perf: Improve MGR recipes for thermal

[victorapm]

Depends on GEOS-DEV/thirdPartyLibs#332

[castelletto1]

What does "Partial column-lumped approximation" mean?

[victorapm]

It means W_{TS} = 0 instead of doing colsum^{-1}(A_{TS}) * colsum^{-1}(A_{SS}) as done by blockColLumped. W_{TS} being the submatrix of the restriction operator coupling temperature and densities.

The pressure part remains the same: W_{PS} = colsum^{-1}(A_{PS}) * colsum^{-1}(A_{SS})

So, I called it a "partial" column-lumped. Let me know if you have other names in mind!

[castelletto1]

So for ThermalCompositionalMultiphaseFVM we have:

| A_SS A_SP A_ST |
| A_PS A_PP A_PT |
| A_TS A_TP A_TT |

With "Partial column-lumped approximation", the correction to block:

| A_PP A_PT |
| A_TP A_TT |

is

- | colsum(A_{PS})  | * colsum^{-1}(A_{SS}) * | A_SP A_ST |
  |       0         |

Correct?

[victorapm]

That's correct, so essentially the energy equation isn't perturbed

[castelletto1]

I like the use of the Galerkin projection instead of the non-Galerkin. Again, I am not sure I fully understand what partial column lumping is.

[castelletto1]

Same comment applies also to the the other two modified strategies

[paveltomin]

would be interesting to discuss this during our solvers meeting on Thursdays