Tight Mechanisms: Precise Performance Bounds

Updated 14 July 2026

Tight mechanisms are defined by achieving exact performance benchmarks—matching lower bounds, tight approximation ratios, or certified privacy parameters—in various models.
In auction design and budget-feasible procurement, for example, mechanisms have been developed that exactly close the gap with randomized 2-approximations and deterministic 3-approximations, proving no further improvement is possible.
In differential privacy and auditability, tightness guarantees precise privacy loss characterizations and exact audit indices, enabling rigorous accounting and empirical validation of mechanism performance.

Searching arXiv for recent and relevant papers on “tight mechanisms” across mechanism design and adjacent technical uses of “tight” mechanism terminology. Tight mechanisms are mechanisms or mechanism classes for which the relevant performance statement is exact: an approximation ratio matches a lower bound, an impossibility frontier is met, an “if and only if” characterization closes the feasible design space, or a numerical accountant returns certified privacy bounds whose residual gap is only discretization or truncation error. Across the literature, the phrase is used in several related but non-identical senses, most prominently in approximation-based mechanism design, auditability theory, and differential privacy (Gravin et al., 2019, Jin et al., 2018, Grigoryan et al., 2023, Koskela et al., 2020).

1. Core meanings of tightness

The term “tight” is not tied to a single model. In budget-feasible procurement, it means matching lower bounds against the relevant benchmark. In posted pricing and auction design, it means identifying the exact supremum of a revenue ratio. In facility location, it means determining the exact worst-case approximation ratio of a strategyproof mechanism. In auditability, it means exact characterization of the smallest coalition needed to expose any deviation. In differential privacy, it means exact privacy parameters or rigorously certified lower and upper bounds for the true privacy loss.

Domain	Tight quantity	Representative result
Budget-feasible procurement	Approximation ratio	Randomized ratio $2$; deterministic ratio $3$ against $\textsf{Fractional Knapsack}$ (Gravin et al., 2019)
Bayesian pricing	Exact revenue gap	$OPT/AP = C \approx 2.6202$ and $BUPP/UP = C$ (Jin et al., 2018)
Multi-unit auctions	Asymptotic revenue gap	Within-group $1+\Theta(1/\sqrt{k})$ ; cross-group $\Theta(\log k)$ (Jin et al., 2021)
Euclidean facility location	Exact approximation ratio	$\sqrt{2}$ for $1\le p\le 2$ , $2^{1-\frac{1}{p}}$ for $3$0 for coordinate-wise median (Chan et al., 7 Jun 2026)
Auditability	Auditability index	$3$1, $3$2 (Grigoryan et al., 2023)
Differential privacy	Exact privacy characterization	Exact iff conditions and tight zCDP formulas (Holohan et al., 2015, Harrison et al., 29 Oct 2025)

This suggests that “tight mechanism” is best understood as a closure notion: the mechanism attains the best provable bound available for the model and benchmark under study, or the paper proves that no better mechanism of the relevant type can exist.

2. Budget-feasible procurement with additive valuations

In "Optimal Budget-Feasible Mechanisms for Additive Valuations" (Gravin et al., 2019), the setting is a reverse auction with $3$3 agents, private costs $3$4, public buyer values $3$5, additive valuation $3$6, and a hard budget $3$7. The offline benchmark is knapsack, and the stronger fractional benchmark is

$3$8

The paper’s main randomized mechanism is a two-stage composition. A pruning step removes items with low value-per-cost ratio $3$9, producing a threshold $\textsf{Fractional Knapsack}$ 0 and surviving set $\textsf{Fractional Knapsack}$ 1 such that

$\textsf{Fractional Knapsack}$ 2

and

$\textsf{Fractional Knapsack}$ 3

The second stage uses randomized posted prices for the highest-value item $\textsf{Fractional Knapsack}$ 4 and the remaining set $\textsf{Fractional Knapsack}$ 5. The analysis hinges on the identity

$\textsf{Fractional Knapsack}$ 6

together with lower bounds on the expected contribution of $\textsf{Fractional Knapsack}$ 7 and each $\textsf{Fractional Knapsack}$ 8, yielding

$\textsf{Fractional Knapsack}$ 9

The resulting mechanism is a $OPT/AP = C \approx 2.6202$ 0-approximation against $OPT/AP = C \approx 2.6202$ 1, hence also against the integral knapsack benchmark, and the ratio is tight because there is already a matching lower bound of $OPT/AP = C \approx 2.6202$ 2 for randomized budget-feasible mechanisms even against the weaker integral benchmark (Gravin et al., 2019).

