CIS failed RSPVEC convergence + negative polarizabilities?

Find answers or ask questions regarding Dalton calculations.
Please upload an output file showing the problem, if applicable.
(It is not necessary to upload input files, they can be found in the output file.)

Post Reply
m775b097
Posts: 18
Joined: 26 Apr 2016, 03:55
First name(s): Matthew
Last name(s): Barclay
Affiliation: University of Kansas
Country: United States

CIS failed RSPVEC convergence + negative polarizabilities?

Post by m775b097 » 26 May 2016, 21:45

Hello again,

So I am attempting to run Excited-State Polarizability calculations, and I came across some curious results.

I've attached three separate files....no wait I'm sorry I guess the forum isn't letting me attach multiple files?
Whatever - anyway the file I do have attached is for a calculation of H2 under CIS/aug-cc-pVDZ.

Assuming I am using the correct input for a calculation of an excited state polarizability term in CIS,

Code: Select all

**WAVE FUNCTIONS
.HF
**RESPONSE
.TDA
*CUBIC
.DOUBLE RESIDUE
.DIPLNZ
I come across two problematic results:
(1) the polarizability element is a negative value (-114.4)
(2) there are multiple warnings pointing out that the RSPVEC have not converged

While my questions will be the same regardless, I will also mention that the other two files I was planning to upload were of H2O at:
-- aug-cc-pVDZ (which also gave a negative polarizability element and had RSPVEC convergence failure)
-- and again at cc-pVDZ (which now gave a positive polarizability element but still had the same convergence troubles)

So my first question is....is the negative excited state polarizability correct? I thought that they could only be positive for the diagonal elements? Is there something I am doing wrong in my input?

And would you happen to know the reason why my RSPVECs are not converging like they should? I've tested this geometry and model chemistry on another QM package and it converges just fine, so is there some keyword I am omitting?

Thanks so much,
~Matt
Attachments
h2augccpvdz.out
(107.89 KiB) Downloaded 246 times

kennethruud
Posts: 252
Joined: 27 Aug 2013, 16:42
First name(s): Kenneth
Last name(s): Ruud
Affiliation: UiT The Arctic University of Norway
Country: Norway

Re: CIS failed RSPVEC convergence + negative polarizabilitie

Post by kennethruud » 27 May 2016, 06:33

Hi!

As regards the negative polarizability: Recall that what you get from the double residue of the cubic response function is not the excited-state polarizability but rather the difference between the excited-state polarizability and the ground-state polarizability. Thus, a negative number suggests your excited-state polarizability is smaller than that of ground state.

As regards the lack of convergence, it is difficult for me at least to give a clear-cut answer, but I note two things: The reason for lack of convergence is linear dependencies in the trial vectors. I note you run at a rather strength bond length (116 pm), for which it is not obvious that a single-reference method would be able to do particularly well. Consider doing a small CAS calculation instead (if indeed you have the geometry you actually want). Whether it would help to increase the basis set further and possibly using point group symmetry I am not sure, though the latter is always recommended unless there is some particular reason you do not want to use symmetry.


Best regards,

Kenneth

m775b097
Posts: 18
Joined: 26 Apr 2016, 03:55
First name(s): Matthew
Last name(s): Barclay
Affiliation: University of Kansas
Country: United States

Re: CIS failed RSPVEC convergence + negative polarizabilitie

Post by m775b097 » 27 May 2016, 19:10

Hi Dr. Ruud,

Thank you so much for clarifying the Polarizability issue - I'm afraid I'm still learning the finer points of the theory and so end up making some pretty ignorant mistakes, but that explains a great deal.

As for the lack of convergence....well I chose the geometry because I had previously run an excited-state optimization (albeit in DFT) which resulted in that bond length.
[for what it's worth, I just now ran the same optimization in CIS, yielding a bond length of 124pm, but that, too, failed to converge RSPVEC in the subsequent polarizability calculation]

Now in all honesty, I'm only using these small molecules as precursors to quickly ensure that I know what I'm doing before I expand these calculations to the larger conjugated organic molecules of interest (which is why I choose to stick with aug-cc-pVDZ).
And, truth told, I've already determined that CIS isn't accurate enough for my eventual purposes and thus have no real need to fix this issue (and so probably won't have time to go for the CAS calculation), but I am curious why it keeps picking up this error.

