From genghis &$at$& darkwing.uoregon.edu Thu Sep 12 10:11:13 1996 Received: from darkwing.uoregon.edu for genghis "-at-" darkwing.uoregon.edu by www.ccl.net (8.7.5/950822.1) id JAA16055; Thu, 12 Sep 1996 09:54:32 -0400 (EDT) Received: (from genghis %-% at %-% localhost) by darkwing.uoregon.edu (8.7.5/8.7.3) id GAA21139; Thu, 12 Sep 1996 06:54:31 -0700 (PDT) Date: Thu, 12 Sep 1996 06:54:31 -0700 (PDT) From: Dale Braden To: cclpost Subject: Summary: ECPs Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Dear CCL, I was very pleased with the immediate responses to my questions regarding effective core potentials. My original post is below, followed by the responses. Thanks to all of you who responded! Dear CCL, I read the postings of a year or two ago regarding the use of effective core potentials in ab initio calculations, and I would like to submit a few questions on this subject. As I understand it, ECPs were developed primarily to speed up calculations by reducing the number of electrons that needed to be considered independently. An added benefit was that relativistic energy corrections could then be made for the electrons in the core. Since core electrons aren't thought to participate much in bonding, this was justified. So my first question is, If I am doing a *geometry optimization* (and don't care about what the energy is) on an organometallic complex, and can afford to use an all-electron basis, should I do so? In other words, does the use of an ECP generally have a neglible effect upon the geometry? I know that relativistic effects increase with atomic number. Are such effects already drastic on first-row transition metals, or do they become significant (in terms of energy or geometry) only by the second-row? After looking over the literature on this subject, I am confused as to whether there is a difference between an "effective core potential" and a "pseudopotential" for the core. I use Gaussian 94/Rev. C.3, and when I tried to input the PPs of Dolg, et al. [JCP 86 (1987) 866], the program would not accept them, but seemed to be expecting information on the different "projections", like P-D, S-D, and so forth. I'm not sure what these are, and in any case, they are not explicity described by Dolg, et al., so I don't know whether the problem is one of format or not. I shall collect and post all responses! Thank you, Dale Braden Dept. of Chemistry University of Oregon genghis.,at,.darkwing.uoregon.edu *1**************** >From schrecke ^at^ zinc.chem.ucalgary.ca Mon Sep 9 15:24 PDT 1996 Hi Dale, here are my $0.03 or so to your questions. Interesting subject ... > > I know that relativistic effects increase with atomic number. Are such > effects already drastic on first-row transition metals, or do they become > significant (in terms of energy or geometry) only by the second-row? It seems that you can get away without relativity in the first transition row. In a few studies from some years ago we found the following relativistic bond contractions for transition metal complexes: Cr(CO)6: <<0.001Angstrom for M-C and C-O Fe(CO)5: 0.002Ang (M-C_ax); 0.003Ang (M-C_eq); <<0.001Ang (C-O_ax); 0.001Ang (C-O_eq) Ni(CO)4: 0.001Ang (M-C); <<0.001Ang (C-O) etc. Cf. J. Li, G. Schreckenbach, T. Ziegler, J.Phys.Chem. 1994,98,4838 JACS 1995,117,486 Inorg.Chem. 1995,34,3245 J. Li, R.M. Dickson, T. Ziegler, JACS 1995,117,11482 We don't use pseudopotentials, but the result should be independent of this. See also reviews: P.Pyykko, Chem.Rev.1988,88,563 (I had cited this paper wrong in a recent CCL posting) P.Pyykko, in: THe Effects of Relativity on Atoms, Molecules and the Solid State (S.Wilson et al., eds.) Plenum, New York, 1991 > After looking over the literature on this subject, I am confused as to > whether there is a difference between an "effective core potential" and a > "pseudopotential" for the core. In my opinion, the two are exactly the same. I am looking forward to