CCL: TDHF

[Ulrike Salzner <salzner^at^fen.bilkent.edu.tr>]

 Many thanks to Stefan Grimme, Artem Masumov, and Doug Fox. 
 
 I asked whether it is possible to calculate and optimize the ground state
 of a system with TDHF. The answer is no. Like CIS, TDHF does not provide
 a correction to the ground state energy. More important I was warned that
 the TDHF geometry optimization for excited states in G98 is an approximation
 and should not be used. The option has been removed in G03.
 
 One comment of mine to the statement that TDDFT is superior to TDHF: I am
 aware that TDDFT is quite successful in general. Unfortunately, the only
 excited state that is currently of interest to me (the 1Bu excited state
 of polyenes) is not well reproduced, especially with increasing size and
 conjugation length, which is exactly what I need to investigate. TDHF, in
 contrast, gives numbers similar to CASMP2. Therefore, it is bad news indeed
 that the geometries can not be optimized at this level of theory.
 
 The original question and the answers follow below:
 
Hello,
 > the time-dependent Hartree-Fock
 method is described as a method for
 > calculating excited states. I would
 like to know whether one can also
 > optimize the ground state including
 correlation with TDHF. In other
 > words, if I use the keywords "fopt"
 and "nroot=0" in Gaussian what am I
 > calculating? I know that CIS does not
 make a correction to the ground
 > state because of Brillouin's theorem
 but I am not sure what TDHF
 > includes exactly.  The reason why I
 am considering this is that I would
 > like to compare the excitet state and
 the ground state geometries on
 > equal footing. Would it be better to
 compare the excited state TDHF
 > geometry to the HF ground state
 geometry?
 > Thanks in advance,
 > Ulrike Salzner
 
Dear Ulrike,
 in fact TDHF is not defined for the ground state but there is a close
 analogy between CIS and TDHF (expanded in singles only,
 CIS equations can be derived from the non-Hermitian TDHF
 problem by neglecting the so-called B-matrix).
 >From that I would argue that the ground state analogue of TDHF is just HF.
 Regards
 Stefan
 _________________________________________________________
 Prof. Dr. Stefan Grimme
 Organisch-Chemisches Institut (Abt. Theoretische Chemie)
 Westfaelische Wilhelms-Universitaet, Corrensstrasse 40
 D-48149 Muenster, Tel (+49)-251-83 36512/33241/36515(Fax)
 
Ulrike,
 Just like CIS, TDHF does not improve the ground state.
 So correct comparison would be TDHF optimized excited state and HF
 optimized ground state.
 In fact, TDHF only differs from CIS in that it has nonzero V-O block in
 hamiltonian (and thus, in transition density) matrix.
 In g98 this block is missing from the routine evaluating gradients, so
 excited state optimization with analytical gradients gives incorrect
 (approximate) results. That is why in g03 this option is blocked. In
 Turbomol TDHF opt. is coded fine.
 To optimize the excited state at TDHF level you need to do Opt=Numer,
 which is much slower.
 Please keep in mind that TDHF is inferior to TDDFT at the same
 computational cost.
 Hope this helps,
 Artem
 ________________________________________________________
      __    ___________        Artem.Masunov^at^LANL.gov
     /  \  /  __   __  \   www.t12.lanl.gov/home/amasunov
    /    \/\  \ \  \ \  \  505.665.2635, Fax:505.665.3909
   /  /\  \ \  \ \  \ \  \  Theoretical Division, MS B268
  /  ____  \ \  \ \  \ \  \    Los Alamos National Lab
 /__/\ _/\ _\ \ _\ \ _\ \ _\     Los Alamos NM 87545
 \ _\/  \/__/\ __/\ __/\ __/ ____________________________
 Dr. Salzner,
    G98/03 does not have gradients implement analytic gradients for
 TDHF or TDDFT so optimizations need to be performed with only
 energy values OPT=(EF,EnOnly) and a symbolic Z-matrix.  But it
 can be done for medium to small systems.  Or at least a few degrees
 of freedom.
    The TDHF and CIS methods both use the HF solution as the reference
 and neither of them improve on the ground state description.
    In the sense that neither the Excited state or the Ground state
 is really correlated it has been our experience that structures in
 CIS are similar in quality to HF although often a slightly larger basis
 is needed.    To go beyond this you might want to consider SACCI
 which is in G03 or CASSCF, both of which can include correlation,
 dynamic vs. static, and treat ground and excited state on a equal
 footing.