"Fossies" - the Fresh Open Source Software Archive

Member "gsl-2.7.1/NEWS" (1 Dec 2021, 73866 Bytes) of package /linux/misc/gsl-2.7.1.tar.gz:


As a special service "Fossies" has tried to format the requested text file into HTML format (style: standard) with prefixed line numbers. Alternatively you can here view or download the uninterpreted source code file. See also the latest Fossies "Diffs" side-by-side code changes report for "NEWS": 2.7_vs_2.7.1.

    1 * What is new in gsl-2.7.1:
    2 
    3 ** update libtool version numbers
    4 
    5 * What was new in gsl-2.7:
    6 
    7 ** fixed doc bug for gsl_histogram_min_bin (lhcsky at 163.com)
    8 
    9 ** fixed bug #60335 (spmatrix test failure, J. Lamb)
   10 
   11 ** fixed bug #36577
   12 
   13 ** clarified documentation on interpolation accelerators (V. Krishnan)
   14 
   15 ** fixed bug #45521 (erroneous GSL_ERROR_NULL in ode-initval2, thanks to M. Sitte)
   16 
   17 ** fixed doc bug #59758
   18 
   19 ** fixed bug #58202 (rstat median for n=5)
   20 
   21 ** added support for native C complex number types in gsl_complex
   22    when using a C11 compiler
   23 
   24 ** upgraded to autoconf 2.71, automake 1.16.3, libtool 2.4.6
   25 
   26 ** updated exponential fitting example for nonlinear least squares
   27 
   28 ** added banded LU decomposition and solver (gsl_linalg_LU_band)
   29 
   30 ** New functions added to the library:
   31       - gsl_matrix_norm1
   32       - gsl_spmatrix_norm1
   33       - gsl_matrix_complex_conjtrans_memcpy
   34       - gsl_linalg_QL: decomp, unpack
   35       - gsl_linalg_complex_QR_* (thanks to Christian Krueger)
   36       - gsl_vector_sum
   37       - gsl_matrix_scale_rows
   38       - gsl_matrix_scale_columns
   39       - gsl_multilarge_linear_matrix_ptr
   40       - gsl_multilarge_linear_rhs_ptr
   41       - gsl_spmatrix_dense_add (renamed from gsl_spmatrix_add_to_dense)
   42       - gsl_spmatrix_dense_sub
   43       - gsl_linalg_cholesky_band: solvem, svxm, scale, scale_apply
   44       - gsl_linalg_QR_UD: decomp, lssolve
   45       - gsl_linalg_QR_UU: decomp, lssolve, QTvec
   46       - gsl_linalg_QR_UZ: decomp
   47       - gsl_multifit_linear_lcurvature
   48       - gsl_spline2d_eval_extrap
   49 
   50 ** bug fix in checking vector lengths in gsl_vector_memcpy (dieggsy@pm.me)
   51 
   52 ** made gsl_sf_legendre_array_index() inline and documented
   53    gsl_sf_legendre_nlm()
   54 
   55 * What was new in gsl-2.6:
   56 
   57 ** add BLAS calls for the following functions:
   58      - gsl_vector_memcpy
   59      - gsl_vector_scale
   60      - gsl_matrix_memcpy
   61      - gsl_matrix_transpose_memcpy
   62      - gsl_matrix_tricpy
   63      - gsl_matrix_transpose_tricpy
   64 
   65 ** deprecated functions gsl_linalg_complex_householder_hm and
   66    gsl_linalg_complex_householder_mh
   67 
   68 ** add unit tests for gsl_linalg_symmtd and gsl_linalg_hermtd
   69 
   70 ** multilarge TSQR algorithm has been converted to use the new Level 3 QR decomposition
   71 
   72 ** nonlinear least squares Cholesky solver now uses the new Level 3 BLAS
   73    method; the old modified Cholesky solver is still available under
   74    gsl_multifit_nlinear_solver_mcholesky and gsl_multilarge_nlinear_solver_mcholesky
   75 
   76 ** implemented Level 3 BLAS versions of several linear algebra routines:
   77      - Triangular matrix inversion
   78      - Cholesky decomposition and inversion (real and complex)
   79      - LU decomposition and inversion (real and complex)
   80      - QR decomposition (courtesy of Julien Langou)
   81      - Generalized symmetric/hermitian eigensystem reduction to standard form
   82 
   83 ** removed deprecated function gsl_linalg_hessenberg()
   84 
   85 ** renamed gsl_interp2d_eval_e_extrap() to gsl_interp2d_eval_extrap_e()
   86    to match documentation (reported by D. Lebrun-Grandie)
   87 
   88 ** renamed some of the gsl_sf_hermite functions to be more consistent
   89    with rest of the library, and deprecated old function names
   90 
   91 ** updated gsl_sf_hermite_func() to use a newer algorithm
   92    due to B. Bunck which is more stable for large x; also added
   93    gsl_sf_hermite_func_fast() which uses the faster Cauchy integral
   94    algorithm in the same paper by Bunck
   95 
   96 ** add gsl_vector_axpby()
   97 
   98 ** add un-pivoted LDLT decomposition and its banded
   99    variant (gsl_linalg_ldlt_* and gsl_linalg_ldlt_band_*)
  100 
  101 ** add binary search tree module (gsl_bst); based on GNU libavl
  102 
  103 ** remove -u flag to gsl-histogram
  104 
  105 ** updated spmatrix module
  106    - added routines and data structures for all types (float,uint,char,...)
  107    - added gsl_spmatrix_scale_columns() and gsl_spmatrix_scale_rows()
  108    - added gsl_spmatrix_add_to_dense()
  109    - more efficient reallocation of COO/triplet matrices (no longer rebuilds binary tree)
  110    - enhanced test suite
  111    - added gsl_spmatrix_min_index()
  112 
  113 ** add routines for banded Cholesky decomposition (gsl_linalg_cholesky_band_*)
  114 
  115 ** documented gsl_linalg_LQ routines and added gsl_linalg_LQ_lssolve()
  116 
  117 * What was new in gsl-2.5:
  118 
  119 ** doc bug fix in binomial distribution figure (Damien Desfontaines)
  120 
  121 ** added Wishart distribution (Timothée Flutre)
  122 
  123 ** added new module for digital filtering (gsl_filter); current filters include:
  124      Gaussian filter
  125      median filter
  126      recursive median filter
  127      impulse detection filter
  128 
  129 ** added new module for moving window statistics (gsl_movstat)
  130 
  131 ** added statistics functions:
  132      gsl_stats_median()
  133      gsl_stats_select()
  134      gsl_stats_mad()
  135      gsl_stats_mad0()
  136      gsl_stats_Sn_from_sorted_data()
  137      gsl_stats_Qn_from_sorted_data()
  138      gsl_stats_gastwirth_from_sorted_data()
  139      gsl_stats_trmean_from_sorted_data()
  140 
  141 ** added Romberg integration (gsl_integration_romberg)
  142 
  143 ** bug fix in deprecated functions gsl_multifit_wlinear_svd and
  144    gsl_multifit_wlinear_usvd (reported by Vlad Koli)
  145 
  146 ** documentation corrected to state that gsl_sf_legendre functions do
  147    not include Condon-Shortley phase by default
  148 
  149 ** bug fix in exponential fitting example when using larger number
  150    of points (reported by Anna Russo)
  151 
  152 ** changed internal workspace inside gsl_spmatrix to a union to
  153    avoid casting (suggested by Manuel Schmitz)
  154 
  155 ** bug fixes in ode-initval2 for very rare solver crashing cases:
  156    #52230 in msadams (reported by Michael Kaufman) and
  157    #52336 in msbdf (reported by Andrew Benson). As a fix,
  158    the maximum scaling of controlled step length was decreased
  159    from 5.0 to 4.9.
  160 
  161 ** add histogram2d figure to manual (was missing in 2.4)
  162 
  163 ** bug fix in gsl_spmatrix_add for duplicate input arguments
  164    (reported by Alfredo Correa)
  165 
  166 ** add support for negative arguments nu in gsl_sf_bessel_Jnu and
  167    gsl_sf_bessel_Ynu (Konrad Griessinger)
  168 
  169 ** better texinfo documentation for gsl_sf_hyperg functions
  170 
  171 ** fix vector and matrix fread/fwrite testing on windows systems
  172    when tmpfile() fails
  173 
  174 ** fix for rstat/test.c on PPC64 (reported by Adam Majer)
  175 
  176 * What was new in gsl-2.4:
  177 
  178 ** migrated documentation to Sphinx software, which has built-in
  179    support for latex equations and figures in HTML output
  180 
  181 ** add const to declaration of appropriate gsl_rstat routines
  182 
  183 ** bug fix for #45730: change gsl_sf_sin/cos to libm sin/cos
  184 
  185 ** fix Cholesky documentation regarding upper triangle on output
  186 
  187 ** added routines to compute integrals with fixed-point quadrature,
  188    based on IQPACK (Konrad Griessinger)
  189 
  190 ** added routines for Hermite polynomials, gsl_sf_hermite_*
  191    (Konrad Griessinger)
  192 
  193 ** added new nonlinear least squares example for fitting
  194    a Gaussian to data
  195 
  196 ** deprecated routines:
  197      gsl_sf_coupling_6j_INCORRECT
  198      gsl_sf_coupling_6j_INCORRECT_e
  199 
  200 ** deprecated routine 'gsl_linalg_hessenberg' (replaced
  201    by gsl_linalg_hessenberg_decomp)
  202 
  203 ** removed routines which were deprecated in v2.1:
  204      gsl_bspline_deriv_alloc
  205      gsl_bspline_deriv_free
  206 
  207 ** changed COD expression to Q R Z^T instead of Q R Z to
  208    be consistent with standard texts
  209 
  210 ** added check for nz == 0 in gsl_spmatrix_get
  211    (reported by Manuel Schmitz)
  212 
  213 ** permit zero-dimension blocks, vectors, matrics, subvectors,
  214    submatrices, and views of the above (bug #49988)
  215 
  216 ** added routine gsl_linalg_COD_lssolve2 for regularized
  217    least squares problems
  218 
  219 * What was new in gsl-2.3:
  220 
  221 ** bug fix in documentation for gsl_linalg_LU_refine
  222    (bug #49728, Joey De Pauw)
  223 
  224 ** added gsl_multifit_linear_tsvd and gsl_multifit_wlinear_tsvd
  225    to give user more control over cutoff for truncated SVD
  226 
  227 ** added routines for Generalized Cross Validation for
  228    regularized linear least squares
  229 
  230 ** improved rstat example program and added documentation for
  231    gsl_rstat_sd_mean (Jonathan Leto)
  232 
  233 ** added function gsl_multifit_linear_rank
  234 
  235 ** bug fix in nonlinear least squares when using data weights with
  236    finite-difference Jacobian
  237 
  238 ** add 2D subspace method for large systems (multilarge_nlinear)
  239 
  240 ** bug fix in gsl_ran_beta for small parameters
  241    (bug #47646, Yu Liu)
  242 
  243 ** bug fix in gsl_complex_tan for negative imaginary arguments
  244    (bug #47347, Yu Liu)
  245 
  246 ** doc bug fix: value of golden ratio
  247 
  248 ** fixed scaling issue in 2D subspace nonlinear least squares
  249    method
  250 
  251 ** optimize dogleg methods to calculate Gauss-Newton point
  252    only when needed
  253 
  254 * What was new in gsl-2.2.1:
  255 
  256 ** reverted gsl_linalg_cholesky_decomp to its previous behavior
  257    so it is backward compatible; new cholesky routine is
  258    gsl_linalg_cholesky_decomp1
  259 
  260 * What was new in gsl-2.2:
  261 
  262 ** updated gsl_linalg_cholesky_invert to use Level-2 BLAS
  263    and added function gsl_linalg_pcholesky_invert
  264 
  265 ** added functions gsl_linalg_tri_*_invert for inverting
  266    triangular matrices
  267 
  268 ** fix GSL_EIGEN_SORT_VAL_{ASC,DESC} for nonsymmetric
  269    eigensystems (Victor Zverovich)
  270 
  271 ** added complete orthogonal decomposition routines
  272    (gsl_linalg_COD)
  273 
  274 ** bug fix where median calculation wasn't reset in
  275    gsl_rstat_reset(); added gsl_rstat_quantile_reset() function
  276    (reported by Pedro Donato)
  277 
  278 ** added multivariate Gaussian random distribution
  279    gsl_ran_multivariate_gaussian (Timothée Flutre)
  280 
  281 ** added functions to estimate the 1-norm reciprocal condition
  282    number for various matrix factorizations:
  283      * gsl_linalg_cholesky_rcond
  284      * gsl_linalg_QRPT_rcond
  285 
  286 ** added functions gsl_linalg_QRPT_{lssolve,lssolve2} to
  287    compute least squares solutions with the QRPT decomposition
  288 
  289 ** added function gsl_permute_matrix()
  290 
  291 ** added modified Cholesky factorization (gsl_linalg_mcholesky)
  292    to handle symmetric indefinite matrices
  293 
  294 ** added pivoted Cholesky factorization (gsl_linalg_pcholesky)
  295    for ill-conditioned matrices
  296 
  297 ** rewrote (real) Cholesky decomposition to use
  298    a Level-2 blas algorithm instead of Level-1. Flop
  299    count is about the same but the code is much simpler
  300    and easier to follow
  301 
  302 ** completely rewritten nonlinear least squares module,
  303    including support for large problems; the user may
  304    now control the linear solver used, the trust region
  305    updating strategy, and the scaling method. In addition,
  306    support has been added for the geodesic acceleration
  307    step (Transtrum 2011) which can speed up convergence
  308    on a wide class of problems.
