RSIC COMPUTER CODE PSR-066

1. NAME AND TITLE

RNGP: Random Number Generator Package.

RNGP includes the following programs:

GEORGE: The Statistical Wizard. (Statistical Analysis Based on MIRAN).

MIRAN: Machine Independent Uniform Random Number Generator.

SFLRAF: IBM 360 Random Number Generator.

FLTRNF: CDC 6600 Random Number Generator.

2. CONTRIBUTORS

Memory Horatio Turner, III, Irving, Texas.

Science Applications, Inc., La Jolla, California.

Oak Ridge National Laboratory, Oak Ridge, Tennessee.

3. CODE LANGUAGE AND COMPUTER

Fortran IV and Assembler Language; IBM 360/370.

4. NATURE OF PROBLEM SOLVED

GEORGE: The Statistical Wizard is designed to provide the researcher with an easy method for generation of random data from 15 different statistical distributions and can also provide simple descriptive statistics on the generated data. The user can also submit his own data to GEORGE for analysis. The simple statistics which can be determined are mean, standard deviation, coefficient of variation, variance, sample mean standard deviation, skewness, and kurtosis. In addition, a histogram, as well as a tabular depiction of the data, may be obtained.

MIRAN: This program is a means of generating uniform random numbers independently of the computer for which it was designed.

5. METHOD OF SOLUTION

GEORGE: Standard statistical formulas are used for the generation of the random numbers based on the Uniform Random Number Generator MIRAN.

MIRAN: This program uses an explicit multiple precision representation. The integers and operations involved in the Random Number Generator (RNG) algorithm are separated into component parts in such a way that all operations are kept within a single computer word and no overflows into the sign bit are made, thus avoiding the sign-clearing problem. A RNG which is machine-independent greatly facilitates the exchange and checkout of Monte Carlo programs between different computers.

6. RESTRICTIONS OR LIMITATIONS

GEORGE: The maximum number of datapoints which can be generated or analyzed is 200. The ERRSET routine of the Fortran compiler should be utilized if at all possible to avoid arithmetic underflows and overflows, although it is not required.

MIRAN: This is a much slower method of producing random numbers. However, since it can generate several thousand random numbers in one second, the time difference will not be noticed in most Monte Carlo applications.

7. TYPICAL RUNNING TIME

GEORGE will generate 200 random normal numbers in about 20 seconds.

8. COMPUTER HARDWARE REQUIREMENTS

RNGP is operable on the IBM 360/370 computers.

9. COMPUTER SOFTWARE REQUIREMENTS

Fortran IV and Assembler Language compilers are required.

10. REFERENCES

M. H. Turner, III, "Abstract of George."

Techniques for Efficient Monte Carlo Simulation: Volume II -- Random Number Generation for Selected Probability Distributions; Appendix B: MIRAN -- A Machine Independent Package for Generating Uniform Random Numbers, SAI-72-590-LJ Vol. II.

11. CONTENTS OF CODE PACKAGE

Included are the referenced documents and one (1.2MB) DOS diskette which contains the source code and sample problem input and output for GEORGE.

12. DATE OF ABSTRACT

August 1974; revised and updated February 1984.