Coupled Stretched Exponential Distributions

• Coupled stretched exponential distributions are non-exponential probability laws that separate linear (scale) uncertainty from nonlinear tail-heaviness using parameters μ, Σ, α, and κ.
• This family includes well-known distributions like generalized Pareto and Student’s t, with decay behaviors transitioning from exponential to power law as κ and α vary.
• The coupled entropy Hκ uniquely quantifies uncertainty at an informational scale, ensuring composability and extensivity in complex systems such as particle collisions and glassy materials.

Coupled stretched exponential distributions are a family of non-exponential probability laws on $\mathbb R^d$ parameterized by location $\boldsymbol\mu$, positive-definite scale matrix $\Sigma$, stretching exponent $\alpha>0$, and coupling parameter $\kappa>-\tfrac1d$. In the formulation of "On the uniqueness of the coupled entropy," they are presented as distributions that uniquely parameterize linear uncertainty with the scale and nonlinear uncertainty with the tail shape for a broad class of complex systems. Their density is organized by a Mahalanobis-like radius, includes the generalized Pareto and Student’s $t$ distributions as special cases, and is paired with a corresponding coupled entropy $H_\kappa$ that is claimed to be uniquely determined by uncertainty at the informational scale $\sigma$ (Nelson, 21 Nov 2025).

1. Formal definition and geometric parameterization

Let $d$ denote dimension, $\boldsymbol\mu\in\mathbb R^d$ a location vector, $\boldsymbol\mu$0 a positive-definite scale matrix, $\boldsymbol\mu$1 the stretching exponent, and $\boldsymbol\mu$2 the coupling parameter. The defining radial coordinate is

$\boldsymbol\mu$3

The coupled stretched exponential distribution has density

$\boldsymbol\mu$4

where $\boldsymbol\mu$5 can be omitted for $\boldsymbol\mu$6, while for $\boldsymbol\mu$7 it indicates truncation at zero. The exponent can be written as

$\boldsymbol\mu$8

which ensures that for large $\boldsymbol\mu$9 the density decays as the power law $\Sigma$0 (Nelson, 21 Nov 2025).

Normalization is defined by the partition function

$\Sigma$1

After transforming to spherical coordinates with surface area $\Sigma$2, the normalization reduces to a one-dimensional radial integral, and after the substitution $\Sigma$3 the paper states that the result becomes a Beta-integral involving $\Sigma$4. This formulation makes explicit that the family is controlled by radial stretching through $\Sigma$5 and tail-shape coupling through $\Sigma$6.

2. Specializations and limiting regimes

The family includes several standard distributions as exact parameter specializations. For $\Sigma$7 and $\Sigma$8, the density reduces to the Type I or Type II Pareto distribution, depending on support, with tail exponent $\Sigma$9. For $\alpha>0$0 and $\alpha>0$1, the distribution becomes Student’s $\alpha>0$2 after the reparameterization $\alpha>0$3 (Nelson, 21 Nov 2025).

Parameters	Resulting distribution	Statement in the source
$\alpha>0$4	Type I or Type II Pareto	Tail exponent $\alpha>0$5
$\alpha>0$6	Student’s $\alpha>0$7	Identical under $\alpha>0$8
$\alpha>0$9	Exponential-tail limit	Recovers the Shannon case

These specializations clarify the scope of the family. When $\kappa>-\tfrac1d$0, exponential tails are recovered. When $\kappa>-\tfrac1d$1, the source states that the asymptotic form approaches a pure power law $\kappa>-\tfrac1d$2. This suggests that the coupled stretched exponential family interpolates between exponential-type behavior and heavy-tailed behavior within a single parameterization.

The same paper summarizes the family as a four-parameter class $\kappa>-\tfrac1d$3 that interpolates between stretched exponentials near the center and pure power laws in the tails. In that sense, the term “stretched exponential” refers not to a single global decay law, but to a radial body-tail construction in which the central region and far-tail regime are controlled separately.

3. Interpretation of the parameters

The source assigns distinct operational roles to the scale, coupling, and stretch parameters. In the univariate notation $\kappa>-\tfrac1d$4, the scale controls the informational scale at which the distribution transitions from central body to tail. It is defined by the property

$\kappa>-\tfrac1d$5

Shifting $\kappa>-\tfrac1d$6 for a coupled Gaussian changes the width but leaves the tail exponent fixed at $\kappa>-\tfrac1d$7 (Nelson, 21 Nov 2025).

The coupling parameter $\kappa>-\tfrac1d$8 governs tail heaviness. As $\kappa>-\tfrac1d$9, one recovers exponential tails; as $t$0, one approaches the stated pure power-law regime. The source gives as an example that particle transverse-momentum distributions in proton-proton collisions often fit best with $t$1–0.3. The stretching exponent $t$2 controls the shape near $t$3: $t$4 yields “stretched” bodies with sharper central peaks, while $t$5 yields “compressed” bodies. As an example, the source states that relaxation processes in glassy materials sometimes follow $t$6–0.8.