  309 
  310 ** added gsl_rstat_rms() for root mean square
  311 
  312 ** optimized lmniel nonlinear least squares solver
  313    (bug #46369)
  314 
  315 ** improved precision in Bessel K0/K1 near x = 2
  316    (Pavel Holoborodko, bug #47401)
  317 
  318 ** added support for compressed row storage sparse
  319    matrices (Alexis Tantet)
  320 
  321 ** bug fix in convergence check of hypergeometric 2F1
  322    function (bug #45926)
  323 
  324 ** added gsl_multilarge_linear_lcurve() to compute
  325    the L-curve for large linear systems
  326 
  327 ** updated multilarge normal equations method to use
  328    new Cholesky scaling for better numerical stability
  329 
  330 ** added scaling to Cholesky routines to reduce the
  331    condition number prior to factorization
  332 
  333 * What was new in gsl-2.1:
  334 
  335 ** added test suite for example programs
  336 
  337 ** bug fix when compiling with #undef GSL_DISABLE_DEPRECATED
  338 
  339 ** bug fix in setting libtool age versioning
  340 
  341 ** bug fix in gsl_multifit_wlinear()
  342 
  343 ** added gsl_multifit_linear_rcond() to compute reciprocal
  344    condition number of least squares matrix
  345 
  346 ** added gsl_multilarge module for large linear least squares
  347    systems
  348 
  349 * What was new in gsl-2.0:
  350 
  351 ** fixed bug #43258 for hypergeometric functions (Raymond Rogers)
  352 
  353 ** added L-curve analysis routines for linear Tikhonov regression
  354 
  355 ** add running statistics module
  356 
  357 ** added bilinear and bicubic interpolation (David Zaslavsky)
  358 
  359 ** added function gsl_multifit_robust_residuals to compute robust
  360    fit residuals
  361 
  362 ** added Steffen monotonic interpolation method (Jean-François Caron)
  363 
  364 ** added new nonlinear least squares solver 'lmniel' suitable for
  365    systems with large numbers of data
  366 
  367 ** nonlinear least squares solver now tracks the number of function
  368    and Jacobian evaluations, see example program for details
  369 
  370 ** the 'fdf' field of gsl_multifit_function_fdf is now deprecated
  371    and does not need to be specified for nonlinear least squares
  372    problems
  373 
  374 ** added extensive test suite to nonlinear least squares module,
  375    resulting in a few minor bug fixes; the routine
  376    gsl_multifit_fdfsolver_driver has been rewritten (with API change)
  377    to handle the various error codes of the lmsder iterate
  378    routine, resulting in a high level caller which is highly robust
  379    for a wide class of problems
  380 
  381 ** added support for sparse matrices, including a GMRES
  382    iterative linear solver
  383 
  384 ** added routines gsl_linalg_givens and gsl_linalg_givens_gv
  385    for Givens rotations
  386 
  387 ** added Tikhonov (ridge) regularization to least squares module
  388    (linear and nonlinear)
  389 
  390 ** removed unused argument 'n' from gsl_sf_ellint_D
  391 
  392 ** merged bspline_deriv_workspace into bspline_workspace to simplify
  393    bspline API; the functions
  394      gsl_bspline_deriv_alloc
  395      gsl_bspline_deriv_free
  396    are now deprecated and will be removed in a future release.
  397 
  398 ** merged ALF extension into GSL for associated Legendre functions;
  399    api has changed; consequently the functions:
  400      gsl_sf_legendre_Plm_array
  401      gsl_sf_legendre_Plm_deriv_array
  402      gsl_sf_legendre_sphPlm_array
  403      gsl_sf_legendre_sphPlm_deriv_array
  404      gsl_sf_legendre_array_size
  405    are now deprecated and will be removed in a future release.
  406 
  407 ** added function gsl_multifit_robust_weights to allow user to
  408    access the various weighting functions
  409 
  410 * What was new in gsl-1.16:
  411 
  412 ** fixed error in gsl_rng_fwrite where uninitialized padding
  413    bytes were being written (bug #39104)
  414 
  415 ** fixed error in gsl_block_alloc where padding bytes were not
  416    properly initialized (bugs #39101,#39102,#39103)
  417 
  418 ** fixed error in ntuple/test.c where padding bytes were not
  419    properly initialized (bug #39105)
  420 
  421 ** fixed triangle selection bug in gsl_sf_coupling_6j_e and
  422    gsl_sf_coupling_9j_e (bugs #39466 and #29606) (Håkan Johansson and
  423    Alexey Illarionov)
  424 
  425 ** added higher level wrapper routine gsl_multifit_fdfsolver_driver
  426 
  427 ** converted gsl_multifit_linear_residuals to use dgemv to improve
  428    efficiency (bug #39153)
  429 
  430 ** added functions gsl_stats_spearman and gsl_sort_vector2 to compute
  431    Spearman rank correlation
  432 
  433 ** added function gsl_poly_dd_hermite_init for Hermite interpolation
  434 
  435 ** Added support for robust linear least squares
  436 
  437 ** Added function gsl_linalg_SV_leverage for computing statistical
  438    leverages from SVD decomposition
  439 
  440 ** Added support for approximating the Jacobian of nonlinear least
  441    squares fits using forward finite differences
  442 
  443 ** Extended gsl_sf_coupling_3j to allow larger range and to handle
  444    the special case (ja jb jc; 0 0 0)=0 when ja+jb+jc is odd
  445 
  446 ** Fixed gsl_sf_mathieu_se_array to return zero when the order is zero
  447    [bug #33679].
  448 
  449 ** Fixed overflow in gsl_sf_lncosh for large negative x (x<-354).
  450 
  451 ** Improved gsl_ran_negative_binomial_pdf to avoid underflow/overflow
  452    for large arguments.
  453 
  454 ** Multisets now allow k strictly greater than n.
  455 
  456 ** Fixed gsl_matrix_complex_fwrite/fread failure for noncontiguous
  457    matrices (Matthias Sitte).
  458 
  459 * What was new in gsl-1.15:
  460 
  461 ** Added Tuomo Keskitalo's new ode branch ode-initval2 with a
  462    gsl_odeiv2 prefix. This provides proper support for implicit
  463    solvers.  It is intended to be the new default for differential
  464    equations.  The existing gsl_odeiv routines will be retained for
  465    binary compatibility but their interface will be deprecated.
  466 
  467 ** Added new gsl_integrate_cquad routines for robust integration of
  468    difficult functions using the doubly-adaptive CQUAD algorithm
  469    (Pedro Gonnet).
  470 
  471 ** Added error checking to CBLAS functions (José Luis García Pallero)
  472 
  473 ** Added a new function gsl_integration_glfixed_point to return
  474    ordered Gauss-Legendre points and weights contained within a
  475    gsl_integration_glfixed_table [bug #32237].
  476 
  477 ** Added a new function gsl_interp_type_min_size to return the size of
  478    an interpolation type.
  479 
  480 ** Added a function gsl_pow_uint(x,n) to compute x^n for unsigned
  481    exponents (needed when n exceeds the range of signed integers).
  482 
  483 ** Added new routine gsl_linalg_complex_cholesky_invert to handle the
  484    matrix inversion for complex Cholesky decompositions (Huan Wu).
  485 
  486 ** Added the functions gsl_vector_equal(x,y) and gsl_matrix_equal(x,y)
  487    for testing equality of two vectors or matrices.
  488 
  489 ** Added function gsl_eigen_nonsymmv_params to control the balancing
  490    transformation for eigenvector calculations. Balancing is now
  491    turned off by default for gsl_eigen_nonsymmv.
  492 
  493 ** It is now possible to choose an alternative cblas library via
  494    pkg-config using the GSL_CBLAS_LIB environment variable or
  495    the pkg-config --define-variable option.
  496 
  497 ** The jacobi method gsl_eigen_jacobi now uses the norm of the
  498    off-diagonal elements for its convergence criterion, as in
  499    algorithm 8.4.3 of Golub and van Loan.
  500 
  501 ** The newton multiroot solvers now return an error when a singular
  502    jacobian is detected.
  503 
  504 ** The interpolation functions now return NaN and when x is out of range,
  505    instead of extrapolating.
  506 
  507 ** The gsl_multimin_fdfsolver multidimensional minimisers now return
  508    GSL_ENOPROG immediately if the generated trial point does not
  509    differ from the initial point (to machine precision), avoiding
  510    unnecessary further iterations.
  511 
  512 ** Extended the range of gsl_sf_bessel_lnKnu_e by rescaling
  513    intermediate results to avoid internal overflows [bug #31528].
  514 
  515 ** Improved the result of gsl_sf_atanint_e for large arguments by
  516    adding the first order 1/x correction term. [bug #29562]
  517 
  518 ** Fixed the gsl_rng_ranlxs generators to enforce a maximum seed value
  519    of 2^31-1.  Larger seed values corrupted the random number state
  520    and caused out of range values to be returned.
  521 
  522 ** Fixed gsl_ran_chisq_pdf(x,nu) to return correct result of 1/2
  523    instead of 0 when x=0 and nu=2, and +inf when x=0 and nu<2.
