A note on rounding fractional matchings with constant-factor strong negative correlation

Abstract: We describe new dependent-rounding algorithms for bipartite graphs. Given a fractional matching $x$ of graph $G = (U \cup V, E)$, the algorithms return an integral solution $X$ such that each right-node $v \in V$ has at most one edge, and where the variables $X_e$ also satisfy broad non-positive correlation properties. In particular, for any edges $e_1, e_2$ sharing a left-node $u \in U$, the variables $X_{e_1}, X_{e_2}$ have \emph{strong} negative-correlation, i.e. the expectation of $X_{e_1} X_{e_2}$ is significantly below $x_{e_1} x_{e_2}$. Dependent rounding schemes with these properties have been used for a approximation algorithms for job-scheduling on unrelated machines to minimize weighted completion times, among other applications. Our new algorithm achieves simpler and qualitatively stronger bounds compared to prior algorithms. In particular, we achieve a negative-correlation property $$ \E[X_{e_1} X_{e_2}] \leq 0.79751 \ x_{e_1} x_{e_2}, $$ which is a significant constant-factor improvement over Baveja, Qu & Srinivasan (2023).

• The paper presents a novel dependent rounding algorithm that significantly improves the strong negative correlation constant to 0.79751 for edge pairs sharing a left-node.
• It employs a Brownian-motion based random walk combined with Dirichlet rounding to optimally balance resource allocation in bipartite graphs.
• The approach provides tighter theoretical guarantees and practical benefits for approximation algorithms in scheduling and resource allocation.

Summary of "A note on rounding fractional matchings with constant-factor strong negative correlation"

Problem Formulation and Motivation

This paper investigates dependent rounding schemes for bipartite graphs under the fractional matching setting. Given a bipartite graph $G = (U \cup V, E)$ and a fractional matching $x$, the goal is to obtain an integral assignment $X$ such that (i) each right-node $v \in V$ is incident to at most one selected edge, and (ii) the random variables $X_e$ representing the rounded edge indicators satisfy strong negative correlation properties—specifically for edge pairs that share a left-node.

Such rounding techniques are fundamental in approximation algorithms, especially in scheduling and resource allocation problems where balancing load (avoiding over-allocation to any left-node) is critical. Independent rounding fails to provide control over correlations, which can lead to poor load balancing. Prior works, including Bansal et al. [bansal2016lift], Baveja et al. [baveja2024], and Harris [harris2025], have progressively improved the guarantees on negative correlation, but typically face trade-offs between simplicity and quantitative strength of the correlation bounds.

Main Contributions

The paper introduces a new random-walk-based dependent rounding algorithm for bipartite graphs, building on and generalizing Baveja et al. [baveja2024]. The algorithm achieves a significantly improved strong negative correlation constant for edge pairs sharing a left-node:

• For all $x_e, x_f$, $[X_e X_f] \leq 0.79751 x_e x_f$, which is substantially tighter than the previous best bound $\frac{26}{27} \approx 0.96$.

A hybrid scheme is also developed that combines this Brownian motion–based rounding with Dirichlet rounding (from Harris [harris2025]), further improving the constant across all input regimes by leveraging the respective strengths of each method in different ranges of $x_e, x_f$ (large and infinitesimal, respectively).

Technical Approach

Brownian-motion Dependent Rounding Algorithm

The core rounding algorithm executes a correlated random walk in the fractional matching polytope. A set of protected edges $P$ is maintained and updated via two types of randomized steps. The primary goal is to iteratively enforce that each left-node is incident to at most one fractional edge in $x$0, after which independent rounding is performed per right-node.

A crucial innovation is the careful manipulation of the protected set $x$1 and the random prioritization/ranking of left-nodes, enabling tight control over dependencies and facilitating the analytical proof of stronger negative correlations.

Analysis and Guarantees

Four main properties are established for the resulting rounded variables $x$2:

1. Per right-node coverage (A1): Each right-node $x$3 has exactly one selected edge.
2. Marginal preservation (A2): Each $x$4 (matching the fractional marginal).
3. Generalized negative correlation for stable edge sets (A3): For any stable edge set $x$5, $x$6.
4. Strong negative correlation for left-node sharing pairs (A4): For any pair $x$7 sharing a left-node, $x$8.

The strong negative correlation bound is established via an intricate potential function analysis and iterative application of conditioning and convexity arguments, ultimately leading to an explicit computation of the optimal correlation constant.

Hybridization with Dirichlet Rounding

To further improve the negative correlation for small marginal values, a hybrid rounded scheme is proposed: With probability $x$9, the new Brownian procedure is executed; with probability $X$0, Dirichlet rounding from [harris2025] is used. Parameters $X$1 and $X$2 yield the claimed universal strong negative correlation bound of $X$3 for all relevant arguments.

Numerical and Comparative Results

The key numerical achievement is the decrease in the strong negative correlation constant from $X$4 (Baveja et al.) to $X$5. The hybrid method ensures that the constant holds regardless of the magnitudes of $X$6 and $X$7, remedies a deficiency in prior contention-resolution schemes (which have poor constants for non-infinitesimal marginals), and does so in a manner that simplifies algorithmic analysis and avoids the need for complex computer-assisted bounds.

Implications and Future Directions

Theoretical Implications

• The work tightens the theoretical limits on dependent rounding for fractional matchings with respect to strong negative correlation—a property central to advanced randomized approximation algorithms for assignment and scheduling problems.
• The hybridization principle exemplifies an effective aggregation of algorithmic techniques, achieving uniform improvements across different input domains without complicated case analyses or obscure non-algebraic expressions.
• The simplicity of the improved constant may lead to new lower bounds and refined understanding in the study of randomized rounding, as well as potentially closing the gap to the best-possible negative dependence constants for similar combinatorial structures.

Practical Impact

Stronger negative correlation in rounding translates to better load balancing when integerizing LP-based relaxations. For instance, in scheduling and allocation, this directly leads to better approximation guarantees and fewer pathological cases where resources are heavily overloaded due to uncontrolled correlations. The presented algorithm is also efficient (polynomial-time) and practical for implementation in settings requiring careful resource allocation with fairness or balance constraints.

Future Developments

The main open avenues include:

• Further tightening of the negative correlation constant, including the possibility of achieving the optimal theoretical bound or proving its unattainability.
• Generalizing the principles to broader classes of dependent rounding problems, including those involving general graphs or higher-order interactions beyond pairs.
• Applying the developed techniques in new domains within approximation algorithms, especially where strong guarantees on negative dependence directly impact worst-case or average-case performance.

Conclusion

This paper delivers a substantial advance in the theory and practice of dependent rounding for bipartite fractional matchings, pushing the frontier on strong negative correlation bounds. By designing a refined Brownian-motion based scheme and hybridizing with Dirichlet rounding, the author closes a longstanding gap and offers both a simplification and a quantitative improvement over prior art, with immediate benefits for approximation algorithms in scheduling and resource allocation. The results open both theoretical and practical avenues for future research in negative dependence rounding mechanisms.

Reference:

"A note on rounding fractional matchings with constant-factor strong negative correlation" (2606.07820)