Interior Point Method (Barrier Method)
Solves LP by following central path through interior of feasible region. Main variant is Mehrotra predictor-corrector:
- Affine scaling direction: solve Newton system for complementarity
- Centering direction: correct toward central path with adaptive μ
- Combined step: predictor + corrector with step length limiting
- Update (x, y, s) and reduce barrier parameter μ
Per iteration solves normal equations: (ADA^T)Δy = rhs where D is diagonal scaling matrix from current iterate. Cholesky factorization of ADA^T is computational bottleneck.
Converges in O(√n log(1/ε)) iterations (polynomial complexity). Each iteration O(m²n + m³) for forming/factoring normal equations.
Mathematical Formulation
Barrier problem: min c^Tx - μ∑log(x_j-l_j) - μ∑log(u_j-x_j) KKT conditions: Ax=b, A^Ty+s=c, XSe=μe (complementarity) Normal equations: (ADA^T)Δy = A(D*r_c - X^{-1}r_xs) + r_b where D = X/S diagonal, r_b, r_c, r_xs are residuals.
Complexity
O(√n log(1/ε)) iterations for ε-optimality. Per iteration: O(nnz(A)m) to form ADA^T, O(m³) for dense Cholesky, or O(nnz(L)²) for sparse Cholesky. Total typically O(n^{1.5}) to O(n³).
Implementations
Clp
- ClpInterior.hpp - Interior point (barrier) method for LP
References
- Mehrotra, "On the implementation of a primal-dual interior point method", SIAM Journal on Optimization 2 (1992) 575-601
- Andersen & Andersen, "The MOSEK interior point optimizer for LP", High Performance Optimization, Kluwer (2000)
Implements interior point methods as an alternative to simplex for LP. Interior point follows a central path through the interior of the feasible region, converging to optimality. Often faster than simplex for very large or dense problems.
Two main variants supported:
- Predictor-Corrector: Standard Mehrotra-style algorithm
- PDCO: Primal-Dual Barrier method with regularization
Requires Cholesky factorization for the normal equations ADA'. Multiple Cholesky backends available (dense, MUMPS, TAUCS, etc.)