NAME
Math::Pari  Perl interface to PARI.
SYNOPSIS
use Math::Pari;
$a = PARI 2;
print $a**10000;
or
use Math::Pari qw(Mod);
$a = Mod(3,5);
print $a**10000;
DESCRIPTION
This package is a Perl interface to famous library PARI for numerical/scientific/numbertheoretic calculations. It allows use of most PARI functions as Perl functions, and (almost) seamless merging of PARI and Perl data. In what follows we suppose prior knowledge of what PARI is (see ftp://megrez.math.ubordeaux.fr/pub/pari, or Math::libPARI).
EXPORTed functions
 DEFAULT

By default the package exports functions PARI(), PARIcol(), PARIvar(), PARImat(), PARImat_tr() and parse_as_gp() which convert their argument(s) to a PARI object. (In fact PARI() is just an alias for
new Math::Pari
). The function PARI() accepts following data as its arguments One integer

Is converted to a PARI integer.
 One float

Is converted to a PARI float.
 One string

Is executed as a PARI expression (so should not contain whitespace).
 PARI object

Is passed unchanged.
 Reference to a Perl array

Each element is converted using the same rules, PARI vectorrow with these elements is returned.
 Several of above

The same as with a reference to array.
 Conflicts of rules in PARI()

In deciding what rule of the above to apply the preference is given to the uppermost choice of those available now. If none matches, then the string rule is used. So
PARI(1)
returns integer,PARI(1.)
returns float,PARI("1")
evaluates1
as a PARI expression (well, the result is the same asPARI(1)
, only slower).Note that for Perl these data are synonymous, since Perl freely converts between integers, float and strings. However, to PARI() only what the argument is now is important. If $v is
1
in the Perl world,PARI($v)
may convert it to an integer, float, or to the result of evaluating the PARI program1
(all depending on how $v was created and accessed in Perl).This is a fundamental limitation of creating an interface between two systems, both with polymorphic objects, but with subtly different semantic of the flavors of these objects. In reality, however, this is rarely a problem.
 PARIcol(), PARImat() and PARImat_tr()

PARIcol() behaves in the same way as PARI() unless given several arguments. In the latter case it returns a vectorcolumn instead of a vectorrow.
PARImat() constructs a matrix out of the given arguments. It will work if PARI() will construct a vector of vectors given the same arguments. The internal vectors become columns of the matrix. PARImat_tr() behaves similarly, but the internal vectors become rows of the matrix.
Since PARI matrices are similar to vectorrows of vectorcolumns, PARImat() is quicker, but PARImat_tr() better corresponds to the PARI input and output forms of matrices:
print PARImat [[1,2], [3,4]]; # prints [1,3;2,4] print PARImat_tr [[1,2], [3,4]]; # prints [1,2;3,4]
 parse_as_gp()

