🔢 mathematics

Spectral theory of non-local Ornstein-Uhlenbeck operators

This paper presents an in-depth spectral analysis of non-local Ornstein-Uhlenbeck operators driven by Lévy processes, establishing explicit formulas for eigenfunctions and co-eigenfunctions via a key intertwining relationship with diffusion semigroups, while also deriving conditions for spectral expansion and semigroup compactness.

Original authors: Rohan Sarkar

Published 2026-04-10

📖 5 min read🧠 Deep dive

CC0 1.0

Original authors: Rohan Sarkar

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are watching a drop of ink spreading in a glass of water. In a perfectly still glass, the ink just diffuses randomly. But now, imagine the water is also swirling in a specific pattern (like a whirlpool) while the ink spreads. This swirling motion is the "Ornstein-Uhlenbeck" part, and the random spreading is the "diffusion."

For decades, mathematicians have studied this specific type of swirling, spreading ink when the randomness comes from Brownian motion (tiny, continuous jiggles, like pollen grains in water). They knew exactly how the ink would eventually settle and how fast it would get there.

This paper is about what happens when the "jiggles" are not continuous, but come in sudden, unpredictable jumps.

Think of it like this: instead of the ink spreading smoothly, imagine a mischievous fairy occasionally teleporting a chunk of ink to a random spot in the glass. This is a Lévy process. The math gets much harder because the ink isn't just spreading; it's teleporting.

Here is a breakdown of what the author, Rohan Sarkar, discovered, using simple metaphors:

1. The Problem: The "Ghost" in the Machine

In the smooth, continuous world, mathematicians have a perfect map (a "spectrum") that tells them exactly how the system behaves. It's like having a sheet of music that tells you every note the system will play.

But when you add the "teleporting jumps" (the non-local part), the system becomes non-normal.

Analogy: Imagine a symphony orchestra. In a normal system, every instrument plays in perfect harmony, and you can easily predict the sound. In this new system, the instruments are playing slightly out of sync with each other. The "music" (the math) is messy, and the usual maps don't work anymore. We didn't know if we could even write down the "sheet music" (eigenfunctions) for this chaotic system.

2. The Big Discovery: The "Magic Mirror" (Intertwining)

The author's main trick is something called Intertwining.

The Metaphor: Imagine you have a complex, chaotic machine (the Jumping Ink). You want to understand it, but it's too hard. So, you build a Magic Mirror (a link operator).
When you look at the chaotic machine through this mirror, it transforms into a simple, smooth machine (the classic, non-jumping swirling ink) that mathematicians already understand perfectly.
The author proved that you can translate the behavior of the complex, jumping system into the language of the simple, smooth system. Once you solve the problem for the simple system, you can use the mirror to translate the answer back to the complex one.

3. The Results: What We Learned

Using this "Magic Mirror," the paper reveals several surprising things:

The "Notes" are the Same: Even though the ink is teleporting, the fundamental "notes" (the spectrum) it plays are exactly the same as the smooth ink. The jumps don't change the pitch of the system, only how the notes are arranged.
The "Sheet Music" Exists: The author wrote down explicit formulas for the "notes" (eigenfunctions) and their "mirror images" (co-eigenfunctions).
- Analogy: In the smooth world, the notes are like Hermite Polynomials (a specific family of curves). The author found that for the jumping world, the notes are a generalized version of these curves. They are still polynomials, but they have been "warped" by the jumps.
The "Biorthogonal" Dance: In a normal system, the notes are like perfect partners in a dance (orthogonal). In this chaotic system, they aren't perfect partners, but they are biorthogonal.
- Analogy: Imagine a dance where Partner A steps left, and Partner B steps right, but they don't quite touch. They are "just right" for each other to make the dance work, even if they aren't perfectly aligned. The author proved that these specific "jumping notes" and their "mirror notes" fit together perfectly to describe the system.

4. The Twist: When the Dance Breaks Down

The paper also found a limit to this magic.

