Standard Tests for Calculator Accuracy?

[Steve S]

Does anyone here know if there is a standard or protocol for assessing the accuracy of calculating devices? What I'm thinking of is a series of problems, for each of the functions that one would typically find on a technical calculator, that could be used to assess the acuracy of the function's algorithm (or at least it's implementation).

If such a standard or protocol exists, where would I look for it?

Thanks in advance for any information...

[db (martinez, ca.)]

The only standard one i know of is Mike Sebastian's "calculator forensics project". It is limited to what was his begining interest, trig accuracy. A couple of people here have pointed out (perhaps rightly) it's flaws, but have not done the work to design an improvement or implement a data base for a better one so.....

[Gerson W. Barbosa]

There are also Hugh Steers's torture tests.

[db (martinez, ca.)]

Interesting site Gerson, i liked "The Secret Life of the MOD Function" too. I must admit to clicking on the link with a bit or trepidation though. With that name i was expecting long falls onto concrete and big hammers.

[Steve S]

Gerson & dB - Thanks for your responses. I already know about both of the webpages that you point to. What I was hoping to find was a more formal set of tests, for example something from NIST or the IEEE. Does anyone know of such a thing?

[John Limpert]

You might find this of interest:

http://www.netlib.org/paranoia/

[Steve S]

John - Thanks for your comment. Because the tests on this webpage appear to be in C and Fortran, I'm not sure that they would be of much use for a calculator. The underlying algorithm, however, might be worthwhile if I can I can decode it. If nothing else turns up, I may try that. Thanks again for this reference...!

[Mike (Stgt)]

See 20.1 Diagnosing Machine Parameters in Numerical Recipes. You only have to migrate that FORTRAN to your calculator (but you know, a good FORTRAN porgammer may do FORTRAN in any language).

Ciao.....Mike

[Steve S]

Mike - I'll take a look at that section. Thanks for the reference.

I've also been looking through the IEEE website; lots of hits, but I haven't yet stumbled onto quite the right set of keywords...

[John Smitherman]

Here is a link to NIST's software page which contains data sets which can be used for calibrating algorithms for statistics and sparsely populated matrices:

http://math.nist.gov/

This may help but is only part of what I think you are looking for.

Regards,

John

[Steve S]

John - Thanks! I'll take a look at this. Still haven't found much at the IEEE website...