A Sharma–Mittal Geometry of Evenness: Representation, Cardinal Resolution, and Relative Merits
Abstract
The proliferation of evenness indices creates an interpretive problem whenever several admissible measures assign different numerical values to the same assemblage. We study that problem by normalizing Sharma–Mittal entropy to its maximum at complete uniformity. Written in terms of the Hill number of order α, the resulting surface is Eα,β(p;S)=(Dα(p)1−β−1)/(S1−β−1), Eα,1=log Dα/log S. Five canonical Hill-based evenness families are exact restrictions of this surface. More generally, for every α > 0 and finite β, the family satisfies Requirements 1a–3b of Chao and Ricotta: transfer, continuity and symmetry, non-increase under addition of a vanishingly rare species, a richness-independent limiting range, and scale invariance. At fixed richness and Hill order, every section is an invertible monotone transformation of the same Dα, so the represented measures contain the same ordinal information and differ in numerical spacing. The parameter β controls that spacing: local sensitivity is proportional to Dα−β and the ratio of endpoint slopes is exactly Sβ. Over the canonical range 0≤β≤2, the largest possible disagreement between any two sections is (√S−1)/(√S+1), attained at Dα=√S. When richness differs, ordinal equivalence need not survive: if the linear section and Dα rank two assemblages in opposite directions, continuity guarantees a finite-β ranking reversal. Coverage-standardized Miocene and Pliocene data from Chao et al. (2020) provide a published example, with a crossing near β ≈ 1.40 using their tabulated values. Finally, under K-fold replication, β < 1 leaves a positive asymptotic evenness deficit, β = 1 approaches one logarithmically, and β > 1 approaches one polynomially; at β = 0 the replication limit is exactly Hill evenness Dα/S. The same parameter also governs independent-product composition and large-richness behavior. These results make the relative merits of alternative evenness geometries explicit without selecting a universal winner.
Related
- Strain resolution and wide relative abundance quantification of bacterial microbiomes with nanopore sequencing of the ribosomal RNA operon
- Temporal resolution shapes inferred higher-order interactions and information representation in neural population activity: A data-driven approach using minimally complex models
- OmniScore: Universal Scoring of Diverse Biomolecular Complexes via Equivariant Geometry-Aware Discrete Representation Learning
- A spin glass representation of hierarchical gene regulation and the partial resolution of frustration
- muDM: A Queryable Multi-Resolution Data Model for Vector Geometry in Spatial Bioimaging
BibTeX
@misc{delvecchio2026sharmamittal,
title = {{A Sharma–Mittal Geometry of Evenness: Representation, Cardinal Resolution, and Relative Merits}},
author = {Luca Delvecchio},
howpublished = {EcoEvoRxiv},
year = {2026},
month = sep,
doi = {10.32942/x2011b},
url = {https://ecoevorxiv.org/repository/view/14879/},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - A Sharma–Mittal Geometry of Evenness: Representation, Cardinal Resolution, and Relative Merits
AU - Luca Delvecchio
PY - 2026
DA - 2026/09/11
PB - EcoEvoRxiv
DO - 10.32942/x2011b
UR - https://ecoevorxiv.org/repository/view/14879/
AB - The proliferation of evenness indices creates an interpretive problem whenever several admissible measures assign different numerical values to the same assemblage. We study that problem by normalizing Sharma–Mittal entropy to its maximum at complete uniformity. Written in terms of the Hill number of order α, the resulting surface is E_α,β(p;S)=(D_α(p)^1−β−1)/(S^1−β−1), E_α,1=log D_α/log S. Five canonical Hill-based evenness families are exact restrictions of this surface. More generally, for every α > 0 and finite β, the family satisfies Requirements 1a–3b of Chao and Ricotta: transfer, continuity and symmetry, non-increase under addition of a vanishingly rare species, a richness-independent limiting range, and scale invariance. At fixed richness and Hill order, every section is an invertible monotone transformation of the same D_α, so the represented measures contain the same ordinal information and differ in numerical spacing. The parameter β controls that spacing: local sensitivity is proportional to D_α^−β and the ratio of endpoint slopes is exactly S^β. Over the canonical range 0≤β≤2, the largest possible disagreement between any two sections is (√S−1)/(√S+1), attained at D_α=√S. When richness differs, ordinal equivalence need not survive: if the linear section and D_α rank two assemblages in opposite directions, continuity guarantees a finite-β ranking reversal. Coverage-standardized Miocene and Pliocene data from Chao et al. (2020) provide a published example, with a crossing near β ≈ 1.40 using their tabulated values. Finally, under K-fold replication, β < 1 leaves a positive asymptotic evenness deficit, β = 1 approaches one logarithmically, and β > 1 approaches one polynomially; at β = 0 the replication limit is exactly Hill evenness D_α/S. The same parameter also governs independent-product composition and large-richness behavior. These results make the relative merits of alternative evenness geometries explicit without selecting a universal winner.
ER -