comppi



Am 4 Nov 2007 um 21:53 hat Robert Dodier geschrieben:

> Let's put some tests for this new code in the test suite.

The problem we had with ceiling(%i^%i), which I believe is fixed by your change in 
nummod.lisp, showed that there might be some more unexpected interferences. 

I hope, I wasn't too quick. Definitely all interferences with PI and E should be tested. 
Which is the test file for things concerning this subject?

> I suppose this means we will need to identify a canonical source
> of pi and e digits .... Also we need to be careful about bigfloat
> comparison in the test code. At present the test code allows a
> nonzero difference between expected and actual results.
> Perhaps we'll need to phrase some tests as is(expected=actual),
> i.e. test for literal equality. That may be too strict.

As written in my last mail (on the list) the results aren't exactly the same. There is a nonzero 
difference.

> Wikipedia claims that Bill Gosper computed pi to 17*10^6 digits
> via a Ramanujan formula in 1985. I think that was after his work
> on Macsyma so I guess that code must not be in Maxima.

Ramanujan and Chudnovsky are of type O(N^2). So with my machine (Pentium 600 Mhz) I 
expect 48 hours of computing for 17*10^6 digits.

Volker