strange behavior of LRF

[JordiVillaFreixa]

in two different systems, late installations of Q6 (since September 2023 at least) are creating memory problems (segmentation fault). The problem has been tracked with the debug compilation and we have identified that it is the LRF calculation which produces the errors. Before going into more detail, is that an identified issue by the developers?

[acmnpv]

Hello, I have been told about something like this as well a few weeks back, but have not been able to dig too much into it due to work and family issues. Can you add the debug trace here so I can compare it to the other one I received?

[ND7996]

Dear Paul ,

I am sending the files with the debug while compiling

outputmakedebug.txt - com - make debug outputmakealldebug.txt - com - make all debug outputmakeall.txt - com - make all COMP=gcc

initial error when running relax.

error.txt

[acmnpv]

Hello, I had a quick glance at the log file, and can already see that you have an issue with your system charge, as your Q/classical system charge is not an integer value. This shouldn't cause any crash, but will create issues with the physical representation of the system later.

Can you run your input with the debug build of Q6 as well?

[ND7996]

Thank you for your response , will check the issues with charge but now I only tried compiling with the command - make debug which shows an error as below makedebug_error.txt

Then i tried make all debug which is as follows

makealldebug_output.txt

and the output as below with the debug build

relax_001.log

[acmnpv]

hello, your latest run is still from the non-debug build.

I need to see if I can try to reproduce the issue with the debug build

[ND7996]

Apologies, here is the file with Debug built

relax_001.log

[acmnpv]

thanks! The issue is the same as I have seen before, where an invalid charge group is being used for the calculation (dr = lrf(ic)%cgp_cent - x(i), where as far as I can see the lrf(ic) field points to uninitialized memory)

The thing is that I honestly don't know what changed there that makes this bug appear now after all this time, and I don't currently have the time to run this through a debugger the investigate further.

If I find the time over the weekend I might be able to do some more checks

[JordiVillaFreixa]

Thanks @acmnpv , in the meanwhile we'll try to check the issue of the charge, which surely does not help having a smooth run. It would be good having a check in Q6 that prevents reaching the segmentation fault problem when issues like this happen. Thanks a lot.

[ND7996]

Hi Paul

we have been looking for the error and found that it is in the execution in linux. We tried with two compilers (gcc and ifort) but the segmentation error still stays when running Qdyn. While in Mac the Qdyn runs well. i am sending my relax inputs and output error with debug built and using the compiler ifort, with the fep and topology file as well. mousecysrelax_ifort.zip

[acmnpv]

did you use an ARM Mac, or an x86_64 processor one? It would be another datapoint for me to look at.

[JordiVillaFreixa]

intel x86_64