TDDFT: Too small a subspace error?

Clicked A Few Times

9:15:38 PM PST - Wed, Nov 13th 2019

I am trying to use the TDDFT module in NWChem 6.8 and after the SCF procedure I am getting an error that I don't quite understand. I looked through the previous posts and I found that someone else had a similar problem using TDDFT in NWChem 6.1. The suggestion was that perhaps the number of roots requested was too high - but from reading the post it was clear if this truly was the problem. Any help would be greatly appreciated.

Kristina

My input file:

memory total 16 Gb



 restart tddft_S2exc

 dft

 xc b3lyp

 maxiter 500

 direct

 end

 tddft

 cis

 nroots 50

 singlet

 triplet

 civecs

 end

 task tddft energy

And then the error that I receive at the end of the SCF procedure.

        Spin multiplicity :       1

   Number of AO functions :    8558

       Use of symmetry is : off

     Symmetry adaption is : off

        Schwarz screening :  0.10D-07

             XC Information

             --------------

                 B3LYP Method XC Potential

             Hartree-Fock (Exact) Exchange   0.20          

                Slater Exchange Functional   0.80 local    

            Becke 1988 Exchange Functional   0.72 non-local

      Lee-Yang-Parr Correlation Functional   0.81          

          VWN I RPA Correlation Functional   0.19 local

            TDDFT Information

            -----------------

         Calculation type : Tamm-Dancoff TDDFT

        Wavefunction type : Restricted singlets & triplets

         No. of electrons :    3680

          Alpha electrons :    1840

           Beta electrons :    1840

             No. of roots :      50

         Max subspacesize :   14000

           Max iterations :     100

              Target root :       1

          Target symmetry : none

     Symmetry restriction : off

                Algorithm : Optimal

       Davidson threshold :  0.10D-03

           Memory Information

           ------------------

         Available GA space size is        4221727932 doubles

         Available MA space size is        1072942916 doubles

         Length of a trial vector is     12361120

         Algorithm : Disk-based single tensor contraction

         Max subspace size reduced from     14000 to       33

Too small a subspace                  33

------------------------------------------------------------------------

------------------------------------------------------------------------

Too small a subspace                  33

------------------------------------------------------------------------

------------------------------------------------------------------------

 current input line : 

   22:   task tddft energy

------------------------------------------------------------------------

------------------------------------------------------------------------

------------------------------------------------------------------------

Forum Regular

7:15:12 AM PST - Thu, Nov 14th 2019
The problem in your case is that you did not allocate enough memory for the calculation to fit in memory so the code switched to the disk-based algorithm. This algorithm is limited to 10 GB of disk space, so the code reduced the maximum subspace to 33 vectors in order to be able to store the necessary quantities. Since you requested 50 roots, the code crashed because you can't solve for 50 roots with only 33 vectors in your subspace. You can increase your allocated GA memory and/or decrease the number of roots requested to get the calculation to run. I believe that you need about 5 times the amount of GA memory that is indicated in your input, so you can increase the GA memory by a factor of 5 or decrease the number of roots by the same factor.

Clicked A Few Times

9:23:38 PM PST - Thu, Nov 14th 2019
Thank you for the information! This is helpful! K

Clicked A Few Times

5:46:34 AM PST - Fri, Nov 15th 2019
Quote:Sean Nov 14th 2:15 pm The problem in your case is that you did not allocate enough memory for the calculation to fit in memory so the code switched to the disk-based algorithm. This algorithm is limited to 10 GB of disk space, so the code reduced the maximum subspace to 33 vectors in order to be able to store the necessary quantities. Since you requested 50 roots, the code crashed because you can't solve for 50 roots with only 33 vectors in your subspace. You can increase your allocated GA memory and/or decrease the number of roots requested to get the calculation to run. I believe that you need about 5 times the amount of GA memory that is indicated in your input, so you can increase the GA memory by a factor of 5 or decrease the number of roots by the same factor. I have another question on this topic. How can I change the number of roots in the restart file? I did change the value in the input file - but I tried to restart the job and it still reads the number of roots as 50? Is there a particular syntax that I have to use? Thank you, Kristina

Forum Regular

5:17:39 AM PST - Mon, Nov 18th 2019
I would just not run it as a restart. You can add vectors input <filename>.movecs to your DFT block so that you read your already calculated SCF solution. This gives you the benefit of a restart without having to worry about which parameters are being set from the previous calculation.

Forum >> NWChem's corner >> General Topics