Indeed, I did attempt to incorporate symmetry (although for my eventual purposes I will require jobs that do not alter the molecular coordinates via symmetry), but I received an even more confusing error that:
XYVEC: Response label EXCITLAB with frequency ******* and symmetry 4 not found on file RSPVEC

-- for which I've attached the corresponding output file.

So...I suppose my final question (if you don't mind answering further) is... do you still think that the single-reference method is the cause of the convergence failure? Could it be anything else?

Thanks so much again for your time and help,
~Matt
Attachments
cispol_symh2.out
(106.23 KiB) Downloaded 267 times

taylor
Posts: 545
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: CIS failed RSPVEC convergence + negative polarizabilitie

Post by taylor » 27 May 2016, 20:11

It appears you have tried to get an excited-state bond-length for H2 from some sort of "CIS" (we do not use this terminology in Dalton so I am not sure why you use it, although I know what it means, and the results are really not very good for things like excited-state geometries). Which state are you trying to study? I ask because there are many interesting problems with excited states in H2 and virtually all of them are extremely multiconfigurational.

H2 is a terrible "example calculation" to use for testing out methods. He is perhaps worse, for reasons I will not list here, but while very small systems are undoubtedly useful for learning to use some features of a program, they are utterly unrepresentative of the problems one encounters with larger systems. As one unphysical "for instance" your LUMO (looking at your output from the calculation with symmetry) is a Rydberg orbital, in symmetry 5, which is an artefact of the basis set.

The equilibrium bond length in the H2 ground state is 74pm. A reasonable rule of thumb is that if you stretch that too much, your ground-state wave function will be useless. What does "too much" mean? If you're very lucky, maybe 50%, more likely less. You are stretching the bond by 60+%, and the chances that your ground-state wave function, which is simply SCF, is any good, are basically zero. You need as a minimum a two-configuration MCSCF.

Now, if you are looking at a more complicated molecule things may be a bit brighter. For example, if the excited-state geometry you optimize is not so very different from the ground-state geometry maybe a response calculation starting from the ground state will give you some useful results. But excited states are always a crapshoot... Your best bet is to try to find something, as a test system, that is not so small and which displays some features of your real system. For example, if you are interested in n -> pi* states in carbonyl systems like cholesterols, start with methanal and then try perhaps acrolein. Obviously, if your interest is in excited states of fullerenes, well, not straightforward to find a useful "model system"...

Best regards
Pete

kennethruud
Posts: 252
Joined: 27 Aug 2013, 16:42
First name(s): Kenneth
Last name(s): Ruud
Affiliation: UiT The Arctic University of Norway
Country: Norway

Re: CIS failed RSPVEC convergence + negative polarizabilitie

Post by kennethruud » 29 May 2016, 14:00

Hi, Matt!

Pete has already given you a lot of feedback, so just very shortly on my part:
- Note that many excited states may be dissociative, so optimizing them may actually lead to the molecule breaking apart, the fact that you still get a bonded molecule possibly a consequence of basis-set superposition errors or similar things.

The best way to test whether this is a general problem with the code or the method, is to make things simpler, and one way of for instance removing the issue of multiconfigurational nature of the wave function, would be to do the calculation at the geometry of the electronic ground state, which for sure is well behaved from a multiconfigurational point of view. Nevertheless, as Pete points out, H2 is weird for a multiple of reasons, some physical and some computational, and my first ever ab initio calculation was also done on H2 and failed catastrophically (read, the program terminated without completing the calculation), in that case because of variational freedom in the basis set.

As regards your comment on not wanting to use symmetry because you do not want to change the coordinates. It of course all depends on what you mean by this statement, but if your main problem is that the molecule is rotated etc. when the program tries to detect symmetry, this can be avoided by giving explicitly the symmetry generators yourself, see Chapter 23 and Section 26.2 in the manual.


Best regards,

Kenneth

Post Reply

Who is online

Users browsing this forum: Google [Bot] and 6 guests