Weighted Fisher informations, their derivation and use

Fisher information has been used to derive many laws of physics, including its differential equations. In many of these Fisher-based derivations the information in the time measurement is required to be negative. Yet by its expression in the standard derivation of the Cramer-Rao (CR) inequality, a negative Fisher seems impossible. To the contrary, we show that in fact the standard CR derivation allows the Fisher to be meaningfully regarded as either negative or positive. The mathematics allow it, and the choice to be made depends upon the physical parameter whose information level is sought. For example, the rules of special relativity require an imaginary time parameter ict, , with c the speed of light and t the time. The squared time parameter is then negative, requiring (as shown) the Fisher information to be negative as well. Further generalizations of the CR derivation are also made, which allow the estimation of general parameters using arbitrarily weighted mean-squared error criteria. The weights are found to define a family of correspondingly weighted Fisher informations. Finally, a condition is found for achieving efficient estimation in the presence of any weight function.