Did you notice that when taking a string, PARI() requires that there is no whitespace there (outside of string constants)? This is exactly as the
PARI
library parses strings. However, to simplify human interaction, thegp
calculator allows whitespace, comments, breaking into multiple lines, many independent expressions (such as function definitions).We do not include the corresponding C code from the calculator, but provide a Perl clone. It supports whitespace,
\\
 and/* */
comments, and, for multiline arguments, it supports line continuation via trailing\
, trailing binary ops, comma, opening parenthesis/bracket; moreover, group of lines in{}
are joined into one line. (Additionally,\q
and\p
are recognized, as well as trailing allocatemem().\e
is tolerated.)Keep in mind that this is just a convenience function, and no attempt was performed to make it particularly quick. Moreover, the PARI user functions (or maybe it is better to call them user macros?) are currently not automatically importable into Perl, so to access functions defined in parse_as_gp()' argument may be awkward. (The temporary fix is to use a temporary convenience function __wrap_PARI_macro():
parse_as_gp <<EOP; add2(x) = x + 2 EOP *add2 = Math::Pari::__wrap_PARI_macro 'add2'; print add2(17);
but keep in mind that the generated this way wrapper is also not designed to be quick.)
With the optional second argument
'quote'
, it would return an unevaluated array of strings instead of the result of evaluation. Special strings\q
etc. are replaced by references to appropriate (undocumented) Perl subroutines.With the optional third argument
'echo'
, would "echo" the commands (preceded by"? "
) before execution. WithTRUE
optional fourth argument (command counter), would "echo" the result too (preceded by"%C = "
, withC
being the command counter, which is incremented). use
with arguments
If arguments are specified in the
use Math::Pari
directive, the PARI functions appearing as arguments are exported in the caller context. In this case the function PARI() and friends is not exported, so if you need them, you should include them into export list explicitly, or include:DEFAULT
tag:use Math::Pari qw(factorint PARI); use Math::Pari qw(:DEFAULT factorint);
or simply do it in two steps
use Math::Pari; use Math::Pari 'factorint';
The other recognized tags are
:PARI
,:all
,prec=NUMBER
, overloaded constants tags (:int
,:float
,:hex
) and "section names" tags.Additionally, the number tags (e.g.,
:4
) export functions from the PARI library from the given "section" (moreover,:PARI
exports all of the "sections"). Tag:all
exports all of the exportable symbols and:PARI
.With older versions of PARI, giving
?
command togp
(the PARI calculator) lists the following sections:1: Standard monadic or dyadic OPERATORS 2: CONVERSIONS and similar elementary functions 3: TRANSCENDENTAL functions 4: NUMBER THEORETICAL functions 5: Functions related to ELLIPTIC CURVES 6: Functions related to general NUMBER FIELDS 7: POLYNOMIALS and power series 8: Vectors, matrices, LINEAR ALGEBRA and sets 9: SUMS, products, integrals and similar functions 10: GRAPHIC functions 11: PROGRAMMING under GP
Starting with GP/PARI version 2.9.0, this list depends significantly on this version; for backward compatibility, we follow this older list of section numbers (to avoid confusion, better use symbolic names below). For compatibility, we assign arbitrary numbers to newer sections:
100: LFUNCTIONS 101: MODULAR SYMBOLS 102: Associative and central simple ALGEBRAS 103: functions related to COMBINATORICS 104: MODULAR FORMS
One can use section names instead of number tags. Recognized names are
:standard :conversions :transcendental :number :elliptic :fields :polynomials :vectors :sums :graphic :programming :l_functions :modular_symb :algebras :combinatorics :modular
One can get the list of all of the functions accessible by
Math::Pari
, or the accessible functions from the given section using listPari() function.Starting from version 5.005 of Perl, three constantoverload tags are supported:
:int
,:float
,:hex
. If used, all the integer/float/hexoroctalorbinary literals in Perl will be automatically converted to became PARI objects. For example,use Math::Pari ':int'; print 2**1000;
is equivalent to
print PARI(2)**PARI(1000);
(The support for this Perl feature is buggy before the Perl version 5.005_57  unless Perl uses mymalloc options; you can check for this with
perl V:usemymalloc
.) Note also that (at least with some versions of Perl) one should enable':float'
for conversion of long integer literals (Perl may consider them as floats, since they won't fit into Perl integers); note that it is PARI which determines which PARI subtype is assigned to each such literal:use Math::Pari ':float', 'type_name'; print type_name 22222222222222222222222;
prints
t_INT
.
Available functions
Directly accessible from Perl
This package supports all the functions from the PARI library with a signature which can be recognized by Math::Pari. This means that when you update the PARI library, the newly added functions will we available without any change to this package; only a recompile is needed. In fact no recompile will be needed if you link libPARI dynamically (you need to modify the Makefile manually to do this).
You can "reach" unsupported functions via going directly to PARI parser using the string flavor of PARI() function, as in
3 + PARI('O(x^17)');
For some "unreachable" functions there is a special wrapper functions, such as O(variable,power)
).
The following functions are specific to GP calculator, thus are not available to Math::Pari in any way:
default error extern input print print1 printp printp1
printtex quit read system whatnow write write1 writetex
whatnow() function is useless, since Math::Pari does not support the "compatibility" mode (with older PARI library). The functionality of print(), write() and variants is available via automatic string translation, and pari_print() function and its variants (see "Printout functions").
default() is the only important function with functionality not supported by the current interface. Note however, that four most important default() actions are supported by allocatemem(), setprimelimit(), setprecision() and setseriesprecision() functions. (When called without arguments, these functions return the current values.)
allocatemem($bytes) should not be called from inside Math::Pari functions (such as forprimes()).
Arguments
Arguments to PARI functions are automatically converted to long
or a PARI object depending on the signature of the actual library function. The arguments are forced into the given type, so even if gp
rejects your code similar to
func(2.5); # func() takes a long in C
arguing that a particular argument should be of type T_INT
(i.e., a Pari integer), the corresponding code will work in Math::Pari
, since 2.5 is silently converted to long
, per the function signature.
Return values
PARI functions return a PARI object or a Perl's integer depending on what the actual library function returns.
Additional functions
Some PARI functions are available in gp
(i.e., in PARI
calculator) via infix notation only. In Math::Pari
these functions are available in functional notations too. Some other convenience functions are also made available.
 Infix, prefix and postfix operations