  524 
  525 ** Fixed gsl_pow_int(x,n) to avoid an infinite loop when n=INT_MIN due
  526    to wrapping of signed integers.
  527 
  528 ** Fixed gsl_sf_hyperg_2F1(a,b,c,x) to avoid returning NaN for
  529    arguments |a|>10. [bug #24812]
  530 
  531 ** Avoid spurious underflow return code in gsl_sf_beta_inc_e when
  532    intermediate underflow does not affect the result. [bug #30933]
  533 
  534 ** Avoid segfault in Chebyshev series derivatives gsl_cheb_calc_deriv
  535    for n=1. [bug #29139]
  536 
  537 * What was new in gsl-1.14:
  538 
  539 ** Upgraded to latest libtool, autoconf and automake (libtool-2.2.6b,
  540    autoconf-2.65, automake-1.11.1).  Fixes security hole in 'make
  541    dist' (see Automake CVE-2009-4029).
  542 
  543 ** Added support for "multisets", which are similar to permutations
  544    and combinations.  A multiset is an array of k integers in the
  545    range 0 to n-1 where each value may occur more than once. Multisets
  546    can be used to iterate over the indices of a k-th order symmetric
  547    tensor in n-space. (Rhys Ulerich)
  548 
  549 ** Added gsl_integrate_glfixed routines for performing fixed order
  550    Gauss-Legendre integration 
  551    (Pavel Holoborodko, Konstantin Holoborodko, Rhys Ulerich)
  552 
  553 ** In the LMDER multi-dimensional fitting routines, the return code
  554    GSL_ENOPROG is now used instead of GSL_CONTINUE for the case where
  555    10 or more attempts to find a suitable trial step have been made
  556    without success. [bug #25383]
  557 
  558 ** The confluent hypergeometric function gsl_sf_hyperg_U (a,b,x) has
  559    been extended to support x < 0 for cases not involving
  560    singularities in 1F1(a,b,x). [bug #27859]
  561 
  562 ** The F-distribution gsl_ran_fdist_pdf now avoids unnecessary
  563    underflow and overflow and handles cases where n > 248. [bug
  564    #28500]
  565 
  566 ** The SVD routines now use a scaled version of the implicit QR method
  567    and compute the shift more reliably for values at the limit of
  568    double precision, avoiding potential NaNs. [bug #28767]
  569 
  570 ** A compile time error is emitted if the configuration stage prevents
  571    the functions gsl_isnan and gsl_finite from being defined.
  572 
  573 ** Added missing dereference in GSL_VECTOR_COMPLEX when not using
  574    GSL_RANGE_CHECK [bug #28017]
  575 
  576 ** Improved the range of gsl_sf_hyperg1F1(a,b,x) when a<0,b>0. [bug
  577    #28718]
  578 
  579 ** Added macros GSL_MAJOR_VERSION and GSL_MINOR_VERSION in
  580    <gsl/gsl_version.h>
  581 
  582 ** Improved gsl_eigen_symmv and gsl_eigen_symm to avoid a potential
  583    infinite loop when the tridiagonalised matrix contains very small
  584    values. [bug #28096]
  585 
  586 ** Added functions gsl_multifit_linear_usvd and
  587    gsl_multifit_wlinear_usvd for multilinear fitting without
  588    column-scaling of the fit matrix.
  589 
  590 * What was new in gsl-1.13:
  591 
  592 ** Upgraded to latest autoconf and automake (autoconf-2.64,
  593    automake-1.11)
  594 
  595 ** Fixed the rk4 and bspline allocators to avoid invalid free()
  596    calls under out of memory conditions. [bug #27194, #27236]
  597 
  598 ** Fixed a bug in gsl_multimin_fminimizer_nmsimplex2 where the center
  599    and size of the simplex were not updated on contract-by-best steps,
  600    causing failures in convergence. [bug #27180]
  601 
  602 ** Added new functions to set MISER and VEGAS Monte Carlo integration
  603    parameters, and to examine VEGAS chi-squared value and intermediate
  604    results.
  605 
  606 ** Added the function gsl_bspline_greville_abscissa to compute
  607    Greville abscissae for B-splines.
  608 
  609 ** The cumulative distribution functions gsl_cdf_gumbel1_{P,Q} should
  610    now handle a larger range of parameters without underflow and
  611    overflow.
  612 
  613 ** The header file gsl_const_cgs.h no longer defines values for
  614    electromagnetic units.  Applications should use gsl_const_cgsm.h
  615    instead to obtain the values in the CGS-Magnetic system.  The
  616    previous values for these units given in gsl_const_cgs.h were
  617    ill-defined as the type of CGS electromagnetic system was
  618    unspecified (the values were a mixture of CGS units with the Ampere
  619    of the MSKA system).  The affected constants are
  620    GSL_CONST_CGS_BOHR_MAGNETON, GSL_CONST_CGS_ELECTRON_CHARGE,
  621    GSL_CONST_CGS_ELECTRON_MAGNETIC_MOMENT, GSL_CONST_CGS_FARADAY,
  622    GSL_CONST_CGS_GAUSS, GSL_CONST_CGS_NUCLEAR_MAGNETON,
  623    GSL_CONST_CGS_PROTON_MAGNETIC_MOMENT, and GSL_CONST_CGS_ROENTGEN.
  624 
  625 ** The Pochhammer functions gsl_sf_poch(a,x) and gsl_sf_lnpoch(a,x) now
  626    handle the special cases where a and a+x are zero or negative
  627    integers.
  628 
  629 ** The confluent hypergeometric function gsl_sf_hyperg_U (a,b,x) now
  630    handles some cases where x=0. The case where 1+a-b is a negative
  631    integer no longer returns an error [bug #22859] and the incorrect
  632    termination of the series in certain cases is fixed [bug #26706].
  633 
  634 ** Added a new function gsl_poly_eval_derivs to evaluate a polynomial
  635    and its derivatives simultaneously.
  636 
  637 ** Added a new univariate minimisation algorithm
  638    gsl_min_fminimizer_quad_golden which is a variant of Brent's
  639    algorithm with safeguarded step-length adjustment.
  640 
  641 ** Added a new Nelder-Mead minimiser gsl_multimin_fminimizer_nmsimplex2rand
  642    which uses a randomly oriented simplex rather than one fixed on
  643    the coordinate axes [bug #25077]
  644 
  645 ** The texinfo file now uses the dircategory "Software libraries" from
  646    the Free Software Directory, as recommended in the Texinfo manual.
  647 
  648 ** The function gsl_ran_exponential now includes zero in its output
  649    range. [bug #25039]
  650 
  651 ** All functions for freeing allocated memory now accept a NULL
  652    pointer, following the standard C convention for free(). [bug
  653    #25319]
  654 
  655 ** The function gsl_sum_levin_u_accel now handles the special case
  656    c_0 + 0 + 0 + 0 + .... that occurs when summing power series
  657    c_n*x^n with x=0. [bug #26807]
  658 
  659 ** The functions gsl_linalg_LU_solve, gsl_linalg_LU_svx,
  660    gsl_linalg_LU_refine, gsl_linalg_LU_invert and their complex
  661    equivalents now return an error for singular matrices.
  662 
  663 ** The multifit LMDER hybrid solvers now check the return code of the
  664    user-supplied function in the gsl_multifit_fdfsolver_set
  665    method. [bug #26871]
  666 
  667 ** Improved the implementation of gsl_ran_discrete_preproc to avoid
  668    internal errors due to inconsistencies from excess precision on
  669    some platforms. [bug #26502]
  670 
  671 ** Corrected gsl_sf_hyperg_2F1(a,b,c,x) to not give a domain error in
  672    the case where c is a negative integer and the series terminates
  673    with a finite result.
  674 
  675 ** The C99 inline keyword is now supported, in addition to the
  676    previously supported GNU-style inline.
  677 
  678 ** Modified gsl_poly_complex_solve_cubic and gsl_poly_solve_cubic to
  679    avoid returning NaNs in cases where excess precision causes a
  680    change in the number of roots. 
  681 
  682 ** Fixed incorrect length check in gsl_blas_drotm. [bug #26503]
  683 
  684 ** Fixed gsl_odeiv_step_gear2 to restore y on step failure
  685 
  686 ** gsl_odeiv_evolve_apply now restores the correct value of t on step
  687    failures [bug #26255].
  688 
  689 ** Using make install prefix=DIR now puts correct paths in package
  690    config files gsl-config and gsl.pc
  691 
  692 ** Modified gsl_monte_vegas to work around pow() function inaccuracies
  693    on MinGW [bug #25413].
  694 
  695 ** Increased the number of terms in gsl_sf_mathieu_a and
  696    gsl_sf_mathieu_b to improve convergence in difficult regions [bug
  697    #25075]
  698 
  699 * What was new in gsl-1.12:
  700 
  701 ** Upgraded to latest libtool, autoconf and automake (libtool-2.2.6,
  702    autoconf-2.63, automake-1.10.2)
  703 
  704 ** Improved the convergence of gsl_sf_gamma_inc_P for x/a ~=~ 1 and
  705    large x,a.  Fixes problems with large arguments in cdf functions
  706    such as gsl_cdf_chisq_Pinv(x,nu) [bug 24704].
  707 
  708 ** Fixed gsl_ran_gamma_knuth to handle the case of a >= UINT_MAX [bug
  709    #24897]
  710 
  711 ** Added gsl_bspline_eval_deriv to compute bspline derivatives
  712    (Rhys Ulerich)
  713 
  714 ** Added a faster simplex mininimser gsl_multimin_fminimizer_nmsimplex2
  715    which is O(N) instead of O(N^2) [bug #24418]
  716 
  717 ** Improved the original chi-squared formula in gsl_monte_vegas to
  718    avoid catastrophic cancellation [bug #24510].  The previous formula
  719    could return incorrect or negative values for relative errors <
  720    1e-8, which could occur when integrating very smooth functions.
  721 
  722 ** Added new auxiliary functions gsl_cheb_order, gsl_cheb_size,
  723    gsl_cheb_coeffs for Chebyshev series [bug #21830]
  724 
  725 ** Updated license of the reference manual to GNU FDL version 1.3.
  726 
  727 ** Fixed a bug where the gsl_isinf function would return +1 for -Inf
  728    on systems where isinf(-Inf) returns the non-standard value +1.
  729    [bug #24489]
  730 
  731 ** Added missing functions gsl_vector_complex_{isnonneg,add,sub,mul,
  732    div,scale,add_constant} and gsl_matrix_complex_float_isnonneg [bug
  733    #22478]
  734 
  735 ** Cross compilation should now work for x86 hosts.
  736 
  737 ** Fixed a bug in gsl_interp_accel_find() where values lying on the
  738    upper boundary between interpolation points could return the index
  739    from the lower side. [bug #24211]
  740 
  741 ** Fixed gsl_linalg_solve_cyc_tridiag so that its output respects the
  742    solution vector's stride. Previously the x_stride value was ignored
  743    causing the output to be incorrect for non-unit stride. [bug #24162]
  744 
  745 ** Corrected a bug in the series calculation of gsl_sf_ellint_Kcomp
  746    for k close to 1. [bug #24146]
  747 
  748 ** Extended gsl_linalg_QRPT_update to handle rectangular matrices.
  749    Corrected definition of the update formula in the manual for
  750    both gsl_linalg_QR_update and gsl_linalg_QRPT_update.
