Exit code 2 (bad value during floating point read)

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
kaushikhatua
Posts: 35
Joined: 10 Jan 2014, 07:05
First name(s): kaushik
Last name(s): hatua
Affiliation: IIEST India
Country: India

Exit code 2 (bad value during floating point read)

Post by kaushikhatua » 25 Feb 2016, 08:03

Hi Dalton users
During DFT cubic response calculation I found a problem which exits with exit code 2 usually lies in the 610 line of rspcr6.F module. I think I have to change the value F20.3 to somehow larger value and recompiling. By the way I must say that the system is multiconfigurational and DFT is not suitable however MCSCF (8,8) also have same problem. It is a second hyperpolarizability calculation at basis set 6-311++G** and the problem arise at YYYY component. Any help would be appreciable

arnfinn
Posts: 231
Joined: 28 Aug 2013, 08:02
First name(s): Arnfinn
Middle name(s): Hykkerud
Last name(s): Steindal
Affiliation: UiT
Country: Norway
Location: UiT The Arctic University of Norway

Re: Exit code 2 (bad value during floating point read)

Post by arnfinn » 25 Feb 2016, 09:11

Please attach your output file (and maybe also your stdout and stderr) to your post. Makes it much easier to help you.

kaushikhatua
Posts: 35
Joined: 10 Jan 2014, 07:05
First name(s): kaushik
Last name(s): hatua
Affiliation: IIEST India
Country: India

Re: Exit code 2 (bad value during floating point read)

Post by kaushikhatua » 25 Feb 2016, 18:24

Please find the attached file. Terminal has been closed due to power cut.
Attachments
new file.txt
(17.46 KiB) Downloaded 178 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: Exit code 2 (bad value during floating point read)

Post by taylor » 25 Feb 2016, 22:28

You were already asked to post your output. I am not sure what "terminal closed" means in your message, but you have not posted your output, just some piece of it, and not the stdout/stderr files you were asked to post either. As I have posted many times before, while variously talented, the Dalton developers are not telepathic. If you do not post your output so that we know what you are trying to do, it is difficult to see how to help you.

That said, it is patently obvious that you are getting read errors because you are generating ludicrously large numbers in your calculation. The fields of asterisks in your output confirm this: probably the program is trying to read a data item that has overflowed the output field when it was written. In your original posting you wrote about editing a format F20.3 and recompiling. In my copy of the code I can find no F20.3 format statement at all, but in any event F20.3 allows a positive number up to 10^16 or a negative of -10^15. In terms of significant digits, such numbers are actually beyond machine accuracy (14 digits). If your calculation is really generating numbers in the range of 10^15, I would strongly suspect you have a problem in your calculations, not with the program. A look at your output shows alarmingly large numbers even when the fields are not all asterisks. You appear (this is not entirely clear because you did not post your output, as you were asked to do...) to be running static properties, so the issue of near-resonances in a frequency-dependent calculation does not seem to be the problem, but in any event the numbers the program is printing suggest something(s) odd and I am very much of the opinion that the problem is with your calculation, not the program.

But we can't decide, nor can we help, if you don't post your output! In full...

Best regards
Pete

kaushikhatua
Posts: 35
Joined: 10 Jan 2014, 07:05
First name(s): kaushik
Last name(s): hatua
Affiliation: IIEST India
Country: India

Re: Exit code 2 (bad value during floating point read)

Post by kaushikhatua » 26 Feb 2016, 04:48

Obviously Dalton is a splendid suite. May be I am missing something. Output file is attached.
Attachments
gama-bhhlyp.txt
(63.84 KiB) Downloaded 178 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: Exit code 2 (bad value during floating point read)

Post by taylor » 26 Feb 2016, 06:57

I still do not see what your actual error is because this is in stdout/stderr, which you did not post. But I repeat what I posted earlier: I think your problem is with the molecule. The code produces ludicrously large values for properties, is it in fact sensible that a hyperpolarizability is more than 2,000,000? And why are you trying to extract 20 roots of the response equations? Somewhere between about 10 and 19 of these will be garbage. I would start by trying another functional, or to avoid issues with DFT try Hartree-Fock, and see whether you get the same error.

