pith. sign in

arxiv: 2605.21209 · v1 · pith:G3DZ4FJZnew · submitted 2026-05-20 · 🧮 math.PR

Sensitivity analysis of Stochastic Fluid Models: Stationary and transient quantities with applications

Anna Aksamit , Ma{\l}gorzata M. O'Reilly , Zbigniew Palmowski This is my paper

Pith reviewed 2026-05-21 01:43 UTC · model grok-4.3

classification 🧮 math.PR
keywords sensitivity analysisstochastic fluid modelsstationary distributionstransient quantitiesMarkovian modulationqueueing systemsinsurance risk processesnumerical examples
0
0 comments X

The pith

Stochastic fluid models now have explicit expressions for how their stationary and transient quantities respond to parameter changes.

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

The paper sets out to provide the first sensitivity analysis for stochastic fluid models by deriving expressions that quantify how stationary distributions and time-dependent quantities shift when parameters vary. A sympathetic reader would care because these models represent practical systems such as queues with fluctuating rates and insurance risk processes, where knowing the effect of changes in rates or claim sizes aids design and risk assessment. The authors include numerical examples in deteriorating systems and risk processes to illustrate use, and position the work as a foundation for sensitivity analysis in broader classes of Markovian modulated models.

Core claim

We establish results for the first sensitivity analysis of the stochastic fluid models (SFMs). We derive expressions for the sensitivity analysis of the key stationary and transient (time-dependent) quantities of this class of models. We also construct numerical examples to demonstrate the application potential of our methodology in queueing systems, such as deteriorating systems and insurance risk processes. This work forms foundation for the sensitivity analysis of other Markovian modulated models, which are generalisations of the SFMs, and have widespread applications.

What carries the argument

Expressions for the derivatives of stationary and transient quantities with respect to parameters in stochastic fluid models under Markovian modulation.

If this is right

  • Parameter optimization becomes feasible in queueing systems modeled by stochastic fluid models through direct computation of sensitivities.
  • Impacts of changes in claim sizes or premium rates can be quantified efficiently in insurance risk processes.
  • The approach serves as a base for extending sensitivity analysis to other Markovian modulated models.
  • Numerical examples demonstrate practical application in deteriorating systems and risk processes.

Where Pith is reading between the lines

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

  • The sensitivity expressions could support gradient-based optimization routines for controlling parameters in fluid models.
  • Possible connections exist to perturbation analysis methods used in simulation of Markov modulated processes.
  • The method might be tested on fluid approximations for communication network traffic models.

Load-bearing premise

Stochastic fluid models have well-defined stationary distributions and transient quantities that can be differentiated with respect to parameters.

What would settle it

For a simple two-state stochastic fluid model, compute the derived analytical sensitivities for a chosen parameter and compare them to finite-difference estimates obtained by direct simulation of the model at nearby parameter values; a large mismatch would falsify the expressions.

Figures

Figures reproduced from arXiv: 2605.21209 by Anna Aksamit, Ma{\l}gorzata M. O'Reilly, Zbigniew Palmowski.

Figure 1
Figure 1. Figure 1: Derivatives of the stationary quantities of the process [PITH_FULL_IMAGE:figures/full_fig_p022_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Derivatives of the stationary quantities of the process [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Derivatives of the transient quantities of the process [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Derivatives of the transient quantities of the process [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Derivatives of the transient quantities of the process [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Density h(t, θ) of the lifetime and its derivatives in Example 5.2: (top row) h(t, θ) evaluated using parameters θ in (15); and θi × ∂ ∂θi h(t; θ) for i = 1, . . . , 6 assuming the process starts from the idle phase φ(0) = 5 at time zero, according to α = [0, 0, 0, 0, 1]. 5.3 Insurer ruin example Gerber in [18] considered the classical insurance risk process Rt = x + ct − X Nt k=1 Uk, where x ≥ 0 is the in… view at source ↗
Figure 7
Figure 7. Figure 7: Example 5.3: (top row) µ, ∂ ∂θ1 ψ(x), and ∂ ∂θ2 ψ(x); and (bottom row) ψ(x). The numerical results are presented in [PITH_FULL_IMAGE:figures/full_fig_p030_7.png] view at source ↗
read the original abstract

We establish results for the first sensitivity analysis of the stochastic fluid models (SFMs). We derive expressions for the sensitivity analysis of the key stationary and transient (time-dependent) quantities of this class of models. We also construct numerical examples to demonstrate the application potential of our methodology in queueing systems, such as deteriorating systems and insurance risk processes. This work forms foundation for the sensitivity analysis of other Markovian modulated models, which are generalisations of the SFMs, and have widespread applications.

