CC3

Use this forum for general discussions of topics that don't fit in the other forums
Post Reply
mwb
Posts: 4
Joined: 29 Jul 2016, 16:06
First name(s): mirza
Middle name(s): wasif
Last name(s): baig
Affiliation: J Heyrovsky
Country: Czech Republic

CC3

Post by mwb » 01 Aug 2016, 18:55

For couple of months I am trying to do CC3 calculation on a organic molecule. But I am unable to fix this problem.

*******************************************************************************
*******************************************************************************
* *
* *
* START OF COUPLED CLUSTER CALCULATION *
* *
* *
*******************************************************************************
*******************************************************************************



CCR12 ANSATZ = 0

CCR12 APPROX = 0



*******************************************************************
* *
*---------- >---------*
*---------- OUTPUT FROM COUPLED CLUSTER ENERGY PROGRAM >---------*
*---------- >---------*
* *
*******************************************************************


The Direct Coupled Cluster Energy Program
-----------------------------------------


Number of t1 amplitudes : 78958
Number of t2 amplitudes : ********
Total number of amplitudes in ccsd : ********


--- SEVERE ERROR, PROGRAM WILL BE ABORTED ---
Date and time (Linux) : Thu Jul 28 16:47:06 2016
Host name : k10

Reason: Insufficient spaces in CCSD_ENERGY

Total CPU time used in DALTON: 37 hours 15 minutes 15 seconds
Total wall time used in DALTON: 48 hours 12 minutes 29 seconds


QTRACE dump of internal trace stack

========================
level module
========================
5 CCSD_ENERGY
4 CC_DRV
3 CC
2 DALTON
1 DALTON main
========================


Can any one kindly help me out. How to figure out this problem.

Wasif

kennethruud
Posts: 241
Joined: 27 Aug 2013, 16:42
First name(s): Kenneth
Last name(s): Ruud
Affiliation: UiT The Arctic University of Norway
Country: Norway

Re: CC3

Post by kennethruud » 01 Aug 2016, 19:06

Hi!

Without seeing the entire output file, it would be a bit of a guesswork as to what the problem is. However, entering into the guessing game, I note that the number of t2 amplitudes exceed what has been assumed by those implementing the code to be the range one can normally expect, my guess is that you molecule is quite large, possibly so large it is not realistic to do this calculation at the CCSD level, and even less so at the CC3 level of theory.

But it is just guessing based on the information provided. Could you either upload the output file or at least tell us the molecular composition and the basis set you are using?


Best regards,

Kenneth

mwb
Posts: 4
Joined: 29 Jul 2016, 16:06
First name(s): mirza
Middle name(s): wasif
Last name(s): baig
Affiliation: J Heyrovsky
Country: Czech Republic

CC3

Post by mwb » 01 Aug 2016, 19:14

I am using aug-cc-pVDZ basis set. i need this one only for CC3 calculation. I am attaching the whole output file.

I hope for your prompt reply.

Wasif
Attachments
cc3_laurdan_laurdan.out
(206.18 KiB) Downloaded 151 times

kennethruud
Posts: 241
Joined: 27 Aug 2013, 16:42
First name(s): Kenneth
Last name(s): Ruud
Affiliation: UiT The Arctic University of Norway
Country: Norway

Re: CC3

Post by kennethruud » 01 Aug 2016, 19:21

Thanks!

I think my guessing is correct: There is no way you are going to be able to run a CC3 calculation (and probably also not CCSD) with more than 900 basis functions with Dalton (and if at all successful, you would be waiting for a very long time).

I am also puzzled by your specification of the model as .SOPPA(CC3), which I believe is a non-existing model, so I am uncertain what you want to achieve. There are also other things I wonder about regarding your Dalton input file, but I think you just have to bite the bullet on this one and find a different computational model to calculate the properties you want for a molecule and basis set of the size you are interested in.


Best regards,

Kenneth

mwb
Posts: 4
Joined: 29 Jul 2016, 16:06
First name(s): mirza
Middle name(s): wasif
Last name(s): baig
Affiliation: J Heyrovsky
Country: Czech Republic

Re: CC3

Post by mwb » 01 Aug 2016, 22:41

For me time is not issue. But I want to do CC3 calculation with aug-cc-pVDZ basis set. If it is possible please let me know. I again say time is not issue. I have no other choice.

Kindly let me know about this.

Wasif

kennethruud
Posts: 241
Joined: 27 Aug 2013, 16:42
First name(s): Kenneth
Last name(s): Ruud
Affiliation: UiT The Arctic University of Norway
Country: Norway

Re: CC3

Post by kennethruud » 01 Aug 2016, 23:17

Hi!

