Asymptotic Error Distributions for Euler-Maruyama Schemes with Additive Semimartingale Noise
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.
Ada: We're looking at the paper: "Asymptotic Error Distributions for Euler-Maruyama Schemes with Additive Semimartingale Noise".
Emmy: Here's the gist of the paper. Asymptotic Error Distributions for Euler–Maruyama Schemes with Additive Semimartingale Noise establishes a joint stable limit theorem for integrated discretization errors of finite-dimensional càdlàg semimartingales and applies it to the Euler–Maruyama scheme for SDEs with additive noise, providing a stable identification of the limiting error process.
Ada: Integrated Discretization Limits The paper introduces a joint stable limit theorem for integrated discretization errors of finite-dimensional semimartingales whose predictable characteristics are absolutely continuous in time and satisfy a pathwise integrability condition.
Emmy: This theorem states that the normalized integrated error, evaluated on the discretization grid, converges stably jointly with the discretized semimartingale in the vector-valued Skorokhod J1-topology.
Ada: The limiting error process is defined by a linear integral equation involving an explicit inhomogeneous term and an auxiliary Brownian motion and independent uniform marks attached to the jumps.
Title and authors: Ada: So we’re looking at this paper now, "Asymptotic Error Distributions for Euler-Maruyama Schemes with Additive Semimartingale Noise." It’s dealing with how the errors in those common Euler-Maruyama schemes behave when you add noise that isn't just simple Brownian motion.
Emmy: Exactly. The main idea here is establishing a joint stable limit theorem for these integrated discretization errors of finite dimensional semimartingales, which is a big deal because it applies to SDEs with additive noise without needing to assume any specific martingale representation property upfront.
Kolya: So, what does that mean for the practical application? What’s the actual result when you look at the error on your computer?
Ada: Well, they state that if you have a drift function that is C one with a locally Lipschitz derivative and it has a nonexplosive solution, then after normalizing by n, the grid point error converges stably jointly with the discretized driver.
Emmy: That convergence lands in this vector valued Skorokhod J1 topology which is a very specific way of measuring convergence for these càdlàg processes. The limiting error process itself solves a linear integral equation that has an explicit inhomogeneous term involving the driver, an auxiliary Brownian motion, and independent uniform marks attached to the jumps.
Kolya: An auxiliary Brownian motion and uniform marks... that sounds like it’s capturing both the smooth parts and the jumps simultaneously in a structured way. What kind of constraints are they putting on those drift functions?
Ada: They provide two alternative regularity conditions for those drift functions, one based on small-jump integrability where the continuous local martingale characteristic is absolutely continuous, and another involving Lévy smoothing where the driver itself is an F-Lévy process and the function f belongs to a Sobolev space.
Emmy: And they use some pretty heavy machinery to get there. The proof relies on applying this general theorem to an augmented semimartingale which is defined as the driver Y and the difference between f of X at time t and f of x zero.
Kolya: That sounds like a lot of technical setup before you get to the actual convergence rate, right? What about how they prove that this convergence actually holds for continuous drivers?
Ada: For continuous drivers, if they satisfy either those two conditions on the drift function, say theorem three point two or theorem three point three, then the continuous time Euler error converges stably in the uniform topology <ref:2610.20225#pg2>.
Emmy: That uniform topology is essentially just saying that the paths of the error process converge uniformly as you increase your mesh size. They show this by proving that there's a unique càdlàg solution to a linear integral equation with respect to time, and then showing that the perturbation term goes to zero in probability uniformly as you increase the mesh size.
Title and authors: Kolya: So if the driver is continuous, we get uniform convergence for continuous time errors. But what happens when we introduce jumps? I mean, those are often crucial in real models.
Ada: When you have jump activity, they show a way to analyze how simultaneous vector jumps produce a nonzero conditional covariance between the coordinates of the limiting Euler error through that common uniform mark.
Emmy: They use the discretized processes Y and U bar when Y has jumps because the exact solution and its Euler approximation have those same jumps, which is important for identifying those jump components separately from the continuous Gaussian ones.
Kolya: I’m curious about how they handle these bounds on the error terms. Are they just showing that something stays finite, or are there specific constraints on those bounds that matter for stability?
Ada: They establish some Gaussian smoothing bounds that can be checked without needing a density theorem for the drifted equation, using Minkowski’s inequality and the time scaling T t equals t to the power of two divided by alpha times T one.
Emmy: Those bounds are integrable in time when alpha is greater than one and p is greater than the maximum of one and d divided by alpha, which they assert in subsection three point three. They also show that convolution with the law of an independent jump does not increase those norm estimates using Young's inequality.
Kolya: So, if we are looking at a specific class of noise, like alpha stable noise where one is less than two, what’s the condition on the drift function? Is C one enough there?