are available under names
gneg, gadd, gsub, gmul, gdiv, gdivent, gmod, gpui, gle, gge, glt, ggt, geq, gne, gegal, gor, gand, gcmp, gcmp0, gcmp1, gcmp_1.
gdivent
means euclidean quotient,gpui
is power,gegal
checks whether two objects are equal,gcmp
is applicable to two real numbers only,gcmp0
,gcmp1
,gcmp_1
compare with 0, 1 and 1 correspondingly (see PARI user manual for details, or Math::libPARI). Note that all these functions are more readily available via operator overloading, so instead ofgadd(gneg($x), $y)
one can write
$x+$y
(as far as overloading may be triggered, see overload, so we assume that at least one of $x or $y is a PARI object).
 Conversion functions

pari2iv, pari2nv, pari2num, pari2pv, pari2bool
convert a PARI object to an integer, float, integer/float (whatever is better), string, and a boolean value correspondingly. Most the time you do not need these functions due to automatic conversions.
 Printout functions

pari_print, pari_pprint, pari_texprint
perform the same conversions to strings as their PARI counterparts, but do not print the result. The difference of pari_print() with pari2pv() is the number of significant digits they output, and whitespace in the output. pari2pv(), which is intended for "computerreadable strings", outputs as many digits as is supported by the current precision of the number; while pari_print(), which targets humanreadable strings, takes into account the currently specified output precision too.
 Constant functions

Some mathematical constants appear as function without arguments in PARI. These functions are available in Math::Pari too. If you export them as in
use Math::Pari qw(:DEFAULT Pi I Euler);
they can be used as barewords in your program:
$x = Pi ** Euler;
 Lowlevel functions

For convenience of lowlevel PARI programmers some lowlevel functions are made available as well (all except type_name() and changevalue() are not exportable):
typ($x) lg($x) lgef($x) lgefint($x) longword($x, $n) type_name($x) changevalue($name,$newvalue)
Here longword($x,$n) returns
$n
th word in the memory representation of $x (including noncode words). type_name() differs from the PARI function type(): type() returns a PARI object, while type_name() returns a Perl string. (PARI objects of string type behave very nonintuitive w.r.t. string comparison functions; remember that they are compared using lex() to the results of evaluation of other argument of comparison!)The function listPari($number) outputs a list of names of PARI functions in the section $number. Use listPari(1) to get the list across all of the sections. (_listPari() behaves likewise, with the versionspecific section numbers.)
 Uncompatible functions

