Hi Mike,
thanks. Maybe fixed in SVN 667. I am now using ⎕CT
(instead of a fixed constant 1E¯10) as
the condition number in ⌹. This number determines
if a matrix is considered singular or not.
Neither the ISO standard nor the IBM APL2 manual says anything
about that aspect, so I hope
it is OK now.
/// Jürgen
On 08/25/2015 08:30 PM, Mike Duvos
wrote:
GNU APL's domino gives a domain error for matrices
that are close to singular, but aren't.
)clear
CLEAR WS
A←100 100⍴101↑1
A[1;1]←0
[IBM APL2]
10 10 ↑⌹A+⎕CT×100
1.000000001E11 ¯1 ¯1 ¯1 ¯1 ¯1 ¯1 ¯1 ¯1 ¯1
¯1.000000000E0 1 0 0 0 0 0 0 0 0
¯1.000000000E0 0 1 0 0 0 0 0 0 0
¯1.000000000E0 0 0 1 0 0 0 0 0 0
¯1.000000000E0 0 0 0 1 0 0 0 0 0
¯1.000000000E0 0 0 0 0 1 0 0 0 0
¯1.000000000E0 0 0 0 0 0 1 0 0 0
¯1.000000000E0 0 0 0 0 0 0 1 0 0
¯1.000000000E0 0 0 0 0 0 0 0 1 0
¯1.000000000E0 0 0 0 0 0 0 0 0 1
[GNU APL]
10 10 ↑⌹A+⎕CT×100
DOMAIN ERROR
10 10↑⌹A+⎕CT×100
^^
A matrix whose determinant is 1E¯11 is obviously not singular.
|
