Syntax to Split DFT Geometry Optimization from TDDFT Main Calculation

[Andrew Salij]

In reviewing the documentation for TDDFT and for DFT, it seems unclear how to perform a geometry optimization under one basis/functional combination and then do tddft with another functional/basis combination on that converged geometry without doing a second geometry optimization. Setting iterations to 1 or 0 in the second dft block causes convergence to fail. 

Sample input:

geometry autosym units angstrom

# Cartesian Coordinates in the form xyz

SAMPLE GEOMETRY

end

ecce_print ecce.out

#Aim to do a ground-state optimization in 3-21G, B3LYP

basis spherical

  * library "3-21g"

END

driver 

class

  default

end

dft

  mult 1    # spin multiplicity. 1=singlet, 3=triplet

  xc b3lyp

  mulliken  # Method to calculate charges

end

task dft optimize 

# Aim to do excited state calculations/enter Davidson iterations in 6-31G, Cam-BL3YP

basis spherical

   * library "6-31g"

end 

dft

  mult 1    # spin multiplicity. 1=singlet, 3=triplet

  XC xcamb88 1.00 lyp 0.81 vwn_5 0.19 hfexch 1.00

  cam 0.33 cam_alpha 0.19 cam_beta 0.46

  direct

  mulliken  # Method to calculate charges

end

tddft

  nroots 5

  cis

  civecs

end

task tddft energy

[Edoardo Aprà]

It's hard to guess what goes wrong in your case since you neither provide a full input file nor upload a full output file.

Anyhow, the attached input file does a dft optmization followed by a tddft calculation (with the same changes in xc functional and basis set you reported)

[Andrew Salij]

  Dr. Aprà  , I appreciate how fuller documentation can aid with things, and I've attached the resultant output from the input that you sent.

I should probably explain the question in further detail. Typically, I've been working in a single basis for TDDFT calculations, but the initial DFT convergence has been taking much longer than desired. The idea would be to split such a task into separate functionals for both the optimization and then the subsequent excited state calculations, only using the better basis for the latter as has been done in the literature (e.g., Computational Details in SI at c7qm00233e1.pdf (rsc.org)). 

When structuring sample codes as done above, I get a comparatively quick initial DFT optimization (line 468), porting of solution vectors into the next module, and then a subsequent DFT optimization (line 3090) in the new basis. This has a now longer total runtime that if the initial DFT optimization were omitted, obviating the purpose of the exercise. It is possible that I am completely misunderstanding how NWChem and TDDFT Calculations work, but can one just not go directly to the NWChem TDDFT Module (line 3625) with the current geometry? 

Sincerely, Andrew Salij 

[Andrew Salij]

Upon further reflection, I have realized that the question posed is rather unsound as it entails separating the functionals between ground state energy calculation and excited state calculation. The thought was to get around this from the documentation:

"The TDDFT module first invokes DFT module for a ground-state calculation (regardless of whether the calculations uses a HF reference as in CIS or TDHF or a DFT functional), and hence there is no need to perform a separate ground-state DFT calculation prior to calling a TDDFT task. "  

Naturally, an initial DFT ground state energy calculation in the basis chosen is a requisite for future calculations and one can't just get around this.