* What is new in gsl-2.5:

** doc bug fix in binomial distribution figure (Damien Desfontaines)

** added Wishart distribution (Timothée Flutre)

** added new module for digital filtering (gsl_filter); current filters include:

     Gaussian filter

     median filter

     recursive median filter

     impulse detection filter

** added new module for moving window statistics (gsl_movstat)

** added statistics functions:

     gsl_stats_median()

     gsl_stats_select()

     gsl_stats_mad()

     gsl_stats_mad0()

     gsl_stats_Sn_from_sorted_data()

     gsl_stats_Qn_from_sorted_data()

     gsl_stats_gastwirth_from_sorted_data()

     gsl_stats_trmean_from_sorted_data()

** added Romberg integration (gsl_integration_romberg)

** bug fix in deprecated functions gsl_multifit_wlinear_svd and

   gsl_multifit_wlinear_usvd (reported by Vlad Koli)

** documention corrected to state that gsl_sf_legendre functions do

   not include Condon-Shortley phase by default

** bug fix in exponential fitting example when using larger number

   of points (reported by Anna Russo)

** changed internal workspace inside gsl_spmatrix to a union to

   avoid casting (suggested by Manuel Schmitz)

** bug fixes in ode-initval2 for very rare solver crashing cases:

   #52230 in msadams (reported by Michael Kaufman) and

   #52336 in msbdf (reported by Andrew Benson). As a fix,

   the maximum scaling of controlled step length was decreased

   from 5.0 to 4.9.

** add histogram2d figure to manual (was missing in 2.4)

** bug fix in gsl_spmatrix_add for duplicate input arguments

   (reported by Alfredo Correa)

** add support for negative arguments nu in gsl_sf_bessel_Jnu and

   gsl_sf_bessel_Ynu (Konrad Griessinger)

** better texinfo documentation for gsl_sf_hyperg functions

** fix vector and matrix fread/fwrite testing on windows systems

   when tmpfile() fails

** fix for rstat/test.c on PPC64 (reported by Adam Majer)

* What was new in gsl-2.4:

** migrated documentation to Sphinx software, which has built-in

   support for latex equations and figures in HTML output

** add const to declaration of appropriate gsl_rstat routines

** bug fix for #45730: change gsl_sf_sin/cos to libm sin/cos

** fix Cholesky documentation regarding upper triangle on output

** added routines to compute integrals with fixed-point quadrature,

   based on IQPACK (Konrad Griessinger)

** added routines for Hermite polynomials, gsl_sf_hermite_*

   (Konrad Griessinger)

** added new nonlinear least squares example for fitting

   a Gaussian to data

** deprecated routines:

     gsl_sf_coupling_6j_INCORRECT

     gsl_sf_coupling_6j_INCORRECT_e

** deprecated routine 'gsl_linalg_hessenberg' (replaced

   by gsl_linalg_hessenberg_decomp)

** removed routines which were deprecated in v2.1:

     gsl_bspline_deriv_alloc

     gsl_bspline_deriv_free

** changed COD expression to Q R Z^T instead of Q R Z to

   be consistent with standard texts

** added check for nz == 0 in gsl_spmatrix_get

   (reported by Manuel Schmitz)

** permit zero-dimension blocks, vectors, matrics, subvectors,

   submatrices, and views of the above (bug #49988)

** added routine gsl_linalg_COD_lssolve2 for regularized

   least squares problems

* What was new in gsl-2.3:

** bug fix in documentation for gsl_linalg_LU_refine

   (bug #49728, Joey De Pauw)

** added gsl_multifit_linear_tsvd and gsl_multifit_wlinear_tsvd

   to give user more control over cutoff for truncated SVD

** added routines for Generalized Cross Validation for

   regularized linear least squares

** improved rstat example program and added documentation for

   gsl_rstat_sd_mean (Jonathan Leto)

** added function gsl_multifit_linear_rank

** bug fix in nonlinear least squares when using data weights with

   finite-difference Jacobian

** add 2D subspace method for large systems (multilarge_nlinear)

** bug fix in gsl_ran_beta for small parameters

   (bug #47646, Yu Liu)