your summary! Yours, Georg -- ============================================================================== Georg Schreckenbach Tel: (Canada)-403-220 8204 Department of Chemistry FAX: (Canada)-403-289 9488 University of Calgary Email: schrecke(-(at)-)zinc.chem.ucalgary.ca 2500 University Drive N.W., Calgary, Alberta, Canada, T2N 1N4 ============================================================================== *2************* >From hrusak { *at * } ims.ac.jp Mon Sep 9 22:33 PDT 1996 Dear Dale, with respect to your questions in the recent posting to CCL 1) In fact for the first row atoms there is not a real speedup when using ECP and the relativistic contributions are quite small and thus it seems that there is no real reason to favor them over AE. However, there are some points of particular importance. I) Sometimes one is interested in comparisons among the different metals in a column of the periodic table and thus a consistent description is favorable. II) Quite often the ECP calculations converges much faster compared to AE. III) In some cases when doing MCSCF/AE based calculations you may suffer from orbital rotations between the active and inactive spaces, which can not occur when using ECP. IV) There are many papers showing the effect of ECP on structure and energetics is negligible small. I think that the effect of the valence basis is much more important than the use of the frozen core approximation. 2) The RECPs from the Stuttgart group (Dolg) seems to be quite good eventhough the others (maybe beside the old Hay&Wadt once) are also comparable. To generate a Gaussian input is quite easy just following the handbook and combine the components for the individual (L) in the L-L(Max) way. If You ask directly by the Stuttgart I think they have also compiled their ECP to a Gaussian readable form. Regards Jan Hrusak *3************* >From ehlers[ AT ]chem.vu.nl Tue Sep 10 00:24 PDT 1996 Dear Dale, To make it short, for the first row transition metal complexes it should make almost no difference (for the results od a geometry optimisation) if you use an all electron basis or an ECP. This is not true for the second and third row transition metals due to the pseudo relativistic treatment of the ECP's. In fact, using an all electron basis you should also make, somehow, a relativistic calculation to reach the same results. But in any case, i recommend the use of ECP's for all kind of transition metals, because you get your results faster. Be aware that you'l have to calculate correlation enegergy, depending on your system. We made the experience that you have to make at least MP2 geometry optimisations for transition metal complexes in low oxidation states. And thats difficult enough using ECP's, dont think about all electron calculations. On your last question, the difference between ECP's and PP's is mainly in the way the authors derived the potentials. We also used the PP's of Dolg, et al. [JCP 86 (1987) 866] in G92, so i assume g94 will make it too. But you can also try the ECPs of Hay and Wadt (LANL2 not LANL1 in Gaussian.) If you are interested in some numbers giving answers to your question, you should read our article in JACS, Vol. 116, No. 4, 1994 p 1514. Succes wiyh your work Andreas =========================================================================== = - Andreas Ehlers = = - http://www.chem.vu.nl/Staf/ehlers/index.html = = (__) ____ - Afdeling Theoretische Chemie, Faculteit Scheikunde = = (oo)/ \/~`- Vrije Universiteit Amsterdam = = U (__)_____|| - De Boelelaan 1083, 1081 HV Amsterdam U U U = = \|/ || W|| - [ ehlers (+ at +) chem.vu.nl \|/ \|/ \|/ = =========================================================================== *4************ >From krause -AatT- chemie.uni-hamburg.de Tue Sep 10 05:37 PDT 1996 Hi Dale, A nice review article concerning relativistic effects is: P. Pyykkoe, Relativistic Effects in Structural Chemistry, Chem. Rev. 88 (1988) 563-594. I