An operation that maps a (convex, proper, lower semicontinuous) function f(x) of a variable x to its convex conjugate f*(p) = sup_x (p·x − f(x)), exchanging variable and slope and producing a dual representation.

Principle

Realize a duality between 'coordinates' and 'conjugate momenta' (or slopes) by taking a supremal linearization; the Legendre transform converts variational descriptions and often turns constrained optimization into an unconstrained dual problem.

Demonstration

For f(x)=½ax^2 (a>0), the Legendre transform yields f*(p)=½ p^2 / a; in mechanics, the Legendre transform of a regular Lagrangian with respect to velocities produces the Hamiltonian in momentum coordinates.

Misapplication

Applying the classical Legendre transform to nonconvex functions without accounting for multiple local suprema; ignoring lower semicontinuity can lead to transforms that correspond to convex envelopes rather than an involutive inverse.

Consequence

Provides a bridge between primal and dual formulations: solves optimization by passing to the convex conjugate, yields thermodynamic potentials from free energies, and produces Hamiltonian from Lagrangian under regularity.

Reversal

The inverse Legendre transform recovers the original function when f is proper, convex and lower semicontinuous; for nonconvex f the inverse may produce the convex hull rather than the original function.

Boundary

Standard Legendre transform requires convexity and appropriate growth (properness, lsc) for involutivity; for nondifferentiable points replace gradient by subgradient and for nonconvex settings use generalized convexification.

Semantic Tension

Confused with Fourier or Laplace transforms; Legendre transform is a convex duality via supremum of linear functionals, not an integral transform — it exchanges geometrical slope information rather than frequency content.

Synthesis

The Legendre transform assigns to a function its convex conjugate by supremizing linear forms p·x−f(x), effecting a coordinate‑to‑slope duality that underpins dual optimization, thermodynamic potentials and passage from Lagrangian to Hamiltonian formalisms.