R14.1.1 MPP full-deck restart (n=d3full01): corrupted thermal state

[Amrith Mariappan]

Hi all,

 

I am running a repeated-impact simulation chain and I cannot get the

full-deck restart to work in R14.1.1. I would appreciate any insight into

whether thermal-state restore on full restart is expected to work for a

coupled S-ALE multi-material model with a phase-change thermal material,

and whether the boundary-condition crash described below is a known issue.

 

SOLVER / ENVIRONMENT

- ANSYS 2025 R2 (v252): lsdyna_mpp_dp_msmpi.exe and lsdyna_mpp_dp_impi.exe

- Both report: LS-DYNA MPP double precision, Revision R14.1.1-16-gfadfb49b17

  (MS MPI V10.0 and Intel MPI builds respectively)

- Windows 10/11 x64, 16 MPI processes, double precision

- Both the initial run and the restart use the same revision and the same

  number of processes, so the d3full version/rank-count requirements are met.

 

MODEL (initial run, "Impact 01")

- 2D axisymmetric S-ALE, coupled structural-thermal:

    *SECTION_ALE2D (ALEFORM=11, ELFORM=14)

    *ALE_STRUCTURED_MESH + *ALE_STRUCTURED_MESH_CONTROL_POINTS

    *ALE_STRUCTURED_MESH_VOLUME_FILLING (ALL / BOXCOR / ELLIPSOID with

     velocity vector via VID)

    *ALE_MULTI-MATERIAL_GROUP (3 AMMGs)

- Materials:

    *MAT_JOHNSON_COOK (MAT_015) + *EOS_GRUNEISEN for the HSLA-100 target

    *MAT_ELASTIC_PLASTIC_HYDRO (MAT_010) + *EOS_GRUNEISEN for the WC

     projectile (2220 m/s)

    *MAT_ALE_VACUUM for the surrounding void

- Thermal:

    *MAT_THERMAL_ISOTROPIC_PHASE_CHANGE for the target (T09)

    *MAT_THERMAL_ISOTROPIC for projectile and vacuum

    *CONTROL_SOLUTION SOLN=2 (coupled)

    *CONTROL_THERMAL_SOLVER SOLVER=12, EQHEAT=1, FWORK=0.9

    *CONTROL_THERMAL_TIMESTEP / *CONTROL_THERMAL_NONLINEAR

    *INITIAL_TEMPERATURE_SET (NSID=0, 296.15 K)

- Boundary conditions via *BOUNDARY_SALE_MESH_FACE: SYM on the axis face

  (negx), NONREFL on the posx/negy/posy faces.

- *DATABASE_BINARY_D3DUMP writes d3dump01/d3full01 at t = 5.3e-5 s.

- *CONTROL_MPP_DECOMPOSITION_DISTRIBUTE_ALE_ELEMENTS is used.

 

The initial run terminates NORMALLY at t = 5.3e-5 s (4845 cycles) with a

physically sound thermal state (final temperature norm ~1.0e5, nonlinear

thermal residual norms ~1e4-1e6). d3dump01 and d3full01.0000-0015 are

written as expected.

 

RESTART RUN ("Impact 02")

Full-deck restart deck = full model + *STRESS_INITIALIZATION (PIDO=10,

PIDN=0, the S-ALE mesh part) + a new *ALE_STRUCTURED_MESH_VOLUME_FILLING

(ELLIPSOID, with VID) introducing a second projectile. Executed with:

    mpiexec -np 16 lsdyna_mpp_dp_xxxmpi.exe i=restart.k n=d3full01 memory=200m

The restart reads the dump correctly (problem time 5.3000e-5, cycle 4845,

"material 10 in restart file initializes"), but then fails at the very

first cycle in TWO distinct ways:

 

FAILURE 1 - thermal solver divergence, Error 40359 (SOL+359):

    "the minimum time step has been reached and the maximum temperature

     change is greater than specified"

    solution time 5.3000e-5, node 6234 (inside the previously impacted,

    hot region of the target), temperature change 5.2976e+12.

  The first thermal solve after restart has an initial residual norm of

  7.4432e+22, i.e. 18 orders of magnitude larger than any thermal solve in

  the healthy initial run (~1e4-1e6). During the Newton iterations the

  trial temperatures drop to ~112 K in material that ended the initial run

  hot. The residual is bit-identical whether or not the restart deck

  contains *INITIAL_TEMPERATURE_SET, so the initial-temperature card is not

  the cause - the thermal state after the restore appears inconsistent/

  corrupted regardless of the deck.

 

FAILURE 2 - access violation in the 2D ALE non-reflecting boundary:

  When the NONREFL *BOUNDARY_SALE_MESH_FACE card is active, the run instead

  crashes at the first mechanical cycle with

    forrtl: severe (157): Program Exception - access violation

    NRBCALE2D (dyn60.F, line 36922) <- APPLY_LOADS <- FEM3D <- SOLTN

  With the NONREFL card commented out, this crash disappears (the run then

  reaches Failure 1 instead). The identical NONREFL setup runs without any

  problem in the initial (non-restart) run.

 

QUESTIONS

1. On a full-deck restart with *STRESS_INITIALIZATION + n=d3full01, is the

   nodal temperature field (and phase-change history) of a coupled

   structural/thermal S-ALE model supposed to be restored from the dump?

   Any known limitations for MAT_THERMAL_ISOTROPIC_PHASE_CHANGE on restart?

2. Is the NRBCALE2D access violation on the restart path a known issue for

   2D S-ALE non-reflecting boundaries? (I noticed the R14.1.0 release note

   about an incorrect shear modulus in 2D S-ALE non-reflecting boundary

   elements, and the R14.0.0 fix to *BOUNDARY_NON_REFLECTING_2D for

   discontinuous boundary lines.)

3. Is there a documented workaround other than abandoning the restart and

   running both impacts as one continuous simulation (which is my current

   fallback)?

 

For context: under R16.1.1 the same model family crashes at cycle 0 with an

access violation inside *MAT_THERMAL_ISOTROPIC_PHASE_CHANGE (already

reported separately), which is why this chain was migrated to R14.1.1.

 

I can provide the keyword decks, d3hsp/mes files, and (if useful) the

d3full01 family as a reproducer.

 

Thanks in advance,

Amrith