Variational calculation of Laplace transforms via entropy on Wiener space and applications (Q457613): Difference between revisions

Consider the standard Wiener space \((W,H,\mu)\) for \(W=C([0,1],R^d)\), \(H\) the Cameron-Martin space, and \(\mu\) the Wiener measure. Heuristically, for a functional \(f\), there is equivalence between the minimisation problems \[ \inf_u E_\mu\Bigl[f\circ(I_W+u)+\frac{1}{2}|u|_H^2\Bigr] \] for adapted \(u\) in \(H\), and \[ \inf_\gamma\Bigl(\int f\,d\gamma+H(\gamma|\mu)\Bigr) \] for probability measures \(\gamma\) on \(W\), where \(H(\cdot|\cdot)\) denotes the relative entropy. The question is to know whether this heuristic equivalence holds true. It is proved (under adequate assumptions) that the first infimum is not smaller than the second one, and \(f\) is said to be a tamed functional if the two infima are equal. The class of tamed functionals is characterised, and for these functionals, it is proved that the existence of a process \(u_0\) realizing the infimum is equivalent to the almost sure invertibility of \(U_0=I_W+u_0\); this invertibility can be interpreted in terms of the existence of a strong solution to some stochastic differential equation.