Example programs

The examples directory (https://github.com/fredrik-johansson/arb/tree/master/examples) contains several complete C programs, which are documented below. Running:

make examples

will compile the programs and place the binaries in build/examples.

pi.c

This program computes \(\pi\) to an accuracy of roughly n decimal digits by calling the arb_const_pi() function with a working precision of roughly \(n \log_2(10)\) bits.

Sample output, computing \(\pi\) to one million digits:

> build/examples/pi 1000000
computing pi with a precision of 3321933 bits... cpu/wall(s): 0.58 0.586
virt/peak/res/peak(MB): 28.24 36.84 8.86 15.56
[3.14159265358979323846{...999959 digits...}42209010610577945815 +/- 3e-1000000]

The program prints an interval guaranteed to contain \(\pi\), and where all displayed digits are correct up to an error of plus or minus one unit in the last place (see arb_printn()). By default, only the first and last few digits are printed. Pass 0 as a second argument to print all digits (or pass m to print m + 1 leading and m trailing digits, as above with the default m = 20).

hilbert_matrix.c

Given an input integer n, this program accurately computes the determinant of the n by n Hilbert matrix. Hilbert matrices are notoriously ill-conditioned: although the entries are close to unit magnitude, the determinant \(h_n\) decreases superexponentially (nearly as \(1/4^{n^2}\)) as a function of n. This program automatically doubles the working precision until the ball computed for \(h_n\) by arb_mat_det() does not contain zero.

Sample output:

> build/examples/hilbert_matrix 200
prec=20: 0 +/- 5.5777e-330
prec=40: 0 +/- 2.5785e-542
prec=80: 0 +/- 8.1169e-926
prec=160: 0 +/- 2.8538e-1924
prec=320: 0 +/- 6.3868e-4129
prec=640: 0 +/- 1.7529e-8826
prec=1280: 0 +/- 1.8545e-17758
prec=2560: 2.955454297e-23924 +/- 6.4586e-24044
success!
cpu/wall(s): 9.06 9.095
virt/peak/res/peak(MB): 55.52 55.52 35.50 35.50

keiper_li.c

Given an input integer n, this program rigorously computes numerical values of the Keiper-Li coefficients \(\lambda_0, \ldots, \lambda_n\). The Keiper-Li coefficients have the property that \(\lambda_n > 0\) for all \(n > 0\) if and only if the Riemann hypothesis is true. This program was used for the record computations described in [Joh2013] (the paper describes the algorithm in some more detail).

The program takes the following parameters:

keiper_li n [-prec prec] [-threads num_threads] [-out out_file]

The program prints the first and last few coefficients. It can optionally write all the computed data to a file. The working precision defaults to a value that should give all the coefficients to a few digits of accuracy, but can optionally be set higher (or lower). On a multicore system, using several threads results in faster execution.

Sample output:

> build/examples/keiper_li 1000 -threads 2
zeta: cpu/wall(s): 0.4 0.244
virt/peak/res/peak(MB): 167.98 294.69 5.09 7.43
log: cpu/wall(s): 0.03 0.038
gamma: cpu/wall(s): 0.02 0.016
binomial transform: cpu/wall(s): 0.01 0.018
0: -0.69314718055994530941723212145817656807550013436026 +/- 6.5389e-347
1: 0.023095708966121033814310247906495291621932127152051 +/- 2.0924e-345
2: 0.046172867614023335192864243096033943387066108314123 +/- 1.674e-344
3: 0.0692129735181082679304973488726010689942120263932 +/- 5.0219e-344
4: 0.092197619873060409647627872409439018065541673490213 +/- 2.0089e-343
5: 0.11510854289223549048622128109857276671349132303596 +/- 1.0044e-342
6: 0.13792766871372988290416713700341666356138966078654 +/- 6.0264e-342
7: 0.16063715965299421294040287257385366292282442046163 +/- 2.1092e-341
8: 0.18321945964338257908193931774721859848998098273432 +/- 8.4368e-341
9: 0.20565733870917046170289387421343304741236553410044 +/- 7.5931e-340
10: 0.22793393631931577436930340573684453380748385942738 +/- 7.5931e-339
991: 2.3196617961613367928373899656994682562101430813341 +/- 2.461e-11
992: 2.3203766239254884035349896518332550233162909717288 +/- 9.5363e-11
993: 2.321092061239733282811659116333262802034375592414 +/- 1.8495e-10
994: 2.3218073540188462110258826121503870112747188888893 +/- 3.5907e-10
995: 2.3225217392815185726928702951225314023773358152533 +/- 6.978e-10
996: 2.3232344485814623873333223609413703912358283071281 +/- 1.3574e-09
997: 2.3239447114886014522889542667580382034526509232475 +/- 2.6433e-09
998: 2.3246517591032700808344143240352605148856869322209 +/- 5.1524e-09
999: 2.3253548275861382119812576052060526988544993162101 +/- 1.0053e-08
1000: 2.3260531616864664574065046940832238158044982041872 +/- 3.927e-08
virt/peak/res/peak(MB): 170.18 294.69 7.51 7.51

