Exact bill splitting and minimum-transfer debt settlement: an integer model and a prediction-robust block decomposition
Year: 2026
Type: preprint
Abstract: Preprint. Submitted to the Open Journal of Mathematical Optimization (November 2026). Code and seeded benchmarks: https://github.com/Gautam5514/exact-bill-splitting Shared-expense ledgers must divide each bill in indivisible currency units and settle the resulting net balances with few transfers. We give one integer model for both problems. Every split rule is reduced to exact rational quotas followed by a single largest-remainder rounding, which is exact in sum, within one unit of every quota, and minimises every separable convex measure of rounding error. For settlement we restate the classical facts with complete proofs: the minimum number of transfers is n − Z, exact opposite pairs can be removed without loss, and a subset dynamic program solves m members in O(m·2^m) time. For large groups we analyse zero-sum-safe block decompositions: whatever a predictor proposes, the plan conserves all balances, moves the least possible amount, and uses at most min(OPT + η, (3/2)·OPT) transfers, where η counts the optimal zero-sum parts cut by the blocks. On seeded synthetic clustered groups of 30 to 200 members a learned instance used 3.0–25.0% fewer transfers than greedy. Learned edge weights improved on Louvain blocks of the co-spend-weighted co-participation graph by only 0.8% on average (at most 2.0%), and the mean gap to a certified lower bound was 12.8–43.1%.
Author: Pandit, Gautam
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7237337243
Cites:
Cited by:
Related to:
