Is Witsenhausens counterexample a relevant toy?

This paper answers a question raised by Doyle on the relevance of the Witsenhausen counterexample as a toy decentralized control problem. The question has two sides, the first of which focuses on the lack of an external channel in the counterexample. Using existing results, we argue that the core difficulty in the counterexample is retained even in the presence of such a channel. The second side questions the LQG formulation of the counterexample. We consider alternative formulations and show that the understanding developed for the LQG case guides the investigation for these other cases as well. Specifically, we consider 1) a variation on the original counterexample with general, but bounded, noise distributions, and 2) an adversarial extension with bounded disturbance and quadratic costs. For each of these formulations, we show that quantization-based nonlinear strategies outperform linear strategies by an arbitrarily large factor. Further, these nonlinear strategies also perform within a constant factor of the optimal, uniformly over all possible parameter choices (for fixed noise distributions in the Bayesian case). Fortuitously, the assumption of bounded noise results in a significant simplification of proofs as compared to those for the LQG formulation. Therefore, the results in this paper are also of pedagogical interest.

💡 Research Summary

The paper addresses two criticisms raised by Doyle concerning the relevance of the Witsenhausen counterexample as a toy problem in decentralized control. The first criticism points out the absence of an external communication channel, while the second questions the reliance on a linear‑quadratic‑Gaussian (LQG) formulation. By leveraging existing results and extending the analysis to new settings, the authors demonstrate that the core difficulty of the counterexample persists even when an external channel is introduced, and that the insights gained from the LQG case remain valuable for more general formulations.

In the first part, the authors revisit the signaling structure that underlies the Witsenhausen problem. They argue that if an external channel has limited capacity (e.g., a few bits per time step), the optimal use of that channel does not eliminate the need for nonlinear signaling. In fact, a quantized signal transmitted directly by the first controller can be more cost‑effective than trying to encode the same information through a constrained external link. Consequently, the presence of an external channel does not fundamentally change the problem’s non‑convex nature; the advantage of nonlinear strategies over linear ones remains.

The second part tackles the LQG assumption by introducing two alternative models. The first model replaces the Gaussian noise with an arbitrary but bounded distribution. By aligning the quantization intervals with the support of the noise, the authors construct a family of quantization‑based nonlinear policies that can make the cost of any linear policy arbitrarily larger. Moreover, these policies achieve a constant‑factor approximation of the optimal cost uniformly over all system parameters, provided the noise distribution is fixed.

The second model adds an adversarial disturbance with a known bound, while retaining quadratic costs. In a worst‑case (min‑max) setting, the same quantization‑based strategies are shown to suppress the disturbance effectively, causing any linear policy to incur unbounded cost. Yet the nonlinear policies stay within a constant factor of the optimal robust cost, demonstrating that the “quantize‑then‑act” principle is robust to hostile environments as well.

A key technical contribution is the simplification of proofs that arises from the bounded‑noise assumption. Because the noise support is compact, expectations reduce to finite sums over quantization cells rather than integrals over unbounded Gaussian tails. This makes the analysis more transparent and pedagogically attractive, allowing the core ideas of signaling and quantization to be taught without the heavy machinery of Gaussian measure theory.

The paper concludes that the Witsenhausen counterexample is far from a trivial toy. Its essential feature—nonlinear signaling that outperforms any linear scheme—survives under realistic extensions such as limited‑capacity external links, non‑Gaussian bounded disturbances, and even adversarial perturbations. The quantization‑based nonlinear policies not only provide provable performance guarantees (constant‑factor optimality) but also lend themselves to intuitive explanations and simpler proofs. Hence, the counterexample serves both as a deep research benchmark and as an excellent educational vehicle for illustrating the subtleties of decentralized control, signaling, and robust design.