Re: HyperChem

[slee-0at0-hyper.hyper.com (Thomas Slee)]

 Dear CCL readers,
 Declaration of interest: I work for Hypercube, developers of HyperChem.
 It seems appropriate, with the current round of discussion of
 HyperChem experiences going on, that I do two things:
 1.  Avoid unnecessary interference in this discussion, as an
     interested party, but:
 2.  Update and correct purely factual observations where necessary.
 With this in mind, here is an update on Dr. Bliznyuk's comments, which
 referred to Release 3 of HyperChem.  The current release is Release 4.
 |
 | Last year we had a chance to try HyperChem v 3.0. This program has
 | good graphics and very good molecular editor/builder.
 |
 	Thank you! (couldn't resist leaving it in there).
 |
 | Unfortunately, it is very slow, at least for semiempirical calculations.
 | For example, it takes 35 min. on 486DX2 50MHz with 16 MB of memory to do
 | MNDO 1SCF and gradients for C60 (compare to Sibiq - 10 min.).
 |
 Dr. Bliznyuk will be happy to know we have significantly improved the
 performance of this kind of calculation.
 Calculations on a 486DX2 50 MHz with 20MB of memory show the following:
 (C_60, AM1, SCF convergence to 1e-4 kcal/mol, + gradient evaluation),
         HyperChem Release 3.    33 min (my calculation)
         HyperChem Release 3.    35 min (Dr. Bliznyuk's calculation)
         Sibiq                   10 min (Dr. Bliznyuk's calculation)
         HyperChem Release 4.    7 min 30 sec. (my calculation.)
 C_60 is a very memory-intensive system for HyperChem's analytical
 energy derivative computation.  Hence some of this time is taken
 up with disk access (I suspect this is the cause of the 2 min
 discrepancy between the Release 3 calculations).  For a system with
 the same number of atomic orbitals but more hydrogen atoms (C_40H_80)
 the memory requirements go down by 3Mb and the Release 4 time for
 SCF plus gradient becomes 6 min 30 sec.  I have no data from Sibiq
 for this system, as we don't have that program in house.
 We did do a couple of small calculations checking a demo disk (small
 molecules only) of Sibiq against HyperChem and the results came out
 for C6H10 that HyperChem takes 3.0 seconds for this small system, while
 Sibiq takes 5.3 sec.  Please do not take this as a claim for superior
 performance overall for HyperChem compared to Sibiq; it is quite possible
 that in other calculations Sibiq would be faster than HyperChem.  It
 is certainly not my intent to compare the two programs, merely to show
 that the timings I give are consistent with Dr. Bliznyuk's values, and
 that we have indeed speeded up the program significantly.
 On the subject of timings, we have carried out test calculations
 similar to the above on a couple of different platforms comparing HyperChem
 and MOPAC 6.  Depending on molecule and compiler, sometimes MOPAC is faster,
 sometimes HyperChem.
 The other point on which I wish to reassure CCL readers is the implicit
 suggestion in an earlier posting that HyperChem may give results
 inconsistent with MOPAC or AMPAC for NDDO-type calculations.
 Our formal testing of HyperChem involves over 1,000 separate calculations.
 This testing, and two years of customer experience with the program, have yet
 to show a case where HyperChem and MOPAC produce different results
 for the same calculation.  The insignificant differences mentioned by
 Antti Poso come, we believe, from MOPAC 6's slightly inconsistent values
 for conversion factors in different parts of the program.
 	Tom Slee
 --
 Tom Slee, Hypercube Inc., 419 Phillip St., Waterloo, Ont., Canada, N2L 3X2
 E-Mail: slee-0at0-hyper.com,  Tel. (519)725-4040,  Fax. (519)725-5193.