QuantumATK Forum

QuantumATK => Scripts, Tutorials and Applications => Topic started by: 395235863 on May 17, 2015, 05:47

Title: The computed multigrid residual is greater than the required accuracy
Post by: 395235863 on May 17, 2015, 05:47

in the .e file
** Back Engine Exception : Singular value decomposition failed
** Location of Exception : mathutils.cpp:1604

in the log file
################################################################################
# WARNING                                                                      #
#                                                                              #
# The computed multigrid residual is greater than the required accuracy.       #
#                                                                              #
# Computed residual :         inf                                            #
# Required accuracy :   1.00000e-12                                            #
#                                                                              #
################################################################################

i attached the input.py out.log job.e files.

what the problems?

desired for your help!

thank you very much!

Title: Re: The computed multigrid residual is greater than the required accuracy
Post by: Jess Wellendorff on May 18, 2015, 17:27

Hi there!

The problem is that the two metallic regions on top and below the dielectrics in your device are too thin. If you increase the thickness of both metallic regions from 0.1 Å to at least 1.0 Å, the problem disappears.

The reason is that the Multi-grid Poisson solver does not work well when the metallic regions become extremely thin (e.g. 0.1 Å, less than a monolayer of metal).
You could also consider using the Direct solver, which does not rely on iterative solution of the Poisson equation. That algorithm uses significantly more memory, but might be more efficient if your calculation uses many SCF steps.

Title: Re: The computed multigrid residual is greater than the required accuracy
Post by: Anders Blom on May 18, 2015, 17:28

The technical reason for this is that any "region" must be thick enough to actually contain a grid point for the electrostatic potential and electron density grids, otherwise the region is actually not included in the calculation, effectively.