O
Since implementing
O(7**6)
would be very tedious, we provide a twoargument formO(7,6)
instead (meaning the same asO(7^6)
in PARI). Note that with polynomials there is no problem like this one, bothO($x,6)
andO($x**6)
work.ifact(n)
integer factorial functions, available from
gp
asn!
.
Looping functions
PARI has a big collection of functions which loops over some set. Such a function takes two special arguments: loop variable, and the code to execute in the loop.
The code can be either a string (which contains PARI code to execute  thus should not contain whitespace), or a Perl code reference. The loop variable can be a string giving the name of PARI variable (as in
fordiv(28, 'j', 'a=a+j+j^2');
or
$j= 'j';
fordiv(28, $j, 'a=a+j+j^2');
), a PARI monomial (as in
$j = PARI 'j';
fordiv(28, $j, sub { $a += $j + $j**2 });
), or a "delayed Math::Pari variable" (as in
$j = PARIvar 'j';
fordiv(28, $j, 'a=a+j+j^2');
). If none of these applies, as in
my $j; # Have this in a separate statement
fordiv(28, $j, sub { $a += $j + $j**2 });
then during the execution of the sub
, Math::Pari would autogenerate a PARI variable, and would put its value in $j; this value of $j is temporary only, the old contents of $j is restored when fordiv() returns.
Note that since you have no control over this name, you will not be able to use this variable from your PARI code; e.g.,
$j = 7.8;
fordiv(28, $j, 'a=a+j+j^2');
will not make j
mirror $j (unless you explicitly set up j
to be a noargument PARI function mirroring $j, see "Accessing Perl functions from PARI code").
Caveats. There are 2 flavors of the "code" arguments (string/sub
), and 4 types of the "variable" arguments (string/monomial/PARIvar
/other). However, not all 8 combinations make sense. As we already explained, an "other" variable cannot work with a "string" code.
Useless musing alert! Do not read the rest of this section! Do not use "string" variables with sub
code, and do not ask why!
Additionally, the following code will not do what you expect
$x = 0;
$j = PARI 'j';
fordiv(28, 'j', sub { $x += $j } ); # Use $j as a loop variable!
since the PARI function fordiv
localizes the PARI variable j
inside the loop, but $j will still reference the old value; the old value is a monomial, not the index of the loop (which is an integer each time sub
is called). The simplest workaround is not to use the above syntax (i.e., not mixing literal loop variable with Perl loop code, just using $j as the second argument to fordiv
is enough):
$x = 0;
$j = PARI 'j';
fordiv(28, $j, sub { $x += $j } );
Alternately, one can make a delayed variable $j which will always reference the same thing j
references in PARI now by using PARIvar
constructor
$x = 0;
$j = PARIvar 'j';
fordiv(28, 'j', sub { $x += $j } );
(This problem is similar to
$ref = \$_; # $$ref is going to be old value even after
# localizing $_ in Perl's grep/map
not accessing localized values of $_ in the plain Perl.)
Another possible quirk is that
fordiv(28, my $j, sub { $a += $j + $j**2 });
will not work too  by a different reason. my
declarations change the meaning of $j only after the end of the current statement; thus $j inside sub
will access a different variable $j (typically a nonlexical, global variable $j) than one you declared on this line.
Accessing Perl functions from PARI code
Use the same name inside PARI code:
sub counter { $i += shift; }
$i = 145;
PARI 'k=5' ;
fordiv(28, 'j', 'k=k+counter(j)');
print PARI('k'), "\n";
prints
984
Due to a difference in the semantic of variablenumberofparametersfunctions between PARI and Perl, if the Perl subroutine takes a variable number of arguments (via @
in the prototype or a missing prototype), up to 6 arguments are supported when this function is called from PARI. If called from PARI with fewer arguments, the rest of arguments will be set to be integers PARI 0
.
Note also that no direct import of Perl variables is available yet (but you can write a function wrapper for this):
sub getv () {$v}
There is an unsupported (and undocumented ;) function for explicitly importing Perl functions into PARI, possibly with a different name, and possibly with explicitly specifying number of arguments.
PARI objects
Functions from PARI library may take as arguments and/or return values the objects of C type GEN
. In Perl these data are encapsulated into special kind of Perl variables: PARI objects. You can check for a variable $obj
to be a PARI object using
ref $obj and $obj>isa('Math::Pari');
Most the time you do not need this due to automatic conversions and overloading.
PARI monomials and Perl barewords
If very lazy, one can code in Perl the same way one does it in PARI. Variables in PARI are denoted by barewords, as in x
, and in the default configuration (no warnings, no strict) Perl allows the same  up to some extent. Do not do this, since there are many surprising problems.
Some bareletters denote Perl operators, like q
, x
, y
, s
. This can lead to errors in Perl parsing your expression. E.g.,
print sin(tan(t))tan(sin(t))asin(atan(t))+atan(asin(t));
may parse OK after use Math::Pari qw(sin tan asin atan)
. Why?
After importing, the word sin
will denote the PARI function sin(), not Perl operator sin(). The difference is subtle: the PARI function implicitly forces its arguments to be converted PARI objects; it gets 't'
as the argument, which is a string, thus is converted to what t
denotes in PARI  a monomial. While the Perl operator sin() grants overloading (i.e., it will call PARI function sin() if the argument is a PARI object), it does not force its argument; given 't'
as argument, it converts it to what sin() understands, a float (producing 0.
), so will give 0.
as the answer.
However
print sin(tan(y))tan(sin(y))asin(atan(y))+atan(asin(y));
would not compile. You should avoid lowercase barewords used as PARI variables, e.g., do
$y = PARI 'y';
print sin(tan($y))tan(sin($y))asin(atan($y))+atan(asin($y));
to get
1/18*y^9+26/4725*y^1141/1296*y^13+328721/16372125*y^15+O(y^16)
(BTW, it is a very good exercise to get the leading term by hand).
Well, the same advice again: do not use barewords anywhere in your program!
Overloading and automatic conversion
Whenever an arithmetic operation includes at least one PARI object, the other arguments are converted to a PARI object and the corresponding PARI library functions is used to implement the operation. Currently the following arithmetic operations are overloaded:
unary 
+  * / % ** abs cos sin exp log sqrt
<< >>
<= == => < > != <=>
le eq ge lt gt ne cmp
 & ^ ~
Numeric comparison operations are converted to gcmp
and friends, string comparisons compare in lexicographical order using lex
.
Additionally, whenever a PARI object appears in a situation that requires integer, numeric, boolean or string data, it is converted to the corresponding type. Boolean conversion is subject to usual PARI pitfalls related to imprecise zeros (see documentation of gcmp0
in PARI reference).
For details on overloading, see overload.
Note that a check for equality is subject to same pitfalls as in PARI due to imprecise values. PARI may also refuse to compare data of different types for equality if it thinks this may lead to counterintuitive results.
Note also that in PARI the numeric ordering is not defined for some types of PARI objects. For string comparison operations we use PARIlexicographical ordering.
PREREQUISITES
Perl
In the versions of perl earlier than 5.003 overloading used a different interface, so you may need to convert use overload
line to %OVERLOAD
, or, better, upgrade.
PARI
Starting from version 2.0, this module comes without a PARI library included.
For the source of PARI library see ftp://megrez.math.ubordeaux.fr/pub/pari.
Perl vs. PARI: different syntax
Note that the PARI notations should be used in the string arguments to PARI() function, while the Perl notations should be used otherwise.
^

