2011-10-10  Rhys Ulerich  <rhys.ulerich@gmail.com>

	* hyperg_U.c (gsl_sf_hyperg_U_int_e,gsl_sf_hyperg_U_e):

	Always initialize the gsl_sf_result_e10 instances.

	Thanks to Victor Zverovich for the bug report and patch.

2011-10-10  Brian Gough  <bjg@network-theory.co.uk>

	* coupling.c (gsl_sf_coupling_3j_e): compute 3j factors using logs

	to allow larger range

2011-09-20  Brian Gough  <bjg@network-theory.co.uk>

	* coupling.c (triangle_selection_fails): extend triangle selection

	to handle all permutations

	(gsl_sf_coupling_3j_e): special case for (ja jb jc; 0 0 0) = 0

	when ja+jb+jc=odd

2011-08-10  Brian Gough  <bjg@network-theory.co.uk>

	* mathieu_radfunc.c (gsl_sf_mathieu_Mc): set odd functions to zero

	for order=0, initialise fn to zero before use

2011-07-21  Brian Gough  <bjg@gnu.org>

	* mathieu_angfunc.c (gsl_sf_mathieu_se_array): handle the case

	where q=0

2011-07-15  Brian Gough  <bjg@network-theory.co.uk>

	* trig.c (gsl_sf_lncosh_e): handle x symmetrically for middle

	region

2010-11-11  Brian Gough  <bjg@gnu.org>

	* ellint.c (gsl_sf_ellint_RC_e, gsl_sf_ellint_RD_e)

	(gsl_sf_ellint_RF_e, gsl_sf_ellint_RJ_e): introduce a limit of

	10000 iterations to avoid infinite loops

	* bessel_Knu.c (gsl_sf_bessel_Knu_scaled_e10_e): alternative

	version of Knu_scaled function to allow greater range for

	gsl_sf_bessel_lnKnu_e.

2010-10-29  Brian Gough  <bjg@gnu.org>

	* hyperg_U.c (hyperg_U_small_a_bgt0): corrected result->err for

	case where a==0.0

2010-08-31  Brian Gough  <bjg@network-theory.co.uk>

	* beta_inc.c (gsl_sf_beta_inc_e): ignore underflow error when term

	is subtracted from 1

2010-08-27  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_2F1.c (gsl_sf_hyperg_2F1_e): use ap and bp consistently

	in large c and large b cases, previously mixed a,b and ap,bp.

2010-04-15  Brian Gough  <bjg@network-theory.co.uk>

	* atanint.c (gsl_sf_atanint_e): added missing term 1/ax for large x

2010-02-25  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_U.c (hyperg_U_negx): handle the case where x<0

2010-01-23  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_1F1.c (gsl_sf_hyperg_1F1_e): use Kummer transformation

	for larger range of x when b>a and a<0

2009-07-17  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_U.c (hyperg_U_series): use a rearrangement of the factor

	in front of the series to avoid incorrect termination when the

	the leading term is zero.

2009-07-16  Brian Gough  <bjg@network-theory.co.uk>

	* poch.c (gsl_sf_lnpoch_sgn_e): handle negative and zero values of

	a, e.g. where gamma(a) and/or gamma(a+x) is infinite

	(gsl_sf_poch_e): handle the case where lnpoch_sgn returns

	GSL_NEGINF when the Pochhammer ratio is zero.

	* test_gamma.c (test_gamma): added tests for poch(a,x) with

	negative arguments

2009-07-12  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_U.c (hyperg_U_int_bge1): avoid using the series when

	1+a-b is zero or a negative integer.

2009-07-11  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_U.c (gsl_sf_hyperg_U_int_e10_e, gsl_sf_hyperg_U_e10_e)

	(hyperg_U_origin, hyperg_U_int_origin): added special case for

	U(a,b,z=0).

