Name the cell when the IMPES storage matrix is singular

[hnil]

A cell whose storage-derivative matrix is singular makes getTrueImpesWeights() throw, and since solver setup is wrapped in OPM_CATCH_AND_RETHROW_AS_CRITICAL_ERROR the run dies pointing at the linear-solver JSON, which is not the problem:

Original error: getTrueImpesWeights() failed: FMatrixError [luDecomposition]: matrix is singular
Error hint: This is likely due to a faulty linear solver JSON specification.

Now:

Original error: getTrueImpesWeights() failed: Singular storage matrix when forming the CPR
pressure weights for cell 688 (Cartesian index 688).  The storage derivatives of that cell
are rank deficient, so the pressure equation cannot be formed there.

Diagnostic only — it still fails, but says where. Reproduced on spe1_precsalt/GASCONDENSATE_VAPWAT_PRECSALT; why that cell is rank deficient is still open.

An earlier version substituted bweights = rhs instead of throwing. That was wrong: bweights is indexed by equation and rhs by primary variable, so it put unit weight on whichever equation shares the pressure variable index — and that fed a singular pressure system into the CPR setup, which is exactly the follow-on failure the first version ran into.

Separately: the "faulty linear solver JSON specification" hint is misleading for every non-configuration failure caught by that macro — worth revisiting, not touched here.

[atgeirr]

I do not understand this sentence, what is meant by "Weight the pressure equation alone"? And it seems this would not throw and thereby chop the step, also what does that have to do with degenerate cells?

[hnil]

Fair — two of the three claims in that comment were wrong. Catching the exception means nothing throws, so the step is not chopped, and the "degenerate cell" framing was hand-waving.

Cut it back to what the code does: on a singular storage matrix, fall back to rhs, which is the unit vector at pressureVarIndex — the unweighted pressure equation. Warning text matches now.

I do not have a real explanation for why the storage matrix is singular on that deck, so I have not tried to write one.

[hnil]

Reworked: it now throws with the cell identified instead of substituting a weight.

The fallback was wrong, not just badly commented — bweights is indexed by equation and rhs by primary variable, so bweights = rhs put unit weight on whichever equation happens to share the pressure variable's index. That is what made the pressure system singular further down, which is the follow-on failure the first version hit.

So this is diagnostic only now: same failure, but it names the cell rather than blaming the solver JSON.