Vaishnavi_198
New member
Title: Reproducible SIGSEGV in Thompson microphysics at the same model time in WRF-Chem v4.6.1
Hello WRF community,
I am running regional future climate-air-quality simulations over India using WRF-Chem v4.6.1. The simulations dynamically downscale bias-corrected CESM2 SSP3-7.0 meteorology using two nested domains. The present case runs from 13 February to 31 May 2049 and is part of a multi-year experiment using the same WRF configuration and emissions for the surrounding years.
chem_opt = 201 (MOZCART)
mp_physics = 8, 8 (Thompson)
time_step = 180 s
parent_time_step_ratio = 1, 5
nproc_x = 8
nproc_y = 7
MPI ranks = 56
OMP_NUM_THREADS = 1
The 2049 simulation fails on 1 March at approximately 11:57 UTC. I have run it three times, including a fresh run after correcting the emission settings, and it terminates at exactly the same model time each time. Equivalent simulations for 2048, 2050, 2051, and 2052 use the same executable, model configuration, emissions, chemistry, and MEGAN inputs (with different corresponding dates) and run successfully.
forrtl: severe (174): SIGSEGV, segmentation fault occurred
The traceback points to:
module_mp_thompson_mp_mp_thompson_
module_mp_thompson_mp_mp_gt_driver_
module_microphysics_driver_mp_microphysics_driver_
SLURM subsequently reports processes being killed by signal 9, but this appears to happen after rank 37 encountered the segmentation fault.
Both domains reach approximately 11:57 UTC in the logs before the failure. The traceback does not identify the domain so I cannot yet determine whether the failure originates in d01 or d02.
The failure occurs at the same model time in three independent runs (stop exactly at the same time stamp even after two restarts), while the surrounding years complete successfully, I am unable to find out what is causing this segmentation fault and eventually the termination.
I have attached namelist.input, the relevant rank-37 rsl.error and rsl.out
Thank you for your help.
Hello WRF community,
I am running regional future climate-air-quality simulations over India using WRF-Chem v4.6.1. The simulations dynamically downscale bias-corrected CESM2 SSP3-7.0 meteorology using two nested domains. The present case runs from 13 February to 31 May 2049 and is part of a multi-year experiment using the same WRF configuration and emissions for the surrounding years.
configuration
WRF-Chem version = 4.6.1chem_opt = 201 (MOZCART)
mp_physics = 8, 8 (Thompson)
time_step = 180 s
parent_time_step_ratio = 1, 5
nproc_x = 8
nproc_y = 7
MPI ranks = 56
OMP_NUM_THREADS = 1
The 2049 simulation fails on 1 March at approximately 11:57 UTC. I have run it three times, including a fresh run after correcting the emission settings, and it terminates at exactly the same model time each time. Equivalent simulations for 2048, 2050, 2051, and 2052 use the same executable, model configuration, emissions, chemistry, and MEGAN inputs (with different corresponding dates) and run successfully.
Error
The first actual failure appears in rsl.error.0037:forrtl: severe (174): SIGSEGV, segmentation fault occurred
The traceback points to:
module_mp_thompson_mp_mp_thompson_
module_mp_thompson_mp_mp_gt_driver_
module_microphysics_driver_mp_microphysics_driver_
SLURM subsequently reports processes being killed by signal 9, but this appears to happen after rank 37 encountered the segmentation fault.
Both domains reach approximately 11:57 UTC in the logs before the failure. The traceback does not identify the domain so I cannot yet determine whether the failure originates in d01 or d02.
Checks already completed
I compared the failed 2049 case with the successful 2048, 2050, 2051, and 2052 cases.- The 2049 namelist differs from the successful cases only in the year and output path.
- met_em files for both domains around the failure time were scanned: no NaN, or broad physical-range violations were found.
- All geo_em numeric variables are identical across years except the intentionally modified annual SOILTEMP, which contains reasonable values and no NaN.
- wrfinput_d01 and wrfinput_d02 have identical structures across years and contain no NaN.
- Derived pressure, temperature, water vapour, and air density in the 2049 wrfinput files are physically reasonable.
- Initial hydrometeor fields are zero in every simulation.
- The required 1 March 12Z record exists in wrfbdy_d01, and the stored boundary fields contain no NaN or obvious isolated extreme.
- No CFL, NaN, Inf, or floating-point warning appears before the segmentation fault.
The failure occurs at the same model time in three independent runs (stop exactly at the same time stamp even after two restarts), while the surrounding years complete successfully, I am unable to find out what is causing this segmentation fault and eventually the termination.
I have attached namelist.input, the relevant rank-37 rsl.error and rsl.out
Thank you for your help.