Deterministic Analogue of Border's Theorem

• The paper shows that deterministic DSIC implementability in a two-bidder auction is fully characterized by a single functional inequality, replacing an uncountable family of constraints.
• It employs critical-value structures and quantile coordinate transformations to rigorously define and verify interim allocation rules.
• The work highlights a clear separation between deterministic DSIC and randomized BIC modalities, offering new insights for mechanism design in auctions.

The deterministic analogue of Border’s theorem establishes a precise and remarkably succinct characterization of when a pair of interim allocation rules in a two-bidder single-item auction is implementable by a deterministic dominant-strategy incentive compatible (DSIC) mechanism. Whereas the original (randomized) Border’s theorem relies on an uncountable family of linear inequalities to specify implementability for randomized Bayesian incentive-compatible (@@@@1@@@@) auctions, the deterministic restriction for DSIC mechanisms yields a single functional inequality that is both necessary and sufficient. This result, developed by Liu et al. (2025), demonstrates not only a structural elegance but also a sharp separation between the expressiveness of randomized BIC and deterministic DSIC modalities in multi-agent mechanism design (Liu et al., 19 Dec 2025).

1. Formal Model and Key Definitions

The framework is a single-item auction with two buyers, indexed by $i=1,2$, each with independent private values $v_i \sim F_i$ drawn from atomless distributions. The mechanism selects, for each bid profile $(b_1, b_2)$, a deterministic allocation $x_i(b_1, b_2) \in\{0,1\}$ and payments $p_i(b_1, b_2)\ge 0$, with feasibility constraint $\sum_i x_i \le 1$.

A mechanism is DSIC if truth-telling is optimal for each participant, regardless of opponents' reports. Deterministic DSIC allocations for buyer $i$ admit a critical-value form: $$x_i(v_i, v_{-i}) = 1\quad \text{iff}\quad v_i \ge c_i(v_{-i}),$$ for some threshold function $c_i$ dependent only on opponents' bids. The interim allocation for buyer $i$ at value $v_i$ is given by $x_i(v_i) = \mathbb{E}_{v_{-i}}[x_i(v_i, v_{-i})]$. Transforming to quantile coordinates $q_i=F_i(v_i)\in[0,1]$ yields interim rules $$\hat x_i(q_i) = \mathbb{E}_{q_{-i}}[x_i(F_i^{-1}(q_i), F_{-i}^{-1}(q_{-i}))]$$.

A pair $(\hat x_1, \hat x_2)$ is deterministic DSIC implementable if there exists a deterministic DSIC mechanism that induces precisely these interim allocations.

2. The Classical Randomized Border’s Theorem

Border’s theorem provides a characterization for (possibly randomized) BIC implementability in terms of linear constraints. In quantile terms, for all $(q_1, q_2)\in[0,1]^2$,

$$\int_{t_1=q_1}^1 \hat x_1(t_1) \,dt_1 + \int_{t_2=q_2}^1 \hat x_2(t_2) \,dt_2 \le 1 - q_1 q_2.$$

This yields an infinite, uncountable system of constraints parameterized by threshold pairs $(q_1, q_2)$. These conditions are tight for randomized BIC implementability in the two-bidder case (Liu et al., 19 Dec 2025).

3. Deterministic DSIC Analogue: Single Functional Inequality

In the deterministic DSIC setting for two bidders, the aforementioned infinite constraint system collapses to a single, order-type functional condition:

Theorem (Deterministic DSIC $\Leftrightarrow$ Single-Constraint):

For strictly increasing interim allocation curves $\hat x_1, \hat x_2 : [0,1] \to [0,1]$, deterministic DSIC implementability is equivalent to

$$\forall q \in [0,1]:\quad \hat x_2\bigl(\hat x_1(q)\bigr)\;\le\;q.$$

If the auction always allocates the item (i.e., $\hat x_1(q) + \hat x_2(q) = 1$ for all $q$), the condition tightens to equality: $\forall q: \;\; \hat x_2(\hat x_1(q)) = q.$ Strictly increasing can be weakened to weakly increasing curves with measure-zero ties resolved appropriately. This single-constraint result fully characterizes deterministic DSIC implementability for interim rules in the two-bidder setting (Liu et al., 19 Dec 2025).

4. Structural Proof Sketch and Critical-Value Colorings

Deterministic DSIC allocations correspond to a three-coloring of the quantile square $[0,1]^2$, with each point $(q_1, q_2)$ colored by the winner ($0$: no sale; $1$: buyer 1; $2$: buyer 2). The critical-value structure dictates, for each fixed $q_2$, a threshold $c_1(q_2)$ such that buyer 1 wins precisely on $\{q_1 \ge c_1(q_2)\}$, and symmetrically for buyer 2.

The proof constructs a "hierarchical allocation" via rearrangement:

1. For buyer 1: Rearrangement moves all "1"-colored area in each vertical column to the bottom, so buyer 1 wins exactly on $\{q_2 \le \hat x_1(q_1)\}$, preserving $\hat x_1$.
2. For buyer 2: Similarly, leftward rearrangement in each row allots buyer 2 wins to $\{q_1 \le \hat x_2(q_2)\}$.
3. Feasibility necessitates that these regions do not overlap. For each $q_1$, the set $\{q_2 \le \hat x_1(q_1)\}$ must lie below $q_2 = \hat x_2^{-1}(q_1)$, which is precisely the single constraint (†).

If (†) is satisfied, the coloring yields a deterministic DSIC mechanism with the desired interim rules (Liu et al., 19 Dec 2025).

5. Separation of Deterministic DSIC and Randomized BIC

The reduction to a single-constraint enables explicit demonstration of separation between deterministic DSIC and randomized BIC implementation power. Liu et al. construct interim-allocation rules $(\hat x_1, \hat x_2)$ which satisfy randomized-BIC (Border’s) constraints but violate the deterministic DSIC functional inequality—thereby implementable by deterministic BIC but not by any deterministic DSIC mechanism.

Example: With probability $p$, run VCG (highest-value wins); with probability $1-p$, allocate for free to buyer 1. In quantile coordinates,

$$\hat x_1(q) = (1-p) + 2p\,q, \qquad \hat x_2(q) = 2p\,q - p,$$

with $\hat x_1(q) + \hat x_2(q) = 1$. However,

$\hat x_2(\hat x_1(q)) = (4p^2) q + p(1-3p).$

Choosing $p=3/4$, $\hat x_2(\hat x_1(q)) > q$ for some $q$, violating the deterministic DSIC constraint, although the randomized BIC constraints are met (Liu et al., 19 Dec 2025).

6. Implications and Concluding Observations

The deterministic analogue of Border’s theorem in the two-bidder case yields a compact and tight order-type condition, replacing the infinite Border family with a single functional inequality on interim-allocation rules. This result uncovers surprising limitations and distinct separation phenomena for deterministic dominant-strategy implementation. The tightness of the result and explicit counterexamples underscore intrinsic differences between deterministic and randomized implementation paradigms in mechanism design, even in simple auction settings (Liu et al., 19 Dec 2025).