Transition Dipole Moment from ADC2/DIIS Algorithm is different from ADC2/Davidson

Questions
#1

Dear Q-Chem users,

I ran ADC2 excited-state calculation (Q-Chem 5.3) for the same structure one time using the DIIS algorithm and another using the Davidson algorithm. The transition dipole moment I get from them is different by a significant magnitude (~0.7 Debye). Any thoughts about that?

The results are as follows:

1-ADC2/Davidson
Excitation energy: 2.300379 eV

Osc. strength: 1.295937
Trans. dip. moment [a.u.]: [ -4.795271, -0.000007, 0.000128]

2-ADC2/DIIS
Excitation energy: 2.300153 eV

Osc. strength: 1.345358
Trans. dip. moment [a.u.]: [ -5.122114, -0.000004, 0.010318]

Your thoughts are highly appreciated!!

Sincerely,
M.Sorour

#2

Could you please post an input file that shows this behavior?

#3

Yeah sure!!!

Here is the input for ADC2/Davidson

$rem
THREADS 16
METHOD adc(2)
BASIS def2-SVP
MEM_TOTAL 60000
MEM_STATIC 5000
ADC_PROP_ES true
ADC_PRINT 1
EE_SINGLETS 1
ADC_DAVIDSON_MAXITER 100
$end

Here is the input for ADC2/DIIS:

$rem
THREADS 16
METHOD adc(2)
BASIS def2-SVP
MEM_TOTAL 60000
MEM_STATIC 5000
ADC_DO_DIIS true
ADC_PROP_ES true
ADC_PRINT 1
EE_SINGLETS 1
ADC_DIIS_MAXITER 100
$end

#4

I am able to reproduce this with a simple molecule input. Unfortunately I don’t have sufficient understanding of the ADC theory to fully explain the behavior, but here’s a rough idea:

The Davidson solver solves for the full ADC(2) problem producing a set of singles and doubles ADC amplitudes.

The DIIS solver first finds the solution in the singles space, then uses DIIS to fold the ADC doubles contributions in a state-specific manner. The result is a vector of singles amplitudes that reproduces the same state energy as the full ADC(2) solution.

Hopefully someone with better knowledge of ADC will help us understand why the density matrices (and thus state properties) are different with the DIIS approach.

#5

Thank you so much for taking the time to try to help with this. I agree with you, it seems it has something to do with the density matrices.

I will share my findings here if I managed to get more insights into this issue.