Dear nwchem community,
I am having trouble to convergence a wavefunction for a carbon nanotube with 2 hydrogen atoms using DFT+PBE0.
Here is (part of) the input
basis
* library 3-21g
end
dft
iterations 2000
xc pbe0
direct
grid fine
convergence gradient 8E-4 diis 4
end
driver
xyz geo
end
task dft optimize
I have two systems (quite similar except for different positioning of the hydrogen atoms.
The structures are optimized PBE geometries from CPMD.
Here is one that works
2b/output: d= 0,ls=0.0,diis 1 -9108.4700250578 -8.16D+04 8.10D+00 2.44D+01 5609.3
2b/output: d= 0,ls=0.5,diis 2 -9034.9757357418 7.35D+01 4.66D-01 8.74D+01 17541.8
2b/output: d= 0,ls=0.5,diis 3 -9041.3269753269 -6.35D+00 2.02D-01 7.76D+01 29413.3
2b/output: d= 0,ls=0.5,diis 4 -9043.6904442580 -2.36D+00 1.21D-01 7.56D+01 41248.0
2b/output: d= 0,ls=0.5,diis 5 -9044.9817332625 -1.29D+00 7.70D+00 7.43D+01 53002.8
2b/output: d= 0,ls=0.5,diis 6 -9095.2034257819 -5.02D+01 3.46D-02 5.85D+01 59436.9
2b/output: d= 0,ls=0.5,diis 7 -9105.7684680531 -1.06D+01 4.17D-03 1.89D+01 65563.8
2b/output: d= 0,ls=0.5,diis 8 -9104.4813523533 1.29D+00 1.18D-03 3.24D+01 71684.0
2b/output: d= 0,ls=0.5,diis 9 -9105.8769727898 -1.40D+00 7.15D-04 2.18D+01 77663.8
2b/output: d= 0,ls=0.5,diis 10 -9107.4244567859 -1.55D+00 1.16D-03 1.31D+01 83650.6
2b/output: d= 0,ls=0.5,diis 11 -9109.2070951994 -1.78D+00 1.27D-03 4.41D+00 92383.6
2b/output: d= 0,ls=0.5,diis 12 -9110.2622692031 -1.06D+00 5.06D-04 6.65D-01100948.2
2b/output: d= 0,ls=0.5,diis 13 -9110.3633724795 -1.01D-01 2.90D-04 3.92D-01108950.1
2b/output: d= 0,ls=0.5,diis 14 -9110.4234724909 -6.01D-02 2.45D-04 2.25D-01114918.9
2b/output: d= 0,ls=0.5,diis 15 -9110.4774325073 -5.40D-02 1.56D-04 6.57D-02120876.2
2b/output: d= 0,ls=0.5,diis 16 -9110.5041094244 -2.67D-02 1.04D-04 2.15D-02126837.3
2b/output: d= 0,ls=0.5,diis 17 -9110.5187859539 -1.47D-02 9.39D-05 1.60D-02132823.1
2b/output: d= 0,ls=0.5,diis 18 -9110.5298692886 -1.11D-02 9.20D-05 1.29D-02138816.4
2b/output: d= 0,ls=0.5,diis 19 -9110.5394020628 -9.53D-03 7.53D-05 9.69D-03144784.6
2b/output: d= 0,ls=0.5,diis 20 -9110.5469022106 -7.50D-03 6.75D-05 7.56D-03150754.4
And one that doesn't
2a/output: d= 0,ls=0.0,diis 1 -9108.5023183806 -8.16D+04 8.27D+00 2.43D+01 5616.0
2a/output: d= 0,ls=0.5,diis 2 -9034.6634884197 7.38D+01 4.65D-01 8.50D+01 17581.9
2a/output: d= 0,ls=0.5,diis 3 -9040.9435208185 -6.28D+00 1.96D-01 7.60D+01 29477.9
2a/output: d= 0,ls=0.5,diis 4 -9043.2585279689 -2.32D+00 1.22D-01 7.42D+01 41316.6
2a/output: d= 0,ls=0.5,diis 5 -9044.5516229056 -1.29D+00 7.86D+00 7.31D+01 53169.2
2a/output: d= 0,ls=0.5,diis 6 -9093.4199801709 -4.89D+01 2.87D-02 6.70D+01 59590.5
2a/output: d= 0,ls=0.5,diis 7 -9080.1167325488 1.33D+01 3.59D-03 1.43D+02 65637.3
2a/output: d= 0,ls=0.5,diis 8 -9095.7736875505 -1.57D+01 1.38D-03 6.66D+01 71659.0
2a/output: d= 0,ls=0.5,diis 9 -9099.3601973387 -3.59D+00 7.82D-04 4.62D+01 77709.0
2a/output: d= 0,ls=0.5,diis 10 -9102.5658530250 -3.21D+00 1.79D-02 3.06D+01 83711.3
2a/output: d= 0,ls=0.5,diis 11 -9030.5701065372 7.20D+01 4.80D-03 3.48D+02 93114.0
2a/output: d= 0,ls=0.5,diis 12 -9035.9333262450 -5.36D+00 3.85D-03 2.94D+02102218.4
2a/output: d= 0,ls=0.5,diis 13 -9039.6119263338 -3.68D+00 3.68D-03 2.71D+02110320.0
2a/output: d= 0,ls=0.5,diis 14 -9042.2300588374 -2.62D+00 1.59D-01 2.66D+02116583.4
2a/output: d= 0,ls=0.5,diis 15 -942.0039711970 8.10D+03 7.09D-02 1.99D+04123442.3
2a/output: d= 0,ls=0.5,diis 16 -3417.0117307364 -2.48D+03 4.12D-02 1.52D+04130265.1
2a/output: d= 0,ls=0.5,diis 17 -4366.1889341364 -9.49D+02 3.40D-02 1.61D+04137121.0
2a/output: d= 0,ls=0.5,diis 18 -4847.9189437237 -4.82D+02 5.71D+00 1.96D+04144007.5
2a/output: d= 0,ls=0.5,diis 19 43359.2082533070 4.82D+04 7.22D+00 3.62D+03152453.2
2a/output: d= 0,ls=0.5,diis 20 35898.8558078528 -7.46D+03 4.11D+00 1.53D+04160518.8
2a/output: d= 0,ls=0.5,diis 21 24338.8539746737 -1.16D+04 4.64D-01 2.30D+04167974.2
2a/output: d= 0,ls=0.5,diis 22 22345.8899119936 -1.99D+03 6.14D-01 2.31D+04175136.6
It seems to me that the initial step size is too large and the DIIS algorithm gets lost for the latter geometry.
Is there a way to choose the initial DIIS step size?
Or is there another, algorithm like steepest descent or conjugate gradient that I can try for the SCF convergence?
(I could not find it in the part of manual about dft)
Thank you in advance for any suggestions.
Best, Jaap
|