** bug fix in gsl_complex_tan for negative imaginary arguments

   (bug #47347, Yu Liu)

** doc bug fix: value of golden ratio

** fixed scaling issue in 2D subspace nonlinear least squares

   method

** optimize dogleg methods to calculate Gauss-Newton point

   only when needed

* What was new in gsl-2.2.1:

** reverted gsl_linalg_cholesky_decomp to its previous behavior

   so it is backward compatible; new cholesky routine is

   gsl_linalg_cholesky_decomp1

* What was new in gsl-2.2:

** updated gsl_linalg_cholesky_invert to use Level-2 BLAS

   and added function gsl_linalg_pcholesky_invert

** added functions gsl_linalg_tri_*_invert for inverting

   triangular matrices

** fix GSL_EIGEN_SORT_VAL_{ASC,DESC} for nonsymmetric

   eigensystems (Victor Zverovich)

** added complete orthogonal decomposition routines

   (gsl_linalg_COD)

** bug fix where median calculation wasn't reset in

   gsl_rstat_reset(); added gsl_rstat_quantile_reset() function

   (reported by Pedro Donato)

** added multivariate Gaussian random distribution

   gsl_ran_multivariate_gaussian (Timothée Flutre)

** added functions to estimate the 1-norm reciprocal condition

   number for various matrix factorizations:

     * gsl_linalg_cholesky_rcond

     * gsl_linalg_QRPT_rcond

** added functions gsl_linalg_QRPT_{lssolve,lssolve2} to

   compute least squares solutions with the QRPT decomposition

** added function gsl_permute_matrix()

** added modified Cholesky factorization (gsl_linalg_mcholesky)

   to handle symmetric indefinite matrices

** added pivoted Cholesky factorization (gsl_linalg_pcholesky)

   for ill-conditioned matrices

** rewrote (real) Cholesky decomposition to use

   a Level-2 blas algorithm instead of Level-1. Flop

   count is about the same but the code is much simpler

   and easier to follow

** completely rewritten nonlinear least squares module,

   including support for large problems; the user may

   now control the linear solver used, the trust region

   updating strategy, and the scaling method. In addition,

   support has been added for the geodesic acceleration

   step (Transtrum 2011) which can speed up convergence

   on a wide class of problems.

** added gsl_rstat_rms() for root mean square

** optimized lmniel nonlinear least squares solver

   (bug #46369)

** improved precision in Bessel K0/K1 near x = 2

   (Pavel Holoborodko, bug #47401)

** added support for compressed row storage sparse

   matrices (Alexis Tantet)

** bug fix in convergence check of hypergeometric 2F1

   function (bug #45926)

** added gsl_multilarge_linear_lcurve() to compute

   the L-curve for large linear systems

** updated multilarge normal equations method to use

   new Cholesky scaling for better numerical stability

** added scaling to Cholesky routines to reduce the

   condition number prior to factorization

* What was new in gsl-2.1:

** added test suite for example programs

** bug fix when compiling with #undef GSL_DISABLE_DEPRECATED

** bug fix in setting libtool age versioning

** bug fix in gsl_multifit_wlinear()

** added gsl_multifit_linear_rcond() to compute reciprocal

   condition number of least squares matrix

** added gsl_multilarge module for large linear least squares

   systems

* What was new in gsl-2.0:

** fixed bug #43258 for hypergeometric functions (Raymond Rogers)

** added L-curve analysis routines for linear Tikhonov regression

** add running statistics module

** added bilinear and bicubic interpolation (David Zaslavsky)

** added function gsl_multifit_robust_residuals to compute robust

   fit residuals

** added Steffen monotonic interpolation method (Jean-François Caron)

** added new nonlinear least squares solver 'lmniel' suitable for

   systems with large numbers of data

** nonlinear least squares solver now tracks the number of function

   and Jacobian evaluations, see example program for details

** the 'fdf' field of gsl_multifit_function_fdf is now deprecated

   and does not need to be specified for nonlinear least squares

   problems

** added extensive test suite to nonlinear least squares module,

   resulting in a few minor bug fixes; the routine

   gsl_multifit_fdfsolver_driver has been rewritten (with API change)

   to handle the various error codes of the lmsder iterate

   routine, resulting in a high level caller which is highly robust

   for a wide class of problems

** added support for sparse matrices, including a GMRES

   iterative linear solver

** added routines gsl_linalg_givens and gsl_linalg_givens_gv

   for Givens rotations

** added Tikhonov (ridge) regularization to least squares module

   (linear and nonlinear)

