|
== Revision history of libcerf, maintained by Joachim Wuttke ==
libcerf-1.2, released 16jul13: - test programs no longer install to $bindir (reported by M. Miclavec). libcerf-1.1, released 12may13: - Added Fortran binding by Antonio Cervellino. libcerf-1.0, released 31jan13 by Joachim Wuttke: - Based on http://ab-initio.mit.edu/Faddeeva as of 28jan13. - Verified accuracy using double-exponential transform. - Simplified function names; use leading 'c' for complex functions (except in w_of_z). - Added function voigt(x,sigma,gamma). - Added configure.ac, Makefile.am &c to allow for autotools standard installation (commands ./configure, make, sudo make install). - Splitted source code into directories lib/ and test/. - Eliminated unused alternate code (!USE_CONTINUED_FRACTION). - Eliminated relerr arguments. - Replaced "complex" by "_Complex" for C++ compatibility. - Wrote man pages w_of_z(3), dawson(3), voigt(3), cerf(3), erfcx(3), erfi(3). - Created project home page http://apps.jcns.fz-juelich.de/libcerf. - Registered project "libcerf" at sourceforge.net. == Revision history of Faddeeva.cc by Steven G. Johnson == Project at http://ab-initio.mit.edu/Faddeeva 4 October 2012: Initial public release (SGJ) 5 October 2012: Revised (SGJ) to fix spelling error, start summation for large x at round(x/a) (> 1) rather than ceil(x/a) as in the original paper, which should slightly improve performance (and, apparently, slightly improves accuracy) 19 October 2012: Revised (SGJ) to fix bugs for large x, large -y, and 15<x<26. Performance improvements. Prototype now supplies default value for relerr. 24 October 2012: Switch to continued-fraction expansion for sufficiently large z, for performance reasons. Also, avoid spurious overflow for |z| > 1e154. Set relerr argument to min(relerr,0.1). 27 October 2012: Enhance accuracy in Re[w(z)] taken by itself, by switching to Alg. 916 in a region near the real-z axis where continued fractions have poor relative accuracy in Re[w(z)]. Thanks to M. Zaghloul for the tip. 29 October 2012: Replace SLATEC-derived erfcx routine with completely rewritten code by me, using a very different algorithm which is much faster. 30 October 2012: Implemented special-case code for real z (where real part is exp(-x^2) and imag part is Dawson integral), using algorithm similar to erfx. Export ImFaddeeva_w function to make Dawson's integral directly accessible. 3 November 2012: Provide implementations of erf, erfc, erfcx, and Dawson functions in Faddeeva:: namespace, in addition to Faddeeva::w. Provide header file Faddeeva.hh. 4 November 2012: Slightly faster erf for real arguments. Updated MATLAB and Octave plugins. 27 November 2012: Support compilation with either C++ or plain C (using C99 complex numbers). For real x, use standard-library erf(x) and erfc(x) if available (for C99 or C++11). #include "config.h" if HAVE_CONFIG_H is #defined. 15 December 2012: Portability fixes (copysign, Inf/NaN creation), use CMPLX/__builtin_complex if available in C, slight accuracy improvements to erf and dawson functions near the origin. Use gnulib functions if GNULIB_NAMESPACE is defined. 18 December 2012: Slight tweaks (remove recomputation of x*x in Dawson) |