Skip to main content
added more detail, made the notation a bit more consistent with the links, and directly addressed the question asked by the OP
Source Link

The reason a Langevin diffusion leaves $\nu(x)=e^{-U(x)}$ invariant is because it is reversible with respectsymmetric or reversible with respect to $\nu$. In comparison to general diffusion processes, the ergodic properties of symmetric diffusions are not only easier to analyze, but it is often possible to obtain an explicit formula for the stationary density $\rho(x)=e^{-U(x)}$of a symmetric diffusion. This symmetry property is a bit more transparent when the SDELangevin equation is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is$$ dX_t = \nabla \log \nu(X_t) dt + \sqrt{2} dW_t \;. \tag{$\star$} $$ The symmetry property can then be easily verified by checking that the basisinfinitesimal generator of MCMC methods based onthe Langevin diffusions; see ediffusion is symmetric in an $L^2$-inner product weighted by $\nu$.g This property of the Langevin diffusion is analogous to the detailed balance condition of a Markov chain.

Exponential convergence of Langevin diffusions and their discrete approximations

It is also the basis of standard MCMC methods based on Langevin diffusions. In fact, many random-walk based MCMC methods (like random walk Metropolis) with target probability density proportional to $\rho(x)$$\nu(x)$, can be viewed as weak approximations toweakly approximate the corresponding Langevin diffusion; seediffusion ($\star$). This is remarkable because, e.g. Theorem 5.1 of

Metropolis integration schemes for self-adjoint diffusions

In particular, the random walk Metropolis doesn’t even explicitlyalgorithm does not involve the gradient of $U$, but by virtue of being reversible.

To read more about this connection, can weakly approximate the Langevin diffusionsee the introduction and Theorem 5.2 of this paper.

The Langevin diffusion is reversible with respect to the density $\rho(x)=e^{-U(x)}$. This property is more transparent when the SDE is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is the basis of MCMC methods based on Langevin diffusions; see e.g.

Exponential convergence of Langevin diffusions and their discrete approximations

In fact, random-walk based MCMC methods (like random walk Metropolis) with target density $\rho(x)$, can be viewed as weak approximations to the corresponding Langevin diffusion; see e.g. Theorem 5.1 of

Metropolis integration schemes for self-adjoint diffusions

In particular, random walk Metropolis doesn’t even explicitly involve the gradient of $U$, but by virtue of being reversible, can weakly approximate the Langevin diffusion.

The reason a Langevin diffusion leaves $\nu(x)=e^{-U(x)}$ invariant is because it is symmetric or reversible with respect to $\nu$. In comparison to general diffusion processes, the ergodic properties of symmetric diffusions are not only easier to analyze, but it is often possible to obtain an explicit formula for the stationary density of a symmetric diffusion. This symmetry property is a bit more transparent when the Langevin equation is written as $$ dX_t = \nabla \log \nu(X_t) dt + \sqrt{2} dW_t \;. \tag{$\star$} $$ The symmetry property can then be easily verified by checking that the infinitesimal generator of the Langevin diffusion is symmetric in an $L^2$-inner product weighted by $\nu$. This property of the Langevin diffusion is analogous to the detailed balance condition of a Markov chain. It is also the basis of standard MCMC methods based on Langevin diffusions. In fact, many random-walk based MCMC methods (like random walk Metropolis) with target probability density proportional to $\nu(x)$, can weakly approximate the corresponding Langevin diffusion ($\star$). This is remarkable because, e.g., the random walk Metropolis algorithm does not involve the gradient of $U$.

To read more about this connection, see the introduction and Theorem 5.2 of this paper.

added 506 characters in body
Source Link

The Langevin diffusion is reversible with respect to the density $\rho(x)=e^{-U(x)}$. This property is more transparent when the SDE is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is the basis of MCMC methods based on Langevin diffusions; see e.g.

Exponential convergence of Langevin diffusions and their discrete approximations

In fact, random-walk based MCMC methods (like random walk Metropolis) with target density $\rho(x)$, can be viewed as weak approximations to the corresponding Langevin diffusion; see e.g. Theorem 5.1 of

Metropolis integration schemes for self-adjoint diffusions

In particular, random walk Metropolis doesn’t even explicitly involve the gradient of $U$, but by virtue of being reversible, can weakly approximate the Langevin diffusion.

The Langevin diffusion is reversible with respect to the density $\rho(x)=e^{-U(x)}$. This property is more transparent when the SDE is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is the basis of MCMC methods based on Langevin diffusions; see e.g.

Exponential convergence of Langevin diffusions and their discrete approximations

The Langevin diffusion is reversible with respect to the density $\rho(x)=e^{-U(x)}$. This property is more transparent when the SDE is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is the basis of MCMC methods based on Langevin diffusions; see e.g.

Exponential convergence of Langevin diffusions and their discrete approximations

In fact, random-walk based MCMC methods (like random walk Metropolis) with target density $\rho(x)$, can be viewed as weak approximations to the corresponding Langevin diffusion; see e.g. Theorem 5.1 of

Metropolis integration schemes for self-adjoint diffusions

In particular, random walk Metropolis doesn’t even explicitly involve the gradient of $U$, but by virtue of being reversible, can weakly approximate the Langevin diffusion.

added 153 characters in body
Source Link

The Langevin diffusion is reversiblereversible with respect to the density $\rho(x)=e^{-U(x)}$. This property is more transparent when the SDE is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is the basis of MCMC methods based on Langevin diffusions; see e.g.

Exponential convergence of Langevin diffusions and their discrete approximations

The Langevin diffusion is reversible with respect to the density $\rho(x)=e^{-U(x)}$. This property is more transparent when the SDE is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is the basis of MCMC methods based on Langevin diffusions; see e.g.

Exponential convergence of Langevin diffusions and their discrete approximations

The Langevin diffusion is reversible with respect to the density $\rho(x)=e^{-U(x)}$. This property is more transparent when the SDE is written as $$ dX_t = \nabla \log \rho(X_t) dt + \sqrt{2} dW_t $$ and is the basis of MCMC methods based on Langevin diffusions; see e.g.

Exponential convergence of Langevin diffusions and their discrete approximations

Source Link
Loading