2009-07-09  Brian Gough  <bjg@network-theory.co.uk>

	* mathieu_workspace.c (gsl_sf_mathieu_free): handle NULL argument

	in free

2009-05-13  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_2F1.c (gsl_sf_hyperg_2F1_e): fix condition on a and b

	when c is a negative integer (either a or b must cause

	cancellation of the series)

2009-01-14  Brian Gough  <bjg@network-theory.co.uk>

	* mathieu_workspace.c (gsl_sf_mathieu_alloc): increase number of

	terms

	* mathieu_charv.c (gsl_sf_mathieu_a): increase number of terms

	(gsl_sf_mathieu_b): increase number of terms

2008-12-04  Brian Gough  <bjg@network-theory.co.uk>

	* gamma_inc.c (gamma_inc_D): propagate cancellation error

	in (x-a)/x for x close to a

2008-12-03  Brian Gough  <bjg@network-theory.co.uk>

	* exp.c (exprel_n_CF): changed N to double, to allow non-integer

	usage for gamma_inc, double error factor to allow for two

	parameters in recurrence.

	(gsl_sf_exprel_n_CF_e): exported function to allow calls to

	exprel_n_CF

	* gamma_inc.c (gamma_inc_P_series): improved convergence condition

	using estimate of the remainder of the series, added continued

	fraction as a fallback, increased nmax to 10000

2008-08-26  Brian Gough  <bjg@network-theory.co.uk>

	* ellint.c (gsl_sf_ellint_Kcomp_e): corrected taylor expansion for

	k close to 1.

2008-07-03  Brian Gough  <bjg@hp2.network-theory.co.uk>

	* Makefile.am (INCLUDES): use top_srcdir instead of top_builddir

2008-06-27  Brian Gough  <bjg@network-theory.co.uk>

	* legendre_poly.c (gsl_sf_legendre_array_size): removed inline

	version of this function in gsl_sf_legendre.h

2008-03-17  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_2F1.c (gsl_sf_hyperg_2F1_e): replace duplicate error

	check on stat3 by stat4

2008-03-15  Brian Gough  <bjg@network-theory.co.uk>

	* bessel.c (gsl_sf_bessel_Jnu_asympx_e): use full asymptotic

	series

2008-03-14  Brian Gough  <bjg@network-theory.co.uk>

	* bessel_j.c (gsl_sf_bessel_jl_e): increase error estimate by

	factor of 2

2008-02-09  Brian Gough  <bjg@network-theory.co.uk>

	* exp.c (gsl_sf_exp_e10_e): only use exponent e10 when standard

	exp() is out of range

	(gsl_sf_exp_mult_e10_e): add full set of error terms

2008-01-10  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_2F1.c (gsl_sf_hyperg_2F1_e): handle case of x==1

2007-10-25  Brian Gough  <bjg@network-theory.co.uk>

	* bessel.c (gsl_sf_bessel_J_CF1): handle underflow by rescaling in

	recurrence

2007-09-13  Brian Gough  <bjg@network-theory.co.uk>

	* ellint.c (gsl_sf_ellint_Pcomp_e): remove test for n <1, no

	restriction needed

2007-09-10  Brian Gough  <bjg@network-theory.co.uk>

	* expint.c (gsl_sf_expint_En_e): added for support En(x)

	(gsl_sf_expint_En_scaled_e): added for support En(x) scaled

	* gamma_inc.c (gamma_inc_CF): include finite precision of log term

2007-08-30  Brian Gough  <bjg@network-theory.co.uk>

	* psi.c (psi_complex_rhp): accumulate error, to allow for

	cancellation effects

	* beta.c (gsl_sf_lnbeta_sgn_e): added missing factor of 2 for

	error.