Ada: For a rotationally symmetric alpha stable driver, they say that C one regularity is sufficient for alpha greater than one and less than two.
Emmy: And when they look at the jump integral itself in the case of a continuous two case, substitution into their representation of Z shows that the jump integral is absolutely convergent because of the local quadratic bound and that finite large-jump sum.
Kolya: That sounds like they’ve managed to tame those potentially infinite sums by showing they are controlled by some local bound, which is good. But what about the multidimensional drift case? Does this work when it’s not globally Lipschitz?
Ada: They do have an example of a multidimensional Sobolev drift where the drift is globally Lipschitz and C one, but it isn't locally C one point one at the origin. This example helps separate those two sufficient conditions without needing that extra argument about occupation time in one dimension.
Title and authors: Emmy: And they also show a way to solve some of these linear equations exactly when all the event times are available, which is useful for their analysis involving common uniform marks in a linear equation.
Kolya: So, if we put all this together, what’s the real impact on how we build these numerical schemes? What does this paper change for the people who are actually trying to implement this stuff?
Ada: It gives them a way to choose the right regularity condition for their drift function based on whether they have small-jump integrability or if they can use Lévy smoothing.
Emmy: It also provides a rigorous framework for identifying the limiting error when you have jumps, by showing that you can jointly identify the continuous Gaussian and marked jump components using that vector valued Skorokhod J1 topology.
Kolya: I think what stands out is how they connect the discretization error directly to an explicit linear integral equation, which gives them a concrete target for analysis instead of just abstract convergence statements.
Ada: Right, and they show how stability of discretized linear equations translates into the stable convergence of the continuous time solution U, provided you have those specific bounds on the coefficient H.
Emmy: And they provide a consistent framework for analyzing these error processes across different regularity classes for both continuous and jump-driven SDEs.
Kolya: So, to wrap up this paper, the paper "Asymptotic Error Distributions for Euler-Maruyama Schemes with Additive Semimartingale Noise" gives us a joint stable limit theorem that explicitly defines the limiting error as solving a linear integral equation with terms involving an auxiliary Brownian motion and uniform marks attached to jumps.
Ada: It shows that we can get this convergence stably jointly with the discretized driver under standard drift assumptions, and they provide two specific regularity conditions—small-jump integrability and Lévy smoothing—to achieve it.
Emmy: And for continuous drivers, they prove stable convergence in the uniform topology, which is a strong statement about how fast the error shrinks as you refine your grid.
Kolya: For me, what I find most useful is that they provide these tools to actually identify the continuous Gaussian and marked jump contributions in a single limit process using that vector valued Skorokhod J1 topology.
Ada: It's a solid paper for anyone building numerical schemes for SDEs with additive noise, because it gives you concrete convergence targets and methods to handle those tricky jump components.
Emmy: It’s quite deep work, connecting discretization error theory to the structure of the limiting stochastic integral equation.
Kolya: Thanks for walking us through that complex stuff today.
The paper's summary: Ada: So, to recap, this paper basically gives us a joint stable limit theorem for those integrated errors when you use an Euler Maruyama scheme on equations with additive semimartingale noise.
Emmy: That’s right, Ada. The main takeaway is that we can actually figure out the exact shape of the limiting error process. It's not just some abstract convergence statement; it’s a specific linear integral equation that involves an auxiliary Brownian motion and these marks attached to the jumps.
Kolya: So, for someone who only listens to this, what does this mean practically? Like, how does this help someone who is actually building a numerical simulation?
Ada: It means you don't have to guess the limiting behavior. Instead, you get a concrete target—that linear integral equation—to solve for the error. You can see exactly how those continuous Gaussian parts and the marked jump parts combine in that vector valued Skorokhod J1 topology.
Emmy: And that topology is key because it lets us treat the continuous and discontinuous pieces as a single unified object, which is really structured mathematically. It’s not just two separate processes converging; they are being identified together by this specific rule.
Kolya: I’m interested in the constraints on the drift functions again. They mentioned two ways to make this work, small-jump integrability or Lévy smoothing for the driver itself. Which one is actually easier to satisfy in a real-world model?
Ada: Well, it depends entirely on your noise structure. If you know your noise has certain small jumps and you can bound the continuous part nicely, then that’s one path. If the driver itself has some inherent smoothness—like being an F-Lévy process—then Lévy smoothing is the other option.
Emmy: And they show that for continuous drivers, if you meet either of those conditions, the error converges stably in the uniform topology. That means your numerical solution gets closer and closer to the true continuous solution uniformly as your time steps get smaller.