logistic.c

This program computes the n-th iterate of the logistic map defined by \(x_{n+1} = r x_n (1 - x_n)\) where \(r\) and \(x_0\) are given. It takes the following parameters:

logistic n [x_0] [r] [digits]

The inputs \(x_0\), r and digits default to 0.5, 3.75 and 10 respectively. The computation is automatically restarted with doubled precision until the result is accurate to digits decimal digits.

Sample output:

> build/examples/logistic 10
Trying prec=64 bits...success!
cpu/wall(s): 0 0.001
x_10 = [0.6453672908 +/- 3.10e-11]

> build/examples/logistic 100
Trying prec=64 bits...ran out of accuracy at step 18
Trying prec=128 bits...ran out of accuracy at step 53
Trying prec=256 bits...success!
cpu/wall(s): 0 0
x_100 = [0.8882939923 +/- 1.60e-11]

> build/examples/logistic 10000
Trying prec=64 bits...ran out of accuracy at step 18
Trying prec=128 bits...ran out of accuracy at step 53
Trying prec=256 bits...ran out of accuracy at step 121
Trying prec=512 bits...ran out of accuracy at step 256
Trying prec=1024 bits...ran out of accuracy at step 525
Trying prec=2048 bits...ran out of accuracy at step 1063
Trying prec=4096 bits...ran out of accuracy at step 2139
Trying prec=8192 bits...ran out of accuracy at step 4288
Trying prec=16384 bits...ran out of accuracy at step 8584
Trying prec=32768 bits...success!
cpu/wall(s): 0.859 0.858
x_10000 = [0.8242048008 +/- 4.35e-11]

> build/examples/logistic 1234 0.1 3.99 30
Trying prec=64 bits...ran out of accuracy at step 0
Trying prec=128 bits...ran out of accuracy at step 10
Trying prec=256 bits...ran out of accuracy at step 76
Trying prec=512 bits...ran out of accuracy at step 205
Trying prec=1024 bits...ran out of accuracy at step 461
Trying prec=2048 bits...ran out of accuracy at step 974
Trying prec=4096 bits...success!
cpu/wall(s): 0.009 0.009
x_1234 = [0.256445391958651410579677945635 +/- 3.92e-31]

real_roots.c

This program isolates the roots of a function on the interval \((a,b)\) (where a and b are input as double-precision literals) using the routines in the arb_calc module. The program takes the following arguments:

real_roots function a b [-refine d] [-verbose] [-maxdepth n] [-maxeval n] [-maxfound n] [-prec n]

The following functions (specified by an integer code) are implemented:

  • 0 - \(Z(x)\) (Riemann-Siegel Z-function)
  • 1 - \(\sin(x)\)
  • 2 - \(\sin(x^2)\)
  • 3 - \(\sin(1/x)\)
  • 4 - \(\operatorname{Ai}(x)\) (Airy function)
  • 5 - \(\operatorname{Ai}'(x)\) (Airy function)
  • 6 - \(\operatorname{Bi}(x)\) (Airy function)
  • 7 - \(\operatorname{Bi}'(x)\) (Airy function)

The following options are available:

  • -refine d: If provided, after isolating the roots, attempt to refine the roots to d digits of accuracy using a few bisection steps followed by Newton’s method with adaptive precision, and then print them.
  • -verbose: Print more information.
  • -maxdepth n: Stop searching after n recursive subdivisions.
  • -maxeval n: Stop searching after approximately n function evaluations (the actual number evaluations will be a small multiple of this).
  • -maxfound n: Stop searching after having found n isolated roots.
  • -prec n: Working precision to use for the root isolation.