** removed unused argument 'n' from gsl_sf_ellint_D

** merged bspline_deriv_workspace into bspline_workspace to simplify

   bspline API; the functions

     gsl_bspline_deriv_alloc

     gsl_bspline_deriv_free

   are now deprecated and will be removed in a future release.

** merged ALF extension into GSL for associated Legendre functions;

   api has changed; consequently the functions:

     gsl_sf_legendre_Plm_array

     gsl_sf_legendre_Plm_deriv_array

     gsl_sf_legendre_sphPlm_array

     gsl_sf_legendre_sphPlm_deriv_array

     gsl_sf_legendre_array_size

   are now deprecated and will be removed in a future release.

** added function gsl_multifit_robust_weights to allow user to

   access the various weighting functions

* What was new in gsl-1.16:

** fixed error in gsl_rng_fwrite where uninitialized padding

   bytes were being written (bug #39104)

** fixed error in gsl_block_alloc where padding bytes were not

   properly initialized (bugs #39101,#39102,#39103)

** fixed error in ntuple/test.c where padding bytes were not

   properly initialized (bug #39105)

** fixed triangle selection bug in gsl_sf_coupling_6j_e and

   gsl_sf_coupling_9j_e (bugs #39466 and #29606) (Håkan Johansson and

   Alexey Illarionov)

** added higher level wrapper routine gsl_multifit_fdfsolver_driver

** converted gsl_multifit_linear_residuals to use dgemv to improve

   efficiency (bug #39153)

** added functions gsl_stats_spearman and gsl_sort_vector2 to compute

   Spearman rank correlation

** added function gsl_poly_dd_hermite_init for Hermite interpolation

** Added support for robust linear least squares

** Added function gsl_linalg_SV_leverage for computing statistical

   leverages from SVD decomposition

** Added support for approximating the Jacobian of nonlinear least

   squares fits using forward finite differences

** Extended gsl_sf_coupling_3j to allow larger range and to handle

   the special case (ja jb jc; 0 0 0)=0 when ja+jb+jc is odd

** Fixed gsl_sf_mathieu_se_array to return zero when the order is zero

   [bug #33679].

** Fixed overflow in gsl_sf_lncosh for large negative x (x<-354).

** Improved gsl_ran_negative_binomial_pdf to avoid underflow/overflow

   for large arguments.

** Multisets now allow k strictly greater than n.

** Fixed gsl_matrix_complex_fwrite/fread failure for noncontiguous

   matrices (Matthias Sitte).

* What was new in gsl-1.15:

** Added Tuomo Keskitalo's new ode branch ode-initval2 with a

   gsl_odeiv2 prefix. This provides proper support for implicit

   solvers.  It is intended to be the new default for differential

   equations.  The existing gsl_odeiv routines will be retained for

   binary compatibility but their interface will be deprecated.

** Added new gsl_integrate_cquad routines for robust integration of

   difficult functions using the doubly-adaptive CQUAD algorithm

   (Pedro Gonnet).

** Added error checking to CBLAS functions (José Luis García Pallero)

** Added a new function gsl_integration_glfixed_point to return

   ordered Gauss-Legendre points and weights contained within a

   gsl_integration_glfixed_table [bug #32237].

** Added a new function gsl_interp_type_min_size to return the size of

   an interpolation type.

** Added a function gsl_pow_uint(x,n) to compute x^n for unsigned

   exponents (needed when n exceeds the range of signed integers).

** Added new routine gsl_linalg_complex_cholesky_invert to handle the

   matrix inversion for complex Cholesky decompositions (Huan Wu).

** Added the functions gsl_vector_equal(x,y) and gsl_matrix_equal(x,y)

   for testing equality of two vectors or matrices.

** Added function gsl_eigen_nonsymmv_params to control the balancing

   transformation for eigenvector calculations. Balancing is now

   turned off by default for gsl_eigen_nonsymmv.

