A distributed binary detection problem with multimessage (greater-than
-or-equal-to 1 bit) communications is considered, wherein the nodes (s
ensors, decision makers (DMs)) of the system are organized in the form
of a tree with multiple root nodes. A numerical algorithm is develope
d for determining the optimal decision rule at each node assuming mono
tone cost functions imposed only on the root nodes. It is assumed that
the observations of each node are conditionally independent of those
of the other nodes. It is shown that the problem is equivalent to solv
ing a nonlinear optimal control problem, and the necessary conditions
of optimality using Bayes' risk as the optimization criterion are deri
ved. The optimal control approach provides an interpretation of certai
n functions of the co-state variables in terms of thresholds, and lead
s to a computationally efficient min-H algorithm to solve for the opti
mal decision rule at each node. The numerical algorithm provides a too
l to investigate the organizational issues of adaptation, structure, a
nd robustness.