The paper examines a class of optimal control problems where you want to maximize some objective while satisfying a constraint that involves an integral over time, called an isoperimetric constraint. Think of it like optimizing a rocket's trajectory while requiring that the total fuel burned equals a fixed amount. The authors approach this using tools from convex analysis, treating the controls as functions living in abstract mathematical spaces. For problems with linear dynamics and concave objectives, they prove that the optimal value changes smoothly and predictably as you adjust the constraint level, and they give a precise formula for how sensitive that value is to changes in the constraint.
The key payoff is connecting two different objects that both act as "prices" for the constraint. One comes from convex duality, a general mathematical technique, and the other is the familiar Lagrange multiplier that appears in the Pontryagin maximum principle, which is the standard tool engineers and mathematicians use to solve optimal control problems. The paper shows these two multipliers agree under mild conditions, validating the sensitivity formula in practical terms. The authors also handle a trickier case involving quadratic objectives and quadratic equality constraints, where they show the sensitivity result can still hold even when certain standard solution methods break down, meaning the formula is more robust than previously appreciated.