[Pw_forum] Convergence threshold
Carlo Sbraccia
sbraccia at sissa.it
Thu Apr 7 20:03:12 CEST 2005
Dear Sergey,
when you do a structural relaxation (as, I think, you are doing) the
convergence threshold is set automatically by the algorithm according to
the level of convergence achieved in the structural relaxation. In
particular, in the input file is set the minimal threshold for scf. In the
namelist &IONS you can set the upscale variable (default is 10.D0) whose
pourpose is to decrease the threshold down to conv_thr / upscale when the
structural relaxation is converged. The idea is that when you are far from
the relaxed structure you don't need a very accurate scf convergence. If
you think that 1.D-8 is a good scf threshold, you cas set conv_thr = 1.D-6
and upscale = 100.D0. The procedure is also explained in the INPUT_PW
file.
Let me know if this clarifies your concerns.
Best,
carlo
On Thu, 7 Apr 2005, Sergey Lisenkov wrote:
> Dear Carlo,
>
> Thank you very much for your response.
>
>> do the dots mean that the code performs other iterations after the message
>> "End of self-consistent calculation" ??? Can you post the complete output
>> file and the details of your input ?
>
> I mean that code performs other iterations but the convergence is already achieved:
>
> Please, look in output:
>
> convergence threshold = 1.0E-08
>
> After first scf loop:
>
> total energy = -304.04955368 ryd
> estimated scf accuracy < 0.00000003 ryd
>
> iteration # 12 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 3.33E-11, avg # of iterations = 2.0
>
> total cpu time spent up to now is 663.40 secs
>
> End of self-consistent calculation
>
> k = 0.0625 0.0000 0.0000 (69280 PWs) bands (ev):
>
> -14.3985 -14.1020 -14.1014 -13.4128 -13.4127 -12.9699 -12.8594 -12.8592
> -12.8567 -12.8560 -8.9009 -8.6354 -8.6344 -8.2400 -8.2399 -7.4494
> -7.4493 -6.9773 -6.7669 -6.7669 -6.5989 -6.5972 -6.5658 -6.5612
> -5.8985 -5.8984 -5.6890 -5.4801 -5.4794 -5.2608 -4.7019 -4.7017
> -4.5606 -4.5575 -4.5170 -4.5165 -4.4484 -4.3031 -4.1997 -4.1993
>
> k = 0.1875 0.0000 0.0000 (69344 PWs) bands (ev):
>
> -14.3210 -14.0308 -14.0302 -13.3701 -13.3701 -13.1341 -12.9893 -12.9886
> -12.8983 -12.8982 -8.6196 -8.4586 -8.4574 -8.3730 -8.3729 -7.7287
> -7.7285 -6.9513 -6.7173 -6.7131 -6.4254 -6.4241 -6.3997 -6.3996
> -5.6185 -5.5852 -5.5850 -5.5652 -5.5640 -5.3769 -4.7220 -4.6945
> -4.6914 -4.6360 -4.6358 -4.5810 -4.4900 -4.4896 -4.2428 -4.2424
>
> k = 0.3125 0.0000 0.0000 (69432 PWs) bands (ev):
>
> -14.1698 -13.8926 -13.8920 -13.3965 -13.2910 -13.2910 -13.2086 -13.2080
> -12.9727 -12.9725 -8.4728 -8.4727 -8.1564 -8.1550 -8.1399 -8.0547
> -8.0546 -7.0152 -7.0120 -6.9517 -6.0979 -6.0968 -5.8711 -5.8710
> -5.5557 -5.5555 -5.5537 -5.4814 -5.2734 -5.2598 -5.2596 -4.9107
> -4.9078 -4.8003 -4.6664 -4.6659 -4.5314 -4.5311 -4.3216 -4.3212
>
> k = 0.4375 0.0000 0.0000 (69512 PWs) bands (ev):
>
> -13.9537 -13.6973 -13.6967 -13.6846 -13.4578 -13.4572 -13.1886 -13.1885
> -13.0721 -13.0720 -8.4579 -8.4579 -8.3125 -8.3124 -7.7879 -7.7860
> -7.5834 -7.3939 -7.3914 -7.1285 -5.7016 -5.6654 -5.6642 -5.6545
> -5.4581 -5.4557 -5.3745 -5.3744 -5.2871 -5.1394 -5.1393 -5.1283
> -5.1258 -5.0494 -5.0430 -5.0422 -4.4686 -4.4684 -4.4092 -4.4088
>
> ! total energy = -304.04955370 ryd
> estimated scf accuracy < 2.8E-09 ryd
>
>
> But after third and next scf loop:
>
> ...