The Compactness Issue: Sometimes, a system is "compact," meaning it settles down nicely and predictably. The author found that if the jumps are too wild (specifically, if they don't have a "finite moment," meaning the jumps can be infinitely huge), the system refuses to settle down in a predictable way.
The Spectral Expansion Failure: Even if the system has a list of "notes," you can't always add them up to get the full picture if the jumps are too extreme.
- Analogy: Imagine trying to describe a storm by adding up the sound of individual raindrops. If the raindrops are normal, the sum works. If the raindrops are occasionally giant boulders falling from the sky, the math of adding them up breaks down. The "sum of the notes" diverges and becomes infinite.

5. Why Does This Matter?

This isn't just abstract math. These "jumping swirls" model real-world chaos:

Finance: Stock prices don't just wiggle; they crash and spike (jumps).
Physics: Particles in a plasma or fluid can collide and bounce unpredictably.
Biology: Animal populations can suddenly boom or crash due to external events.

In summary: Rohan Sarkar took a very messy, chaotic mathematical problem (swirling ink with teleporting jumps) and built a bridge to a clean, simple problem we already understood. He showed that despite the chaos, the system still has a hidden order (the same "notes" as the smooth version), provided the jumps aren't too crazy. He gave us the "sheet music" for this chaotic dance, allowing scientists to better predict how these complex systems will behave over time.

1. Problem Statement and Context

The paper investigates the spectral theory of Non-Local Ornstein-Uhlenbeck (OU) operators, which are generators of OU processes driven by general Lévy processes (including jumps). These operators generalize the classical diffusion OU operator ( $A_{OU} = \frac{1}{2}\text{tr}(\Sigma\nabla^2) + \langle Bx, \nabla \rangle$ ) by adding a non-local integro-differential term representing the jump component.

Key Challenges:

Non-Normality: Unlike the self-adjoint case (where $\Sigma B = B^*\Sigma$ ), these operators are generally non-normal in the $L^2$ space weighted by the invariant distribution $\mu$ . This complicates spectral analysis as eigenfunctions are not necessarily orthogonal.
Non-Locality: The presence of jumps (pure jump processes or mixed diffusion-jump) introduces non-local terms, making standard PDE techniques insufficient.
Degeneracy: Existing literature often assumes the non-degeneracy of the Gaussian component ( $\Sigma_\infty$ is invertible). This paper aims to remove this restriction, allowing for cases where the process is purely jump-driven ( $\Sigma=0$ ) or has a degenerate Gaussian part.
Spectral Expansion: It is unclear whether the semigroup admits a spectral expansion (convergence of eigenfunction series) in the non-local, non-normal setting, especially given potential super-exponential growth of eigenfunction norms.

2. Methodology

The author employs a novel approach based on intertwining relationships rather than second quantization or direct PDE analysis.

Intertwining Relations: The core technique involves proving that the non-local Lévy-OU semigroup ( $P_t$ $P_{t}$ ) is intertwined with a simpler diffusion OU semigroup ( $Q_t$ $Q_{t}$ ) or a drift-only operator ( $L = \langle Bx, \nabla \rangle$ $L = ⟨ B x, \nabla ⟩$ ).
- Specifically, the paper establishes a link operator $\Lambda$ (a Markov operator) such that $L\Lambda = \Lambda A$ (relating the non-local operator to the drift) and $Q_t V = V P_t$ (relating the non-local operator to a normal diffusion operator).
Approximation Techniques: To handle cases where the Gaussian component is degenerate ( $\Sigma_\infty$ $Σ_{\infty}$ singular) or the Lévy measure lacks exponential moments, the author uses:
- Regularization: Adding a small diffusion term ( $\epsilon \Delta$ ) to approximate the non-local operator by a non-degenerate diffusion operator.
- Truncation: Truncating the Lévy measure to ensure finite exponential moments, proving results for the truncated case, and then taking limits.
Pseudo-Differential Operators: The link operators are constructed using Fourier multipliers and pseudo-differential representations of the semigroups.
Biorthogonal Systems: Since the operators are non-normal, the paper constructs biorthogonal sequences of eigenfunctions ( $H_n$ ) and co-eigenfunctions ( $G_n$ ) such that $\langle H_n, G_m \rangle = \delta_{nm}$ .

