Problems with a CCSD calculation (oscillations in convergence)

[Evgen Gr]

Hello,

I have a CCSD calculation that exhibits oscillations in the energy and density during the CCSD iterations and cannot converge. I wonder what options are available to reach convergence in such a case? Can one get some improvement by changing the DIIS directive?

Best regards,

Evgeniy

[Peterson, Kirk]

Dear Evgeniy,

I’ve found that denominator shifts often are a big help in converging difficult ccsd calculations.  Try  shift,0.5,0.5  and go up from there.

best regards,

-Kirk

--
You received this message because you are subscribed to the Google Groups "molpro-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to molpro-user...@googlegroups.com.
To post to this group, send email to molpr...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/molpro-user/0f23e178-21e2-4740-9198-669954d58b7c%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Evgen Gr]

Dear Kirk,

Many thanks for your suggestion. Indeed, shifts and shiftp both with a value 0.9 have made CCSD converge!

Best regards,

Evgeniy

четверг, 11 октября 2018 г., 21:37:00 UTC+2 пользователь Kirk Peterson написал:

[marc.ri...@gmail.com]

Hello! I know this thread is a bit old, but maybe you can give me a hand. I have tried different values of shifts up to 3, and the calculation still does not seem to converge. It is a carbonate water system (CO3(2-)-H2O) at about 8 Angstrom of distance, with the total charge set to -2. Is there anything else that can be done to see if it converges? Otherwise I will take it as "impossible to converge". I have seen that for SCF one can change and tweak the DIIS, and it seems like those options are also there for CCSD. however, I am not sure what is safe to do and what is not.

Just in case, I will paste the input in here. The calculation is failing already for the avdz dimer calculation (run with MOLPRO 2020.2). Thanks!

------

memory,512,M

gthresh,zero=1.0e-16,twoint=3.0e-15,energy=1.0e-8,gradient=1.0e-6

gprint,orbitals

! Extrapolation from peterson's paper

! J. Chem. Phys. 131, 194105 (2009); http://dx.doi.org/10.1063/1.3265857

PROC EXTRAP

  A=(HF_HIGH - HF_LOW)/(EXP(-ALPHA*SQRT(X_HIGH)) - EXP(-ALPHA*SQRT(X_LOW)))

  E_HF_CBS=HF_LOW-A*EXP(-ALPHA*SQRT(X_LOW))

  B=(X_LOW/X_HIGH)^(-BETA)

  E_CCSD_CBS=(CCSD_LOW - B*CCSD_HIGH)/(1-B)

  C=(X_LOW/X_HIGH)^(-GAMMA)

  E_T_CBS=(T_LOW - C*T_HIGH)/(1-C)

