The common wisdom is that Ay is more accurate than inv(A)*y and I believe that to be true, but when is it also faster? Say the matrix A is well-conditioned so I don’t really care about the ability of to find a least-squares solution.
Am I doing something misleading here? The takes much longer.
A = randn(20);
A = A*A’;
s = randn(20,1);
timeit(@() inv(A)*s)
timeit(@() A s)
From the documentation for mldivide, it sounds like it should be using the Cholesky solver since A is Hermitian, why is that not faster than an inv and matrix multiplication?
ishermitian(A)The common wisdom is that Ay is more accurate than inv(A)*y and I believe that to be true, but when is it also faster? Say the matrix A is well-conditioned so I don’t really care about the ability of to find a least-squares solution.
Am I doing something misleading here? The takes much longer.
A = randn(20);
A = A*A’;
s = randn(20,1);
timeit(@() inv(A)*s)
timeit(@() A s)
From the documentation for mldivide, it sounds like it should be using the Cholesky solver since A is Hermitian, why is that not faster than an inv and matrix multiplication?
ishermitian(A) The common wisdom is that Ay is more accurate than inv(A)*y and I believe that to be true, but when is it also faster? Say the matrix A is well-conditioned so I don’t really care about the ability of to find a least-squares solution.
Am I doing something misleading here? The takes much longer.
A = randn(20);
A = A*A’;
s = randn(20,1);
timeit(@() inv(A)*s)
timeit(@() A s)
From the documentation for mldivide, it sounds like it should be using the Cholesky solver since A is Hermitian, why is that not faster than an inv and matrix multiplication?
ishermitian(A) mldivide, inv, slow MATLAB Answers — New Questions

​