CCSD convergence

  • negoty
  • Topic Author
  • Visitor
  • Visitor
6 years 1 month ago - 6 years 1 month ago #636 by negoty
CCSD convergence was created by negoty
Dear users and developers,
I'm very new to using MRCC and already think it's quite cool. I came across an issue and couldn't find anything about it in the manual so far.
I am currently trying to compute CCSD(T) of (closed shell) [Fe(H2O)6]2+ and am encountering convergence issues in both the SCF and CCSD iterations. I was able to solve the SCF issues. Are there ways to speed up or help CCSD convergence along aside from increasing the number of iterations? It is increasingly slow and I am unsure why.

Thanks in advance!

Input:
Code:
# CCSD(T) calc=CCSD(T) mem=100000MB mult=1 charge=2 scfmaxit=50 scfext=15 scfdamp=0.7 scflshift=0.25 scfiguess=ao #qscf=AugHessL basis=atomtype Fe:cc-pwCVTZ O:cc-pCVDZ H:cc-pVDZ core=corr unit=angs geom=xyz 19 Fe 0.000000000 0.000000000 0.000000000 O 0.000000000 0.000000000 1.989595000 H -0.789066000 0.002150000 2.555337000 H 0.789066000 -0.002150000 2.555337000 O -0.036065000 1.989327000 0.000000000 H -0.007113000 2.554548000 0.788946000 H -0.007113000 2.554548000 -0.788946000 O -1.989619000 -0.009611000 0.000000000 H -2.547156000 0.785416000 0.000000000 H -2.563405000 -0.792705000 0.000000000 O 0.000000000 0.000000000 -1.989595000 H 0.789066000 -0.002150000 -2.555337000 H -0.789066000 0.002150000 -2.555337000 O 0.036065000 -1.989327000 0.000000000 H 0.007113000 -2.554548000 -0.788946000 H 0.007113000 -2.554548000 0.788946000 O 1.989619000 0.009611000 0.000000000 H 2.547156000 -0.785416000 0.000000000 H 2.563405000 0.792705000 0.000000000

Output snippet:
Code:
Starting CCSD iterations... Residual norm Energy Iteration 1 4.274088814384 -1719.550464222440 Iteration 2 6.484609056622 -1719.842522232327 Iteration 3 2.193209171674 -1720.033668644362 Iteration 4 1.980510710749 -1720.209594806344 Iteration 5 2.002652643973 -1720.345925538991 Iteration 6 1.239898116240 -1720.413527453012 Iteration 7 0.726006191715 -1720.461212885563 Iteration 8 0.565702744661 -1720.507344600944 Iteration 9 0.484284282073 -1720.543777687296 Iteration 10 0.371546782703 -1720.565270711646 Iteration 11 0.287244841891 -1720.578718665980 Iteration 12 0.245685775346 -1720.588653602694 Iteration 13 0.196006045394 -1720.595537426124 Iteration 14 0.147666442792 -1720.599459234976 Iteration 15 0.116515619156 -1720.601980166463 Iteration 16 0.101014030733 -1720.603678087391 Iteration 17 0.083979696661 -1720.604643284147 Iteration 18 0.066517805798 -1720.605170933782 Iteration 19 0.052176645444 -1720.605607293077 Iteration 20 0.044106502967 -1720.606030554397 Iteration 21 0.037077658969 -1720.606359059045 Iteration 22 0.030265135499 -1720.606527785491 Iteration 23 0.024271261951 -1720.606576605529 Iteration 24 0.020425249901 -1720.606588491923 Iteration 25 0.017028052329 -1720.606612637738 Iteration 26 0.013649564313 -1720.606645937079 Iteration 27 0.011383525498 -1720.606666868290 Iteration 28 0.009686951664 -1720.606677923058 Iteration 29 0.008139587728 -1720.606693398128 Iteration 30 0.006869743029 -1720.606716792576 Iteration 31 0.005807696774 -1720.606740604099 Iteration 32 0.004866826676 -1720.606754073281 Iteration 33 0.004171591320 -1720.606755069804 Iteration 34 0.003652780662 -1720.606748737499 Iteration 35 0.003078183214 -1720.606742609954 Iteration 36 0.002677137847 -1720.606744787171 Iteration 37 0.002440727451 -1720.606756613783 Iteration 38 0.002151469817 -1720.606768501056 Iteration 39 0.001986561378 -1720.606773601665 Iteration 40 0.001918707880 -1720.606774806203 Iteration 41 0.001929114539 -1720.606777488375 Iteration 42 0.001784663811 -1720.606782974503 Iteration 43 0.001730907738 -1720.606791820029
Last edit: 6 years 1 month ago by negoty.

