Hi,
I am running quite a lot of calculations concerning transition moments between excited states these days and every now and then I am experiencing problems with this. The problem is that sometimes during the solution of the linear equation systems the calculations just stop and don't do anything anymore. There is no crash and no error message and the calculations are still listed as running by the queueing system, however there is not I/O activity anymore. The calculaions are running on 6x12 CPUs and use the following input
**DALTON INPUT
.RUN RESPONSE
.DIRECT
**INTEGRALS
.NOSUP
.DIPLEN
.ANGMOM
**WAVE FUNCTIONS
.DFT
CAMB3LYP
**RESPONSE
.MAXRM
2000
*LINEAR
.SINGLE RESIDUE
.DIPMAG
.DIPLEN
.DIPVEL
*QUADRATIC
.DOUBLE RESIDUES
.DIPMAG
.DIPVEL
.ROOTS
15
*END OF INPUT
and the following molecule (it is large, I admit)
BASIS
augccpVDZ
C6CN0H2
Atomtypes=3 Nosymmetry
Charge=7.0 Atoms=1
N 4.41188898284932 6.49444969122519 4.96140905474819
Charge=6.0 Atoms=27
C 1.77006330871262 0.25944636705728 3.43125315284364
C 2.80345148083015 0.86921524218232 1.38665677877990
C 5.57483216245681 0.57722957984171 0.99296263538144
C 6.82494361063880 1.36227167595663 2.09375643850366
C 5.43759727382103 3.24732584251416 3.69891013465326
C 3.31438879815929 1.91748661421325 5.15210341768039
C 1.40278415842258 2.66033591510973 0.25412552008909
C 2.83291442734503 4.61842352975007 1.35586060377899
C 5.57741686615823 4.45481823244718 1.47462489492428
C 6.86162271086875 2.50371022542751 0.46605555109218
C 1.61791171082055 6.78141863230014 2.34468539546960
C 0.94464888078449 7.06967812100209 2.09853737654975
C 2.44482910453671 5.07091866226285 1.15583040605212
C 1.30109201301674 2.70516897472647 0.50884149666236
C 3.00222051591017 0.56935268476162 0.13289698652095
C 5.56369215645855 1.07329814407755 0.45330021931201
C 6.55213403033355 3.57020391872331 0.20365768100636
C 5.08434741560349 5.44835613781992 0.75740889551285
C 2.33246918089289 2.05206381497006 0.59320691413063
C 4.03578711880460 4.02023954033214 0.13219377513280
C 6.45321453374936 3.38733236258072 1.14042489881170
C 7.21598400165323 0.93318385631905 1.19078099517469
C 3.38170731277689 6.58094947402892 0.29509728731933
C 1.19018293219713 7.23030488280853 1.51922386821545
C 0.42467296117706 5.28527575728076 2.40375785592782
C 0.13345393960794 2.76907942781964 1.92904465568990
C 2.63436453862716 5.93444948667585 3.82022340647026
Charge=1.0 Atoms=17
H 4.12604412581761 0.74337806746685 6.67149548928406
H 6.72815987613464 4.19986282474858 5.00086363496860
H 1.86083452082312 8.82153925221195 2.64158155903457
H 8.84677401782978 1.56297618491705 1.80793186888900
H 2.76136858869137 8.27167765298395 3.16641579387480
H 8.89544550006357 2.35652417843996 0.68363849589997
H 6.56352969073982 5.93760097961701 2.49159400478269
H 0.18632529533873 0.08416678094206 3.92781450568165
H 0.71072647034186 9.18868065925883 1.87078243176265
H 5.87485698441622 7.29716822804330 1.15943978237162
H 4.68518149475709 8.03998783514680 0.31762716229977
H 8.52790359871620 3.89571238539074 0.64270093352284
H 7.70933855955356 4.89510908281696 1.73228274331912
H 9.11975021379412 0.44754156219738 1.77936245584791
H 1.11619981417322 1.33700098283579 2.67054392122060
H 2.09878638407135 3.30487355295295 6.08568952541859
H 4.62882584379568 4.73164082563988 2.48379491982007
This example has this problem while other calculations with the same input and a slightly different molecule (same size, different structure) went through. Is this an issue in the code or should I maybe contact our IT guys concerning this?
Transition moments between excited states: Calculations die

 Posts: 23
 Joined: 09 Mar 2015, 15:28
 First name(s): Daniel Henrik
 Last name(s): Friese
 Affiliation: Scientist
 Country: Norway

 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: Transition moments between excited states: Calculations
It is rare for Dalton to go into an infinite loop, at least in my experience, but that's certainly what this looks like. Does scancel (assuming you're SLURM) or qdel or whatever succeed in killing it? If so it does not seem to be a problem with the queueing system interfering, so I don't think the IT guys will be much help. (If you can't get the queueing system to terminate the job, then it might be a system issue.)
This is always risky with large calculations but you could try putting up print levels in the region where it seems to get stuck, or more tediously going into the code and inserting your own debugging output statements. Possibly also one of the response people have some thoughts?
Best regards
Pete
This is always risky with large calculations but you could try putting up print levels in the region where it seems to get stuck, or more tediously going into the code and inserting your own debugging output statements. Possibly also one of the response people have some thoughts?
Best regards
Pete

 Posts: 23
 Joined: 09 Mar 2015, 15:28
 First name(s): Daniel Henrik
 Last name(s): Friese
 Affiliation: Scientist
 Country: Norway
Re: Transition moments between excited states: Calculations
Hi Pete,
Thanks for your reply. I could always successfully kill the jobs when this occurred. But however it is quite difficult to reproduce the error as it happens sometimes and sometimes it does not.
But if I can try to restart the calculations with a higher printlevel for the double residues region.
Thanks for your reply. I could always successfully kill the jobs when this occurred. But however it is quite difficult to reproduce the error as it happens sometimes and sometimes it does not.
But if I can try to restart the calculations with a higher printlevel for the double residues region.
Re: Transition moments between excited states: Calculations
Please note that this is a much larger calculation than you think. Specifying 15 roots means 120 matrix elements and you are thus calculating 15 roots of an eigenvalue equation followed by 120 linear response equations (see JCP 82, 3235 (1985) Equation 6.13). This presents tremendous difficulties for a system like this. The algorithm will try to solve all these equations simultaneously if I am not mistaken, but it may very well happen that you don't see any output for a very long time. Most of these higher roots will be meaningless at this level of theory anyway, so you will be wasting a lot of computer time.
Regards,
Olav
Regards,
Olav

 Posts: 23
 Joined: 09 Mar 2015, 15:28
 First name(s): Daniel Henrik
 Last name(s): Friese
 Affiliation: Scientist
 Country: Norway
Re: Transition moments between excited states: Calculations
If I understood the output correctly, the linear response equations are not solved completely simultaneous. They are solved in batches e.g. for XDIPLEN for all frequencies, for YDIPLEN for all frequencies and for ZDIPLEN for all frequencies.
What do you actually mean by higher roots? I do not see why it should be a problem to determine 15 excitation energies.
Best regards,
Daniel
What do you actually mean by higher roots? I do not see why it should be a problem to determine 15 excitation energies.
Best regards,
Daniel
Re: Transition moments between excited states: Calculations
Yes you are right in that each operator is calculated separately, but each "batch" will still be 120 equations. Do you see these equations converging? After that is the contraction of the eigenvectors and the response vectors,
since you have 9 operators, you have over 1000 matrix elements, so if you want to estimate the time for that think in terms of 1000 independent SCF iterations.
You can of course calculate 15 excitation energies in principle, but what I am saying is that within the DFT approximation, the 15th root has most likely very little in common with the exact solution.
Olav
since you have 9 operators, you have over 1000 matrix elements, so if you want to estimate the time for that think in terms of 1000 independent SCF iterations.
You can of course calculate 15 excitation energies in principle, but what I am saying is that within the DFT approximation, the 15th root has most likely very little in common with the exact solution.
Olav

 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: Transition moments between excited states: Calculations
The first problem with DFT response is that, like SCF response, it is really only capable of describing well those states that are completely dominated by single excitations from the groundstate occupation. The more multipleexcitation character an excited state has, the worse it will be described by DFT response. I have posted before about stetrazine, where a state that in nature is at less than 5eV above the ground state is predicted to be more than 10eV by DFT response. It seems very doubtful that any system will have all 15 of the lowest excited states dominated by single excitations!
Second, Rydberg states. These are always a problem, and although using an augmented basis set is a help, it is not a cure. It is unlikely that for any (neutral) system all 15 of the lowest roots are valencelike, and if you cannot properly describe their Rydberg character (e.g., with a specialized basis for such states) you will not see the valenceRydberg mixing.
Best regards
Pete
Second, Rydberg states. These are always a problem, and although using an augmented basis set is a help, it is not a cure. It is unlikely that for any (neutral) system all 15 of the lowest roots are valencelike, and if you cannot properly describe their Rydberg character (e.g., with a specialized basis for such states) you will not see the valenceRydberg mixing.
Best regards
Pete

 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: Transition moments between excited states: Calculations
It occurs to me now that it is perhaps important to amplify the response to the question (paraphrasing) "why should there be a problem calculating 15 excitation energies"? Neither Olav nor I are making a comment about the operation of the program: given enough time and memory it can certainly calculate the necessary roots and solutions to the response equations. The point we are making is to what extent the 15 solutions eventually obtained can be connected with physical reality (see the comments about multiple excitations in excited states, Rydberg states, etc.). Yes, the program can generate them, but their connection to the "real world" will be progressively more tenuous as we calculate more roots.
Best regards
Pete
Best regards
Pete
Who is online
Users browsing this forum: No registered users and 5 guests