Extending the alias Monte Carlo sampling method to general distributions

Conference · Mon Jan 07 04:00:00 EST 1991

OSTI ID:6023539

Edwards, A L; Rathkopf, J A ^[1]; Smidt, R K ^[2]

---

1. Lawrence Livermore National Lab., CA (USA)
2. California Polytechnic State Univ., San Luis Obispo, CA (USA)

The alias method is a Monte Carlo sampling technique that offers significant advantages over more traditional methods. It equals the accuracy of table lookup and the speed of equal probable bins. The original formulation of this method sampled from discrete distributions and was easily extended to histogram distributions. We have extended the method further to applications more germane to Monte Carlo particle transport codes: continuous distributions. This paper presents the alias method as originally derived and our extensions to simple continuous distributions represented by piecewise linear functions. We also present a method to interpolate accurately between distributions tabulated at points other than the point of interest. We present timing studies that demonstrate the method's increased efficiency over table lookup and show further speedup achieved through vectorization. 6 refs., 12 figs., 2 tabs.

• Research Organization: Lawrence Livermore National Lab., CA (USA)
• Sponsoring Organization: DOE/DP
• DOE Contract Number: W-7405-ENG-48
• OSTI ID: 6023539
• Report Number(s): UCRL-JC-104791; CONF-910414--16; ON: DE91007619
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
DISTRIBUTION
INTERPOLATION
MONTE CARLO METHOD
NUMERICAL SOLUTION
SAMPLING