From owner-chemistry |-at-| ccl.net Wed May 26 00:06:00 2010 From: "Jiabo Li jiaboli:_:yahoo.com" To: CCL Subject: CCL:G: Different single-point DFT energy between GAMESS and Gaussian Message-Id: <-41967-100525224826-12764-cyn5crIZaOSk0BpA6k/20w-*-server.ccl.net> X-Original-From: Jiabo Li Content-Type: multipart/alternative; boundary="0-46328946-1274842094=:80568" Date: Tue, 25 May 2010 19:48:14 -0700 (PDT) MIME-Version: 1.0 Sent to CCL by: Jiabo Li [jiaboli::yahoo.com] --0-46328946-1274842094=:80568 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable The most likely scenario is the basis set linear dependency. GAMESS remove = the linear dependent (or more precisely, the near linear dependent) compone= nt in the basis set space. This certainly will slightly change the energies= . Here is an example output from GAMESS. You can check your log file for si= milar information: =A0 =A0NUMBER OF CARTESIAN ATOMIC ORBITALS=3D=A0=A0=A0=A0=A0=A0=A0 166 =A0NUMBER OF SPHERICAL CONTAMINANTS DROPPED=3D=A0=A0=A0=A0 0 =A0NUMBER OF LINEARLY DEPENDENT MOS DROPPED=3D=A0=A0=A0=A0 2 =A0TOTAL NUMBER OF MOS IN VARIATION SPACE=3D=A0=A0=A0=A0 164 =A0THE INPUT BASIS SET CONTAINS APPROXIMATE LINEAR DEPENDENCE. =A0THE SMALLEST EIGENVALUE OF THE OVERLAP MATRIX IS=A0=A0 2.898259E-09 =A0THERE ARE=A0=A0=A0 2 EIGENVALUES LESS THAN QMTTOL=3D=A0 1.00E-07 =A0EIGENVECTORS BELOW -QMTTOL- ARE DROPPED FROM THE MO SPACE, =A0IN ORDER TO ELIMINATE THE APPROXIMATE LINEAR DEPENDENCE. =A0THIS WILL MOST LIKELY CHANGE THE ENERGY ON THE ORDER OF TENS OF MICROHAR= TREES, =A0DEPENDING ON THE NUMBER OF ORBITALS REMOVED.=A0 THE VALUE OF QMTTOL CAN = BE =A0CHANGED IN $CONTRL: JOBS WITH EIGENVALUES BELOW 1.0E-07 PROBABLY WON'T =A0CONVERGE.=A0 EIGENVALUES BETWEEN 1.0E-07 AND 1.0E-06 MAY REQUIRE LOOSENI= NG OF =A0THE DENSITY CRITERION (CONV IN $SCF), MORE ACCURATE INTEGRAL EVALUATION =A0(ICUT=3D11 IN $CONTRL), MORE ACCURATE DIRECT SCF FOCK MATRIX =A0CONSTRUCTION (FDIFF=3D.FALSE. IN $SCF), OR CHANGING THE SCF CONVERGER =A0(DIIS=3D.TRUE. IN $SCF). =A0 Hope this solves the mystery. =A0 Jiabo --- On Tue, 5/25/10, Gustavo Mercier gamercier+/-yahoo.com wrote: > From: Gustavo Mercier gamercier+/-yahoo.com Subject: CCL:G: Different single-point DFT energy between GAMESS and Gaussi= an To: "Li, Jiabo " Date: Tuesday, May 25, 2010, 3:32 PM Hi! In addition to the many suggestions that you have received I would add: 1. GAMESS direct scf has several short cuts. Look at the control keyword gr= oup for this. I would get rid of it for both programs. 2. DIIS also has adjustable parameters, including how much 'history' you ke= ep. 3. With large basis sets there may be some linear dependencies and the prog= rams may automatically make changes in the basis set. Check your output. Review the GAMESS documentation and check out the fine print. Look at the f= ile "refs.pdf", for example. The developers have made a variety of choices = in the implementation of different algorithms and they may tell you that th= eir choices may result in small differences. However, these can easily be i= n the millihartree or less. You may conclude that getting the exactly the s= ame result is not reasonable. This does not even include such details as co= mpiler flags and optimization schemes used with different compilers. =A0 -- Gustavo A. Mercier, Jr., MD, PhD gamercier|*|yahoo.com Gustavo.Mercier|*|bmc.org gumercie|*|bu.edu 617-414-6440 > From: Piotr Nowak piotrnowak[*]student.uw.edu.pl To: "Mercier, Gustavo, " Sent: Tue, May 25, 2010 1:53:20 PM Subject: CCL:G: Different single-point DFT energy between GAMESS and Gaussi= an First of all, thanks everyone for response. Some questions and suggestions = appeared; I'll try to answer them briefly: -I've been using exactly the same structures for the single point energy ca= lculations; -I have been using spherical harmonics in both programs. Gaussian uses them= by default, and I have ensured their use in GAMESS with "ISPHER=3D+1" keyw= ord. The number of cartesian basis fuinctions is the same; -Gaussian manual states that default grid uses 75 radial shells and 302 ang= ular points/shell. I have been using the same grid in GAMESS thanks to "NRA= D=3D75" and =A0"NLEB=3D302" keywords. I also suspected that grid handling m= ight be implemented differently in both programs, therefore I tried some "s= uper-ultra-extra-fine" grid with 250 radial shells and 974 angular points/s= hell (using "Int(Grid=3D250974)" keyword in Gaussian). Unluckily, the energ= y difference remained within the same order of magnitude as it was with for= mer grid; -The relative energies are still different. If you compare e.g. different g= eometries of the same molecule, or activation energies, the error is still = 10^(-4) hartree. I would agree with Soren - there must be some "hidden" adjustable parameter= s, but I have no idea which one can cause these differences. I still hope i= t is possible to get the same results using both programs.=A0 Kind regards, Piotr On Tue, May 25, 2010 at 12:50 AM, Piotr Nowak piotrnowak!^!student.uw.edu.p= l wrote: Sent to CCL by: "Piotr =A0Nowak" [piotrnowak~!~student.uw.edu.pl] Dear CCL users, I'm trying to reproduce single point energy obtained with Gaussian 03 using GAMESS US. Hartree-Fock energy is almost exactly the same e.g. Gaussian: -1849.26414782 GAMESS: =A0 -1849.2641478646 Unfortunately my attempts to get the same results using DFT failed. The energy differences between both programs are unreasonably huge. Here are so= me examples of results for different functionals (the same case as above- mentioned HF example): M05-2X Gaussian: -1855.79754118 GAMESS: =A0 -1855.7976587495 SVWN5 Gaussian: -1845.45112047 GAMESS: =A0 -1845.4510666810 Slater (also known as Dirac, one of the simplest LDA functionals, so I'm su= re it has the same definition in both programs) Gaussian: -1833.20351470 GAMESS: =A0 -1833.2034704727 I have done those calculations using the same grid, using tight convergence criteria. I've found out that Gaussian uses slightly different cc-pVDZ basi= s set than the one present in Basis Set Exchange, but using this basis set wi= th GAMESS has left the results unchanged. I have also tried different guesses, and SCF algorithms, but without success. Here are keywords used in inputs for above calculations. Gaussian: #p m05/cc-pvdz nosymm iop(6/7=3D3) scf=3Dtight GAMESS: =A0$BASIS EXTFIL=3D.TRUE. GBASIS=3DCCPVDZGN $END =A0$CONTRL ISPHER=3D+1 SCFTYP=3DRHF RUNTYP=3DENERGY DFTTYP=3DM05 $END =A0$SYSTEM PARALL=3D.TRUE. MWORDS=3D200 $END =A0$SCF DIRSCF=3D.TRUE. DIIS=3D.TRUE. $END =A0$DFT NRAD=3D75 NLEB=3D302 $END I would appreciate any kind of help. Best regards, Piotr Nowak E-mail to subscribers: CHEMISTRY]^[ccl.net or use: =A0 =A0 =A0http://www.ccl.net/cgi-bin/ccl/send_ccl_message E-mail to administrators: CHEMISTRY-REQUEST]^[ccl.net or use =A0 =A0 =A0http://www.ccl.net/cgi-bin/ccl/send_ccl_message=A0 =A0 =A0http://www.ccl.net/chemistry/sub_unsub.shtml=A0 =A0 =A0http://www.ccl.net/spammers.txt--0-46328946-1274842094=:80568 Content-Type: text/html; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable
The most likely scenario is the basis se= t linear dependency. GAMESS remove the linear dependent (or more precisely,= the near linear dependent) component in the basis set space. This certainl= y will slightly change the energies. Here is an example output from GAMESS.= You can check your log file for similar information:
 

