In this paper, a dynamic system-optimal traffic assignment model is fo
rmulated as a convex control problem for a congested general network w
ith many origins and many destinations. Analytical and computational d
ifficulties caused by the nonconvexity of the previous models are elim
inated. The modeling of are traffic dynamics is improved to prohibit i
nstantaneous flow propagation on each are even though the concave exit
functions are still employed to represent the physical process of tra
ffic congestion. An economic interpretation of the optimality conditio
ns is given as a dynamic assignment principle which requires equilibra
tion of actual marginal costs on all the paths that are used. A numeri
cal example is also presented. (C) 1998 Elsevier Science Ltd. All righ
ts reserved.