Double-Exponential Capacity Measure

• Double-Exponential Capacity Measure is a concept in real-stable polynomial theory where capacity lower bounds decay as exp(-2^(Θ(n))), highlighting a weak guarantee.
• It arises in the analysis of strongly Rayleigh distributions and metric-TSP, controlling coefficients and aggregated-event probabilities.
• Innovative productization methods have replaced these double-exponential bounds with improved, degree-independent singly exponential estimates.

Searching arXiv for recent and directly relevant papers on the term and neighboring capacity notions. A double-exponential capacity measure is not a single standardized invariant across mathematics and theoretical computer science. In the clearest explicit usage, it denotes a capacity-related lower bound whose worst-case dependence on the number of variables is doubly exponential, typically of the form $\exp(-2^{\Theta(n)})$ or $2^{-2^{\Theta(n)}}$, rather than singly exponential $\exp(-\Theta(n))$. That usage is most sharply articulated in the theory of real stable polynomials and strongly Rayleigh distributions, where capacity controls permanents, coefficients, and aggregated-event probabilities, and where earlier bounds used in metric TSP analysis were “doubly exponential in the number of variables” before being replaced by singly exponential gradient-based estimates via productization (Gurvits et al., 2020). In adjacent literatures, however, the phrase is absent or explicitly rejected as a natural description: biologically plausible associative memories and dynamic energy networks exhibit exponential, not double-exponential, storage capacity in the relevant network size, while measure-theoretic and geometric capacity theories remain in logarithmic, polynomial-scale, or exponential-time regimes (Kafraj et al., 2 Jan 2026, Karuvally et al., 28 Oct 2025, Morales, 23 Apr 2025, Karak et al., 2014).

1. Terminological scope and principal meaning

The most direct arXiv-era meaning of the phrase arises from capacity bounds for real stable polynomials. For $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$, the capacity at exponent $\mathbf{1}$ is

$\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$

and more generally

$\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$

In the strongly Rayleigh and metric-TSP setting, the relevant probabilities are controlled through such a capacity-type quantity, and the older bounds discussed there are described as “doubly exponential in the number of variables,” with asymptotic behavior $\exp(-2^{\Theta(n)})$. The replacement bounds obtained by Gurvits and Leake are degree-independent and only singly exponential in $n$, namely $\exp(-\Theta(n))$, so in this context a “double-exponential capacity measure” means a capacity bound whose guaranteed lower scale deteriorates as a double exponential in the dimension parameter (Gurvits et al., 2020).

This usage is fundamentally different from the way “capacity” is used in associative memory, dynamical systems, or metric geometry. In those areas, capacity may mean the number of stable memories, an upper-capacity dimension of a measure, or a Sobolev-type condenser capacity; none of those papers adopts double-exponential growth as the natural asymptotic regime. This suggests that the expression functions less as a universal definition than as a comparative descriptor for an especially weak asymptotic lower bound.

2. Stable-polynomial capacity and the origin of the doubly exponential regime

Real stable polynomials are multivariate polynomials $2^{-2^{\Theta(n)}}$0 such that

$2^{-2^{\Theta(n)}}$1

They include determinant polynomials and partition functions of negatively dependent distributions, and they are strongly log-concave. Within this class, capacity is tied to normalization, Hall-type obstructions, and coefficient control. In particular,

$2^{-2^{\Theta(n)}}$2

and for $2^{-2^{\Theta(n)}}$3 with $2^{-2^{\Theta(n)}}$4, vanishing capacity is equivalent to vanishing permanent and to the existence, up to permutation, of a bottom-left $2^{-2^{\Theta(n)}}$5 zero block with $2^{-2^{\Theta(n)}}$6. If $2^{-2^{\Theta(n)}}$7 and $2^{-2^{\Theta(n)}}$8, then $2^{-2^{\Theta(n)}}$9; in the special case $\exp(-\Theta(n))$0, the polynomial is called doubly stochastic and satisfies $\exp(-\Theta(n))$1. This makes $\exp(-\Theta(n))$2 a multiplicative robustness measure around a target exponent $\exp(-\Theta(n))$3 (Gurvits et al., 2020).

