LAECHG=.TRUE. problem

[hunterdai]

In VASP 4. 6 31,

As suggested in the website, I put LAECHG=.TRUE.
in the INCAR file. It runs well for some systems. But for other systems, I got some error messages like following.

What was the reason could be? How can I solve it?

Thanks

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
vasp 00000000005C90E5 Unknown Unknown Unknown
vasp 0000000000682AC1 Unknown Unknown Unknown
vasp 0000000000420A87 Unknown Unknown Unknown
vasp 000000000040612A Unknown Unknown Unknown
libc.so.6 00002ADC922445AA Unknown Unknown Unknown
vasp 000000000040606A Unknown Unknown Unknown
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
vasp 00000000005C90E5 Unknown Unknown Unknown
vasp 0000000000682AC1 Unknown Unknown Unknown
vasp 0000000000420A87 Unknown Unknown Unknown
vasp 000000000040612A Unknown Unknown Unknown
libc.so.6 00002B91948E55AA Unknown Unknown Unknown
vasp 000000000040606A Unknown Unknown Unknown

[graeme]

It's most likely a memory limitation; this flag requires lots of memory. Try running the job in parallel so that it is distributed across machines with a greater total memory.

[hunterdai]

I realize that it works with PAW, but not for USPP.

How can I make it work with USPP?

Thanks

[graeme]

I don't think that you can. The point of this flag is to write the core charges, which are well defined in the PAW (frozen core) approximation, but not in the USPP.

[hunterdai]

ok, thanks so much!

[graeme]

So you got the segfault with a small system (to avoid any memory problems) using a PAW potential? Also, check that you did a clean build of vasp. If you still have problems, you could take it up with the vasp folk -- the writing of core charges is standard in vasp now and does not involve our code.