The same paper gives a deterministic mechanism with a $OPT/AP = C \approx 2.6202$ 3-approximation against the fractional optimum. It reuses pruning but replaces randomized prices with an adaptive posted-price rule that distinguishes the cases $OPT/AP = C \approx 2.6202$ 4, $OPT/AP = C \approx 2.6202$ 5, and $OPT/AP = C \approx 2.6202$ 6. In the intermediate range it offers

$OPT/AP = C \approx 2.6202$ 7

and adjusts prices for $OPT/AP = C \approx 2.6202$ 8 depending on whether $OPT/AP = C \approx 2.6202$ 9 accepts. The guarantee improves the prior deterministic bound of $BUPP/UP = C$ 0 to $BUPP/UP = C$ 1, and the paper also proves that no deterministic truthful IR budget-feasible mechanism can beat $BUPP/UP = C$ 2 against $BUPP/UP = C$ 3, even with only $BUPP/UP = C$ 4 items. In this model, both the randomized and deterministic frontiers are therefore closed (Gravin et al., 2019).

3. Revenue-optimality gaps and exact pricing frontiers

A second major use of tightness concerns exact revenue gaps between simple and optimal mechanisms. In "Tight Approximation Ratio of Anonymous Pricing" (Jin et al., 2018), the single-item Bayesian setting compares anonymous posted pricing

$BUPP/UP = C$ 5

with the Myerson auction. The paper proves

$BUPP/UP = C$ 6

where

$BUPP/UP = C$ 7

and

$BUPP/UP = C$ 8

The same constant governs the unit-demand single-buyer comparison between optimal item pricing and uniform pricing: $BUPP/UP = C$ 9 The lower bound is matched by explicit triangular-distribution instances, so the constant is not only an upper bound but the exact worst-case ratio (Jin et al., 2018).

The paper also identifies a continuous worst-case instance $1+\Theta(1/\sqrt{k})$ 0 with revenue

$1+\Theta(1/\sqrt{k})$ 1

which is decreasing in $1+\Theta(1/\sqrt{k})$ 2, so the worst case occurs at $1+\Theta(1/\sqrt{k})$ 3. As a corollary, the bound on Myerson relative to a second-price auction with anonymous reserve improves from $1+\Theta(1/\sqrt{k})$ 4 to $1+\Theta(1/\sqrt{k})$ 5, although tightness for that specific comparison is left open (Jin et al., 2018).

In the multi-unit setting, "Tight Revenue Gaps among Multi-Unit Mechanisms" (Jin et al., 2021) studies four basic mechanisms: Myerson auction, sequential posted-pricing, $1+\Theta(1/\sqrt{k})$ 6-th price auction with anonymous reserve, and anonymous pricing. The paper divides them into a discriminating group and an anonymous group. Within one group, the revenue gap is asymptotically $1+\Theta(1/\sqrt{k})$ 7. Across groups, any pair has asymptotically tight gap $1+\Theta(1/\sqrt{k})$ 8. For anonymous reserve versus anonymous pricing, the paper gives an exact representation,

$1+\Theta(1/\sqrt{k})$ 9

with

$\Theta(\log k)$ 0

Here tightness is asymptotic rather than finite-instance exactness, but the upper and lower bounds match at the correct order (Jin et al., 2021).

4. Strategyproof and bribeproof mechanisms

In Euclidean facility location under $\Theta(\log k)$ 1-norm social cost, "Strategyproof Mechanisms for Euclidean Facility Location Problems under $\Theta(\log k)$ 2-norm Social Cost" (Chan et al., 7 Jun 2026) resolves the exact deterministic frontier for the coordinate-wise median: $\Theta(\log k)$ 3 This confirms the Goel–Hann-Caruthers conjecture for $\Theta(\log k)$ 4. The paper then studies two randomized mechanisms. Uniformly rotated coordinate-wise median (URCM) satisfies

$\Theta(\log k)$ 5

so it strictly improves over the deterministic $\Theta(\log k)$ 6 bound for $\Theta(\log k)$ 7, gives $\Theta(\log k)$ 8 at $\Theta(\log k)$ 9, and cannot improve over CM for $\sqrt{2}$ 0. Centroid random dictatorship (CRD) improves over CM and URCM for every finite $\sqrt{2}$ 1, with tight endpoint bounds at $\sqrt{2}$ 2 and $\sqrt{2}$ 3, and at $\sqrt{2}$ 4 for every even $\sqrt{2}$ 5 (Chan et al., 7 Jun 2026).

