Convex Optimization — Duality and KKT Conditions
Takeaway
Convex problems have no bad local minima; duality and KKT conditions provide certificates of optimality and efficient algorithms.
The problem (before → after)
- Before: Nonconvex landscapes trap local methods.
- After: Convex structure guarantees global optima and strong theoretical tools.
Mental model first
Think of a bowl-shaped valley; any path downhill leads to the bottom. Constraints are fences; the optimal point touches fences where forces balance.
Just-in-time concepts
- Convex sets/functions, epigraphs.
- Lagrangian L(x, λ, ν) and dual function g(λ, ν).
- KKT conditions: stationarity, primal/dual feasibility, complementary slackness.
First-pass solution
Formulate as minimize f(x) s.t. g_i(x) ≤ 0, h_j(x) = 0. Write L; derive KKT; solve via interior-point or first-order methods.
Iterative refinement
- Strong duality via Slater’s condition.
- Proximal methods and ADMM for large-scale problems.
- Conic formulations unify many problems (LP, QP, SDP).
Principles, not prescriptions
- Choose convex relaxations when exact problems are hard.
- Dual variables interpret constraints as prices.
Common pitfalls
- Violating convexity assumptions; KKT become necessary but not sufficient.
- Poor scaling without preconditioning.
Connections and contrasts
- See also: [/blog/interior-point-methods], [/blog/branch-and-bound], [/blog/optimal-transport].
Quick checks
- Why convex? — Guarantees global optimality and tractable algorithms.
- What is complementary slackness? — Active constraints have positive multipliers and zero slack.
- When strong duality? — Slater’s condition holds.
Further reading
- Boyd & Vandenberghe (book above)