Incidentally, many years ago at NASA Ames we were calculating Be/C compounds and had terrible trouble with (near-)symmetry-breaking in the wave function. This seems to be a problem endemic in such compounds. If you use Dalton to calculate the Cioslowski populations (by-product of frequency calculations and intensities) in such cases they are always unphysical, sometimes wildly so.

Best regards
Pete

lyzhao
Posts: 74
Joined: 11 Nov 2013, 00:36
First name(s): Youzhao
Last name(s): Lan
Affiliation: Institute of Physical Chemistry
Country: China

Re: Exit code 2 (bad value during floating point read)

Post by lyzhao » 26 Feb 2016, 08:05

Dear kau,
If I understand your question correctly, please do the following, then recompile the dalton.

in file \DALTON\rsp\rspcr6.F
change WRITE (LUPRI,'(A,F20.12)') 'Contribution from C4FOCK:' to WRITE (LUPRI,'(A,F30.12)') 'Contribution from C4FOCK:'
change WRITE(LUPRI,'(/A,F20.8)') ' << A; B, C, D >> = ', CUBVAL to WRITE(LUPRI,'(/A,F30.8)') ' << A; B, C, D >> = ', CUBVAL

in file \DALTON\rsp\rspcr3.F
change WRITE(LUPRI,'(/2(A),F16.6)') '@ Averaged gamma parallel ', to WRITE(LUPRI,'(/2(A),F30.8)') '@ Averaged gamma parallel ',

Good luck
Lan

kaushikhatua
Posts: 35
Joined: 10 Jan 2014, 07:05
First name(s): kaushik
Last name(s): hatua
Affiliation: IIEST India
Country: India

Re: Exit code 2 (bad value during floating point read)

Post by kaushikhatua » 26 Feb 2016, 08:27

Please see the the error log. I will shortly try lyzhao's suggestion. I repeat Dalton is a superb program, no doubt. By the way, same calculation when carried out at CAS(8,8) converge smoothly, but obviously with a huge magnitude. However I always prefer CCSD which I will do later.
Attachments
error-log.txt
(3.17 KiB) Downloaded 178 times

lyzhao
Posts: 74
Joined: 11 Nov 2013, 00:36
First name(s): Youzhao
Last name(s): Lan
Affiliation: Institute of Physical Chemistry
Country: China

Re: Exit code 2 (bad value during floating point read)

Post by lyzhao » 26 Feb 2016, 08:49

a huge magnitude possibly means you have met a resonance.

Lan

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: Exit code 2 (bad value during floating point read)

Post by taylor » 26 Feb 2016, 09:43

I do not understand the need/urge to rebuild the program with a different format for the floating-point read. What is confusing about the simple observation that results in these ranges must be total garbage? The numbers are completely unphysical! They make no sense. Is getting a number like 100,000,000 for a hyperpolarizability any more useful/meaningful than *********.*? I do not see how the calculation is "at a resonance" since it is a static property, or? What is most likely happening is that the wave function for the system, which will necessarily display the high symmetry of the molecule even if the user specifies "Nosymmetry", is close to a critical point at which the wave function needs/wants to break the spatial symmetry, for example by localizing certain orbitals or charges. I have already mentioned that this is a known problem in very electron-deficient species like beryllium or boron.

Changing the program will not fix a problem that is inherent in Nature. It just wastes more computer time rebuilding the program, and then more when the job is rerun to produce garbage with digits in it rather than asterisks. The project is broken, not the program. I do not fault Lan for providing a patch, but if indeed the calculation is flawed, changing the program will not get rid of the garbage!

I repeat my earlier suggestion: calculate the Cioslowski populations. If the results look weird, like Be with a charge of 4+ or C with a charge of 2-, or anything else that flies in the face of common sense, then you already have the answer. The wave function is at a point of incipient symmetry-breaking. Another test would be to slightly distort the geometry.

