pari-2.13 (STABLE) released

[Karim Belabas <Karim.Belabas@math.u-bordeaux.fr>]

Dear PARI lovers,

I am pleased to announce the release of pari-2.13.0 (STABLE) !
The sources and binary distributions for macOS and Windows are available at
the usual address

  http://pari.math.u-bordeaux.fr/download.html

This is a major STABLE release ending a development cycle which started in
July 2018. For those still using pari-2.11.4 or earlier versions, it is
time to upgrade.

Have fun !

    K.B.

PARI-2.13.0 HIGHLIGHTS vs PARI-2.11.4: see the end for COMPATIBILITY ISSUES.
=====================================

[The GP language and the gp calculator]
  - support call by reference in GP functions: f(~L,x) = L[1] = x;
    This allows to modify the contents of a vector / list argument and avoids
    unnecessary copies of arguments.
    For clarity, built-in functions handling lists/maps should now use the ~
    prefix to indicate mutable containers, e.g. listput(~L,x), mapput(~M,x)
  - lexical variables names can now reuse builtin functions names as in
      f(theta) = my(gamma=1/3);
      sum(omega=1,10, exp(2*I*Pi*gamma*theta*omega);
  - GP closures improvement: no longer copy the full context, only the
    variables actually accessed by the function.
  - allow holes in multi-assignements: [a,,c]=[1,2,3].
  - new parallel GP function parplothexport, parforprimestep, parforeach.
  - new GP functions export, unexport, exportall, unexportall simplify
    writing code for parallel execution (allowing simple access to "global"
    variable in parallel threads)
  - new GP functions getlocalprec, getlocalbitprec: to replace obscure
    idiomatic constructs such as bitprecision(1.)
  - new GP functions to simplify string handling: strsplit (split into a
    vector of strings with respect to a given separator), strjoin (join a
    vector of strings using a given separator), strtime (format a 'time')
  - new GP function arity, returns the number of arguments of a function
  - improve extended help: ??? output now include double quotes when needed
    (e.g. ???conductor); support all symbolic operators using the _
    placeholder syntax: ?? _+_, ?? !_, ?? _+=_, ?? _++, etc.
  - natural extensions to the syntax of t_LISTS: allow L[a..b] and L[^i];
    allow listinsert(L, x, n) for n > #L, as in listput
  - allow forsquarefree(n = a, b) with a*b < 0
  - allow subst(e, x, vector v) for vector(subst(e, x, v[i]))
  - new GP function matreduce, to convert a multiset to a vector of distinct
    elements together with their multiplicities:
    ? matreduce([x,y,z,x,x,y])
    %2 =
    [x 3]

    [y 2]

    [z 1]

    Can also be used to merge factorizations:
    F1 = factor(12);
    F2 = factor(20);
    matreduce(matconcat([F1,F2]~)) -> factor(12 * 20)
  - support plotinit/plotexport even when no graphic engine is available
    (Configure --graphic=none).

[Number Fields, Local Fields, Finite Fields]
  - Number fields functions now allow compact representations of algebraic
    numbers, as factorization matrices: products of "small" elements possibly
    raised to huge powers. This is very useful for units and S-units or
    principal ideals generators, which may be huge in expanded form. Functions
    transparently process such objects but, for compatibility reasons, they are
    not created by default, this must be specified as follows:
    * bnf = bnfinit(pol, 1) now constructs an algebraic bnf structure
      containing all data (including units) in exact compact form.
      This is the critical step. Class field theoretic functions involving bnf
      will process units in compact form, allowing them to succeed
      independently of their size or the working precision.
    * new GP function bnfunits(bnf, optional S) returns the fundamental
      S-units in compact form [ bnf.fu returns the units in expanded form
      which can be expensive ].
    * bnfisprincipal(bnf,id,4) or bnrisprincipal(bnr,id,4) return
      the principal part of the discrete logarithm in compact form.
    Without optional flags, nothing changes: old programs should run
    without modification and fail when units are large. New programs should
    use the new flags.

  - The old bnrinit(, 1) flag [ include explicit bnr generators in the output ]
    was mandatory in many class field theoretic functions. Not anymore: their
    internal compact representation is always sufficient. You can still use
    them for debugging but they are expensive and no longer needed elsewhere.

