top of page
Moacyr do Nascimento - Confirmed Speaker - NMR 2026

Moacyr do Nascimento

Categorical Diversity Optimization: A framework for inversion and component analysis of NMR data

Classification of probed nuclei in NMR relaxometry via relaxation rate spectra is presented as a paradigmatic Categorical Inverse Problem (CIP)—a mixture model problem aiming to find a distribution of known types proportionally linked to a dataset. Here, the categorical set is a partitioned interval, mapping relaxation rate bands onto signal amplitudes via a discretized continuous response model related to the Laplace transform.

I define CIP generally to emphasize its normalization condition and present proper data preconditioning and denoising strategies for applying the principle of Categorical Diversity Optimization: Selection of either the most or least diverse distributions representing a solution set.

By varying diversity measures, the framework establishes a continuum of well-posed inverse problems for any dataset, proving a continuum of uniquely characterized distributions exists as special CIP solutions. This quality unifies traditional inversion methods like Entropy Maximization, Non-negative Least Squares, and the Minimax Principle under a single banner.

Effective 1D and 2D computational algorithms to compute these solutions are advanced, alongside theoretical results on their categorical support and the influence of noise.

As diversity measures differ mainly on how they weigh the contribution of rarer categories, their choice yields distinct qualitative and quantitative descriptions, yet numerical maximal diversity solutions display a remarkable consistency of signatures. Comparisons of minimal and maximal diversity solutions for polyexponential signal decay exemplify the impact of optimization principle.

Finally, the framework's versatility is demonstrated by unmixing problems based on NMR relaxometric data, where mixed signals of unknown proportions are decomposed using a few referential component decays. Potential applications in Quality Control and Petrophysical Evaluation are considered together with current methodological limitations.

​

1) Hansen, Per Christian. Discrete inverse problems: insight and algorithms. Society for Industrial and Applied Mathematics, 2010.

​

2) Polyanskiy, Yury, and Yihong Wu. Information theory: From coding to learning. Cambridge university press, 2025.

​

3) Jost, Lou. "Entropy and diversity." Oikos 113.2 (2006): 363-375.

bottom of page