Best regards
Pete

kaushikhatua
Posts: 35
Joined: 10 Jan 2014, 07:05
First name(s): kaushik
Last name(s): hatua
Affiliation: IIEST India
Country: India

Re: Exit code 2 (bad value during floating point read)

Post by kaushikhatua » 27 Feb 2016, 16:00

Hi
Sorry for the late reply. Apart from rebuilding the program, I did population analysis as Peter suggest. Be has 0.44 a.u charge in NBO of g09 at HF density with C-Be bond of occupancy 1.7867.
On the other hand GAPT charge in Dalton found -0.151 a.u at HF level and 0.088 a.u at BHHLYP. By the way YYYY component which troubles me comes out as 10*8 order at CAS(8,8) level and 10*7 at HF level. I think I should discard DFT from my calculation because of high degree of multiconfigurational nature of the wave function (C0=0.33) and go for post HF. I am sure Peter have definitely some answer.

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: Exit code 2 (bad value during floating point read)

Post by taylor » 28 Feb 2016, 10:59

Could I ask you to please post your outputs! The issue seems to have migrated from a problem with precision in a Dalton format statement to an issue with the multiconfigurational nature of the wave function. What, specifically, for instance, is "C0" here? You talk about CAS(8,8): are you telling us that the coefficient for the leading term in the wave function is 0.33? That is, that the weight (square of the amplitude) of this configuration is only 11% of the final wave function? If so, I would wonder whether you are actually looking at the ground state of the system. I cannot think of anything, other than perhaps the notorious ground state of Cr2, where numbers like this are seen. In such a case you are right: the DFT methods will yield nothing of value. Except perhaps fortuitously...

I cannot comment on the NBO numbers from Another Program, partly because this risks violating your licence for the Other Program (please read the licence terms before posting results from it --- I don't care since I don't have a licence for that program, but you or your group/department/university might care...), and partly because I have zero faith in NBO analyses anyway. But if you ran the Cioslowski populations for your system with Dalton, you probably ran vibrational frequencies as well (if you are calculating the dipole gradients the only other term needed for frequencies is the integral second derivatives and for this calculation this should be close to trivial), and these would also be of considerable interest: are they physically sensible? You only give the Be population, which is close enough to zero that I am already suspicious...

I am happy to keep thinking about all this, but not without a program output. Excerpts just lead to messages back and forth "what about this? what about that?".
Best regards
Pete

kaushikhatua
Posts: 35
Joined: 10 Jan 2014, 07:05
First name(s): kaushik
Last name(s): hatua
Affiliation: IIEST India
Country: India

Re: Exit code 2 (bad value during floating point read)

Post by kaushikhatua » 29 Feb 2016, 02:40

Sorry for the NBO and G09. By the way I have attached the files. Could you please point the odd, it troubles me lot.
Attachments
CAS88.txt
(112 KiB) Downloaded 181 times
BHHLYP-GAPT.txt
(58.48 KiB) Downloaded 178 times
HF-GAPT.txt
(56.96 KiB) Downloaded 190 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: Exit code 2 (bad value during floating point read)

Post by taylor » 29 Feb 2016, 14:16

There is undoubtedly some sort of pathological problem here and I am still convinced it is with this particular molecule, not with the program. I say this because I observe that in your calculation that generates the Cioslowski populations, you also calculate the static dipole polarizability, and if you look at the listing, the component alpha_{yy} is negative! This occurs for both HF and BHLYP. This must surely be completely unphysical? It implies that if the molecule is subjected to a static homogeneous field in the y direction, the resulting induced dipole will have the opposite sign from that expected from electrostatics. Alternatively, it implies that if the molecule is placed in such a field the energy will go up.

The polarizability is being calculated here in the routines from the ABACUS properties module the code in RESPONSE. If indeed there is an issue with the program, it seems there must be problems with the polarizability calculation in ABACUS and also, given the issues with the hyperpolarizability, the cubic response routines in RESPONSE, which seems to me to be possible but really unlikely. Certainly the polarizability code in ABACUS has been in use for well over 25 years and it is difficult for me to see how a bug that gives a negative polarizability component (in the principal axis system) could hide for that length of time.

So I return to my earlier contention: there is something pathological about this molecule. I looked also at your CAS(8,8) output and while I agree there is considerable multiconfigurational character I think you have misread the output. The leading coefficient in the wave function is not necessarily the first value printed in the list: this depends on how the program constructs the configuration list. And you can see that the second listed configuration actually has a coefficient of 0.88, or a weight of about 75%. As I say, this is multiconfigurational, but not as extreme as your earlier concerns. That said, I would not like to be relying on a single-configuration method here, at least not SCF or DFT.