** It is now possible to choose an alternative cblas library via

   pkg-config using the GSL_CBLAS_LIB environment variable or

   the pkg-config --define-variable option.

** The jacobi method gsl_eigen_jacobi now uses the norm of the

   off-diagonal elements for its convergence criterion, as in

   algorithm 8.4.3 of Golub and van Loan.

** The newton multiroot solvers now return an error when a singular

   jacobian is detected.

** The interpolation functions now return NaN and when x is out of range,

   instead of extrapolating.

** The gsl_multimin_fdfsolver multidimensional minimisers now return

   GSL_ENOPROG immediately if the generated trial point does not

   differ from the initial point (to machine precision), avoiding

   unnecessary further iterations.

** Extended the range of gsl_sf_bessel_lnKnu_e by rescaling

   intermediate results to avoid internal overflows [bug #31528].

** Improved the result of gsl_sf_atanint_e for large arguments by

   adding the first order 1/x correction term. [bug #29562]

** Fixed the gsl_rng_ranlxs generators to enforce a maximum seed value

   of 2^31-1.  Larger seed values corrupted the random number state

   and caused out of range values to be returned.

** Fixed gsl_ran_chisq_pdf(x,nu) to return correct result of 1/2

   instead of 0 when x=0 and nu=2, and +inf when x=0 and nu<2.

** Fixed gsl_pow_int(x,n) to avoid an infinite loop when n=INT_MIN due

   to wrapping of signed integers.

** Fixed gsl_sf_hyperg_2F1(a,b,c,x) to avoid returning NaN for

   arguments |a|>10. [bug #24812]

** Avoid spurious underflow return code in gsl_sf_beta_inc_e when

   intermediate underflow does not affect the result. [bug #30933]

** Avoid segfault in Chebyshev series derivatives gsl_cheb_calc_deriv

   for n=1. [bug #29139]

* What was new in gsl-1.14:

** Upgraded to latest libtool, autoconf and automake (libtool-2.2.6b,

   autoconf-2.65, automake-1.11.1).  Fixes security hole in 'make

   dist' (see Automake CVE-2009-4029).

** Added support for "multisets", which are similar to permutations

   and combinations.  A multiset is an array of k integers in the

   range 0 to n-1 where each value may occur more than once. Multisets

   can be used to iterate over the indices of a k-th order symmetric

   tensor in n-space. (Rhys Ulerich)

** Added gsl_integrate_glfixed routines for performing fixed order

   Gauss-Legendre integration 

   (Pavel Holoborodko, Konstantin Holoborodko, Rhys Ulerich)

** In the LMDER multi-dimensional fitting routines, the return code

   GSL_ENOPROG is now used instead of GSL_CONTINUE for the case where

   10 or more attempts to find a suitable trial step have been made

   without success. [bug #25383]

** The confluent hypergeometric function gsl_sf_hyperg_U (a,b,x) has

   been extended to support x < 0 for cases not involving

   singularities in 1F1(a,b,x). [bug #27859]

** The F-distribution gsl_ran_fdist_pdf now avoids unnecessary

   underflow and overflow and handles cases where n > 248. [bug

   #28500]

** The SVD routines now use a scaled version of the implicit QR method

   and compute the shift more reliably for values at the limit of

   double precision, avoiding potential NaNs. [bug #28767]

** A compile time error is emitted if the configuration stage prevents

   the functions gsl_isnan and gsl_finite from being defined.

** Added missing dereference in GSL_VECTOR_COMPLEX when not using

   GSL_RANGE_CHECK [bug #28017]

** Improved the range of gsl_sf_hyperg1F1(a,b,x) when a<0,b>0. [bug

   #28718]

** Added macros GSL_MAJOR_VERSION and GSL_MINOR_VERSION in

   <gsl/gsl_version.h>

** Improved gsl_eigen_symmv and gsl_eigen_symm to avoid a potential

   infinite loop when the tridiagonalised matrix contains very small

   values. [bug #28096]

** Added functions gsl_multifit_linear_usvd and

   gsl_multifit_wlinear_usvd for multilinear fitting without

   column-scaling of the fit matrix.

