Groups | Search | Server Info | Login | Register

Groups > comp.lang.forth > #23463

Re: FSL contribution available for review

Followups directed to: comp.lang.forth

Show all headers | View raw

Marcel Hendrix mhx@iae.nl wrote:

> C G Montgomery <cgm@physics.utoledo.edu> writes:
> 
>> From Hans Bezemer, a version of the error function which is
>> intermediate, in precision and speed, between the versions provided in
>> Algorithm #62. (The algorithm used led to some discussions in
>> comp.lang.forth in October, 2011.)
> 
> The submitted code is definitely and unnecessarily weird.
> I suggest:
> 
> create _L   7.618009172947146e1   f,  -8.650532032941677e1  f,
> 2.401409824083091e1   f,  -1.231739572450155e   f,
> 0.1208650973866179e-2 f,  -0.5395239384953e-5   f,
> 
> : gammaln_new                          ( f1 -- f2)
>   1.000000000190015e fover
>   6 0 do  1e f+ _L i floats + f@ fover f/ frot f+ fswap loop fdrop
>   fover fdup 5.5e f+ fdup fln
>   frot 0.5e f+  f*  f- fnegate fswap
>   2.5066282746310005e f* frot f/ fln f+ ;
> 
> A test, comparing gammaln_new to the zgamma results:
> 
> : test ( -- )
>    1e 0e FLOCALS| sum arg |
>    #30 0  DO  	CR ." \ "
>    arg +e. 2 spaces
>    arg 0e R,I->Z zgamma REAL FLN fdup   arg gammaln_new  f- fswap f/ +e.
>    1e +TO arg
>         LOOP ;
> \                  x                      gammaln_new
> \  1.0000000000000000000e+0000  -2.6619999999999999994e+0002
> \  2.0000000000000000000e+0000   1.4566666666666666671e+0002
> \  3.0000000000000000000e+0000   1.5156837315413355634e-0016
> \  4.0000000000000000000e+0000   5.8634650642649096784e-0017
> \  5.0000000000000000000e+0000   3.2136599976015110008e-0017
> \  6.0000000000000000000e+0000   2.0019558716152115772e-0017
> \  7.0000000000000000000e+0000  -2.7457496325686340266e-0015
> \  8.0000000000000000000e+0000  -1.0006774080483571122e-0014
> \  9.0000000000000000000e+0000  -2.2116712234461730942e-0014
> \  1.0000000000000000000e+0001  -3.8521827653372058092e-0014
> \  1.1000000000000000000e+0001  -5.8262560404684105464e-0014
> \  1.2000000000000000000e+0001  -8.0298384254259491232e-0014
> \  1.3000000000000000000e+0001  -1.0367055194912764301e-0013
> \  1.4000000000000000000e+0001  -1.2757593891011606686e-0013
> \  1.5000000000000000000e+0001  -1.5138223371142210353e-0013
> \  1.6000000000000000000e+0001  -1.7461517233283822270e-0013
> \  1.7000000000000000000e+0001  -1.9693448721728147118e-0013
> \  1.8000000000000000000e+0001  -2.1811326046325099310e-0013
> \  1.9000000000000000000e+0001  -2.3800667844593437190e-0013
> \  2.0000000000000000000e+0001  -2.5653866524181842926e-0013
> \  2.1000000000000000000e+0001  -2.7367993864864199716e-0013
> \  2.2000000000000000000e+0001  -2.8944059304657091458e-0013
> \  2.3000000000000000000e+0001  -3.0385158276675865048e-0013
> \  2.4000000000000000000e+0001  -3.1696484957374142414e-0013
> \  2.5000000000000000000e+0001  -3.2884041033705138642e-0013
> \  2.6000000000000000000e+0001  -3.3954736559091735848e-0013
> \  2.7000000000000000000e+0001  -3.4915706165453221168e-0013
> \  2.8000000000000000000e+0001  -3.5774134954325300864e-0013
> \  2.9000000000000000000e+0001  -3.6537242328887360372e-0013
> \  3.0000000000000000000e+0001  -3.7211905235738305660e-0013 ok
> 
> -marcel

Let me clarify a possible confusion.  The code contribution I was 
referring to is not the code recently posted for the gamma and beta 
functions. It is an implementation of the error function, and also differs 
somewhat from the code discussed in 2011.  (A reviewer for the error 
function submission has not yet been selected, although I don't hesitate 
to say that Marcel would certainly be well qualified to do that review.)

I think it is possible that Hans Bezemer may decide to submit a 
contribution for the beta function algorithms that would be related to the 
code posted here on June 8.  It may include a version of the log-gamma 
function.  He has not yet submitted such a contribution.  If he does so, I 
will be inviting a reviewer for that submission.  

Of course, discussion of the posted code for the log of the gamma function 
is certainly on-topic for comp.lang.forth and is appreciated.  But it is 
not code that is involved in my invitation for a reviewer, and I felt I 
should make that clear.

regards to all   cgm

Back to comp.lang.forth | Previous | Next — Previous in thread | Find similar

Thread

FSL contribution available for review C G Montgomery <cgm@physics.utoledo.edu> - 2013-06-09 19:43 -0400
  Re: FSL contribution available for review mhx@iae.nl (Marcel Hendrix) - 2013-06-11 23:29 +0200
    Re: FSL contribution available for review C G Montgomery <cgm@physics.utoledo.edu> - 2013-06-11 19:38 -0400

csiph-web