  751 
  752 ** Added routine gsl_linalg_cholesky_invert
  753 
  754 ** Fixed a bug the simplex algorithm which caused the second highest
  755    point to be incorrectly equal to the first when the first value was
  756    the highest, which could cause suboptimal convergence. [bug #23192]
  757 
  758 ** Fixed a problem with convergence for inverse gamma and chisq
  759    distribitions, gsl_cdf_gamma_{P,Q}inv and gsl_cdf_chisq_{P,Q}inv.
  760    [bug #23101]
  761 
  762 ** Improved the handling of constant regions in Vegas by eliminating
  763    spurious excess precision when computing box variances.
  764 
  765 ** Fixed a potential division by zero in gsl_monte_miser when the
  766    left/right weight factors decrease below 1.
  767 
  768 ** Fixed incorrect dimensions check in gsl_matrix_sub{row,column}
  769 
  770 * What was new in gsl-1.11:
  771 
  772 ** The GSL repository and bug database are now hosted at Savannah
  773    http://savannah.gnu.org/projects/gsl/
  774 
  775 ** Upgraded to latest libtool, autoconf and automake (libtool-2.2,
  776    autoconf-2.61, automake-1.10.1)
  777 
  778 ** Fixed underflow in ODE adaptive step size controller that could
  779    cause step size to decrease to zero (bug #21933).
  780 
  781 ** Improved the handling of the asymptotic regime in gsl_sf_bessel_jl.
  782 
  783 ** Improved the handling of large arguments in cumulative distribution
  784    functions using the incomplete beta function, such as gsl_cdf_fdist_P.
  785 
  786 ** Fixed overflow bug in gsl_cdf_hypergeometric_{P,Q} for large
  787    arguments (bug #22293).
  788 
  789 ** gsl_ran_gaussian_ziggurat now handles generators with different
  790    ranges explicitly, to minimise the number of function calls
  791    required (bug #21820).  Also fixes bug #22446 (rng limit in
  792    gsl_ran_chisq()).
  793 
  794 ** Added missing error terms in gsl_sf_exp_mult_e10_e to prevent
  795    the error being underestimated (bug #22041).  
  796 
  797 ** Updated some constants to the CODATA 2006 values.
  798 
  799 ** The hypergeometric function gsl_sf_hyperg_2F1 now handles the case
  800    where x==1.
  801 
  802 ** Fixed a bug in the brent minimiser which prevented optimal convergence.
  803 
  804 ** Added functions for evaluating complex polynomials
  805 
  806 ** The convergence condition for gsl_multiroots_test_delta now accepts
  807    dxi == 0.
  808 
  809 ** Improved functions gsl_ldexp and gsl_frexp to handle the full range
  810    of double precision numbers in all cases.
  811 
  812 ** Added new quasi random generators gsl_qrng_halton and
  813    gsl_qrng_reversehalton which support dimensions up to 1229.
  814 
  815 ** Added function gsl_multifit_linear_residuals for computing the
  816    residuals of the fit
  817 
  818 * What was new in gsl-1.10:
  819 
  820 ** License updated to GNU GPL version 3.
  821 
  822 ** Added support for generalized eigensystems
  823 
  824 ** Added function gsl_stats_correlation to compute Pearson correlation
  825 of two datasets
  826 
  827 ** Added the new function gsl_sf_expint(n,x) for computing the n-th
  828 order exponential integral.
  829 
  830 ** Added functions gsl_vector_isnonneg and gsl_matrix_isnonneg.
  831 
  832 ** Added functions gsl_matrix_subrow and gsl_matrix_subcolumn
  833 
  834 ** Extended Cholesky routines to complex matrices
  835 
  836 ** Added support in gsl_ieee_set_mode for controlling SSE exceptions
  837 and rounding through the MXCSR control word on x86 processors.
  838 
  839 ** The autoconf macro AM_PATH_GSL has been renamed to AX_PATH_GSL, to
  840 avoid conflicts with the autoconf namespace.
  841 
  842 ** Improved handling of underflow in gsl_eigen_symm.
  843 
  844 ** The function gsl_multiroot_fdjacobian now returns the error code
  845 GSL_ESING if any of the columns of the computed jacobian matrix are
  846 zero.  This may occur if the step size of the derivative is too small.
  847 
  848 ** Extended the function gsl_sf_beta_inc(a,b,x) to handle cases where
  849 a<0 or b<0.
  850 
  851 ** Fixed the round-off error estimate in gsl_deriv_{central,backwards,
  852 forward} to correctly account for numerical error in the step-size h.
  853 
  854 ** Fixed gsl_cdf_beta_Pinv, gsl_cdf_gamma_Pinv, gsl_cdf_beta_Pinv to
  855 avoid returning spurious values for large parameters when the
  856 iteration did not converge.  If the iteration cannot converge, GSL_NAN
  857 is returned.
  858 
  859 ** gsl_ran_dirichlet now handles smaller values of alpha[] without
  860 underflow, avoiding a NaN in the returned value.
  861 
  862 ** The SVD routines now avoid underflow in the Schur decomposition for
  863 matrices with extremely small values <O(1e-150).
  864 
  865 ** gsl_complex_pow now returns 0^0=1 (instead of zero) to match the usual
  866 pow function, and handles z^(1,0) and z^(-1,0) as special cases.
  867 
  868 ** Fixed a bug in the set function for multifit lmder solver so that
  869 previous values are cleared correctly.
  870 
  871 ** Improved the function gsl_log1p to prevent possible loss of
  872 accuracy caused by optimisations.
  873 
  874 ** Improved the convergence test in the Lambert functions to take
  875 account of finite precision and avoid possible failure to converge.
  876 
  877 ** The configure script no longer redefines finite() to isfinite() as
  878 a workaround for missing finite(), as this caused problems on Solaris.
  879 If finite() is not present gsl_finite() will be used instead.
  880 
  881 ** Improved the accuracy of the generalised laguerre polynomials for
  882 large n when alpha=0.
  883 
  884 ** The functions gsl_{isnan,isinf,finite} will now always use the
  885 system versions of isnan, isinf and finite if they are available.
  886 Previously the portable GSL implementations were used whenever the
  887 platform supported IEEE comparisons.  The portable gsl_isinf function
  888 now returns -1 instead of +1 for -Inf, in line with recent versions of
  889 the GNU C Library.
  890 
  891 * What was new in gsl-1.9:
  892 
  893 ** Fixed the elliptic integrals F,E,P,D so that they have the correct
  894 behavior for phi > pi/2 and phi < 0.  The angular argument is now
  895 valid for all phi.  Also added the complete elliptic integral
  896 gsl_sf_ellint_Pcomp.
  897 
  898 ** Added a new BFGS minimisation method gsl_multimin_fdfminimizer_vector_bfgs2
  899 based on the algorithm given by R.Fletcher in "Practical Methods of
  900 Optimisation" (Second edition).  This requires substantially fewer
  901 function and gradient evaluations, and supercedes the existing BFGS
  902 minimiser.
  903 
  904 ** The beta functions gsl_sf_beta_e(a,b) and gsl_sf_lnbeta_e(a,b) now
  905 handle negative arguments a,b.  Added new function gsl_sf_lnbeta_sgn_e
  906 for computing magnitude and sign of negative beta values, analagous to
  907 gsl_sf_lngamma_sgn_e.
  908 
  909 ** gsl_cheb_eval_mode now uses the same error estimate as
  910 gsl_cheb_eval_err.
  911 
  912 ** Improved gsl_sf_legendre_sphPlm_e to avoid underflow with large
  913 arguments.
  914 
  915 ** Added updated Knuth generator, gsl_rng_knuthran2002, from 9th
  916 printing of "The Art of Computer Programming".  Fixes various
  917 weaknesses in the earlier version gsl_rng_knuthran.  See
  918 http://www-cs-faculty.stanford.edu/~knuth/news02.htm
  919 
  920 ** The functions gsl_multifit_fsolver_set, gsl_multifit_fdfsolver_set
  921 and gsl_multiroot_fsolver_set, gsl_multiroot_fdfsolver_set now have a
  922 const qualifier for the input vector x, reflecting their actual usage.
  923 
  924 ** gsl_sf_expint_E2(x) now returns the correct value 1 for x==0,
  925 instead of NaN.
  926 
  927 ** The gsl_ran_gamma function now uses the Marsaglia-Tsang fast gamma
  928 method of gsl_ran_gamma_mt by default.
  929 
  930 ** The matrix and vector min/max functions now always propagate any
  931 NaNs in their input.
  932 
  933 ** Prevented NaN occuring for extreme parameters in
  934 gsl_cdf_fdist_{P,Q}inv and gsl_cdf_beta_{P,Q}inv
  935 
  936 ** Corrected error estimates for the angular reduction functions
  937 gsl_sf_angle_restrict_symm_err and gsl_sf_angle_restrict_pos_err.
  938 Fixed gsl_sf_angle_restrict_pos to avoid possibility of returning
  939 small negative values.  Errors are now reported for out of range
  940 negative arguments as well as positive.  These functions now return
  941 NaN when there would be significant loss of precision.
  942 
  943 ** Corrected an error in the higher digits of M_PI_4 (this was beyond
  944 the limit of double precision, so double precision results are not
  945 affected).
  946 
  947 ** gsl_root_test_delta now always returns success if two iterates are
  948 the same, x1==x0.
  949 
  950 ** A Japanese translation of the reference manual is now available
  951 from the GSL webpage at http://www.gnu.org/software/gsl/ thanks to
  952 Daisuke TOMINAGA.
  953 
  954 ** Added new functions for basis splines, see the "Basis Splines"
  955 chapter in the GSL Reference Manual for details.
  956 
  957 ** Added new functions for testing the sign of vectors and matrices,
  958 gsl_vector_ispos, gsl_vector_isneg, gsl_matrix_ispos and
  959 gsl_matrix_isneg.
  960 
  961 ** Fixed a bug in gsl_sf_lnpoch_e and gsl_sf_lnpoch_sgn_e which caused
  962 the incorrect value 1.0 instead of 0.0 to be returned for x==0.
  963 
  964 ** Fixed cancellation error in gsl_sf_laguerre_n for n > 1e7 so that
  965 larger arguments can be calculated without loss of precision.
  966 
  967 ** Improved gsl_sf_zeta_e to return exactly zero for negative even
  968 integers, avoiding less accurate trigonometric reduction.
  969 
  970 ** Fixed a bug in gsl_sf_zetam1_int_e where 0 was returned instead of
  971 -1 for negative even integer arguments.
  972 
  973 ** When the differential equation solver gsl_odeiv_apply encounters a
  974 singularity it returns the step-size which caused the error code from
  975 the user-defined function, as opposed to leaving the step-size
  976 unchanged.
  977 
  978 ** Added support for nonsymmetric eigensystems
  979 
  980 ** Added Mathieu functions
  981 
  982 * What was new in gsl-1.8:
  983 
  984 ** Added an error check to trap multifit calls with fewer observations
  985 than parameters.  Previously calling the multifit routines with n<p
  986 would cause invalid memory access.
  987 
  988 ** Added the Debye unit to physical constants.
  989 
  990 ** Added cumulative distribution functions for the discrete
  991 distributions, including binomial, poisson, geometric, negative
  992 binomial, pascal and hypergeometric.
  993 
  994 ** Added the functions gsl_cdf_beta_{Pinv,Qinv} and
  995 gsl_cdf_fdist_{Pinv,Qinv} for computing the inverse of the cumulative
  996 beta and F distributions.