I may have been somewhat diplomatic in my answer: I think there is no way you are going to get this calculation through at the CC3 level. CC3 is a N^7 scaling method. Thus, if it takes you a day to do a calculation with 100 basis functions, your calculation (which has about 10 times as many basis functions) will need 10 000 000 days to complete (i.e. 27000 years) (ok, it is not quite as bad, but we are talking a very long time here). Admittedly, I do not know how long a CC3 calculations with 100 basis functions will take for the properties you are interested, but even if we can do 200 basis functions in a day (which I think is optimistic), we are still talking about some 300 years. The only way around this (if at all solvable) is to reduce the scaling of the approach. Whereas much as been done on reduced-scaling CC methods in the literature as well as parallelization (but that can help you only so much compared to reducing the scaling), not much has been done in this respect yet in the context of Dalton, though developments are ongoing in LSDalton.

I do not know what you mean when you say have no other choice, because I believe you have to find another choice, because this is an unsolvable problem with present technology. The strategy would normally be to find smaller model systems related to your molecule of interest, benchmark more approximate methods against a high-level method such as CC3 to lend support to the quality of the more approximate methods, and then run the calculation of the interest with the approximate method, estimating the reliability of your results based on the benchmark studies done on the smaller system(s).

Thus, either explore lower-scaling CC implementations in other programs (but personally I do not think you will find a CC3 solution out there for the system you are interested in, but I may not be well informed about all developments, or attack this problem in a systematic manner by benchmarking and using more approximate methods.

Best regards,

Kenneth

taylor
Posts: 525
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: CC3

Post by taylor » 02 Aug 2016, 01:27

Kenneth has indeed been very diplomatic in his answers, even in the later one where he stated he may have been too diplomatic first time round... If you do not understand the scaling of computational work at different levels, you should read up on this before running anything but near-trivial calculations. CCSD, for example, is a method that scales as roughly N6 for N basis functions. This is obtained because the major work in the CCSD calculation involves steps that go as Nocc3Nvirt3 and Nocc2Nvirt4, where Nocc is the number of occupied orbitals and Nvirt is the number of unoccupied (virtual) orbitals. This N6 scaling is for an iterative calculation --- the work is done in each iteration of a method to obtain the optimized amplitudes of the singles and doubles.

If you now go from CCSD to CCSD(T), the triples step scales as Nocc3Nvirt4 but this is non-iterative: it is one-shot. But with CC3 it is now iterative and the time requirement will increase significantly. Note that very little advantage is possible from system size (that is, an extended system for which a number of integrals are zero by distance) in Dalton, although LSDalton is different here.

With all due respect, I suggest you read about using calibration calculations on small systems that have the sort of features your desired system has, in order to find a lower-scaling method that is practical. Saying "I do not care about the time" may be a reasonable statement if the calculation takes a week, or even a couple of months, but as Kenneth says not if it takes years!

Best regards
Pete

mwb
Posts: 4
Joined: 29 Jul 2016, 16:06
First name(s): mirza
Middle name(s): wasif
Last name(s): baig
Affiliation: J Heyrovsky
Country: Czech Republic

Re: CC3

Post by mwb » 02 Aug 2016, 01:36

I do not care about for time. I mean that I can wait for months like 3 to 4. Of course not for years. Actually in literature I found following paper:

http://scitation.aip.org/content/aip/jo ... /1.2889385

in which authors have performed CC3 calculations on some set of organic molecules.

My molecule of interest is of about 61 atoms. What I can do is I can reduce its atoms to 34 by replacing alkyl chain with H atom.

Due to some concerns we only need CC3 energy for its ground state and first 5 excited states.

Previously I tried with CFOUR but it was not successful. In paper mentioned above I found that CC3 method is implemented in Dalton.

So I intended to do with this. I haven't used DALTON before.

Well its nice to have prompt replies from its users.

Wasif

vitoversace
Posts: 3
Joined: 04 Sep 2014, 20:05
First name(s): vito
Last name(s): versace
Affiliation: Student
Country: Italy

Re: CC3

Post by vitoversace » 16 Aug 2016, 20:15

Dear all,

I am just a new user and would like to know whether it is
possible to freeze core and virtual orbitals in eom-CC2 or eom-CC3 calculations
along with use of Density Fitting to speed up calculation for closed and open shell molecules ?
I am looking for something similar to CASSCF but at CC level.

Best
Vito

taylor
Posts: 525
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: CC3

Post by taylor » 17 Aug 2016, 08:48

Dalton does only closed-shell CC (anything), although by using triplet response you can do various response calculations on triplet excited states. There is no "density fitting". Of any sort, that I am aware. A much more efficient procedure is in LSDalton, but I am not sure how much of that functionality (linear-scaling CC) is in the current release.

Best regards
Pete

Post Reply

Who is online

Users browsing this forum: No registered users and 5 guests