Solution space characterisation of perturbed linear discrete and continuous stochastic Volterra convolution equations: the $\ell^p$ and $L^p$ cases

This paper characterizes the solution spaces of perturbed linear stochastic Volterra equations in discrete and continuous time, establishing that while $p$-summable perturbations are necessary and sufficient for almost sure $p$-summability in the discrete case, the continuous case allows for almost sure $p$-integrability even with non-integrable perturbations, a result proven via discretization and extended to analyze asymptotic convergence and broader functional differential equations.

The Gist

Imagine you are trying to predict the weather in a city that has a very strange memory.

In this city, the temperature today isn't just determined by what happened yesterday. It's influenced by the weather from the last week, the last month, and even the last year, but with a twist: the further back you go, the less influence that old weather has. This is a Volterra equation. It's a mathematical way of describing systems with "memory."

Now, imagine this city is also subject to random, chaotic gusts of wind (the noise or stochastic part) and some external forces like a heating system turning on and off (the perturbation).

The paper you asked about is essentially a rulebook for answering one big question: "Will the temperature in this city eventually settle down and stay within a reasonable, manageable range, or will it go wild and explode?"

Here is the breakdown of their findings using simple analogies:

1. The Two Types of "Wildness"

The authors are looking at two ways a system can go wrong:

• The Discrete Case (The Ticking Clock): Imagine checking the temperature once every hour. The question is: If you add up all the temperature readings over the next 100 years, will the total be a finite number? (This is called being p-summable).
• The Continuous Case (The Flowing River): Imagine the temperature is a smooth, flowing river. The question is: If you measure the "energy" of the river over an infinite time, is the total energy finite? (This is called being p-integrable).

2. The Big Surprise: Discrete vs. Continuous

The most interesting discovery in this paper is that the rules are different for the "Ticking Clock" and the "Flowing River."

• For the Ticking Clock (Discrete):
  Think of this like a budget. If you want your total spending over a year to be under control, every single expense you make must be reasonable. If you make one massive, crazy purchase, your whole budget is ruined.

  • The Rule: The external forces (the heating system) and the random wind gusts must both be "well-behaved" (small enough) for the system to stay stable. If the input is messy, the output is messy.

• For the Flowing River (Continuous):
  This is where it gets magical. Imagine a river that is usually calm, but occasionally has a massive, violent wave.

  • The Rule: Even if the "waves" (the random noise) are huge and chaotic, the river can still stay calm overall!
  • Why? Because in the continuous world, the system has a "smoothing" effect. It can absorb a few massive spikes without the total energy exploding. You can have a "bad" input (a huge wave) and still get a "good" output (a calm river), provided the bad input doesn't happen too often or in a specific pattern.

3. The "Memory" Factor

The paper also looks at how the system's memory affects things.

• Infinite Memory (Volterra): The system remembers everything from the beginning of time. The authors found that even with this heavy memory, the "smoothing" effect of the continuous river still works.
• Finite Memory (Functional Equations): Imagine the system only remembers the last hour. The authors found that in this case, the rules become even stricter and clearer. If the system has a short memory, it's easier to prove exactly when it will settle down.

4. The "Diagonal" Secret

There is a special case the authors looked at involving the "Diagonal Noise."

• Imagine the random wind gusts hitting the city. If the wind hits the North-South axis and the East-West axis completely independently (like two separate fans blowing in different rooms), the system is much easier to predict.
• If the wind is chaotic and mixes everything up, it's harder. But if the chaos is "diagonal" (separated), the authors can give you a precise checklist to guarantee the temperature will eventually drop to zero and stay there.

5. Why Does This Matter?

In the real world, we use these equations to model:

• Finance: How stock prices react to past trends and sudden news.
• Engineering: How bridges vibrate under wind and traffic loads over decades.
• Biology: How populations grow based on past resources and random environmental shocks.

