Abstract:We propose and study a new multilevel method for the numerical approximation of a Gibbs distribution $\pi$ on $\mathbb{R}^d$, based on (overdamped) Langevin diffusions. This method inspired by \cite{mainPPlangevin} and \cite{giles_szpruch_invariant} relies on a multilevel occupation measure, $i.e.$ on an appropriate combination of $R$ occupation measures of (constant-step) Euler schemes with respective steps $\gamma_r = \gamma_0 2^{-r}$, $r=0,\ldots,R$. We first state a quantitative result under general assumptions which guarantees an \textit{$\varepsilon$-approximation} (in a $L^2$-sense) with a cost of the order $\varepsilon^{-2}$ or $\varepsilon^{-2}|\log \varepsilon|^3$ under less contractive assumptions.
We then apply it to overdamped Langevin diffusions with strongly convex potential $U:\mathbb{R}^d\rightarrow\mathbb{R}$ and obtain an \textit{$\varepsilon$-complexity} of the order ${\cal O}(d\varepsilon^{-2}\log^3(d\varepsilon^{-2}))$ or ${\cal O}(d\varepsilon^{-2})$ under additional assumptions on $U$. More precisely, up to universal constants, an appropriate choice of the parameters leads to a cost controlled by ${(\bar{\lambda}_U\vee 1)^2}{\underline{\lambda}_U^{-3}} d\varepsilon^{-2}$ (where $\bar{\lambda}_U$ and $\underline{\lambda}_U$ respectively denote the supremum and the infimum of the largest and lowest eigenvalue of $D^2U$). We finally complete these theoretical results with some numerical illustrations including comparisons to other algorithms in Bayesian learning and opening to non strongly convex setting.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about how to effectively numerically approximate the invariant distribution (i.e., Gibbs distribution) of the diffusion process, especially in high - dimensional cases. Specifically, the author proposes a new multi - layer method, based on (over - damped) Langevin diffusion, to approximate the numerical solution of the Gibbs distribution on \(\mathbb{R}^d\). This method is inspired by [PP18] and [GMS+20] and relies on multi - layer occupation measures, that is, the appropriate combination of the occupation measures of the constant - step - size Euler scheme with step sizes \(\gamma_r=\gamma_0 2^{-r}\), where \(r = 0,\ldots,R\).
### Main problems
1. **Complexity of numerical approximation**:
- The author aims to provide a method that guarantees \(\varepsilon\)-approximation (in the \(L^2\) sense) under general assumptions, and the complexity of this method is \(\varepsilon^{-2}\) or \(\varepsilon^{-2}|\log\varepsilon|^3\) under weaker contraction assumptions.
- For the over - damped Langevin diffusion with strongly convex potential \(U:\mathbb{R}^d\rightarrow\mathbb{R}\), the author obtains the \(\varepsilon\)-complexity results of \(O(d\varepsilon^{-2}\log^3(d\varepsilon^{-2}))\) or \(O(d\varepsilon^{-2})\), depending on the additional assumptions of the potential \(U\).
2. **Parameter selection**:
- By appropriately selecting parameters, the cost can be controlled to be dominated by \((\bar{\lambda}_U\vee 1)^2\lambda_U^{-3}d\varepsilon^{-2}\), where \(\bar{\lambda}_U\) and \(\lambda_U\) represent the supremum and infimum of the maximum and minimum eigenvalues of \(D^2U\), respectively.
3. **Theoretical and numerical verification**:
- The author not only provides theoretical results but also verifies them through numerical experiments, including comparison with other algorithms in Bayesian learning and exploration of applications in non - strongly convex settings.
### Specific objectives
- **Multi - layer method for general diffusion processes**: Under general assumptions, provide a multi - layer method that can achieve \(\varepsilon\)-approximation with a complexity of \(\varepsilon^{-2}\) or \(\varepsilon^{-2}|\log\varepsilon|^3\).
- **Optimization under strongly convex potential**: For the over - damped Langevin diffusion with strongly convex potential \(U\), optimize the parameter selection so that the complexity is \(O(d\varepsilon^{-2}\log^3(d\varepsilon^{-2}))\) or \(O(d\varepsilon^{-2})\).
- **Numerical verification**: Verify the validity of theoretical results through numerical experiments and compare with existing methods.
### Method overview
- **Multi - layer occupation measures**: Use multi - layer occupation measures to reduce bias and variance by appropriately combining the occupation measures of Euler schemes with different step sizes.
- **Parameter selection**: By selecting appropriate parameters, such as step size \(\gamma_r\) and time \(T_r\), ensure that the complexity of the algorithm is optimized.
- **Numerical experiments**: Through specific numerical experiments, verify the effectiveness of the algorithm and compare it with other methods.
### Theoretical contributions
- **General results**: Under general assumptions, provide \(\varepsilon\)-complexity analysis of the multi - layer method.
- **Optimization under strongly convex potential**: For the strongly convex potential \(U\), provide a more refined complexity analysis and optimize the parameter selection.
- **Numerical verification**: Verify the correctness and effectiveness of theoretical results through numerical experiments.
### Numerical experiments
- **Different