Power is denoted by
**
in Perl. \
and\/

There are no such operators in Perl, use the word forms
gdivent(x,y)
andgdivround(x,y)
instead. ~

There is no postfix
~
Perl operator. Use mattranspose() instead. '

There is no postfix
'
Perl operator. Use deriv() instead. !

There is no postfix
!
Perl operator. Use factorial()/ifact() instead (returning a real or an integer correspondingly).  big integers

Perl converts big literal integers to doubles if they could not be put into C integers (the particular flavor can be found in the output of
perl V
in newer version of Perl, look forivtype
/ivsize
). If you want to input such an integer, usewhile ($x < PARI('12345678901234567890')) ...
instead of
while ($x < 12345678901234567890) ...
Why? Because conversion to double leads to precision loss (typically above 1e15, see perlnumber), and you will get something like 12345678901234567168 otherwise.
Starting from version 5.005 of Perl, if the tag
:int
is used on the 'use Math::Pari' line, all of the integer literals in Perl will be automatically converted to became PARI objects. E.g.,use Math::Pari ':int'; print 2**1000;
is equivalent to
print PARI(2)**PARI(1000);
Similarly, large integer literals do not lose precision.
This directive is lexically scoped. There is a similar tag
:hex
which affects hexadecimal, octal and binary constants. One may also need to use tag:float
for autoconversion of large integer literals which Perl considers as floating point literals (see "use
with arguments" for details).  doubles

