A set of necessary conditions for optimality in control problems that couples the state trajectory and adjoint (costate) variables through a Hamiltonian function; optimal controls minimise the Hamiltonian almost everywhere and satisfy costate (adjoint) differential equations with transversality conditions.

Principle

The organising idea is to convert a dynamic optimisation problem into pointwise stationarity/minimisation conditions on a Hamiltonian built from the instantaneous cost, dynamics, and adjoint multipliers so that candidate optimal trajectories satisfy a boundary-value Hamiltonian system.

Demonstration

In the minimum‑time control of a double integrator (bounded acceleration), Pontryagin's conditions produce bang‑bang controls: the Hamiltonian is minimised by choosing the control at its bounds and the costate dynamics determine switching times that yield a time‑optimal trajectory.

Misapplication

Using the principle as a sufficient test for optimality without verifying convexity, regularity, or second‑order conditions, or applying the classical statement blindly to problems with state constraints or non-smooth dynamics can lead to incorrect acceptance of non-optimal candidates.

Consequence

Applied correctly, PMP yields a boundary-value problem whose solutions are strong candidates for optima and exposes structure (e.g., bang‑bang or singular arcs) that guides numerical shooting, homotopy, or indirect methods to compute optimal controls.

Reversal

Reversing the minimisation sign (maximising the Hamiltonian) corresponds to problems formulated as maximisation instead of minimisation; failing to enforce the minimisation leads to trajectories that violate optimality conditions and typically increase the cost.

Boundary

Provides necessary conditions under mild regularity assumptions for a broad class of deterministic optimal control problems; it does not by itself guarantee global optimality, and extensions or modifications are required for state constraints, nonsmooth costs, stochastic controls, or infinite-dimensional systems.

Semantic Tension

There is a persistent tension between Pontryagin's variational/adjoint viewpoint and the dynamic‑programming/Hamilton–Jacobi–Bellman approach: PMP gives pointwise necessary conditions and structural insight, whereas HJB offers sufficient conditions and value function characterization but can be harder to solve.

Synthesis

The Pontryagin Minimum Principle transforms an optimal control problem into a coupled state–costate Hamiltonian system plus pointwise minimisation of the Hamiltonian; solving that boundary‑value system yields candidate optimal trajectories whose validity must be checked against regularity and sufficiency conditions.