RYSROT problem summary

 Dear netters,
 Thanks to all who replied to my problem.
 However this was, unfortunately not a case of more is better.
 Pg 93 of the g92 manual states that 2,000,000 words is sufficient for
 up to 110 basis functions and 16 atoms.  The limit of 2,800,000 was
 the most a user on our system is (on average) supposed to take when the
 system is loaded with jobs (as it usually is).
 The word from Gaussian is:
   Dr. Koyanagi,
    The Rys algorithm is a quadrature method and requires the roots to a
 one dimensional polynomial on the region (0.0,1.0).  This error occurs
 when the root finding code fails to find appropriate roots.  It is
 quite un-common and is a perverse combination or coefficients which are
    I am re-running your case to make sure I understand where it is failing.
 With Gaussian 92 I am not sure there is much we can do but I will see a
 tweak is helpful.
 Doug Fox
 help { *at * } gaussian.com
    I have run the test case you sent, using the geometry in the OPT input
 and run an MP2 FREQ on my HP system here without problem.  This may be
 an SGI specific problem.  I will send a copy of this to SGI for
 Doug Fox
 help { *at * } gaussian.com
 Which doesn't suprise me because we've been experiencing
 O/S problems for the last 5-6 weeks.  Which I may as well ask
 about that as well:
 Has anyone (on an R3000 SGI, O/S 4.0.5) experienced problems where the
 O/S after a time, does not release memory back to the user programs?
 After about ten days of operation the System Activity Reporter will
 show a steady drop of FREE MEMORY until there is about slightly less
 than 2Megs and then swapping takes over in a big way.
 I ask this here in this forum because ~95% of the CPU time on our
 system goes to ab initio computations.
 Greg Koyanagi                      koyanagi { *at * } watsci.uwaterloo.ca
 Department of Chemistry            519-885-1211x6388
 Universtiy of Waterloo
 Waterloo, Ontario, CANADA
 N2L 3G1