The phrase “double-exponential capacity measure” becomes relevant when this robustness measure is used to certify probabilities of aggregated events for strongly Rayleigh measures. In that setting, one studies a generating polynomial $\exp(-\Theta(n))$4 obtained from a strongly Rayleigh distribution after aggregating coordinates into $\exp(-\Theta(n))$5 blocks, and seeks lower bounds on coefficients

$\exp(-\Theta(n))$6

The older bounds invoked in the metric TSP analysis were degree-independent but “doubly exponential in the number of variables,” so the relevant probability or its governing capacity quantity could decay like

$\exp(-\Theta(n))$7

The double-exponential feature therefore does not mean that the polynomial itself has doubly exponential degree or state space; it means that the best certified lower bound on a capacity-controlled quantity deteriorates at that rate as a function of the number of aggregated variables.

3. Productization and the reduction to singly exponential bounds

The decisive change is the productization technique. Its structural statement is that any real stable polynomial with prescribed normalization at $\exp(-\Theta(n))$8 can, at any positive point $\exp(-\Theta(n))$9, be matched by a product of linear or affine forms with the same value and the same gradient at $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$0. In the homogeneous case, if $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$1 and $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$2, then for any $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$3 there exists a nonnegative matrix $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$4 with row sums $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$5, column sums $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$6, and

$p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$7

This reduces the minimization of capacity over all real stable polynomials to a minimization over products of linear forms, where convex-geometric and Hall-type arguments are available (Gurvits et al., 2020).

The resulting bounds are explicit. For a homogeneous real stable polynomial of degree $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$8 in $p \in \mathbb{R}_+^n[x_1,\ldots,x_n]$9 variables, if $\mathbf{1}$0 and

$\mathbf{1}$1

then

$\mathbf{1}$2

For a general real stable polynomial and $\mathbf{1}$3, if $\mathbf{1}$4 and

$\mathbf{1}$5

then

$\mathbf{1}$6

Both bounds are degree-independent and singly exponential in $\mathbf{1}$7. The threshold $\mathbf{1}$8 is best possible in the homogeneous setting because it is equivalent to the Hall-type condition that every matrix in $\mathbf{1}$9 has strictly positive permanent (Gurvits et al., 2020).

For strongly Rayleigh measures this yields explicit probability bounds. If $\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$0 is strongly Rayleigh on $\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$1, $\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$2, and

$\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$3

then

$\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$4

In the special case $\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$5,

$\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$6

The practical significance is exact: these estimates replace a double-exponential lower scale by a singly exponential one while preserving independence from the polynomial degree.

4. Why exponential memory capacity is not a double-exponential capacity measure

In associative-memory theory, “capacity” usually means the number of stable fixed points or retrievable memories, not a lower bound on a coefficient or capacity functional. Krotov and Hopfield’s earlier two-layer dense associative memory had winner-takes-all hidden dynamics and capacity at most linear in the number of hidden neurons. The threshold-based two-layer model of 2026 removes that competition: the hidden indicators satisfy

$\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$7

and with $\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$8 in the ideal Gaussian-weight regime, all $\cpc_{\mathbf{1}}(p)=\inf_{x_1,\ldots,x_n>0}\frac{p(x)}{x_1\cdots x_n},$9 binary hidden states become stable fixed points. The resulting capacity is

$\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$0

provided $\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$1 is sufficiently larger than $\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$2. The paper states explicitly that its capacity is exponential in $\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$3 and “not” double-exponential in any standard network-size metric (Kafraj et al., 2 Jan 2026).

A parallel conclusion holds for the Exponential Dynamic Energy Network. EDEN defines capacity through a bitwise retrieval criterion at transition time,

$\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$4

and derives

$\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$5

whereas the reference sequence network has

$\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$6

The paper does not use the term “double-exponential capacity,” and its own discussion emphasizes that exponential in $\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$7 is the relevant scaling for the number of distinct stored patterns in a binary $\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$8-neuron state space (Karuvally et al., 28 Oct 2025).

