsilon}b\left(\frac{x}{\varepsilon}\right)\cdot\nabla u_{\varepsilon}+\frac{1}{\varepsilon^{2}}c\left(\frac{x}{\varepsilon}\right)u_{\varepsilon}&=0\qquad\mbox{ in }(0,T)\times\Omega,\\ u_{\varepsilon}(t,x)&=0\qquad\mbox{ on }(0,T)\times\partial\Omega,\\ u_{\varepsilon}(0,x)&=u_{\varepsilon}^{0}(x)\qquad\mbox{ in }\Omega.\end{aligned}\right. (1.1) Here the parameter \displaystyle\varepsilon\in(0,1) . The bounded domain \displaystyle\Omega\subset\mathbb{R}^{d} is assumed to be open and with Lipschitz boundary \displaystyle\partial\Omega . The coefficients \displaystyle A,b,c are assumed to be bounded in \displaystyle\mathrm{L}^{\infty} and to be \displaystyle Y -periodic with the standard notation \displaystyle Y:=[0,1)^{d} for the unit cell. Furthermore, \displaystyle A is assumed to be uniformly elliptic, i.e. A(y)\xi\cdot\xi\geq\mu\left|\xi\right|^{2}\quad\forall\xi\in\mathbb{R}^{d}\,\mbox{ and for a.e. }y\in Y, for some \displaystyle\mu>0 . We will be commenting more on the initial datum \displaystyle u^{0}_{\varepsilon} later. Our present work is inspired by • a result due to Grégoire Allaire and Rafael Orive [ 3 ] dealing with the parabolic problems with large drift and potential terms • an article of Grégoire Allaire and Anne-Lise Raphael [ 4 ] which dealt with parabolic problems with large drift and potential terms in periodic porous media • a work of Yves Capdeboscq [ 7 ] addressing the periodic homogenization of eigenvalue problems for elliptic operators involving large drift terms We also recall a work due to Grégoire Allaire and co-authors [ 2 ] dealing with parabolic problems with large potential terms (and no drift terms). Our present work is likewise related to the work of Patrizia Donato and Andrey Piatnitski [ 10 ] . In this work, we demonstrate that the solution to the initial boundary value problem ( 1.1 ) can be factorized as follows: u_{\varepsilon}(t,x)=e^{\frac{-\lambda t}{\varepsilon^{2}}}v_{\varepsilon}(t,x)\psi\left(\frac{x}{\varepsilon}\right) (1.2) where \displaystyle(\lambda,\psi) are an eigenpair associated with a certain exponential \displaystyle Y -periodic eigenvalue problem. Factorization similar to ( 1.2 ) are already available in [ 4 , 3 , 10 ] . The use of exponential \displaystyle Y -periodic eigenvalue problems in homogenization can be traced back to [ 6 , page 383] . Allaire and Orive in [ 3 , Lemma 2.1] (see also [ 7 , Lemma 3] , where Capdeboscq treats operators with no lower order terms) show the existence of a unique \displaystyle\theta\in\mathbb{R}^{d} such that the eigenvalue problems (with \displaystyle A^{\ast} denoting the adjoint of \displaystyle A ): \left\{\begin{aligned} -\dv_{y}({A}(y)\nabla\psi(y))+b(y)\cdot\nabla_{y}\psi(y)+c(y)\psi(y)=\lambda\psi(y)\qquad\mbox{ in }(0,1)^{d}\\ y\mapsto\psi(y)e^{-2\pi\theta\cdot y}\quad\mbox{ is }Y\mbox{-periodic},\end{aligned}\right. (1.3) and \left\{\begin{aligned} -\dv_{y}({A}^{\ast}(y)\nabla\psi^{\ast}(y))-\dv_{y}(b(y)\psi^{\ast}(y))+c(y)\psi^{\ast}(y)=\lambda^{\ast}\psi^{\ast}(y)\qquad\mbox{ in }(0,1)^{d}\\ y\mapsto\psi^{\ast}(y)e^{2\pi\theta\cdot y}\quad\mbox{ is }Y\mbox{-periodic},\end{aligned}\right. (1.4) have the following properties: • Both the problems ( 1.3 ) and ( 1.4 ) have the same first eigenvalue which is real, positive and simple. • The first eigenfunctions of ( 1.3 ) and ( 1.4 ) can be chosen such that \min_{Y}\{\psi,\psi^{\ast}\}\geq a>0\qquad\mbox{ and }\qquad\int_{Y}\psi(y)\psi^{\ast}(y)\,{\rm d}y=1. (1.5) • The function \displaystyle\beta:Y\to\mathbb{R}^{d} defined by \beta(y):=\psi(y)\psi^{\ast}(y)b(y)+\psi(y)A^{\ast}(y)\nabla_{y}\psi^{\ast}(y)-\psi^{\ast}(y)A(y)\nabla_{y}\psi(y)\quad\mbox{ for }y\in Y, (1.6) is divergence-free and is of zero average over \displaystyle Y , i.e. \int_{Y}\beta(y)\,{\rm d}y=0. The factorization ( 1.2 ) implies that \displaystyle u_{\varepsilon}(0,x) and \displaystyle v_{\varepsilon}(0,x) are equal to each other up to a factor of the first eigenfunction \displaystyle\psi\left(\frac{x}{\varepsilon}\right) mentioned above. Observe further that, for the factorization ( 1.2 ) to hold, the factor \displaystyle v_{\varepsilon} should necessarily satisfy the following initial boundary value problem: \left\{\begin{aligned} \sigma\left(\frac{x}{\varepsilon}\right)\frac{\partial v_{\varepsilon}}{\partial t}-\operatorname{div}\left(\alpha\left(\frac{x}{\varepsilon}\right)\nabla v_{\varepsilon}\right)+\frac{1}{\varepsilon}\beta\left(\frac{x}{\varepsilon}\right)\cdot\nabla v_{\varepsilon}&=0\qquad\mbox{ in }(0,T)\times\Omega,\\ v_{\varepsilon}&=0\qquad\mbox{ on }(0,T]\times\partial\Omega,\\ v_{\varepsilon}(0,x)&=u_{0}(x)\qquad\mbox{ in }\Omega,\end{aligned}\right. (1.7) where the coefficient vector field \displaystyle\beta is defined in ( 1.6 ) and the coefficients \displaystyle\sigma and \displaystyle\alpha are \displaystyle Y -periodic and are defined as follows: \sigma(y):=\psi(y)\psi^{\ast}(y)\quad\mbox{ and }\quad\alpha(y):=\sigma(y)A(y)\quad\mbox{ for a.e. }y\in Y. (1.8) Remark 1.1 . The coefficient \displaystyle\sigma is \displaystyle Y -periodic because \sigma(y)=\psi(y)\psi^{\ast}(y)=\psi(y)e^{-2\pi\theta\cdot y}e^{2\pi\theta\cdot y}\psi^{\ast}(y) Remark 1.2 . We have made a choice of taking the initial datum for \displaystyle v_{\varepsilon} in ( 1.7 ) to be \displaystyle\varepsilon -independent. The factorization ( 1.2 ), then, forces the initial datum for \displaystyle u_{\varepsilon} in ( 1.1 ) to be of the form u_{\varepsilon}(0,x)=u^{0}_{\varepsilon}(x)=u_{0}(x)\psi\left(\frac{x}{\varepsilon}\right) This is sometimes referred to as the well prepared initial datum. Similar well preparedness assumptions on the data have been made in the literature [ 2 , 3 ] . This choice results in strong convergence of the factor \displaystyle v_{\varepsilon} to its homogenized limit. If one were to keep the data \displaystyle u_{\varepsilon}(0,x) independent of the parameter \displaystyle\varepsilon , then the factorization forces the initial datum for \displaystyle v_{\varepsilon} to be v_{\varepsilon}(0,x)=\frac{u_{0}(x)}{\psi\left(\frac{x}{\varepsilon}\right)}. (1.9) It should be noted that the equation ( 1.7 ) can still be homogenized with ( 1.9 ) as the initial datum. The homogenized equation