d4-journals-jair-2003-0026

The strongest case I believe is a continuous population of participants in the paper’s supply chain market. A type is a complete role in the task network: a consumer’s desired goods and values, or a producer’s input recipe, output, and cost. A society is a distribution \(\mu\) over those types. In a large procurement market, millions of small suppliers and buyers could share a relatively small number \(\tau\) of such roles.

A producer type of mass \(\mu_\pi\) can have any mass \(x_\pi\in[0,\mu_\pi]\) active. If it is active, each unit of mass consumes its full input bundle and supplies its output; this preserves the paper’s production complementarities at the level of each infinitesimal participant. A consumer type can allocate mass \(y_{c,g}\) to goods, with \(\sum_g y_{c,g}\leq\mu_c\). Goods are measured in per-capita quantities, and prices and costs retain their per-unit meanings. The welfare benchmark is
\[ W(x,y)=\sum_{c,g}v_c(g)y_{c,g}-\sum_\pi \kappa_\pi x_\pi, \]
subject to mass balance for each good. Fractional aggregate activity means a fraction of a type participates; each participant still receives an integral bundle. The discretization bridge is \(n_t\) identical participants of type \(t\) corresponding to \(\mu_t=n_t/n\). How closely the continuous flow approximates that finite market is a further question, not an assumed identity.

My lead is Atomless SAMP-SB Quiescence, anchored on Theorem 10, proved in this paper (Appendix A.6). Theorem 10 says that SAMP-SB reaches a quasi-quiescent state after a number of bids by consumers and active producers bounded by a polynomial in the network size and the maximum consumer value. Quasi-quiescence means that no consumer or active producer would change its bids in response to the quotes it has received; Theorem 8 says the allocation and prices then stop changing. This is a bid-complexity result for the protocol. The paper separately notes in Observation 6 that an asynchronous run can contain exponentially many bids, so the mirror should count the mass of bid revisions up to quasi-quiescence, rather than treating every delayed or redundant message as meaningful progress.

The continuous problem is: given a finite acyclic goods-and-types network, binary rational type masses, costs and consumer values, and bid increments \(\delta_b,\delta_s\), characterize the worst-case aggregate SAMP-SB behavior over fair asynchronous schedules. Each auction receives mass-valued offers; at a tied marginal price it may split mass, and each infinitesimal agent follows the paper’s bidding policy. Ask whether every fair run reaches aggregate quasi-quiescence and bound
\[ U=\sum_{t\in T}\int_{A_t} r(a)\,da, \]
where \(A_t\) is the mass of type \(t\) and \(r(a)\) is that participant’s number of bid revisions. A solution is a terminal price and allocation profile, together with a bound on \(U\) and a certificate that no positive mass would revise its bids. The main parameters are the goods-and-type network size, support size \(\tau\), input bit length \(L\), and \(R=\max_{t,g}v_t(g)\). The expected direction is Class A when values and price increments are numerically bounded: the finite-type aggregate process looks capable of inheriting a finite event bound. Since Theorem 10’s bound uses numerical consumer value, a result polynomial in \(R\) may be pseudo-polynomial under binary encoding; the bit-complexity question remains open.

A useful second anchor is Theorem 14, also proved here (Appendix A.10). It says that for networks with no input complementarities—each producer has at most one input—the quiescent state of safe SAMP-SB is a \(\lambda\)-\(\delta\)-competitive equilibrium. Here the equilibrium conditions require nonnegative surplus, consumers to be within \(\delta_b\) of their best available surplus, producers to be within the input price tolerances \(\lambda\) and output increment \(\delta_s\) of their best feasible surplus, and goods to balance. The corresponding problem, Atomless Safe-SAMP-SB Equilibrium, asks whether the aggregate safe protocol has a finite terminal state and whether almost every participant in it satisfies those same surplus conditions, with aggregate goods balanced. I expect the guarantee to extend for one-input producer types, giving a restricted Class A result. It would establish an outcome guarantee as well as the lead question’s event bound.

The cited NP-completeness statement about finding a feasible supply chain appears in prose in this paper and refers to Walsh et al. (2003); it has no numbered result here, so I would not use it as an anchor. It does motivate a further question: does allowing type mass to split make feasible aggregate flow tractable, and what discrete-to-continuous rounding gap remains? Multi-input complementarities may still obstruct equilibrium, so continuization need not remove every source of difficulty.

The weak point is that the paper’s task network gives many agents distinct roles, and some supply chains may have little repeated structure. The proposed regime needs many interchangeable suppliers and buyers with identical recipes, values, and costs, plus capacities that scale with population. Aggregate mass-valued auctions also require a tie-splitting convention and a careful account of asynchronous updates. That makes the mirror a genuine extension of the market protocol rather than a direct restatement. Still, it keeps the paper’s central object—the decentralized formation of supply chains—and makes the participants, rather than the goods or time, the continuous population.