hope this helps. Yours Knut -------------------------------------- Knut Krause Institut fuer Physikalische Chemie Bundesstrasse 45 20146 Hamburg Germany Tel: # 040/4123-3428 Fax: # 040/4123-3452 E-mail: krause[ AT ]chemie.uni-hamburg.de -------------------------------------- *5************* >From Igor Shamovsky Tue Sep 10 05:59 PDT 1996 Dear Dale: If you can afford the all-electron basis set, then the answer is very simple. Use it. Forget about "effective core potentials" or "pseudopotentials", whatever you prefer. Stick to ab initio calculations if possible. Pseudopotentials have some semiempirical smell. Contribution of relativistic effect for the first and second raw elements is negligible. With best regards, Igor. *6*********** >From ashutosh "-at-" sol.acs.unt.edu Tue Sep 10 07:20 PDT 1996 Hi Dale, I have some experience working with ECP's on iodine compounds.. usually it takes some effort to correctly input the ECP's in a format accepatble by G94, and usually the ECP listing has all the information necessary to be able to enter them correctly in G94 input file. If you are willing, I can help you some with whatever little experience I have with ECP's. Provide me the ECP listing if you can and I will take a look at it. Regards, Ashutosh Misra *7************ >From FAU ^at^ ps1515.chemie.uni-marburg.de Wed Sep 11 07:10 PDT 1996 Hi, there is a fault in my 1st mail (in the 2nd line of the 1st Br ECP) which is now corrected. The input format for ECP's of the Stuttgart-group (Stoll, Preuss et al.) is free format and as follows: line 1: -: optional, if you want to combine several ECP's in one file atomic symbol 0 line 2: name of the ECP number of projections number of replaced electrons line 3: comment for general term line 4: number of lines for the general term line 5: (here: a single line) 2: indicates gaussian functions exponent coefficient lines 3 to 5 are "repeated" for each of the projections. important: do not terminate the ECP by ++++ or **** as is stated in the manual. At the end of the mail are three examples of Stuttgart-ECP's. -Cl 0 cl-ecp10-mwb 3 10 f (and higher) 1 2 1. 0. s-f 2 2 6.3943 33.13663196 2 3.1971 16.27072783 p-f 2 2 5.6207 24.41699269 2 2.8103 7.68304978 d-f 1 2 5.3381 -8.58764865 -Br 0 br-ecp28-mwb 4 28 g (and higher) 1 2 1. 0. f-g 1 2 2.7207 -8.16149293 s-g 2 2 5.0218 61.51372099 2 2.5109 9.02149299 p-g 2 2 4.2814 53.87586402 2 2.1407 4.62940227 d-g 2 2 2.8800 20.84967744 2 1.4400 2.96544431 This example shows how an alternative way to describe f-projections: simply add it to the general term and subtract it from the lower terms. -Br 0 br-ecp28-mwb 3 28 f (and higher) 1 2 2.7207 -8.16149293 s-f 3 2 5.0218 61.51372099 2 2.5109 9.02149299 2 2.7207 8.16149293 p-f 3 2 4.2814 53.87586402 2 2.1407 4.62940227 2 2.7207 8.16149293 d-f 3 2 2.8800 20.84967744 2 1.4400 2.96544431 2 2.7207 8.16149293 The results for the different descriptions differ in the 6th or 7th decimal digit of the energy. __________________________________________________________ Stefan Fau, fau "at@at" ps1515.chemie.uni-marburg.de FB Chemie der Philipps-Universitaet Marburg, Hans-Meerwein-Str. D-35032 Marburg *8************ >From ashutosh (+ at +) sol.acs.unt.edu Wed Sep 11 09:13 PDT 1996 Dear Dale, Here is what I can explain regarding the ECP's you mentioned in your previous e-mail. I havent read the reference you have cited, but if you look closely in the original paper, it will probably give you an idea if this ECP is of the Hay-Wadt type or the Stuttgart Group's type. I would say by looking at the ECP you are working on, is that it is a Stuttgart Group's ECP.You might also want to take a look at the following reference: Bergner, A., Dolg,M., et. al., Molecular Physics, 1993, Vol 80, No. 6, 1431-1441. Your ECP matches the format given in the paper above.Now, having resolved that this ECP is probably that of Stuttgart type, we have to enter it in a format acceptable to