Spatially Resolved Entropy

• Spatially resolved entropy is a measure that quantifies local heterogeneity by decomposing global entropy into spatially distinct components using thermodynamic and information-theoretic methods.
• It applies across domains such as molecular solvation, urban analytics, and quantum mechanics, employing techniques like k-nearest neighbor estimators and Bayesian spatial models.
• Key challenges include selecting optimal spatial resolution and ensuring additivity and consistency, which are critical for accurately interpreting local and global entropy dynamics.

Spatially resolved entropy quantifies heterogeneity, disorder, or uncertainty at specific spatial locations or scales, distinguishing it from traditional, globally averaged entropy measures. It plays a critical role in physics, statistical mechanics, molecular and quantum systems, urban science, information theory, thermodynamics, and numerous applied domains. Spatial resolution introduces unique methodological challenges and requires principled definitions to ensure physical and statistical consistency.

1. Formal Definitions of Spatially Resolved Entropy

A rigorous spatial decomposition of entropy must yield a field $\overline{S}(x)$ or discrete set $\{\overline{S}_i\}$ that integrates (or sums) to the global entropy. In equilibrium thermodynamics, spatially resolved entropy density is defined via a spatial analogy to partial molar quantities:

$$\overline S(x) = \rho(x)\left(\frac{\partial S}{\partial N(x)}\right)_{T,P}$$

where $N(x)$ is the infinitesimal particle number at $x$ and $\rho(x)$ the local density. This construction ensures

$S = \int d^3x\,\overline S(x)$

In information-theoretic settings, spatially resolved entropy involves the local or binned Shannon entropy of a conditional or marginal distribution at each site or region. For example, for a categorical variable $X$ over spatial regions $s$, the resolved entropy is

$H(s) = -\sum_k p_k(s) \log p_k(s)$

where $p_k(s)$ is the local probability of category $k$ at $s$. In quantum systems, spatial resolution is achieved via local density matrices $\hat\rho(x,t)$, yielding local entropy densities

$$s(x,t) = \mathrm{Tr}[-\hat\rho(x,t)\log\hat\rho(x,t)]$$

These formulations generalize across physical, biological, and socio-economic systems, underpinning both theoretical and data-driven analyses (Persson, 2016, Altieri et al., 2019, Marijan et al., 30 Oct 2025).

2. Spatially Resolved Entropy in Thermodynamic and Molecular Systems

First-order grid inhomogeneous solvation theory (GIST) provides a physically rigorous spatial decomposition of the translational entropy in inhomogeneous fluids. For a solvent in discrete grid cells $i$:

$$\Delta\overline S_i^\mathrm{GIST} = -\frac{N_i}{v} \ln\left(\frac{N_i}{v\rho_\infty}\right)$$

where $N_i$ is the mean particle number in cell $i$, $v$ the cell volume, and $\rho_\infty$ the bulk density. In the continuum, this yields

$$\Delta\overline S(x) = -\rho(x)\ln\frac{\rho(x)}{\rho_\infty}$$

This assignment strictly satisfies the requirement $\overline S(x)=\rho(x)\partial S/\partial N(x)$, ensuring additivity and correct equilibrium behavior. Grid-cell theory (GCT), by contrast, does not consistently yield a spatial decomposition matching the thermodynamic definition, particularly failing for the ideal gas in an external field (Persson, 2016).

Spatially resolved rotational solvent entropies can be computed using nonparametric $k$-nearest neighbor (kNN) density estimators on $SO(3)$, assigning to each spatial voxel $r$ a local rotational entropy:

$$S_\mathrm{rot}(r) = -k_B \int_{SO(3)} p(\Omega|r)\ln p(\Omega|r)\,d\Omega$$

Empirical application to atomistic water simulations achieves sub-nanometer spatial resolution, providing detailed entropy maps critical for understanding solvation and hydrophobicity (Heinz et al., 2019).

3. Statistical, Urban, and Information-Theoretic Approaches

Spatial entropy in urban and ecological settings extends the classical Shannon entropy to account for spatial arrangement and correlation:

• Bayesian spatial entropy surfaces: Using spatial multinomial–CAR models, posterior distributions for local entropies $H(s)$ are generated, incorporating spatial autocorrelation and borrowing strength across locations (Altieri et al., 2019).
• Distance-conditioned residual entropy: By examining unordered pairs $(i,j)$ at distance classes $D$, one defines the residual entropy $H(Z|D)$ and the mutual information $I(Z;D)$:

$$H(Z|D) = -\sum_{k=1}^K p(d_k)\sum_{r=1}^R p(z_r|d_k)\log p(z_r|d_k)$$

$I(Z;D) = H(Z) - H(Z|D)$

The mutual information $I(Z;D)$ quantifies the spatial contribution to heterogeneity, with partial contributions at each distance class indicating relevant spatial scales (Altieri et al., 2017, Altieri et al., 2018).

• Urban sprawl and complexity metrics: Entropy of visitor origins, calculated from large-scale mobile phone datasets, serves as a spatially resolved indicator of urban attractiveness and socio-economic complexity. For unit $j$ and time $t$:

$$S_j(t) = -\frac{1}{\ln N}\sum_{i=1}^N p_{j\leftarrow i}(t)\ln p_{j\leftarrow i}(t)$$

High entropy correlates with diverse, job-rich, and affluent urban subregions (Lenormand et al., 2020).