* What was new in gsl-1.13:

** Upgraded to latest autoconf and automake (autoconf-2.64,

   automake-1.11)

** Fixed the rk4 and bspline allocators to avoid invalid free()

   calls under out of memory conditions. [bug #27194, #27236]

** Fixed a bug in gsl_multimin_fminimizer_nmsimplex2 where the center

   and size of the simplex were not updated on contract-by-best steps,

   causing failures in convergence. [bug #27180]

** Added new functions to set MISER and VEGAS Monte Carlo integration

   parameters, and to examine VEGAS chi-squared value and intermediate

   results.

** Added the function gsl_bspline_greville_abscissa to compute

   Greville abscissae for B-splines.

** The cumulative distribution functions gsl_cdf_gumbel1_{P,Q} should

   now handle a larger range of parameters without underflow and

   overflow.

** The header file gsl_const_cgs.h no longer defines values for

   electromagnetic units.  Applications should use gsl_const_cgsm.h

   instead to obtain the values in the CGS-Magnetic system.  The

   previous values for these units given in gsl_const_cgs.h were

   ill-defined as the type of CGS electromagnetic system was

   unspecified (the values were a mixture of CGS units with the Ampere

   of the MSKA system).  The affected constants are

   GSL_CONST_CGS_BOHR_MAGNETON, GSL_CONST_CGS_ELECTRON_CHARGE,

   GSL_CONST_CGS_ELECTRON_MAGNETIC_MOMENT, GSL_CONST_CGS_FARADAY,

   GSL_CONST_CGS_GAUSS, GSL_CONST_CGS_NUCLEAR_MAGNETON,

   GSL_CONST_CGS_PROTON_MAGNETIC_MOMENT, and GSL_CONST_CGS_ROENTGEN.

** The Pochhammer functions gsl_sf_poch(a,x) and gsl_sf_lnpoch(a,x) now

   handle the special cases where a and a+x are zero or negative

   integers.

** The confluent hypergeometric function gsl_sf_hyperg_U (a,b,x) now

   handles some cases where x=0. The case where 1+a-b is a negative

   integer no longer returns an error [bug #22859] and the incorrect

   termination of the series in certain cases is fixed [bug #26706].

** Added a new function gsl_poly_eval_derivs to evaluate a polynomial

   and its derivatives simultaneously.

** Added a new univariate minimisation algorithm

   gsl_min_fminimizer_quad_golden which is a variant of Brent's

   algorithm with safeguarded step-length adjustment.

** Added a new Nelder-Mead minimiser gsl_multimin_fminimizer_nmsimplex2rand

   which uses a randomly oriented simplex rather than one fixed on

   the coordinate axes [bug #25077]

** The texinfo file now uses the dircategory "Software libraries" from

   the Free Software Directory, as recommended in the Texinfo manual.

** The function gsl_ran_exponential now includes zero in its output

   range. [bug #25039]

** All functions for freeing allocated memory now accept a NULL

   pointer, following the standard C convention for free(). [bug

   #25319]

** The function gsl_sum_levin_u_accel now handles the special case

   c_0 + 0 + 0 + 0 + .... that occurs when summing power series

   c_n*x^n with x=0. [bug #26807]

** The functions gsl_linalg_LU_solve, gsl_linalg_LU_svx,

   gsl_linalg_LU_refine, gsl_linalg_LU_invert and their complex

   equivalents now return an error for singular matrices.

** The multifit LMDER hybrid solvers now check the return code of the

   user-supplied function in the gsl_multifit_fdfsolver_set

   method. [bug #26871]

** Improved the implementation of gsl_ran_discrete_preproc to avoid

   internal errors due to inconsistencies from excess precision on

   some platforms. [bug #26502]

** Corrected gsl_sf_hyperg_2F1(a,b,c,x) to not give a domain error in

   the case where c is a negative integer and the series terminates

   with a finite result.

** The C99 inline keyword is now supported, in addition to the

   previously supported GNU-style inline.

** Modified gsl_poly_complex_solve_cubic and gsl_poly_solve_cubic to

   avoid returning NaNs in cases where excess precision causes a

   change in the number of roots. 

