WARNING: error in the number of electrons¶
Sometimes you may see the following warning:
number of grid points = 54836
DFT exchange-correlation energy: = -230.0368431371773283
number of electrons from numerical integration = 107.9973057254599667
number of electrons from orbital occupations = 108
WARNING: error in the number of electrons = -0.0026942745400333
is larger than 1.0d-3
this can happen when starting from coefficients
from a different geometry
or it can mean that the quadrature grid is inappropriate
What does this mean?
When you run a Kohn–Sham (KS) density functional theory (DFT) calculation, the exchange-correlation (XC) energy and XC potential contributions are integrated numerically. This means that for KS DFT calculations the numerical quadrature grid enters as an additional parameter which you control.
In every iteration, when integrating the XC contributions, the code will also integrate the density at the same time and compare the number of electrons from numerical integration with the number of electrons from orbital occupations as a sanity check and issue the above warning if the difference is larger than a threshold (here 0.001).
If you see this warning, then the number of electrons from numerical integration deviates significantly from the target number of electrons. This is a sanity check and it does not necessarily mean that your results will be useless, since the integrated number of electrons is not a sufficient measure for the quality of the grid for the property that you calculate. If you see this warning in every iteration of your calculation, then it may be a good idea to investigate the quality of the grid. You can either select a finer (and more expensive) grid (see **GRID), and/or tighten the screening threshold (.SCREENING under *DFT).
When restarting from a
DFCOEF file from a different geometry, the warning
can appear during the first few iterations and then this warning disappears
(because the density of the old geometry is then displaced with respect to the
grid at new geometry).
If the integrated number of electrons is significantly off, then there is a serious problem, possibly a bug. In this case please contact the developers and send your output.
If you do not see the above warning, then it does not guarantee that your integration grid is useful for the property that you study. It is always a good practice to verify that your results are converged with respect to the grid parameters. When you calculate a table of functionals, it is a good practice for one of the results to re-run with the best integration grid and to check that your results do not change.
What to do if your DFT run is not converging ? Instead of DFT try first to perform the SCF method. If the previous SCF step does converge, save the MO-coeffient file and use them as starting set for your DFT method.
This trick can help because the HOMO-LUMO gap is higher in the SCF than in the DFT. Restarting the DFT from converged SCF coefficients makes the DFT method converging in many cases.