WALK-IMAGE_Restart

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
cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

WALK-IMAGE_Restart

Post by cymantren » 08 Aug 2020, 14:52

Hello everybody,

I tried to improve the results on WALk-IMAGE transition -state calculation, but it faliled.
The Manual says, that one simply add
...
*WALK
.RESTART
.IMAGE
...
I did two things:
the former calculation ----> WALK-IMAGE_TiB5_bipyr.out
a) Copying the former geometry to a new mol file- TiB5_bipyramid_1.mol --> WALK-IMAGE_TiB5_bipyr_1.out
b) Simply inserted the former DALTON.wlk file into the new directory, which is always produced by the dalton.-script and changed the dal-file.
---> WALK-IMAGE_restrt_TiB5_bipyr.out

Nothing really worked.

The b) approach was my idea and it worked, but it failed because of the wrong occupations. At least the new calculation can use the former DALTON.WLK file and can use it. Maybe I should try it without ".RESTART" ?

How one can improve the results of a former transition-state calculations by the WALK-IMAGE-method ?
Please tell me, how I can improve my results.

Thank you !

Best regards.

Alfred
Attachments
WALK-IMAGE_restrt_TiB5_bipyr.out
(82.56 KiB) Downloaded 22 times
WALK-IMAGE_TiB5_bipyr_1.out
(39.81 KiB) Downloaded 23 times
WALK-IMAGE_TiB5_bipyr.out
(1.65 MiB) Downloaded 19 times

taylor
Posts: 597
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: WALK-IMAGE_Restart

Post by taylor » 09 Aug 2020, 09:39

If you are using .WALK then the calculation is full second-order, calculating the Hessian exactly at each step. The only benefit I can see in restarting in such a case is if the wave function calculation is so expensive that saving the cost of that would be desirable. And then that benefit could be obtained by a "normal" restart, supplying the tar.gz file with the current wave function in it.

We use the image surface method quite a bit and we never use any restart option: if the calculation was heading in (what looks like) the right direction then we simply start a new walk from the last geometry step. Otherwise we may rethink our starting geometry. We have been using a new method in our group in Tianjin which has proved very effective (Field-Theodore and Taylor, JCTC, in press) and we hope that this method will be part of the next Dalton release.

Best regards
Pete

cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

Re: WALK-IMAGE_Restart

Post by cymantren » 10 Aug 2020, 08:52

Dear Prof.Taylor,

thank you very much for you quick reply.

A) I just wanted to try this option and, by the way, the restart option does'nt work, even for a simple scf-calculation. I remember, that many years ago, one had to supply the new calculation with the old SIRIUS.RST-file.
As a try, I did this with one of the test examples: energy_restart. But it did'nt work. --> Appendix

How one can do a restart on a similar simply calculation,e.g SCF,MP or MCSCF ?

B) Concerning my molecule, I wanted to improve the results, rather obtain results at all, because it did'nt converge. Yes, I know that it is always possible to use the not always last step of an optimization, in order to improve the calculation. I already did this in the ..TiB5_1-file. But the result was not very encouraging- it stops after few steps, telling me
--------------------------------------------------------------------
SEVERE ERROR, PROGRAM WILL BE ABORTED ---
Date and time (Linux) : Sat Aug 8 15:12:47 2020
......
Reason: FATAL ERROR: problems with automatic SCF occupation
.....................................................................
?????

Please tell me what kind of information is collected in the DALTON.WLK and why it is not sensibel to use it. Actually it should be advantageous having this file for the new calculation, because of the former calculated Hessian, which I think is a part of this file.

So I will don't use the restart option in the walk module, but will use the new geometry and a normal WALk-IMAGE calculation -- at least if the automatic occupation is correct.

A new calculation produces always a new directory, with a new process-number.
It is possible to make a restart, from the old directory ?


Looking forward to your answer.

Thank you very much.


Best regards

Alfred
Attachments
WALK-IMAGE_TiB5_bipyr_1.out
(39.81 KiB) Downloaded 20 times
energy_restart__1_H2O_ccpVDZ.out
(47.65 KiB) Downloaded 26 times
energy_restart__2_H2O_ccpVDZ.out
(25.42 KiB) Downloaded 19 times

taylor
Posts: 597
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: WALK-IMAGE_Restart

Post by taylor » 10 Aug 2020, 15:35

I accept your statement that RESTART does not work: it is certainly many years since I last used it, although it definitely worked at some point!