A different frontier appears in bribeproofness. In "Bribeproof mechanisms for two-values domains" (Mihalák et al., 2015), each agent’s type lies in $\sqrt{2}$ 6, and the central construction is the $\sqrt{2}$ 7-linear payment rule

$\sqrt{2}$ 8

For $\sqrt{2}$ 9, bribeproofness is equivalent to the bounded-influence condition

$1\le p\le 2$ 0

In binary allocation problems, this becomes monotonicity plus non-bossiness, and the $1\le p\le 2$ 1-linear mechanism is then strongly bribeproof. The path-auction results are especially rigid: for the path auction on general networks, the $1\le p\le 2$ 2-linear mechanism is the only bribeproof mechanism, while on three-values domains no bribeproof mechanism exists (Mihalák et al., 2015).

These results show two distinct forms of tightness. Facility location closes an approximation-ratio question for a canonical strategyproof rule, whereas bribeproofness closes a feasibility question by showing that the nontrivial positive result survives only on the boundary of very restricted domains.

"A Theory of Auditability for Allocation Mechanisms" (Grigoryan et al., 2023) defines auditability through the smallest coalition needed to prove that an observed outcome could not have been generated by the promised mechanism. For a mechanism $1\le p\le 2$ 3, the problem-specific auditability index is

$1\le p\le 2$ 4

and the worst-case index is $1\le p\le 2$ 5. The benchmark values are $1\le p\le 2$ 6 for maximally auditable and $1\le p\le 2$ 7 for minimally auditable mechanisms.

Within DA-representable priority mechanisms, the paper proves that $1\le p\le 2$ 8 exactly when either the induced stable allocation is unique or every alternative stable allocation is sufficiently undesirable for some pair of individuals. This yields a stark contrast: $1\le p\le 2$ 9 and more generally

$2^{1-\frac{1}{p}}$ 0

For hierarchical exchange, maximally auditable mechanisms are exactly vice ownership mechanisms; for sequential dictatorships, maximally auditable ones are exactly vice dictatorships; and mechanisms with $2^{1-\frac{1}{p}}$ 1 are exactly those with a sequential clinching implementation (Grigoryan et al., 2023). In this line of work, tightness means exact structural characterization of the maximally auditable class.

"Tight Samurai Accountant" (Kattwinkel et al., 29 Sep 2025) uses the word differently but no less precisely. In the Border–Sobel auditing/plunder model, a mechanism $2^{1-\frac{1}{p}}$ 2 specifies an advance-payment-dependent audit probability and refunds. The paper defines the deviation-loss function

$2^{1-\frac{1}{p}}$ 3

and says that $2^{1-\frac{1}{p}}$ 4 is tighter than $2^{1-\frac{1}{p}}$ 5 if $2^{1-\frac{1}{p}}$ 6. The main theorem states that if $2^{1-\frac{1}{p}}$ 7 is tight or efficient, then

$2^{1-\frac{1}{p}}$ 8

where $2^{1-\frac{1}{p}}$ 9 is the set of weakly increasing, weakly concave $3$00 with $3$01 and $3$02. The paper further proves

$3$03

so tightness refines efficiency without loss of optimality (Kattwinkel et al., 29 Sep 2025). Here the tight object is not an approximation ratio but a dominance order over profit and incentive-threat profiles.

6. Exact differential privacy mechanisms

In finite categorical data, "Differentially Private Response Mechanisms on Categorical Data" (Holohan et al., 2015) studies the discrete exponential mechanism

$3$04

A central contribution is a minimal verification principle: for neighboring databases $3$05, it is enough to check privacy on the sufficient set

$3$06

The main privacy theorem is exact: $3$07 if and only if the mechanism is $3$08-differentially private. The paper also derives a tight lower bound on maximal expected Hamming error and shows that the optimal product sanitisation mechanism uses

$3$09

This is an instance where privacy characterization, verification burden, and utility lower bound all line up exactly (Holohan et al., 2015).