  997 
  998 ** Added the multilinear fit estimator function gsl_multifit_linear_est
  999 for computing model values and their errors.
 1000 
 1001 ** Avoid division by zero in gsl_multimin_fdfminimizer_vector_bfgs
 1002 if the step-size becomes too small.
 1003 
 1004 ** Users on DEC Alpha systems will need to specify their desired IEEE
 1005 arithmetic options via CFLAGS when building the library, as these are
 1006 no longer added automatically.
 1007 
 1008 ** Added new random variate generators gsl_ran_gaussian_ziggurat
 1009 and gsl_ran_gamma_mt for the Gaussian and Gamma distributions based on
 1010 the Marsaglia-Tsang ziggurat and fast gamma methods.
 1011 
 1012 ** Improved the speed of the exponential power distribution
 1013 gsl_ran_exppow.
 1014 
 1015 ** Improved the speed of the Gaussian ratio method by adding quadratic
 1016 bounds in gsl_ran_gaussian_ratio_method.
 1017 
 1018 ** Added an extra term to the taylor series of the synchrotron
 1019 functions gsl_sf_synchrotron_1 and gsl_sf_synchrotron_2 for small x to
 1020 ensure smooth matching with the chebyshev expansion.
 1021 
 1022 ** The binomial pdf gsl_ran_binomial_pdf now handles the cases p=0
 1023 and p=1 and is more accurate for the case of small p with k=0.
 1024 
 1025 ** Fixed the spherical bessel function gsl_sf_bessel_jl_e) to limit
 1026 the use of gsl_sf_bessel_Jnu_asympx_e to the range x>100*l*l to
 1027 satisfy he requirement x>>l*l in the asymptotic expansion.
 1028 
 1029 ** The scaled bessel function gsl_sf_bessel_In_scaled now handles
 1030 larger arguments x > 1e7 correctly for n < 150 using the uniform
 1031 asymptotic expansion instead of the continued fraction expansion.
 1032 
 1033 ** The functions gsl_stats_min/max now return NaN if the data contains
 1034 NaN. Similarly, the functions gsl_stats_min/max_index return the index
 1035 of the first occurring NaN in the data when it contains a NaN.
 1036 
 1037 ** Fixed an invalid memory access that caused incorrect results for
 1038 the special case in periodic cubic spline interpolation of 3 points.
 1039 
 1040 ** Added Debye functions for n=5 and n=6
 1041 
 1042 ** Added the missing functions gsl_spline_name() and
 1043 gsl_spline_min_size()
 1044 
 1045 ** The function gsl_rng_uniform_int(r,n) now returns an error for n=0,
 1046 which can occur when passing an unsigned integer value of 2^32.
 1047 
 1048 * What was new in gsl-1.7:
 1049 
 1050 ** Switched gsl_randist_binomial to use the faster binomial random
 1051 variate TPE algorithm by default.  The previous binomial variate
 1052 algorithm is available as gsl_randist_binomial_knuth.  This will
 1053 result in a different sequence of binomial variates in programs using
 1054 this function.
 1055 
 1056 ** Improved the algorithm for gsl_sf_elljac_e to avoid cancellation
 1057 errors near quarter periods.
 1058 
 1059 ** Fixed the branch selection in gsl_sf_gamma_inc_Q_e to avoid
 1060 inaccurate results for large a,x where x~=~a.
 1061 
 1062 ** The multilinear fitting functions now have forms which accept a
 1063 user-specified tolerance for the SVD cutoff and return the
 1064 corresponding effective rank of the design matrix.
 1065 
 1066 ** The quadratic solvers in poly/ now handle linear equations
 1067 gracefully (i.e. quadratrics with a leading coefficient of zero).
 1068 
 1069 ** The output of "make check" now only shows test failures by default,
 1070 to reduce the amount of output.  Set the environment variable
 1071 GSL_TEST_VERBOSE=1 to display all the output.  To assist debugging,
 1072 the test number of each failure is shown in square brackets at the
 1073 line-end [NNNN].
 1074 
 1075 ** Fixed bugs in gsl_linalg_SV_decomp_jacobi which caused
 1076 incorrect results for some input matrices.
 1077 
 1078 ** Bessel, coulomb, dilogarithm and legendre_H3d functions now use
 1079 hypot internally to avoid overflow when computing terms like
 1080 sqrt(1+x*x).
 1081 
 1082 ** The 'Usage' chapter of the reference manual now explains how to
 1083 handle deprecated functions using the GSL_DISABLE_DEPRECATED macro.
 1084 
 1085 ** The conflicting enum definitions for 'forward' and 'backward' in
 1086 gsl_ftt.h and gsl_wavelet.h are deprecated.  User code should switch
 1087 to the new definitions gsl_fft_forward, gsl_fft_backward,
 1088 gsl_wavelet_forward and gsl_wavelet_backward. Selectively define
 1089 GSL_DISABLE_DEPRECATED before including the headers to use the new
 1090 definitions on either or both modules.
 1091 
 1092 ** Fixed an error in the the brent minimisation algorithm.  Iterations
 1093 should now follow Brent's original description correctly.
 1094 
 1095 ** The bound coulomb function gsl_sf_hydrogenicR_e no longer reports
 1096 an underflow for exact zeroes of the wavefunction.
 1097 
 1098 ** gsl_linalg_SV_decomp_jacobi now reports an error for the
 1099 unimplemented case M<N correctly.
 1100 
 1101 ** Fixed conformance test for the SYRK and HERK blas functions
 1102 gsl_blas_{s,d,c,z}syrk and gsl_blas_{c,z}herk for non-square matrices.
 1103 
 1104 ** Configure now checks for presence of ieeefp.h if needed.
 1105 
 1106 ** Differential equation solvers now propagate error codes returned
 1107 from user-defined functions to the top-level in all cases.
 1108 
 1109 ** Sort functions now avoid an infinite loop if Nans are present in
 1110 the input vector.  The order of nans in the output is undefined,
 1111 although other elements will be sorted correctly.
 1112 
 1113 * What was new in gsl-1.6:
 1114 
 1115 ** Added a new wavelet directory, with 1-dimensional and 2-dimensional
 1116 discrete wavelet transforms.
 1117 
 1118 ** Added support for LQ and P^T LQ decompositions.  To find the QR
 1119 decomposition of large systems (M>>N) use the LQ decomposition,
 1120 solving the transpose of the original system.  This allows more
 1121 efficient memory access, and is useful for solving large least-squares
 1122 problems.
 1123 
 1124 ** Fixed a bug in the SYRK and HERK blas functions gsl_blas_{s,d,c,z}syrk 
 1125 and gsl_blas_{c,z}herk which caused invalid memory access for non-square 
 1126 matrices.
 1127 
 1128 ** Fixed a bug in gsl_swap_vectors which caused it to return incorrect
 1129 results when swapping vectors with different strides.
 1130 
 1131 ** Corrected the error estimate for gsl_cheb_eval_n_err to use
 1132 evaluation order instead of the approximation order.
 1133 
 1134 ** Improved the reliability of the gsl_sf_gamma_inc family of
 1135 functions.
 1136 
 1137 ** Equal abscissae are now handled gracefully in the cspline and
 1138 periodic cspline interpolations.
 1139 
 1140 ** Removed potential cancellation error in calculation of uniform
 1141 histogram ranges.
 1142 
 1143 ** Improved numerical stability of integration for akima and cspline
 1144 interpolation.
 1145 
 1146 ** Differential equation solvers now handle error codes returned from
 1147 user-defined functions.
 1148 
 1149 ** Improved error estimates in ode-initval solvers, and provide exact
 1150 derivatives on output.  Added new semi-implicit ode-initval solver,
 1151 gsl_odeiv_step_rk2simp.
 1152 
 1153 ** Added missing function definition for gsl_sf_psi_1.
 1154 
 1155 ** Fixed the function gsl_sf_expint_Ei_scaled to call
 1156 gsl_sf_expint_Ei_scaled_e instead of gsl_sf_expint_Ei_e.
 1157 
 1158 ** Added cumulative distribution function for exponential power
 1159 distribution.
 1160 
 1161 ** The functions gsl_cdf_beta_P and gsl_cdf_beta_Q now return
 1162 consistent results of 0 or 1 for out of range values, x<0 and x>1,
 1163 rather than 0 for left and right tails simultaneously.
 1164 
 1165 ** The Jacobi eigensolvers gsl_eigen_jacobi and gsl_eigen_jacobi_invert 
 1166 have new implementations from Golub and Van Loan.
 1167 
 1168 ** The standard output and standard error streams are now flushed by
 1169 the default error handler before the program aborts, in order to
 1170 ensure that error messages are properly displayed on some platforms.
 1171 
 1172 * What was new in gsl-1.5:
 1173 
 1174 ** Multifit routines now handle iterations where |f| is already
 1175 minimised to zero, without division by zero.
 1176 
 1177 ** Fixed the singular value tolerance test in the multifit covariance
 1178 calculation from < to <= to match the original MINPACK code.
 1179 
 1180 ** The macro HAVE_INLINE is now tested with #ifdef instead of #if as
 1181 in versions prior to 1.4, to match the documentation, and the macro
 1182 GSL_RANGE_CHECK_OFF now works correctly.  An alternative macro
 1183 GSL_RANGE_CHECK={0,1} can be used to control range-checking.
 1184 
 1185 ** Fixed a potential array overflow in gsl_ran_landau.
 1186 
 1187 ** Fixed a small discrepancy in the tolerance calculation of the
 1188 one-dimensional brent minimiser.
 1189 
 1190 ** Numerical derivatives should now be calculated using the
 1191 gsl_deriv_forward, gsl_deriv_central and gsl_deriv_backward functions,
 1192 which accept a step-size argument in addition to the position x.  The
 1193 original gsl_diff functions (without the step-size) are deprecated.
 1194 
 1195 ** Corrected documentation for gsl_ran_hypergeometric_pdf()
 1196 
 1197 ** The tridiagonal matrix solvers gsl_linalg_solve_symm_tridiag,
 1198 gsl_linalg_solve_tridiag, gsl_linalg_solve_symm_cyc_tridiag,
 1199 gsl_linalg_solve_cyc_tridiag now use the GSL_ERROR macro to report
 1200 errors, instead of simply returning an error code.  The arguments to
 1201 these functions must now use exact lengths with no additional
 1202 elements.  For cyclic systems all vectors must be of length N, for
 1203 tridiagonal systems the offdiagonal elements must be of length N-1.
 1204 
 1205 ** The singular value decomposition routines gsl_linalg_SV_decomp and
 1206 gsl_linalg_SV_decomp_mod now handle the SVD of a column vector (N=1,
 1207 arbitrary M), which can occur in linear fitting.
 1208 
 1209 ** Restored missing header files gsl_const_mks.h and gsl_const_cgs.h.
 1210 The incorrect values of the electrical units for gsl_const_cgs
 1211 (VACUUM_PERMEABILITY and VACUUM_PERMITTIVITY) have been removed.
 1212 
 1213 ** Fixed gsl_linalg_SV_decomp() to avoid an infinite loop when
 1214 computing the SVD of matrices containing Inf and Nan.
 1215 
 1216 ** Fixed gsl_linalg_balance_columns() to avoid an infinite loop when
 1217 rescaling matrices containing Inf and NaN.
 1218 
 1219 ** Fixed header file <gsl/gsl_sf_log.h> to include declarations for 
 1220 error codes in inline versions of gsl_sf_log functions
 1221 
 1222 ** Fixed header file <gsl/gsl_const.h> to include new MKSA and CGSM
 1223 header files.
