MathGroup Archive 2009

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

Search the Archive

Re: Maintaining a Mathematica bug list

Dear all,
Murray's observation that "one person's bug is another person's
feature" made me think that insisting on the word "bug" is probably

To explain what I think, I will resort to an example.

A few days ago, I used Minimize on a numeric function and, after some
testing, I discovered that NMinimize with method "Simulated Annealing"
found a better solution, so that Minimize's result was wrong.

I wrote this to Daniel Lichtblau (Wolfram) and he kindly told me: this
is documented: Minimize on a numeric function calls NMinimize. (What
is not stated is that it call NMinimize with Method->Automatic).
So I replied: ok, but than it should be possible to specify a Method
also in Minimize, in case this happens (this is not possible now).
So Daniel told me: use SetOptions[NMinimize]={Method->"Simulated
Annealing"} and you get what you want.

The moral of this is manifold:
1) Now I know how to handle the situation
2) Daniel is right: this is not a bug and there is a way out in
Mathematica as it is
3) In general terms, I am not very happy, because if I did not make
some tests I would never realized that Minimize returned a wrong value
4) I tend to consider symbolic functions as a way to provide the best
possible result: when this does not happen I start trembling...
4) Like me, other users may have fallen in the same trap.

Now, calling this a bug is misleading.
It should more probably called a "Caveat when using Minimize on
numerical function".

What is difficult now is to find a place where caveats (and genuine
bugs) are collected and can be quickly searched before starting
complex or serious activities. This place should also collect the
discussions which ended with the words "You should report this as a
bug to Wolfram".

I think that we really need is this: a searchable wiki collecting
caveats and genuine bugs.

How to create this in practice I do not know. Perhaps is the solution or there is a better one. In any
case, I believe it is useful and, if required, I will contribute to
this achievement.


  • Prev by Date: Re: Compile for different parameter types
  • Next by Date: Re: Mathematica and some General Comments
  • Previous by thread: Re: Maintaining a Mathematica bug list
  • Next by thread: Re: Maintaining a Mathematica bug list