Could it be....?
- To: mathgroup at smc.vnet.net
- Subject: [mg33746] Could it be....?
- From: "Steve Gray" <stevebg at adelphia.net>
- Date: Wed, 10 Apr 2002 00:50:27 -0400 (EDT)
- Sender: owner-wri-mathgroup at wolfram.com
I have been doing some symbolic matrix computations in which the
product of a bunch of square matrices may or may not have all rows
equal. To avoid a very slow FullSimplify of the final product (this program
has to run for hundreds of hours) to see if all the rows are equal, I gave
the only variable in the computation a numeric value after the product
was computed. In other words, the way I wrote it, the matrix products
are done symbolically but the final product is tested numerically.
I observe some odd behavior in the timing of the matrix multiplies.
Is it possible that Mathematica, after noting that the final
product is used only for numeric evaluation, decides on its own to do
the multiplies numerically to speed things up? If so, is it possible
that it decides this only after the code has been run a few times?
Sounds nuts, maybe, but this would explain a lot. (When I look at
the matrix products, they are symbolic.)
I would love to solve this mystery. Thanks for any information.
Steve