  - The bnrinit function computes general ray class groups G = Cl_f(K) but we
    are often interested in simpler quotients G/H, e.g., G/G^N for abelian
    extensions of exponent N. The function bnrinit now accepts an optional
    argument to specify N; later calls to bnrisprincipal solve the discrete
    logarithm in this quotient group and are now only correct modulo N, which
    is adequate if we actually want to work in G/H for any congruence subgroup
    of index dividing N. This trick avoids expensive discrete logarithms if the
    modulus f is large. This is used intensively in PARI internals, speeding
    up all class field theoretic functions.

    The rnfconductor function, which computes a pair (f,H) attached to an
    abelian extension K[X]/(T) of the base field K together with a bnr
    structure for G = Cl_f(K), now accepts a new flag to return G / G^deg(T),
    which is less expensive and usually sufficient.

    The same trick is available in idealstar for the group G = (Z_K/f)^*,
    allowing to compute in G/G^N; ideallog then returns results which are
    correct modulo N.

  - new GP function bnrclassfield, generalizing rnfkummer (which only
    supported relative extensions of prime degree)
  - new GP function idealdown(nf,A) to return A \cap Q
  - new GP function idealismaximal
  - new GP function bnrmap to map ray class groups Cl_A -> Cl_B when B | A.
    This also works with ray class groups modulo N provided the exponent of
    Cl_A divides N

  - nfsubfields: add van Hoeij-Klueners-Novocin algorithm to the existing
    algorithms of Klueners (doesn't run in polynomial time) and Allombert
    (not always applicable). The running time is polynomial in the output size.
    New GP functions nfsubfieldsmax (maximal subfields) and nfsubfieldscm
    (maximal CM subfield), both of which always run in polynomial time
    because their output is provably small.

  - new GP member function nf.p returning rational ramified primes;
    rnf.p returns rational primes dividing the norm of the relative
    discriminant ideal.

  - allow nfinit([T, Z_K basis, ramified primes]) when all arithmetic
    invariants are known.

  - galoisinit: the definition of "weakly super-solvable" groups, supported
    by the function, is less restrictive. For instance, there are no
    exceptions in order < 48 and 23 exceptions in order < 144, 40% of them in
    order 96. There were 29 such exceptions in 2.12.0.

  - nfbasis now accepts an optional argument &dK to recover the field
    discriminant
  - rnfinit(nf, [T,B]) for orders that are maximal at a certain set of primes
    only; allow B a list of primes and maximal ideals (same for rnfbasis,
    rnfdisc, rnfconductor, rnfpseudobasis)
  - new GP function nfdiscfactors to return the field discriminant in
    factored form
  - new GP function polteichmuller to define nice models of p-adic extensions
    (see ??polteichmuller)
  - new GP function ffmaprel(m, t) to express the finite field element t as
    an algebraic element over the codomain of the map m; this allows for
    instance to compute the relative minimal polynomial, trace and norm of t


[Transcendental functions]
  - new GP function hypergeom() for general hypergeometric functions;
    airy() for Airy Ai and Bi functions. As a consequence, hyperu(), the
    U-confluent hypergeometric function is now implemented for arbitrary
    complex parameters.
  - new GP functions ellK and ellE (complete elliptic integrals of the first
    and second kind)
  - new GP function dirpowers(n,x) = [1^x, ..., n^x]
  - faster exp(small rational number) using binary splitting. E.g, at 10^6
    decimal digits of accuracy, exp(1/3) is now computed in 0.2s, about 50
    times faster than in pari-2.11.
  - faster formulas for Catalan() and log(2)
  - x^s: faster and more precise when s is a t_COMPLEX
  - improve gamma-related functions:
    . n!, factorial(n), lngamma(n), psi(n) for integral n
    . gamma and lngamma(t_FRAC) use direct formula when denom | 24
    . lngamma and psi near 1 [ cache values of zeta(odd integers) ]
  - fast algorithm for zeta([a + b*n | n<-[0..N]]). Compare:
    ? \p2000
    ? zeta(vector(1000,i,i+1));
    time = 787 ms.
    ? vector(1000,i,zeta(1+i));
    time = 3,334 ms.
    Also fix tuning problems for zeta(small even integer), slow at huge
    accuracy.
  - add optional argument to sqrtint to obtain the remainder as well
  - gammamellininvinit(A): allow A to be an L-function; this also allowed
    in gammamellininv and gammamellininvasymp.
  - besseli, besselj, besselk, besseln(nu, z), eint1(z): now use asymptotic
    expansion for large z
  - fix cancellation in acosh / asinh for huge t_COMPLEX and in cos(z) and
    sin(z) when Im(z) is nonzero but tiny
  - improved zetahurwitz: faster & use less memory.

