Abstract:To date, most methods for simulating conditioned diffusions are limited to the Euclidean setting. The conditioned process can be constructed using a change of measure known as Doob's $h$-transform. The specific type of conditioning depends on a function $h$ which is typically unknown in closed form. To resolve this, we extend the notion of guided processes to a manifold $M$, where one replaces $h$ by a function based on the heat kernel on $M$. We consider the case of a Brownian motion with drift, constructed using the frame bundle of $M$, conditioned to hit a point $x_T$ at time $T$. We prove equivalence of the laws of the conditioned process and the guided process with a tractable Radon-Nikodym derivative. Subsequently, we show how one can obtain guided processes on any manifold $N$ that is diffeomorphic to $M$ without assuming knowledge of the heat kernel on $N$.
What problem does this paper attempt to address?
### Problems the paper attempts to solve
This paper aims to solve the problem of simulating conditional diffusion processes on manifolds. Specifically, most existing methods are limited to conditional diffusion processes in Euclidean spaces, and this paper extends these methods to make them applicable to conditional diffusion processes on manifolds.
### Main contributions
1. **Extension of simulation methods for conditional diffusion processes**:
- By introducing a guiding function \( g \) based on the manifold heat kernel, the author extends the guiding process method in Euclidean space to the manifold \( M \).
- This method allows the construction of a guiding process through a comparison manifold \( M \) that is diffeomorphic to \( M \), without assuming the heat kernel of the target manifold \( N \).
2. **Avoidance of numerical and analytical problems**:
- By using a smooth guiding term, the discontinuities at the secant of the target point are avoided, thus simplifying numerical calculations and analysis.
- The need to verify complex Lyapunov conditions is avoided, making the method more practical.
3. **Theoretical results**:
- The equivalence between the conditional process and the guiding process is proved, and an explicit expression for the Radon - Nikodym derivative between them is given.
- A Feynman - Kac - type formula for representing the untraceable function \( h \) is provided.
4. **Numerical simulation and parameter estimation**:
- The effectiveness of the method is demonstrated through numerical simulation, especially in the simulation of diffusion bridges on tori embedded in three - dimensional space.
- The Metropolis - Hastings algorithm is applied for parameter estimation, demonstrating the method of estimating the parameters of the diffusion process under discrete - time observations.
### Mathematical formulas
- **Heat kernel equation**:
\[
\frac{\partial \kappa(t, x, y)}{\partial t}=\frac{1}{2} \Delta_x \kappa(t, x, y)
\]
where \(\Delta_x\) is the Laplace - Beltrami operator on the manifold \( M \).
- **Guiding function**:
\[
g(t, x)=\kappa(T - t, x, x_T)
\]
- **Radon - Nikodym derivative**:
\[
\frac{dP_h^T}{dP_g^T}(X)=\frac{g(0, x_0)}{h(0, x_0)} \Psi_{T,g}(X)
\]
where
\[
\Psi_{T,g}(X)=\exp\left(\int_0^T V(\log g)(s, X_s) \, ds\right)
\]
### Conclusion
By introducing a guiding function based on the manifold heat kernel, this paper successfully extends the simulation method of conditional diffusion processes, making it applicable to a wider range of manifolds. This method is not only of great significance in theory, but also demonstrates its effectiveness and practicality in practical applications.