[molpro-user] SACASSCF in MOLPRO & ORCA
Davood Alizadeh
alizadeh_davood at yahoo.com
Wed Jan 14 20:57:00 GMT 2015
Dear Dr. Hill,
thank you very much for your attention.
I just tried rotating the MOs [those optimized at the CAS (19,12)] in order to produce the desired active space using:
+++++++++++++++ ROTATE,14.1,16.1,0;
ROTATE,6.2,7.2,0;
ROTATE,6.3,7.3,0;+++++++++++++++
But, it didn`t help and the re-optimized active space is similar to the initial guess without rotation!
When I use the MOs optimized at the CAS(9,7) level, which are the desired ones, as initial guess for the subsequent CAS(19,12), it seems that the optimizer converges towards another local minimum, as can be seen from here:
+++++++++++++++
ITER. MIC NCI NEG ENERGY(VAR) ENERGY(PROJ) ENERGY CHANGE GRAD(0) GRAD(ORB) GRAD(CI) STEP TIME
1 40 177 0 -6582.87557885 -6582.87558251 -0.00000366 0.00081842 0.00012663 0.00005381 0.18D-01 6.14
2 50 156 0 -6582.87599307 -6582.87895778 -0.00296472 0.00794192 0.00010914 0.02988693 0.91D-01 13.23
3 50 182 0 -6582.88373013 -6582.88385640 -0.00012627 0.01894572 0.00000005 0.00182617 0.99D-02 19.38
4 23 90 0 -6582.88385665 -6582.88385666 0.00000000 0.00004359 0.00000096 0.00001591 0.50D-04 22.57
** WVFN **** CONVERGENCE REACHED, FINAL GRADIENT: 0.10D-05+++++++++++++++
The same occurs when rotating the MOs:
+++++++++++++++ ITER. MIC NCI NEG ENERGY(VAR) ENERGY(PROJ) ENERGY CHANGE GRAD(0) GRAD(ORB) GRAD(CI) STEP TIME
1 10 60 0 -6582.87555809 -6582.87560513 -0.00004704 0.00646904 0.00000020 0.00000008 0.45D+00 4.83
2 13 69 0 -6582.87568695 -6582.87545761 0.00022935 0.00115422 0.00000007 0.00000013 0.85D+00 10.00
3 13 68 0 -6582.87561557 -6582.87551002 0.00010555 0.00090474 0.00000005 0.00000014 0.60D+00 15.06
4 13 69 0 -6582.87561503 -6582.87587317 -0.00025814 0.00055427 0.00000004 0.00000007 0.32D+01 20.08
5 15 52 0 -6582.87931724 -6582.88786934 -0.00855210 0.00561534 0.00000052 0.00000097 0.29D+01 24.34
6 17 59 0 -6582.87701680 -6582.86990141 0.00711539 0.01343667 0.00000039 0.00000009 0.33D+01 29.07
7 12 57 0 -6582.88278094 -6582.88389759 -0.00111665 0.05034000 0.00000131 0.00000071 0.50D+00 33.61
8 10 57 0 -6582.88381208 -6582.88385534 -0.00004326 0.00751312 0.00000015 0.00000007 0.56D-01 38.10
9 10 47 0 -6582.88385651 -6582.88385663 -0.00000013 0.00064537 0.00000001 0.00000011 0.24D-02 42.21
10 9 39 0 -6582.88385665 -6582.88385665 0.00000000 0.00009879 0.00000001 0.00000007 0.14D-03 45.91
** WVFN **** CONVERGENCE REACHED, FINAL GRADIENT: 0.26D-04+++++++++++++++
But I think, according to the state-averaging, orbital and CI gradients of the first iteration etc., the initial guess is good enough and need not be changed as much as it is changed by the optimazer. Is there any way for fettering the freedom of the optimizer?
Sincerely yours,Davood.