[Multiple Zeta Values]
  - new GP function polylogmult for multiple polylogarithm values and their
    Yamamoto interpolation.
  - new GP function zetamultdual: zetamultdual([4]) -> [2,1,1]
    (the corresponding Muliple Zeta Value is the same).
  - zetamult now accepts an optional parameter t for Yamamoto's
    interpolation; in particular for t = 1, we obtain star MZVs and for
    t = 'x, Yamamoto's polynomial.
  - zetamultall: add optional flag to compute only values of given weight,
    and/or up to duality, and/or star values.

[Numerical summation and integration]
  - limitnum/asympnum: allow closures of the form N->[u(1),...,u(N)],
    which allows to handle efficiently sums, recursions, continued
    fractions, etc.
  - new GP function asympnumraw: extrapolate the first terms of the asymptotic
    expansion of some expression; more robust and versatile than asympnum
    (which only supports rational expansions). E.g.,
      f(n) = n! / (sqrt(n) * (n/exp(1))^n);
      \p100
      v = asympnumraw(f, 10);
      bestappr(v/v[1])
    returns 10 terms of Stirling's expansion without needing to identify
    v[1] = sqrt(2*Pi); asympnum would fail because of the v[1] 'period'.
  - new GP function derivn for (formal or numerical) n-th derivative
  - faster derivnum; f'...' (n quotes) now always uses derivnum which is
    about 2^n/n faster, and more accurate.
  - fix numerical derivative of functions with optional argument: e.g.,
    f(x, y = 1) = x*y;
    f'(1) used to return 0 [ instead of 1 ].
  - improve intnumgaussinit (use direct Newton instead of calling polrootsreal)

[Elementary Number Theory]
  - improvements to the MPQS factoring algorithm: better sieve, faster
    linear algebra (block Lanczos), better tunings, no longer write temporary
    files to disk
  - rewrite the Bernoulli number cache using zeta(2n) = (*) B_{2n}
    and power recycling for zeta(2n) + additive Clausen-von Staudt:
    simpler and orders of magnitude faster than the old iteration. We now
    always store Bernoulli numbers in rational form. The GP function
    bernvec() is no longer obsolete (now faster than computing Bernoulli
    numbers individually).
  - qfbsolve(Q,n) now supports arbitrary integer n (was limited to primes)
    and return all solutions modulo units.
  - new GP function eulerianpol (eulerian polynomials), eulerpol (Euler
    polynomials), eulerfrac (Euler numbers), eulervec ([E_0, ..., E_2n]).
  - new GP function halfgcd: given polynomials or integers [x,y], compute an
    invertible matrix M such that M * [x,y]~ = [a,b]~ with b small (half the
    "size" of [x,y]).
  - qfbsolve: new optional flag to ask for all solutions modulo units (vs. a
    single solution) or to allow non-primitive solutions.
  - randomprime: new optional argument to specify a congruence class.
    E.g. randomprime([2^31, 2^32-1], 7) -> a 32-bit prime congruent to 1 mod 7.
         randomprime([a, b], Mod(2,3)) -> a prime = 2 mod 3 in [a,b]

[Polynomials and Power series]
  - new GP function rootsof1(N): complex roots of unity of order N.
  - new GP functions fft, fftinv: Fast Fourier Transform of order 2^n and its
    inverse, given roots of unity. Supports general base rings,
    e.g., finite rings.
  - new GP function pollaguerre for Laguerre polynomials. Add an
    optional flag to pollegendre and polhermite to return P_{n-1}(a) and P_n(a)
  - polinterpolate over Z/pZ is now asymptotically fast
  - polrootsreal: faster due to asymptotically fast P(X) -> P(X+a) algorithm
  - power series and truncated polynomials: systematic use of Newton division
  - polrootspadic: allow computing roots in unramified extensions of Q_p
    E.g. T = y^2+y+1; p = 2;
         polrootspadic(x^3-x^2+64*y, [T,p], 5);

