[SciPy-dev] scipy.optimize.nonlin rewrite

Bill Baxter wbaxter@gmail....
Mon Dec 8 14:01:40 CST 2008


On Tue, Dec 9, 2008 at 4:51 AM, Ondrej Certik <ondrej@certik.cz> wrote:
> On Mon, Dec 8, 2008 at 8:38 PM, Robert Kern <robert.kern@gmail.com> wrote:
>> On Mon, Dec 8, 2008 at 12:42, Ondrej Certik <ondrej@certik.cz> wrote:
>>> On Mon, Dec 8, 2008 at 7:11 PM, Pauli Virtanen <pav@iki.fi> wrote:
>>>> Hi Ondrej,
>>>>
>>>> Mon, 08 Dec 2008 18:43:47 +0100, Ondrej Certik wrote:
>>>>> First let me apologize for taking me so long to reply. I wrote this code
>>>>> in the first place and I am happy that Pauli has rewritten it. I agree
>>>>> with the general direction, but I think this change should not go into
>>>>> 0.7.0, as it changes the interface and it is not well tested yet.
>>>>> Also, you renamed all the working broyden implementations that I use as
>>>>> BadBroyden, so I suggest to name them GoodBroyden, more below.
>>>>
>>>> Quick comment (I'll post a more thorough reply later on). The "good" and
>>>> "bad" Broyden's method are names referring to specific ways to update the
>>>> Jacobian (at least these were the names I learned here in a univ.
>>>> course), cf. also [1]; they do not really refer to goodness or badness of
>>>> the methods, and definitely not to quality of implementation. (If you
>>>> meant I had mislabeled one of these, please correct me.)
>>>>
>>>> .. [1] http://en.wikipedia.org/wiki/Broyden%27s_method
>>>
>>> Ah, ok ---  that wiki didn't exist yet when I wrote this. I only knew
>>> first Broyden method and a second Broyden method. Well, still I think
>>> it's weird to call something that works well by BadBrodyen, but if
>>> that's what people are used to, then ok. Do you have some good
>>> reference of this, besides wiki?
>>>
>>> And in any case, all of this should be explained in the docstrings.
>>
>> If there's an alternate set of names, I would suggest going with
>> those. "Bad" and "Good" are simply going to confuse people.
>
> In fact, I would suggest to use or even invent alternate names, rather
> than to use Bad and Good, it's really confusing.

This page seems to suggest that "bad Broyden" is used to refer to the
inverse update using Sherman Morrison, which is what the Wikipedia
calls "good Broyden".
http://www.math.buffalo.edu/~pitman/courses/mth437/na3/node3.html

There's also a paper called "On The Discovery of the "Good Broyden"
method by Broyden himself which I presume would be the final authority
on which method is which:
http://www.springerlink.com/content/cvvygkeyv8blm96r/
(Needs Springer access to download ... which I don't have.)

--bb


More information about the Scipy-dev mailing list