Thermodynamics of data

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Stosic, Borko
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912826011942912
author Stosic, Borko
author_facet Stosic, Borko
contents The recently introduced concept of generalized thermodynamics is explored here in the context of 1d, 2d and 3d data analysis, performed on samples drawn from a 3d X-ray soil sample image. Different threshold levels are used to binarize the 3d sample, wherefrom relative frequencies of binary patterns are found and then used to address finite size scaling behavior of the response functions as a function of the disorder parameter (equivalent of temperature in thermodynamics). It is found that for different threshold levels response functions for increasing sample sizes approach the thermodynamic limit from different directions, with a crossover reminiscent of a transition from open to periodic boundaries of the Ising model, implying existence of a characteristic correlation scale. It is argued here that this characteristic scale corresponds to the "natural" properties of the data, where correlations within finite size samples are neither underestimated nor overestimated. In the current context of soil this scale may be related to the so-called Representative elementary volume (REV), while in other situations this characteristic scale should be interpreted in the context of the phenomenon under study.
format Preprint
id arxiv_https___arxiv_org_abs_2510_06818
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Thermodynamics of data
Stosic, Borko
Statistical Mechanics
Data Analysis, Statistics and Probability
The recently introduced concept of generalized thermodynamics is explored here in the context of 1d, 2d and 3d data analysis, performed on samples drawn from a 3d X-ray soil sample image. Different threshold levels are used to binarize the 3d sample, wherefrom relative frequencies of binary patterns are found and then used to address finite size scaling behavior of the response functions as a function of the disorder parameter (equivalent of temperature in thermodynamics). It is found that for different threshold levels response functions for increasing sample sizes approach the thermodynamic limit from different directions, with a crossover reminiscent of a transition from open to periodic boundaries of the Ising model, implying existence of a characteristic correlation scale. It is argued here that this characteristic scale corresponds to the "natural" properties of the data, where correlations within finite size samples are neither underestimated nor overestimated. In the current context of soil this scale may be related to the so-called Representative elementary volume (REV), while in other situations this characteristic scale should be interpreted in the context of the phenomenon under study.
title Thermodynamics of data
topic Statistical Mechanics
Data Analysis, Statistics and Probability
url https://arxiv.org/abs/2510.06818