This paper has so far concentrated on algorithms for posterior marginals. Most algorithms presented in this paper extend readily to expected utility by simple inclusion of the utility functions [10].
Calculation of lower and upper variances is more complex than expected utility because variances are non-linear functionals of the distributions. We must reduce calculation of variances to an iterative calculation of expected utilities [37, Theorem G2,] in order to solve this problem (the method is presented in [10]).
Fri May 30 15:55:18 EDT 1997