remains the same, but with the initial datum for the limit problem being \left(\int_{Y}\frac{1}{\psi(y)}\,{\rm d}y\right)u_{0}(x). All of this can be argued using the notion of two-scale convergence [ 1 ] , for example. In this case, the convergence will be weak and hence cannot talk about rates. Our main result is the following quantitative homogenization result for the factor \displaystyle v_{\varepsilon} in ( 1.2 ) which solves the initial boundary value problem ( 1.7 ). Theorem 1.3 . Let \displaystyle v_{\varepsilon} be the solution to ( 1.7 ) and let \displaystyle v_{0} be the solution to the corresponding homogenized problem ( 1.12 ). Then, there exists a positive constant \displaystyle C , independent of \displaystyle\varepsilon , such that the following estimate holds: \displaystyle\left\|v_{\varepsilon}-v_{0}\right\|_{\mathrm{L}^{2}(\Omega_{T})} \displaystyle\leq C\varepsilon^{\frac{1}{4}}\left\{\left\|\partial_{t}v_{0}\right\|_{L^{2}(0,T;W^{1,d}(\Omega))}+\left\|v_{0}\right\|_{\mathrm{L}^{2}(0,T;\mathrm{H}^{2}(\Omega))}+\left\|u_{0}\right\|_{L^{2}(\Omega)}\right\}. (1.10) In the above result, there is an inherent assumption that the terms on the right hand side of the bound are finite. This essentially translates to the data being sufficient smooth to guarantee adequate regularity for the solution \displaystyle v_{0} of the homogenized equation. The proof of the above theorem is inspired by the ideas in the work of Jun Geng and Zhongwei Shen [ 11 ] that dealt with \displaystyle\mathrm{L}^{2} convergence rates in the context of periodic homogenization of parabolic problems. Remark 1.4 . Our estimate ( 1.10 ) is not sharp as in [ 11 ] . In there, the authors have obtained an \displaystyle\mathcal{O}(\varepsilon) rate. Upon deriving a \displaystyle\mathcal{O}(\sqrt{\varepsilon}) rate for \displaystyle\left\|v_{\varepsilon}-v_{0}\right\|_{\mathrm{L}^{2}(\Omega_{T})} , the authors employ a duality argument to improve the rate to \displaystyle\mathcal{O}(\varepsilon) . We emphasise here that the same strategy can be followed in our case as well. Thus, improving the rate in Theorem 1.3 to \displaystyle\mathcal{O}(\sqrt{\varepsilon}) . Before we comment further on the proof of Theorem 1.3 , we return to the equation ( 1.7 ). As \displaystyle\beta (defined by ( 1.6 )) is periodic, divergence-free and is of zero average, there exists a periodic skew-symmetric matrix-valued function \displaystyle y\mapsto B(y) such that (see [ 7 , Lemma 1, p. 809] ) \beta(y)=-\dv_{y}B(y)\qquad\mbox{ in }Y. Taking a \displaystyle Y -periodic matrix-valued function \displaystyle y\mapsto M(y):=\alpha(y)+B(y) , the evolution equation in ( 1.7 ) can be rewritten as \sigma\left(\frac{x}{\varepsilon}\right)\frac{\partial v_{\varepsilon}}{\partial t}-\dv\left(M\left(\frac{x}{\varepsilon}\right)\nabla v_{\varepsilon}\right)=0\qquad\mbox{ in }(0,T)\times\Omega. (1.11) Observe that \displaystyle M is also elliptic: M(y)\xi\cdot\xi=\alpha(y)\xi\cdot\xi\geq b\left|\xi\right|^{2}\quad\forall\xi\in\mathbb{R}^{d}\,\mbox{ and for a.e. }y\in Y, where the ellipticity constant \displaystyle b depends on