** Fixed incorrect length check in gsl_blas_drotm. [bug #26503]

** Fixed gsl_odeiv_step_gear2 to restore y on step failure

** gsl_odeiv_evolve_apply now restores the correct value of t on step

   failures [bug #26255].

** Using make install prefix=DIR now puts correct paths in package

   config files gsl-config and gsl.pc

** Modified gsl_monte_vegas to work around pow() function inaccuracies

   on MinGW [bug #25413].

** Increased the number of terms in gsl_sf_mathieu_a and

   gsl_sf_mathieu_b to improve convergence in difficult regions [bug

   #25075]

* What was new in gsl-1.12:

** Upgraded to latest libtool, autoconf and automake (libtool-2.2.6,

   autoconf-2.63, automake-1.10.2)

** Improved the convergence of gsl_sf_gamma_inc_P for x/a ~=~ 1 and

   large x,a.  Fixes problems with large arguments in cdf functions

   such as gsl_cdf_chisq_Pinv(x,nu) [bug 24704].

** Fixed gsl_ran_gamma_knuth to handle the case of a >= UINT_MAX [bug

   #24897]

** Added gsl_bspline_eval_deriv to compute bspline derivatives

   (Rhys Ulerich)

** Added a faster simplex mininimser gsl_multimin_fminimizer_nmsimplex2

   which is O(N) instead of O(N^2) [bug #24418]

** Improved the original chi-squared formula in gsl_monte_vegas to

   avoid catastrophic cancellation [bug #24510].  The previous formula

   could return incorrect or negative values for relative errors <

   1e-8, which could occur when integrating very smooth functions.

** Added new auxiliary functions gsl_cheb_order, gsl_cheb_size,

   gsl_cheb_coeffs for Chebyshev series [bug #21830]

** Updated license of the reference manual to GNU FDL version 1.3.

** Fixed a bug where the gsl_isinf function would return +1 for -Inf

   on systems where isinf(-Inf) returns the non-standard value +1.

   [bug #24489]

** Added missing functions gsl_vector_complex_{isnonneg,add,sub,mul,

   div,scale,add_constant} and gsl_matrix_complex_float_isnonneg [bug

   #22478]

** Cross compilation should now work for x86 hosts.

** Fixed a bug in gsl_interp_accel_find() where values lying on the

   upper boundary between interpolation points could return the index

   from the lower side. [bug #24211]

** Fixed gsl_linalg_solve_cyc_tridiag so that its output respects the

   solution vector's stride. Previously the x_stride value was ignored

   causing the output to be incorrect for non-unit stride. [bug #24162]

** Corrected a bug in the series calculation of gsl_sf_ellint_Kcomp

   for k close to 1. [bug #24146]

** Extended gsl_linalg_QRPT_update to handle rectangular matrices.

   Corrected definition of the update formula in the manual for

   both gsl_linalg_QR_update and gsl_linalg_QRPT_update.

** Added routine gsl_linalg_cholesky_invert

** Fixed a bug the simplex algorithm which caused the second highest

   point to be incorrectly equal to the first when the first value was

   the highest, which could cause suboptimal convergence. [bug #23192]

** Fixed a problem with convergence for inverse gamma and chisq

   distribitions, gsl_cdf_gamma_{P,Q}inv and gsl_cdf_chisq_{P,Q}inv.

   [bug #23101]

** Improved the handling of constant regions in Vegas by eliminating

   spurious excess precision when computing box variances.

** Fixed a potential division by zero in gsl_monte_miser when the

   left/right weight factors decrease below 1.

** Fixed incorrect dimensions check in gsl_matrix_sub{row,column}

* What was new in gsl-1.11:

** The GSL repository and bug database are now hosted at Savannah

   http://savannah.gnu.org/projects/gsl/

** Upgraded to latest libtool, autoconf and automake (libtool-2.2,

   autoconf-2.61, automake-1.10.1)

** Fixed underflow in ODE adaptive step size controller that could

   cause step size to decrease to zero (bug #21933).

** Improved the handling of the asymptotic regime in gsl_sf_bessel_jl.

** Improved the handling of large arguments in cumulative distribution

   functions using the incomplete beta function, such as gsl_cdf_fdist_P.

