Skip to content
An implementation of Takuji Nishimura's and Makoto Matsumoto's Mersenne Twister pseudo random number generator in Go.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
COPYING initial commit Oct 21, 2013
README.rst doc fix Oct 23, 2013
doc.go initial commit Oct 21, 2013
mt19937.go Speed up the .Read() method. Jul 15, 2018
mt19937_test.go Make some mention of the fact MT19937 implements the rand.Source64 in… Jul 15, 2018


The Mersenne Twister in Go

An implementation of Takuji Nishimura's and Makoto Matsumoto's Mersenne Twister pseudo random number generator in Go.

Copyright (C) 2013 Jochen Voss

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

The homepage of this package is at <>. Please send any comments or bug reports to the program's author, Jochen Voss <>.


The Mersenne Twister is a pseudo random number generator (PRNG), developed by Takuji Nishimura and Makoto Matsumoto. The Mersenne Twister is, for example, commonly used in Monte Carlo simulations and is the default random number generator for many programming languages, e.g. Python and R. This package implements the 64bit version of the algorithm.


This package can be installed using the go get command:

go get


Detailed usage instructions are available via the package's online help, either on or on the command line:

go doc

The class MT19937 represents instances of the Mersenne Twister. New instances can be allocated using the mt19937.New() function. A seed can be set using the .Seed() or .SeedFromSlice() methods. MT19937 implements the rand.Source interface from the math/rand package. Typically the PRNG is wrapped in a rand.Rand object as in the following example:

rng := rand.New(mt19937.New())

Comparison to the Go Default PRNG

Go has a built-in PRNG provided by the math/rand package. I did not find any information about this built-in PRNG except for a comment in the source code which says "algorithm by DP Mitchell and JA Reeds". In contrast, the MT19737 generator provided in this package is a well-understood random number generator. Relevant references include [Ni2000] and [MatNi1998].

[Ni2000]T. Nishimura, Tables of 64-bit Mersenne Twisters, ACM Transactions on Modeling and Computer Simulation 10, 2000, pages 348-357.
[MatNi1998]M. Matsumoto and T. Nishimura, Mersenne Twister: a 623-dimensionally equidistributed uniform pseudorandom number generator, ACM Transactions on Modeling and Computer Simulation 8, 1998, pages 3--30.

The unit tests for the mt19937 package verify that the output of the Go implementation coincides with the output of the references implementation.

The mt19937 generator is slightly slower than the Go default PRNG. A speed comparison can be performed using the following command:

go test -bench=.

On my (64bit) system I get the following results:

method time per call thoughput
MT19937.Uint64 14.9 ns/op 537.48 MB/s
MT19937.Int63 15.0 ns/op 533.54 MB/s
builtin Int63 11.9 ns/op 674.07 MB/s

This shows that, on my system, a call to the Int63() method of the built-in PRNG takes about 80% of the time that MT19937.Int63() takes.

You can’t perform that action at this time.