C. Lung et al., WAVE OPERATOR AND ARTIFICIAL-INTELLIGENCE CONTRACTION ALGORITHMS IN QUANTUM DYNAMICS - APPLICATION TO CD3H AND C6H6, The Journal of chemical physics, 98(9), 1993, pp. 6722-6734
We have established in this study the capabilities of the wave operato
r (WO) algorithm to extract from a huge primitive space a smaller subs
pace (the active space) containing all of the zero order states which
play an active role during the intramolecular vibrational energy redis
tribution (IVR) from an initial state \i]0. While exact methods such a
s the recursive residue generation method (RRGM) or the Chebychev algo
rithms can only be applied in a primitive space containing less than a
bout 200 000 states, the WO algorithm can be used efficiently in ultra
large basis sets containing billions of states. The recursive residue
generation method (RRGM) or Chebychev methods can then be applied in t
his active space which typically contains less than 10 000 states. In
order to draw general conclusions on the efficiency of such a method a
nd on the main features of IVR phenomena, we have concurrently studied
IVR from the fifth CH overtone in the nine mode CD3H molecule and fro
m the second CH overtone in the 16 mode C6H6 system. We have analyzed
the main features of the active space and have shown that the WO algor
ithm selects the important states. A very broad energy distribution of
states in the active space has been obtained for these two systems. W
e have also shown that C6H6 is a very complex system to study; while o
nly a few hundred states are effectively populated during the IVR from
the fifth CH overtone in CD3H, about 8000 states have to be considere
d in order to accurately study IVR from the second CH overtone in C6H6
. However, we have shown that the WO method is able to reproduce corre
ctly both the survival probability of the initial state and the intric
ate energy flow through the molecule during the first picosecond. Fina
lly, we have shown that the WO algorithm builds a far more efficient a
ctive space than a more traditional artificial intelligence (Al) tree
pruning procedure.