[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
fun with random numbers
- Subject: fun with random numbers
- From: Brian Jackel <bjackel(at)phys.ucalgary.ca>
- Date: Wed, 01 Mar 2000 10:04:54 -0700
- Newsgroups: comp.lang.idl-pvwave
- Organization: University of Calgary
- Xref: news.doit.wisc.edu comp.lang.idl-pvwave:18735
We've just spent a couple hours figuring out why a Monte-Carlo
simulation was giving peculiar (ie. wrong) results.
The "test" procedure creates two random variables and prints
them. Called twice, you might think that the results would be
totally different. Here's an example result:
y 0.649213 0.577647 0.139354 0.745442
y 0.577647 0.139354 0.745442 0.942317
The problem is that for the first random number, "seed" is
undefined, so a generic state is used. From the docs:
"In addition to states maintained by the user in variables,
the RANDOMU and RANDOMN functions contain a single shared
generic state that is used if a named variable is not supplied
as the Seed argument or the named variable supplied is undefined.
The generic state is initialized once using the time-of-day, and
may be re-initialized by providing a Seed argument that is a
constant or expression."
Which is all fine. This generic state is used to produce the
first random number, the seed is updated, then apparently used
to overwrite the generic state. Another four random numbers are
produced, with the seed changing every time. However, none
of these values are used to overwrite the generic state.
Consequently, on the next call, we start with a generic state
that is only advanced *one* from the previous call. Hence the
sequence of "y" values starts at the second position of the
previous "y" sequence.
I would assert that this is not good behaviour. The "generic
1) always be overwritten with the last seed OR
2) should never be changed without an explicit call by the user