| paper | d4-conf-wine-2021-0027 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
other→P discrete result → its continuous mirror
Answer: Both \(L_a\) and \(L_b\) exist for every permitted input. Computing both to additive error \(2^{-k}\) is possible in deterministic time polynomial in \(L+k\), hence polynomial in \(m,\tau,L,k\) on the specified domain \(m=\tau=2\). · discrete side: other · adjudication: proved
full proof & adversarial review →
Theorem 3.2
statement extracted from the paper’s text layer
Given \(d_a\ge d_b>0\), take the population limit of bipartite markets with arrival rates \(Nd_a,Nd_b\), compatibility probability \(1/N\), and unit-rate exponential departures. Under Patient2, determine or bound the limiting unmatched-mass fractions \(L_a,L_b\), including whether the bounds of Theorem 3.2 hold.
A distribution of mass over the two sides and their evolving local compatibility states, with compatibility scaled to keep each agent's expected neighborhood finite. Arrivals and departures evolve that distribution; Patient2 matches critical mass along existing compatibility links when possible, and the objective is the long-run expected unmatched fraction on each side.
The sparse limit must retain evolving local neighborhoods and their correlations, potentially producing a measure-valued state with no specified finite input or complexity parameters.
fatal: False
The mirror covers the Patient2 loss guarantee in Theorem 3.2; it leaves the Greedy, 1-sided, and optimal-algorithm results unaddressed.
The strongest mirror I believe is the large-population loss of the paper’s Patient2 policy. Consider a labor market with many workers and positions arriving over time, grouped into two ex-ante types by side. Agents within each type share the same arrival process, departure hazard, and compatibility law. Let each agent carry mass \(1/N\). In the \(N\)-agent approximation, arrivals occur at rates \(N d_a\) and \(N d_b\), lifetimes are exponential with rate \(1\), and each opposite-side pair is compatible with probability \(1/N\). Thus arrival mass per unit time is \(d_a,d_b\), while an agent has only a finite expected number of compatible neighbors as \(N\) grows. The population limit must retain those sparse random neighborhoods and their evolution.
A continuous version has instance \(d_a\ge d_b>0\), an initially empty market, and the limiting population process of that family of markets. Its state is a distribution of mass across sides and current local compatibility states. Its decision variable is matching flow along existing compatibility edges, limited by the available mass at each endpoint. Under the patient policy, critical mass is matched to compatible opposite-side mass when possible; unmatched critical mass departs. The objective is the long-run expected unmatched fraction on each side. The problem is to determine or bound the limiting loss vector \((L_a,L_b)\) as \(N,T\to\infty\). Time and compatibility uncertainty remain in the model; the population is what becomes continuous.
My anchor is Theorem 3.2, proved in this paper. It states that for \(d_a\ge d_b\), as \(\lambda_a,\lambda_b,T\to\infty\), the Patient2 losses satisfy
\[
L_a(\mathrm{Patient2})\le \frac{d_a-d_b}{d_a}+\frac{\log(d_b+3)}{d_a},
\qquad
L_b(\mathrm{Patient2})\le
e^{-\max\{d_a-d_b,\;d_a/(1+d_b)\}}.
\]
Consequently,
\[
L(\mathrm{Patient2})\le
\frac{d_a-d_b}{d_a+d_b}
+\frac{\log(d_b+3)}{d_a+d_b}
+e^{-\max\{d_a-d_b,\;d_a/(1+d_b)\}}.
\]
In the balanced case \(d_a=d_b\), it further gives \(L(\mathrm{Patient2})\le 2e^{-C d_a}\) for some constant \(C\). The mirror asks for this same performance guarantee in the continuous sparse-population process. I expect that policy-evaluation problem to be tractable by stationary-state and concentration arguments. The paper’s large-market analysis already supplies much of the answer; the exact limiting loss and extensions to heterogeneous compatibility laws are natural follow-up questions.
This is a credible population mirror because the paper itself studies the regime of many agents per side, with only two ex-ante types and fixed sparse compatibility density. A plausible regime is a large, standardized labor or freight market in which many agents share the same role and operating conditions. Arrival time and realized neighbors affect an agent’s current state, so the continuum model must track local compatibility states as they evolve; treating all mass as freely interchangeable would change the problem. High multiplicity here comes from many exchangeable agents per role, not from making time or outcomes continuous.
One qualification is central: this is an algorithmic performance theorem, not a theorem classifying a problem as polynomial-time solvable, NP-hard, or similar. Under a strict requirement for complexity-class anchors, this paper provides no compliant anchor. Theorem 3.2 is the strongest positive anchor available because it is a named result about a matching algorithm’s performance. The weakest point in the mirror is preserving the paper’s sparse, individual-level compatibility structure in an atomless mass model: a simple fractional matching flow would erase the local randomness that drives the bound. The paper’s own high-arrival-rate analysis makes the proposed regime plausible, while also meaning this mirror is chiefly a continuous formulation and generalization of an asymptotic result the authors already prove.
Theorem 3.2 is the proponent’s strongest anchor, and the proposed scaling is sensible. In fact, setting \(\lambda_a=Nd_a\), \(\lambda_b=Nd_b\), and \(p=1/N\) gives exactly the paper’s large market regime: \(d_i=\lambda_i p\) stays fixed as arrival rates grow. So the specific Patient2 guarantee they want in the limit is already what the paper bounds. An exact limiting loss or heterogeneous compatibility laws would be further analytic results, but they are not new complexity questions.
The stronger version would be to optimize over matching policies in a continuous population model. That could be a worthwhile research problem, but it changes the target. The paper compares two specified policies and derives stochastic performance bounds; it does not ask how hard it is to compute an optimal policy, or give a finite input and output whose exact, approximate, or parameterized computation could be classified. A continuous-time policy optimization problem may be interesting, but it moves toward stochastic control—the kind of analytic question ChoCo distinguishes from its computational focus.
There is also a modeling cost to making the population continuous here. With \(p=1/N\), a type-level compatibility kernel sees vanishing pair density, even though agents retain a finite expected number of neighbors. Rescaling compatibility to preserve matching capacity leads to a sparse local limit: the state must retain distributions of evolving neighborhoods and their correlations. That is a coherent model, and the authors might recognize it as a natural extension. But a simple flow between masses of types would discard the structure responsible for Patient2’s bound; retaining that structure requires a different, generally infinite-dimensional computational problem whose input and complexity parameters the paper does not specify.
I would not argue that high multiplicity is implausible here: the paper’s own scaling supports it. The negative case is narrower. The proposed mirror either restates the paper’s existing asymptotic guarantee or becomes a new sparse stochastic-control problem, rather than a continuous mirror of a computational-complexity result. The weak point in that case is that the latter problem could still prove valuable if ChoCo expands to include computational questions about dynamic policies.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.