 NUMBER OF CARTESIAN ATOMIC ORBITALS=3D   &nbs= p;    166
 NUMBER OF SPHERICAL CONTAMINANTS DROPPED= =3D     0
 NUMBER OF LINEARLY DEPENDENT MOS DRO= PPED=3D     2
 TOTAL NUMBER OF MOS IN VARIATION= SPACE=3D     164
 THE INPUT BASIS SET CONTAINS APPROXIMATE LINEAR DEPENDENCE.
&= nbsp;THE SMALLEST EIGENVALUE OF THE OVERLAP MATRIX IS   2.898259E= -09
 THERE ARE    2 EIGENVALUES LESS THAN QMTT= OL=3D  1.00E-07
 EIGENVECTORS BELOW -QMTTOL- ARE DROPPE= D FROM THE MO SPACE,
 IN ORDER TO ELIMINATE THE APPROXIMATE LINEAR = DEPENDENCE.
 THIS WILL MOST LIKELY CHANGE THE ENERGY ON THE ORDER OF TENS OF = MICROHARTREES,
 DEPENDING ON THE NUMBER OF ORBITALS REMOVED.  = THE VALUE OF QMTTOL CAN BE
 CHANGED IN $CONTRL: JOBS WITH EIGENVALU= ES BELOW 1.0E-07 PROBABLY WON'T
 CONVERGE.  EIGENVALUES BETWEE= N 1.0E-07 AND 1.0E-06 MAY REQUIRE LOOSENING OF
 THE DENSITY CRITERI= ON (CONV IN $SCF), MORE ACCURATE INTEGRAL EVALUATION
 (ICUT=3D11 IN= $CONTRL), MORE ACCURATE DIRECT SCF FOCK MATRIX
 CONSTRUCTION (FDIF= F=3D.FALSE. IN $SCF), OR CHANGING THE SCF CONVERGER
 (DIIS=3D.TRUE.= IN $SCF).
 
Hope this solves the mystery.
 
Jiabo

--- On Tue, 5/25/10, Gustavo Mercier gamercier+/-yahoo= .com <owner-chemistry^^^ccl.net> wrote:

From: Gustavo Mercier gamercier+/-yahoo.com <o= wner-chemistry^^^ccl.net>
Subject: CCL:G: Different single-point DFT en= ergy between GAMESS and Gaussian
To: "Li, Jiabo " <jiaboli^^^ya= hoo.com>
Date: Tuesday, May 25, 2010, 3:32 PM

Hi!

In addition to the many suggestions that you have rece= ived I would add:

1. GAMESS direct scf has several short cuts. Look = at the control keyword group for this. I would get rid of it for both progr= ams.
2. DIIS also has adjustable parameters, including how much 'history= ' you keep.
3. With large basis sets there may be some linear dependenci= es and the programs may automatically make changes in the basis set. Check = your output.

Review the GAMESS documentation and check out the fine = print. Look at the file "refs.pdf", for example. The developers have made a= variety of choices in the implementation of different algorithms and they = may tell you that their choices may result in small differences. However, t= hese can easily be in the millihartree or less. You may conclude that getti= ng the exactly the same result is not reasonable. This does not even include such details as compiler flags and optimization schemes used with = different compilers.
 

--
Gustavo A. Mercier, Jr., MD, PhD
gamercier|*|= yahoo.com
Gustavo.Mercier|*|bmc.org
gumercie|*|bu.edu
617-414-6440=



From: Piotr Nowak piotrnowa= k[*]student.uw.edu.pl <owner-chemistry|*|ccl.net>
To: "Mercier, Gustavo, " <gamercier|*|= yahoo.com>
Sent: Tue,= May 25, 2010 1:53:20 PM
Subject: CCL:G: Different single-point DFT energy between GAMESS and Gauss= ian

First of all, thanks everyone for response. Some questions and suggest= ions appeared; I'll try to answer them briefly:

