A Log-Barrier Newton-CG Method for Bound Constrained Optimization with Complexity Guarantees
We describe an algorithm based on a logarithmic barrier function, Newton’s method, and linear conjugate gradients, that obtains an approximate minimizer of a smooth function over the nonnegative orthant. We develop a bound on the complexity of the approach, stated in terms of the required accuracy and the cost of a single gradient evaluation of … Read more