 1224 
 1225 ** Added Stefan-Boltzmann constant and Thomson cross section to
 1226 physical constants
 1227 
 1228 * What was new in gsl-1.4:
 1229 
 1230 ** Added cumulative distribution functions and their inverses for the
 1231 continuous random distributions including: gaussian, lognormal, gamma,
 1232 beta, cauchy, laplace, chisq, exponential, gumbel, weibull,
 1233 F-distribution, t-distribution, logistic, pareto and rayleigh.
 1234 
 1235 ** Added faster binomial random variates using the TPE rejection
 1236 algorithm, in the function gsl_randist_binomial_tpe.
 1237 
 1238 ** Added new functions gsl_rng_fwrite and gsl_rnd_fread for storing
 1239 the state of random number generators in a file.
 1240 
 1241 ** Added a new function gsl_combination_memcpy()
 1242 
 1243 ** Corrected values of electrical constants in CGS units.  To take
 1244 account of different electrical systems of units the values are now
 1245 prefixed by GSL_CONST_MKSA (for the SI Metre, Kilogram, Second, Ampere
 1246 system) or GSL_CONST_CGSM (for the Centimetre, Gram, Second, Magnetic
 1247 system with the Gauss as the fundamental unit of magnetic field
 1248 strength).  The previous GSL_CONST_MKS and GSL_CONST_CGS prefixes have
 1249 been removed, as have the permeability and permittivity constants in
 1250 the CGS system since this uses different defining equations.
 1251 
 1252 ** Fixed bugs in the random number generators gsl_rng_fishman18,
 1253 gsl_rng_fishman2x, and gsl_rng_knuthran2 which caused them to return
 1254 incorrect results.  Minor corrections were made to the parameters in
 1255 the other Knuth generators borosh13, coveyou, fishman20, lecuyer21,
 1256 and waterman14.
 1257 
 1258 ** Fixed a missing transpose bug in the gsl_linalg_QR_QRsolve
 1259 and gsl_linalg_QRPT_QRsolve routines which were computing the
 1260 solution to Q^T R x = b instead of Q R x = b.
 1261 
 1262 ** Fixed gsl_sf_gammainv to return zero instead of a domain
 1263 error for arguments corresponding to singularities in gamma.
 1264 
 1265 ** Fixed a bug in the simplex minimization algorithm which
 1266 caused it to fail to find the second highest point correctly
 1267 when searching the set of simplex points.
 1268 
 1269 ** Fixed a bug in the conjugate gradient minimizers conjugate_pr,
 1270 conjugate_fr and vector_bgfs which caused the search
 1271 directions to be updated incorrectly.
 1272 
 1273 ** Fixed a bug in gsl_sf_psi_1_int(1) which caused it to
 1274 return the incorrect sign for psi(1,1).
 1275 
 1276 ** Fixed the simulated annealing routine gsl_siman_solve to use the
 1277 parameter iters_fixed_T for the number of iterations at fixed
 1278 temperature instead of n_tries.
 1279 
 1280 ** Fixed a bug in gsl_combination_valid which caused it to return the
 1281 incorrect status.
 1282 
 1283 ** Fixed a bug in gsl_permutation_canonical_to_linear which caused the
 1284 output to always be zero, and the input permutation to be incorrectly
 1285 replaced by the output.
 1286 
 1287 ** Fixed a bug is gsl_ran_discrete which could cause uninitialised
 1288 data to be returned for some distributions.
 1289 
 1290 ** Fixed the dependencies for gsl_chebyshev.h to include gsl_math.h.
 1291 
 1292 ** Fixed a bug in gsl_complex_arccsc_real which caused it to return 
 1293 the incorrect sign for the imaginary part when -1<x<0.
 1294 
 1295 ** Fixed a bug in the QAWC Cauchy integration routine which could
 1296 allow the singularity to fall on an interval boundary, leading to
 1297 division by zero.
 1298 
 1299 ** Improved gsl_sf_gamma_inc_P(a,x) to avoid a domain error for x<<a
 1300 when a>10.
 1301 
 1302 ** Improved the accuracy of gsl_sf_coupling_3j for large arguments.
 1303 
 1304 ** Improved the performance of gsl_sf_choose(m,n) by separating the
 1305 calculations for small and large arguments.
 1306 
 1307 ** On platforms without IEEE comparisons gsl_{isnan,isinf,finite} will
 1308 fall back to the system versions of isnan, isinf and finite if
 1309 available.
 1310 
 1311 ** gsl_linalg_householder_hv now uses BLAS routines internally
 1312 
 1313 ** The script configure.in is now compatible with autoconf-2.50 and
 1314 later.
 1315 
 1316 ** Reduced the memory usage of the multifit algorithms from MxM to MxN
 1317 for large M by performing the QR decomposition of the Jacobian
 1318 in-place.
 1319 
 1320 ** IEEE modes now use the C99 fenv.h functions when platform spectific
 1321 functions are not available.
 1322 
 1323 * What was new in gsl-1.3:
 1324 
 1325 ** Changed interface for gsl_sf_coupling_6j...(...). The old functions
 1326 actually calculated 6j for a permutation of the arguments (that
 1327 related to Racah W). This was incorrect and not consistent with
 1328 the documentation. The new versions calculate < {a,b,c}, {d,e,f} >,
 1329 as stated in the documentation. The old versions are still available
 1330 as gsl_sf_coupling_6j_INCORRECT...(...), though they are deprecated
 1331 and will be removed at some point in the future.
 1332 
 1333 ** Added new functions for computing Em(x)=exp(-x)*Ei(x), the modified
 1334 (scaled) form of the exponential integral, gsl_sf_expint_E1_scaled,
 1335 gsl_sf_expint_E2_scaled, gsl_sf_expint_Ei_scaled.
 1336    
 1337 ** Fixed compilation problems with gcc -ansi and other ANSI compilers.
 1338 
 1339 ** Fixed uninitialized memory access in the Niederreiter quasi-random
 1340 number generator.
 1341 
 1342 ** Fixed the eigenvalue routines to prevent an infinite loop for Inf
 1343 or NaN entries in matrix.
 1344 
 1345 ** Fixed a bug in the multifit and multiroots allocation routines
 1346 which cause them to fail to report some out of memory conditions.
 1347 
 1348 ** Fixed a bug in the seeding for the random number generator
 1349 gsl_rng_taus2 which affected a small number of seeds.
 1350 
 1351 ** Modified the complex householder transforms to avoid division by
 1352 zero, which could cause NaNs to be returned by the gsl_eigen_hermv
 1353 eigenvalue decomposition.
 1354 
 1355 ** The Nelder-Mead simplex algorithm for multidimensional 
 1356 minimisation has been added.
 1357 
 1358 ** The random number distributions now include the Dirichlet and
 1359 Multinomial distributions.
 1360 
 1361 ** Added a new function gsl_fcmp for approximate comparison of
 1362 floating point numbers using Knuth's algorithm.
 1363 
 1364 ** Added new functions gsl_ldexp and gsl_frexp as portable
 1365 alternatives to ldexp() and frexp().
 1366 
 1367 ** Fixed a bug in gsl_linalg_bidiag_unpack_B which was returning
 1368 incorrect results for the superdiagonal.
 1369 
 1370 ** Fixed a bug in the acceptance condition for simulated annealing
 1371 
 1372 ** Ordinary differential equations can now be solved using a different
 1373 absolute error for each component with gsl_odeiv_control_scaled_new().
 1374 
 1375 ** Upgraded to libtool-1.4.3
 1376 
 1377 * What was new in gsl-1.2:
 1378 
 1379 ** Added new functions for combining permutations, converting between
 1380 cyclic and linear representations, and counting cycles and inversions.
 1381 
 1382 ** New multiroot functions now allow access to the current values of f
 1383 and dx.
 1384 
 1385 ** The default error handler now outputs a explanatory message before
 1386 aborting.
 1387 
 1388 ** Extended gsl_linalg_SV_decomp to handle exact zeroes in the
 1389 singular values, and added tests for 3x3 matrices.
 1390 
 1391 ** Fixed a bug in gsl_linalg_SV_decomp which caused singular values to
 1392 be sorted incorrectly.
 1393 
 1394 ** Fixed a bug in gsl_linalg_solv_symm_cyc_tridiag which caused it to
 1395 produce incorrect results.
 1396 
 1397 ** Added nonsymmetric tridiagonal solvers gsl_linalg_solve_tridiag and
 1398 gsl_linalg_solve_cyc_tridiag.
 1399 
 1400 ** The declarations used to export static objects can now be
 1401 controlled through a macro GSL_VAR and the header file
 1402 <gsl/gsl_types.h>.
 1403 
 1404 ** The simulated annealing routine gsl_siman_solve now keeps track of
 1405 the best solution so far.
 1406 
 1407 ** The values of the physical constants have been updated to the
 1408 CODATA 1998 recommendations.
 1409 
 1410 ** Added new physical constants, newton, dyne, joule, erg and 
 1411 power-of-ten prefixes, Mega, Giga, Tera, etc.
 1412 
 1413 ** The error estimate for the elliptic function gsl_sf_ellint_Kcomp_e
 1414 has been improved to take account of numerical cancellation for small
 1415 arguments.
 1416 
 1417 ** The domain of gsl_sf_psi_1piy has been extended to negative y.
 1418 
 1419 ** Fixed memory leak in the Chebyshev module.
 1420 
 1421 ** The seeding procedure of mt19937 has been updated to the latest
 1422 version from Makoto Matsumoto and Takuji Nishimura (Jan 2002). The
 1423 original seeding procedure is available through the generator
 1424 gsl_rng_mt19937_1999.
 1425 
 1426 ** A new random number generator gsl_rng_taus2 has been added to
 1427 correct flaws in the seeding procedure of gsl_rng_taus, as described
 1428 in an erratum to the original paper of P. L'Ecuyer.
 1429 
 1430 ** Added missing declaration for the generator gsl_rng_mt_19937_1998.
 1431 
 1432 ** Added missing quasi-random number generator function gsl_qrng_init.
 1433 
 1434 ** Removed unnecessary endpoint subtraction in chebyshev-based
 1435 QUADPACK routines to avoid possible loss of precision.
 1436 
 1437 ** Fixed bug in gsl_interp_cspline_periodic which caused a
 1438 discontinuity in the derivative near the boundary.
 1439 
 1440 ** The function gsl_min_fminimizer_minimum has been renamed to
 1441 gsl_min_fminimizer_x_minimum for consistency (the old function name is
 1442 still available but is deprecated).  Additional functions have been
 1443 added for accessing the function values at the minimum and endpoints
 1444 of the bounding interval.
 1445 
 1446 ** The KNOWN-PROBLEMS file of "make check" failures has been replaced
 1447 by a BUGS file, since we now require "make check" to work correctly
 1448 for stable releases.
 1449 
 1450 * What was new in gsl-1.1.1:
 1451 
 1452 ** Fixes to histogram2d stat functions
 1453 
 1454 ** Added missing prototypes for complex LU determinant functions
 1455 
 1456 ** Improved error handling in multifit routines
 1457 
 1458 ** Added check to avoid division by zero for rank-deficient matrix in
 1459 multifit iteration
 1460 
 1461 * What was new in gsl-1.1:
 1462 
 1463 ** The permutation module now includes a copy function
 1464 gsl_permutation_memcpy
 1465 
 1466 ** The implementation of gsl_sf_gamma_inc has been improved and now
 1467 avoids problems caused by internal singularities which occurred in the
 1468 series expansion for some combinations of parameters.
