Entropy-Invariant Scaling

Updated 24 January 2026

Entropy-Invariant Scaling is a framework where entropy properties remain constant under transformations like rescaling and group actions.
It unifies diverse areas, including dynamical systems, quantum many-body physics, statistical mechanics, and complex networks, by revealing universal scaling behaviors.
The concept offers practical insights into phase transitions, system classification, and robustness by linking invariant entropy measures with structural complexity.

Entropy-invariant scaling refers to a class of scaling laws and patterns in the growth, distribution, or invariance properties of entropy—broadly understood as information, disorder, or complexity—that persist under group actions, rescalings, or parameter changes in mathematical, physical, and information-theoretic systems. The concept permeates several domains, including dynamical systems, probability measures on combinatorial or relational structures, quantum information, statistical mechanics, and complex networks. Crucially, entropy-invariant scaling entails that the entropy—appropriately defined—exhibits scaling behaviors or functional forms that are preserved (invariant) under transformations such as relabeling, system size increase, metric rescaling, or group operations.

1. Entropy-Invariant Scaling in Dynamical and Measure-Theoretic Systems

In ergodic theory and dynamical systems, entropy-invariant scaling is formalized via the notion of scaling entropy, introduced by A. M. Vershik and collaborators (Zatitskiy, 2014, Vershik et al., 2023). For a measure-preserving transformation $T$ of a probability space $(X, \mu)$ and an admissible semimetric $\rho$ on $X$ , define the averaged semimetric over $n$ iterates as

$\rho_n(x,y) = \frac{1}{n}\sum_{k=0}^{n-1} \rho(T^k x, T^k y).$

The $\varepsilon$ -entropy $H_\varepsilon(X,\mu,\rho_n)$ is the logarithm of the minimal number of $\rho_n$ -balls of radius $\varepsilon$ needed to cover almost all of $X$ . A sequence $h_n$ is a scaling-entropy sequence if $H_\varepsilon(X,\mu,\rho_n)/h_n$ remains bounded above and below for small $\varepsilon$ as $n \to \infty$ ; its equivalence class is a metric invariant.

Entropic scaling distinguishes systems:

Positive Kolmogorov–Sinai entropy $\Rightarrow$ linear growth $h_n \sim n$ (e.g., Bernoulli shifts).
Pure-point spectrum $\Rightarrow$ bounded $h_n$ (e.g., circle rotations).
Substitution or intermediate-complexity systems can exhibit logarithmic growth or other sublinear regimes.
For general $\mathbb{Z}$ - or amenable-group actions, the scaling-entropy class is a robust isomorphism invariant (Veprev, 2023).

This construction generalizes Shannon entropy, refining the notion of dynamical complexity, especially in regimes where Kolmogorov–Sinai entropy vanishes but the system exhibits nontrivial information growth.

2. Scaling Laws for Symmetric Invariant Measures

In combinatorics and logic, entropy-invariant scaling describes the growth rate of the entropy function $H(n)$ of $S_n$ -invariant probability measures $\mu$ on spaces of $L$ -structures (relational models) on $n$ elements (Ackerman et al., 2018). For a countable relational language $L$ —possibly with infinitely many relation symbols—the entropy function

$H(n) = -\sum_{A \in \mathrm{Str}_L([n])} \mu_n(A) \log_2 \mu_n(A)$

(where $\mu_n$ is the restriction of $\mu$ to $\mathrm{Str}_L([n])$ ) exhibits universal scaling regimes:

If $L$ is finite with all relations of fixed arity $k\geq 1$ and $\mu$ is non-redundant, then

$H(n) = C n^k + o(n^k)$

for some constant $C\geq 0$ , mirroring Erdős–Rényi and random-free phases.

For $k\geq 2$ , one can realize all sub-polynomial growth $H(n) = o(n^k)$ , with arbitrary intermediate rates.
For possibly infinite $L$ , $H(n)$ can grow faster than any fixed polynomial.

This establishes precise entropy scaling bounds for symmetric measures, with a full continuum of scaling behaviors in the $o(n^k)$ regime and the leading term governed by hypergraphon-structure integrals (Ackerman et al., 2018).

3. Entanglement Entropy and Scale-Invariant Universal Forms

In quantum many-body systems and field theory, entropy-invariant scaling refers to universal finite-size scaling forms of entanglement entropy, exhibiting invariance under block/system rescalings and parameter transformations:

For scale-invariant (but non-conformal) states emerging from spontaneous symmetry breaking with type-B Goldstone modes—realized in ferromagnetic spin chains and higher- $d$ lattices—the entanglement entropy of a block of size $n$ in a system of size $L$ is universally (Zhou et al., 2023, Zhou et al., 2024):

$S_f(L, n) = \frac{N_B}{2} \log_2 \left[ \frac{n(L-n)}{L} \right] + S_{f0}$

where $N_B$ is the number of type-B Goldstone modes and $S_{f0}$ is a non-universal constant. This form arises from three physical requirements: block-environment symmetry, extensivity in the thermodynamic limit, and scale homogeneity.

