[Date Index] [Thread Index] [Author Index]

Re: Machine-precision Exp[] function

• To: mathgroup at smc.vnet.net
• Subject: [mg93624] Re: Machine-precision Exp[] function
• From: Roman <rschmied at gmail.com>
• Date: Tue, 18 Nov 2008 07:22:20 -0500 (EST)
• References: <gfma98$g54$1@smc.vnet.net>

Daniel Lichtblau kindly pointed out a very simple solution: with
  SetSystemOptions["CatchMachineUnderflow" -> False]
the default behavior of switching to arbitrary-precision numbers can
be suppressed. After the above command, we get
  In[1] := Exp[-1`*^6]
  Out[1] = 0.`
and the execution speed is identical to that of the compiled command.

There are some hints in this direction in the help file under
$MinMachineNumber, but they seem outdated (Version 6.0.2.0) as they
invoke the command
  Developer`SetSystemOption["CatchMachineUnderflow" -> False]
which no longer exists.

Roman.