Sensitivity Analysis of User Equilibrium Flows Revisited

This paper addresses certain misconceptions regarding what is known and what may be expected when performing sensitivity analyses of network user equilibrium flow patterns. Our presentation relies on a simple observation: any given user equilibrium sensitivity analysis technique should be employed only when the regularity conditions on which it is based are satisfied. Violating regularity, as we show through previously published numerical examples, as well as new examples presented here for the first time, may well lead to incorrect results when the Tobin-Friesz sensitivity analysis method is applied. This is especially so when the most critical regularity assumption of the Tobin-Friesz method, namely that the unperturbed solution must be a nondegenerate extreme point, is violated. We also illustrate how a degenerate unperturbed solution may sometimes be modified to obtain an appropriate nondegenerate solution, thereby allowing the Tobin-Friesz method to be applied.