A failure of automatic SCF occupation implies that you are running with symmetry? As I have said before, this is unwise in transition-state optimizations because saddle points seem to dislike being symmetric! But if you want to run with symmetry it is straightforward to avoid at least the occupation error: simply specify the SCF occupation explicitly. The program will not deviate from that.

Unless the calculation of the Hessian is very expensive (which of course it can be!) I do not see what is the advantage of using the Hessian from the last step? It will only be used once and then discarded on a WALK. So you save one Hessian evaluation. Our experience with polyatomic molecules is that TS searches started from other stationary points can take 30 or more steps to converge, and in such a situation I would see saving one Hessian as a meagre benefit. Note that I am not saying one shouldn't do this, and moreover I agree that if the manual says it works, it should work! But even if it were working, I do not see much is gained with such a restart.

Best regards
Pete

cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

Re: WALK-IMAGE_Restart

Post by cymantren » 11 Aug 2020, 09:25

Dear Mr. Taylor,


you said, that transition states seems to dislike symmetry and of course I believe you, but it would be interesting to rationlize this.
For exmaple, the ground-state of H2O has C2v sym. One transition- state of it, e.g. H-O-H, would have Dinf.h sym. which has higher sym. than the ground state. The same is true for Ammonia NH3. It's transition-state has D3h sym., whereas itself is of C3v sym. I don't know, wheter these few examples are exception of the above mentioned rule, or transition states always have a higher sym., than their corresponding ground state. Nature seems to dislike the latter fact, but the Woddward-Hoffmann Rules has as a basis the conservation of the orbital-sym. I am fond of the JT-(Renner-Teller)-effect and this theorem seems uses the reduction of sym. This is however only valid, if there are degenerate orbitals, which are not equal occupied. Of course, this is not always the case.

However, I managed to get rid of the imaginary frequencies, by using "modfol, mod " and "INDEX 0" - as Mr. Jensen suggested - in the WALK-module and the molecule remains 2D, which suprised me a little bit, because the ground -state is of the same sym. as the transition state !? My expection was, that the sym. changes, maybe from 2D to 3D- but it did'nt.

Regarding to the TiB5-bibyramidal structure I nevertheless will turn the sym. off.


Finally, what do you mean with "RESTART does'nt work" ? Only in the case of using the WALK module or even in a simple scf-calc. ?


Kind regards

Alfred
Attachments
WALK_modfol-0_12_TiB5_631gdp_optim.ps
(6.75 KiB) Downloaded 24 times
WALK_modfol-0_12_TiB5_631gdp_optim.out
(322.29 KiB) Downloaded 23 times

cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

Re: WALK-IMAGE_Restart

Post by cymantren » 14 Aug 2020, 14:04

Hello Dalton-Developers and Users,

I had a deeper look at the manual and there it is described, how one can do a RESTART from a former calculation:

$DALTON15/dalton -t your-scratch-directory -put SIRIUS.RST file.dal file.mol &
AND
In the file.dal there mus be:
...
**WAVE FUNCTION
.RESTART
.HF
.MP2
.MCSCF
...
So it is not neccessary to quickly copy the Rst-file into the new directory.

It works.

Kindly regards

Alfred

taylor
Posts: 597
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: WALK-IMAGE_Restart

Post by taylor » 14 Aug 2020, 19:09

Sorry about the delay in replying, but I have been taking care of some other matters this week.

With respect to RESTART, all I had meant to imply was that I was not disagreeing with your statement that it did not work: I accepted that and merely said that at some point in the past it did work. Apparently you have it working now which is good to know, although I still do not understand how restarting a second-order walk, as opposed to starting from the last geometry step, confers any useful benefits.

It is certainly true that there are many transition states that have symmetry, such as the examples you give, or indeed (say) rotation around the CC bond in ethane. However, these are conformational transition states. I recognize that in your XY5 investigations this is what you are interested in, but most transition state searches are directed towards chemical reactions, and here I stand by my assertion that Nature seems to be less of a fan of symmetry than I am. As simple examples, the transition state for dissociation of C2H2 into two CH fragments does not have D\inftyh (nor C\inftyv) symmetry: it is actually C2v with a cis orientation of the CH bonds about the CC bond, as I recall. The TS for insertion of CH2 into H2 to give methane could certainly potentially have C2v symmetry, but this is not what is observed. In general, it seems that for chemical reactions it is the exception, rather than the rule, that symmetrical TSs are observed. And while I agree that e.g. the Woodward-Hoffmann rules are very successful, it is nevertheless the case that when detailed calculations are performed on the pathways for most reactions to which the rules are applied, the expected symmetry does not appear.