** Fixed overflow bug in gsl_cdf_hypergeometric_{P,Q} for large

   arguments (bug #22293).

** gsl_ran_gaussian_ziggurat now handles generators with different

   ranges explicitly, to minimise the number of function calls

   required (bug #21820).  Also fixes bug #22446 (rng limit in

   gsl_ran_chisq()).

** Added missing error terms in gsl_sf_exp_mult_e10_e to prevent

   the error being underestimated (bug #22041).  

** Updated some constants to the CODATA 2006 values.

** The hypergeometric function gsl_sf_hyperg_2F1 now handles the case

   where x==1.

** Fixed a bug in the brent minimiser which prevented optimal convergence.

** Added functions for evaluating complex polynomials

** The convergence condition for gsl_multiroots_test_delta now accepts

   dxi == 0.

** Improved functions gsl_ldexp and gsl_frexp to handle the full range

   of double precision numbers in all cases.

** Added new quasi random generators gsl_qrng_halton and

   gsl_qrng_reversehalton which support dimensions up to 1229.

** Added function gsl_multifit_linear_residuals for computing the

   residuals of the fit

* What was new in gsl-1.10:

** License updated to GNU GPL version 3.

** Added support for generalized eigensystems

** Added function gsl_stats_correlation to compute Pearson correlation

of two datasets

** Added the new function gsl_sf_expint(n,x) for computing the n-th

order exponential integral.

** Added functions gsl_vector_isnonneg and gsl_matrix_isnonneg.

** Added functions gsl_matrix_subrow and gsl_matrix_subcolumn

** Extended Cholesky routines to complex matrices

** Added support in gsl_ieee_set_mode for controlling SSE exceptions

and rounding through the MXCSR control word on x86 processors.

** The autoconf macro AM_PATH_GSL has been renamed to AX_PATH_GSL, to

avoid conflicts with the autoconf namespace.

** Improved handling of underflow in gsl_eigen_symm.

** The function gsl_multiroot_fdjacobian now returns the error code

GSL_ESING if any of the columns of the computed jacobian matrix are

zero.  This may occur if the step size of the derivative is too small.

** Extended the function gsl_sf_beta_inc(a,b,x) to handle cases where

a<0 or b<0.

** Fixed the round-off error estimate in gsl_deriv_{central,backwards,

forward} to correctly account for numerical error in the step-size h.

** Fixed gsl_cdf_beta_Pinv, gsl_cdf_gamma_Pinv, gsl_cdf_beta_Pinv to

avoid returning spurious values for large parameters when the

iteration did not converge.  If the iteration cannot converge, GSL_NAN

is returned.

** gsl_ran_dirichlet now handles smaller values of alpha[] without

underflow, avoiding a NaN in the returned value.

** The SVD routines now avoid underflow in the Schur decomposition for

matrices with extremely small values 

** gsl_complex_pow now returns 0^0=1 (instead of zero) to match the usual

pow function, and handles z^(1,0) and z^(-1,0) as special cases.

** Fixed a bug in the set function for multifit lmder solver so that

previous values are cleared correctly.

** Improved the function gsl_log1p to prevent possible loss of

accuracy caused by optimisations.

** Improved the convergence test in the Lambert functions to take

account of finite precision and avoid possible failure to converge.

** The configure script no longer redefines finite() to isfinite() as

a workaround for missing finite(), as this caused problems on Solaris.

If finite() is not present gsl_finite() will be used instead.

** Improved the accuracy of the generalised laguerre polynomials for

large n when alpha=0.

** The functions gsl_{isnan,isinf,finite} will now always use the

system versions of isnan, isinf and finite if they are available.

Previously the portable GSL implementations were used whenever the

platform supported IEEE comparisons.  The portable gsl_isinf function

now returns -1 instead of +1 for -Inf, in line with recent versions of

the GNU C Library.

* What was new in gsl-1.9:

** Fixed the elliptic integrals F,E,P,D so that they have the correct

behavior for phi > pi/2 and phi < 0.  The angular argument is now

valid for all phi.  Also added the complete elliptic integral

gsl_sf_ellint_Pcomp.

