Calling free_memory() on a solver that has not solved anything yet raises PyPardisoError with error code -1.
from pypardiso.scipy_aliases import pypardiso_solver
pypardiso_solver.free_memory()
# PyPardisoError: The Pardiso solver failed with error code -1.
If any solve happens first, the same call works:
import numpy as np, scipy.sparse as sp, pypardiso
from pypardiso.scipy_aliases import pypardiso_solver
pypardiso.spsolve(sp.csr_matrix(np.array([[2.0, 1.0], [1.0, 3.0]])), np.array([1.0, 2.0]))
pypardiso_solver.free_memory() # fine
Why this happens
free_memory() sets phase 0, which tells Pardiso to release the memory holding the L and U factors:
https://github.com/haasad/PyPardiso/blob/master/pypardiso/pardiso_wrapper.py#L335-L343
On a fresh solver there are no factors to release, so Pardiso gets a request it cannot fulfil and returns -1 ("input inconsistent"). Older MKL versions seem to have let this pass; this shows up with MKL 2026.1.0 from conda-forge.
Suggested fix
free_memory() reads as cleanup, and cleanup when there is nothing to clean up should do nothing rather than raise. Skipping the Pardiso call when no factorization exists would do it, for example:
def free_memory(self, everything=False):
self.remove_stored_factorization()
if self.factorized_A.shape[0] == 0 and not everything:
return
...(everything=True uses phase -1, which releases all internal memory and may still be worth calling on a fresh solver — I have not tested that case.)
Environment
|
|
pypardiso |
0.4.7 |
mkl |
2026.1.0 (conda-forge) |
numpy / scipy |
1.26.4 / 1.13.1 |
| Python |
3.11.15 |
| OS |
Windows 11 (10.0.26100), x64 |
Context
This came up in bw2calc, which calls free_memory() to clear solver state between Monte Carlo iterations. Some of our solver classes never call spsolve, so they hit this on the first iteration. Reported by @KarinTreyer in brightway-lca/brightway2-calc#157. We are also handling it on our side, so this is not urgent for us.
Calling
free_memory()on a solver that has not solved anything yet raisesPyPardisoErrorwith error code -1.If any solve happens first, the same call works:
Why this happens
free_memory()sets phase 0, which tells Pardiso to release the memory holding the L and U factors:https://github.com/haasad/PyPardiso/blob/master/pypardiso/pardiso_wrapper.py#L335-L343
On a fresh solver there are no factors to release, so Pardiso gets a request it cannot fulfil and returns -1 ("input inconsistent"). Older MKL versions seem to have let this pass; this shows up with MKL 2026.1.0 from conda-forge.
Suggested fix
free_memory()reads as cleanup, and cleanup when there is nothing to clean up should do nothing rather than raise. Skipping the Pardiso call when no factorization exists would do it, for example:(
everything=Trueuses phase -1, which releases all internal memory and may still be worth calling on a fresh solver — I have not tested that case.)Environment
pypardisomklnumpy/scipyContext
This came up in
bw2calc, which callsfree_memory()to clear solver state between Monte Carlo iterations. Some of our solver classes never callspsolve, so they hit this on the first iteration. Reported by @KarinTreyer in brightway-lca/brightway2-calc#157. We are also handling it on our side, so this is not urgent for us.