Sub-Poisson Distributions: Theory & Applications

Updated 8 July 2026

Sub-Poisson distributions are statistical models defined by exhibiting less variability than Poisson benchmarks, as measured by MGF domination, variance-to-mean ratios, or directional-convex order.
They find applications across probability theory, count-data models, spatial point processes, and spectral statistics, influencing fields like gene regulation and random-matrix theory.
Methodologies such as precise moment bounds, concentration inequalities, and dcx comparisons provide actionable insights for modeling regularity and controlling fluctuations.

Sub-Poisson distributions are not a single universally standardized class. Across probability, stochastic modeling, spatial statistics, and random-matrix theory, the term denotes objects that are compared to a Poisson benchmark and found to be more regular, less variable, or otherwise “below Poisson” in a precise technical sense. In one probabilistic usage, a non-negative random variable with mean $\mu$ is sub-Poissonian when its MGF is dominated by that of a Poisson law of mean $\mu$ ; in count-data modeling, sub-Poisson often means underdispersed, typically $\mathrm{Var}(X)<\E X$ or Fano factor $<1$ ; in point-process theory, it refers to directionally-convex domination by a Poisson point process with the same mean measure; and in spectral statistics it designates level-spacing laws of the form $a s e^{-b s^c}$ with $c<1$ (Ahle, 2021, Blaszczyszyn et al., 2010, Kumar et al., 2017).

1. Terminological scope and Poisson benchmarks

The cited literature uses “sub-Poisson” in several non-equivalent ways. What is common is the choice of Poisson behavior as a reference object: the comparison may be made at the level of MGFs, raw moments, variance-to-mean ratios, spatial ordering, or spectral spacing laws. The technical consequences therefore depend on the ambient category of random object.

Domain	Defining comparison	Poisson feature being benchmarked
Non-negative random variables	MGF domination	moments and tails
Count distributions	$\mathrm{Var}(X)<\E X$ or $\mathrm{FF}<1$	dispersion
Point processes	$\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$	clustering and void structure
Spectral statistics	$p(s)=a s e^{-b s^c}$ with $\mu$ 0	level-spacing profile

In the variance-based count-model literature, “sub-Poissonian” is often synonymous with underdispersion. In the gene-regulation setting, the defining condition is $\mu$ 1 (Iyer-Biswas et al., 2011). In the point-process setting, by contrast, the notion is order-theoretic: a point process is sub-Poisson if it is directionally-convexly dominated by a Poisson point process with the same mean measure (Blaszczyszyn et al., 2010). In random-matrix theory, the term is used for intermediate nearest-level-spacing statistics with weaker repulsion than Wigner-Dyson, and specifically for the parameter regime $\mu$ 2 inside the family $\mu$ 3 (Kumar et al., 2017).

A further source of ambiguity is that Poisson also appears in adjacent but distinct phrases such as “compound Poisson” and “subexponential densities.” The class $\mu$ 4 of subexponential densities for compound Poisson sums concerns convolution asymptotics such as $\mu$ 5, not sub-Poissonianity in any of the senses above (Shimura et al., 2020). Likewise, a study of the Poisson benchmark for the minimum of $\mu$ 6 explicitly notes that it does not define sub-Poissonianity via variance or moment comparison, although it uses Poisson concentration around the mean as a reference point (Li et al., 2022).

2. MGF domination and raw-moment theory

A central probabilistic definition is the sub-Poissonian MGF bound for a non-negative random variable $\mu$ 7 with mean $\mu$ 8: $\mu$ 9 This class includes the Poisson distribution, the Binomial distribution $\mathrm{Var}(X)<\E X$0, and sums of independent $\mathrm{Var}(X)<\E X$1-bounded random variables. It excludes distributions such as the geometric and negative binomial, whose MGFs exceed the Poisson bound (Ahle, 2021).

For such $\mathrm{Var}(X)<\E X$2, the normalized raw moments admit the uniform bound

$\mathrm{Var}(X)<\E X$3

valid for all $\mathrm{Var}(X)<\E X$4 (Ahle, 2021). The paper presents this as a sharp and simple inequality, improving previous uniform bounds by a factor exponential in $\mathrm{Var}(X)<\E X$5.

The asymptotic structure is two-regime. When $\mathrm{Var}(X)<\E X$6, the normalized raw moment approaches $\mathrm{Var}(X)<\E X$7, and the estimate is asymptotically tight: $\mathrm{Var}(X)<\E X$8 for Poisson $\mathrm{Var}(X)<\E X$9, and also for Binomial $<1$ 0 with suitable $<1$ 1 (Ahle, 2021). This is supported by the lower bound

$<1$ 2

for Poisson, and similarly for suitable binomial laws.

In the opposite regime $<1$ 3, the optimal moment growth satisfies

$<1$ 4

with matching lower bounds from Poisson moments and Bell numbers up to constant-order corrections of the same logarithmic type (Ahle, 2021). This establishes asymptotic sharpness both for moderate $<1$ 5 and for large $<1$ 6.