The Takeaway:
This paper tells engineers and scientists: "Don't panic if your input data looks messy or chaotic. In the continuous world (like real-time physics), the system might naturally smooth out those chaos spikes. However, if you are looking at data in discrete chunks (like daily reports), you need to be much more careful, because one bad day can ruin the whole average."

They provide a mathematical "checklist" (the conditions on $f$ and $\sigma$) that tells you exactly when your system will be safe and when it will go off the rails.

Technical Summary

Here is a detailed technical summary of the paper "Solution space characterisation of perturbed linear discrete and continuous stochastic Volterra convolution equations: the $\ell^p$ and $L^p$ cases" by John A. D. Appleby and Emmet Lawless.

1. Problem Statement

The paper addresses the qualitative behavior of solutions to perturbed linear stochastic Volterra equations in both discrete and continuous time. Specifically, the authors seek to establish necessary and sufficient conditions on the forcing terms (deterministic perturbation $f$ and stochastic perturbation $\sigma$) such that the solution trajectories belong to specific $L^p$ spaces almost surely.

The primary equations studied are:

• Discrete Case: Stochastic Volterra summation equations:
  $$X(n+1) = X(n) + \sum_{j=0}^n K(n-j)X(j) + f(n) + \sigma(n)\xi(n+1)$$
• Continuous Case: Stochastic Volterra integro-differential equations:
  $$dX(t) = \left( f(t) + \int_{[0,t]} \nu(ds)X(t-s) \right) dt + \sigma(t)dB(t)$$

The core question is: Under what conditions on $f$ and $\sigma$ does the solution $X$ satisfy $\|X\| \in L^p$ (continuous) or $\ell^p$ (discrete) almost surely? This is distinct from merely finding sufficient conditions; the authors aim for a complete characterization of the solution space.

2. Methodology

The authors employ a multi-faceted approach combining probabilistic analysis, functional analysis, and discretization techniques:

• Discretization of Continuous Problems: A key methodological innovation is the use of an auxiliary discretization to prove converse results for the continuous case. By analyzing the discrete analogue, they derive necessary conditions for the continuous system.
• Reduction to Ornstein-Uhlenbeck (OU) Type Processes: The authors introduce a lemma (Lemma 3) showing that the $p$-integrability of the Volterra equation's solution is equivalent to the $p$-integrability of a simpler, memory-less OU-type process:
  $$dY(t) = (f(t) - Y(t))dt + \sigma(t)dB(t)$$
  This reduction allows them to bypass the complex memory kernel $\nu$ when determining integrability, proving that the memory structure does not alter the fundamental integrability criteria of the perturbation terms.
• Probabilistic Tools:
  • Borel-Cantelli Lemmas: Used extensively in the discrete case to prove that if a sum of random variables converges, the coefficients must satisfy specific summability conditions.
  • Itô's Isometry: Crucial for handling the $p \ge 2$ case in the continuous setting, linking the $L^p$ norm of the stochastic integral to the $L^{p/2}$ norm of the quadratic variation.
  • Kolmogorov's Two-Series Test: Applied to sequences of independent random variables to establish almost sure convergence.
• Class of Noise Distributions: The paper defines a specific class of random variables ($\mathcal{D}$) to handle general noise distributions in the discrete case, relaxing the requirement for Gaussian noise in certain contexts, though Gaussian noise is required for general matrix structures.

3. Key Contributions and Results

A. Discrete Case Characterization

For the discrete equation (1.7), the authors prove that for $p \in [1, \infty)$, the solution $X$ is almost surely $p$-summable ($X \in \ell^p$) if and only if:

1. The deterministic forcing term $f \in \ell^p$.
2. The noise coefficient matrix $\sigma \in \ell^p$.

Key Insight: Unlike some stabilization-by-noise phenomena, noise cannot "stabilize" a system with non-summable deterministic or stochastic forcing. If the solution is $p$-summable, the perturbations must be $p$-summable.