 1469 
 1470 ** IEEE comparisons of infinities and NaNs are tested during the
 1471 configure stage and the functions gsl_isnan, gsl_isinf and gsl_finite
 1472 are only compiled on platforms which support the necessary tests.
 1473 
 1474 ** The histogram routines now include a sum function,
 1475 gsl_histogram_sum for computing the total bin sum, and additional
 1476 statistics functions for 2d histograms.
 1477 
 1478 ** Internal error checking of user-defined functions has been improved
 1479 in the multiroots functions.
 1480 
 1481 ** Constants now include the Bohr Radius and Vacuum Permittivity.
 1482 
 1483 ** Range checking is now turned off when building the library, but is
 1484 still on by default when compiling user applications.
 1485 
 1486 ** A combinations directory has been added for generating combinations (n,k).
 1487 
 1488 ** The gamma function now returns exact values for integer arguments.
 1489 
 1490 ** Fixed bugs in gsl_sf_hyperg_1F1_int and gsl_sf_hyperg_1F1.
 1491 
 1492 ** Fixed internal error handling in gsl_sf_laguerre_n to allow
 1493 recovery from overflow.
 1494 
 1495 ** Several routines for handling divided difference polynomials have
 1496 been added to the poly/ directory.
 1497 
 1498 ** The interpolation routines now include polynomial interpolation,
 1499 based on divided-differences.
 1500 
 1501 ** Added new random number generators from Knuth's Seminumerical
 1502 Algorithms, 3rd Edition: borosh13, coveyou, fishman18, fishman20,
 1503 fishman2x, knuthran, knuthran2, lecuyer21, waterman14.
 1504 
 1505 ** Changed divisor in random number generator gfsr4 from 2^32-1 to
 1506 2^32 to prevent exact value of 1.0 from being returned, as specified
 1507 in the documentation.
 1508 
 1509 * What was new in gsl-1.0:
 1510 
 1511 ** First general release.
 1512 
 1513 ** Increased the maximum number of iterations in gsl_poly_complex_solve()
 1514 from 30 to 60.
 1515 
 1516 * What was new in gsl-0.9.4:
 1517 
 1518 ** Reorganized the multmin functions to use the same interface as the
 1519 other iterative solvers.
 1520 
 1521 ** Added histogram _alloc functions for consistency, in addition to the
 1522 existing _calloc functions.
 1523 
 1524 ** Renamed all the gsl_multimin functions to be consistent with the
 1525 rest of the library.  An underscore has been removed from _minimizer
 1526 in all the function names.
 1527 
 1528 ** Renamed the function gsl_sf_coulomb_CL_list to gsl_sf_coulomb_CL_array
 1529 
 1530 ** A bug in the multimin functions where the function parameters
 1531 (params) were omitted has been fixed.
 1532 
 1533 ** A bug in the nonlinear minimization routines has been fixed, which
 1534 could prevent the algorithms from converging.  Additional tests from
 1535 the NIST reference datasets have been added and these now agree with
 1536 MINPACK.
 1537 
 1538 ** All the physical constants and conversion factors are now defined as
 1539 real numbers to avoid potential problems with integer arithmetic.
 1540 
 1541 ** The ODE evolution routines now allow for negative step sizes, and
 1542 integrating backwards as well as forwards.
 1543 
 1544 ** The implicit Burlisch-Stoer ODE algorithm 'bsimp' now detects
 1545 singularities and forces a reduction in step size, preventing runaway
 1546 instabilities.
 1547 
 1548 ** Fixed a bug in the ODE evolution function gsl_odeiv_evolve_apply
 1549 which could cause an erroneous value to be returned if the step size
 1550 is reduced on the last step.
 1551 
 1552 * What was new in gsl-0.9.3:
 1553 
 1554 ** Routines for complex LU decomposition are now available, allowing
 1555 the solution of systems of equations with complex coefficients.
 1556 
 1557 ** Matrix views of vectors now correctly require a unit stride for the
 1558 original vector.
 1559 
 1560 ** Permutations can now be applied to complex arrays and vectors.
 1561 
 1562 ** gsl_sf_pow_int now handles the case x = 0, n < 0
 1563 
 1564 ** The static versions of inline functions can now be hidden by
 1565 defining the preprocessor macro HIDE_INLINE_STATIC.  This is needed
 1566 for some compilers.
 1567 
 1568 ** The original seeding procedure of mt19937 is available through the
 1569 generator gsl_rng_mt19937_1998.  The seeding procedure was flawed, but
 1570 is available for compatibility.
 1571 
 1572 ** Added missing functions gsl_complex_div_real and
 1573 gsl_complex_div_imag.
 1574 
 1575 ** Missing functions for constant vector and matrix views have now been
 1576 added.
 1577 
 1578 ** Statistical calculations for histograms are now available, and the
 1579 gsl-histogram command also displays the histogram mean and standard
 1580 deviation.
 1581 
 1582 ** The behavior of GSL_IEEE_MODE for denormalized exceptions has been
 1583 fixed on Openbsd and Netbsd.
 1584 
 1585 ** A pkg-config file gsl.pc is included in the distribution
 1586 
 1587 ** The reference manual can now be printed in @smallbook format without
 1588 overflow.
 1589 
 1590 * What was new in gsl-0.9.2:
 1591 
 1592 ** Vector and matrix views are now compliant with the ANSI standard.
 1593 
 1594 ** Added Lambert functions gsl_sf_lambert_W0, gsl_sf_lambert_Wm1.
 1595 
 1596 ** The reference manual now uses the GNU Free Documentation License.
 1597 
 1598 ** Fixed a couple of bugs in the SVD routines.
 1599 
 1600 ** Macros for Infinity and Nan now work correctly with Microsoft Visual
 1601 C++, and a bug in the config.h file for the finite() function has been
 1602 fixed.
 1603 
 1604 ** Redundant entries in the test suite for the complex math functions
 1605 have been removed, making the distribution size smaller.
 1606 
 1607 ** Installed programs gsl-randist and gsl-histogram now use shared
 1608 libraries.
 1609 
 1610 * What was new in gsl-0.9.1:
 1611 
 1612 ** The single precision ffts now uses float throughout, rather than
 1613 mixing float and double.
 1614 
 1615 ** The random number distributions now include the Landau distribution.
 1616 
 1617 ** The fft function interface has been reorganized, with workspaces
 1618 separate from wavetables to eliminate unnecessary recomputation of
 1619 trigonometric factors.
 1620 
 1621 ** The gsl_interval type has been eliminated and replaced by two double
 1622 arguments for simplicity.
 1623 
 1624 ** The order of the arguments to the minimization routines is no more
 1625 logical, with function values assocatied with x-values.
 1626 
 1627 ** Modified initialization of vector and matrix views to work with the
 1628 SunPro compiler.
 1629 
 1630 ** Renamed gsl_Efunc_t to gsl_siman_Efunc_t, in accordance with
 1631 namespace conventions.
 1632 
 1633 ** Improved accuracy and fixed bugs in gsl_sf_hyperg_1F1,
 1634 gsl_sf_bessel_I0_scaled, gsl_sf_erfc, gsl_sf_log_erfc,
 1635 gsl_sf_legendre_Q0 and gsl_sf_legendre_Q1, and gsl_sf_zeta.
 1636 
 1637 ** Improved IEEE compliance of special functions, overflows now return
 1638 Inf and domain errors return NaN.
 1639 
 1640 ** Improved checking for underflows in special functions when using
 1641 extended precision registers
 1642 
 1643 * What was new in gsl-0.9:
 1644 
 1645 ** There is a new system of vector and matrix views.  Any code using
 1646 vector and matrix views will need to be updated.  
 1647 
 1648 ** The order of arguments of the view functions involving strides have
 1649 been changed to be consistent with the rest of the library.
 1650 
 1651 ** The ode solvers have been reorganized.
 1652 
 1653 ** There are new eigensystem routines for real symmetric and complex
 1654 hermitian matrices.
 1655 
 1656 ** The linear algebra directory now includes functions for computing
 1657 symmetric tridiagonal decompositions and bidiagonal decompositions.
 1658 
 1659 ** The svd routines now include the Golub-Reinsch and Modified
 1660 Golub-Reinsch algorithms in addition to the Jacobi algorithm.
 1661 
 1662 ** The interpolation directory has been reorganized and a higher-level
 1663 "spline" interface has been added which simplifies the handling of
 1664 interpolation arguments.
 1665 
 1666 ** IEEE support is now available on OpenBSD.
 1667 
 1668 * What was new in gsl-0.8:
 1669 
 1670 ** The build process now uses the latest libtool and automake.
 1671 
 1672 ** The library should now compile with Microsoft Visual C++.
 1673 
 1674 ** Portable versions of the isinf, isnan and finite functions are
 1675 available as gsl_isinf(x), gsl_isnan(x) and gsl_finite(x).
 1676 
 1677 ** The definitions of GSL_POSINF, GSL_NEGINF and GSL_NAN no longer
 1678 cause divisions by zero during compilation.
 1679 
 1680 ** The gsl_interp_obj has been renamed to gsl_interp.
 1681 
 1682 ** The poly_eval and pow_int functions have been moved from the
 1683 specfunc directory to the poly and sys directories.
 1684 
 1685 ** The Chebyshev functions are now available as an independent module
 1686 in their own directory.
 1687 
 1688 ** The error handling conventions have been unified across the
 1689 library.  This simplifies the use of the special functions.
 1690 
 1691 ** A full CBLAS implementation is now included for systems where ATLAS
 1692 has not been installed. The CBLAS library can also be used
 1693 independently of GSL.  The organisation of the BLAS directories has been
 1694 simplified.
 1695 
 1696 ** IEEE support for HPUX-11, NetBSD, Apple Darwin and OS/2 are now
 1697 included.
 1698 
 1699 ** The library now includes implementations of log1p, expm1, hypot,
 1700 acosh, asinh, atanh for platforms which do not provide them.
 1701 
 1702 ** The convention for alloc and set functions has changed so that they
 1703 are orthogonal. After allocating an object it is now necessary to
 1704 initialize it.
 1705 
 1706 ** There is a new module for estimating numerical derivatives of functions
 1707 
 1708 ** There is a new module for handling data with ntuples
 1709 
 1710 ** The histogram lookup functions are now optimized for the case of
 1711 uniform bins, and include an inline binary search for speed.
 1712 
 1713 ** The Chebyschev coefficients for the QAWO algorithm are now
 1714 precomputed in a table for efficiency, rather than being computed on
 1715 the fly.
 1716 
 1717 ** There are several new sorting functions for selecting the k-th
 1718 smallest or largest elements of a dataset.
 1719 
 1720 ** Iterator functions are now available for permutations,
 1721 gsl_permutation_next and gsl_permutation_prev.
 1722 
 1723 ** The function gsl_complex_xy has been renamed gsl_complex_rect
 1724 
 1725 ** The API for simulated annealing has been changed to support search
 1726 spaces in which the points cannot be represented as contiguous-memory
 1727 data structures.  gsl_siman_solve() now takes three extra arguments: a
 1728 copy constructor, a copy function and a destructor, allowing
 1729 gsl_siman_solve() to do its work with linked data structures.  If all
 1730 three of these function pointers are NULL, then the traditioanl
 1731 approach of using malloc(), memcpy(), and free() with the element size
 1732 is used.
