SeedRandom¶

Status: Stable

documented, exercised by the test suite and/or worked examples, with no known limitations recorded.

Description¶

SeedRandom[n]

seeds the pseudorandom generator with the integer n.

SeedRandom[]

reseeds the pseudorandom generator from system entropy.

Examples (6)¶

Every input below was run against the current Mathilda build and its output recorded.

Basic examples (2)¶

In[1]:= SeedRandom[42]; {RandomInteger[], RandomInteger[], RandomInteger[]}
Out[1]= {1, 1, 0}

In[2]:= SeedRandom[42]; {RandomInteger[], RandomInteger[], RandomInteger[]}
Out[2]= {1, 1, 0}

Applications (4)¶

Seed, then draw

In[3]:= SeedRandom[42]; RandomReal[]
Out[3]= 0.814305

The same seed reproduces the same draw

In[4]:= SeedRandom[42]; RandomReal[]
Out[4]= 0.814305

One seed fixes the whole sequence

In[5]:= SeedRandom[1]; RandomInteger[{1, 6}, 5]
Out[5]= {5, 5, 1, 5, 2}

A bignum seed is allowed

In[6]:= SeedRandom[123456789012345678901234567890]; RandomInteger[100]
Out[6]= 6

Implementation notes¶

builtin_seedrandom (in src/random.c) reseeds the single global GMP Mersenne Twister state g_rand_state shared by all Random* builtins. SeedRandom[n] calls gmp_randseed_ui for a machine integer or gmp_randseed for a bignum seed, making subsequent random draws reproducible; SeedRandom[] reseeds from system entropy (time(NULL) ^ clock()). Both first call ensure_rand_init to lazily construct the state with gmp_randinit_mt. Returns Null; a non-integer seed returns NULL (unevaluated).

Protected.
After SeedRandom[n], the sequence of pseudorandom numbers generated will be the same each time.
Accepts bignums as seeds.

Attributes: Protected.

References¶

Source: src/random.c
Specification: docs/spec/builtins/random-number-generation.md
Tests: tests/test_convolutions.c
Tests: tests/test_correlations.c
Tests: tests/test_erf.c
Tests: tests/test_erfc.c

Notes & additional examples¶

Notes¶

SeedRandom[n] reseeds the single global generator shared by every Random* builtin — RandomInteger, RandomReal, RandomComplex, RandomChoice, RandomSample, and the RandomVariate/RandomImage family — so one seed makes all of them reproducible together, which is why the two SeedRandom[42] draws above are equal. SeedRandom[] with no argument reseeds from system entropy (time ^ clock), making subsequent draws non-reproducible again.

n may be a machine integer or an arbitrary-precision bignum. Reseeding also drops any cached Gaussian deviate held by the distribution sampler, so RandomVariate is reproducible after a reseed and not just RandomReal. SeedRandom returns Null; a non-integer seed is left unevaluated.

The stream is reproducible within a build, but is not promised stable across Mathilda versions or across builds with and without 128-bit integer support, so a seed pins a sequence for a given binary rather than forever.