dgc@ucla-cs.UUCP (01/16/85)
The following article has been distributed to the numerical analysis
community by the well-known Stanford University Numerical Analyst
and Applied Mathematician Gene Golub. Since it presents quite a
different view from the "Murray Hill hype", it appears to be worth
redistributing.
Please note that I have no personal knowledge of the issues involved and
am simply submitting this as a matter of public interest -- dgc.
------------------------------------------------------------------------
Return-path: <GOLUB@SU-SCORE.ARPA>
Date: Tue 15 Jan 85 11:57:36-PST
From: Irvin Lustig <OR.LUSTIG@SU-SIERRA.ARPA>
Subject: "Facts" on Karmarkar
To: su-bboards@SU-SIERRA.ARPA
A recent question was posted as to why there was a mild disturb-
ance at the talk given last Friday by Narendra Karmarkar. The
community should realize the motivation for the attendees to
question Karmarkar's computational results.
As most people know, Karmarkar has claimed that his algorithm
has run 50 times faster than the Simplex Method. However, this
has NOT been substantiated at ANY talk given by Karmarkar. No
person outside of AT&T Bell Labs has seen the nature of the
problems solved (i.e., number of rows, number of columns, or
sparsity structure) nor the CPU time consumed by solving any of
these problems.
What this all means is that the jury is still out on his
algorithm. No one has yet substantiated his claims, and
Karmarkar refused to show any results to members of the O.R.
department here at Stanford in private sessions on Friday
morning. His talk last Friday was very similar to ones that
have occurred all over the country.
Another interesting point is that Karmarkar has ignored some
of the past research done in similar areas. In repeated
conversations with members of the O.R. community, he has refused
to acknowledge some of the early results done in the field.
Some of us feel that there is a loss of scholarship here, and
we're not too happy about it.
Irvin Lustig
O.R. Dept.
------------------------------------------------------------------------
David G. Cantor
ARPA: dgc@ucla-locus.arpa
UUCP: ...!{ihnp4, randvax, sdcrdcf, ucbvax}!ucla-cs!dgc