Abstract:We propose a general method to identify nonlinear Fokker--Planck--Kolmogorov equations (FPK equations) as gradient flows on the space of probability measures on $\mathbb{R}^d$ with a natural differential geometry. Our notion of gradient flow does not depend on any underlying metric structure such as the Wasserstein distance, but is derived from purely differential geometric principles. We explicitly identify the associated energy functions $E$ and show that these are Lyapunov functions for the FPK solutions. Moreover, we show restricted uniqueness results for such gradient flows, and we also prove that the gradient of $E$ is a gradient field on $\mathbb{R}^d$, which can be approximated by smooth gradient fields. These results cover classical and generalized porous media equations, where the latter have a generalized diffusivity function and a nonlinear transport-type first-order perturbation.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to identify the nonlinear Fokker - Planck - Kolmogorov equation (FPK equation) as a gradient flow on the probability measure space by proposing a general method. Specifically, the authors present a method that does not rely on any underlying metric structure (such as the Wasserstein distance), but starts from pure differential - geometric principles. They explicitly identify the energy function \(E\) associated with these FPK equations and prove that these energy functions are Lyapunov functions of the FPK solutions. In addition, they also prove the restricted uniqueness of this gradient flow and show that the gradient of the energy \(E\) is a gradient field that can be approximated by a smooth gradient field. These results cover the classical and generalized porous media equations, where the latter has a generalized diffusion function and a first - order perturbation of the nonlinear transport type.
### Main contributions of the paper
1. **Identification of the gradient flow structure**:
- The authors propose a method to identify the nonlinear FPK equations as gradient flows on the probability measure space.
- They define the energy function \(E\) associated with these equations and prove that these energy functions are Lyapunov functions, which helps to analyze the asymptotic behavior of the gradient flow solutions.
2. **Generalized porous media equations**:
- The authors extend their method to make it applicable to a wider range of nonlinear FPK equations, especially the generalized porous media equations (gPME), which have a generalized diffusion function and a first - order perturbation of the nonlinear transport type.
- They prove that under appropriate assumptions, the unique probability solution of these equations is the restricted unique solution of the gradient flow equation.
3. **Differential - geometric construction**:
- The authors introduce a natural differential - geometric structure for the probability measure space \(P\) and construct the tangent bundle from the base space \(M\) to \(P\) by the method of "lifting" geometry.
- This method is different from the gradient flow method on the Wasserstein space in the optimal transport theory, as it does not rely on any metric structure.
### Main results
- **Theorem 1**: For the classical porous media equation (PME), with arbitrary probability measure \(\mu_0\in P\) as the initial data, the unique probability solution \(\mu_t\) is the restricted unique solution of the gradient flow equation of the energy \(E(vdx)=\frac{1}{m - 1}\int_{\mathbb{R}^d}v(x)^m dx\).
- **Theorem 2**: For the generalized porous media equation (gPME), under appropriate assumptions, the unique probability solution \(\mu_t\) is the restricted unique solution of the gradient flow equation of the energy \(E(v(x)dx)=\int_{\mathbb{R}^d}\eta(v(x))dx+\int_{\mathbb{R}^d}\Phi(x)v(x)dx\), where \(\eta(r)=\int_0^r\int_1^s\frac{\beta'(w)}{w b(w)}dw ds\).
### Methodology
- **Definition of the gradient flow**: The authors define the gradient flow by considering the nonlinear function class \(F\) on the probability measure space \(P\). Specifically, they calculate the derivative of \(F(\mu_t)\) and compare it with the derivative of the gradient flow equation, thereby finding the energy function \(E\).
- **Differential - geometric construction**: The authors construct the tangent bundle \(T_\nu P\) from the base space \(M\) to the probability measure space \(P\) by the method of "lifting" geometry and define the weighted inner product \(\langle\cdot,\cdot\rangle_b\).
### Conclusion
This paper provides a new method to identify the nonlinear FPK equations as gradient flows on the probability measure space. This method does not rely on any metric structure but is based on differential - geometric principles. This method is applicable not only to the classical porous media equation but also to a wider range of nonlinear FPK equations, providing a new tool for the analysis of the asymptotic behavior of these equations.