Please Log in or Create an account to join the conversation.

  • kallay
  • Offline
  • Administrator
  • Administrator
  • Mihaly Kallay
More
6 years 1 month ago #637 by kallay
Replied by kallay on topic CCSD convergence
You should increase the maximum number of CC iterations, see keyword ccmaxit in the manual. But please be careful as your systems seems to be multireference.

Best regards,
Mihaly Kallay

Please Log in or Create an account to join the conversation.

  • kallay
  • Offline
  • Administrator
  • Administrator
  • Mihaly Kallay
More
6 years 1 month ago #639 by kallay
Replied by kallay on topic CCSD convergence
I have taken a closer look at the problem. The occupation changes over the iterations. You should manually set the occupation:
occ=15,6,7,14
or turn off symmetry:
symm=off

Best regards,
Mihaly Kallay

Please Log in or Create an account to join the conversation.

  • nagypeter
  • Offline
  • Premium Member
  • Premium Member
  • MRCC developer
More
6 years 1 month ago #640 by nagypeter
Replied by nagypeter on topic CCSD convergence
Hello!

I have done some tests too.
I used the same input file and the cc-pwCVTZ basis set from the EMSL database.
As suggested by Mihály, increasing the ccmaxit variable will help, the CCSD iteration converged for me in 79 iterations.

Reference energy [au]: -1718.212045286579
CCSD correlation energy [au]: -2.394753844090
CCSD(T) correlation energy [au]: -2.418205473490

At this point there seems to be an issue with the MP2 guess of the CCSD iteration, but the energies are correct (except for the MP2). That is why it needs abnormally large number of iterations. We will try to look at it.

I think you should check your structure. The Fe-O bond length in your input matches closely the [Fe(H2O)6]3+ value and it is quite far from the [Fe(H2O)6]2+ equilibrium distance. I have tried a longer Fe-O bond length value, but the slow convergence remained.

Is there any other case where this situation occurred for you?

Thank you,
Peter

Please Log in or Create an account to join the conversation.

  • nagypeter
  • Offline
  • Premium Member
  • Premium Member
  • MRCC developer
More
6 years 1 month ago - 6 years 1 month ago #641 by nagypeter
Replied by nagypeter on topic CCSD convergence
With the help of Mihály we have found the solution.

You should set only the following in your input:

calc=CCSD(T)
mem=7000MB
mult=1
charge=2
occ=15,6,7,14

basis=atomtype
Fe:cc-pwCVTZ
O:cc-pCVDZ
H:cc-pVDZ
core=corr

unit=angs
geom=xyz
...

The slow convergence in the HF was caused by the frequent changes in occupation. This can be fixed by setting occ, or turning off the spacial symmetry via symm=off, as suggested before. There is no need to change any other setting of the scf program, it converges in 22 steps nicely.
If you also set verbosity=3, you can follow the above in more details.

The problem in the MP2 and CCSD was originated from a bug that emerges only if damping is set in the input for the scf. So you should avoid damping, the bug will be fixed in the next release.
With this input the CCSD converges in 18 steps to the final result of:

Reference energy [au]: -1718.212045286700
MP2 energy [au]: -1720.570336329711
CCSD correlation energy [au]: -2.394753972428
CCSD(T) correlation energy [au]: -2.444087223663

I hope this solves all of your problems.
Thanks for pointing out this issue and for using our program,
Peter
Last edit: 6 years 1 month ago by nagypeter.

Please Log in or Create an account to join the conversation.

  • negoty
  • Topic Author
  • Visitor
  • Visitor
6 years 1 month ago - 6 years 1 month ago #642 by negoty
Replied by negoty on topic CCSD convergence
Hi and thanks a ton for your help! I'm currently on vacation but it seems that all my problems have been solved in my absence (which is awesome). I'll see to it as soon as I'm back.
Concerning the occupation numbers: I noticed that they change during the SCF but I wasn't sure how to determine the correct numbers. I'm not used to utilising symmetry so is there a way to know the numbers I should aim for?
Thanks again!
Last edit: 6 years 1 month ago by negoty.

Please Log in or Create an account to join the conversation.

Time to create page: 0.043 seconds
Powered by Kunena Forum