While running the test suite with cmucl, I've come across a couple of
minor bugs.
First, in test 248 in rtest14, I get the failure:
********************** Problem 284 ***************
Input:
2.0
test_wronskian(lambda([n, z], w_jy(n, z %i)), lambda([n, z], --------),
z %i %pi
1.0e-8, 10)
Result:
4
[7.312360852040531e-6, [[- -, - 8], .07957015918509569 %i
5
- 2.91038304567337e-11, .07957747154594767 %i,
7.312360852040531e-6]]
Basically, this fails at w_jy(-4/5,-8*%i). Given that the terms are on
the order of 141309-102667*%i, I think it would be pretty hard to expect
an absolute accuracy of 1e-8.
The second bug is in bessel_y with clisp.
(%i17) float(bessel_y(-4/5,-8*%i));
(%o17) 389.5687067045917 - 126.5784634720102 %i
(%i18) :lisp $%o17
((MPLUS SIMP) 389.5687067045917291L0 ((MTIMES SIMP)
-126.57846347201015544L0 $%I))
Thus, %o17 is printed out correctly, but the actual internal
representation is using clisp's long-float support. I guess technically
this is not wrong, but is probably not quite what we want. There's
probably a plain pi floating around, since pi is a long-float on clisp.
Ray