Exact Information Prices in a Full-Partition LP: Joint Weight and Tolerance Sensitivity
Year: 2026
Type: preprint
Abstract: We determine the exact value of a full-partition linear programming relaxation as two objective weights and two observation tolerances vary over the nonnegative reals. The model has eight vertices, three soft terminals and six local probability laws, each supported on all 4,140 set partitions. The laws share terminal moments and are coupled by three selected pair observations, the third of which has zero tolerance. The value has six parameter regimes. Five rational dual certificates bound every original partition column; explicit partition laws attain the bounds through convex interpolation and weight-cone decompositions. After normalizing the total objective weight, the formula is the upper concave envelope of four priced points. This representation explains the transition lines between regimes. We also give exact examples in which local relay prices, and then a combined cell-triple certificate, miss a positive joint gain. Finally, we prove additivity for any finite number of modules with disjoint private vertices, three common terminals and no cross-module private-pair observations. The results describe the specified relaxation and certificate libraries; they provide neither a five-parameter formula nor exactness for arbitrary overlapping networks.
Authors: Zhang, Shuai, Zhang, Shuhao
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7233788886