B. Continuous Case Characterization

For the continuous equation (1.1), the characterization is more nuanced due to the nature of integration versus summation. The solution $X$ is almost surely $p$-integrable ($X \in L^p$) if and only if the following hold (assuming the resolvent of the measure $\nu$ is integrable):

• For the deterministic term $f$:
  $$t \mapsto \int_t^{t+\theta} f_i(s) ds \in L^p(\mathbb{R}^+) \quad \forall \theta > 0$$
  Note: This condition is weaker than $f \in L^p$. $f$ itself need not be integrable; only its local averages over intervals of length $\theta$ need to be.

• For the stochastic term $\sigma$:

  • Case $p \ge 2$:
    $$t \mapsto \int_t^{t+\theta} \sigma_{ij}^2(s) ds \in L^{p/2}(\mathbb{R}^+) \quad \forall \theta > 0$$
  • Case $1 \le p < 2$:
    $$n \mapsto \int_n^{n+1} \sigma_{ij}^2(s) ds \in \ell^{p/2}(\mathbb{N})$$
    Note: Similar to $f$, $\sigma^2$ does not need to be in $L^{p/2}$ globally; only its moving averages (or discrete block sums) need to be.

C. Asymptotic Behavior

The paper investigates the almost sure convergence of solutions to zero ($X(t) \to 0$ a.s.):

• General Case: The asymptotic behavior is determined by the convolution of the resolvent $r$ and the forcing term $f$. Specifically, $\|X(t) - (r * f)(t)\| \to 0$ a.s.
• Diagonal Noise Case: For diagonal diffusion matrices, the authors provide a sharp characterization for $X(t) \to 0$ a.s. This involves a specific series condition on $\sigma^2$ (related to the law of the iterated logarithm) and the decay of the moving averages of $f$.
• Comparison with Mean Convergence: The authors highlight that $L^p$ integrability of paths is not sufficient for almost sure convergence to zero, nor is it necessary in the same way mean convergence is. They demonstrate that standard Lyapunov functional approaches often yield only sufficient conditions, whereas their method provides necessary and sufficient characterizations.

D. Stochastic Functional Differential Equations (SFDEs)

The results are extended to SFDEs with finite delay. A significant finding here is that for finite memory systems, the condition that the resolvent is integrable ($r \in L^1$) becomes a necessary condition for the existence of $p$-integrable solutions, whereas in the infinite memory (Volterra) case, it was an assumption. This leads to a "cleaner" and stronger characterization for SFDEs.

4. Significance and Implications

• Sharpness of Conditions: The paper resolves a gap in the literature by moving from sufficient conditions (often derived via Lyapunov functions) to necessary and sufficient conditions. This allows researchers to precisely determine the boundary of the solution space.
• Ill-Behaved Perturbations: The authors demonstrate that solutions can be well-behaved (integrable) even when the perturbation functions $f$ and $\sigma$ are highly irregular (e.g., unbounded or having infinite $L^p$ norms), provided their local averages satisfy the derived conditions. Section 5 provides explicit examples of such "spiky" functions.
• Methodological Shift: By reducing the complex Volterra problem to a simpler OU-type process for integrability questions, the paper offers a powerful tool for analyzing systems with memory without getting bogged down in the kernel's complexity for these specific qualitative questions.
• Distinction between Discrete and Continuous: The paper clarifies a fundamental difference: in the discrete case, $p$-summability of the solution strictly requires $p$-summability of the noise coefficients. In the continuous case, the "smoothing" effect of integration allows for non-integrable noise coefficients as long as their local energy averages are controlled.

5. Conclusion

Appleby and Lawless provide a comprehensive framework for understanding the solution spaces of perturbed stochastic Volterra equations. Their work establishes that the integrability of trajectories is dictated primarily by the local averaging properties of the forcing terms rather than their global integrability, and that the memory kernel's influence is secondary to these forcing terms regarding $L^p$ properties. The results significantly advance the theoretical understanding of stochastic systems with memory, offering precise criteria for stability and integrability that were previously unknown.