Dannberg et al. [2019] Ridge melt migration model failing in SolverCG

ASPECT
1

Hello,

ASPECT 3.1.0-pre is not solving timestep 0 of the Dannberg et al. [2019], mid-ocean ridge melting model. It produces solutions, but when it gets down to the last refinement level the BlockSchurPreconditioner fails. I’ve tried relaxing the Linear solver tolerance to 1e-6, and reducing the global refinement from 6 to 5, but it ends up crashing in the same solver block (in timestep 1 or 2) . I’ve attached my input file (adapted from “mid_ocean_ridge_adaptive_porosity.prm”), std out, and error output. There was no solver_history.txt file produced. A few of the original input parameters have changed or gone away (noted by “G.Ito”) so I wonder if those contribute to the issues.

Thanks in advance,

Garrett

input_Lst1e-6.prm (11.3 KB)

log_Lst1e-6.out.txt (9.4 KB)

log_Lst1e-6.err.txt (20.2 KB)

2

Ah! The input file was trying to solve the problem with 6 and 6 or 5 and 6 Initial global and adaptive refinement levels. This looks to be beyond the limits of resolution shown in the paper, and hence what set the limit on the number of rows in Table 4. It solves fine with the resolution I surmise the problem needs in practice. Note to self: failure in the inner CGSolve is a symptom of too many dofs for the problem. Comments welcome, though.