[CPMD-list] Problem doing transition state search

[Christian Tuma]

Dear CPMD developers,

before CPMD release 3.7 I used an external optimiser to locate transition
states with CPMD. Now I was trying to do this using version 3.7 and its
recently implemented internal transition search algorithm. The system under
study was a small zeolite system (37 atoms per unit cell).
First I did a frequency calculation, the structure was already VERY close
to the transition state:

      -1273.4964        -31.8808         34.0303         45.3821
         82.6421         84.8253         94.0645        105.0102 ...

translation projected:

      -1272.7988           .0000           .0000           .0000
         80.7846         82.4328         89.3480        104.7284 ...

Then I tried a transition state search using the calculated Hessian:

&CPMD
 OPTIMIZE GEOMETRY
 CONVERGENCE ORBITALS
  1.0E-06
 CONVERGENCE GEOMETRY
  1.0E-04
 RESTART HESSIAN LATEST
 RFO ORDER=1
&END
...

The first relaxation steps looked like this:

 *** GNMAX=  1.132939E-04                ETOT=   -432.246618  ***
 *** GNMAX=  2.066742E-04 [9.24E-03]     ETOT=   -432.246615  ***
 *** GNMAX=  3.516007E-04 [9.36E-03]     ETOT=   -432.246610  ***
 *** GNMAX=  5.045955E-04 [9.47E-03]     ETOT=   -432.246602  ***
 *** GNMAX=  6.773491E-04 [1.13E-02]     ETOT=   -432.246591  ***
 *** GNMAX=  1.041622E-03 [2.61E-02]     ETOT=   -432.246554  ***
 *** GNMAX=  1.026987E-03 [5.18E-03]     ETOT=   -432.246553  ***
 *** GNMAX=  9.818658E-04 [6.93E-03]     ETOT=   -432.246550  ***
 *** GNMAX=  9.562766E-04 [4.51E-03]     ETOT=   -432.246550  ***
 *** GNMAX=  9.000179E-04 [4.27E-03]     ETOT=   -432.246547  ***
 *** GNMAX=  8.270440E-04 [7.04E-03]     ETOT=   -432.246550  ***
 *** GNMAX=  8.178444E-04 [1.79E-03]     ETOT=   -432.246550  ***
 *** GNMAX=  7.637884E-04 [4.46E-03]     ETOT=   -432.246546  ***
 *** GNMAX=  7.829667E-04 [7.01E-03]     ETOT=   -432.246533  ***
 *** GNMAX=  8.246068E-04 [8.47E-03]     ETOT=   -432.246510  ***
...

Then it continued with almost the same values for GNMAX and ETOT. 
But I expected that it would take only a few steps to get from
GNMAX=  1.132939E-04 to GNMAX < 1.0E-04 to fulfil my given convergence
criterion that I usually use with ordinary geometry optimisations.
Instead GNMAX got bigger.

Did I make something wrong in my input file? 
Any recommendations for changes?


One more thing that puzzled me were statements in the output like the
following (example after first step):

 RFO| MODES: NEGATIVE=   1     ZERO=   9     POSITIVE= 101
 LOWEST HESSIAN EIGENVALUES:
 -.095510 -.000729  .000858  .001499
  .004318  .004675  .005900  .007038

Why ZERO=   9 ? From the source code (rrfo.f) I got the answer, every
eigenvalue between -0.01 and 0.01 is treated as 0. I do not know the details
of the RFO scheme, but after projection of the translation there are only
3 eigenvalues exactly 0. Does this somehow influence the performance of
the transition state search?


Thank you, any advice is welcome.

Christian.

-- 
Christian Tuma             Humboldt-Universitaet Berlin
ct at chemie.hu-berlin.de     Arbeitsgruppe Quantenchemie (Prof. Sauer)
phone: +49-30-20937140     Brook-Taylor-Str. 2, 12489 Berlin, GERMANY
fax:   +49-30-20937136     http://www.chemie.hu-berlin.de/ag_sauer