Skip to content
  1. Apr 11, 2012
  2. Apr 09, 2012
  3. Apr 08, 2012
  4. Mar 24, 2012
  5. Mar 23, 2012
  6. Mar 22, 2012
  7. Mar 21, 2012
  8. Dec 06, 2011
    • Raymond Toy's avatar
      Better exp-integral-e computation and fix for incomplete-gamma-tail. · c388f817
      Raymond Toy authored
      For exp-integral-e, use the series for small z and the
      incomplete-gamma-tail for near the negative real axis.  Otherwise, use
      the continued fraction.
      
      In incomplete-gamma-tail, we were using the continued fraction instead
      of the incomplete-gamma function for the region just below the negative
      real axis.  We should use the cf except in that region.
      c388f817
  9. Dec 05, 2011
  10. Mar 29, 2011
    • Raymond Toy's avatar
      Make gamma accurate for integers; add precision test for · af9fc6fc
      Raymond Toy authored
      incomplete-gamma-tail. 
      
      qd-gamma.lisp:
      o For integer values, just compute the gamma value directly by
        multiplication.  This works around the problem that the current
        algorithm is not as accurate as we would like.
      
      rt-test.lisp:
      o Reduce required accuracy in gamma-incomplete-tail.3.d.
      o Add precision test for gamm incomplete tail near the negative real
        axis.
      af9fc6fc
    • Raymond Toy's avatar
      Make INCOMPLETE-GAMMA-TAIL more accurate. · 08d25437
      Raymond Toy authored
      o Use the new continued fraction for the incomplete-gamma when the
        argument z is close enough to the negative real axis.
      08d25437
    • Raymond Toy's avatar
      More accurate incomplete-gamma function, add debugging to lentz, and · c31b1cdd
      Raymond Toy authored
      some random clean ups.
      
      o Add *DEBUG-CF-EVAL* to enable debugging prints in LENTZ.
      o Modify LENTZ to terminate with an error if *MAX-CF-ITERATIONS* is
        reached.
      o Modify LENTZ to return the function value, the number of iterations,
        and the number of times a zero value had to be replaced.
      o Adjust cf-incomplete-gamma and cf-incomplete-gamma-tail not to
        signal overflow prematurely when calculating z^a*exp(-z).
      o Fix doc bug in reference for continued fraction for (original)
        cf-incomplete-gamma.
      o Add new version of cf-incomplete-gamma using a different continued
        fraction.  This appears to converge faster and to be more accurate
        than the original, especially for points near the negative real
        axis.
      c31b1cdd
  11. Mar 25, 2011
    • Raymond Toy's avatar
      Add DOMAIN-ERROR condition; fix bug in FRESNEL-S-SERIES. · 7fa2a4c3
      Raymond Toy authored
      qd-methods.lisp:
      o Define DOMAIN-ERROR condition to allow signaling errors for
        incorrect domains.
      
      qd-gamma.lisp:
      o Signal domain error in CF-INCOMPLETE-GAMMA-TAIL if necessary. 
      o Fix bug in FRESNEL-S-SERIES.  We were comparing a real against a
        complex.
      7fa2a4c3
  12. Mar 24, 2011
    • Raymond Toy's avatar
      Fix incomplete-gamma-tail for negative reals, use series for Fresnel S · bed5763d
      Raymond Toy authored
      for small arg and update tests.
      
      qd-gamma.lisp:
      o INCOMPLETE-GAMMA-TAIL was hanging for arguments on the negative real
        axis.  Use INCOMPLETE-GAMMA in this case too.
      o Add the series expansion for Fresnel S and use it for evaluating it
        for small arguments.  We were losing accuracy with the existing
        algorithm.
      
      rt-tests.lisp:
      o Update thresholds for elliptic-pi-n0.d, elliptic-pi.n2.q,
        theta3.1.d.
      o Fix typo in test name.  gamma-incomplete-tail.1.q should have been
        2.q.
      o Add tests for gamma-incomplete-tail for arguments on the negative
        real axis.
      o Add tests for Fresnel S.
      bed5763d
  13. Mar 18, 2011
  14. Mar 17, 2011
  15. Mar 16, 2011
    • Raymond Toy's avatar
      Add gamma and log-gamma functions; work in progress. · a40062c6
      Raymond Toy authored
      oct.asd:
      o Add qd-gamma.lisp.  The implementations need some work.  The
        accuracy is less than desired because gamma(2.0) /= 1.  It's close
        but not quite right.
      
      rt-tests.lisp:
      o Basic tests of the gamma function.  Accuracy is not as good as we
        would ike.
      
      qd-gamma.lisp:
      o New file for implementation of gamma function.
      a40062c6