2007-08-27  Brian Gough  <bjg@network-theory.co.uk>

	* beta_inc.c (gsl_sf_beta_inc_e): handle cases where a<=0 or b<=0

2007-04-27  Brian Gough  <bjg@network-theory.co.uk>

	* lambert.c (halley_iteration): increase tolerance to prevent

	exceeding max iters due to finite precision

2007-04-04  Brian Gough  <bjg@network-theory.co.uk>

	* laguerre.c (gsl_sf_laguerre_n_e): use recursion for a=0 in

	addition to a>0

2007-02-17  Brian Gough  <bjg@network-theory.co.uk>

	* log.c (gsl_sf_log_e): removed HIDE_INLINE_STATIC

	* exp.c (gsl_sf_exp_e): removed HIDE_INLINE_STATIC

2007-02-14  Brian Gough  <bjg@network-theory.co.uk>

	* mathieu_charv.c: made solve_cubic static

2007-02-12  Brian Gough  <bjg@network-theory.co.uk>

	* mathieu_charv.c (figi): ensure that e[ii] is set when

	e2[ii]==0.0 and there is no error condition, as per the original

	eispack routine

2007-02-09  Brian Gough  <bjg@network-theory.co.uk>

	* ellint.c (gsl_sf_ellint_F_e): do argument reduction for phi>pi/2

	(gsl_sf_ellint_E_e): do argument reduction for phi>pi/2

	(gsl_sf_ellint_P_e): do argument reduction for phi>pi/2

	(gsl_sf_ellint_D_e): do argument reduction for phi>pi/2

	(gsl_sf_ellint_Dcomp_e): added complete D integral

	(gsl_sf_ellint_Pcomp_e): added complete P integral

2007-01-31  Brian Gough  <bjg@network-theory.co.uk>

	* beta.c (gsl_sf_lnbeta_sgn_e): added to support calculations with

	negative a,b

	(gsl_sf_lnbeta_e): rewritten in terms of gsl_sf_lnbeta_sgn_e

	(gsl_sf_beta_e): handle negative a,b

	* gamma.c (gsl_sf_lngamma_sgn_e): make error calculations an exact

	copy of gsl_sf_lngamma_e (these functions could be merged to avoid

	duplication)

2007-01-29  Brian Gough  <bjg@network-theory.co.uk>

	* test_legendre.c (test_legendre): added extra test cases for

	underflow

2007-01-26  Brian Gough  <bjg@network-theory.co.uk>

	* expint.c (expint_E2_impl): handle x==0.0 as a special case

	(expint_E2_impl): corrected error term

	* gsl_sf_log.h: removed inline version of log

	* gsl_sf_exp.h: removed inline version of exp

2007-01-23  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_1F1.c (hyperg_1F1_1_series): increase accuracy by factor

	of 4 in sum, tighter convergence condition, increase error

	estimate to allow for accumulated roundoff

	(hyperg_1F1_1): use series when |x| > |b|

	(gsl_sf_hyperg_1F1_e): only use Kummer when |x| < 100 otherwise

	exponential takes extreme value

	* hyperg.c (gsl_sf_hyperg_1F1_series_e): allow 10000 iterations in

	series to extend valid range

	(gsl_sf_hyperg_1F1_series_e): increase accuracy by factor of 4 in

	sum, tighter convergence condition

2007-01-19  Brian Gough  <bjg@network-theory.co.uk>

	* laguerre.c (laguerre_large_n): use the second term in the

	asymptotic expansion from Slater, p.73.

2007-01-17  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_1F1.c (hyperg_1F1_largebx): asymptotic expansion for

	large b and x, with |x|<|b| from Slater 4.3.7

	(hyperg_1F1_1): use new asymptotic expansion for |x|<|b|

	(hyperg_1F1_small_a_bgt0): use new asymptotic expansion for |x|<|b|

	(hyperg_1F1_renorm_b0): add neglected terms in expansion for AS13.3.7

2007-01-14  Brian Gough  <bjg@network-theory.co.uk>

	* legendre_poly.c (gsl_sf_legendre_sphPlm_e): added explicit

	computation of error term to allow for case when final term is

	zero.

2007-01-12  Brian Gough  <bjg@network-theory.co.uk>

	* trig.c (gsl_sf_angle_restrict_symm_err_e): compute edge cases

	more reliably, return NaN when total loss of precision

	(gsl_sf_angle_restrict_pos_err_e): as above

	* legendre_poly.c (gsl_sf_legendre_Pl_e): improve error estimate

	for large l by including rounding error at each step of recurrence

2006-10-03  Brian Gough  <bjg@network-theory.co.uk>

	* poch.c (gsl_sf_lnpoch_e, gsl_sf_lnpoch_sgn_e): corrected

	result->val to 0.0 for x==0, previously returned incorrect value

	1.0

2006-09-24  Brian Gough  <bjg@network-theory.co.uk>

	* laguerre.c (laguerre_large_n): work with small angles to avoid

	cancellation error, computer angular reduction exactly for integer

	eta.