-I've been using exactly the same structures for the single point ener= gy calculations;
-I have been using spherical harmonics in both programs. Gaussian uses= them by default, and I have ensured their use in GAMESS with "ISPHER=3D+1"= keyword. The number of cartesian basis fuinctions is the same;
-Gaussian manual states that default grid uses 75 radial shells and 30= 2 angular points/shell. I have been using the same grid in GAMESS thanks to= "NRAD=3D75" and  "NLEB=3D302" keywords. I also suspected that grid ha= ndling might be implemented differently in both programs, therefore I tried= some "super-ultra-extra-fine" grid with 250 radial shells and 974 angular = points/shell (using "Int(Grid=3D250974)" keyword in Gaussian). Unluckily, t= he energy difference remained within the same order of magnitude as it was = with former grid;
-The relative energies are still different. If you compare e.g. differ= ent geometries of the same molecule, or activation energies, the error is s= till 10^(-4) hartree.

I would agree with Soren - there must be some "hidden" adjustable para= meters, but I have no idea which one can cause these differences. I still h= ope it is possible to get the same results using both programs. 

Kind regards,
Piotr


On Tue, May 25, 2010 at 12:50 AM, Piotr Nowak piot= rnowak!^!student.uw.edu.pl <owner-chemistry= ]^[ccl.net> wrote:

Sent to CCL by: "P= iotr  Nowak" [piotrnowak~!~student.uw.edu.pl]
Dear CCL users,
I'm trying to reproduce single point energy obtained with Gaussian 03 usi= ng
GAMESS US. Hartree-Fock energy is almost exactly the same e.g.
Gau= ssian: -1849.26414782
GAMESS:   -1849.2641478646

Unfortunate= ly my attempts to get the same results using DFT failed. The
energy diff= erences between both programs are unreasonably huge. Here are some
examp= les of results for different functionals (the same case as above-
mentio= ned HF example):
M05-2X
Gaussian: -1855.79754118
GAMESS:   -1= 855.7976587495
SVWN5
Gaussian: -1845.45112047
GAMESS:   -1845= .4510666810
Slater (also known as Dirac, one of the simplest LDA functio= nals, so I'm sure
it has the same definition in both programs)
Gaussian: -= 1833.20351470
GAMESS:   -1833.2034704727

I have done those c= alculations using the same grid, using tight convergence
criteria. I've = found out that Gaussian uses slightly different cc-pVDZ basis
set than t= he one present in Basis Set Exchange, but using this basis set with
GAME= SS has left the results unchanged. I have also tried different guesses,
= and SCF algorithms, but without success.

Here are keywords used in i= nputs for above calculations.
Gaussian:
#p m05/cc-pvdz nosymm iop(6/7= =3D3) scf=3Dtight

GAMESS:
 $BASIS EXTFIL=3D.TRUE. GBASIS=3DC= CPVDZGN $END
 $CONTRL ISPHER=3D+1 SCFTYP=3DRHF RUNTYP=3DENERGY DFTT= YP=3DM05 $END
 $SYSTEM PARALL=3D.TRUE. MWORDS=3D200 $END
 $= SCF DIRSCF=3D.TRUE. DIIS=3D.TRUE. $END
 $DFT NRAD=3D75 NLEB=3D302 $= END

I would appreciate any kind of help.
Best regards,
Piotr Nowak




E-mail to subscribers: CHEMISTRY]^[ccl.net= or use:
     http://www.ccl.net/cg= i-bin/ccl/send_ccl_message

E-mail to administrators: CHEMISTRY-REQUEST]^[ccl.net or use
    &nbs= p;http://www.ccl.net/cgi-bin/ccl/send_ccl_message
Subscribe/Unsubscribe:
     http://www.ccl.net/chemistry/sub_unsub.shtml
Before posting, check wait time at: http://www.ccl.net

Job: = http://www.ccl.net/jobs
Conferences: http://server.ccl.net/chemistry/announcements/conferences/

Search Messages: http://www.ccl.net/ch= emistry/searchccl/index.shtml

     <= A href=3D"http://www.ccl.net/spammers.txt" rel=3Dnofollow target=3D_blank>h= ttp://www.ccl.net/spammers.txt

RTFI: http://www.ccl.net/chemistry/aboutccl/instructions/




--0-46328946-1274842094=:80568--