Best regards
Pete

cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

Re: WALK-IMAGE_Restart

Post by cymantren » 16 Aug 2020, 15:44

Dear Mr. Taylor,

I carefully read your statement and it is very interesting.
Everyone tries to find a pattern and TS may have or don't have any symmetry properties, it's up to oneself to find a pattern. You also mentioned that there a exmaples and counter-examples for the "TS <----> Symmetry/Nosym." - topic.
There are also other theories who try to explain the difficult behaviour of such systems, e.g. Woodward-Hoffmann and JT-distortion, but sometimes it is not easy to find a pattern or a rule. The energy of a TS is higher than that of the groundstate, but if it were vice versa - the energy of the deeper lying TS is not accessible for some reasons - and if some of the Conformational Transition States, are purely fiction, it could maybe happen, that the energy of this fictional TS, is deeper than that of the ground state ?

What do you think ?

You said that a really TS, which is a part of a chemical reaction, eg. NH +H2 --> NH-H2 ---> NH3
should have a specie, e.g. NH-H2. which is in princible observable, whereas that of a "conformational TS", is not ?! Did I understand you correctly ?

I think, the last species -NH-H2- could be followed by a WALK - IRC.

So, if one uses e.g WALK-IMAGE --> Appendix, then one is searching for a conformational TS, otherwise one has to use IRC ?

I found, that it is sometimes sensibel to start from the TS and use MOD, MODFOL, to find its groundstate. Of course, going directly via OPTIMIZE, would give the same, but I also try to get an idea of the PES. The form of the PES -a trajectory- is also possible by
-----------
....
*WALK
.DYNAMI
..
---------
It is possible to visualize the trajectories in DALTON ?

There are other programs who do this, but use a different approach - the wigner-distrb.
However, a basis for that is a VIBRATION- analysis.

Finally I just want to mention what I understand by the term " state".
I think a "state", is every possible subject relating to a electronic configuration, which "name" is governed by its constituent orbitals.
E.G.

DOCC 2 2 0 0
SOCC 0 1 1 0

So one would get a singlet, or a triplet A_2 state: (1,3)^ A_2.

Now, one has to find the ground-state and because of the rule of max. Multiplicity, this should be a triplet.

This is very known, but for me, the problem starts right here, if one is not sure about the configurations. Are they correct ?? What, if more groundstates for different geometries exist - What is the real one ?
Moreover, the above derivation was a bit theoretically - what's about Uncertaintity or active space by using the program etc. ?
----

Sometimes I rediscover the very simple and basic fact: the Schrödinger-equation could not be solved exactly.
The approximations can not be better, that the reality of a molecule ? At least, I think so.

Best and kindly regards

Alfred

cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

Re: WALK-IMAGE_Restart

Post by cymantren » 18 Aug 2020, 06:51

Hello everybody,


...of course " I want understand..... " is a typo, because stating "states" means: electronic states.

kindly regards

Alfred

taylor
Posts: 597
Joined: 15 Oct 2013, 05:37
First name(s): Peter
Middle name(s): Robert
Last name(s): Taylor
Affiliation: Tianjin University
Country: China

Re: WALK-IMAGE_Restart

Post by taylor » 22 Aug 2020, 13:42

With respect to the image surface method, it has been my experience over the years that it has an impressively large radius of convergence to the lowest-energy TS on a given surface. Even from quite weird starting geometries it seems to be able to do this. Now, in a very small system, like say NH + H2, it is very likely that the lowest energy TS is a reactive TS, perhaps to NH2 and H, say, which would be fine (note that since NH is a triplet ground state, formation of ammonia in its ground state would be spin-forbidden). But a tetratomic is so small that other than a possible rotational TS, like in H2O2, the appearance of conformational TSs is very unlikely. On the other hand, by the time we get to say 10-12 atoms there will likely be several conformational modes, and the TSs on these modes will almost certainly be the lowest-energy TSs on the surface. Unless we have a good guess at the TS for a reaction of interest, methods like the image surface will find one of these conformational TSs, which will then give us no information on the reaction of interest. This is why in our latest method we search for higher-order stationary points first, and then walk downhill, so that we eventually find (at least in principle) a TS that lies above the conformational TSs. We usually find that if we do this at a computational level that is adequate (e.g., a small active space MCSCF in a modest basis set) we can use the result to start directly a TS optimization at a higher level, such as CCSD(T) in a large basis.

