Budget-Feasible Mechanisms

Updated 7 November 2025

Budget-feasible mechanisms are rules for procurement auctions that ensure truthfulness and keep total payments within a fixed budget, balancing incentives with efficient allocations.
They employ methods like greedy, threshold, and random sampling techniques to achieve near-optimal approximations across valuation domains such as additive, submodular, and XOS functions.
These mechanisms are applied in diverse fields including crowdsourcing, IoT assignments, and experimental design, with ongoing research addressing online and multidimensional challenges.

Budget-Feasible Mechanisms are a class of mechanisms in algorithmic mechanism design for procurement auctions and related combinatorial settings, where the buyer (auctioneer) faces both incentive constraints (truthfulness) and a hard payment budget. Unlike classic mechanisms, where feasibility is determined only by the allocation structure, budget-feasible mechanisms require that the sum of payments to agents never exceeds a specified budget. This constraint profoundly affects the design and performance of mechanisms, introducing new technical challenges and stimulating a rich body of research on truthful, efficient, and approximately optimal allocation rules across various valuation domains and feasibility constraints.

1. Formal Model and Definitions

A budget-feasible mechanism $M=(f,p)$ consists of:

Allocation rule $f: \mathbb{R}^n\to 2^{[n]}$ , selecting a subset $S$ of agents based on bids (cost reports).
Payment rule $p: \mathbb{R}^n\to \mathbb{R}^n$ , issuing individual payments $p_i\geq 0$ .
Budget-feasibility: $\sum_{i=1}^n p_i\leq B$ for any cost profile.
Truthfulness (dominant strategy incentive compatibility, DSIC): An agent cannot gain by misreporting her true cost. For single-parameter domains, truthfulness is equivalent to monotone allocation and threshold payments (Myerson's Lemma).
Individual Rationality: $p_i\geq c_i$ for allocated agents.
Approximation: For some $\alpha\geq 1$ , $v(f(c_1,\ldots,c_n))\geq \frac{1}{\alpha}\cdot OPT$ , with $OPT$ the algorithmic optimum under the true cost vector.

Budget-feasible mechanism design applies to a diverse array of valuation domains: additive (knapsack), monotone and non-monotone submodular, XOS (fractionally subadditive), and general subadditive functions, frequently under additional feasibility constraints such as matroids, $p$ -systems, or independence systems.

2. Core Mechanism Design Techniques

The field has developed several paradigms and mechanisms, often matched to specific valuation domains. Key general techniques include:

Greedy and Proportional Share Mechanisms: For additive and monotone submodular functions, mechanisms that order agents/items by marginal value per unit cost and incrementally select items while respecting the budget deliver strong guarantees (Singer, 2010, Chen et al., 2010, Jalaly et al., 2017).
Threshold and Oracle Mechanisms: Threshold rules (e.g., adding items if their cost/marginal value meets a moving threshold), and oracle-based mechanisms that leverage submodular maximization algorithms as a black-box, allow for flexible trade-offs between mechanism simplicity and approximation ratio (Jalaly et al., 2017).
Random Sampling and Posted Price Mechanisms: Randomly partitioning agents into groups to estimate optimal benchmarks and set posted prices for remaining agents, a paradigm that enables universal truthfulness and near-optimal guarantees in many settings (Bei et al., 2011, Bei et al., 2012, Gravin et al., 2019).
Local Search and Quasi-Monotonicity: For symmetric and non-monotone submodular objectives, mechanisms leverage local search to identify "almost monotone" regions and apply monotone submodular budget-feasible mechanisms to achieve constant factors (Amanatidis et al., 2017, Amanatidis et al., 2019).
LP and Integrality Gap Reductions: Construction of fractional set cover relaxations ties the approximation ratio of budget-feasible mechanisms to the integrality gap of a linear program ("approximate core"), especially for subadditive valuations (Bei et al., 2011, Bei et al., 2012).

3. Guarantees by Valuation Class and Feasibility Constraints

Submodular and Additive Functions

Additive valuations (Knapsack):
- Randomized mechanisms achieve tight $2$-approximations (Gravin et al., 2019).
- Deterministic mechanisms have tight $3$-approximation bounds.
- Impossibility: no DSIC deterministic mechanism can beat $1+\sqrt{2}\approx 2.41$ (Chen et al., 2010).
Monotone submodular valuations:
- Improved deterministic guarantee: $5$ (Jalaly et al., 2017), with parameterized trade-off using black-box maximization oracles.
- In large markets, best known deterministic ratio: $2.58$ (Jalaly et al., 2017).

XOS and Subadditive Functions

XOS (fractionally subadditive):
- Constant-factor approximation mechanisms exist, with random sampling and LP-based reductions (Bei et al., 2011, Bei et al., 2012).
- Mechanisms apply to independence system knapsack and combinatorial structures, e.g., matchings and matroids (Amanatidis et al., 2016).
Subadditive valuations:
- Best polytime approximation factor is $O(\log\log n)$ (Neogi et al., 5 Jun 2025), improving on previous $O(\log n/\log\log n)$ (Bei et al., 2011, Bei et al., 2012).
- For the Bayesian setting, constant-factor universal truthfulness is attainable given agent costs drawn from known distributions (Bei et al., 2012).

