Equivalence of equilibrium markets and optimal central control

Prove that, under the preconditions (i) agents act competitively (price-taking), (ii) a market equilibrium exists, and (iii) the systems performance index can be written as a linear combination of local contributions, multi-agent equilibrium markets produce optimal decentralized solutions to distributed resource allocation problems that match the quality—measured by the given overall systems performance index—of solutions produced by the optimal central multi-input, multi-output systems controller with access to all relevant local data (i.e., the total system state and control vectors).

Background

The paper compares multi-agent market mechanisms with conventional centralized control for building climate regulation, developing both improved control schemes and a decentralized market design (MARKET-B) that achieves performance comparable to centralized control without relying on global information. This leads to the qualitative conclusion that local data combined with market communication can yield global control.

Building on these results, the authors propose a broader claim beyond the specific application: that equilibrium markets can provide optimal decentralized solutions equivalent in quality to those from optimal central controllers, given specific preconditions. They explicitly frame this broader claim as a conjecture and call for a rigorous mathematical proof grounded in dynamic systems and optimal control theory.

References

We believe that this indeed can be done in a formal way, and as a conjecture we state the following general result. Multi-agent equilibrium markets yield optimal decentralized solutions to distributed resource allocation problems that are of the same quality, in terms of a given overall systems performance index, as solutions given by the optimal central (multi-input, multi-output) systems controller that has access to all relevant local data. These local data involve the total system state and control vectors (in building control these are the vectors formed from the difference between the actual and setpoint temperatures for each office, and the cooling power for each office, respectively). Preconditions for this to be true are: (i) agents act competitively; (ii) the equilibrium exists; (iii) the systems performance index can be written as a linear combination of local contributions (implying diagonality of certain matrices; if not, agents are not independent).

Decentralized Markets versus Central Control: A Comparative Study  (1106.0223 - Akkermans et al., 2011) in Section 10 (Conclusions), final paragraph