Brent Worden wrote:
Gamma.java specifically mentioned that is was based on "Numerical
Recipes in C" which is not available under a license compatible with the
ASL.


What is specifically mentioned was "is based on the formulas and
descriptions" found in NR and it clearly states in
http://www.nr.com/com/info-copyright.html that "Of course, the mathematical
algorithms that underly the programs are not copyrightable..."

This is a good point. The full text of the paragraph that starts with that sentence is:


"Of course, the mathematical algorithms that underly the programs are not copyrightable. So, the test of whether a similar program is actually infringing on the Numerical Recipes copyright is whether the expression of the underlying algorithm is so similar to that in Numerical Recipes as to make clear that the infringing program was in fact derived from the Numerical Recipes program. It is usually very easy to tell if something was copied from Numerical Recipes, even if variable names are changed (for example), since all programs are full of arbitrary stylistic choices as to the order in which things are done, the way expressions are written, etc."

I am not a lawyer and I have learned that lawyers often see things very differently than I do, but to me it looks like as long as all we are using are the formulas in NR, we *might* be OK, as long as:

* "our" implementation does not end up looking like it was "derived from
  the NR program"
* the formulas that we use are part of the "mathematical algorithms"
  rather than part of the "expression of the algorithms"

The best situation,IMHO, is to have non-NR references that describe the algorithms. If NR itself provides these, we should refer to the original source (cf earlier discussion of corrected mean computations). I would suggest that given the potential legal entanglements, we should try to avoid formulas or algorithms that we can find *only* in NR.

Phil




1) logGamma is derived using (6.1.3) and (6.1.5) from NR and the Paul Godfrey's paper on Lanczos coefficients found at http://my.fit.edu/~gabdo/gamma.txt.

2) regularizedGammaP is derived using (6.2.1) and (6.2.5) from NR.  I also
have hand-written notes on how I derived the implementation from those
formulas that I would be happy to provide for your viewing.

3) ChiSquaredDistribution and GammaDistribution are based on (6.2.1) and
(6.2.18) from NR and the fact that ChiSquared(v) ~ Gamma(v/2, 2) which can
be found in Casella and Berger.

The t-distribution patch has several modifications to the comments of all
the mentioned classes with links to all the cited material.

Also, the t patch makes use of other mathematical algorithms found in NR.

A) The continued fraction class is based on (5.2.1) and the modified Lentz
algorithm defined in that same section and accredited to Thompson and
Barnett.

B) logGamma uses the cf class and (6.2.7) from NR to provide better
convergence than the series.

C) logBeta uses (6.1.9) from NR.

D) regularizedBeta is derived using the cf class and (6.4.1), (6.4.5) and
(6.4.6) from NR.

E) TDistribution is based on (6.4.9) from NR.

Brent Worden
http://www.brent.worden.org


--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]





---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Reply via email to