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.

Referee Report

2 major / 3 minor

Summary. The paper claims to establish the first sensitivity analysis of stochastic fluid models (SFMs) by deriving explicit expressions for the sensitivities of key stationary and transient quantities with respect to model parameters under Markov modulation. It constructs numerical examples applying these formulas to queueing systems such as deteriorating systems and insurance risk processes, and positions the results as a foundation for sensitivity analysis of more general Markovian modulated models.

Significance. If the derivations hold, the work provides a useful explicit framework for computing parameter sensitivities in SFMs without simulation, which is valuable for optimization and analysis in queueing and risk models. Credit is due for the direct differentiation approach applied to the linear system for stationary distributions and the matrix exponential/ODE for transients, both of which are differentiable on finite-dimensional spaces under standard positive-recurrence assumptions. The numerical examples demonstrate practical applicability, strengthening the contribution.

major comments (2)
  1. [§3.2, Eq. (12)] §3.2, Eq. (12): the stationary sensitivity formula is obtained by differentiating the linear balance equations, but the manuscript should explicitly verify that the matrix remains invertible in a neighborhood of the nominal parameter values used in the numerical examples of §5.1; without this, the local differentiability claim is not fully load-bearing.
  2. [§4.1, Eq. (25)] §4.1, Eq. (25): the transient sensitivity via the matrix exponential is correctly differentiated, yet the paper omits any discussion of how the formulas extend when the fluid level process hits a boundary, which is central to the risk-process application in §5.2.
minor comments (3)
  1. [Abstract] The abstract states the results are 'the first' but does not briefly indicate the differentiation technique; adding one sentence would improve clarity.
  2. [§2 and §3] Notation for the phase generator matrix is introduced in §2 but reused with slight variations in §3; a single consolidated definition would aid readability.
  3. [Figure 2] Figure 2 caption does not specify the parameter values perturbed in the sensitivity plots; this makes it harder to reproduce the numerical results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation and the constructive comments, which have helped us improve the clarity and rigor of the manuscript. We address each major comment below and indicate the revisions made.

read point-by-point responses

  1. Referee: [§3.2, Eq. (12)] §3.2, Eq. (12): the stationary sensitivity formula is obtained by differentiating the linear balance equations, but the manuscript should explicitly verify that the matrix remains invertible in a neighborhood of the nominal parameter values used in the numerical examples of §5.1; without this, the local differentiability claim is not fully load-bearing.

    Authors: We agree that an explicit statement on local invertibility strengthens the differentiability argument. In the revised manuscript we have added a remark immediately after Equation (12) recalling that the coefficient matrix is nonsingular under the positive-recurrence hypothesis maintained throughout the paper. Because the entries of this matrix are continuous (in fact, affine) functions of the modulating rates and drifts, nonsingularity persists in an open neighborhood of any nominal parameter vector at which recurrence holds. For the concrete parameter values used in §5.1 we have inserted a short numerical verification (now reported in the text) confirming that the smallest singular value remains bounded away from zero under small perturbations, thereby making the local differentiability claim fully rigorous for the examples. revision: yes

  2. Referee: [§4.1, Eq. (25)] §4.1, Eq. (25): the transient sensitivity via the matrix exponential is correctly differentiated, yet the paper omits any discussion of how the formulas extend when the fluid level process hits a boundary, which is central to the risk-process application in §5.2.

    Authors: We acknowledge that the transient derivation in §4.1 is presented for the free process. In the revised version we have inserted a short subsection (now §4.2) that explains how the same differentiation technique carries over to processes with reflecting or absorbing boundaries. The key step is to replace the unrestricted generator by the boundary-adjusted generator (standard in the SFM literature) before differentiating the matrix exponential; the resulting sensitivity ODE remains well-posed. This extension is then used without change in the insurance-risk example of §5.2, where the fluid level is reflected at zero, and we have added a brief consistency check confirming that the computed sensitivities respect the boundary condition. revision: yes