** Added a new BFGS minimisation method gsl_multimin_fdfminimizer_vector_bfgs2

based on the algorithm given by R.Fletcher in "Practical Methods of

Optimisation" (Second edition).  This requires substantially fewer

function and gradient evaluations, and supercedes the existing BFGS

minimiser.

** The beta functions gsl_sf_beta_e(a,b) and gsl_sf_lnbeta_e(a,b) now

handle negative arguments a,b.  Added new function gsl_sf_lnbeta_sgn_e

for computing magnitude and sign of negative beta values, analagous to

gsl_sf_lngamma_sgn_e.

** gsl_cheb_eval_mode now uses the same error estimate as

gsl_cheb_eval_err.

** Improved gsl_sf_legendre_sphPlm_e to avoid underflow with large

arguments.

** Added updated Knuth generator, gsl_rng_knuthran2002, from 9th

printing of "The Art of Computer Programming".  Fixes various

weaknesses in the earlier version gsl_rng_knuthran.  See

http://www-cs-faculty.stanford.edu/~knuth/news02.htm

** The functions gsl_multifit_fsolver_set, gsl_multifit_fdfsolver_set

and gsl_multiroot_fsolver_set, gsl_multiroot_fdfsolver_set now have a

const qualifier for the input vector x, reflecting their actual usage.

** gsl_sf_expint_E2(x) now returns the correct value 1 for x==0,

instead of NaN.

** The gsl_ran_gamma function now uses the Marsaglia-Tsang fast gamma

method of gsl_ran_gamma_mt by default.

** The matrix and vector min/max functions now always propagate any

NaNs in their input.

** Prevented NaN occuring for extreme parameters in

gsl_cdf_fdist_{P,Q}inv and gsl_cdf_beta_{P,Q}inv

** Corrected error estimates for the angular reduction functions

gsl_sf_angle_restrict_symm_err and gsl_sf_angle_restrict_pos_err.

Fixed gsl_sf_angle_restrict_pos to avoid possibility of returning

small negative values.  Errors are now reported for out of range

negative arguments as well as positive.  These functions now return

NaN when there would be significant loss of precision.

** Corrected an error in the higher digits of M_PI_4 (this was beyond

the limit of double precision, so double precision results are not

affected).

** gsl_root_test_delta now always returns success if two iterates are

the same, x1==x0.

** A Japanese translation of the reference manual is now available

from the GSL webpage at http://www.gnu.org/software/gsl/ thanks to

Daisuke TOMINAGA.

** Added new functions for basis splines, see the "Basis Splines"

chapter in the GSL Reference Manual for details.

** Added new functions for testing the sign of vectors and matrices,

gsl_vector_ispos, gsl_vector_isneg, gsl_matrix_ispos and

gsl_matrix_isneg.

** Fixed a bug in gsl_sf_lnpoch_e and gsl_sf_lnpoch_sgn_e which caused

the incorrect value 1.0 instead of 0.0 to be returned for x==0.

** Fixed cancellation error in gsl_sf_laguerre_n for n > 1e7 so that

larger arguments can be calculated without loss of precision.

** Improved gsl_sf_zeta_e to return exactly zero for negative even

integers, avoiding less accurate trigonometric reduction.

** Fixed a bug in gsl_sf_zetam1_int_e where 0 was returned instead of

-1 for negative even integer arguments.

** When the differential equation solver gsl_odeiv_apply encounters a

singularity it returns the step-size which caused the error code from

the user-defined function, as opposed to leaving the step-size

unchanged.

** Added support for nonsymmetric eigensystems

** Added Mathieu functions

* What was new in gsl-1.8:

** Added an error check to trap multifit calls with fewer observations

than parameters.  Previously calling the multifit routines with n

would cause invalid memory access.

** Added the Debye unit to physical constants.

** Added cumulative distribution functions for the discrete

distributions, including binomial, poisson, geometric, negative

binomial, pascal and hypergeometric.

** Added the functions gsl_cdf_beta_{Pinv,Qinv} and

gsl_cdf_fdist_{Pinv,Qinv} for computing the inverse of the cumulative

beta and F distributions.