What to explore next is a rather open question. Have you tried calculating the hyperpolarizability at the CAS level? What is the situation with the beta hyperpolarizability, does that also show problems? Your gamma values and the contributions from the various terms seem to me to be wildly implausible. Now it could be, still, that there is some incipient symmetry-breaking in the wave function. I agree that the Cioslowski populations look plausible, but it is clear your input gives the molecule exact C2v symmetry and possibly distorting this so that the symmetry reduces to Cs (i.e., in the molecular plane) would show whether symmetry-breaking is an issue. Of course, simply putting "Nosymmetry" in the input does not affect this because the geometry still has C2v symmetry: the only way to test it is to force the geometry to a lower symmetry. At what level was the geometry optimized? It is also possible (although I think it is unlikely) that it is, say, this combination of methods and basis set that is at the root of the problem: easily tested by running with a different set.

Finally, having mentioned basis sets, I would regard 6-311++G** as unlikely to give good (hyper)polarizabilities with any method. It is not valence triple-zeta in quality to start with, and although the diffuse s and p may help, diffuse higher angular functions will be needed for good results. A better choice, still at a modest level, might be aug-cc-pVDZ. Further, BHLYP cannot possibly give useful higher polarizabilities, other than purely fortuitously. It was developed early on when focus was on good energies and some of the right physics at smaller interelectronic distances, and this has little relation to the delicate perturbation of the outermost part of the electron density needed for these electric properties. I am not an expert on DFT but possibly others may be able to make suggestions here.

Best regards
Pete
P.S. You should update to the latest version of Dalton. We cannot maintain support for older versions, although I do not think the Dalton version is responsible for any of your problems here.

kaushikhatua
Posts: 35
Joined: 10 Jan 2014, 07:05
First name(s): kaushik
Last name(s): hatua
Affiliation: IIEST India
Country: India

Re: Exit code 2 (bad value during floating point read)

Post by kaushikhatua » 29 Feb 2016, 15:33

Well, it's a real problem here. Let my try this thing
1. Improving the basis set
2. better correlation, CC2 ,CCSD
3. Behavior if other tensor beta
3. little distorted geometry

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: Exit code 2 (bad value during floating point read)

Post by taylor » 01 Mar 2016, 11:41

I have played a bit with this myself and have drawn a few conclusions: I posted here rather than PM'ing because the findings may be useful to other users. Briefly, although perhaps the choice of the word "resonance" was not quite right, Lan was heading in the right direction with his suggestion, I think. Note that in what I describe here I use C2v notation, and use the Mulliken convention that B2 is symmetric with respect to the plane of the molecule (when the system is planar). Your y-direction is thus B2, for example. I repeat, this is conventional: the program may or may not adhere to this convention if you run with symmetry, so don't be misled by the labels in the program. Your y is B2 here.

The ground state is 1A1 and as you have already noted is rather multiconfigurational (crudely: 75% a12 and 25% b22, where I have just listed the symmetry of the orbitals that show considerable differences from 2 or 0 in occupation). But the ground state seems well behaved and as you checked the Cioslowski populations are reasonable (and all the harmonic frequencies are real at the C2v geometry and for me all polarizability components are positive). Given the two orbitals a1 and b2 noted above we would also expect low-lying, or at least not very high-lying 3B2 and 1B2 states derived from the occupation a11b21. The triplet is not relevant here for your calculations. But the 1B2 state turns out to be quite unusual. Although all frequencies are real and reasonable, the Cioslowski populations and the yy polarizability are not reasonable. The latter is negative, and the populations are absurd (values of -2.3 for C, for example). Consistent with the latter, the IR intensities are physically implausible too. The 1B2 state seems to be very close in energy to the ground state.

If we now think of sum-over-states expression for the (hyper)polarizability --- this is of course not how the response program works internally but we can view things this way --- this low-lying 1B2 state will have (see the above comments about dipole gradients etc.) very large transition dipole matrix elements in the sum-over-states numerators and very small energy differences in the denominators. I suspect this near-degeneracy and strong coupling via the dipole operator (both of which make me suspicious still that the wave function might localize given the chance, but I can't spend all my time on this!) are what is giving the odd results you see for the ground-state properties.

