Version Molpro97 has the full functionality of molpro96. It
includes all patches of molpro96.4. 

Note: Due to a modified record structure (see below), 
molpro96 files cannot be restarted with molpro97.

Molpro97 has the following new features:

(1) Improved scf, in particular for large-scale direct calculations.
It is now possible to do initial calculations with a small basis set
(or semi-empirically) and use the resulting orbitals as
starting guess for larger calculations. This considerably reduces
the number of scf iterations in large calculations. In order to
make this as robust as possible, it was necessary to change the
record structure. The scf orbital records now hold information
about the corresponding basis set and wavefunctions, as well as
the density and fock matrices. The program recognizes automatically
whether the starting orbitals have the correct dimension, and if not,
a maximum overlap criterium is used to generate orbitals for
the new basis. This works in the following simple way:

basis=sto-3g
int
hf

basis=vdz
int
hf

basis=vtz
int
hf

will do the three calculations in one run and use in each case
the orbitals of the previous scf as starting guess. The previous
orbitals can also be computed in different symmetries. However,
the nuclear charges must be the same as in the current calculations,
otherwise either eigenfunctions of H0 are generated (scf, mcscf),
or an error occurs (other programs). 

This feature can be used with the scf and the mcscf programs.

Since now several orbital sets can be stored in one record, the
input for starting orbitals has been generalized. In all programs,
the format of the START (scf, multi) or ORBITAL (mrci, ccsd etc)
cards is as follows:

START,record,[type(s)],[STATE=istate],[SYM=isym],[NELEC=nelec],[SPIN=spin],[MS2=ms2]

where TYPE can be one of the following: CANONICAL, ALPHA, BETA, NATURAL, DIABATIC, LOCAL

If several types are given, they are searched in the given order. 

Normally, appropriate defaults are used, and none of these epcifications are necessary.

Similarly, density matrix input has the general form

DENSITY,record,[type],[STATE=istate],[SYM=isym],[NELEC=nelec],[SPIN=spin],[MS2=ms2]

where type can be CHARGE, SPIN, or TRANSITION (only one type can be specified).
Density matrices are now stored in the same record as the orbitals.

(2) RHF and HF are now the same program. HF and RHF are
aliases and can both be used for closed and open-shell cases.

(3) UHF has been improved, and now has DIIS convergence 
acceleration. Orbital sets for UNO-CAS can be generated
automatically.

(4) Separate level shifts are now available for the closed-
and open shells in the scf. 

(5) Direct SCF has more efficient prescreening now. Furthermore,
direct calculations start with a relatively crude integral threshold
(1.d-8), which is lowered automatically during the calculation.
This saves considerable CPU in large calculations.

(6) DFT now includes B3LYP hybrid functional. DFT calculations
are now possible up to nuclear charge 86 (previous version: 36).

(7) Different equivalent MCSCF/CASSCF orbitals (e.g. natural and
canonical) are now stored in one record. This simplifies the
input and avoids bookeeping of many records.

(8) Geometry optimization is further improved. In particular,
transition state optimization is now available. 

(9) Frequencies and thermodynamic properties can be computed
automatically.

(10) Integral-direct correlation methods are improved (still in
further development).

(11) UHF and DFT gradients are now available.

(12) MP2 one-electron properties computed as analytical energy
derivatives are now available. MP2 gradients will be available
soon.

(13) Property calculations are simplified. Property matrices are
computed automatically when needed; thus, it is no longer necessary
to specify to operators after int. Furthermore, a global EXPEC
(or GEXPEC) card is available. All properties are now stored in
one special record (1700.1), and the use must not worry about 
record numbers any more. Note that there are subtle changes in
the input for properties: the center number now refers to the
z-matrix row, and not to the printed atom numbers. This avoids
confusion when the z-matrix is reordered automatically. The 
following input possibilities exist for expectation values:

EXPEC,oper1,[center],[origin],oper2,...

where center can be a z-matrix row number or symbol. If center
is 0 or blank, the operator is computed at the given origin (x,y,z).

(14) The orbital diabatization procedure DIAB in multi is simplified and
more robust. It is no longer necessary to store previous geometries 
and to to the DISPL calculation; the overlap matrices between the
neighboring geometries are computed automatically when needed. The
geometry information is taken from the dump records.

(15) Similarly, the DDR procedure is simplified. Similar to DIAB, all
overlap matrices are computed automatically, and no DISPL and GEOM
cards are necessary any more. Moreover, the DDR can be done after
the displacements, i.e., it is not necessary any more to do the
reference geometry last.