SPECT 3.1.0-pre: Nonlinear Solver Non-convergence and Abnormal Process Termination When Running .prm

1

Problem Description

I encountered nonlinear solver non-convergence and abnormal process termination issues while running an ASPECT-related model. The detailed information is as follows:

Running Environment

  • ASPECT Version: 3.1.0-pre (main, eff1d6fbf)

  • Dependencies Versions: deal.II 9.7.0 (32-bit indices, vectorization level 1), Trilinos 14.4.0, p4est 2.3.6, Geodynamic World Builder 1.0.0

  • Operating System: Ubuntu (inferred from the code execution environment, with reference to relevant compatibility discussion scenarios in the community)

  • Running Mode: OPTIMIZED mode

  • Number of MPI Processes: 2

Running Command and Related Configuration

  • Command: mpirun -np 2 aspect-release prms/009_model_50_y.prm

  • Plugin: Loaded the shared library /home/ubuntu/aspect/cookbooks/free_surface_with_crust/plugin_zenodo/libriftplugin_d10.release.so

  • Model Configuration: Adopted the configuration with free surface and crust (under the path cookbooks/free_surface_with_crust/plugin_zenodo)

Error Phenomena

  1. The program initially created the output directory output/, then loaded the mesh and degree-of-freedom information normally, and gradually increased the mesh refinement level (from level 1 to level 3):

    • Level 1: 3,125 active cells, 261,555 degrees of freedom

    • Level 2: 25,000 active cells, 1,956,600 degrees of freedom

    • Level 3: 200,000 active cells, 15,122,190 degrees of freedom

  2. Nonlinear solver non-convergence:

    • At timestep 0 (t=0 years, dt=0 years), after solving the Stokes system, the warning “WARNING: The nonlinear solver in the current timestep failed to converge” appeared twice. The program continued to the next timestep according to the parameter settings, but the solution did not fully converge.

    • Key residual data:

      • First Stokes initial residual: 5.25909e+18 (v=5.24215e+18, p=4.21687e+17), relative residual was 1 after 1 nonlinear iteration

      • Second Stokes initial residual: 8.15629e+18 (v=8.1558e+18, p=8.91025e+16), relative residual was 0.999982 after 1 nonlinear iteration

  3. Abnormal process termination:

    • After the mesh was refined to level 3 and the mesh displacement system was solved, the program terminated with the prompt “Primary job terminated normally, but 1 process returned a non-zero exit code”, and process rank 1 exited due to signal 15 (Terminated).

Questions and Help Requests

  1. What are the possible core causes of the nonlinear solver non-convergence? Is it related to the model initial parameter settings, plugin compatibility, or dependency library version matching?

  2. Is the process termination due to signal 15 related to the solver non-convergence? How to troubleshoot and resolve this termination issue?

  3. For the current ASPECT version (3.1.0-pre) and model configuration, are there any suggestions for adjusting nonlinear solver parameters (such as iteration count, convergence threshold, etc.) or other configurations to solve this problem?