Same input but slightly different solution

We recently noticed that with the same input (A, P, and etc), the solver can return slightly different solutions (the values’ differences are usually less than 1e-2). Is this normal? If so, is there anything we can setup to avoid it?


What is the tolerance you are solving to? Do the solutions become more similar as you make the tolerance smaller (e.g. 1e-6 instead of the default 1e-3)?

If not, is there an example you can provide that shows this?

The solver does not have completely deterministic behaviour because the interval used to update the value \rho is based on the observed factorisation and iteration time. This means that rho might update at a slightly different point (and hence to a slightly different value) between successive solves with the same data. To get completely deterministic behaviour, set the rho update interval to a fixed value.