2006-09-22  Brian Gough  <bjg@network-theory.co.uk>

	* zeta.c (gsl_sf_zeta_e): make sin_term exactly zero for negative

	even integers

	(gsl_sf_zetam1_int_e): return value is -1 for zetam1_int with

	negative even integers

2006-03-26  Brian Gough  <bjg@network-theory.co.uk>

	* fermi_dirac.c (fd_neg): initialize s to zero (avoid spurious

	warning from compiler)

2006-02-23  Brian Gough  <bjg@network-theory.co.uk>

	* coulomb.c (gsl_sf_coulomb_wave_FG_e): fixed sign of F_lam_min,

	covers case when k_lam_G is nonzero and F_lam_min and F_lam differ

	in sign.

2006-01-21  Brian Gough  <bjg@network-theory.co.uk>

	* synchrotron.c (gsl_sf_synchrotron_1_e): added first order

	correction term for the taylor expansion

	(gsl_sf_synchrotron_2_e): added first order correction term for

	the taylor expansion

2006-01-20  Brian Gough  <bjg@network-theory.co.uk>

	* bessel_j.c (gsl_sf_bessel_jl_e): limit the use of

	gsl_sf_bessel_Jnu_asympx_e to the range x>100*l*l to satisfy

	the requirement x>>l*l in the asymptotic expansion

	* bessel_In.c (gsl_sf_bessel_In_scaled_e): limit the use of

	gsl_sf_bessel_I_CF1_ser to the region where the continued

	converges with the allowed 20000 terms (|x|<1e7)

2005-12-20  Brian Gough  <bjg@network-theory.co.uk>

	* debye.c (gsl_sf_debye_5_e, gsl_sf_debye_6_e): added n=5,6

	(R. J. Mathar)

2005-11-15  Brian Gough  <bjg@network-theory.co.uk>

	* dilog.c (dilog_xge0): fix calculation of error estimate

2005-08-04  Brian Gough  <bjg@network-theory.co.uk>

	* coulomb.c (gsl_sf_coulomb_wave_sphF_array): fix warning about

	variable shadowing for k

2005-07-29  Brian Gough  <bjg@network-theory.co.uk>

	* gamma_inc.c (gsl_sf_gamma_inc_Q_e): use continued fraction close

	to peak of the integrand x > a - sqrt(a)

2005-07-28  Brian Gough  <bjg@network-theory.co.uk>

	* elljac.c (gsl_sf_elljac_e): use separate iterations to avoid

	division by zero, new algorithm from Bulirsch avoids inverse trig

	functions.

2005-07-03  Brian Gough  <bjg@network-theory.co.uk>

	* legendre_poly.c (gsl_sf_legendre_sphPlm_e): compute exponential

	internally to avoid underflow error when calling gsl_sf_exp_err

2005-07-02  Brian Gough  <bjg@network-theory.co.uk>

	* bessel_i.c (gsl_sf_bessel_il_scaled_array): handle x==0 as a

	special case

	* bessel_k.c (gsl_sf_bessel_kl_scaled_array): added lmax==0 as a

	special case

	* bessel_y.c (gsl_sf_bessel_yl_array): added lmax==0 as a special

	case

	* transport.c (gsl_sf_transport_2_e): improve error estimate for

	small x

2005-06-26  Brian Gough  <bjg@network-theory.co.uk>

	* test_sf.c (test_jac): added tests over the full period for

	elljac functions

2005-05-23  Brian Gough  <bjg@network-theory.co.uk>

	* test_bessel.c (test_bessel): added test for steed(99,1,...)

	* legendre_H3d.c (legendre_H3d_CF1): use hypot

	(legendre_H3d_CF1_ser): use hypot

	(gsl_sf_legendre_H3d_e): use hypot

	(gsl_sf_legendre_H3d_e): use hypot

	(gsl_sf_legendre_H3d_array): use hypot

	* dilog.c (dilogc_unitdisk): use hypot

	* coulomb.c (gsl_sf_coulomb_CL_array): use hypot

	(coulomb_F_recur): use hypot

	(coulomb_G_recur): use hypot

	(coulomb_jwkb): use hypot

	(gsl_sf_coulomb_wave_F_array): use hypot

	(gsl_sf_coulomb_wave_FG_array): use hypot

	(gsl_sf_coulomb_wave_FGp_array): use hypot

	* bessel_j.c (gsl_sf_bessel_jl_steed_array): use hypot

	* bessel.c (gsl_sf_bessel_Inu_scaled_asymp_unif_e): use hypot

	(gsl_sf_bessel_Knu_scaled_asymp_unif_e): use hypot

2005-03-02  Brian Gough  <bjg@network-theory.co.uk>

	* coulomb_bound.c (gsl_sf_hydrogenicR_e): don't check for

	underflow when function is known to be zero (e.g. r=0 l>0 or at

	zeroes of the laguerre polynomial).

2004-12-29  Brian Gough  <bjg@network-theory.co.uk>

	* dilog.c (gsl_sf_complex_dilog_e): use const consistently in

	arguments of declaration and definition

	(gsl_sf_complex_dilog_xy_e): as above

2004-12-26  Brian Gough  <bjg@network-theory.co.uk>

	* gamma_inc.c (gamma_inc_D): improve error estimate for case of

	u=x/a to include cancellation errors and only use it when x < 0.5*a

	since the cancellation errors are significant for x/a ~ 1

2004-12-23  Brian Gough  <bjg@network-theory.co.uk>

	* gsl_sf_coupling.h: fixed declaration to

	gsl_sf_coupling_6j_INCORRECT instead of

	gsl_sf_coupling_INCORRECT_6j

2004-11-12  Brian Gough  <bjg@network-theory.co.uk>

	* psi.c (gsl_sf_psi_1): added missing function definition

	gsl_sf_psi_1

2004-10-11  Brian Gough  <bjg@network-theory.co.uk>

	* expint.c (gsl_sf_expint_Ei_scaled): fixed call to incorrect

	function gsl_sf_expint_Ei_e

2004-06-03  Brian Gough  <bjg@network-theory.co.uk>

	* psi.c (psi_n_xg0): added missing return type int

2003-11-30  Brian Gough  <bjg@network-theory.co.uk>

	* gsl_sf_log.h: added missing headers for inline functions

2003-08-11  Brian Gough  <bjg@network-theory.co.uk>

	* test_sf.c: added preprocessor definitions TEST_FACTOR and

	TEST_SIGMA to allow larger tolerances on released versions (to

	reduce the number of spurious bug reports).

2003-07-24  Brian Gough  <bjg@network-theory.co.uk>

	* gamma.c (gsl_sf_choose_e): declare k as unsigned int instead of

	int

	(gsl_sf_gamma_e): avoid shadowed declaration of sgn

	* bessel_sequence.c (gsl_sf_bessel_sequence_Jnu_e): declare i as

	size_t instead of int

2003-06-11  Brian Gough  <bjg@network-theory.co.uk>

	* coupling.c (gsl_sf_coupling_6j_INCORRECT_e): fixed typo in test

	for two_jf < 0

	(gsl_sf_coupling_6j_e): moved the implementation from the

	INCORRECT function into the correct function

2003-06-05  Brian Gough  <bjg@network-theory.co.uk>

	* test_sf.c (test_coupling): added some regression tests for

	coupling functions

2003-06-02  Brian Gough  <bjg@network-theory.co.uk>

	* gamma_inc.c (gamma_inc_D): handle x<a, a>10 as a special case

2003-04-18  Brian Gough  <bjg@network-theory.co.uk>

	* gamma.c (gsl_sf_gammainv_e): handle any singularities in

	gamma(x) up front and return zero directly

2003-04-12  Brian Gough  <bjg@network-theory.co.uk>

	* psi.c: changed value of psi(1,1) in table psi_1_table to be

	positive ((-)^2 * 1! * zeta(2)), in accordance with Abramowitz &

	Stegun 6.4.2.

2003-04-08  G. Jungman

    * erfc.c, gsl_sf_erf.h: added gsl_sf_hazard_e() and

    gsl_sf_hazard() functions

2003-04-07  G. Jungman

    * gamma_inc.c, gsl_sf_gamma.h: added gsl_sf_gamma_inc_e() and

    gsl_sf_gamma_inc(), implmentations of the non-normalized incomplete

    gamma function

2003-03-09  Brian Gough  <bjg@network-theory.co.uk>

	* gsl_sf_legendre.h: added missing const to formal parameter on

	gsl_sf_legendre_sphPlm_deriv_array

2003-01-25  Brian Gough  <brian.gough@network-theory.co.uk>

	* test_gamma.c (test_gamma): added a test for

	gamma_inc_P(10,1e-16) (BUG#12)

Sat Sep  7 15:56:15 2002  Brian Gough  <bjg@network-theory.co.uk>

	* test_sf.h (TEST_FACTOR): include an overall factor in the test

 	tolerances, otherwise there are too many bug reports for minor

 	failures.

