Has anybody seen a version of the saddle point method for a real-value multivariate function with multiple local minima ?
For a real-value function with a single variable which has a unique minimum the approximation result is usually presented as follow.
We want to approximate a integral of the form $I_\alpha = \int_\mathbb{R}e^{-f_\alpha (x)dx}$ where $f_\alpha$ is twice differentiable. For every $f_\alpha$, $f_\alpha$ has a unique minimum reached at $x_\alpha$. As $\alpha$ goes to $\infty$, $f_\alpha$ is more localised around $x_\alpha$: $$ \frac{d^2 f}{dx^2}(x_\alpha) \xrightarrow[\alpha \rightarrow \infty]{} + \infty $$ And we use the approximation: $$ I_\alpha \approx e^{-m_\alpha} \sqrt{\frac{2\pi}{\frac{d^2 f}{dx^2}(x_\alpha)}} $$
Thanks