Closure Dependent Entropy Hierarchy

• Closure dependent entropy hierarchies are defined as stratified frameworks where successive coarse-graining yields monotonic entropy reduction and enforces model-specific realizability constraints.
• Methodologies such as maximum entropy and φ-divergence minimization are used to construct these hierarchies, ensuring conservation laws and robust kinetic modeling.
• These frameworks extend across disciplines—from kinetic theory and plasma physics to symbolic dynamics and computational information theory—providing unified entropy bounds and structural insights.

A closure dependent entropy hierarchy is a structural and ordering principle that arises in the mathematical description of reduced or approximate representations of physically or informationally complex systems, where coarse-graining or moment closure schemes induce a stratification of entropy-like quantities, with each level of closure yielding a monotonic reduction in entropy and introducing model-specific realizability constraints. Such hierarchies are central in statistical mechanics, kinetic theory, plasma physics, symbolic dynamics, combinatorics, and computational information theory. Each closure-dependent entropy hierarchy encodes, via the notion of “closure” (in the algebraic, topological, probabilistic, or operational sense), both a chain of entropy functionals and a mechanism for quantifying and delimiting the information loss, irreversibility, and statistical admissibility intrinsic to the reduction.

1. Mathematical Definition and Hierarchical Structure

Closure dependent entropy hierarchies are most explicitly constructed in kinetic theory and plasma physics via moment hierarchies and their entropic projections (Tevzadze, 1 Feb 2026, Abdel-Malik et al., 2015). Given a high-dimensional distribution $f(z)$ (e.g., solution of the Vlasov or Boltzmann equation), one defines the full Gibbs entropy

$S^{(\infty)}[f] = -\int f(z)\ln f(z)\,dz,$

which is an invariant under the conservative dynamics (if collisionless).

A reduced (closure) model specifies a finite set of moments $\{M^{(k)}\}_{k=0}^n$ and reconstructs a distribution $f^{(n)}$ that reproduces these moments—typically via maximum entropy or generalized divergence minimization. The projected entropy at closure order $n$ is then

$S^{(n)} = -\int f^{(n)}\ln f^{(n)}\,dz,$

with the monotonicity property

$$S^{(\infty)} \geq S^{(1)} \geq S^{(2)} \geq \cdots \geq S^{(n)} \geq \cdots$$

guaranteed by the contraction of admissible distributions under successive moment constraints. The resulting hierarchy quantifies the increasing information loss as more microscopic detail is excluded.

In settings with closure operators (e.g., matroids, network coding), closure-dependent entropy is captured by chains of rank functions and their associated inequalities (Gadouleau, 2013), forming a sharply ordered sequence of entropy bounds between the inner, outer, lower, upper ranks, and the true entropy function.

2. Closure Mechanisms in Kinetic Theory and Statistical Physics

Moment closures of the Boltzmann equation via $\phi$-divergence minimization furnish a formal closure-dependent entropy hierarchy parameterized by both the choice of divergence and the moment space dimension (Abdel-Malik et al., 2015). For two phase-space densities $\mu_1,\mu_2$, the $\phi$-divergence is

$$D_\phi(\mu_1 \|\mu_2) = \int \mu_2(v)\,\phi\left(\frac{\mu_1(v)}{\mu_2(v)}\right) dv,$$

with convex $\phi$ satisfying $\phi(1)=\phi'(1)=0,\;\phi''(1)>0$.

Given fixed moments $m_i = \int \psi_i(v) f(v) dv$, the closure $f_\phi$ is determined by minimizing $D_\phi(f\|M)$ (with $M$ a reference Maxwellian) subject to moment constraints. The entropy associated to each closure is

$$H_\phi(f_\phi) = \int M \phi\left(\frac{f_\phi}{M}\right) dv,$$

establishing a two-parameter hierarchy in $(\phi, M)$ which interpolates between the classical Grad closure and the maximum entropy (Levermore) closure.

This approach guarantees the inheritance of conservation laws, Galilean invariance, and entropy dissipation for any collision operator that dissipates the selected $\phi$-divergence, and systematically avoids pathologies such as non-realizability and singular fluxes at the closure boundary (Abdel-Malik et al., 2015).

3. Residual Entropy and Statistical Realizability

In reduced models, the nature of the closure—specifically, whether it is "hard" (conservative, invariant under time-reversal) or "soft" (introduces modeled or approximated fluxes at the truncation order)—determines the temporal behavior of the reduced entropy (Tevzadze, 1 Feb 2026). For hard closures, the projected entropy $S^{(n)}$ is constant and the closure is reversible. In contrast, soft (non-conservative) closures yield

$\frac{d}{dt} S^{(n)} \geq 0,$

with the excess $R^{(n)} = S^{(n)}_{soft} - S^{(n)}_{hard}$ interpreted as a residual entropy quantifying the irretrievable information leakage to unresolved degrees of freedom.