The significance of this formulation is that it yields explicit bounds depending on $<1$ 7 rather than only on universal constants. Earlier bounds of the form

$<1$ 8

lose an exponential factor in $<1$ 9 relative to the refined estimate above (Ahle, 2021). This suggests that MGF-based sub-Poissonianity is especially effective when precise moderate-moment control is required.

3. Centered sub-Poisson concentration and variance proxies

A more recent framework extends the MGF-domination idea to centered random variables and develops a nonasymptotic concentration theory. With

$a s e^{-b s^c}$ 0

a random variable $a s e^{-b s^c}$ 1 is upper sub-Poisson with variance proxy $a s e^{-b s^c}$ 2 if

$a s e^{-b s^c}$ 3

Lower sub-Poisson means that $a s e^{-b s^c}$ 4 is upper sub-Poisson, and two-sided sub-Poisson means the same bound holds for all $a s e^{-b s^c}$ 5 (Leskelä et al., 16 Aug 2025).

Within this framework, the optimal sub-Poisson variance proxy is the smallest $a s e^{-b s^c}$ 6 for which the centered MGF inequality holds for all $a s e^{-b s^c}$ 7. The paper states that finiteness of this proxy is equivalent to sub-Poissonianity, that the proxy dominates the ordinary variance, and that it is zero if and only if $a s e^{-b s^c}$ 8 is deterministic (Leskelä et al., 16 Aug 2025). It also derives bounds connecting the proxy to sub-Gaussian and sub-exponential Orlicz norms.

The principal concentration result is a Bennett-type inequality without boundedness assumptions. If $a s e^{-b s^c}$ 9 is upper sub-Poisson with variance proxy $c<1$ 0, then for $c<1$ 1,

$c<1$ 2

A Bernstein-type corollary is

$c<1$ 3

The paper presents this as a tail theory tailored to Bernoulli and Poisson variables, and to signed versions of these variables, in their natural tail regime (Leskelä et al., 16 Aug 2025).

The same work proves that sub-Poissonianity is closed under independent sums and convex combinations, but not under all linear operations such as scalar multiplication (Leskelä et al., 16 Aug 2025). This suggests a structural analogy with sub-Gaussian theory, but with $c<1$ 4 replacing the quadratic log-MGF.

4. Underdispersion in discrete count models

In several count-data literatures, a sub-Poisson distribution is an underdispersed count law, typically characterized by variance smaller than the mean. In stochastic auto-regulation models, the relevant quantity is the Fano factor

$c<1$ 5

with sub-Poisson behavior defined by $c<1$ 6 (Iyer-Biswas et al., 2011). The steady-state protein count distribution is represented as a Poisson mixture,

$c<1$ 7

If $c<1$ 8 is a positive, normalizable density, then the mixture is super-Poisson, with $c<1$ 9; sub-Poisson behavior therefore requires parameter regimes in which no such positive mixing density exists (Iyer-Biswas et al., 2011).

For the auto-repression model, the effective parameter

$\mathrm{Var}(X)<\E X$0

controls this transition. The sub-Poisson regime occurs when $\mathrm{Var}(X)<\E X$1, whereas auto-activation remains in the $\mathrm{Var}(X)<\E X$2 regime and cannot produce sub-Poisson statistics (Iyer-Biswas et al., 2011). The paper further states that the region of parameter space yielding sub-Poisson behavior under auto-repression is narrow.

A different route to underdispersion is obtained by Markov chain-based generalizations of discrete parent laws, including Poisson-derived models. In that construction, the mean number of successes is

$\mathrm{Var}(X)<\E X$3

and dispersion is controlled by $\mathrm{Var}(X)<\E X$4: overdispersion when $\mathrm{Var}(X)<\E X$5, equidispersion when $\mathrm{Var}(X)<\E X$6, and underdispersion or sub-Poisson behavior when $\mathrm{Var}(X)<\E X$7 (Baker, 2020). The mechanism is explicit: when $\mathrm{Var}(X)<\E X$8, successive successes are negatively correlated and fluctuations are suppressed.

The mean-parametrized Conway–Maxwell–Poisson distribution provides an extreme underdispersion result. For fixed mean $\mathrm{Var}(X)<\E X$9, as the underdispersion parameter $\mathrm{FF}<1$ 0, the limiting law is a point mass at $\mathrm{FF}<1$ 1 when $\mathrm{FF}<1$ 2 is an integer, and a shifted Bernoulli on $\mathrm{FF}<1$ 3 and $\mathrm{FF}<1$ 4 with weights given by the fractional part of $\mathrm{FF}<1$ 5 when $\mathrm{FF}<1$ 6 is non-integer (Huang, 2020). The paper states that this limiting law is the most underdispersed discrete distribution possible for a given mean and that this is currently the only known generalization of the Poisson distribution exhibiting arbitrary underdispersion for any mean.