[Linear Algebra]
  - generic fast linear algebra using CUP decomposition
  - improve linear systems over Q [treat contents sensibly]
  - fix inefficiency in LLL (not present in original fplll code)
  - matsnf(integral matrix): now uses a modular algorithm.
  - matrixqz(,-1 or -2): faster implementation.
  - matrank, matker over the integers: faster implementation for large
    rectangular matrices.
  - qflll(x, 3) in place reduction: return x * qflll(x) [faster].

[Curves]
  - hyperellpadicfrobenius: allow X defined over a number field and
    compute Frobenius over unramified extension of Q_p. E.g.,
      M = hyperellpadicfrobenius(x^5 + 'a  * x + 1, ['a^2+1,3], 10);
    (over Q_9, to 10 3-adic digits of accuracy)
  - new GP function ellpadiclambdamu: Iwasawa invariants lambda and mu for
    the p-adic L function L_p(E), where p is a prime of good or bad
    multiplicative reduction for the elliptic curve E/Q.
  - make ellheight(E) return the Faltings height of E (over a number field)
    make ellheight(E, torsion point) return an exact 0
  - support for non-integral models in ellratpoints and hyperellratpoints

[L-functions]
  - lfuncreate(v) can now correctly handle non-rational data: instead of
    v = [a, asta, Vga, k, N, eps, poles], v can now be a closure (without
    arguments) evaluating to such a vector at current working precision.
  - lfuncreate([G,v]): where G is (Z/NZ)^* or a ray class group and v a
    vector of characters with the same conductor. This vector-valued L-function
    allows fast simultaneous evaluation for multiple characters. E.g.,
      F = [Mat(160), [1]];
      G = bnrinit(bnfinit(x), F); \\ ray class group
      \\ vector of all primitive characters
      v = [ c | c <- bnrchar(G, [1]), bnrconductor(G, c) == F ];
      L = lfuncreate([G, v]);
      lfun(L,1); \\ evaluate all L(chi, 1) simultaneously
  - parallel support for lfuninit.
  - lfun now allows non-integral weights; in particular, lfunqf is no longer
    restricted to even dimensions.
  - new GP function lfundual (to manipulate the dual L-function) and
    lfunshift (to manipulate L(s + a))
  - new GP function dirpowerssum for fast computation of 1^s + 2^s + ... + N^s
  - exact (fast!) rational formulas for quadratic Dirichlet L-functions at
    non-positive integers. E.g.
      lfun(1, -7) -> 1/240 [ = zeta(-7) ]
      lfun(-691, -2) -> -175118   [ = L(chi_{-691}, -2) ]

[Modular forms]
  - mffromlfun now supports forms of half-integral weight, e.g. from lfunqf.
  - lfunmfspec is now implemented in all integral weights
  - new GP function mfgaloisprojrep (projective Artin representation attached
    to a modular form of weight 1 by Deligne-Serre)
  - new GP function mfisetaquo(f): can f be represented as a meromorphic eta
    quotient ?

[Miscellaneous]
  - new GP function permcycles to return the decomposition of a permutation
    as a product of disjoint cycles.
  - new GP function mscosets to construct H\G given generators for
    a group G and a description of a subgroup H

    ? PSL2 = [[0,1;-1,0], [1,1;0,1]];  \\ S and T
    \\ G = PSL2, H = Gamma0(2)
    ? [C] = mscosets(PSL2, g -> g[2,1] % 2 == 0);
    ? C \\ three cosets
    %3 = [[1, 0; 0, 1], [0, 1; -1, 0], [0, 1; -1, -1]]

    The function also returns the right action of G on H\G.

  - new GP function msfarey to build subgroups H of G = PSL_2(Z) (as Farey
    symbols) given an oracle deciding whether an element of G belongs to H.