4. Multiscale and Scale-Dependent Spatial Entropy

Spatially resolved entropy can be constructed at multiple scales to capture the complexity of patterning in, for example, urban morphologies. By aggregating characteristics of each spatial element across a sequence of neighborhood sizes, multiscale entropy is defined as the joint entropy over the vector of values at all considered scales:

$H = -\sum_{s\in S} p(s)\log p(s)$

where $s$ indexes discrete states formed by binning the multiscale feature vector associated with each location. This approach distinguishes patterns (e.g., polycentric sprawl, fractal cascades) that occupy multiscale phase space more uniformly, and aligns with maximum-entropy explanations for emergent spatial complexity (Barner et al., 2017).

Spatial entropy also formalizes connections to fractal dimension via Rényi entropies $M_q(\epsilon)$ (order $q$) at scale $\epsilon$:

$$M_q(\epsilon) = -\frac{1}{q-1}\ln\left[\sum_{i=1}^N p_i(\epsilon)^q\right]$$

with entropy-based indices (normalized filling, redundancy, odds) mirroring those in fractal geometry. Unlike fractal dimension, entropy-based measures remain meaningful even without strict scaling laws and provide direct assessment of spatial disorder and saturation (Chen, 2020).

5. Experimental and Quantum-Mechanical Spatial Entropy

In closed quantum many-body systems, local entropy densities $s(x,t)$ are defined via local density matrices, and their dynamics obey continuity equations:

$\partial_t s(x,t) + \nabla\cdot j_s(x,t) = 0$

Though the total von Neumann entropy $S$ is invariant under unitary evolution, spatially resolved entropy reveals nontrivial transport: empirical data in spinor Bose gases exhibit a dual entropy cascade—entropy decreases macroscopically (infrared, large-scale order) while increasing microscopically (ultraviolet, disorder)—with total entropy conserved. Mode-resolved Boltzmann–Einstein entropy and Shannon entropy of experimentally measured observables further quantify this transport, supporting universal scaling predictions and elucidating the formation of macroscopic order from microscopic dynamics (Marijan et al., 30 Oct 2025).

6. Computational, Methodological, and Practical Considerations

Choice of Spatial Resolution and Discretization

• Mesh refinement and resolution criteria: In numerical wave problems, the Shannon entropy of discretized mode intensities $S(N)$, as a function of the number of mesh points $N$, provides a quantitative indicator of resolved spatial complexity. The optimal mesh range is bounded below by the critical mesh $N_c$ (where salient features appear) and above by $N_\mathrm{max}$ (where entropy saturates) (Park et al., 2019).
• Parametric and nonparametric estimation: kNN density estimators, Bayesian hierarchical models, and explicit spatial binning are widely used to estimate local distributions underlying spatial entropy.
• Additivity and decomposition: Desirable properties such as additivity across scales, distance classes, or partitions ensure interpretability and mathematical consistency (Altieri et al., 2017, Barner et al., 2017).

Table: Representative Approaches to Spatially Resolved Entropy

Domain / Approach	Key Definition or Method	Reference
Thermodynamic (GIST)	$$\overline S(x) = -\rho(x)\ln\frac{\rho(x)}{\rho_\infty}$$	(Persson, 2016)
Quantum Many-Body	$$s(x,t) = \mathrm{Tr}[-\hat\rho(x,t)\log\hat\rho(x,t)]$$	(Marijan et al., 30 Oct 2025)
Urban Socio-Economic	$$S_j = -\frac{1}{\ln N}\sum p_{j\leftarrow i}\ln p_{j\leftarrow i}$$	(Lenormand et al., 2020)
Multiscale Urban	$H = -\sum p(s)\log p(s)$ over multiscale state vectors	(Barner et al., 2017)
Bayesian Spatial Surface	Local entropy $H(s)$ from posterior of multinomial–CAR model	(Altieri et al., 2019)
Shannon–Mutual Information (Pairs)	$H(Z|D)$, $I(Z;D)$ via pairwise distance classes	(Altieri et al., 2017)

7. Applications, Extensions, and Limitations

Spatially resolved entropy supports inference and explanation in molecular solvation thermodynamics, quantum nonequilibrium dynamics, urban structure, remote sensing, ecology, and mobility analysis. It reveals where and at what scale patterns are most or least heterogeneous, providing actionable information for model selection, hypothesis testing, and policy.

Several caveats and open issues remain:

• Sensitivity to spatial discretization (e.g., modifiable areal unit problem, choice of $\epsilon$)
• High-dimensional joint or multiscale vectors require careful regularization or data scaling
• In classical molecular or urban systems, entropy can diverge as resolution increases unless coarse-grained appropriately
• In quantum systems, the operational meaning of spatial entropy depends critically on the choice of local measurements or coarse-graining

Nonetheless, rigorous definitions—anchored in thermodynamics, probability, or information theory—combined with computational advances, increasingly enable detailed spatially resolved entropy analyses across complex systems (Persson, 2016, Barner et al., 2017, Marijan et al., 30 Oct 2025).