From: Grant Hill <grant.hill at sheffield.ac.uk>
To: Davood Alizadeh <alizadeh_davood at yahoo.com>
Cc: "molpro-user at molpro.net" <molpro-user at molpro.net>
Sent: Wednesday, 14 January 2015, 20:06:32
Subject: Re: [molpro-user] SACASSCF in MOLPRO & ORCA
Dear Davood,
Have you tried rotating the orbitals to produce the active space you desire?
See the following manual page: http://www.molpro.net/info/2012.1/doc/manual/node240.html
Best wishes,
Grant
> On 13 Jan 2015, at 09:27, Davood Alizadeh <alizadeh_davood at yahoo.com> wrote:
>
>
> Dear Molpro experts,
>
> Please accept my apologies for my long message. I try to be as concise as possible.
>
> I`m trying to calculate the excitation energies (EEs) of a diatomic molecule (FeAg) using the Molpro and ORCA quantum chemistry packages. What I`m willing to obtain is "identical" excitation energies with these two codes. Despite the slight, at least in my opinion, differences in the SA-CASSCF optimization algorithms implemented in these two codes (e.g.,
> 1) CSF-based optimization of ORCA Vs determinant-based optimization of Molpro in SA-CASSCF calculations
> 2) non-linear coupled optimization of molpro (Werner) Vs uncoupled optimizations (Newton-Rouphson, DIIS, ...) of ORCA
> .
> .
> .)
>
> I`m employing identical basis sets, active spaces, inactive (closed) spaces, etc in these two programs. Thus, I think expecting identical results (either energy deference or absolute energies) might be totally warranted.
>
> The thing is this is the case "just" for a CAS space of (9e,7o) spanned by the atomic {molecular} representations (4s3d^Fe) and (5s^Ag) {sig(3), delta, and pi}, i.e., using this active space, these two codes give identical EEs and absolute energies. However, when the full valence space of (19e,12o), consisting of (4s3d^Fe) and (5s4d^Ag) {sig(4), delta (2), and pi(2)} which is a more adequate/realistic one, is used the Molpro results show considerably different EEs from those of ORCA. I need to confess that these differences which is about 2500 cm^-1 (at the worst case) cannot be considered "huge" but closer results are required for the system at hand.
>
> I think, basically, these discrepancies between EEs of the two codes is due to the following compositions of natural MOs (in irrep A_1) optimized at the CAS(19e,12o) level of theory (from 15.1 to 20.1 is active):
> Orb Occ Energy Coefficients
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>
> 1 1s 1 1s 1 1s 1 1s 1 1s 1 1s 1 1s 1 1s 1 1s 1 1s
> 1 1s 1 1s 1 1s 1 1s 1 2pz 1 2pz 1 2pz 1 2pz 1 2pz 1 2pz
> 1 2pz 1 2pz 1 2pz 1 2pz 1 3d0 1 3d2+ 1 3d0 1 3d2+ 1 3d0 1 3d2+
> 1 3d0 1 3d2+ 1 3d0 1 3d2+ 1 4f0 1 4f2+ 1 4f0 1 4f2+ 1 4f0 1 4f2+
> 1 5g0 1 5g4+ 1 5g2+ 2 1s 2 1s 2 1s 2 1s 2 1s 2 1s 2 1s
> 2 1s 2 1s 2 1s 2 1s 2 1s 2 2pz 2 2pz 2 2pz 2 2pz 2 2pz
> 2 2pz 2 2pz 2 2pz 2 2pz 2 3d0 2 3d2+ 2 3d0 2 3d2+ 2 3d0 2 3d2+
> 2 3d0 2 3d2+ 2 3d0 2 3d2+
> 14.1 2.00000 -0.502146 -0.001015 -0.000011 -0.000088 -0.000423 -0.001134 0.000337 0.004706 0.006334 -0.002624 -0.014412
> -0.011738 0.002893 0.090524 -0.033874 0.003529 0.000112 0.000271 -0.002368 -0.007818 -0.001503
> -0.007113 0.037722 0.036990 -0.018287 0.007173 0.000000 0.006388 0.000000 0.000198 0.000000
