1

##### Ridft, Rdgrad, Dscf, Grad / Re: Unrestricted DFT in H2 Dissociation

« Last post by**Wook**on

*March 27, 2017, 02:52:24 am*»

Thank you for your reply!

Wook

Wook

1

Thank you for your reply!

Wook

Wook

2

Of cause not.... You should assign to each job only 75% of the RAM that is availlable for that particular calculation.

If you run several calculation on one machine you need in advance decide how you want to split the total available RAM for the jobs.

If both have about the same size the best would be to give each of them half of the RAM --> set $maxcor to about 37.5% of the total RAM.

If you run several calculation on one machine you need in advance decide how you want to split the total available RAM for the jobs.

If both have about the same size the best would be to give each of them half of the RAM --> set $maxcor to about 37.5% of the total RAM.

3

I've not used Gaussian since more than 20 years.... No idea what 'guess=mix' in Gaussian does.

That symmetry has to be broken 'by hand' to get the correct UHF solution is a very special problem for higly symmetric molecules. Up to know I had this only for H2 which I use to teach students the bond breaking problem in HF and KS-DFT.

Christof

That symmetry has to be broken 'by hand' to get the correct UHF solution is a very special problem for higly symmetric molecules. Up to know I had this only for H2 which I use to teach students the bond breaking problem in HF and KS-DFT.

Christof

4

I hope you do not expect me to analyze the notation from Dreuw....

The implementation in TURBOMOLE is based on the original formulation by Schirmer and coworker. They used a strictly MP-based approximation which

leads to the occurence of second-order contributions in the expressions for transition moments. That might be interesting for academic purposes,

but for a model that could be used for routine (production) calculations it is for the performance/cost ratio unacceptable to include the O(N^6) scaling

terms second-order terms.

There identification is straightforward if one notes that a) the MP2 energy is computed from the first-order amplitudes and the b) the second-order

amplitudes are those computed in a MPPT program to compute the MP4 energy.

The detailed working expressions are given e.g. by Lunkenheimer in JCTC 9 (2013) 977.

The implementation in TURBOMOLE is based on the original formulation by Schirmer and coworker. They used a strictly MP-based approximation which

leads to the occurence of second-order contributions in the expressions for transition moments. That might be interesting for academic purposes,

but for a model that could be used for routine (production) calculations it is for the performance/cost ratio unacceptable to include the O(N^6) scaling

terms second-order terms.

There identification is straightforward if one notes that a) the MP2 energy is computed from the first-order amplitudes and the b) the second-order

amplitudes are those computed in a MPPT program to compute the MP4 energy.

The detailed working expressions are given e.g. by Lunkenheimer in JCTC 9 (2013) 977.

5

Thanks christof.haettig for answering me my question.

Thinking about the answer that you give me. I have another question.

If I try to set up two calculations at the same machine. Do I need to assign the 75% of RAM memory of that machine in both cases?

Thanks in advance

Thinking about the answer that you give me. I have another question.

If I try to set up two calculations at the same machine. Do I need to assign the 75% of RAM memory of that machine in both cases?

Thanks in advance

6

Dear Christof

Thank you very much for your reply.

I am still wondering whether there is a way to do what you said by a command in Turbomole, like 'guess=mix' in Gaussian.

Is it possible?

Thanks,

Wook

Thank you very much for your reply.

I am still wondering whether there is a way to do what you said by a command in Turbomole, like 'guess=mix' in Gaussian.

Is it possible?

Thanks,

Wook

7

Dear Users and Developers,

I would like to perform ADC(2) transition moment calculations with Turbomole and make comparisons with previous results and benchmarks obtained with other programs.

Could you, please, help to clarify the precise expression for the transition moments printed by Tubromole? The corresponding section of the manual (http://www.turbomole-gmbh.com/manuals/version_6_6/Documentation_html/DOKse43.html) discusses that some terms are neglected in the ground to excited state transition moments:

“...the implementation in the ricc2 program neglects in the calculation of the ground to excited state transition moments the contributions which are second order in ground state amplitudes (i.e. contain second-order amplitudes or products of first-order amplitudes)”

It is not clear to us, which contributions are included and which ones are neglected. Could you, please, explain which terms are computed in more detail, because we were not able to find a more detailed reference/publication covering the implementation in Turbomole. Could you point us to such documentation in the literature that we perhaps missed?