    \\ Gamma_0(N) is built-in
    G0(N) = mspolygon(N);

    \\ Gamma_1(N) via Gamma_0(N)
    G1(N) = msfarey(G0(N), g -> my(a = g[1,1] % N); a == 1 || a == N-1);

    \\ Gamma(N)
    G(N) = msfarey(G1(N), g -> g[1,2] % N == 0);

  - better Windows support: allow install() in all cases + no longer crash
    on deep recursion [ (dive(n) = dive(n+1)); dive(0) ]

COMPATIBILITY ISSUES BETWEEN 2.11.4 and 2.13.0
==============================================

0) Obsoleted functions and interfaces:
  - removed member functions .futu and .tufu [deprecated since 2.2]
  - removed obsolete functions 'bnfcompress' and 'zetamultinit'
  - 'bnfissunit' is obsolete, use bnfisunit
  - 'rnfkummer' is obsolete, use bnrclassfield!
  - bnf.fu returns fundamental units in expanded form and can be very
    expensive. Use bnfunits !
  - bnfsunit()[1] returns fundamental S-units (without units) in expanded
    form and can be very expensive. More generally, bnfsunit contains mostly
    useless information in an expensive form and is now mostly obsolete.
    Use bnfunits !
  - matsnf: remove obsolete binary flag '2' [make it a no no-op as in mathnf]
  - inline / uninline are now obsolete, use export/unexport
  - renamed GP functions Strchr -> strchr, Strexpand -> strexpand,
    Strprintf -> strprintf, Strtex -> strtex
  - renamed GP function besseln -> bessely
  - poldisc(t_COMPLEX) now raises an exception (used to return -4)

1) Output changes:
  - matrixqz(,-1 or -2): no longer return result in HNF (very expensive now
    that the functions have been sped up and often not needed).
  - bnfisunit and bnfissunit: the torsion unit exponent is now a t_INT
    (used to be a t_INTMOD, requiring awkward conversions).
  - bnf.fu now returns 0 when bnf does not contain fundamental units [used to
    raise an exception]
  - qfbsolve now returns a vector of all solutions modulo units [was a single
    solution]
  - mfparams now returns [N,k,CHI,space or P, Phi] where Phi is the
    cyclotomic polynomials defining the field of values of CHI
  - fix compatibility issue between coefficients of a modular form f and
    values of corresponding Dirichlet character via znchar and chareval
    [used different cyclotomic polynomials to define the same cyclotomic field
    when order(chi) = 2 mod 4].
  - polinterpolate(X,Y,t,&e): e is now more precisely defined and the
    error estimate is given as a binary exponent; compared to the value
    dy returned up to this change, we have e = exponent(dy).
  - x ^ t_FRAC: return an exact result if possible; e.g. 4^(1/2) is now 2
  - GP timer and ## now also display the realtime when nbthreads is not 1

2) Input changes:
  - zetamult: the optional 'zetamultinit' argument is gone (the implemented
    algorithm is uniformly faster than the old one which made use of this
    precomputation)
  - parsum: removed optional argument initializing the sum to the given value
    [this wasn't very useful and removing it allows grouping subsums => faster]
  - mfshimura: disallow the discriminant D argument (not well defined);
    only allow squarefree positive t
  - disallow mfeisenstein(k,chi1,chi2) when Q(chi1 chi2) is not its field
    of definition: the resulting modular form would violate asumptions made
    in almost all mf routines, preventing those routines from accepting the
    form.
  - the argument of a GP user member function is now a reference and can be
    modified.
  - lfunconductor(L, [a,b]) is no longer supported to look for conductor in
    interval [a,b]; implement lfunconductor(L, list of potential N) instead
  - limitnum / asympnum: removed the useless optional parameter 'muli/k'
    [now implicitly set to 1, not to 20]; to replicate the old behavior
    extrapolate u(muli*n) .
  - CAVEAT: nfbasis documentation was incorrect: it is not guaranteed that
    nfbasis([T, listP]) returns the maximal order even if listP contains all
    prime divisors of the field discriminant. See ??nfbasis

--
Karim Belabas, IMB (UMR 5251)  Tel: (+33) (0)5 40 00 26 17
Universite de Bordeaux         Fax: (+33) (0)5 40 00 21 23
351, cours de la Liberation    http://www.math.u-bordeaux.fr/~kbelabas/
F-33405 Talence (France)       http://pari.math.u-bordeaux.fr/  [PARI/GP]
`