With function 0, the program isolates roots of the Riemann zeta function on the critical line, and guarantees that no roots are missed (there are more efficient ways to do this, but it is a nice example):

> build/examples/real_roots 0 0.0 50.0 -verbose
interval: [0, 50]
maxdepth = 30, maxeval = 100000, maxfound = 100000, low_prec = 30
found isolated root in: [14.111328125, 14.16015625]
found isolated root in: [20.99609375, 21.044921875]
found isolated root in: [25, 25.048828125]
found isolated root in: [30.419921875, 30.4443359375]
found isolated root in: [32.91015625, 32.958984375]
found isolated root in: [37.548828125, 37.59765625]
found isolated root in: [40.91796875, 40.966796875]
found isolated root in: [43.310546875, 43.3349609375]
found isolated root in: [47.998046875, 48.0224609375]
found isolated root in: [49.755859375, 49.7802734375]
---------------------------------------------------------------
Found roots: 10
Subintervals possibly containing undetected roots: 0
Function evaluations: 3058
cpu/wall(s): 0.202 0.202
virt/peak/res/peak(MB): 26.12 26.14 2.76 2.76

Find just one root and refine it to approximately 75 digits:

> build/examples/real_roots 0 0.0 50.0 -maxfound 1 -refine 75
interval: [0, 50]
maxdepth = 30, maxeval = 100000, maxfound = 1, low_prec = 30
refined root (0/8):
[14.134725141734693790457251983562470270784257115699243175685567460149963429809 +/- 2.57e-76]

---------------------------------------------------------------
Found roots: 1
Subintervals possibly containing undetected roots: 7
Function evaluations: 761
cpu/wall(s): 0.055 0.056
virt/peak/res/peak(MB): 26.12 26.14 2.75 2.75

Find the first few roots of an Airy function and refine them to 50 digits each:

> build/examples/real_roots 4 -10 0 -refine 50
interval: [-10, 0]
maxdepth = 30, maxeval = 100000, maxfound = 100000, low_prec = 30
refined root (0/6):
[-9.022650853340980380158190839880089256524677535156083 +/- 4.85e-52]

refined root (1/6):
[-7.944133587120853123138280555798268532140674396972215 +/- 1.92e-52]

refined root (2/6):
[-6.786708090071758998780246384496176966053882477393494 +/- 3.84e-52]

refined root (3/6):
[-5.520559828095551059129855512931293573797214280617525 +/- 1.05e-52]

refined root (4/6):
[-4.087949444130970616636988701457391060224764699108530 +/- 2.46e-52]

refined root (5/6):
[-2.338107410459767038489197252446735440638540145672388 +/- 1.48e-52]

---------------------------------------------------------------
Found roots: 6
Subintervals possibly containing undetected roots: 0
Function evaluations: 200
cpu/wall(s): 0.003 0.003
virt/peak/res/peak(MB): 26.12 26.14 2.24 2.24

Find roots of \(\sin(x^2)\) on \((0,100)\). The algorithm cannot isolate the root at \(x = 0\) (it is at the endpoint of the interval, and in any case a root of multiplicity higher than one). The failure is reported:

> build/examples/real_roots 2 0 100
interval: [0, 100]
maxdepth = 30, maxeval = 100000, maxfound = 100000, low_prec = 30
---------------------------------------------------------------
Found roots: 3183
Subintervals possibly containing undetected roots: 1
Function evaluations: 34058
cpu/wall(s): 0.032 0.032
virt/peak/res/peak(MB): 26.32 26.37 2.04 2.04

