To handle such problems, we have developed a mechanism for temporal reasoning. The essential functions of the temporal reasoning are to deduce and maintain the causal temporal constraints in hypotheses and to support the possibility of nodes having different values over different time intervals. In the example, the temporal reasoning must allow the LAP to have a value of high for some period after the MI but also be normal by the time of the examination. In addition, it must enforce the restriction that the time interval of pulmonary congestion is after the MI if it is to be explained by the MI. Thus, the temporal reasoning must support the logical constraints of the causal theory allowing some situations that would be ruled out in a purely probabilistic network and disallowing some situations that would satisfy such a network.
Besides the logical constraints provided by the temporal model, the temporal characteristics provide an opportunity to adjust the probabilities when there are different causal situations with different probabilities. For example, if the depression of LV function is sometimes transient but often continues for days, it is appropriate to use a different probability for a causal link involving LV dysfunction a day after an MI than for a link involving LV dysfunction during an MI.