Result changes depending on whether matrix is sparse or not

[John_Gibson]

Hmmm, my suggestion to use pinv is not a practical one. It’s never a good idea to compute an inverse to solve an Ax=b problem.

Is there any special reason in your application why one solution for x is better than another?

Regarding pseudo-inverse and QR, this thread is relevant.

Moore-Penrose Generalized Inverse of Sparse Matrix

You almost never compute the inverse (pseudo or otherwise) of a sparse matrix because the inverse is generally dense. On the other hand, you do compute the application of the inverse to a vector, and you often precompute factorizations that let you apply the inverse more quickly. In the case of the ordinary inverse, you can apply a sparse inverse with A \ b and compute the factorization with lu(A) etcetera. In the case of the pseudoinverse, applying it computes the least-squares solution, and …