QuantumATK Forum
General => News and Announcements => Topic started by: Anders Blom on October 21, 2014, 15:24
-
An error which appears in some rare cases for the FFT2D solver has been detected, which results in the potential going to infinity, causing the calculation to crash or gives nan/inf values. These should be easy to spot in the log file.
This appears to actually be bug in the current Intel Math Kernel Library (MKL) which we use for all matrix operations. It manifests itself when your grid is 128x128, 256x256 or 512x512, in which case it will give completely wrong results.
This has already been posted by someone else on the Intel forum (https://software.intel.com/en-us/forums/topic/509936) and we are waiting for Intel to provide a fix.
The workaround for now is to increase the density grid mesh cut-off slightly - just a Hartree or two will usually do the trick!