Doubles in Perl are typically of precision approximately 15 digits (see perlnumber). When you use them as arguments to PARI functions, they are converted to PARI real variables, and due to intermediate 15digitstobinary conversion of Perl variables the result may be different than with the PARI manydigitstobinary conversion. E.g.,
PARI(0.01)
andPARI('0.01')
differ at 19th place, assetprecision(38); print pari_print(0.01), "\n", pari_print('0.01'), "\n";
shows.
Note that setprecision() changes the output format of pari_print() and friends, as well as the default internal precision. The generic PARI===>string conversion does not take into account the output format, thus
setprecision(38); print PARI(0.01), "\n", PARI('0.01'), "\n", pari_print(0.01), "\n";
will print all the lines with different number of digits after the point: the first one with 22, since the double 0.01 was converted to a lowprecision PARI object, the second one with 41, since internal form for precision 38 requires that many digits for representation, and the last one with 39 to have 38 significant digits.
Starting from version 5.005 of Perl, if the tag
:float
is used on theuse Math::Pari
line, all the float literals in Perl will be automatically converted to became PARI objects. E.g.,use Math::Pari ':float'; print atan(1.);
is equivalent to
print atan(PARI('1.'));
Similarly, large float literals do not lose precision.
This directive is lexically scoped.
 array base

Arrays are 1based in PARI, are 0based in Perl. So while array access is possible in Perl, you need to use different indices:
$nf = PARI 'nf'; # assume that PARI variable nf contains a number field $a = PARI('nf[7]'); $b = $nf>[6];
Now $a and $b contain the same value.
 matrices

Note that
PARImat([[...],...,[...])
constructor creates a matrix with specified columns, while in PARI the command[1,2,3;4,5,6]
creates a matrix with specified rows. Use a convenience function PARImat_tr() which will transpose a matrix created by PARImat() to use the same order of elements as in PARI.  builtin perl functions