The contrast matters because a double-exponential capacity measure in the stable-polynomial sense is a pathology of a bound, not a superior memory law. In memory models, by contrast, exponential capacity is already the claimed achievement.

5. Measure-theoretic and geometric capacities: neighboring notions without double-exponential growth

Morales’ measure-theoretic expansion exponent introduces yet another use of “capacity.” For a measurable map $\cpc_\kappa(p)=\inf_{x>0}\frac{p(x)}{x^\kappa}, \qquad x^\kappa=x_1^{\kappa_1}\cdots x_n^{\kappa_n}.$9 on a metric space and a Borel probability measure $\exp(-2^{\Theta(n)})$0,

$\exp(-2^{\Theta(n)})$1

and the global expansion exponent satisfies

$\exp(-2^{\Theta(n)})$2

With the measure upper capacity

$\exp(-2^{\Theta(n)})$3

the main entropy inequality is

$\exp(-2^{\Theta(n)})$4

The framework is explicitly built from exponential-in-time growth and polynomial-scale covering laws. The exposition surrounding the paper makes clear that nothing in it directly captures double-exponential behavior; rather, any such theory would require modified normalizations such as $\exp(-2^{\Theta(n)})$5 or $\exp(-2^{\Theta(n)})$6, which are not part of the paper’s results (Morales, 23 Apr 2025).

Geometric potential theory gives a different comparison. In a $\exp(-2^{\Theta(n)})$7-doubling metric measure space supporting a $\exp(-2^{\Theta(n)})$8-Poincaré inequality and satisfying a chain condition, if a compact set $\exp(-2^{\Theta(n)})$9 has

$n$0

then its generalized Hausdorff measure vanishes for the logarithmic gauge

$n$1

The paper further derives a quantitative lower bound

$n$2

Here the refined scale is logarithmic, not double-exponential: the theory operates at the borderline between Sobolev capacity and generalized Hausdorff measure, and its advertised optimality is “for logarithmic gauge functions” (Karak et al., 2014).

6. Conceptual synthesis and common misconceptions

Three distinct notions are often conflated under the word capacity. In real-stability and strongly Rayleigh theory, capacity is the infimum $n$3 and directly controls coefficients and probabilities. There, a double-exponential capacity measure refers to a guaranteed lower bound that may be as small as $n$4, and the main achievement is to improve that to $n$5 (Gurvits et al., 2020). In associative-memory models, capacity is the number of stable memories, with scaling $n$6 or $n$7; those papers explicitly frame their achievements as exponential rather than double-exponential (Kafraj et al., 2 Jan 2026, Karuvally et al., 28 Oct 2025). In measure-theoretic and geometric settings, capacity is either a dimension-like covering exponent or a Sobolev condenser functional, and the operative asymptotics are entropy products and logarithmic gauges, not doubly exponential laws (Morales, 23 Apr 2025, Karak et al., 2014).

A common misconception is therefore to treat “double-exponential capacity” as a stronger form of storage. In the stable-polynomial literature it denotes the opposite: an especially poor worst-case lower guarantee. Another misconception is to treat any exponential memory capacity as double-exponential after reparametrizing the network size. The 2026 associative-memory paper addresses this directly: its $n$8 law becomes $n$9 only under an artificial choice $\exp(-\Theta(n))$0, and the paper states that this is not the natural size metric (Kafraj et al., 2 Jan 2026).

The most defensible encyclopedic definition is thus context-sensitive. In current research usage, the term most naturally denotes the doubly exponential deterioration of capacity-based lower bounds in the number of variables, especially in the analysis of real stable polynomials, strongly Rayleigh distributions, and metric-TSP-style aggregated events. Outside that setting, the phrase is usually absent, and neighboring theories instead speak of exponential storage capacity, upper-capacity dimension, or logarithmic Hausdorff gauges.