Pavlo Solntsev
pavlo.solntsev@gmail.com
My two cents.
Do you use diffuse functions? Is geometry really good? In the first case you may drop off some diffuse functions to achieve a convergence. If geometry is not good, you may try to use smaller basis set and then expend it up to desired size. Keep in mind, for DFT you may also need to increase grid accuracy, especially for transition metals.
Pavel.
On Tue Dec 31 '13 3:30pm, David G. wrote
----------------------------------------
>>Dear Thom
>thank you for prompt reply.
>Strictly speaking I have already used the following setting:
>$CONTRL icut=11 INTTYP=HONDO $END
>$SCF DIRSCF=.TRUE. MAXDII=30 FDIFF=.f. $END
>and the energy and density errors of one of the scf procedures are as the following:
> DIRECT SCF CALCULATION, SCHWRZ=T FDIFF=F
> NONZERO BLOCKS
> ITER EX DEM TOTAL ENERGY E CHANGE DENSITY CHANGE DIIS ERROR INTEGRALS SKIPPED
> * * * INITIATING DIIS PROCEDURE * * *
> 1 0 0 -6318.293008038 -6318.293008038 16.874057280 0.033936892 -1 253644024
> 2 1 0 -6318.294422865 -0.001414827 3.414042850 0.002014172 -1 253618487
> 3 2 0 -6318.294479580 -0.000056715 1.496379733 0.002124828 -1 253623176
> 4 3 0 -6318.294549256 -0.000069677 1.891834693 0.000885264 -1 253622891
> 5 4 0 -6318.294584416 -0.000035159 0.559149100 0.000207986 -1 253620301
> 6 5 0 -6318.294586242 -0.000001827 0.553632503 0.000127246 -1 253622189
> 7 6 0 -6318.294586678 -0.000000436 0.327187758 0.000053581 -1 253622878
> 8 7 0 -6318.294586785 -0.000000107 0.110526091 0.000066342 -1 253622412
> 9 8 0 -6318.294586860 -0.000000075 0.065027675 0.000018497 -1 253622147
> 10 9 0 -6318.294586877 -0.000000018 0.054933155 0.000004938 -1 253622337
> 11 10 0 -6318.294586873 0.000000004 0.051449709 0.000003961 -1 253622415
> 12 11 0 -6318.294586883 -0.000000011 0.032116478 0.000001336 -1 253622444
> 13 12 0 -6318.294586878 0.000000005 0.014416976 0.000001326 -1 253622413
> 14 13 0 -6318.294586882 -0.000000004 0.017963579 0.000000746 -1 253622431
> 15 14 0 -6318.294586882 0.000000000 0.029253188 0.000000296 -1 253622410
> 16 15 0 -6318.294586877 0.000000006 0.047571716 0.000000271 -1 253622356
> 17 16 0 -6318.294586878 -0.000000002 0.017608553 0.000000117 -1 253622453
> 18 17 0 -6318.294586886 -0.000000008 0.017771178 0.000000407 -1 253622408
> 19 18 0 -6318.294586879 0.000000007 0.009235945 0.000000112 -1 253622377
> 20 19 0 -6318.294586882 -0.000000002 0.004881919 0.000000117 -1 253622386
> 21 20 0 -6318.294586884 -0.000000002 0.008555893 0.000000179 -1 253622389
> 22 21 0 -6318.294586880 0.000000005 0.001003955 0.000000115 -1 253622386
> 23 22 0 -6318.294586886 -0.000000006 0.001273499 0.000000275 -1 253622384
> 24 23 0 -6318.294586877 0.000000008 0.003755976 0.000000252 -1 253622389
> 25 24 0 -6318.294586878 -0.000000001 0.002233113 0.000000118 -1 253622391
> 26 25 0 -6318.294586878 0.000000000 0.001947346 0.000000111 -1 253622388
> 27 26 0 -6318.294586879 -0.000000001 0.000369859 0.000000172 -1 253622393
> 28 27 0 -6318.294586886 -0.000000006 0.003189816 0.000000452 -1 253622392
> 29 28 0 -6318.294586882 0.000000004 0.000703361 0.000000222 -1 253622389
> 30 29 0 -6318.294586881 0.000000000 0.001907654 0.000000115 -1 253622387
> 31 30 0 -6318.294586881 0.000000001 0.000790576 0.000000106 -1 253622380
> 32 31 0 -6318.294586879 0.000000001 0.000617188 0.000000189 -1 253622381
> 33 32 0 -6318.294586880 -0.000000001 0.000166500 0.000000109 -1 253622379
> 34 33 0 -6318.294586877 0.000000004 0.000368997 0.000000170 -1 253622381
> 35 34 0 -6318.294586881 -0.000000004 0.000245834 0.000000110 -1 253622378
> 36 35 0 -6318.294586876 0.000000005 0.000544505 0.000000218 -1 253622378
> 37 36 0 -6318.294586879 -0.000000003 0.000476620 0.000000113 -1 253622380
> 38 37 0 -6318.294586875 0.000000004 0.000519073 0.000000233 -1 253622378
> 39 38 0 -6318.294586886 -0.000000011 0.000457023 0.000000465 -1 253622379
> 40 39 0 -6318.294586884 0.000000002 0.000099051 0.000000223 -1 253622378
> 41 40 0 -6318.294586881 0.000000003 0.000124343 0.000000167 -1 253622376
>
>
>as I understand there is no large difference between steps 33 and 41 ! so I prefer to force the code stop at step 33 to reduce the calculation cost. Is it possible to make the FF do it ? does it seem reasonable to do ?
>could we adjust the density convergence criteria in FF ?
>would increasing NCONV raise the computation cost more than the current duration ?
>Many Thanks, David
>
>
>
>
>On Tue Dec 31 '13 1:06pm, Thomas Pijper wrote
>---------------------------------------------
>>Dear David,
>>Such a problem is often solved by tightening a few accuracy parameters. I recommend adding the following options to see if it solves the problem:
>>$CONTRL INTTYP=HONDO ICUT=11 $END
>>Switching from SOSCF to DIIS may also help.
>>By the way, depending on what the calculated wavefunction will be used for, you might want to tighten NCONV to 6 or higher. If you're performing an MP2 calculation, I recommend you select NCONV=7.
>>
>>
>>Kind regards,
>>Thom
>>
>>
>