Modified conjugate gradient method to minimise quadratic functional restricted to positive solutions

152 Views Asked by At

Knowing that CG is an efficient method of solving sparse systems of linear equations, I have used it to solve a linear system of equations generated by a finite element algorithm.

In order to impose an additional boundary condition, I have to ensure that the components of the solution are greater than or equal to zero, resulting in a slightly modified version of the minimisation problem

$ x_{*} = \arg \min_{x \in \mathbb{R}^n} \varphi(x), \quad \varphi(x) = \frac{1}{2}x^TAx - b^T x, $

specifically

$ x_{*} = \arg \min_{x \in \mathbb{R}_{+}^n} \varphi(x), \quad \varphi(x) = \frac{1}{2}x^TAx - b^T x. $

How would I go about modifying the conjugate gradient algorithm to include this restriction?

1

There are 1 best solutions below

0
On BEST ANSWER

See this paper for a generalization of CG type methods to non-negative constrained setting: https://www.hindawi.com/journals/jam/2013/986317/