Symmetric and Non-Monotone Submodular Functions

Mechanisms with local search plus greedy knapsack yield $O(1)$ approximations for symmetric submodular and Budgeted Max Cut objectives (Amanatidis et al., 2017).
For general non-monotone submodular objectives, offline and online mechanisms achieve $O(1)$ and $O(p)$ approximations (with $p$ the rank quotient of the system) (Amanatidis et al., 2019).

Beyond Indivisible Procurement: Matroids, Partial Allocations, Multidimensional

Matroid Constraints: Polynomial-time $4$-approximation mechanisms for matroid (Leonardi et al., 2016), $(3\alpha+1)$ for intersections (with $\alpha$ the underlying packer approximation).
Partial Allocations: For divisible agents and multiple levels of service, deterministic mechanisms achieve $2+\sqrt{3}$ (Amanatidis et al., 2023), with linear valuations reaching the tight bound of $2$ (strictly better than indivisible case).
Multidimensional Types: Impossibility of constant-factor approximation against standard benchmark due to monopolist phenomenon; constant-factor guarantees are only attainable via redefined benchmarks ( $OPT_{Bench}$ ) (Neogi et al., 12 Aug 2025).

4. Incentive Compatibility and Variants

Classical budget-feasible mechanisms require DSIC, imposing hard lower bounds (deterministic: $1+\sqrt{2}\approx 2.41$ , randomized: $2$) (Chen et al., 2010, Gravin et al., 2019). Recent work explores relaxed notions:

Non-Obvious Manipulability (NOM), Best-case/Worst-case NOM (BNOM/WNOM): Derivation of tight deterministic $2$-approximation for NOM mechanisms, with randomized universal-BNOM mechanisms achieving expected ratio arbitrarily close to 1 (Keijzer et al., 17 Feb 2025).
Golden Tickets and Wooden Spoons: Mechanistic constructs for realizing BNOM and WNOM, respectively (Keijzer et al., 17 Feb 2025).
Relaxations can yield strictly improved guarantees compared to DSIC, particularly under randomization.

5. Online and Learning-Augmented Mechanism Design

Online budget-feasible mechanism design, typically under the secretary/random-arrival model, demonstrates markedly different behavior:

Without predictions, mechanisms for submodular objectives have high competitive ratios (e.g., $1710$ (Amanatidis et al., 2019)).
With predictions of the offline optimum, competitive ratios are dramatically reduced—for monotone submodular functions, consistent ratio is as low as $6$, robust to $146$ (Amanatidis et al., 30 May 2025).
The effect of predictions is significant online, but negligible for offline mechanism design (Amanatidis et al., 30 May 2025).

6. Impossibility Results, Limitations, and Benchmarks

General superadditive/synergistic value functions: Budget constraint and truthfulness combine to preclude any meaningful approximation (Singer, 2010).
For subadditive class: Polytime mechanisms are limited by integrality gap of fractional LP cover; $O(\log n)$ is tight for worst-case (Bei et al., 2011, Bei et al., 2012).
Multidimensional setting: Standard benchmarks infeasible; new benchmarks (removing unique player dominance) necessary for any guarantees (Neogi et al., 12 Aug 2025).
Strong lower bounds: For $p$ -system constraints, no polynomial-time mechanism can beat $O(p)$ approximation (Amanatidis et al., 2019).

7. Applications, Practical Impact, and Future Directions

Budget-feasible mechanisms underpin theoretical and applied research in crowdsourcing, data acquisition, experimental design, combinatorial procurement, and team formation tasks:

Crowdsourcing/IoT assignment: Mechanisms such as TUBE-TAP guarantee budget feasibility, peer-evaluated quality thresholds, and incentives in multi-task, multi-agent settings (Singh et al., 2018).
Experimental design: Deterministic, polynomial-time, approximate truthful mechanisms constructed for D-optimality (information gain) objectives, with proven impossibility bounds (Horel et al., 2013).
Combinatorial Optimization: Mechanisms for matching, matroids, and independence systems offer improved scalability and approximation (Leonardi et al., 2016, Amanatidis et al., 2016).
Large Markets: Instance-optimal randomized mechanisms, outperform worst-case optimal mechanisms in practical settings; budget-smoothed analysis reveals competitive ratios strictly better than $1-1/e$ in the average case (Rubinstein et al., 2022).

Future research directions include closing the gap for subadditive valuations, developing mechanisms robust to misspecification and learning-augmentation, and extending incentive compatibility relaxations without sacrificing practical budget feasibility. Advances in LP-gaps, online learning, multidimensional types, and behavioral mechanism design continue to shape the evolving landscape of budget-feasible mechanism theory.