I am afraid I don't have a good suggestion as to how to tackle this, but I will be interested to hear of your experiences. I fear the results will depend very strongly both on the basis set and the correlation treatment.

Best regards
Pete

Bernd S.
Posts: 17
Joined: 19 Feb 2015, 13:31
First name(s): Bernd
Last name(s): Schimmelpfennig
Affiliation: KIT
Country: Germany

Re: Exit code 2 (bad value during floating point read)

Post by Bernd S. » 02 Mar 2016, 13:17

I dare to disagree on the CAS to be reasonable. Having an orbital with occupation 1.992822654 in the active space is a clear indication of something to be fishy. Some counting by hand actually suggests a CAS (8/9) with the active orbitals being the 2p on C and the 2s on Li and Be. As much occupations from some MP2 can give some ideas about the active space it should not replace thinking about it to be reasonable. Actually, even with CAS (8/8) and to be admitted some other code, basis set and starting guess I did not run into the 1.992 occupation. This suggest the Hartree-Fock to be perhaps not the best starting guess in this case.

Cheers,

Bernd


[quote="taylor"]
***
So I return to my earlier contention: there is something pathological about this molecule. I looked also at your CAS(8,8) output and while I agree there is considerable multiconfigurational character I think you have misread the output. The leading coefficient in the wave function is not necessarily the first value printed in the list: this depends on how the program constructs the configuration list. And you can see that the second listed configuration actually has a coefficient of 0.88, or a weight of about 75%. As I say, this is multiconfigurational, but not as extreme as your earlier concerns. That said, I would not like to be relying on a single-configuration method here, at least not SCF or DFT.
***

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: Exit code 2 (bad value during floating point read)

Post by taylor » 02 Mar 2016, 14:04

I stand by everything I said in my last post, which is not the one you are responding to! I myself have not run any CAS(8,8) calculations, in part because I run with symmetry when it's there and could not figure out what seemed balanced across irreps. Rather than make an already long post even longer I did not go into details, but I ran 12 electrons in 14 MOs as a CAS --- the full valence would be 12 in 20 which is barely feasible and I was not prepared to put a lot of my own time and resources into it, and also a 12 in 20 RAS, with the RAS1 through 3 spaces as 5, 2, and 13 MOs respectively, and up to two RAS1 holes and up to two RAS3 electrons. The ground state calculations are virtually the same in both cases. In fact, the occupation numbers confirm that it is actually a pretty straightforward two-configuration ground state.

It is the 1B2 excited state contribution that is the problem when it comes to calculating the polarizabilities (see the rest of my last post)!

Best regards
Pete

Bernd S.
Posts: 17
Joined: 19 Feb 2015, 13:31
First name(s): Bernd
Last name(s): Schimmelpfennig
Affiliation: KIT
Country: Germany

Re: Exit code 2 (bad value during floating point read)

Post by Bernd S. » 02 Mar 2016, 14:17

Oops, no offense meant Pete, I know your fully capable of using CASSCF correctly and CAS (12/14) is a fair choice to get most included with reasoanble effort, I am fully with you on that point. I simply wantetd to point out that the CAS(8/8) is doubtful and one reason such calculations regulary show up is the use of MP2 to determine the active space.

Cheers,

Bernd

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: Exit code 2 (bad value during floating point read)

Post by taylor » 02 Mar 2016, 15:58

No worries, Bernd : I wasn't in the least offended! And I should have been more explicit about the active space in the original posting, but as I said it was already getting to be a very long posting!

I think in systems like this the most important thing is to use the symmetry and figure out what makes a plausible active space. I don't disagree with using MP2 to get a picture of some of the occupations but in a system that has symmetry it is not always straightforward to get a proper balance so that, e.g., for two "equivalent" bonds that occur in different irreps, one gets the bonding and antibonding combinations of both. This is much more obvious, i would claim, if one thinks about the symmetries of the various orbitals.

Best regards
Pete

Post Reply

Who is online

Users browsing this forum: No registered users and 7 guests