When we have found our reactive (we hope...) TS we normally then use an IRC walk to find the minima it connects, as you suggest. The one downside to this is that the usual IRC implementations (not just the one in Dalton) can require a very small stepsize (and thus very many steps) to stay on the reaction path. We are currently looking at an approach to reduce this effort. I agree that it is possible to use e.g. mode-following (or in some cases gradient extremals) to walk from a TS to a minimum but it should be borne in mind that mode-following is not reaction-path following and it cannot be guaranteed formally that this approach will correctly locate the minima that the TS connects (although I know of no counterexamples from our own experiences).

I think it is possible to use other software to visualize the results of dynamic walks, but I have never done any so I don't know for sure. Bear in mind that such walks require some starting momenta for the nuclei, so this is not the same as a reaction path in the normal definition.

I am not sure I completely understand your comments about a TS being lower than the ground state? Do you mean a TS on an excited-state surface being lower at some geometries than the ground-state energy at that geometry? Something like this is observed in experiments and usually, thanks to nonadiabatic coupling, leads to nonradiative relaxation after crossing to the ground-state surface. On a single surface, say, the ground state, it may well happen that a TS lies below a local minimum somewhere on the surface (although it cannot directly connect to that minimum, of course!) but be above other minima. We are currently completing some work in which we observe such behaviour.

Best regards
Pete

cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

Re: WALK-IMAGE_Restart

Post by cymantren » 27 Aug 2020, 15:21

Dear Mr. Taylor,

it is due of my unexperience with other methods, that I only know that a saddle point - transition state - must have an imaginary frequency.
I don't know no other methods of proving that feature, but of course they may exist. The IMAGE-Method can have a similar feature by doing " .VIBANA .."
....
*WALK
.IMAGE
.INDEX
1
...
**PROPERTIES
.VIBANA

....

leads to structure which have one imag. vibration. This is correct, at least on B5 and surely for other molecules.

It was my idea - rather a question - if structures may exist, which have a transition-state, which lie not upper, but under a minimum. The book I read, concerning that feature, is " Quantenchemie Vol. 3" in a series of 5 Volumes, written and edited by Prof. Haberditzl and Prof.Zülicke et al.
In this book I found that usually a transition state is the point of the highest energy on the minimum path, but it may happen, that - as an exception - the above feature may appear. One must be carefully with the words "Saddle-point -- > Transition-state", because the former is always the latter, but not every latter is the former. This proves the fact, that a TS may lie under a minimum - it may even happen that this point is the point with the lowest energy.
This fact I don't really understand well and I will keep on the rule, not the exception.
The rule says, if a imaginary freq. appears, then it is a transition state. A transition state is then a saddle-point and there is line which connects the educts and products. This can be done by IRC.

However these things demand a more and deeper understanding, like you said in your example with "NH+H2". One has to obey different things at the same time and even if it should be spin--forbidden, it doesn't means that it could not happen, via a radiationless transition.

Concerning my B5-case, I don't consider educts and products -at least at the moment. Just a transition state -- saddle-point -- should lead to a minimum-structure, which I am looking for and which I already have found, even with better basis-sets. The mode-following-method is good and it leads to the (almost) same structure I found with the usual optimize-module. Of course, on a certain PES there may be more transitions states, but if one is the point with the least energy, things may become more difficult.
But of course, this calculation was only the starting point for further things e.g. - dynamical behaviour- and I noticed your suggestions.

Molecules are sometimes like humans - they do want they want, BUT always keeping on the laws, the laws of nature. There is no place for arbitrariness.


Thank you for your comments !


Best Regards

Alfred

cymantren
Posts: 31
Joined: 11 Jan 2014, 13:59
First name(s): Alfred
Last name(s): Güthler
Affiliation: privat
Country: Germany

Re: WALK-IMAGE_Restart

Post by cymantren » 27 Aug 2020, 18:07

Dear Reader and Administrator,

I beg you to ignore or even delete my last sentence, with " Molecules...".

The sense for humor is relative. It could be misleading and here -actually nowhere- is not the place to make such statements.
Simply delete my last sentence.

Thank you!


Best Regards

Alfred

Post Reply

Who is online

Users browsing this forum: No registered users and 10 guests