 1733 
 1734 * What was new in gsl-0.7:
 1735 
 1736 ** Linux/PowerPC should now be well supported.
 1737 
 1738 ** Header files for common physical constants have been added.
 1739 
 1740 ** Functions linear and nonlinear regression in one or more dimensions
 1741 are now available.
 1742 
 1743 ** Vector and matrix views now have access to the address of the
 1744 underlying block for compatibility with VSIPL (www.vsipl.org).
 1745 
 1746 ** There is a new library for generating low-discrepancy quasi-random
 1747 sequences.
 1748 
 1749 ** The seeding procedure of the default random number generator
 1750 MT19937 has been updated to match the 10/99 release of the original
 1751 code.  This fixes a weakness which occurred for seeds which were
 1752 powers of 2.
 1753 
 1754 ** The blas library libgslblasnative has been renamed libgslblas to avoid
 1755 confusion with system blas library
 1756 
 1757 * What was new in gsl-0.6:
 1758 
 1759 ** The library is now installed as a single shared or static libgsl
 1760 file using libtool.
 1761 
 1762 ** The gsl-config script now works.  There is also a gsl.m4 file which
 1763 people can use in their configure scripts.
 1764 
 1765 ** All header files are now in installed as pkginclude headers in a
 1766 gsl/ subdirectory.
 1767 
 1768 ** The header files now use extern "C" to allow them to be included in
 1769 C++ programs
 1770 
 1771 ** For consistency the following functions have been renamed,
 1772 
 1773     gsl_vector_copy (dest, src) is now  gsl_vector_memcpy (dest, src)
 1774     gsl_rng_cpy (dest, src)     is now  gsl_rng_memcpy (dest, src)
 1775     gsl_matrix_copy_row (v,m,i) is now  gsl_matrix_get_row (v,m,i)
 1776     gsl_matrix_copy_col (v,m,j) is now  gsl_matrix_get_col (v,m,j)
 1777     gsl_vector_swap             is now  gsl_vector_swap_elements
 1778     gsl_vector_swap_cols        is now  gsl_vector_swap_columns
 1779     gsl_vector_swap_row_col     is now  gsl_vector_swap_row_column
 1780 
 1781 and the vector/matrix view allocation functions have been simplified.
 1782 
 1783 ** A new sort directory has been added for sorting objects and vectors.
 1784 
 1785 ** A permutation directory has been added for manipulating permutations
 1786 
 1787 ** Statistics functions now support a stride argument for generality, and
 1788 also support weighted samples and a covariance function.
 1789 
 1790 ** The names of the statistics functions have been reorganized for
 1791 improved clarity. Consult manual for details.
 1792 
 1793 ** The environment variable GSL_IEEE_MODE now uses "," as a separator
 1794 instead of ";"
 1795 
 1796 ** The autogen.sh script, mostly for use by developers who use the CVS
 1797 repository, now does not run configure.
 1798 
 1799 ** The histogram directory now has additional functions for copying
 1800 and comparing histograms, performing arithmetic on histograms and
 1801 finding maximum and minimum values. Corresponding functions have been
 1802 added for vectors and matrices.
 1803 
 1804 ** The linear algebra directory supports additional methods, including
 1805 rectangular QR, rectangular QRPT and Cholesky decomposition.
 1806 
 1807 ** Complex arithmetic (+,-,*,/) and complex elementary functions
 1808 (sqrt, log, exp, sin, cos, tan, arcsin, arccos, arctan, sinh, cosh,
 1809 tanh, arcsinh, arccosh, arctanh) are now supported.
 1810 
 1811 ** Multidimensional minimization methods are now available.
 1812 
 1813 ** The special functions directory now includes a routine for
 1814 computing the value of the incomplete beta function.
 1815 
 1816 * Was new in gsl-0.5:
 1817 
 1818 ** There is now a KNOWN-PROBLEMS file which lists compilation problems
 1819 and test failures which are known to the developers.
 1820 
 1821 ** Many improvements have been made to the special functions directory.
 1822 
 1823 ** The extrapolations from the Levin u-transform are now more reliable.
 1824 
 1825 ** Linear algebra and Eigensystem routines are now available.
 1826 
 1827 ** ODE solvers are now available.
 1828 
 1829 ** Multidimensional root finding algorithms are available.
 1830 
 1831 ** Minimization now keeps track of function values.
 1832 
 1833 ** Matrices and vectors now use a BLAS compatible format, and have a
 1834 separate memory handling layer (gsl_block).
 1835 
 1836 ** Roots of general polynomials can now be found using gsl_poly_complex_solve
 1837 
 1838 ** IEEE modes support on Sparclinux, Tru64, AIX and IRIX
 1839 
 1840 ** We have added the second generation RANLUX generators RANLXS and RANLXD
 1841 
 1842 ** Minimization algorithms are available (one-dimensional)
 1843 
 1844 ** Documentation now works out of the box with the standard Texinfo.
 1845 
 1846 ** Full reimplementation of the QUADPACK integration library
 1847 
 1848 ** Introduced THANKS file.
 1849 We appreciate all patches from people on the net, even those which are
 1850 too small to warrant adding the author to the AUTHORS file.  The
 1851 THANKS file should include everyone who sent in patches.  They should
 1852 also be mentioned in the ChangeLog entry.
 1853 
 1854 * What was new in gsl-0.4.1:
 1855 
 1856 ** Two changes not making their way into the documentation
 1857 A couple of things are not getting into the docs, so here are the
 1858 errata:
 1859 *** The FFT routines now take a stride parameter.  Passing 1 for the
 1860 stride will make them behave as documented.
 1861 *** The complex numbers are now an opaque type, and no assumptions can
 1862 be made about the format in which they are stored (they are not stored
 1863 as a simple structure anymore, since that is not portable).  The type
 1864 is now gsl_complex (or gsl_complex_long_double or gsl_complex_float),
 1865 and the macros to access them are
 1866     GSL_REAL(z)
 1867     GSL_IMAG(z)
 1868     GSL_COMPLEX_P_REAL(zp)
 1869     GSL_COMPLEX_P_IMAG(zp)
 1870     GSL_COMPLEX_EQ(z1,z2)
 1871     GSL_SET_COMPLEX(zp,x,y)
 1872     GSL_SET_REAL(zp,x)
 1873     GSL_SET_IMAG(zp,y)
 1874 
 1875 This change in the complex number API makes it important that you
 1876 start working with 0.4.1 or later.
 1877 
 1878 ** 0.4.1 is being released in occasion of the Red Hat 6.0 release.
 1879 The specfunc module is still in an alpha state; if you run "make
 1880 check" in the specfunc directory you will see that some tests still
 1881 fail.
 1882 
 1883 ** Most Alpha specific problems have been fixed. In particular the
 1884 random number generators rand48 and ranf now work on the Alpha
 1885 
 1886 ** Additional random number distributions: 
 1887    Rayleigh distribution
 1888    n-dimensional spherical distribution 
 1889      (ie, points at random on an n-dimensional sphere)
 1890    Gaussian tail distribution 
 1891      (ie, choosing values from a gaussian distribution subject to a 
 1892       constraint that they be larger than some fixed value, eg 5 sigmas)
 1893    Walker's algorithm for arbitrary discrete distributions
 1894 
 1895 * What was new in gsl-0.4:
 1896 
 1897 ** A single libgsl.a file is built in the top level directory and
 1898 installed, instead of separate .a files for each subdirectory.
 1899 
 1900 ** The parts of the complex struct gsl_complex, .real and .imag, are
 1901 not supported anymore. The macros GSL_REAL(z) and GSL_IMAG(z) do the
 1902 same job. All complex numbers are considered as packed arrays of
 1903 floating point numbers, for portability since the layout of structs or
 1904 arrays of structs is not guaranteed.
 1905 
 1906 ** The interface for matrices and vectors has changed. Vectors now
 1907 support strides, and can be used to access rows and columns of a
 1908 matrix. Many more types are available (float, double, long double,
 1909 int, long, short, char, signed and unsigned, plus complex floats,
 1910 doubles and long doubles) due to improvements in our preprocessor
 1911 template system.
 1912 
 1913 ** The random number generators have a completely new thread-safe
 1914 interface and have moved from the random directory to the rng
 1915 directory. Any program using random numbers will have to be
 1916 updated. You can also choose generators and seeds using the
 1917 environment variables GSL_RNG_TYPE and GSL_RNG_SEED.
 1918 
 1919 ** Some additional random number distributions have been added in the
 1920 randist directory. The available distributiosn are: bernoulli, beta,
 1921 binomial, cauchy, chisq, erlang, exponential, fdist, flat, gamma,
 1922 gauss, geometric, levy, logistic, lognormal, nbinomial, pareto,
 1923 poisson, sphere, tdist, twosidedexp, weibull.
 1924 
 1925 ** The FFT interface has be extended to support strides, but the
 1926 implementation hasn't been finished for all the cases yet, The FFT
 1927 allocation functions now return a pointer to a newly allocated
 1928 wavetable struct, instead of taking the pointer to an existing struct
 1929 as an argument.
 1930 
 1931    e.g. status = gsl_fft_wavetable_alloc(n, w)
 1932    is now    w = gsl_fft_wavetable_alloc(n) in accordance with usual practice
 1933 
 1934 ** The statistics directory now works with all the builtin
 1935 types. It has a new function for computing the lag1-autocorrelation and
 1936 an extra set of numerical accuracy tests from NIST as part of 'make
 1937 check'.
 1938 
 1939 ** The simulated annealing routines no longer set the random number
 1940 seed with the time of day.  You'll need to reseed the generator
 1941 yourself if you want subsequent runs to use different random numbers.
 1942 
 1943 ** Work is in progress on a reimplementation of QUADPACK in the
 1944 `integration' subdirectory, but it is not finished yet.
 1945 
 1946 ** Work is in progress on reimplementations of the VEGAS and
 1947 MISER Monte Carlo algorithms in the monte' subdirectory.  They work
 1948 just fine, but the code is still evolving.
 1949 
 1950 ** Work has started on a portable blas system in the `blas'
 1951 subdirectory.
 1952 
 1953 ** You can now set the IEEE arithmetic mode for your programs from the
 1954 environment variable GSL_IEEE_MODE by calling the function
 1955 gsl_ieee_env_setup(). Currently this only works with the Linux kernel,
 1956 HP-UX, SunOS4 and Solaris.
 1957 
 1958 ** There are some simple spline interpolation functions in the `interp'
 1959 subdir.
 1960 
 1961 ** The NEWS file now uses outline mode, like the Emacs NEWS file
 1962 
 1963 
 1964 * This covers changes made *after* the gsl-0.2 snapshot
 1965 
 1966 ** Added several new modules: histogram, integration, matrix, specfunc
 1967 and vectors.
 1968 
 1969 ** Changed libgsl_statisctics.a to libgslstatistics.a and
 1970 libgsl_siman.a to libgslsiman.a, since most of the packages don't have
 1971 the underscore.  Users will have to remove the old files, unless they
 1972 do a "make uninstall" on the previous release before they install this
 1973 new one.
 1974 
 1975 ** Changes to the random number suite
 1976 Incorporated the gauss, poisson and exponential distributions in
 1977 the standard libgslrandom.a
 1978 
 1979 
 1980 Local variables:
 1981 mode: outline
 1982 paragraph-separate: "[ 	]*$"
 1983 end: