Changes in version 0.6-1 (2025-10-13)                  

NEW FEATURES

    o   pnchisq_ss() gets new optional argument ssr = ss(....)  for
	flexibility.

    o   New gam1() calls mpfr-ized direct code or the double-only C
	code gam1d()

    o   Add Abramowitz & Stegun, p.945, pbeta() approximations 26.5.20
	and 26.5.21 as pbetaAS_eq20() and pbetaAS_eq21().  For
	“didactical” reasons, also provide the simple moment matching
	normal approximation pbetaNorm2().  For now, also low-level
	versions .pbeta.eq20(), ..., .pbetaSeq21().

BUG FIXES

    o   ldexp({}, 1) or pow({}, 2) no longer segfault.

    o   pbetaRv1() for verbose = TRUE and/or mpfr-number arguments.

    o   dnbinomR() and dnbinom.mu(): fix vectorization logic, using
	ifelse().

    o   pbetaRv1() for verbose = TRUE and/or mpfr-number arguments.

    o   dnbinomR() and dnbinom.mu(): fix vectorization logic, using
	ifelse().

                 Changes in version 0.6-0 (2025-07-08)                  

NEW FEATURES

    o   added functions dltgammaInc() and lgammaP11(), both from TOMS
	Algorithm 1006 by Rémy Abergel and Lionel Moisan (2020).

    o   C-level logcf() now uses long double, i.e., R's LDOUBLE, to
	possibly improve for smallish eps.
	
	logcf() also got an optional maxit argument. For the pure-R
	version, renamed the simplistic and slow one from logcfR() to
	logcfR_vec() and the smarter faster one from logcfR.() to
	logcfR().  They have given identical results in test cases, and
	logcfR() should really be the one used practically.

    o   bd0C() has new version = "R_2025_0510" where as bd0() has just
	been updated to the latest version also in R-devel svn
	revisions 88190,1,3 and 88208.
	
	New bd0_p1l1(), also as version in dpois_raw(); bd0(), and the
	now four bd0_<meth>() functions work with x=Inf and properly
	vectorize.

    o   The log1pmx vignette now shows that for dpois(x, 48) at least,
	using the two parts of ebd0() gives clearly the most accurate
	values (used in R since ca. 2022).
	
	On the other hand, new evaluations at the end of
	tests/bd0-tst.R indicate that increasing the (only in DPQ)
	optional cutoff from delta = 0.1 to about delta = 0.4 gets much
	more accurate bd0(x, np) values, notably when |x - np| is
	smallish for relatively large x and np where ebd0() is clearly
	less accurate.

BUG FIXES

    o   rchk PROTECT fix in bd0.c.

    o   Strictly _not_ comparing Rboolean with NA_LOGICAL.

    o   log1pmx(Inf) is correct (= -Inf) now.

    o   dpois_raw(1, {}) now works (returning numeric(0)).

MISC

    o   tests/pow-tst.R also work on "bizarre" platforms where
	osVersion is NULL.

    o   tests/stirlerr-tst.R uses many more if(doExtras) ... to run
	faster without “extras”.

                 Changes in version 0.5-9 (2024-08-23)                  

NEW FEATURES

    o   dbinom_raw() gets a new more accurate version = "R4.4", using
	the new pow1p() function, from R-devel, i.e., R 4.4.0 and
	newer.

    o   stirlerr_simpl() gets argument version, a string specifying the
	exact function to be used, currently with options ("R3",
	"lgamma1p", "MM2", "n0").  The first being the direct formula
	hardwired in R up to versions 4.3.z, "lgamma1p" is to use
	lgamma1p(n) instead of lgamma(n+1), whereas "MM2" is a version
	using lgamma(n) and slightly more accurate for n \in [1, 5],
	approximately.
	
	stirlerr() correspondingly gets argument direct.ver to be
	passed to stirlerr_simpl(., version=*) in the “direct” formula
	case, when n < cutoffs[1].

    o   stirlerr() gets argument order allowing currently up to order
	20 terms (constants S0..S19) in Stirling's series (and _not_
	using cutoffs or direct formula).

    o   New stirlerrC() interfacing to C code which should
	“parallelize” R's own C level stirlerr().
	
	More documentation and tests; experiments added in
	../Misc/stirlerr-trms.R; nice explorations to
	../tests/stirlerr-tst.R.

    o   New functions pow(x,y) and pow_di(x,y) computing x^y as in R's
	C API Rmathlib.

    o   dgamma.R(x, *) gets new optional argument _dpois_r_args_.

    o   New function expm1x() providing exp(x) - 1 - x numerically
	accurately also for small |x|.

    o   New experimental Ixpq() implementing Gil et al.(2023)'s direct
	incomplete Beta function, i.e., pbeta() version.  *TODO*: make
	it work with mpfr() !

    o   New function pntGST23_T6() and pntGST23_1() implementing Gil et
	al.(2023)'s first asymptotic formula (Theorem 6), and based on
	Temme's Maple code for their direct computation (1).
	
	pnt*(): CDF of the non-central t-distribution.

    o   New function pntVW13() implementing Viktor Witkovský(2013)'s
	algorithm, from his published Matlab code.

    o   Data set pt_Witkovsky_Tab1 containing (the free software subset
	of) Viktor Witkovský(2013)'s _Table 1_ of pt(x, df=nu,
	ncp=delta) examples; updated with octave results for both the
	original 2013 matlab code and the 2022 corrected one.

    o   lgamma1p_series(x, k), the Taylor series approximations for
	lgamma1p(x) := \log Gamma(1+x) are now available up to k = 15,
	also for mpfr() numbers x.

    o   New gam1d() and gamln1() providing more TOMS 708 auxiliary
	routines.

BUG FIXES

    o   _fix_ format fixes for when 'long != long long'.

    o   dgamma.R(x, ..) now _does_ vectorize in x, as documented.

    o   minor adjustment of pntR()'s default use.pnorm.

    o   src/bd0.c needed one more PROTECT().

Misc

    o   new tests/dt-ex.R exploring dt() accuracy issues, mostly
	historical, but newly for df=nu << 1.

                 Changes in version 0.5-8 (2023-12-01)                  

BUG FIXES

    o   ebd0() and C level ebd0C() now work with _long_ vectors and
	hence return data.frame(yh = *, yl = *) instead of 2-row
	matrix.

    o   format fixes: "%lld" and (long long) notably for R_xlen_t.

                 Changes in version 0.5-7 (2023-11-03)                  

Misc

    o   fix "lost braces"

    o   replace UTF-8 greek glyps by plotmath

                 Changes in version 0.5-6 (2023-09-07)                  

NEW FEATURES

    o   phyperR() gets lower.tail=TRUE, log.p=FALSE arguments.

BUG FIXES

    o   Correctly vectorize phyperR().

Misc

    o   Documentation (and internal comment) updates for qbinomR() and
	qnbinomR(), notably fixing "empty" argument doc.

    o   Deprecated phyperBinMolenaar(); it has been synonym to
	phyperBinMolenaar.1().

    o   Fix LaTeX warnings/errors by removing amazingly many \cr in
	man/*.Rd files.

                 Changes in version 0.5-5 (2023-06-23)                  

BUG FIXES

    o   qbinomR(p, *), qnbinomR(p, *), and qpoisR(p, *): Fix
	R_Q_P01_boundaries(p, ..) for trivial boundaries, e.g., p \in
	\{0,1\} for log.p=FALSE.

Misc

    o   A new flang Fortran compiler does not know outdated derfc(),
	but needs current standard erfc().

                 Changes in version 0.5-4 (2023-04-14)                  

NEW FEATURES

    o   New R function qnormCappr().

    o   New R function rexpm1() from TOMS 708.

    o   Provide also qnormR(*, version = "1.0.x") and "1.0_noN", for
	history's sake.

    o   New R function rlog1() from TOMS 708.

BUG FIXES

    o   qnormUappr() & qnormUappr6() were _wrongly_ negative for lp in
	[-.693147, 0) (corresponding to p > 1/2).

Misc

    o   Tweaks to the qnorm-asymp vignette, related to the JSS
	submission.

    o   tests/pqnorm_extreme.R plots more about the pnormAsymp()
	relative errors.

    o   Apple clang 14.0.3 needs looser tolerance for logcf{R}() and
	dbinom_raw() comparisons.

    o   Minimally document the _internal_ .D_*() and .DT_*() utilities,
	as we provide them to experts and e.g., package DPQmpfr.

                 Changes in version 0.5-3 (2022-12-01)                  

NEW FEATURES

    o   New R functions qntR() and qtU(), Vectorize()d versions of
	qntR1() and qtU1(), respectively.

    o   New qtR() <- Vectorize(qntR1, *); pure R implementation of R's
	Mathlib C-level qt(), but additionally allowing tweaks, used
	for fixing R's bug PR#18630.  Added (optional) log-scale Newton
	steps via logNewton = log.p, needing more tests.

    o   New qtNappr() - from the remark in R's nmath/qt.c about very
	large df approximation, now with all 4 large-df terms from
	Abramowitz & Stegun's (26.7.5).

    o   New gammaVer() to exemplify R's (partly historical) versions of
	gamma() implementations.

    o   New qnormUappr6(), providing the ‘6 coefficients’-approximation
	of A. & S. to qnorm().

    o   qnormR() gets new version = "2022-08-04" which uses MM's newly
	derived asymptotic approximations to qnorm().

    o   New qnormAsymp() providing new asymptotic approximations to
	qnorm().

    o   Added vignette “Asymptotic Tail Formulas for Gaussian
	Quantiles”.

    o   New chebychevPoly() (and auxiliaries) for evaluation of
	Chebyshev polynomials.

                 Changes in version 0.5-2 (2022-06-10)                  

NEW FEATURES

    o   qbetaAppr() (and the qbetaAppr.{1,3}() auxiliaries) now get a
	log.p argument.

    o   all(?) qbeta.Appr*() functions now accept a lower.tail argument
	(with default TRUE).

    o   New R function dpsifn(), as interface to R's C API dpsifn(),
	the workhorse of all R's psigamma() functions.

    o   dpois_raw(x, lambda, *) gets new cutoff small.x__lambda to
	_not_ use bd0() or ebd0() when x is much smaller than lambda.

    o   Simple functions, mostly for didactical and comparison
	purposes, dpois_simpl(), dpois_simpl0() and stirlerr_simpl(),
	all of which use the “simple” direct formula which leads to
	numerical cancellation typically.

    o   New R function bpser() interfacing to the C function of the
	same in R's Mathlib pbeta() or bratio() in file
	nmath/toms708.c.

BUG FIXES

    o   Updated URL of Abramowitz & Stegun.

    o   Fixed \ escape in one man/*.Rd

Misc

    o   Our tests/*.R no longer rely on the Matrix test-tools.R
	collection but rather use our own.

    o   In tests/chisq-nonc-ex.R (and DESCRIPTION), we've replaced
	akima by interp as the latter has a FOSS licence.

                 Changes in version 0.5-1 (2021-12-13)                  

NEW FEATURES

    o   bd0(x, np, delta, ..) now uses if (|x-np| <= delta * (x+np))
	(less-equal instead of strictly less), and hence setting delta
	= 0 is now allowed, using the series expansion only for x ==
	np, useful e.g., for the case of highly accurate mpfr-numbers.
	
	Similarly, in log1pmx(x, .., eps2, ..), eps2 = 0 is now
	allowed.

BUG FIXES

    o   consistency between ebd0() and ebd0C(): multiplication * e
	coming last ("bug" did not show ..).

    o   okLongDouble() should no longer fail on M1 mac.

    o   ebd0C() now checking |yl| < 5.5 before returning; was
	horrendously wrong on Windows compiled with -mnative as long as
	it included (the default) -mfma (FMA:= Fused-Multiply-Add).

    o   fix rchk issue: "need" PROTECT(.) while calling allocating
	lgamma1p().

                 Changes in version 0.5-0 (2021-11-24)                  

NEW FEATURES

    o   new ebd0C() interface to C version; helps to fix lapsus in pure
	R version ebd0():
	
	ebd0(x, M) for large M now checks for overflow in M/x, and
	notably the case f * 2^-(e+10) =: fg == Inf.

    o   In bd0() work around underflow of (x-np)/(x+np).

    o   new R functions frexp() and ldexp() for getting and setting
	base-2 representations of numbers, and new R function modf() to
	split number into integer and fractional part, all interfacing
	the C99 (math lib) standard functions of the same name.

    o   logcfR.(x) has been “vectorized” in x even though it's
	iterative with different number of iterations for each x[i],
	and is hence considerably faster when x is an "mpfr" vector.
	
	TODO: Consider renaming the two logcfR versions; at least
	neither is deprecated!

BUG FIXES

    o   fixed embarrassing typo (two i's) bug in logcfR()

    o   stirlerr(n) now also works when is.integer(n) and n is large
	enough for n * n to overflow (to NA).

    o   fix typo/thinko in dnbinomR()

    o   ditto in logcfR(*, trace=TRUE) iteration report.

    o   dpois_raw(x, *) now works up to maximal x, now preventing
	overflow in previous 2*pi*x computation.

    o   dnbinomR() and dnbinom.mu() fix for x > size when log=TRUE,
	notably for x >> size.

                 Changes in version 0.4-4 (2021-05-27)                  

NEW FEATURES

    o   new bd0_*() versions of bd0(), based mainly on log1pmx().

REFACTORING

    o   To be more modular, our test-tools.R no longer source() those
	of Matrix.

BUG FIXES

    o   src/bd0.c gave warning with some compiler settings with
	-Wself-assign.

    o   test-tools.R readRDS_() thinko fixed.

                 Changes in version 0.4-3 (2021-05-05)                  

NEW FEATURES

    o   Provide R functions log1mexpC(), log1pexpC(), log1pmxC(), and
	lgamma1pC(), all interfacing to R's C API (Rmath.h), aka
	‘Rmathlib’.

    o   New pnormAsymp() for asymptotic (typically upper tail, log
	scale) approximation of pnorm().  With Rmpfr, we can see how
	accurate these are in detail.

    o   New dnbinomR() finding better code for R itself, dnbinom.mu()
	and dbinom_raw(); also new utility functions bd0() and
	stirlerr(), both vectorized, and also working with
	"mpfr"-numbers, such that dpois_raw() now does so, too.
	
	Additionally, an “extended” version of bd0() called ebd0() in
	pure R, where a C version was proposed by Morten Welinder in
	R's bugzilla, PR#15628.
	
	Experimentally, also provide p1l1() and its Taylor series
	approximations p1l1ser() which could be employed for bd0() /
	ebd0() instead of the current algorithms.

    o   Several C level utilities to be .Call()ed from R, from R's
	mathlib, ldexp() and frexp() even from C math.

    o   Pure R implementations (somewhat experimental) of corresponding
	R Mathlib C functions: qbinomR(), qnbinomR(), and qpoisR() each
	with several tuning parameters for the underlying algorithm,
	notably the root-finding parts.

    o   newly, logcf() now based on C code, perfectly vectorizes; the
	pure R version, now called logcfR(x, *) currently still runs
	the iterations simultaneously for all `x[i]` and hence
	convergence and rescaling happen by “group decision”,
	efficiently but undesirable for strict comparisons.
	
	logcfR(x, *) and log1pmx(x, *) now also work for "mpfr"-numbers
	x, and log1pmx() gets optional eps2 = 1e-2 and minL1 = -0.791
	arguments, the defaults of which _may change_, as I think at
	least the latter to not be perfect.

    o   Now lb_chiAsymp(nu, order) works up to order 8.

    o   Provide the first parts of a new vignette
	(../vignettes/log1pmx-etc.Rnw) on log1pmx(), bd0(), and
	stirlerr(), which should contain part of Loader(2000)'s report
	and new findings of improved bd0() and stirlerr() computations.