3. Key Contributions and Main Results

A. Spectrum and Multiplicities

Spectrum Identification: The paper proves that the point spectrum of the non-local operator $A$ in $L^p(\mu)$ is exactly the set $N(\lambda) = \{ -\sum n_i \lambda_i \mid n_i \in \mathbb{N}_0 \}$ , where $\lambda_i$ are the eigenvalues of $-B$ .
Isospectrality: Crucially, the spectrum and the algebraic/geometric multiplicities of the eigenvalues are independent of the non-local perturbation (the Lévy measure $\Pi$ ). They depend solely on the drift matrix $B$ .
Polynomial Eigenfunctions: The eigenfunctions corresponding to these eigenvalues are polynomials.

B. Explicit Formulas for Eigenfunctions and Co-eigenfunctions

Eigenfunctions ( $H_n$ ): Under the assumption that $B$ is diagonalizable, the paper derives explicit formulas for eigenfunctions. These are generalizations of Hermite polynomials, defined via derivatives of the characteristic function of the invariant distribution (Rodriguez-type formula).
Co-eigenfunctions ( $G_n$ ): Explicit formulas for the eigenfunctions of the adjoint operator $A^*$ are provided. These are generally not polynomials unless the process is purely diffusive ( $\Pi=0$ ).
Biorthogonality: The sequences $(H_n)$ and $(G_n)$ form a biorthogonal system in $L^2(\mu)$ . The paper proves that the existence of such a biorthogonal sequence of polynomials and co-eigenfunctions is equivalent to the diagonalizability of the drift matrix $B$ .

C. Spectral Expansion and Convergence

Diffusion Case ( $\Pi=0$ ): When the process is a standard diffusion, a spectral expansion exists for $t > t_0$ (where $t_0$ depends on the norm of the link operator). The expansion is given by:
$P_t f = \sum_{n} e^{-t\langle n, \lambda \rangle} \langle f, G_n \rangle H_n$
Non-Local Case ( $\Pi \neq 0$ ): The paper demonstrates a surprising phenomenon: even if the semigroup is compact and has the same spectrum, spectral expansion may fail. This is due to the super-exponential growth of the norms of the eigenfunctions $H_n$ in the presence of jumps, which causes the series to diverge for any $t > 0$ .

D. Compactness Conditions

Necessary Condition: The paper proves that for the semigroup $P_t$ to be compact on $L^p(\mu)$ , the invariant distribution $\mu$ must have moments of all orders. Consequently, Lévy-OU semigroups driven by $\alpha$ -stable processes ( $\alpha < 2$ ) are never compact.
Sufficient Condition: A sufficient condition for compactness is provided in terms of the invariant density $\mu$ : if $W = -\log \mu$ satisfies $\lim_{|x|\to\infty} |\nabla W(x)| = \infty$ , then $P_t$ is compact. This generalizes known results for diffusion processes.

4. Significance and Impact

Unification of Theory: The work unifies the spectral theory of diffusion and jump-driven OU processes, showing that the spectral structure is robust against non-local perturbations, provided the drift is well-behaved.
Removal of Degeneracy Assumptions: By handling singular $\Sigma_\infty$ , the results apply to a broader class of processes, including pure jump processes, which were previously difficult to analyze spectrally.
New Special Functions: The derived eigenfunctions represent a new class of special functions (generalized Hermite polynomials for non-Gaussian measures), which may have applications in mathematical finance and physics.
Limitations of Spectral Expansions: The finding that spectral expansions can fail for compact non-local semigroups due to norm growth is a significant theoretical insight, highlighting a fundamental difference between local (diffusion) and non-local operators.
Intertwining as a Tool: The paper establishes intertwining relations as a powerful tool for analyzing non-normal, non-local Markov semigroups, extending the methodology beyond the diffusion case.

In summary, this paper provides a comprehensive spectral analysis of non-local OU operators, establishing that while the spectrum remains invariant under jumps, the functional analytic properties (biorthogonality, spectral expansion, and compactness) are highly sensitive to the nature of the Lévy measure and the geometry of the drift.