The discrete weak-stability literature imposes a further restriction. In the class of Poisson–delayed Sibuya distributions, which includes Poisson and Hermite laws, the paper states that only the Poisson law is sub-Poissonian in the sense of variance not exceeding the mean; there are no other nontrivial sub-Poissonian discrete stable laws in that family (Aldridge, 15 Sep 2025). This corrects a common overgeneralization: Poisson-related families need not contain genuinely underdispersed non-Poisson members.

5. Sub-Poisson point processes and spatial connectivity

For point processes, sub-Poissonianity is defined through the directionally-convex order. A function $\mathrm{FF}<1$ 7 is directionally convex (dcx) if for vectors $\mathrm{FF}<1$ 8 with $\mathrm{FF}<1$ 9 and $\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 0,

$\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 1

For point processes $\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 2, one writes $\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 3 when for every finite collection of bounded Borel sets $\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 4,

$\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 5

It is enough to verify this for mutually disjoint $\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 6 (Blaszczyszyn et al., 2010).

A point process $\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 7 is sub-Poisson if

$\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 8

where $\Phi \le_{dcx} \Phi_{\mathrm{Poi}}$ 9 is a Poisson point process with the same mean measure (Blaszczyszyn et al., 2010). The paper also introduces weaker notions, including weakly sub-Poisson point processes defined by the joint-intensity inequalities

$p(s)=a s e^{-b s^c}$ 0

for all $p(s)=a s e^{-b s^c}$ 1, almost everywhere, in the stationary case of intensity $p(s)=a s e^{-b s^c}$ 2 (Blaszczyszyn et al., 2010).

The geometric interpretation is that point processes smaller in dcx order are less variable and more regular than Poisson; they exhibit less clustering and fewer voids. The order is compatible with comparisons of Ripley’s $p(s)=a s e^{-b s^c}$ 3-function, correlation functions, and shot-noise fields (Blaszczyszyn et al., 2010).

These comparisons have concrete consequences for continuum percolation. For a homogeneous dcx-sub-Poisson point process $p(s)=a s e^{-b s^c}$ 4 of intensity $p(s)=a s e^{-b s^c}$ 5, the critical communication radius in the Boolean or Gilbert model satisfies

$p(s)=a s e^{-b s^c}$ 6

so the classical non-degenerate phase transition for Poisson networks extends to homogeneous sub-Poisson networks (Blaszczyszyn et al., 2010). The paper also extends analogous percolation results for the SINR graph when the interferers form an $p(s)=a s e^{-b s^c}$ 7-sub-Poisson point process or when the backbone nodes form a sub-Poisson point process. As a model class, perturbed lattices are given as examples of sub-Poisson point processes (Blaszczyszyn et al., 2010).

In random-matrix theory, “sub-Poisson” refers neither to MGF domination nor to underdispersion of a count law. The relevant object is the nearest-level-spacing distribution

$p(s)=a s e^{-b s^c}$ 8

introduced for ensembles of real pseudo-symmetric matrices satisfying

$p(s)=a s e^{-b s^c}$ 9

for an appropriate metric $\mu$ 00 (Kumar et al., 2017). The special case $\mu$ 01 is the semi-Poisson law

$\mu$ 02

while the regime $\mu$ 03 is explicitly identified as sub-Poisson (Kumar et al., 2017).

The paper reports numerical evidence that ensembles of large pseudo-symmetric matrices with $\mu$ 04 independent entries in the range

$\mu$ 05

have nearest-level-spacing histograms well fitted by $\mu$ 06, often with $\mu$ 07 (Kumar et al., 2017). Selected examples listed in the paper include $\mu$ 08 and $\mu$ 09 with $\mu$ 10, $\mu$ 11 with $\mu$ 12, and $\mu$ 13 with $\mu$ 14. The authors emphasize that these fits are robust with respect to changes in the distribution of matrix elements.

For $\mu$ 15 pseudo-symmetric matrices, the spacing law can be derived analytically. In the Gaussian case,

$\mu$ 16

where $\mu$ 17 is the modified Bessel function of the second kind (Kumar et al., 2017). The paper states that this form lies close to semi-Poisson and to the broader sub-Wigner/sub-Poisson family, with linear level repulsion but less repulsion than Wigner-Dyson statistics.

The physical contexts cited for these sub-Poisson or sub-Wigner spacing laws include Anderson metal-insulator transitions, topological transitions in Josephson junctions, and PT-symmetric systems near eigenvalue coalescence (Kumar et al., 2017). At the same time, the paper is explicit that for large $\mu$ 18 the claim is conjectural and numerically supported rather than analytically established. This marks an important distinction from the probabilistic and point-process literatures, where sub-Poisson properties are given by direct inequalities or order relations rather than empirical spectral fits.