Sun Sep  1 17:34:27 2002  Brian Gough  <bjg@network-theory.co.uk>

	* test_legendre.c (test_legendre): increased tolerance on test

Sat Jul 13 23:11:37 2002  Brian Gough  <bjg@network-theory.co.uk>

	* ellint.c (gsl_sf_ellint_Kcomp_e): improved error estimate to

 	take cancellation of y=1-k^2 into account near k=1 

Sun Jul  7 21:39:12 2002  Brian Gough  <bjg@network-theory.co.uk>

	* test_bessel.c (test_bessel): increased tolerance on test of

 	bessel_Jn_array

Tue May 28 21:04:29 2002  Brian Gough  <bjg@network-theory.co.uk>

	* psi.c (gsl_sf_psi_1piy_e): changed log(y) to log(ay) since

 	function is even and can be extended to negative y

Mon Jan 28 19:02:42 2002  Brian Gough  <bjg@network-theory.co.uk>

	* gamma_inc.c (gamma_inc_Q_CF): express 'small' constant in terms

 	of GSL_DBL_EPSILON

	(gamma_inc_Q_CF): patch for more reliable series from Hans Plesser

Sat Jan 26 17:33:29 2002  Brian Gough  <bjg@network-theory.co.uk>

	* test_gamma.c (test_gamma): increased tolerance on a test

	* test_hyperg.c (test_hyperg): increased tolerance on a couple of

 	tests

Fri Jan 18 15:12:30 2002  Brian Gough  <bjg@network-theory.co.uk>

	* test_airy.c (test_airy): increased tolerance on test

Tue Jan  8 14:31:04 2002  Brian Gough  <bjg@network-theory.co.uk>

	* test_legendre.c (test_legendre): increased tolerance by one

 	level on _array tests

	* hyperg_1F1.c (hyperg_1F1_small_a_bgt0): fix branch for a==1

Fri Dec  7 12:38:40 2001  Brian Gough  <bjg@network-theory.co.uk>

	* laguerre.c (laguerre_n_cp): catch internal error, not global

 	error

	* error.h (INTERNAL_OVERFLOW_ERROR): added internal error macros

 	which do not call the error handler.

Wed Dec  5 19:25:26 2001  Brian Gough  <bjg@network-theory.co.uk>

	* gamma.c (gamma_xgthalf): return gamma(x) exactly when x is an

 	integer

	* hyperg_1F1.c (hyperg_1F1_ab_posint):  fix bug in recurrence

 	initialisation, as done for hyperg_1F1_ab_pos

Thu Oct 18 11:37:25 2001  Brian Gough  <bjg@network-theory.co.uk>

	* coulomb.c (gsl_sf_coulomb_CL_array): renamed from

 	gsl_sf_coulomb_CL_list for consistency

Sat Oct 13 15:55:56 2001  Brian Gough  <bjg@network-theory.co.uk>

	* cheb_eval.c (cheb_eval_e): keep track of cancellation error to

 	prevent underestimates of total error.

Fri Oct 12 16:39:27 2001  Brian Gough  <bjg@network-theory.co.uk>

	* test_hyperg.c (test_hyperg): increased tolerance from TOL0 to

 	TOL1 on one test to allow it to pass under different optimizations

Thu Oct 11 14:21:34 2001  Brian Gough  <bjg@network-theory.co.uk>

	* gegenbauer.c (gsl_sf_gegenpoly_n_e): initialize variable gk to

 	zero to avoid warning

	* bessel_i.c (gsl_sf_bessel_il_scaled_e): initialize variable

 	iellm1 to zero to avoid warning

	* bessel_Jnu.c (gsl_sf_bessel_Jnu_e): initialize variable Ynp1 to

 	zero to avoid warning

	* legendre_poly.c (gsl_sf_legendre_sphPlm_e): initialize variables

 	p_ell, y_ell to zero to avoid warning

Thu Sep  6 10:38:51 2001  Brian Gough  <bjg@network-theory.co.uk>

	* clausen.c: added missing initialiser to cheb_series struct

	* pow_int.c (gsl_sf_pow_int_e): handle the case x == 0, n < 0

	* legendre_poly.c (gsl_sf_legendre_array_size): added missing

 	static version of this inline function

Wed Aug 15 20:18:43 2001  Brian Gough  <bjg@network-theory.co.uk>

	* hyperg_1F1.c (hyperg_1F1_ab_pos): fix bug in recurrence

 	initialisation

	(hyperg_1F1_ab_neg): if all else fails, try the series.

