arxiv: 2604.26304 · v1 · submitted 2026-04-29 · 🧮 math.PR · cs.NA· math.NA

Recognition: unknown

Optimization-Free Concentrated Matrix-Exponentials

Maria Laura Battagliola , Oscar Peralta

Authors on Pith no claims yet

Pith reviewed 2026-05-07 12:53 UTC · model grok-4.3

classification 🧮 math.PR cs.NAmath.NA

keywords matrix-exponential distributionsFejér kernelErlang boundphase-type distributionsconcentrated delaysprobability densitiesasymptotic varianceclosed-form moments

0 comments

The pith

Raising the Fejér kernel to a logarithmic power produces matrix-exponential distributions whose variance falls below the Erlang limit.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper gives an explicit family of matrix-exponential densities centered on a unit delay. These densities are formed by taking the trigonometric Fejér kernel and raising it to a power proportional to the logarithm of a concentration parameter. The resulting objects come with closed-form parameters and exact moment formulas, removing the need for numerical fitting. This matters for modeling near-deterministic positive times in stochastic systems, because matrix-exponential representations remain analytically tractable while achieving tighter concentration than Erlang or phase-type distributions allow. A reader who accepts the construction gains a concrete, provable way to build low-variance delay models without optimization.

Core claim

The authors prove that the density formed by raising the trigonometric Fejér kernel to a logarithmic power admits an exact matrix-exponential representation. The representation has closed-form parameters, its moments are available in closed form, and its variance tends to zero faster than the Erlang bound as the logarithmic power grows, furnishing the first analytical example of a matrix-exponential class that asymptotically surpasses the Erlang variance limit for the same mean.

What carries the argument

The mechanism is the logarithmic powering of the trigonometric Fejér kernel, which produces a probability density that exactly matches the absorption-time law of a finite-state continuous-time Markov chain.

If this is right

Moments and parameters become available in closed form, eliminating numerical optimization steps in model construction.
Analytical asymptotic statements about variance reduction can now be written directly for this matrix-exponential family.
The distributions can be substituted into Markovian queueing or renewal models while preserving exact solvability.
The construction supplies a concrete benchmark against which other low-variance matrix-exponential proposals can be compared.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar explicit constructions might be obtained by raising other positive trigonometric kernels to logarithmic powers, potentially yielding families with different tail or concentration properties.
In applied settings the closed-form moments could be used to match low-variance empirical data without iterative fitting routines.
The same kernel technique might extend to time-inhomogeneous or marked point processes where concentration around a deterministic delay is required.

Load-bearing premise

The powered Fejér kernel density must exactly equal a matrix-exponential distribution whose parameters produce the claimed closed-form moments and variance behavior.

What would settle it

Compute the matrix-exponential parameters from the explicit density formula and check whether the variance of the resulting distribution lies strictly below the Erlang variance for the same mean once the logarithmic power exceeds a moderate threshold.

read the original abstract

Near-deterministic positive delays require highly concentrated distributions, but phase-type models are constrained by the Erlang variance limit. While matrix-exponential distributions can empirically bypass this barrier, prior low-variance constructions relied entirely on numerical optimization. We propose an explicit family of concentrated matrix-exponential densities for the unit delay, obtained by raising the trigonometric Fej\'er kernel to logarithmic power. With exact moments and closed-form parameters, this gives the first analytical proof of a matrix-exponential class that asymptotically surpasses the Erlang bound.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives an explicit Fejér-kernel construction for matrix-exponentials that claims to beat the Erlang variance bound analytically and without optimization, but the step confirming it stays matrix-exponential needs direct verification from the derivations.

