MATLAB computes Incomplete Elliptic Integrals incorrectly?

Discussion in 'MATLAB' started by Ishan Sharma, Apr 24, 2010.

  1. Ishan Sharma

    Ishan Sharma Guest

    I compute the Incomplete Elliptic Function of the second kind in both MATLAB and Maple. For some values, I obtain an exact match, but not so for others. Please see below:

    CASE 1: Perfect match
    MATLAB output
    MAPLE13 output: 0.9863238216

    CASE 2: No match.
    MATLAB output
    MAPLE13 output: 0.9495803244

    Can someone throw light on this?


    Ishan Sharma, Apr 24, 2010
    1. Advertisements

  2. You are mixing floating point numbers and exact numbers, which often
    leads to problems. You might get a different result (or not) if you were
    to use something like,

    mfun('EllipticE', sqrt(1-sym('4/10')^2),

    I converted a 10th order series of EllipticE with the same first
    argument, and the second near 9/10 (one of your second arguments is
    below that, the other above that). Some large numbers are involved --
    large enough to present precision problems if the second argument was
    not too much bigger (relatively.) Even so, the default 10 digits of
    precision within mfun are enough that one can clearly see that in that
    range of values, the curve approximates linear fairly closely, which the
    Maple 13 answer is consistent with and the MuPad answer is not.

    I would thus say that you have found a bug.
    Walter Roberson, Apr 24, 2010
    1. Advertisements

  3. ----------
    I don't know the parameters MAPLE13 uses for elliptic integrals but you should check on it. Even within Mathworks there are two conflicting definitions used. Here are two excerpts from Mathworks manuals:

    "Some definitions of K and E use the elliptical modulus k instead of the
    parameter m. They are related as k^2 = m."

    "This definition uses modulus k. The numerical ellipke function and the MuPAD functions for computing elliptic integrals use the parameter m = k^2."

    Also I would advise you to display your matlab results to a higher accuracy by using 'format long', so the comparisons will be more valid. You might find that your "perfect match" was not so perfect.

    Roger Stafford
    Roger Stafford, Apr 24, 2010
  4. After consulting some old tables of elliptic integrals I agree with Walter. The answer 1.4182 you quote as being given by 'EllipticE' is way off the correct value for the arguments you state you have given it. That answer would only be correct for a k value well below .6 and an x value much closer than yours to 1. After double checking your answers you should report this to Mathworks.

    Roger Stafford
    Roger Stafford, Apr 24, 2010
  5. Ishan Sharma

    moiseev.igor Guest

    You may try to use the "elliptic12" function from the elliptic package

    Just get the single file elliptic12.m, put it in your PATH and it
    gives the results for incomplete ellip. integrals F and E.

    The only difference with Maple is that it calculates the integrals
    with different to maple representation,
    check the sample code for "asin" and square of "sqrt((1-.6^2)/(1-.
    4^2))" to obtain the same results!

    [F,E] = elliptic12(asin(sqrt(1-.4^2)),(1-.5^2)/(1-.4^2))
    F =
    E =
    moiseev.igor, Apr 27, 2010
  6. Ishan Sharma

    Daniel Guest

    Can I ask how well this package has been validated?
    Being the only (if not the only) full elliptic function package in Matlab which is available to my knowledge, I wanted to know how well it works? ... feedback from users etc...? Couldn't find much about this online...
    Daniel, Dec 7, 2011
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.