Such monotonic residual entropy growth imposes realizability constraints: any macroscopic trajectory that would require $R^{(n)}$ to decrease is inadmissible, thus enforcing a statistical exclusion or “entropy-exclusion” boundary on the accessible phase space. At each hierarchy level, this exclusion boundary can be determined and, in certain cases, compared directly to empirical data, as with plasma anisotropy distributions in the solar wind (Tevzadze, 1 Feb 2026).

4. Closure-Dependent Entropy Hierarchies in Symbolic Dynamics

In symbolic dynamics, closure dependent entropy hierarchies (often termed the "CPE hierarchy") emerge from the iterative application of topological and transitive closures to the Blanchard entropy-pairs relation $EP(X)$ on a subshift $X\subset A^\mathbb{Z}$ (Salo, 2019). The process constructs a transfinite sequence of relations $E^{(\alpha)}\subset X^2$, alternating closure operations:

• For even $\beta$, $E^{(\beta+1)}$ is the topological closure of $E^{(\beta)}$;
• For odd $\beta$, $E^{(\beta+1)}$ is the transitive closure of $E^{(\beta)}$;
• At limit ordinals, $E^{(\alpha)} = \bigcup_{\beta<\alpha} E^{(\beta)}$ (with closure if odd).

The CPE-class (or CPE-rank) is the least ordinal $\lambda$ for which $E^{(\lambda)}=E^{(\lambda+1)}$, and the system is said to be of CPE-class $\lambda$ if $E^{(\lambda)}=X^2$ but $E^{(\lambda-1)}\neq X^2$. Notably, for every countable ordinal $\alpha\geq1$, there exists a subshift with CPE-class $\alpha$, realized through a sequence of constructions embedding arbitrary abstract closure processes into subshift dynamics (Salo, 2019).

5. Hierarchies in Computational Information Theory

In computational entropy theory, the closure-dependent hierarchy organizes several distinct but related notions of pseudoentropy, distinguished by their closure properties under test classes $\mathcal{F}$ (Impagliazzo et al., 2020). The canonical chain is:

• HILL entropy (existence of a "dense model"),
• entropy as “dense in a pseudorandom set”,
• pseudodensity (test-based operational definition).

Without further closure, the implications are strict: $$\text{HILL} \implies \text{Dense-in-Pseudorandom} \implies \text{Pseudodensity},$$ with neither reversal holding generally. When $\mathcal{F}$ is closed under $k$-wise majority, the dense model theorem guarantees equivalence of all three definitions (up to polynomial losses). Absence of such closure, e.g., for $\mathrm{AC}^0$ or low-degree polynomial classes, leads to genuine separations, which are explicitly constructed (Impagliazzo et al., 2020). The corresponding hierarchy thus not only classifies pseudoentropy notions but also imposes strict limitations on their use in cryptography and combinatorics.

6. General Rank and Entropy Bounds for Closure Operators

In algebraic and combinatorial settings, closure-dependent entropy hierarchies manifest through the stratification of entropy via nested rank-type functions associated to closure operators (Gadouleau, 2013). For a closure operator $\text{cl}:2^E\to 2^E$ (e.g., matroids), the hierarchy of ranks—inner, outer, lower, upper—is explicitly ordered: $$\mathrm{lr}(X) \leq \mathrm{ur}(X) \leq H_f(X) \leq \mathrm{or}(X) \leq \mathrm{ir}(X) \leq |X|.$$ Shannon entropy functions must lie between upper and outer ranks, and density theorems show that achievable closure entropies are dense in the permitted range (Gadouleau, 2013). These bounds offer necessary criteria for solvability, coding, or secret sharing applications.

7. Applications and Physical Significance

Closure dependent entropy hierarchies play a determining role in:

• Classifying the reversibility and irreversibility of reduced physical models (Tevzadze, 1 Feb 2026);
• Quantifying realizability and admissibility of macrostates (via entropy exclusion or boundaries);
• Ensuring the well-posedness, hyperbolicity, and admissible closures in kinetic equations (Abdel-Malik et al., 2015, Sadr et al., 2023);
• Organizing the complexity and entropy-structure in symbolic and dynamical systems (Salo, 2019);
• Delineating the mathematical and operational distinctions between pseudoentropy notions critical to cryptographic reductions (Impagliazzo et al., 2020);
• Establishing sharp entropy bounds, density properties, and representability in combinatorial and information-structural frameworks (Gadouleau, 2013).

These hierarchies thus constitute a unifying conceptual tool, structuring the interplay between closure operations, entropy minimization, and statistical or operational realizability across a diverse array of mathematical, physical, and computational systems.