read the letter

The main takeaway is that this work supplies a closed-form family of matrix-exponential densities for unit-mean positive random variables whose variance shrinks faster than the Erlang case. They raise the trigonometric Fejér kernel to a logarithmic power, normalize, and assert that the result has exact moments plus matrix parameters given in closed form. That is the concrete advance over prior numerical fitting approaches for low-variance phase-type or matrix-exponential models. Applications in queueing and reliability that need tight delay distributions would benefit if the formulas hold, since they avoid solving optimization problems at each use. The paper does a clean job stating the construction and the claimed asymptotic improvement. It also supplies the moment expressions directly, which is practical for plugging into larger calculations. The soft spot is the matrix-exponential property itself. Raising a trigonometric polynomial to a non-integer power yields a function whose Fourier series is infinite, so it is not obvious that the normalized version on the positive line has a rational Laplace transform or admits a finite matrix representation T with the stated entries. The abstract does not include the explicit T or the Laplace-transform argument, and the stress-test note correctly flags this mapping as the least secured step. If the full derivations show how the kernel power produces a genuine matrix-exponential with the claimed closed-form parameters and variance asymptotics, then the central claim stands; otherwise the novelty rests on an unverified step. This is for readers working on stochastic modeling who need analytical low-variance positive distributions rather than fitted ones. A queueing theorist or reliability engineer could extract usable formulas if the verification passes. It deserves a serious referee to check the matrix representation, the Laplace transform, and the asymptotic comparison against the Erlang bound.

Referee Report

1 major / 2 minor

Summary. The manuscript claims to provide an explicit, optimization-free family of concentrated matrix-exponential (ME) densities for the unit delay. The construction is based on raising the trigonometric Fejér kernel to a logarithmic power, yielding densities with exact moments, closed-form parameters, and asymptotic variance that surpasses the Erlang bound. This is presented as the first analytical proof of an ME class with this property.

Significance. If the construction is correct and the resulting distributions are indeed matrix-exponential with the claimed closed-form expressions, this would be a notable contribution to the theory of phase-type and matrix-exponential distributions, offering a way to achieve highly concentrated distributions without numerical fitting.

major comments (1)

[Construction and main theorem (body of paper)] The central construction: the claim that raising the trigonometric Fejér kernel to a logarithmic power produces a density that admits a matrix-exponential representation (i.e., exactly equal to α exp(T t) s for finite matrix T with closed-form entries) lacks a derivation of T, a Laplace-transform calculation, or an argument that the resulting function on [0,∞) has a rational Laplace transform. A non-integer power of a trigonometric polynomial is not itself a trigonometric polynomial and has an infinite Fourier series; the mapping to an ME density while preserving the ME property is therefore the load-bearing step whose correctness is not secured.

minor comments (2)

[Abstract] The abstract could be expanded to indicate the specific range of the logarithmic power parameter and the resulting matrix dimension.
[Notation and preliminaries] Notation for the normalized kernel and the matrix parameters should be introduced with explicit definitions before use in moment calculations.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful and constructive review. The major comment correctly identifies a gap in the presentation of the central construction, and we will revise the manuscript to address it directly.

read point-by-point responses

Referee: The central construction: the claim that raising the trigonometric Fejér kernel to a logarithmic power produces a density that admits a matrix-exponential representation (i.e., exactly equal to α exp(T t) s for finite matrix T with closed-form entries) lacks a derivation of T, a Laplace-transform calculation, or an argument that the resulting function on [0,∞) has a rational Laplace transform. A non-integer power of a trigonometric polynomial is not itself a trigonometric polynomial and has an infinite Fourier series; the mapping to an ME density while preserving the ME property is therefore the load-bearing step whose correctness is not secured.

Authors: We agree that the manuscript as submitted does not contain an explicit derivation of the matrix T or the Laplace-transform calculation confirming rationality. The Fejér kernel is a finite trigonometric polynomial, but the non-integer power indeed produces an infinite Fourier series. In the revision we will add a dedicated subsection that (i) writes the powered kernel in exponential form, (ii) performs the change of variables that maps the construction to a density on [0,∞) with unit mean, and (iii) evaluates the Laplace transform by direct integration, showing that the resulting expression is a ratio of polynomials whose coefficients are given in closed form. From this rational transform we will extract the companion matrix T, the initial vector α, and the exit vector s, all with explicit entries. This addition will make the ME property fully rigorous. revision: yes

