Mrg32k3a Class

public sealed class Mrg32k3a : RandomSource<ScalarGenerator32<Mrg32k3aEngine>>, 
	IRandomSourceFactory<Mrg32k3a>, IAdvanceable, IJumpable

Public NotInheritable Class Mrg32k3a
	Inherits RandomSource(Of ScalarGenerator32(Of Mrg32k3aEngine))
	Implements IRandomSourceFactory(Of Mrg32k3a), IAdvanceable, 
	IJumpable

public ref class Mrg32k3a sealed : public RandomSource<ScalarGenerator32<Mrg32k3aEngine>>, 
	IRandomSourceFactory<Mrg32k3a^>, IAdvanceable, IJumpable

[<SealedAttribute>]
type Mrg32k3a = 
    class
        inherit RandomSource<ScalarGenerator32<Mrg32k3aEngine>>
        interface IRandomSourceFactory<Mrg32k3a>
        interface IAdvanceable
        interface IJumpable
    end

Inheritance

Object  →  RandomSource<ScalarGenerator32<Mrg32k3aEngine>>  →  Mrg32k3a

Implements

IAdvanceable, IJumpable, IRandomSourceFactory<Mrg32k3a>

Use the Mrg32k3a class to represent a pseudo-random number generator that uses Pierre L'Ecuyer's MRG32k3a combined multiple recursive generator. MRG32k3a has a six-word state split across two third-order recurrences modulo 4294967087 and 4294944443, and combines the two recurrence values to produce 32-bit output.

This generator is especially useful for simulation and parallel Monte Carlo work because the original RngStreams design partitions the sequence into long, non-overlapping streams and substreams. The classic hierarchy uses streams spaced 2¹²⁷ outputs apart and substreams spaced 2⁷⁶ outputs apart. In Mrg32k3a, LongJump() moves to the next classic stream and Jump() moves to the next classic substream.

When Standard is used, StreamAddress is interpreted as the classic RngStreams address [stream, substream]. The first segment is applied as a 2¹²⁷ stream jump and the second segment is applied as a 2⁷⁶ substream jump. Addresses deeper than two segments are rejected for the standard profile. The default profile uses the general Numerics.NET stream derivation rules unless no stream address is supplied, in which case compatible MRG32k3a state initialization is used.

SimOpt three-level stream addressing is also available through JumpSimOptStream(UInt64), JumpSimOptSubstream(UInt64), and JumpSimOptSubsubstream(UInt64). These methods use jumps of 2¹⁴¹, 2⁹⁴, and 2⁴⁷ outputs respectively, but are explicit operations and are not used by Standard.

The generator supports fixed stream jumps, state persistence, copying, and direct six-word state initialization. It is intended for reproducible simulation work, not for cryptographic use.

Reference: Pierre L'Ecuyer, "Good Parameters and Implementations for Combined Multiple Recursive Random Number Generators," 1999.

Reference