Recently, p-presentation distances for p∈ [1,∞] were introduced for merge trees and multiparameter persistence modules as more sensitive variations of the respective interleaving distances (p=∞). It is well-known that computing the interleaving distance is NP-hard in both cases. We extend this result by showing that computing the p-presentation distance is NP-hard for all p∈ [1,∞) for both merge trees and t-parameter persistence modules for any t≥ 2. Though the details differ, both proofs follow the same novel strategy, suggesting that our approach can be adapted to proving the NP-hardness of other distances based on sums or p-norms.
No takes yet. Share an insight, caveat, or question.
Bjerkevik et al. (2024) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: