Hello Sinan,

I have run the template of the BNS merger with different versions of ET, specifically 202305 and 202211. The difference between the two versions is negligible. However, there is a noticeable difference between my run and the reference data provided on the ET website. Since I used the same files provided by the template, I am wondering what could the difference be and whether it indicates my running has certain faults.

Looking at the plots included and given the relatively low resolution of the gallery example parameter file (dx=8.0 on coarsest level, star is on level 6 so dx_star = 8.0/2**6 = 0.125 = 180m which is only a "low resolution" run for publication data) I think the difference may well be explained due to different compiler options or compilers. In particular use the Intel compiler (with defaults to -ffast value unsafe optimizations) and GNU compiler (by default only uses value save optimizations) can lead to noticeable differences. The directly visible differences in psi4 are all in the post-merger phase of the evolution which is very sensitive to small roundoff errors.

These errors are further amplified in psi4 which is the second time derivative of gravitational wave strain, which in turn is the quantity actually measured by experiments. The two time derivatives boost high frequency signals such as the post-merger signal. Converting to strain would reduce the magnitude of this part of the signal and thus make its (absolute, not relative) error go down.

Could you provide details on the compiler option, processors type and possibly cluster that you used?

Yours, Roland