                 Changes in version 0.4-2 (2020-11-09)                  

NEW FEATURES

    o   New functions pnormL*() and pnormU*() for (mathematically
	proven) lower and upper bound to pnorm(), notably also for
	investigation with log.p=TRUE.

    o   qnormR(), implementing current R's qnorm() in pure R, notably
	with trace and version options.

                 Changes in version 0.4-1 (2020-06-19)                  

TESTING

    o   Reorganize tests; notably to become less platform dependent.

                        Changes in version 0.4-0                        

NEW FEATURES

    o   Many new phyper*() functions and helpers for them, such as
	Bernoulli numbers Bern() and asymptotic lgammaAsymp().

    o   Notably phyperR2() which is a pure R version of R's own (C code
	based) phyper().

                 Changes in version 0.3-5 (2019-10-21)                  

NEW FEATURES

    o   pnbetaAS310() gained a useAS226 option.

    o   New okLongDouble() function, notably for detecting that with a
	valgrinded version of R, long double C arithmetic seems to
	silently switch to (simple) double precision.

BUG FIXES

    o   long double printing from C now should happen correctly on all
	platforms including 32- and 64-bit Windows.

                        Changes in version 0.3-4                        

NEW FEATURES

    o   more efficient (internally vectorized) dntJKBf().
	Consequently, dntJKBf1() is deprecated now.

    o   pntR() (and pntR1()) get new option use.pnorm (the default of
	which had been hard coded previously).

BUG FIXES

    o   fix thinko in any_mpfr() and all_mpfr().

    o   pnchisqRC()'s C code gave severe valgrind warnings; fixed
	printing of long double etc; also added special MinGW deal in
	Windows.

    o   ppoisD() behaves differently in a valgrinded version; for now,
	reproducible only when using valgrind on non-valgrinded
	installed package.

                 Changes in version 0.3-3 (2019-10-02)                  

NEW FEATURES

    o   Renamed (and generalized / merged) many functions to have less
	"." in names.

    o   New pnbetaAS310() function interfacing to my _corrected_ C
	version of 'ASA 310' (2007).

    o   New algdiv() function interfacing to the 'TOMS 708' version of
	our logQab().

    o   New pl2curves() which generalizes (somewhat) previous function
	p.dnchiB().

    o   Made newton() more flexible with new xMin and xMax arguments
	which are notably useful for q*() (quantile function)
	computations.
	
	Correspondingly replaced previous qchisq2() and qchisqA() with
	new qchisqN() (‘N’ewton).

    o   new pnchisqRC(), a version of R's C level non-central
	chi-squared, with additional options.

    o   “new” logspace.add() and logspace.sub().

    o   “new” pnchisqT93() (plus its two auxiliaries), implementing
	Temme(1993)'s approximations.

    o   “new” pnchisqBolKuz() and qnchisqBolKuz() implementing Bol'shev
	and Kuznetzov (1963)'s approximations.

    o   “new” pnchi1sq() and pnchi2sq() with “exact” formulas for the
	special cases df=1 and df=3.

    o   simplified formula in dtWV().

BUG FIXES

    o   qnchisqPearson(pp, df=DF, ncp=100) and qnchisqSankaran_d(*) no
	longer return NaN for very large DF = 1e200.

    o   pnchisq() now also has default verbose = 0 as all other such
	functions with verbose (or trace[.lev] optional argument.

                        Changes in version 0.3-0                        

NEW FEATURES

    o   Move many of my up to 15 years old DPQ computation utilities
	into a package, to become public, “bloggable”, etc.