Fine/coarse transformation coefficients' convergence

Submitted by long9930 on Mon, 09/21/2015 - 10:31

Hello everybody!

I currently have a convergence problem when I was running example SWCNT(8,0). The message is as below

WARNING: there are fine/coarse transformation coefficients with 2-norm < 95%.
To improve convergence, you might want to consider:
- recalculating the kernel using the "extended_kernel" flag
- using the "unrestricted_transformation" flag in the input file
- including more bands from the coarse WFN_co file (number_*_bands_coarse)
- using a coarse WFN_co file with a denser k-mesh

Its from the output of absorption. I was using 64 coarse k-points and 256 fine k-points, 4 c-bands and 4 v-bands. Im trying to increase the bands number right now. But I couldn't find any information regarding to the first two suggestions in the warning message. Is there anybody can help out of this? I'm not sure how important this convergence is.
I really appreciate your guys' help, thanks!

Zhongqu Long

jdeslip's picture

Submitted by jdeslip on Mon, 09/21/2015 - 12:45

Hi Zhongqu,

Please note that the examples do not represent converged parameters. They are unconverged in many ways.

In this particular case, to make the interpolation better, you should generally use more bands on the coarse grid than the on the fine grid. Try with 8 bands or 16 bands.

You could also increase the sampling on the coarse grid. But, I think 32 is OK.

About the first two suggestions, those are new features in BGW 1.1 that we will better explain in the final release (coming very soon). However, I don't think they are relevant for the (8,8) swcnt example. They let you interpolate the conduction bands based on both valence and conduction bands instead of just conduction bands for example.

You can see how well the interpolation is doing by looking at the dccmat and dvvmat files. Usually it is the top and bottom bands that are converged the least.