Thanks for giving more details. I think you have done the simulations and analysis properly. Your HCACF results indicate that your system have complex high frequency modes, which makes the evaluation of the thermal conductivity more difficult. I have one more suggestion (on post-processing) right now:
– Divide your 500 ps heat current data into five blocks (thus each with 100 ps), and evaluate an HCACF for each block of data.
– For each block of data, set the maximum correlation time to about 10 ps.
– Plot the five running thermal conductivity (similar to the blue curve in the second figure) curves in the same figure to see whether they have large differences. This can help to identify the statistical error without doing more simulations.
Hi Bruce and Markus,
Not sure if my email yesterday got through as i got a rejection email afterwards as i ‘wasn’t a member’. Incase it didn’t, the reply was to Markus saying i did changed this the acceleration default ‘mixed’ to ‘double’ in the command line. Either way, I noticed minimal (very subtle differences in flux values) with both modes tested - on very short simulations though.
I finally modified a version of my code (it was quite rigid) to spilt the data into 100 ps chunks with a 10 ps autocorrelation truncation, and then show the running TC as a function of t, as suggested by you. Figures and outputs of airebo and airebo/intel are attached.
I realise the figures too distinguishable with their sections, but the general assumption i have from these graphs/results are that the airebo/intel sections are relatively similar and oscillates closely around zero, and the airebo consist of larger error (expected for short simulations, i assume) but at least isn’t near-zero.
AirInt_outputsSPILT_80000values_5sampl_8000aclength.txt (6.72 KB)
AIREBOoutputsSPILT_80000values_5sampl_8000aclength.txt (6.79 KB)
I am not an expert on the math of the stress/heat/flux part, but IIRC
stress/atom requires per-atom virials, which is one of the things that
AIREBO/intel (and, for that matter, the USER-INTEL package) currently
does not support.
So the virial contribution from the potential is incorrectly zero.
I you should have gotten an error somewhere along the line, but that
check seems to be buggy right now.
Per-atom virials might, however, be added in the future.
This makes perfect sense. Without the virial-related heat current, the heat current only comes from the kinetic (convective) part, which gives essentially zero thermal conductivity for solid phase. This is what Ben observed for the intel version.
Markus Höhnerbach <[email protected]…24…> 于 2019年2月13日周三 20:38写道：
Hi Markus and Bruce,
Thanks a lot for clearing that up! I appreciate all the help in the emails.