Dear MM5 Users,

Please note that setting the MPHYSTBL option to 1 in configure.user,
and thus using the lexmoiss and lexmoisr versions of the exmoiss and
exmoisr routines, is considered unsafe for operational use.  Operational
users of MM5 should use executables compiled with MPHYSTBL = 0 option.

The lexmoiss and lexmoisr routines use lookup tables instead of the EXP
intrinsic, for speed. However, there is no bounds-checking on the
tables, and it can happen where the table lookout goes out of bounds
with sometimes fatal results for the run.  For example:

  Segmentation fault in lexmoisr at 0x100e12b8
  0x100e12b8 (lexmoisr+0x30f0) c0993888        lfs   fr4,0x3888(r25)
  (dbx) where
  lexmoisr() at 0x100e12b8
  solve.__solve_out_31() at 0x10070594
  _xlsmpCall() at 0xd05a94d4
  _xlsmpParDoSetup() at 0xd05a93e8
  solve() at 0x10065734
  mm5() at 0x10000d34

Which occurred on several successive CONUS runs on an IBM SP
Winterhawk-II system in May.  Setting MPHYSTBL back to 0, recompiling,
and rerunning corrected the problem.

The sensitivity to the option appears to be related to conditions in
the particular simulation.  Non-operational users running historical
cases may be able to use MPHYSTBL=1 without incident. But given the
wide variation in scenarios that occurs over time and the requirement
for high reliability in operational settings, it is very strongly
recommended that these users use only MPHYSTBL=0.

Rotang

May 11, 2000