PROBLEMS OF ATOMIC SCIENCE AND TECHNOLOGY
Series: Nuclear and Reactor Constants

since 1971

Русский (РФ)

ISSN 2414-1038 (online)

Authors & Affiliations

Gay E.V.
Russian Federation State Scientific Centre — A.I. Leipunsky Institute of Physics and Power Engineering, Obninsk, Russia

Abstract

The onset of true PPP is considered, i.e., the effect that is not caused by misinterpretation of incompletely documented experimental data. PPP is shown to be accompanied by an unexpectedly low uncertainty associated with the least-squares method (LSM) estimate. This phenomenon is denoted the Small Uncertainty Paradox (SUP). Two measurements are used to demonstrate that the occurrence of PPP and SUP is conditioned by the presence of large values of the experimental systematic errors and by the significance deviation of their shape from the shape of the assumed regression (model) function. An investigation is made of the dependence of PPP and SUP on the value of the covariance for the experimental errors. SUP can occur in nuclear data evaluation without the manifestation of PPP. Covariance (correlation) limitation is proposed as a radical way of preventing these paradoxes. An estimate of the systematic error is considered as a possible benefit if these paradoxes are true. The work is based on classical LSM; the Bayesian approach is not used.

Key words
Small Uncertainty Paradox, least square method, the paradox of small errors, regression function, the value of covariance

Article Text (in Russian, PDF)

References

UDC 539.17

Problems of Atomic Science and Technology. Series: Nuclear Constants", issue 1-2, 2007