Circularity Check

0 steps flagged

Derivation chain is self-contained via direct differentiation

full rationale

The core results follow from differentiating the standard linear algebraic equations for the stationary distribution of a finite-phase Markov-modulated fluid model (invertible under positive recurrence) and the matrix-exponential or linear-ODE representations of the transient quantities. Both the algebraic solution map and the exponential/ODE flow are smooth on finite-dimensional spaces, so their parameter derivatives exist in a neighborhood of any interior point without additional interchange arguments or domination conditions. The numerical examples simply substitute the resulting closed-form sensitivities into standard queueing and risk models; no fitted parameters are renamed as predictions, no self-citation chain is load-bearing, and no ansatz is smuggled in. The derivation therefore reduces to elementary matrix calculus applied to the model definition itself.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No specific free parameters, axioms, or invented entities are described in the abstract.

pith-pipeline@v0.9.0 · 5609 in / 953 out tokens · 22941 ms · 2026-05-21T01:43:33.777536+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Abate and W

    J. Abate and W. Whitt. Numerical inversion of Laplace transforms of probability distributions. ORSA Journal on computing, 7(1):36–43, 1995

    work page 1995

  2. [2]

    Ahn and V

    S. Ahn and V. Ramaswami. Fluid flow models and queues - a connection by stochastic coupling. Stochastic Models, 19(3):325–348, 2003

    work page 2003

  3. [3]

    Aksamit, M

    A. Aksamit, M. M. O’Reilly, and Z. Palmowski. Sensitivities of some performance measures of quasi-birth-and-death processes.Stochastic Models, 0(0):1–0, 2024

    work page 2024

  4. [4]

    Asmussen and H

    S. Asmussen and H. Albrecher.Ruin Probabilities. Advanced Series on Statistical Science & Applied Probability, 14. World Scientific, Second edition, 2010

    work page 2010

  5. [5]

    A. L. Badescu, L. Breuer, S. Drekic, G. Latouche, and D. A. Stanford. The surplus prior to ruin and the deficit at ruin for a correlated risk process.Scandinavian Actuarial Journal, 2005(6):433–445, 2005

    work page 2005

  6. [6]

    N. G. Bean and M. M. O’Reilly. Performance measures of a multi-layer Markovian fluid model. Annals of Operations Research, 160:99–120, 2008

    work page 2008

  7. [7]

    N. G. Bean, M. M. O’Reilly, and Z. Palmowski. Yaglom limit for stochastic fluid models.Advances in Applied Probability, 53(3):649–686, 2021

    work page 2021

  8. [8]

    N. G. Bean, M. M. O’Reilly, and J. E. Sargison. A stochastic fluid flow model of the operation and maintenance of power generation systems.IEEE Transactions on Power Systems, 25(3):1361–1374, 2010

    work page 2010

  9. [9]

    N. G. Bean, M. M. O’Reilly, and P. G. Taylor. Algorithms for return probabilities for stochastic fluid flows.Stochastic Models, 22(1):149–184, 2005

    work page 2005

  10. [10]

    N. G. Bean, M. M. O’Reilly, and P. G. Taylor. Hitting probabilities and hitting times for stochastic fluid flows.Stochastic Processes and their Applications, 115(9):1530–1556, 2005

    work page 2005

  11. [11]

    N. G. Bean, M. M. O’Reilly, and P. G. Taylor. Algorithms for the Laplace-Stieltjes transforms of first return times for stochastic fluid flows.Methodology and Computing in Applied Probability, 10(3):381–408, 2008

    work page 2008

  12. [12]

    Breuer and A

    L. Breuer and A. L. Badescu. A generalised gerber–shiu measure for markov-additive risk processes with phase-type claims and capital injections.Scandinavian Actuarial Journal, 2014(2):93–115, 2014

    work page 2014

  13. [13]

    Cao.Stochastic learning and optimization: A sensitivity-based approach

    X.-R. Cao.Stochastic learning and optimization: A sensitivity-based approach. Springer Science & Business Media, 2007

    work page 2007

  14. [14]

    Da Silva Soares

    A. Da Silva Soares. Fluid queues - building upon the analogy with QBD processes.Doctoral Dissertation, Universite Libre de Bruxelles, Belgium, 2005. 32

    work page 2005

  15. [15]

    da Silva Soares and G

    A. da Silva Soares and G. Latouche. Fluid queues with level dependent evolution.European Journal of Operational Research, 196:1041–1048, 2009

    work page 2009

  16. [16]

    Den Iseger

    P. Den Iseger. Numerical transform inversion using gaussian quadrature.Probability in the Engi- neering and Informational Sciences, 20(1):1–44, 2006

    work page 2006

  17. [17]

    D. C. Dickson and C. Hipp. Ruin probabilities for erlang(2) risk processes.Insurance: Mathematics and Economics, 22(3):251–262, 1998

    work page 1998

  18. [18]