> total energy = -304.04986379 ryd
> estimated scf accuracy < 0.00000238 ryd
>
> iteration # 8 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 2.98E-09, avg # of iterations = 3.3
>
> total cpu time spent up to now is 2657.46 secs
>
> total energy = -304.04986469 ryd
> estimated scf accuracy < 0.00000074 ryd
>
> iteration # 9 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 9.26E-10, avg # of iterations = 2.0
>
> total cpu time spent up to now is 2705.26 secs
>
> total energy = -304.04986483 ryd
> estimated scf accuracy < 0.00000048 ryd
>
> iteration # 10 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 5.97E-10, avg # of iterations = 2.3
>
> total cpu time spent up to now is 2756.29 secs
>
> total energy = -304.04986512 ryd
> estimated scf accuracy < 0.00000001 ryd
>
> iteration # 11 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 1.79E-11, avg # of iterations = 3.5
>
> total cpu time spent up to now is 2806.19 secs
>
> total energy = -304.04986512 ryd
> estimated scf accuracy < 0.00000001 ryd
>
> iteration # 12 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 1.34E-11, avg # of iterations = 2.0
>
> total cpu time spent up to now is 2857.51 secs
>
> total energy = -304.04986513 ryd
> estimated scf accuracy < 4.6E-09 ryd ! it should be enough
>
> iteration # 13 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 5.79E-12, avg # of iterations = 3.0
>
> total cpu time spent up to now is 2911.27 secs
>
> total energy = -304.04986513 ryd
> estimated scf accuracy < 5.1E-09 ryd
>
> iteration # 14 ecut= 38.00 ryd beta=0.30
> Davidson diagonalization (with overlap)
> ethr = 5.79E-12, avg # of iterations = 2.3
>
> total cpu time spent up to now is 2961.01 secs
>
> End of self-consistent calculation
>
> k = 0.0625 0.0000 0.0000 (69280 PWs) bands (ev):
>
> -14.4011 -14.1046 -14.1042 -13.4154 -13.4154 -12.9701 -12.8624 -12.8622
> -12.8577 -12.8571 -8.9136 -8.6463 -8.6455 -8.2410 -8.2408 -7.4448
> -7.4447 -6.9766 -6.7735 -6.7734 -6.5935 -6.5922 -6.5671 -6.5622
> -5.9090 -5.9088 -5.6854 -5.4780 -5.4773 -5.2569 -4.7024 -4.7023
> -4.5561 -4.5529 -4.5289 -4.5286 -4.4408 -4.3170 -4.1982 -4.1979
>
> k = 0.1875 0.0000 0.0000 (69344 PWs) bands (ev):
>
> -14.3235 -14.0333 -14.0328 -13.3727 -13.3726 -13.1349 -12.9907 -12.9901
> -12.9015 -12.9013 -8.6317 -8.4685 -8.4675 -8.3742 -8.3741 -7.7266
> -7.7264 -6.9511 -6.7164 -6.7122 -6.4241 -6.4228 -6.4057 -6.4056
> -5.6147 -5.5927 -5.5925 -5.5644 -5.5633 -5.3739 -4.7342 -4.6914
> -4.6884 -4.6369 -4.6368 -4.5741 -4.4981 -4.4977 -4.2416 -4.2413
>
> k = 0.3125 0.0000 0.0000 (69432 PWs) bands (ev):
>
> -14.1720 -13.8948 -13.8943 -13.3981 -13.2932 -13.2932 -13.2106 -13.2101
> -12.9762 -12.9761 -8.4740 -8.4739 -8.1645 -8.1632 -8.1508 -8.0543
> -8.0542 -7.0168 -7.0136 -6.9530 -6.0974 -6.0963 -5.8767 -5.8767
> -5.5549 -5.5540 -5.5530 -5.4772 -5.2831 -5.2649 -5.2647 -4.9087
> -4.9058 -4.7945 -4.6717 -4.6712 -4.5328 -4.5327 -4.3212 -4.3208
>
> k = 0.4375 0.0000 0.0000 (69512 PWs) bands (ev):
>
> -13.9548 -13.6985 -13.6980 -13.6875 -13.4611 -13.4606 -13.1891 -13.1891
> -13.0773 -13.0772 -8.4586 -8.4586 -8.3133 -8.3132 -7.7937 -7.7919
> -7.5923 -7.3980 -7.3956 -7.1341 -5.7019 -5.6657 -5.6646 -5.6601
> -5.4571 -5.4548 -5.3795 -5.3795 -5.2818 -5.1441 -5.1440 -5.1273
> -5.1249 -5.0463 -5.0456 -5.0450 -4.4675 -4.4673 -4.4118 -4.4115
>
> ! total energy = -304.04986513 ryd
> estimated scf accuracy < 5.0E-10 ryd
>
> band energy sum = -46.52156332 ryd
> one-electron contribution = -947.75987385 ryd
> hartree contribution = 484.42869513 ryd
> xc contribution = -183.18402805 ryd
> ewald contribution = 342.46534165 ryd
>
> convergence has been achieved
>
> Forces acting on atoms (Ry/au):
> .....
>
>
> May be i don't understand the process of scf calculations properly?
>
> Best wishes,
> Sergey
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
More information about the Pw_forum
mailing list