ENDPROC

SYMMETRY,NOSYM

geomtyp=xyz

geometry={

O1 -7.14374406529000e-03 6.41852917239400e-02 -2.03334573306300e-02

H1 4.24679843748490e-01 -6.70494219029260e-01 -4.76448960656650e-01

H1 -3.11294167855020e-01 -3.48256259434400e-01 7.99182079538420e-01

C2 8.99609728114587e+00 4.36689428000000e-06 -4.49456827200000e-05

O2 8.89985735735297e+00 4.09640676830950e-01 1.23839531404131e+00

O2 9.29016482833988e+00 8.50545424950290e-01 -9.49292752277040e-01

O2 8.79807680807427e+00 -1.26018938015908e+00 -2.89068819493180e-01

}

basis={

default=avdz

}

CHARGE=-2,SPIN=0

{hf;start,atdens}

HF_LOW=ENERGY

{ccsd(t)-f12b,maxit=80,THRDEN=1.0e-9,THRVAR=1.0e-11;shift,1.0,1.0}

CCSD_LOW=ENERGC-ENERGR

T_LOW=ENERGY-ENERGC

e_AB_AB_t=energy

SYMMETRY,NOSYM

basis={

default=avtz

}

CHARGE=-2,SPIN=0

{hf;start,atdens}

HF_HIGH=ENERGY

{ccsd(t)-f12b,maxit=80,THRDEN=1.0e-9,THRVAR=1.0e-11;shift,1.0,1.0}

CCSD_HIGH=ENERGC-ENERGR

T_HIGH=ENERGY-ENERGC

e_AB_AB_q=energy

X_LOW=2

X_HIGH=3

ALPHA=4.30

BETA=2.483070

GAMMA=2.790300

! We run the extrapolation procedure

EXTRAP

! We obtain the Complete Basis Set energy

e_AB_AB_tq=E_HF_CBS+E_CCSD_CBS+E_T_CBS

geometry={

O1 -7.14374406529000e-03 6.41852917239400e-02 -2.03334573306300e-02

H1 4.24679843748490e-01 -6.70494219029260e-01 -4.76448960656650e-01

H1 -3.11294167855020e-01 -3.48256259434400e-01 7.99182079538420e-01

C2 8.99609728114587e+00 4.36689428000000e-06 -4.49456827200000e-05

O2 8.89985735735297e+00 4.09640676830950e-01 1.23839531404131e+00

O2 9.29016482833988e+00 8.50545424950290e-01 -9.49292752277040e-01

O2 8.79807680807427e+00 -1.26018938015908e+00 -2.89068819493180e-01

}

dummy,C2,O2

basis={

default=avdz

}

CHARGE=0,SPIN=0

{hf;start,atdens}

HF_LOW=ENERGY

{ccsd(t)-f12b,maxit=80,THRDEN=1.0e-9,THRVAR=1.0e-11;shift,1.0,1.0}

CCSD_LOW=ENERGC-ENERGR

T_LOW=ENERGY-ENERGC

e_A_A_t=energy

SYMMETRY,NOSYM

basis={

default=avtz

}

CHARGE=0,SPIN=0

{hf;start,atdens}

HF_HIGH=ENERGY

{ccsd(t)-f12b,maxit=80,THRDEN=1.0e-9,THRVAR=1.0e-11;shift,1.0,1.0}

CCSD_HIGH=ENERGC-ENERGR

T_HIGH=ENERGY-ENERGC

e_A_A_q=energy

X_LOW=2

X_HIGH=3

ALPHA=4.30

BETA=2.483070

GAMMA=2.790300

! We run the extrapolation procedure

EXTRAP

! We obtain the Complete Basis Set energy

e_A_A_tq=E_HF_CBS+E_CCSD_CBS+E_T_CBS

geometry={

O1 -7.14374406529000e-03 6.41852917239400e-02 -2.03334573306300e-02

H1 4.24679843748490e-01 -6.70494219029260e-01 -4.76448960656650e-01

H1 -3.11294167855020e-01 -3.48256259434400e-01 7.99182079538420e-01

C2 8.99609728114587e+00 4.36689428000000e-06 -4.49456827200000e-05

O2 8.89985735735297e+00 4.09640676830950e-01 1.23839531404131e+00

O2 9.29016482833988e+00 8.50545424950290e-01 -9.49292752277040e-01

O2 8.79807680807427e+00 -1.26018938015908e+00 -2.89068819493180e-01

}

dummy,O1,H1

basis={

default=avdz

}

CHARGE=-2,SPIN=0

{hf;start,atdens}

HF_LOW=ENERGY

{ccsd(t)-f12b,maxit=80,THRDEN=1.0e-9,THRVAR=1.0e-11;shift,1.0,1.0}

CCSD_LOW=ENERGC-ENERGR

T_LOW=ENERGY-ENERGC

e_B_B_t=energy

SYMMETRY,NOSYM

basis={

default=avtz

}

CHARGE=-2,SPIN=0

{hf;start,atdens}

HF_HIGH=ENERGY

{ccsd(t)-f12b,maxit=80,THRDEN=1.0e-9,THRVAR=1.0e-11;shift,1.0,1.0}

CCSD_HIGH=ENERGC-ENERGR

T_HIGH=ENERGY-ENERGC

e_B_B_q=energy

X_LOW=2

X_HIGH=3

ALPHA=4.30

BETA=2.483070

GAMMA=2.790300

! We run the extrapolation procedure

EXTRAP

! We obtain the Complete Basis Set energy

e_B_B_tq=E_HF_CBS+E_CCSD_CBS+E_T_CBS

IE_t=(e_AB_AB_t-e_A_A_t-e_B_B_t)*tokcal

IE_q=(e_AB_AB_q-e_A_A_q-e_B_B_q)*tokcal

IE_tq=(e_AB_AB_tq-e_A_A_tq-e_B_B_tq)*tokcal

~                                           

====

[Peterson, Kirk]

Hi Marc,

 

I don't know what's going on, but I can get it to converge in about 250 iterations :^)

 

regards,  -Kirk

To view this discussion on the web, visit https://groups.google.com/d/msgid/molpro-user/f531ae7e-a8ca-4fb6-9e19-331c594e5c8dn%40googlegroups.com.

[Marc Riera]

Wow. I will increase the max it to 300 then. I did not do it earlier bc if something does jot converge in less than 80 means that usially something is odd. 

Thanks Kirk!

[tibo...@gmail.com]

I think I read a paper a while ago that studied CC convergence, and showed that for a given CC problem there is usually a critical threshold of level shifts, below which the CC iterations never converge and may start to oscillate, whereas for sufficiently large shifts convergence is possible, but may be very slow.