Wed Aug  8 19:55:34 2001  Brian Gough  <bjg@network-theory.co.uk>

	* test_sf.c (test_coupling): used analytic expressions for the

 	exact values to problems with insufficient precision

Thu Jul 12 15:25:04 2001  Brian Gough  <bjg@network-theory.co.uk>

	* test_legendre.c (test_legendre): corrected tests to use exact

	floating point arguments near singularities.

	* legendre_Qn.c (gsl_sf_legendre_Q0_e): removed unnecessary error

 	terms that I added

	(gsl_sf_legendre_Q1_e): removed unnecessary error terms that I

 	added

Wed Jul 11 16:06:45 2001  Brian Gough  <bjg@network-theory.co.uk>

	* error.h: moved domain, overflow and underflow errors into

 	macros, return Nan for domain error, Inf for overflow.

Tue Jul 10 22:00:55 2001  Brian Gough  <bjg@network-theory.co.uk>

	* erfc.c (gsl_sf_log_erfc_e): introduce an additional log(erfc)

 	branch to prevent loss of accuracy for small x

	* test_sf.c (test_zeta): increased tolerance on zeta function test

 	to take into account differences in rounding

2001-07-10  Brian Gough  <bjg@hppav.local>

	* legendre_Qn.c (gsl_sf_legendre_Q0_e): use (log1p(x) - log1p(-x))

	instead of log((1+x)/(1-x)), for accuracy

	(gsl_sf_legendre_Q1_e): use (log1p(x) - log1p(-x)) instead of

	log((1+x)/(1-x)), for accuracy

	(gsl_sf_legendre_Q0_e): improve error estimate near singular

	points

	(gsl_sf_legendre_Q1_e): improve error estimate near singular

	points

	(gsl_sf_legendre_Q0_e): fixed incorrect factor of 2 in asymptotic

	expansion

	* check.h (CHECK_UNDERFLOW): provide macro for checking results

	for underflow

Fri Jul  6 20:16:19 2001  Brian Gough  <bjg@network-theory.co.uk>

	* zeta.c (riemann_zeta1m_slt0): added a function to compute

 	zeta(1-s) without loss of accuracy for s near zero, as used by the

 	reflection formula.  This fixes a bug in the accuracy of results

 	of zeta(-x) for small x, where loss of precision previously

 	occurred by evaluating 1-x.

2001-07-06  Brian Gough  <bjg@network-theory.co.uk>

	* bessel_I0.c (gsl_sf_bessel_I0_scaled_e): added missing scaling

	factor of exp(-x) for case x<2sqrt(epsilon)

Thu Jul  5 16:16:13 2001  Brian Gough  <bjg@network-theory.co.uk>

	* erfc.c (gsl_sf_erfc_e): corrected case of -10

 	account that chebyshev fit is defined in terms of positive x.

Tue Jul  3 12:09:33 2001  Brian Gough  <bjg@network-theory.co.uk>

	* eval.h (EVAL_DOUBLE): avoid returning the status value as a

 	numerical result

2001-06-28  Brian Gough  <bjg@network-theory.co.uk>

	* elementary.c (gsl_sf_multiply_e): use GSL_COERCE_DBL macro to

	deal with extended register problem

	* coupling.c: include stdlib.h for prototype of abs()