Gaussian94. One of the things not mentioned in Gaussian manuals is that the Stuttgart ECP's (as opposed to Hay-Wadt ECP's) are not easily entered into Gaussian input files, and one needs to employ a "trick" to make it work. Here is the trick (I will not go into the mathematics/physics of the procedure), and it has to do with the fact that Hay-Wadt potentials give values for (U_l - U_lmax) terms and Stuttgart ECP's give them for U_l terms. So one needs to ADD a U_lmax term with a zero coefficient while entering Stuttgart ECP's into Gaussian9x. Confused? well, read on, and hopefully this point will become clear. Come to think of it, the ECP you have mentioned is definitely Stuttgart type, since Dolg is from the Stuttgart group!! I am assuming you are using the correct basis set supplied with the ECP you are using. Such basis set should be in the reference (hopefully!!) you are using to obtain the ECP. This basis set will be used for describing the valence electrons (while the ECP takes care of the core electrons). This basis set can be used in the input section of your Gaussian file using the "gen" keyword. If you are unclear about this procedure, let me know.. I will hopefully include a sample G94 file to assist you, in this mail. Ok, so looking at the Co ECP you have supplied below: Co has 27 electrons total. I do not know how many electrons should be in the core and how many in valence- this will be dictated by the basis set you are using for the valence electrons. I assume that 17 electrons are in the core, and 10 in valence orbitals (BUT PLEASE CHECK THIS!! I AM NOT FAMILIAR WITH TRANSITION METAL CHEMISTRY). This is why Q=17 in the ECP you have quoted. Q l k A_kl a_kl 17 0 1 283.960566 23.66 0 2 47.1568459 10.61 1 1 182.212236 25.04 1 2 35.2333515 10.44 2 1 -26.4753327 29.54 2 2 -1.82578723 10.18 The above will be written in the following manner in G94 input Co 0 Dolg 3 17 L=3 component 1 2 1.000000 0.000000 <--- This is what I meant by the trick L=0 component 2 2 23.66 283.960566 2 10.61 47.1568459 L=1 component 2 2 25.04 182.212236 2 10.44 35.2333515 L=2 component 2 2 29.54 -26.4753327 2 10.18 -1.82578723 That's it!! The above should work just fine with a compatible basis set in G94. I do not have the basis set otherwise I would have tested it out. Above, you see a L=3 component with a 0.00 coefficient, and that is the trick about using Stuttgart type ECP's with Gaussian9x. The ECP you gave had a maximum L=2 component, so I added a L=3 component with a zero coefficient. Try this out (after checking two things :(1) the core charge, if that number should be 17 or something else, if you think it should be different, replace 17 with the appropriate number. (2) the correct basis set for this ECP. This basis set can be either included in the input section of the job, or stored as a different file to which you can refer to in your input. I will include the following example, which is that for a bromine atom using Stuttgart ECP's. Location of the basis set in my example is indicated by the lines with a -8 at 8- prefix. You will need to substitute the appropriate directory names where your basis set is located. Feel free to summarize to CCL if you feel appropriate. Let me know if this helps, and if I can be of further assistance. Best wishes, Ashutosh Misra ------------------------------------------------------------------ Sample G94 input file for Br atom ------------------------------------------------------------------ %chk=br #mp4/gen pseudo=read test Br .,at,. mp4/6-311g** 0 2 Br ( ( at ) ) /scratch2/marshap/gbs/ecp/st/br631gst.gbs (-(at)-)/scratch2/marshap/gbs/ecp/st/br6311d.gbs Br 0 Stuttgart 4 28 L=4 component 1 2 1.000000 0.000000 L=0 component 2 2 5.0218 61.513721 2 2.5109 9.021493 L=1 component 2 2 4.2814 53.875864 2 2.1407 4.629402 L=2 component 2 2 2.8800 20.849677 2 1.4400 2.965444 L=3 component 1 2 2.7207 -8.161493 ---------------------------------------------