This does not miss any roots:

> build/examples/real_roots 2 1 100
interval: [1, 100]
maxdepth = 30, maxeval = 100000, maxfound = 100000, low_prec = 30
---------------------------------------------------------------
Found roots: 3183
Subintervals possibly containing undetected roots: 0
Function evaluations: 34039
cpu/wall(s): 0.023 0.023
virt/peak/res/peak(MB): 26.32 26.37 2.01 2.01

Looking for roots of \(\sin(1/x)\) on \((0,1)\), the algorithm finds many roots, but will never find all of them since there are infinitely many:

> build/examples/real_roots 3 0.0 1.0
interval: [0, 1]
maxdepth = 30, maxeval = 100000, maxfound = 100000, low_prec = 30
---------------------------------------------------------------
Found roots: 10198
Subintervals possibly containing undetected roots: 24695
Function evaluations: 202587
cpu/wall(s): 0.171 0.171
virt/peak/res/peak(MB): 28.39 30.38 4.05 4.05

Remark: the program always computes rigorous containing intervals for the roots, but the accuracy after refinement could be less than d digits.

poly_roots.c

This program finds the complex roots of an integer polynomial by calling acb_poly_find_roots() with increasing precision until the roots certainly have been isolated. The program takes the following arguments:

poly_roots [-refine d] [-print d] <poly>

Isolates all the complex roots of a polynomial with
integer coefficients. For convergence, the input polynomial
is required to be squarefree.

If -refine d is passed, the roots are refined to an absolute
tolerance better than 10^(-d). By default, the roots are only
computed to sufficient accuracy to isolate them.
The refinement is not currently done efficiently.

If -print d is passed, the computed roots are printed to
d decimals. By default, the roots are not printed.

The polynomial can be specified by passing the following as <poly>:

a <n>          Easy polynomial 1 + 2x + ... + (n+1)x^n
t <n>          Chebyshev polynomial T_n
u <n>          Chebyshev polynomial U_n
p <n>          Legendre polynomial P_n
c <n>          Cyclotomic polynomial Phi_n
s <n>          Swinnerton-Dyer polynomial S_n
b <n>          Bernoulli polynomial B_n
w <n>          Wilkinson polynomial W_n
e <n>          Taylor series of exp(x) truncated to degree n
m <n> <m>      The Mignotte-like polynomial x^n + (100x+1)^m, n > m
c0 c1 ... cn   c0 + c1 x + ... + cn x^n where all c:s are specified integers

This finds the roots of the Wilkinson polynomial with roots at the positive integers 1, 2, ..., 100:

> build/examples/poly_roots -print 15 w 100
prec=53: 0 isolated roots | cpu/wall(s): 0.42 0.426
prec=106: 0 isolated roots | cpu/wall(s): 1.37 1.368
prec=212: 0 isolated roots | cpu/wall(s): 1.48 1.485
prec=424: 100 isolated roots | cpu/wall(s): 0.61 0.611
done!
(1 + 1.7285178043492e-125j)  +/-  (7.2e-122, 7.2e-122j)
(2 + 5.1605530263601e-122j)  +/-  (3.77e-118, 3.77e-118j)
(3 + -2.58115555871665e-118j)  +/-  (5.72e-115, 5.72e-115j)
(4 + 1.02141628524271e-115j)  +/-  (4.38e-112, 4.38e-112j)
(5 + 1.61326834094948e-113j)  +/-  (2.6e-109, 2.6e-109j)
    ...
(95 + 4.15294196875447e-62j)  +/-  (6.66e-59, 6.66e-59j)
(96 + 3.54502401922667e-64j)  +/-  (7.37e-60, 7.37e-60j)
(97 + -1.67755595325625e-65j)  +/-  (6.4e-61, 6.4e-61j)
(98 + 2.04638822325299e-65j)  +/-  (4e-62, 4e-62j)
(99 + -2.73425468028238e-66j)  +/-  (1.71e-63, 1.71e-63j)
(100 + -1.00950111302288e-68j)  +/-  (3.24e-65, 3.24e-65j)
cpu/wall(s): 3.88 3.893

