My Community

I have a system that does not converge. I tried running 20 iterations with AMIX=0.2, BMIX=1.E-4, AMIX_MAG =0.8, and BMIX_MAG =1.E-4, then restarting with istart=1 and the default parameters. The last two iterations of the first calculation have similar energies, but the restart is very different.
The output suggests the restart worked
reading WAVECAR
the WAVECAR file was read sucessfully
initial charge from wavefunction
entering main loop
If I run small systems, the restart works perfectly, but for a large slab with 109 atoms, I have a problem. I am attaching the last iteration of the first calculation and the first of the restart. Any idea why the restart energy is so different from the previous run?
Is there some additional parameters I need to set for the restart.

Last iteration of fist run

total energy-change (2. order) : 0.1895780E+00 (-0.1161716E-01)
number of electron 582.0001388 magnetization 2.0000039
augmentation part 49.1712487 magnetization 0.7861362

Free energy of the ion-electron system (eV)
---------------------------------------------------
alpha Z PSCENC = 353.12760340
Ewald energy TEWEN = 289088.79829745
-1/2 Hartree DENC = -321287.80685203
-V(xc)+E(xc) XCENC = 896.96981175
PAW double counting = 24882.89705078 -23492.05478448
entropy T*S EENTRO = 0.15179800
eigenvalues EBANDS = -6590.31959519
atomic energy EATOM = 35321.38948610
---------------------------------------------------
free energy TOTEN = -826.84718422 eV

energy without entropy = -826.99898222 energy(sigma->0) = -826.89778355


restart istart=1
eigenvalue-minimisations : 2048
total energy-change (2. order) : 0.3927231E+04 (-0.2262286E+05)
number of electron 582.0000000 magnetization 2.0000000
augmentation part 582.0000000 magnetization 2.0000000

Free energy of the ion-electron system (eV)
---------------------------------------------------
alpha Z PSCENC = 353.12760340
Ewald energy TEWEN = 289088.79829745
-1/2 Hartree DENC = -319984.92051822
-V(xc)+E(xc) XCENC = 795.37877967
PAW double counting = 19677.69668483 -18267.42073329
entropy T*S EENTRO = -0.01223799
eigenvalues EBANDS = -3056.80645250
atomic energy EATOM = 35321.38948610
---------------------------------------------------
free energy TOTEN = 3927.23090944 eV

energy without entropy = 3927.24314743 energy(sigma->0) = 3927.23498877

the only reason I can think of is that the size and shape of the cell, the magnetization,or the space group,... have changed substantially in the first run (the one you continue from), such that the FFT grids (NGX,....), kpoint-sets, or #of bands are not appropriate for the new starting geometry anymore (please mind that ISTART=1 uses the parameters of the previous run 'brute force', no matter whether they are suitable for the current run.
--> therefore please check
-- has magnetization changed substantially (#of bands!)
-- please run on the same #of CPUs (#bands always is an integer multiple of the #of CPUs)
-- if you run with ISIF >=3: please check whether the axis ratios, cell volume and cell geometry (if you have switched off symmetry) have changed (NGX,NGY,NGZ, k-point mesh)