Wed Jun 27 21:20:22 2001  Brian Gough  <bjg@network-theory.co.uk>

	* test_sf.c (test_fermidirac): increased tolerance on test of

 	gsl_sf_fermi_dirac_int(9,500) by factor of two to allow for MSVC

Tue Jun 26 12:10:17 2001  Brian Gough  <bjg@network-theory.co.uk>

	* gsl_sf_gamma.h: added const to prototype, to match function

 	definition

Sun May  6 13:01:01 2001  Brian Gough  <bjg@network-theory.co.uk>

	* test_sf.c: initialize message_buff to null string to prevent

 	random segmentation faults

	* test_sf.h: simplified test macros to reduce stack usage

Mon Apr 30 12:38:15 2001  Brian Gough  <bjg@network-theory.co.uk>

	* airy_zero.c bessel_zero.c: zeros are now counted using an

 	unsigned int

	* poly.c: moved poly_eval function into poly/ directory

Wed Apr 25 17:28:48 2001  Brian Gough  <bjg@network-theory.co.uk>

	* trig.c (gsl_sf_polar_to_rect): dropped _e from polar/rect

 	conversion functions

Tue Apr 24 17:05:22 2001  Brian Gough  <bjg@network-theory.co.uk>

	* split out chebyshev functions into their own cheb/ directory,

 	leaving behind an internal cheb struct and cheb evaluation

 	function.

Mon Apr 23 10:21:06 2001  Brian Gough  <bjg@network-theory.co.uk>

	* changed tests for EFAULT into a commented-out macro, since

 	EFAULT should only apply to invalid non-null pointers, but it

 	might be useful to keep the tests around for debugging in this

 	directory.

	* unified error handling conventions to _e for error handling

 	functions and no suffix for plain functions, so _impl functions

 	are no longer needed.

1999-08-02  Mark Galassi  <rosalia@lanl.gov>

	* fermi_dirac.c: took out the use of some "const int" constants

	which were being used to size arrays, since this is not portable

	(thanks to Bernd Petrovitsch <bernd@ict.tuwien.ac.at> for pointing

	this out).

1999-01-02  Mark Galassi  <rosalia@cygnus.com>

	* trig.c (gsl_sf_rect_to_polar_impl): introduced an #ifdef

	HAVE_HYPOT for systems that do not have the hypot() routine.

Sun Feb 14 20:59:50 1999  Brian Gough  <bjg@netsci.freeserve.co.uk>

	* Makefile.am (include_HEADERS): added gsl_sf_result.h

1998-12-05  Mark Galassi  <rosalia@cygnus.com>

	* Makefile.am: t-erf.c was commented out, which is a problem

	(pointed out by automake-1.3b).  Moved the commented-out t-erf.c

	line to the end of the long list of files.

	* legendre_poly.c (gsl_sf_legendre_Pl_impl):

	* legendre_con.c (gsl_sf_conicalP_xgt1_neg_mu_largetau_impl):

	added an ugly fix to the invocation of gsl_sf_bessel_Jn_impl() to

	give it two bogus arguments so that it builds.  I hope Jerry fixes

	it up soon!

	* gsl_sf_airy.h: added an include of gsl_precision.h so that

	gsl_prec_t is defined.

Tue Nov 17 17:29:31 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* added #include <config.h> to all top-level source files

	* chebyshev.c (gsl_sf_cheb_eval_n): changed local gslMIN and

 	gslMAX to the standard GSL_MIN and GSL_MAX macros

Tue Aug 18 13:36:15 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* coulomb.c (C0sq): changed to using gsl_sf_expm1_impl instead of

 	expm1, since the latter is a GNU extension.

Mon Aug  3 14:43:16 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* bessel_amp_phase.h: undefined consts are now declared extern

Mon Jul 13 12:40:27 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* gamma.c: changed all the factorial functions to take unsigned

 	int arguments, since negative values are not allowed.

	(gsl_sf_choose): fixed off by one error in call to factorial.

Sun Jul 12 13:21:41 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* gsl_sf_legendre.h, gsl_sf_poly.h, gsl_sf_pow_int.h: added

 	HAVE_INLINE around some inline definitions in the header files

	* gamma.c: implemented the natural versions of gsl_sf_lnchoose and

	gsl_sf_choose.

Wed Apr 15 21:27:48 1998  Brian Gough  <bjg@vvv.lanl.gov>

	* added the return code as the second argument of GSL_WARNING, so

 	we can filter on the error number or output a standard message for

 	each one.

1998-02-02  Mark Galassi  <rosalia@cygnus.com>

	* Makefile.am (include_HEADERS): added gsl_specfunc.h so it gets

	into the distribution.

	(INCLUDES): added -I$(top_srcdir) so that bessel.c can find

	gsl_math.h