Parameter	Role	Example stated in the source
$t$7	Informational scale	Width changes, tail exponent fixed
$t$8	Tail heaviness	Proton–proton collisions: $t$9–0.3
$H_\kappa$0	Central-body shape	Glassy relaxation: $H_\kappa$1–0.8

This parameter separation is central to the paper’s interpretation of uncertainty. Scale is treated as the linear contribution, while coupling captures nonlinear uncertainty associated with tail shape. A plausible implication is that the formalism is intended to disentangle body-width effects from genuine heavy-tail effects in settings where both are present.

4. Coupled entropy and uncertainty at scale

A central claim of the 2025 paper is that the coupled entropy $H_\kappa$2 is uniquely determined by requiring that a generalized entropy measure uncertainty at the scale $H_\kappa$3 for a class of non-exponential distributions. For the coupled stretched exponential family, the entropy is expressed as

$H_\kappa$4

where the coupled logarithm is

$H_\kappa$5

In this representation, the term $H_\kappa$6 is identified as the linear contribution from the stretching exponent, while the generalized logarithm of the partition function quantifies the nonlinear uncertainty due to the heavy-tail shape $H_\kappa$7 (Nelson, 21 Nov 2025).

The same source states that, in the limit $H_\kappa$8, $H_\kappa$9 and the generalized logarithm reduces to the ordinary logarithm, recovering the Shannon case. This positioning of $\sigma$0 is broader than a mere entropy redefinition: the paper argues that the entropy is the unique measure compatible with the informational scale for the given class of distributions.

The paper further frames this construction as optimizing the representation of uncertainty due to linear sources. This suggests a conceptual division in which ordinary scale effects are absorbed into the linear component of the entropy, whereas tail nonlinearity is isolated in the generalized logarithmic term.

5. Composability, extensivity, and thermodynamic connections

Two formal properties are emphasized: composability and extensivity. For two statistically independent subsystems $\sigma$1 and $\sigma$2 with the same coupling $\sigma$3, the joint entropy satisfies

$\sigma$4

The proof sketch given in the source relies on the coupled-sum property of the generalized logarithm together with factorization of the partition functions $\sigma$5 (Nelson, 21 Nov 2025).

For $\sigma$6 identical, independent coupled-exponential subsystems, the paper states that if the number of microstates scales as a coupled stretched exponential of $\sigma$7, then extensivity in the sense $\sigma$8 for $\sigma$9 requires the parameter relations $d$0 and $d$1. Under that choice, the entropy recovers linear growth in $d$2.

Thermodynamic relationships are presented as additional support for the uniqueness claim. For the one-dimensional coupled exponential with $d$3 and $d$4, the generalized temperature $d$5 is stated to be equal to the informational scale $d$6. The coupled free energy is given in terms of $d$7, $d$8, and the generalized logarithm of $d$9. The same framework is connected to equilibrium versus non-equilibrium behavior in systems with multiplicative noise or high-energy particle collisions. The paper also reviews applications of the coupled entropy in measuring statistical complexity, training variational inference algorithms, and designing communication channels.

6. Relation to stretched-exponential laws in interacting systems

The phrase “stretched exponential” also appears in a different but related research context: lifetime statistics in interacting multi-species systems. In "A Universal Lifetime Distribution for Multi-Species Systems," the observed lifetime distribution is not formulated as a coupled stretched exponential density on $\boldsymbol\mu\in\mathbb R^d$0, but as an effective lifetime law

$\boldsymbol\mu\in\mathbb R^d$1

That result arises in a dynamical graph model with immigration and extinction cascades, where each species has fitness $\boldsymbol\mu\in\mathbb R^d$2, survives only while $\boldsymbol\mu\in\mathbb R^d$3, and is removed when its fitness becomes negative. After each immigration plus ensuing extinctions, the system returns to a state with all $\boldsymbol\mu\in\mathbb R^d$4, and in the long run self-organizes into a statistical steady state with fluctuating $\boldsymbol\mu\in\mathbb R^d$5 (Murase et al., 2015).

The heuristic explanation is a “modified Red–Queen” hypothesis. Simulations and mean-field arguments indicate that the mortality rate of a typical resident species is independent of age once the species is not extremely young, but depends on the current number of coexisting species $\boldsymbol\mu\in\mathbb R^d$6. In steady state, $\boldsymbol\mu\in\mathbb R^d$7 fluctuates with approximately an exponential stationary distribution, $\boldsymbol\mu\in\mathbb R^d$8, and because extinctions occur roughly one at a time, the characteristic lifetime scales as $\boldsymbol\mu\in\mathbb R^d$9. Mixing exponential conditional lifetimes $\boldsymbol\mu$00 over exponentially distributed time scales $\boldsymbol\mu$01 yields

$\boldsymbol\mu$02

with asymptotic form

$\boldsymbol\mu$03

and therefore effectively $\boldsymbol\mu$04.

This comparison is useful because it distinguishes two meanings of “stretched exponential.” In the coupled stretched exponential distribution of (Nelson, 21 Nov 2025), the term refers to a parametric probability family with explicit coupling, scale, and tail-shape structure. In the multi-species lifetime model of (Murase et al., 2015), the stretched exponential emerges as an effective marginal law produced by a mixture of exponentials with an exponential distribution of time scales. This suggests that similar functional forms can arise from different mechanisms: one as a generalized distribution family with associated entropy, the other as a universal lifetime signature of migration-driven interacting-species systems.