Closed ajkerr0 closed 7 years ago
A small minority of the terms in the double sum are big (large absolute values), we think that this is the source of the errors. The same relative errors on these big terms could swing the total thermal conductivity number.
We have found that increasing the damping (gamma) factor vastly reduces the spread of these outlier values.
We found with a larger stapling to a vacuum the zero modes contributing to the large terms seem to go away.
With an appreciable drag constant we have yet to see unstable kappa values.
Going to close this issue as long as we get stable values.
Especially notable when performing ParamSpaceExplorer.explore, certain values are seemingly 'out of bounds', either absurdly high compared to the bulk of the values or negative.
Is this an error that occurs due to the nature of drivers/stapled atoms being in nodes of modes? Is it an indicator of a fault of the code that only shows itself in certain cases? Is it a fault of the theory?