To assist the discussion I collected here the terms contributing to the ADC(2) transition densities according to the Eqs. (A1)-(A3) in the paper of Dreuw and co-workers (Mol. Phys., 2014, 112, 774-784):

“0th order:

\rho_{ai} = Y_{ia}

1st order:

\rho_{ia} = - \sum_{jb} t_{ij}^{ab} Y_{jb}

2nd order:

\rho_{ij} = - \sum_{a} \rho^{MP2}_{ia} Y_{ja} - \sum_{kab} Y_{ik}^{ab} t_{jk}^{ab}

\rho_{ia} = - \sum_{jb} Y_{jb} t^D_{ij}^{ab}

\rho_{ai} = 1/2 \sum_{jb} t_{ij}^{ab} \sum_{kc} t_{jk}^{bc} Y_{kc} - 1/2 \sum_{b} \rho^{MP2}_{ab} Y_{ib} + 1/2 \sum_{j} \rho^{MP2}_{ij} Y_{ja}

\rho_{ab} = \sum_{i} Y_{ia} \rho^{MP2}_{ib} + \sum_{ijc} Y_{ij}^{ac} t_{ij}^{bc},

where Y denotes the excited state eigenvectors, t_{ij}^{ab} is the MP2 amplitude, t^D_{ij}^{ab} is an O(N^6) scaled intermediate, and \rho^{MP2}_{pq} is the pq part of the MP2 density matrix.”

Could you explain your approximation using the above terminology?

An second source for discrepancies in comparisons could come from different approaches to normalize the ground/excited state wave function. Could you explain how do you normalize these functions in Tubromole?

Thank you very much for your help in advance!

Yours sincerely,

Dávid Mester

I would like to perform ADC(2) transition moment calculations with Turbomole and make comparisons with previous results and benchmarks obtained with other programs.

Could you, please, help to clarify the precise expression for the transition moments printed by Tubromole? The corresponding section of the manual (http://www.turbomole-gmbh.com/manuals/version_6_6/Documentation_html/DOKse43.html) discusses that some terms are neglected in the ground to excited state transition moments:

“...the implementation in the ricc2 program neglects in the calculation of the ground to excited state transition moments the contributions which are second order in ground state amplitudes (i.e. contain second-order amplitudes or products of first-order amplitudes)”

It is not clear to us, which contributions are included and which ones are neglected. Could you, please, explain which terms are computed in more detail, because we were not able to find a more detailed reference/publication covering the implementation in Turbomole. Could you point us to such documentation in the literature that we perhaps missed?

To assist the discussion I collected here the terms contributing to the ADC(2) transition densities according to the Eqs. (A1)-(A3) in the paper of Dreuw and co-workers (Mol. Phys., 2014, 112, 774-784):

“0th order:

\rho_{ai} = Y_{ia}

1st order:

\rho_{ia} = - \sum_{jb} t_{ij}^{ab} Y_{jb}

2nd order:

\rho_{ij} = - \sum_{a} \rho^{MP2}_{ia} Y_{ja} - \sum_{kab} Y_{ik}^{ab} t_{jk}^{ab}

\rho_{ia} = - \sum_{jb} Y_{jb} t^D_{ij}^{ab}

\rho_{ai} = 1/2 \sum_{jb} t_{ij}^{ab} \sum_{kc} t_{jk}^{bc} Y_{kc} - 1/2 \sum_{b} \rho^{MP2}_{ab} Y_{ib} + 1/2 \sum_{j} \rho^{MP2}_{ij} Y_{ja}

\rho_{ab} = \sum_{i} Y_{ia} \rho^{MP2}_{ib} + \sum_{ijc} Y_{ij}^{ac} t_{ij}^{bc},

where Y denotes the excited state eigenvectors, t_{ij}^{ab} is the MP2 amplitude, t^D_{ij}^{ab} is an O(N^6) scaled intermediate, and \rho^{MP2}_{pq} is the pq part of the MP2 density matrix.”

Could you explain your approximation using the above terminology?

An second source for discrepancies in comparisons could come from different approaches to normalize the ground/excited state wave function. Could you explain how do you normalize these functions in Tubromole?

Thank you very much for your help in advance!

Yours sincerely,

Dávid Mester

8

If the structure is "wrong" or not depends on what you want to do...

9

I'm not sure what you want to compute. But I have the impression that you tried to converge the SCF procedure to an excited state by asking the program to occupy for beta spin the 2nd orbital and leaving the lowest energy beta spin orbital empty. This is not supposed to work.

If you want to compute the correct (symmetry broken) spin unrestricted ground state for H2 1.) switch off symmetry and 2.) generate somehow

symmetry broken start MOs. You can do the latter by modifying by hand for one of the occupied orbitals the coefficients in the alpha or beta file

or you can do a prelimary calculation where you change the charge for one of the atoms a little bit.

For a scan of the potential curve start at large distances and then proceed to smaller distances. (Not the other way.)

If you want to compute the correct (symmetry broken) spin unrestricted ground state for H2 1.) switch off symmetry and 2.) generate somehow

symmetry broken start MOs. You can do the latter by modifying by hand for one of the occupied orbitals the coefficients in the alpha or beta file

or you can do a prelimary calculation where you change the charge for one of the atoms a little bit.

For a scan of the potential curve start at large distances and then proceed to smaller distances. (Not the other way.)

10

The comparison of total DFT energies between different codes is difficult, because the total energies are pretty sensitive to the integration grids

for the DFT functional and also sensitive to the details of XC functional. It would be better to compare energy differences.

for the DFT functional and also sensitive to the details of XC functional. It would be better to compare energy differences.