A precision problem on elliptic123

[tangjinchuan]

Dear virogr,
there might be a problem with the precision of elliptic123. I run the following code on 64bit linux version of Matlab:
K>> [~, K1]= elliptic123(pi/2, 1+2.*eps(0.5))

K1 =

  NaN +    NaNi

K>> [~, K1]= elliptic123(pi/2, 1)

K1 =

 1

K>> ellipticE(pi/2, 1+2.*eps(0.5))

ans =

 1

K>> ellipticE(pi/2, 1)

ans =

 1

Thanks!
Best wishes,
Jinchuan

[wspr]

This was my code originally (I think)... I’ll see what I can do. Thanks!

[tangjinchuan]

Dear Will,
Thanks for the response. I wish the problem will be fixed soon.

Best wishes,
Jinchuan

[tangjinchuan]

An update:
I could locate e=0 in elliptic12ic when calling [~, K1]= elliptic123(pi/2, 1+2.*eps(0.5)). This may cause the NaN problem.

Currently, I have added a nasty line to avoid the NaN problem in function elliptic123, which is just a temporary solution.

if nargout<3
  
  if nargin==1
    [F,E] = elliptic12c(a1);  % == elliptic12(m) 
  elseif nargin==2
    a2(abs(a2-1)<4.6*eps(1))=1; % Added by Jinchuan
    [F,E] = elliptic12x(a1,a2); % == elliptic12(b,m)

I hope this could help a little bit.

Best wishes,
Jinchuan