> 0.024728 0.000000 -0.004631 0.000000 0.000080 0.000000 0.000223 0.000000 0.002026 0.000000
> 0.000075 0.000000 0.000000 -0.000033 -0.000004 -0.000017 0.000296 0.000680 -0.000543 -0.001423
> 0.001182 -0.001467 0.004694 -0.002035 -0.035825 0.000509 0.000045 -0.000310 -0.002299 0.003527
> -0.005320 0.011943 -0.011928 0.043778 -0.253219 0.000000 0.010008 0.000000 0.370894 0.000000
> 0.493947 0.000000 0.310074 0.000000
>
> 15.1 2.00000 -0.498811 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
> 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
> 0.000000 0.000000 0.000000 0.000000 0.000000 0.000815 0.000000 -0.000031 0.000000 0.001877
> 0.000000 -0.003179 0.000000 0.013680 0.000000 0.000047 0.000000 -0.000177 0.000000 0.001064
> 0.000000 0.000000 0.000021 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
> 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
> 0.000000 0.000000 0.000000 0.000000 0.000000 -0.254068 0.000000 0.010335 0.000000 0.375990
> 0.000000 0.494604 0.000000 0.320878
>
> 16.1 1.99873 -2.707087 -0.000010 0.000000 -0.000001 -0.000009 -0.000011 -0.000013 0.000187 0.000148 0.000024 -0.001313
> 0.000525 -0.000108 0.001382 0.000143 0.325725 0.004979 0.020291 -0.245961 -0.476994 -0.390804
> -0.094062 0.000806 -0.001811 0.000502 0.000440 0.000000 0.000508 0.000000 -0.000169 0.000000
> 0.000107 0.000000 0.000089 0.000000 0.000020 0.000000 0.000058 0.000000 -0.000115 0.000000
> -0.000008 0.000000 0.000000 -0.000048 -0.000007 -0.000019 0.000164 0.000127 0.000039 -0.000312
> -0.000632 0.000584 0.000997 -0.000840 -0.001190 -0.000553 -0.000036 0.000598 0.001010 0.000225
> -0.001080 -0.002101 -0.000195 0.000712 -0.001950 0.000000 0.000078 0.000000 0.002893 0.000000
> 0.003809 0.000000 0.002121 0.000000
> ...
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> These compositions show that the optimized reference space is spanned by 15.1 (4d^Ag), 16.1 (2p^Fe) ... and the 14.1(4d^Ag) is turned out to be a core orbital whereas it is expected to have this orbital in the active space! The same happens for B_1 and B_2 irreps.
>
> This is my input by the way:
>
>
>
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> SET,DKROLL=1
>
> {rhf
> wf,73,4,5;
> occ,20,8,8,3;
> closed,17,7,7,2
> orbital,2100.2
> }
>
> {casscf;occ,20,8,8,3;closed,16,7,7,2
> { ITERATIONS;
> DO,internal,2,TO,20;
> }
> start,2100.2;
> WF,73,1,3;state,3
> WF,73,2,3;state,3
> WF,73,3,3;state,3
> WF,73,4,3;state,3
> WF,73,1,5;state,3
> WF,73,2,5;state,3
> WF,73,3,5;state,3
> WF,73,4,5;state,3
> expec2,lxx,lyy,lzz;
> natorb,ci,print;
> ORBITAL,2140.2;
> }
>
> {casscf;occ,20,8,8,3;closed,14,6,6,1
> { ITERATIONS;
> DO,internal,2,TO,20;
> }
> start,2140.2;
> WF,73,1,3;state,3
> WF,73,2,3;state,3
> WF,73,3,3;state,3
> WF,73,4,3;state,3
> WF,73,1,5;state,3
> WF,73,2,5;state,3
> WF,73,3,5;state,3
> WF,73,4,5;state,3
> expec2,lxx,lyy,lzz;
> natorb,ci,print;
> ORBITAL,2140.2;
> }
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>
>
> Could someone help me about this problem please?
>
>
>
>
> Thanks a lot in advance,
> sincerely yours.
> Davood.
>
>
>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20150114/4215927c/attachment-0001.html>
More information about the Molpro-user
mailing list