Spectrum management is of central importance in the physical layer of emerging heterogeneous wireless networks.  As cell sizes continue to shrink, traffic variations over time in a given cell become significant.  Adapting spectrum allocation across cells to their traffic conditions is both rewarding and challenging.  In this talk, we describe a framework for modeling the topology of a multi-cell system, the dynamic traffic, and the spectrum allocation in a relatively slow timescale.  We formulate optimization problems that can be solved efficiently to yield spectrum allocation that minimizes the average packet sojourn time.  Simulations suggest significant advantages in terms of both delay and throughput compared to optimized static spectrum allocation schemes.