Lagrange Multiplier: What λ Actually Means, and Its Units

λ is not scaffolding: it is a price with units, and it resurfaces as a shadow price, a solver’s dual, and inverse temperature.

Somewhere in every multivariable calculus course there is a problem that produces two answers along with instructions to keep only one. A shape is to be optimized under a constraint, a Lagrange multiplier λ is introduced so that the algebra closes, and the student solves the system, reports the optimal dimensions, and discards λ without a second thought. It was scaffolding. The building is up.

Discarding it is the most expensive habit the method teaches. λ is not scaffolding, and it is not an artifact of the bookkeeping. It is worth more than the coordinates it was invented to help find, because it answers a different question: not where the optimum sits, but what the constraint is costing, and whether it would be worth paying to loosen it. The Lagrange multiplier is the rate at which the best achievable value of the objective changes when the constraint is relaxed by one unit: a price, quoted in whatever units the objective and the constraint happen to carry. What it is a price of is the harder question. The standard treatment computes that number and then looks away from it. Berkeley’s Dan Klein once wrote a…