Kolya: But what about when things get jumpy? Like when simultaneous vector jumps happen? The paper shows that these jumps create a specific kind of covariance in the limiting error, tied to that common uniform mark. That’s interesting because it gives us a structural reason why those jump coordinates are related.
Ada: Exactly. It tells you how the discrete steps interact with the jumps in a way that isn't obvious just by looking at the individual components separately. It links the jump activity directly into the structure of the error limit itself, which is a big piece of information for modeling complex systems.
Emmy: So, this work moves us from saying "the error goes to zero" to saying "this is exactly what that specific function looks like in the limit." And that structure—the linear integral equation with auxiliary terms—is the new benchmark for analyzing these types of SDE errors.
Kolya: It’s a strong tool for verification, I think. If you can check if your numerical error process matches this limiting integral equation, you’ve really verified your scheme's behavior under those conditions.
Ada: Precisely. And the stability results they show for the discretized linear equations are important too. They confirm that if the initial convergence holds stably with respect to one set of conditions, it will also hold stably with respect another set of conditions on a slightly different way of looking at things.
Emmy: So, this paper is essentially providing a comprehensive toolbox for anyone building numerical schemes for SDEs driven by additive semimartingale noise—it gives you the convergence target and the structural explanation for how those jumps behave in the limit.
The paper's improvements: Kolya: So, we've talked about how the paper sets up that linear integral equation for those discretization errors involving the Brownian motion and those jump marks attached to everything. What are they suggesting to make this framework even better?
Ada: They are suggesting an adaptive numerical scheme. Instead of just picking one fixed step size, you can choose a scheme whose error convergence rate is directly controlled by how smooth or rough the drift function is locally.
Emmy: That ties back to the regularity conditions we discussed earlier, meaning if your drift function has more structure—like being smoother in some sense—you get a better convergence rate for your simulation. It’s about making the numerical method smarter based on where you are in time or space.
Kolya: So, this moves the focus from just finding *an* error bound to finding the *best possible* error bound for a specific type of function you are simulating. That sounds like a real practical win for complex models.
Ada: It is. And they also highlight the power of using Sobolev regularity techniques to estimate those discretization errors when your drift function isn't globally Lipschitz but has certain local smoothness properties, like being in the W two p local space.
Emmy: That is useful because it lets you apply these convergence results to a wider class of equations, not just the nicely behaved ones. It broadens the scope of what this error analysis applies to geometrically.
Kolya: And they also focus on making sure that when you use the discretized linear equations, they are stable—that the error process converges reliably to the continuous-time solution under specific conditions on how fast those coefficients converge.
Ada: Exactly. It’s about ensuring that as your numerical steps get smaller, you don't just get a sequence of answers but a stable convergence to the true continuous process U.
Emmy: And they also give us this idea about removing the need for some extra regularization steps in the proof—they show you can localize and mollify without losing the core information about those continuous Gaussian and marked jump components being identified jointly.
Kolya: So, it’s less about proving convergence in a vacuum and more about providing explicit methods for building better numerical tools that are tailored to the specific noise and drift structure you have on hand.
Ada: That’s the core idea. It turns this theoretical result into a practical guide for developing robust algorithms where you can tune your discretization based on known properties of your system.
Conclusion: Ada: So, to wrap up, this paper "Asymptotic Error Distributions for Euler-Maruyama Schemes with Additive Semimartingale Noise" gives us a joint stable limit theorem for integrated errors and shows exactly how the limiting error process solves a linear integral equation involving those auxiliary Brownian motion and jump marks.
Emmy: That’s right, Ada. The authors establish that under certain regularity conditions on the drift function, you can get stable convergence in the uniform topology for continuous drivers, while jump activity yields a specific nonzero conditional covariance through that common mark.
Kolya: For me, what really sticks is how they connect the discretization error directly to an explicit linear integral equation with terms involving an auxiliary Brownian motion and those marks attached to jumps. That gives us a concrete target for analysis instead of just abstract convergence statements.
Ada: It really does. And they provide specific regularity conditions—like small-jump integrability or Lévy smoothing—that you can check based on the structure of your noise and drift, which is crucial for choosing the right numerical approach in practice.
Emmy: This work solidifies how we can identify those continuous Gaussian and marked jump components together using that vector valued Skorokhod J1 topology, making it a more structured way to look at the limiting error than before.
Kolya: It’s a powerful tool for verification, I think. If you can check if your numerical error process matches this limiting integral equation, you’ve really verified your scheme's behavior under those conditions.
Ada: Exactly. And the stability results they show for the discretized linear equations are important too. They confirm that if the initial convergence holds stably with respect to one set of conditions, it will also hold stably with respect another set of conditions on a slightly different way of looking at things.
Emmy: So, this paper is essentially providing a comprehensive toolbox for anyone building numerical schemes for SDEs driven by additive semimartingale noise—it gives you the convergence target and the structural explanation for how those jumps behave in the limit.
