fastlg error in test_sne

[chongchonghe]

Describe the bug
Resampled cooling causes this error: Assertion failed: (x > 0 && "log divergent for x <= 0"), function fastlg, file FastMath.hpp, line 37.. This is not necessary a bug in the cooling routine, but what could've caused this?

To Reproduce
Steps to reproduce the behavior:

1. Compile this problem 'test_sne' (require Aditi's branch with some bug fix)
2. Run this problem 'test_sne'
3. See error

Coarse STEP 1 at t = 0 (0%) starts ...
...[level 0] estimated hydro timestep: 1.565446e+12
...[level 0] 	hydro timestep limited at cell [31, 31, 255] with signal speed = 9.036961e+06
...[level 0] 	cell density = 1.642355e-29, |v| = 0.000000e+00, cs = 9.036961e+06
...[level 0] timestep limited by HYDRO
...coarse timestep set by level 0
[Level 0 step 1] ADVANCE with time = 0.0000000000e+00 dt = 1.5654460365e+12
	cooling substeps (per cell): avg 1.03125, max 2
	cooling substeps (per cell): avg 1, max 1
The total number of SN gone off= 4
The total number of SN gone off= 4
[Level 0 step 1] Advanced 262144 cells

Coarse STEP 2 at t = 1.565446036e+12 (0.01654455756%) starts ...
...[level 0] estimated hydro timestep: 2.296661e+11
...[level 0] 	hydro timestep limited at cell [0, 0, 116] with signal speed = 6.159757e+07
...[level 0] 	cell density = 6.983390e-25, |v| = 1.623054e+03, cs = 6.159595e+07
...[level 0] timestep limited by HYDRO
...coarse timestep set by level 0
[Level 0 step 2] ADVANCE with time = 1.5654460365e+12 dt = 2.2966611094e+11
	cooling substeps (per cell): avg 1, max 1
Assertion failed: (x > 0 && "log divergent for x <= 0"), function fastlg, file FastMath.hpp, line 37.
SIGABRT
See Backtrace.0 file for details
--------------------------------------------------------------------------
MPI_ABORT was invoked on rank 0 in communicator MPI_COMM_WORLD
  Proc: [[25045,0],0]
  Errorcode: 6

Additional context
Add any other context about the problem here.

[dosubot]

The fastlg assertion error happens when a zero or negative value is passed to the logarithm function, which is undefined for such inputs. In the resampled cooling routines, fastlg is used to compute the log of density (rho) and specific internal energy (eint = Eint/rho) for table interpolation. If either rho or eint is zero, negative, or NaN, you'll hit this assertion. There are currently no explicit checks or floors in the cooling routines to guarantee these values are always positive before calling fastlg—the only safeguard is the assertion itself, which is why you're seeing the crash rather than a more graceful failure or warning see source.

A related issue has shown that NaN values can be produced in the SN feedback routine and then propagate into the cooling routine, causing failures like this see issue #1161. This means the root cause is likely upstream: either the SN feedback or energy deposition is producing non-physical values (zero, negative, or NaN) for internal energy or density, which then get passed to the cooling routines.

Recent changes—such as switching from 'grackle' to 'resampled' cooling tables and modifying SN energy deposition—could have affected the values passed to fastlg see PR #1179. Improving robustness would require adding explicit checks or clamps for physical admissibility (e.g., ensuring rho > 0 and Eint > 0) before log-scaling in the cooling routines.

For debugging, try logging or printing the values of rho and Eint just before the call to fastlg in the cooling routines to catch where the non-physical value first appears. This should help you trace the issue back to its source.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[BenWibking]

@chongchonghe Do you still see this issue with the current version of this problem?

[BenWibking]

I think this should be fixed with release 26.04. Please re-open if it is not fixed.