CHARMM params

[ilya]

Dear cp2k users and developers!!!

I am running an MD of peptide with an ATP molecule inside. There are
PO2 - PO3 groups in ATP and the angle between P - O - P should differ
from 180 deg substantially. When using NAMD it does not change the
angle (which is correct I think), while running MD or GEO_OPT with
CP2K makes it (P-O-P) almost linear after a few steps (all input psf,
pdb and param files are the same). After spending a lot of time
investigating this I noticed a strange thing in CHARMM prm file.
Here is the copy of original charmm param file (angles section)

CN9  ON2  P2      20.0     120.0  35.   2.33 !nad/ppi, jjp1/adm jr.
7/95
P    ON2  P       15.0     140.0 -40.0  2.80 !nad/ppi, jjp1/adm jr.
7/95
P2   ON2  P       15.0     140.0 -40.0  2.80 !nad/ppi, jjp1/adm jr.
7/95
P3   ON2  P       45.0     140.0  40.0  3.15 !PPI2, from nad/ppi, adm
jr., 7/01
P2   ON2  P2      15.0     140.0 -40.0  2.80 !nad/ppi, jjp1/adm jr.
7/95
HN4  ON4  P       30.0     115.0  40.0  2.35 !MP_1, ADM Jr.
HN4  ON4  P3      30.0     120.0  40.0  2.35 !PPI2, from MP_1, adm
jr., 7/01

notice the minuses in Urey-Bradley force const (col 6, value = -40.0)
for lines 2, 3 and 5. What is the point of negative force const (well,
I should ask charmm devs)???? CHARMM uses the form of U_ub = k (s-
s0)^2. As I understand using negative const gives pointless results
(the larger the diff the smaller the energy).
Removing the minuses and changing values to 40.0 makes the fragment
'not so linear' in CP2K but anyway the angle still differs a lot from
that when using NAMD. Maybe the implementation is different in NAMD?

Any ideas?

Best wishes,
Ilya.