Approximation convergence in the inverse first-passage time problem

TL;DR

This paper introduces a new approximation method for the inverse first-passage time problem of Wiener processes, which converges uniformly and accounts for the boundary's starting value, enhancing practical applicability.

Contribution

It proposes a modified approximation that estimates both the boundary and its starting value, with proven uniform convergence under certain smoothness conditions.

Findings

The approximation is well-defined for absolutely continuous boundaries.

A subsequence of the approximation converges uniformly as interval length decreases.

Results extend to reflected Wiener processes.

Abstract

The inverse first-passage time problem determines a boundary such that the first-passage time of a Wiener process to this boundary has a given distribution. An approximation which is based on the starting value of the boundary to a smooth boundary by a piecewise linear boundary is given by equating the probability of the first-passage time to a linear boundary and the increment of the distribution on each interval. We propose a modification of that approximation which also approximates the starting value of the boundary. First, we show that the approximation is well-defined when assuming that the boundary is absolutely continuous. Second, we show that a subsequence of this new approximation uniformly converges to the boundary when the length of each interval of linear approximation goes to 0 asymptotically. The results are obtained using Arzela-Ascoli theorem on any compact space on which we further assume that the boundary admits uniformly dominated derivative. As the starting value of the boundary is unknown, this makes the new approximation more suitable for applications. The results are also proved in the first-passage time problem of a reflected Wiener process.