"Exact zCDP Characterizations for Fundamental Differentially Private Mechanisms" (Harrison et al., 29 Oct 2025) gives analogous exactness results for zCDP. For the $3$10-DP Laplace mechanism, the tight zCDP parameter is

$3$11

For discrete Laplace, the paper gives a sensitivity-dependent exact formula that interpolates between binary randomized response and continuous Laplace. For $3$12-randomized response with $3$13,

$3$14

and for RAPPOR,

$3$15

The paper also proves the exact worst-case zCDP guarantee for $3$16-bounded range mechanisms: $3$17 An important nuance is that the KL-endpoint rule is not universal: for sufficiently large $3$18, the best zCDP bound for $3$19-randomized response is not attained at $3$20 (Harrison et al., 29 Oct 2025).

7. Tight accounting and empirical auditing in differential privacy

A further body of work treats tightness as exact or near-exact privacy accounting under composition. "Tight Differential Privacy for Discrete-Valued Mechanisms and for the Subsampled Gaussian Mechanism Using FFT" (Koskela et al., 2020) works with privacy loss distributions (PLDs) and the exact formula

$3$21

For discrete one-dimensional outputs, the paper places the PLD on a grid, forms left- and right-rounded distributions, and computes repeated convolutions by FFT, obtaining rigorous lower and upper bounds on the true $3$22 guarantee. The analysis gives strict bounds for the subsampled Gaussian mechanism, exact or near-exact accounting for the exponential mechanism on counting queries, and reports that in the binomial setting the accountant allows up to $3$23 reduction in noise variance at equal privacy compared to existing bounds (Koskela et al., 2020).

"Tight Accounting in the Shuffle Model of Differential Privacy" (Koskela et al., 2021) adapts the PLD/Fourier-accountant framework to shuffled mechanisms. For general $3$24-LDP shufflers it uses Feldman et al.’s dominating pair construction; for shuffled $3$25-randomized response it extracts exact privacy-loss random variables from the Balle et al. analysis. The paper then reduces PLD construction from $3$26 to $3$27 terms, up to a logarithmic tolerance factor, by truncating binomial tails with Hoeffding’s inequality, with no noticeable change in the resulting $3$28-upper bounds. It also shows that prior analytic composition bounds are substantially looser, especially for repeated shuffle rounds (Koskela et al., 2021).

"Tight Group-Level DP Guarantees for DP-SGD with Sampling via Mixture of Gaussians Mechanisms" (Ganesh, 2024) avoids black-box conversion from example-level to group-level privacy by modeling a group change directly as a composition of mixture-of-Gaussians mechanisms. Under Poisson sampling, the per-step scalar sensitivity is dominated by $3$29; under fixed batch-size sampling, it is dominated by

$3$30

The paper states that the resulting group-level guarantees are tight up to discretization error in implementation, assuming every intermediate iterate is released, and gives explicit one-dimensional constructions where the dominating pair is exact (Ganesh, 2024).

Finally, "Tight Auditing of Differentially Private Machine Learning" (Nasr et al., 2023) studies empirical privacy lower bounds rather than formal accountants. In the white-box setting with access to all model updates, the audit uses $3$31-DP and GDP rather than only final $3$32 guarantees. Because GDP composes tightly via

$3$33

the method reduces the audit from thousands or millions of training runs to essentially two runs per observation step. The paper reports that around $3$34 observations can suffice for tight GDP-based estimates in the white-box setting, compared with $3$35 million or more observations for comparable tightness under direct $3$36-based auditing, and uses the method to surface implementation bugs such as clipping after gradient averaging and biased noise sampling (Nasr et al., 2023).

Taken together, these privacy papers show that tightness can mean exact mechanism-specific privacy parameters, certified composition theorems, or empirical lower bounds that nearly meet provable upper bounds. A plausible implication is that, in privacy as in mechanism design, the decisive technical issue is often not merely constructing a mechanism but identifying the correct benchmark and proving that no hidden slack remains once composition, sampling, or numerical approximation is accounted for.