    H. U. Gerber and E. S. W. Shiu. On the time value of ruin.North American Actuarial Journal, 2(1):48–78, 1998

    work page 1998

  19. [19]

    He and H

    Q.-M. He and H. Wu. Double-sided queues with marked markovian arrival processes and abandon- ment.Stochastic Models, 37(1):23–58, 2020

    work page 2020

  20. [20]

    He and H

    Q.-M. He and H. Wu. Multi-layer MMFF processes and the MAP/PH/K+GI queue: Theory and algorithms.Queueing Models and Service Management, 3(1):37–87, 2020

    work page 2020

  21. [21]

    Heidergott, A

    B. Heidergott, A. Hordijk, and H. Weisshaupt.Taylor series expansions for elastoplastic materials and Markov processes. Springer, 2006

    work page 2006

  22. [22]

    Horv´ ath, I

    G. Horv´ ath, I. Horv´ ath, S. A.-D. Almousa, and M. Telek. Numerical inverse Laplace transformation using concentrated matrix exponential distributions.Performance Evaluation, 137:102067, 2020

    work page 2020

  23. [23]

    Joyner and B

    J. Joyner and B. Fralix. A new look at markov processes of G/M/1-type.Stochastic Models, 32(2):253–274, 2016

    work page 2016

  24. [24]

    Latouche and V

    G. Latouche and V. Ramaswami.Introduction to Matrix Analytic Methods in Stochastic Modeling. ASA-SIAM Series on Statistics and Applied Probability, Series Number 5. 1996

    work page 1996

  25. [25]

    C. D. Meyer. The condition of a finite Markov chain and perturbation bounds.SIAM Journal on Algebraic Discrete Methods, 1(3):273–283, 1980

    work page 1980

  26. [26]

    Neuts.Matrix-geometric Solutions in Stochastic Models: An Algorithmic Approach

    M. Neuts.Matrix-geometric Solutions in Stochastic Models: An Algorithmic Approach. The Johns Hopkins University Press, Baltimore, 1981

    work page 1981

  27. [27]

    M. M. O’Reilly and W. Scheinhardt. Stationary distributions for a class of Markov-modulated tandem fluid queues.Stochastic Models, 33(4):524–550, 2017

    work page 2017

  28. [28]

    Phung-Duc, H

    T. Phung-Duc, H. Masuyama, S. Kasahara, and Y. Takahashi. A simple algorithm for the rate matrices of level-dependent QBD processes. InProceedings of the 5th International Conference on Queueing Theory and Network Applications, pages 46–52, 2010

    work page 2010

  29. [29]

    Ramaswami.Matrix Analytic Methods: A Tutorial Overview with Some Extensions and New Results, pages 261–296

    V. Ramaswami.Matrix Analytic Methods: A Tutorial Overview with Some Extensions and New Results, pages 261–296. 1996. 33

    work page 1996

  30. [30]

    Ramaswami

    V. Ramaswami. Matrix-analytic methods: A tutorial overview with some extensions and new results. In S. Chakravarthy and A. Alfa, editors,Matrix-Analytic Methods in Stochastic Models, pages 261–296. Marcel Dekker: New York, 1997

    work page 1997

  31. [31]

    Ramaswami

    V. Ramaswami. Passage times in fluid models with application to risk processes.Methodology and Computing in Applied Probability, 8(4):497–515, Dec 2006

    work page 2006

  32. [32]

    Samuelson, A

    A. Samuelson, A. Haigh, M. M. O’Reilly, and N. G. Bean. Stochastic model for maintenance in continuously deteriorating systems.European Journal of Operational Research, 259(3):1169–1179, 2017

    work page 2017

  33. [33]

    P. J. Schweitzer. Perturbation theory and finite Markov chains.Journal of Applied Probability, 5(2):401–413, 1968

    work page 1968

  34. [34]

    Toutain, J

    J. Toutain, J. L. Battaglia, C. Pradere, J. Pailhes, A. Kusiak, W. Aregba, and J. C. Batsale. Numerical inversion of laplace transform for time resolved thermal characterization experiment. Journal of Heat Transfer, 133(4), 2011

    work page 2011

  35. [35]

    Wu.Analysis of Stochastic Models through Multi-Layer Markov Modulated Fluid Flow Processes

    H. Wu.Analysis of Stochastic Models through Multi-Layer Markov Modulated Fluid Flow Processes. PhD thesis, the University of Waterloo, 2021. 34

    work page 2021