In 2D conformally invariant critical points, the celebrated area law receives logarithmic corrections from sharp corners; these corrections are robust entropy invariants, with their coefficients providing direct probes of the central charge or breakdown of unitarity (Zhao et al., 2021).
In 2+1D scale-invariant critical fermion models, the subleading correction to area law is a scale-invariant function $F(u, \tau)$ of geometric ratios, universal across classes of models and insensitive to the overall length scale (Chen et al., 2014).

This universality and scaling-invariance of entropy forms underpins entropic classification of quantum phases and provides diagnostic power for the presence or absence of conformal invariance.

4. Entropy-Invariant Scaling in Statistical Physics and Probability Laws

Fundamental invariance principles—shift, stretch, rotation—completely determine the functional forms of canonical probability distributions (exponential, Gaussian, gamma, power law) and their entropy content (Frank, 2016). Primary scale invariance leads directly to scaling-invariant laws:

Shift and stretch invariance yield the exponential family.
Rotational invariance produces Gaussian laws.
Affine-invariant scale transformations, or the incorporation of measurement scale into maximum entropy, generate broader families (gamma, Weibull, Pareto, Student’s $t$ ):

$p(x) \propto \exp[-\lambda T(x)]$

for appropriate $T(x)$ (linear, log, or interpolating).

The scaling properties are fully determined by invariance, with entropy (self-information) emerging as a secondary quantity.

Entropy-invariant scaling thereby rationalizes the ubiquity of heavy tails, power-law phenomena, and universal statistics in complex systems (Frank et al., 2010).

5. Generalized Entropic Invariants under Scale Transformations

A recent generalization defines scale-invariant entanglement entropies via the unit-invariant singular value decomposition (UISVD), which yields spectra and entropy measures invariant under diagonal rescalings (i.e., change of local "units") of operator bases (Caputa et al., 28 Dec 2025). Given matrix $A$ , the left-, right-, or bi-unit-invariant singular values are constructed by balancing row/column norms or geometric means; entropies built from these UISVD spectra are strictly invariant under arbitrary diagonal transformations, providing well-defined entropy measures in contexts (non-Hermitian quantum mechanics, random matrix theory, topological states) where ordinary entanglement entropy lacks this invariance.

UISVD-based entropies are especially relevant in biorthogonal quantum mechanics and physical settings where local normalizations or metrics are physical gauge freedoms. Their strict entropy-invariant scaling under (unit) rescalings makes them robust descriptors in quantum chaos, holographic contexts, and beyond (Caputa et al., 28 Dec 2025).

6. Scaling Entropy Growth Gaps and Group-Theoretic Constraints

In the context of group actions and amenable group ergodic theory, scaling entropy growth gaps reflect a group-level rigidity: certain amenable groups (notably the group of all finite permutations $S_\infty$ and the Houghton group $\mathcal H_2$ ) force a universal lower bound on the scaling entropy rate of any free probability-measure-preserving action (Veprev, 2023). For these groups, the scaling entropy invariant $h_{sc}$ satisfies

$h_{sc}(G \curvearrowright X, \mu) \succeq [\rho(n)]$

for all free actions, where $\rho(n)$ can be taken as $\log \log|F_n|$ for appropriate Følner sequences. Thus entropy-order invariance of growth rates becomes a group-theoretic invariant, tightly linking combinatorial, spectral, and information-theoretic features of group actions.

This group-induced entropy-invariant scaling distinguishes groups with rigid slow-entropy spectra from those permitting arbitrarily slow scaling entropy growth, charting a classification landscape parallel to that for classical measure-theoretic entropy.

Maximum entropy principles—supplemented by suitable dynamical or combinatorial constraints—yield spatial scaling laws robust under rescalings and invariant with respect to system size. For example, maximizing the Shannon entropy of friendship distance subject to a “cost” constraint in a 2D lattice yields the empirically observed

$P(r) \propto \frac{1}{r}$

for the probability of a friendship at distance $r$ , which is strictly invariant under changes of the metric scale (Hu et al., 2010). This entropy-invariant scaling of the distribution drives both universality in network topology and navigability properties, paralleling the emergence of power-law tails in information-rich complex systems.

Conclusion

Entropy-invariant scaling unifies a spectrum of phenomena in mathematical physics, dynamical systems, probability theory, combinatorics, and complex networks. It reflects the emergence of universal, transformation-invariant entropy laws—captured by scaling sequences, universal scaling functions, and group-theoretic bounds—that underlie complexity, randomness, and order in deterministic, stochastic, and quantum domains. These invariants classify systems more finely than classical entropy; they elucidate universality classes beyond conformal or Markovian limits, and they provide new axes—resilience, robustness, geometric origin—for the structure of probability, information, and entanglement in high-dimensional, interacting, and dynamically evolving structures (Zatitskiy, 2014, Vershik et al., 2023, Ackerman et al., 2018, Zhao et al., 2021, Zhou et al., 2024, Zhou et al., 2023, Carfora et al., 2018, Frank, 2016, Frank et al., 2010, Caputa et al., 28 Dec 2025, Veprev, 2023, Hu et al., 2010).