This finds the roots of a Bernoulli polynomial which has both real and complex roots. Note that the program does not attempt to determine that the imaginary parts of the real roots really are zero (this could be done by verifying sign changes):

> build/examples/poly_roots -refine 100 -print 20 b 16
prec=53: 16 isolated roots | cpu/wall(s): 0 0.007
prec=106: 16 isolated roots | cpu/wall(s): 0 0.004
prec=212: 16 isolated roots | cpu/wall(s): 0 0.004
prec=424: 16 isolated roots | cpu/wall(s): 0 0.004
done!
(-0.94308706466055783383 + -5.512272663168484603e-128j)  +/-  (2.2e-125, 2.2e-125j)
(-0.75534059252067985752 + 1.937401283040249068e-128j)  +/-  (1.09e-125, 1.09e-125j)
(-0.24999757119077421009 + -4.5347924422246038692e-130j)  +/-  (3.6e-127, 3.6e-127j)
(0.24999757152512726002 + 4.2191300761823281708e-129j)  +/-  (4.98e-127, 4.98e-127j)
(0.75000242847487273998 + 9.0360649917413170142e-128j)  +/-  (8.88e-126, 8.88e-126j)
(1.2499975711907742101 + 7.8804123808107088267e-127j)  +/-  (2.66e-124, 2.66e-124j)
(1.7553405925206798575 + 5.432465269253967768e-126j)  +/-  (6.23e-123, 6.23e-123j)
(1.9430870646605578338 + 3.3035377342500953239e-125j)  +/-  (7.05e-123, 7.05e-123j)
(-0.99509334829256233279 + 0.44547958157103608805j)  +/-  (5.5e-125, 5.5e-125j)
(-0.99509334829256233279 + -0.44547958157103608805j)  +/-  (5.46e-125, 5.46e-125j)
(1.9950933482925623328 + 0.44547958157103608805j)  +/-  (1.44e-122, 1.44e-122j)
(1.9950933482925623328 + -0.44547958157103608805j)  +/-  (1.43e-122, 1.43e-122j)
(-0.92177327714429290564 + -1.0954360955079385542j)  +/-  (9.31e-125, 9.31e-125j)
(-0.92177327714429290564 + 1.0954360955079385542j)  +/-  (1.02e-124, 1.02e-124j)
(1.9217732771442929056 + 1.0954360955079385542j)  +/-  (9.15e-123, 9.15e-123j)
(1.9217732771442929056 + -1.0954360955079385542j)  +/-  (8.12e-123, 8.12e-123j)
cpu/wall(s): 0.02 0.02

complex_plot.c

This program plots one of the predefined functions over a complex interval \([x_a, x_b] + [y_a, y_b]i\) using domain coloring, at a resolution of xn times yn pixels.

The program takes the parameters:

complex_plot [-range xa xb ya yb] [-size xn yn] <func>

Defaults parameters are \([-10,10] + [-10,10]i\) and xn = yn = 512.

The output is written to arbplot.ppm. If you have ImageMagick, run convert arbplot.ppm arbplot.png to get a PNG.

Function codes <func> are:

  • gamma - Gamma function
  • digamma - Digamma function
  • lgamma - Logarithmic gamma function
  • zeta - Riemann zeta function
  • erf - Error function
  • ai - Airy function Ai
  • bi - Airy function Bi
  • besselj - Bessel function \(J_0\)
  • bessely - Bessel function \(Y_0\)
  • besseli - Bessel function \(I_0\)
  • besselk - Bessel function \(K_0\)
  • modj - Modular j-function
  • modeta - Dedekind eta function
  • barnesg - Barnes G-function
  • agm - Arithmetic geometric mean

The function is just sampled at point values; no attempt is made to resolve small features by adaptive subsampling.

For example, the following plots the Riemann zeta function around a portion of the critical strip with imaginary part between 100 and 140:

> build/examples/complex_plot zeta -range -10 10 100 140 -size 256 512