Some PARI functions, like
length
andeval
, are Perl (semi)reserved words. To reach these functions, one should either import them:use Math::Pari qw(length eval);
or call them with prefix (like
&length
) or the full name (likeMath::Pari::length
).
Highresolution graphics
If you have Term::Gnuplot Perl module installed, you may use highresolution graphic primitives of PARI. Before the usage you need to establish a link between Math::Pari and Term::Gnuplot by calling link_gnuplot(). You can change the output filehandle by calling set_plot_fh(), and output terminal by calling plotterm(), as in
use Math::Pari qw(:graphic asin);
link_gnuplot(); # automatically loads Term::Gnuplot
plotterm('emtex');
plot_outfile_set('out.tex'); # better do after plotterm()
ploth($x, .5, .999, sub {asin $x});
close FH or die;
libPARI documentation
libPARI documentation is included, see Math::libPARI. It is converted from Chapter 3 of PARI/GP documentation by the gphelp script of GP/PARI.
ENVIRONMENT
No environment variables are used.
BUGS
A few of PARI functions are available indirectly only.
Using overloading constants with the Perl versions below 5.005_57 could lead to segfaults (at least without
D usemymalloc
), as in:use Math::Pari ':int'; for ( $i = 0; $i < 10 ; $i++ ) { print "$i\n" }
It may be possible that conversion of a Perl value which has both the integer slot and the floating slot set may create a PARI integer, even if the actual value is not an integer.
problems with refcounting of array elements and Mod().
Workaround: make the modulus live longer than the result of Mod(). Until Perl version
5.6.1
, one should exercise a special care so that the modulus goes out of scope on a different statement than the result:{ my $modulus = 125; { my $res = Mod(34, $modulus); print $res; } $fake = 1; # A (fake) statement here is required }
Here $res is destructed before the
$fake = 1
statement, $modulus is destructed before the first statement after the provided block. However, if you remove the$fake = 1
statement, both these variables are destructed on the first statement after the provided block (and in a wrong order!).In
5.6.1
declaring $modulus before $res is all that is needed to circumvent the same problem:{ my $modulus = 125; my $res = Mod(34, $modulus); print $res; } # destruction will happen in a correct order.
Access to array elements may result in similar problems. Hard to fix since in PARI the data is not refcounted.
Legacy implementations of dynalinking require the code of DLL to be compiled to be "position independent" code (PIC). This slows down the execution, while allowing sharing the loaded copy of the DLL between different processes. [On contemporary architectures the same effect is allowed without the positionindependent hack.]
Currently, PARI assembler files are not positionindependent. When compiled for the dynamic linking on legacy systems, this creates a DLL which cannot be shared between processes. Some legacy systems are reported to recognize this situation, and load the DLL as a nonshared module. However, there may be systems (are there?) on which this can cause some "problems".
Summary: if the dynaloading on your system requires some kind of
fPIC
flag, using "assembler" compiles (anything butmachine=none
) *may* force you to do a static build (i.e., creation of a custom Perl executable withperl Makefile.PL static make perl make test_static
).
isprime() is a misnomer before PARI version 2.3!
In older versions of PARI, the oneargument variant of the function isprime() is actually checking for probable primes. Moreover, it has certain problems.
POSSIBLE WORKAROUND (not needed for newer PARI): before version 2.3 of PARI, to get probability of misdetecting a prime below 1e12, call isprime() twice; below 1e18, call it 3 times; etc. (The algorithm is probabilistic, and the implementation is such that the result depends on which calls to isprime() were performed ealier.)
The problems: first, while the default algorithm (before version 2.3) gives practically acceptable results in nonadversarial situations, the worstcase behaviour is significantly worse than the average behaviour. The algorithm is looking for socalled "witnesses" (with up to 10 tries) among random integers; usually, witnesses are abundant. However, there are nonprime numbers for which the fraction of witnesses is close to the theoretical minimum, 0.75; with 10 random tries, the probability of missing a witness for such numbers is close to 1e6. (The known worstcase numbers M have phi(M)/4 nonwitnesses, with M=P(2P1), prime P, 2P1 and 4P+1; the proportion of such numbers near K is expected to be const/sqrt(K)log(K)^2. Note that numbers which have more than about 5% nonwitnesses may also be candidates for false positives. Conjecturally, they are of the form (aD+1)(bD+1) with a<b, ab <= const, prime aD+1, and bD+1, and D not divisible by high power of 2 (above a=1, b=2 and D is odd); the proportion of such numbers may have a similar asymptotic const/sqrt(K)log(K)^2.)
Second, the random number generator is "reset to known state" when PARI library is initialized. That means that the behaviour is actually predictable if one knows which calls to isprime() are performed; an adversary can find nonprimes M which will trigger a false positive exactly on the Nth call to isprime(M) (for particular values of N). With enough computing resources, one can find nonprimes M for which N is relatively small (with M about 1e9, one can achieve N as low as 1000). Compare with similar (but less abundant) examples for simpler algorithm, Carmichael numbers; see also numbers with big proportion of nonwitnesses and numbers with many nonwitnesses, and the conjecture about proportion.
INITIALIZATION
When Math::Pari is loaded, it examines variables $Math::Pari::initmem and $Math::Pari::initprimes. They specify up to which number the initial list of primes should be precalculated, and how large should be the arena for PARI calculations (in bytes). (These values have safe defaults.)
Since setting these values before loading requires either a BEGIN
block, or postponing the loading (use
vs. require
), it may be more convenient to set them via Math::PariInit:
use Math::PariInit qw( primes=12000000 stack=1e8 );
use Math::PariInit
also accepts arbitrary Math::Pari import directives, see Math::PariInit.
These values may be changed at runtime too, via allocatemem() and setprimelimit(), with performance penalties for recalculation/reallocation.
AUTHOR
Ilya Zakharevich, ilyaz@cpan.org