Distributed optimization typically requires agents to exchange full gradient vectors, which can be costly in bandwidth-constrained networks. A new paper on arXiv proposes an alternative: each agent communicates only the sign of its local cost gradient, reducing the message to a single bit per directed link per iteration.

The algorithm is designed for constrained optimization and is based on a Laplacian-gradient approach. The authors specifically mention applications to optimal scheduling and allocation, where agents must coordinate under resource limits while minimizing a shared objective.

This one-bit communication scheme could make distributed optimization far more practical in low-bandwidth or energy-limited settings. However, the abstract does not include convergence rates, numerical results, or comparisons with other methods; those details are presumably in the full paper. As the only source, this work stands alone in this digest, with no conflicting or corroborating studies to compare against.