Circularity Check

0 steps flagged

No circularity: explicit analytical construction stands on its own

full rationale

The paper advances an explicit construction of matrix-exponential densities by raising the trigonometric Fejér kernel to a logarithmic power, asserting that this yields closed-form parameters and exact moments that asymptotically beat the Erlang bound. No quoted equation or derivation step reduces the claimed moments, variance behavior, or matrix parameters to a fitted quantity defined by the same construction, nor does any load-bearing premise collapse to a self-citation or self-definition. The derivation is presented as a direct, optimization-free analytical family whose correctness rests on verifying the matrix-exponential representation and moment formulas, which are independent of the target asymptotic claim.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The construction rests on the assumption that the powered Fejér kernel defines a valid density admitting a matrix-exponential representation; no free parameters are mentioned in the abstract.

axioms (1)

domain assumption Raising the trigonometric Fejér kernel to a logarithmic power yields a valid probability density on the positive reals that admits a matrix-exponential representation.
This is the central step asserted in the abstract for obtaining closed-form parameters and the variance result.

pith-pipeline@v0.9.0 · 5373 in / 1174 out tokens · 135822 ms · 2026-05-07T12:53:27.260746+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

8 extracted references · 7 canonical work pages

[1]

The least variable phase type distribution is Erlang.Stochastic Models, 3(3):467–473, 1987

David Aldous and Larry Shepp. The least variable phase type distribution is Erlang.Stochastic Models, 3(3):467–473, 1987. doi:https://doi.org/10.1080/15326348708807067

work page doi:10.1080/15326348708807067 1987
[2]

Concentrated matrix exponential distributions

Ill´ es Horv´ ath, Orsolya S´ af´ ar, Mikl´ os Telek, and Bence Z´ amb´ o. Concentrated matrix exponential distributions. In D. Fiems, M. Paolieri, and A. Platis, editors,Computer Performance Engi- neering. EPEW 2016, Lecture Notes in Computer Science, volume 9951, pages 18–31. Springer, Cham, 2016. doi:https://doi.org/10.1007/978-3-319-46433-6_2

work page doi:10.1007/978-3-319-46433-6_2 2016
[3]

High order concentrated matrix-exponential distributions.Stochastic Models, 36(2):176–192, 2020

G´ abor Horv´ ath, Ill´ es Horv´ ath, and Mikl´ os Telek. High order concentrated matrix-exponential distributions.Stochastic Models, 36(2):176–192, 2020. doi: https://doi.org/10.1080/ 15326349.2019.1702058

work page arXiv 2020
[4]

Concentrated matrix exponential distributions with real eigenvalues.Probability in the Engineering and Informational Sciences, 36(4):1171–1187, 2022

Andr´ as M´ esz´ aros and Mikl´ os Telek. Concentrated matrix exponential distributions with real eigenvalues.Probability in the Engineering and Informational Sciences, 36(4):1171–1187, 2022. doi:https://doi.org/10.1017/S0269964821000309

work page doi:10.1017/s0269964821000309 2022
[5]

Numerical inverse Laplace transformation using concentrated matrix exponential distributions.Performance Evaluation, 137:102067, 2020

G´ abor Horv´ ath, Ill´ es Horv´ ath, Salah Al-Deen Almousa, and Mikl´ os Telek. Numerical inverse Laplace transformation using concentrated matrix exponential distributions.Performance Evaluation, 137:102067, 2020. doi:https://doi.org/10.1016/j.peva.2019.102067

work page doi:10.1016/j.peva.2019.102067 2020
[6]

and Nielsen, B.F

Mogens Bladt and Bo Friis Nielsen.Matrix-Exponential Distributions in Applied Probability. Probability Theory and Stochastic Modelling, volume 81. Springer New York, NY, 2017. doi: https://doi.org/10.1007/978-1-4939-7049-0

work page doi:10.1007/978-1-4939-7049-0 2017
[7]

Ill´ es Horv´ ath, Zs´ ofia Talyig´ as, and Mikl´ os Telek. An optimal inverse Laplace transform method without positive and negative overshoot – an integral based interpretation.Electronic Notes in Theoretical Computer Science, 337:87–104, 2018. doi: https://doi.org/10.1016/j.entcs. 2018.03.035

work page doi:10.1016/j.entcs 2018
[8]

3rd edition, volumes I and II combined

Antoni Zygmund.Trigonometric Series. 3rd edition, volumes I and II combined. Cambridge University Press, Cambridge, UK, 2002. 7

2002