1978Nuclear Science and EngineeringRequires access

Unified Definition of a Class of Monte Carlo Estimators

Iván Lux

Open publisher page 4 citations

Abstract

A unified definition of a wide class of Monte Carlo reaction rate estimators is presented, since most commonly used estimators belong to that class. The definition is given through an integral transformation of an arbitrary estimator of the class. Since the transformation contains an arbitrary function, in principle an infinite number of new estimators can be defined on the basis of one known estimator. It is shown that the most common estimators belonging to the class, such as the track-length and expectation estimators, are special cases of transformation, corresponding to the simplest transformation kernels when transforming the usual collision estimator.A pair of new estimators is defined and their variances are compared to the variance of the expectation estimator. One of the new estimators, called the trexpectation estimator, seems to be appropriate for flux-integral estimation in moderator regions. The other one, which uses an intermediate estimation of the final result and is therefore called the self-improving estimator, always yields a lower variance than the expectation estimator. As is shown, this estimator approximates well to possibly the best estimator of the class.Numerical results are presented for the simplest geometries, and these results indicate that for absorbers that are not too strong, in practical cases the standard deviation of the self-improving estimator is less than that of the expectation estimator by more than 10%. The experiments also suggest that the self-improving estimator is always superior to the track-length estimator as well, i.e., that it is the best of all known estimators belonging to the class.In the Appendices, for simplified cases, approximate conditions are given for which the trexpectation and track-length estimators show a higher efficiency than the expectation estimator.

About this research paper

What this paper is about

A unified definition of a wide class of Monte Carlo reaction rate estimators is presented, since most commonly used estimators belong to that class. The definition is given through an integral transformation of an arbitrary estimator of the class. Since the transformation contains an arbitrary function, in principle an infinite number of new estimators can be defined on the basis of one known estimator. It is shown that the most common estimators belonging to the class, such as the track-length and expectation estimators, are special cases of transformation, corresponding to the simplest transformation kernels when transforming the usual collision estimator.A pair of new estimators is defined and their variances are compared to the variance of the expectation estimator. One of the new estimators, called the trexpectation estimator, seems to be appropriate for flux-integral estimation in moderator regions. The other one, which uses an intermediate estimation of the final result and is therefore called the self-improving estimator, always yields a lower variance than the expectation estimator. As is shown, this estimator approximates well to possibly the best estimator of the class.Numerical results are presented for the simplest geometries, and these results indicate that for absorbers that are not too strong, in practical cases the standard deviation of the self-improving estimator is less than that of the expectation estimator by more than 10%. The experiments also suggest that the self-improving estimator is always superior to the track-length estimator as well, i.e., that it is the best of all known estimators belonging to the class.In the Appendices, for simplified cases, approximate conditions are given for which the trexpectation and track-length estimators show a higher efficiency than the expectation estimator.

Why it matters

OpenAlex reports 4 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

A unified definition of a wide class of Monte Carlo reaction rate estimators is presented, since most commonly used estimators belong to that class. The definition is given through an integral transformation of an arbitrary estimator of the class. Since the transformation contains an arbitrary function, in principle an infinite number of new estimators can be defined on the basis of one known estimator. It is shown that the most common estimators belonging to the class, such as the track-length and expectation estimators, are special cases of transformation, corresponding to the simplest transformation kernels when transforming the usual collision estimator.A pair of new estimators is defined and their variances are compared to the variance of the expectation estimator. One of the new estimators, called the trexpectation estimator, seems to be appropriate for flux-integral estimation in moderator regions. The other one, which uses an intermediate estimation of the final result and is therefore called the self-improving estimator, always yields a lower variance than the expectation estimator. As is shown, this estimator approximates well to possibly the best estimator of the class.Numerical results are presented for the simplest geometries, and these results indicate that for absorbers that are not too strong, in practical cases the standard deviation of the self-improving estimator is less than that of the expectation estimator by more than 10%. The experiments also suggest that the self-improving estimator is always superior to the track-length estimator as well, i.e., that it is the best of all known estimators belonging to the class.In the Appendices, for simplified cases, approximate conditions are given for which the trexpectation and track-length estimators show a higher efficiency than the expectation estimator.

Key concepts: Estimator, Invariant estimator, Minimum-variance unbiased estimator, Efficient estimator, Stein's unbiased risk estimate, Mathematics, Trimmed estimator, Bias of an estimator

Related papers

Back to paper searchBrowse research topicsOriginal source
Unified Definition of a Class of Monte Carlo Estimators — Research Paper | ScholarLens