A performance criterion for information theoretic data analysis

✍ Scribed by J.L. Kinsey; R.D. Levine

• Publisher: Elsevier Science
• Year: 1979
• Tongue: English
• Weight: 357 KB
• Volume: 65
• Category: Article
• ISSN: 0009-2614
• DOI: 10.1016/0009-2614(79)80261-4

✦ Synopsis

A practical criterion which determines the number of slgnificznt parameters (the "constraints") \ hich need be cmplo) ed in an information theoreticdata analysis is derked and applied.

1. Xntroduction

Implicit in the development of information theory [t-4] up to this point has been the assumption that the data given are known exactly_ In practice this is aimost never the case. Rather, in addition to the actual fluctuations there are the many sources of error typical of a real experiment *_ The characteristics of such noise are often unknown or at best poorly character-ized_ Thus the experimental&t typically reports data with a somewhat subjective error limit. The purpose of this letter is to discuss the practical issue of an information theoretic analysis which explicitIy recognizes the existence of scatter in the data_ A statistician would advise that under such circumstances confidence intervals for the estimated parameters be computed (see refs. [5.6] for an actual example)_ If at the desired level of confidence zero is a possibte vaIue for any of the parameters then that param-* Work supsorted by the A'ational Science Foundation, the Air Force Office of Scientific Research and the U.S.-Israel Binational Science Foundation, Jerusalem. Israel. * Similarly, computations. xbhether classical or quanta1 ha=. in prac:ice, an inherenr error term arising mostly from truecation (whether of the number of trajectories or of the size of the basis set)_ Hence our recommendations for the firting of experimental data apply equally well 10 computstional results.