Kolya: I think it’s solid work because it gives us these rigorous methods to handle those tricky jump components, moving beyond just saying things converge in some weak sense.
Ada: It is a deep dive into connecting discretization error theory directly to the structure of the limiting stochastic integral equation, which is very useful for understanding the underlying dynamics of the system.
Emmy: This leads us nicely into how we can apply these ideas to more complex geometries or perhaps even non-linear SDEs next.
Division of Mathematical Science for Social Systems, Department of Systems Innovation, Graduate School of Engineering Science, The University of Osaka
math.PR
Submitted: 2026-10-14
Updated: 2026-10-31
Comments: 44 pages
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 90/100
The gist: Asymptotic Error Distributions for Euler–Maruyama Schemes with Additive Semimartingale Noise establishes a joint stable limit theorem for integrated discretization errors of finite-dimensional…
Key concepts
- Joint Stable Limit Theorem
- This theorem states that when you look at the integrated discretization errors of several semimartingales simultaneously, they converge stably. This means they converge in a way that respects their paths and topology, allowing researchers to understand the joint behavior of multiple errors on a grid.
- Linear Integral Equation
- The limiting error process is mathematically described by a specific linear integral equation. This equation includes terms related to the original driver, an auxiliary Brownian motion, and uniform marks attached to the jumps. Solving this equation reveals the precise nature of the long-term error behavior.
- Vector-Valued Skorokhod J1-topology
- This is a specific mathematical topology used to define convergence for vector-valued semimartingales. It ensures that the convergence observed between discretized and continuous processes respects both their paths and their jump structure, which is crucial for analyzing complex stochastic systems.
- Common Uniform Marks
- When dealing with simultaneous vector jumps, the analysis shows that a specific mark (1, 1) is the only one that contributes to the limiting error's covariance. This construction helps mathematically separate and identify how continuous Gaussian parts and marked jump components interact in the final error.
Terminology
Summary
Asymptotic Error Distributions for Euler–Maruyama Schemes with Additive Semimartingale Noise establishes a joint stable limit theorem for integrated discretization errors of finite-dimensional càdlàg semimartingales and applies it to the Euler–Maruyama scheme for SDEs with additive noise, providing a stable identification of the limiting error process. p. 0
Integrated Discretization Limits
The paper introduces a joint stable limit theorem for integrated discretization errors of finite-dimensional semimartingales whose predictable characteristics are absolutely continuous in time and satisfy a pathwise integrability condition. This theorem states that the normalized integrated error, evaluated on the discretization grid, converges stably jointly with the discretized semimartingale in the vector-valued Skorokhod J1-topology. p. 0 The limiting error process is defined by a linear integral equation involving an explicit inhomogeneous term and an auxiliary Brownian motion and independent uniform marks attached to the jumps. p. 0
Euler-Maruyama Error Analysis
The main result concerns the Euler–Maruyama error for the additive equation (1.1) under assumptions on the drift function. p. 3 The theorem states that for a C 1 drift with locally Lipschitz derivative and a nonexplosive solution, the grid-point error normalized by n converges stably jointly with the discretized driver. p. 0 The limiting error solves a linear integral equation with an explicit inhomogeneous term involving the driver, an auxiliary Brownian motion, and independent uniform marks attached to the jumps. p. 0
Regularity Refinements
The authors present two alternative regularity conditions for the drift function in Theorem 3.3. These alternatives are Jump integrability, where the continuous local martingale characteristic is absolutely continuous and a specific integral bound holds, or Lévy smoothing, where the driver Y is an F-Lévy process and f belongs to a Sobolev space. p. 3
Proof Techniques for Convergence
The proof of Theorem 3.2 relies on applying the general theorem to the augmented semimartingale Ξ = (Y, f(X) − f(x0)). p. 5 The convergence is established by showing that the required absolute continuity of the characteristics of f(X) is obtained through localization and mollification under the C 1,1 loc assumption. p. 5 The final joint convergence with the mean-value coefficient is achieved by applying stable Slutsky’s theorem to the sequence (Y[n], Z[n], H[n]). p. 1
Stability and Consistency Results
Auxiliary results establish consistency and stability for the linear integral equation and the discretized linear equation. These results confirm that if the convergence holds stably with respect to F, then it also holds stably with respect to G. p. 2 The joint limit for the regularized process is shown to be stable with respect to G. p. 1
Continuous Convergence
For continuous drivers satisfying either Theorem 3. p. 42 or Theorem 3. p. 0 p. 0 p. 03, the continuous-time Euler error converges stably in the uniform topology. p. 0 This convergence is established by showing that the limiting process U is a unique càdlàg solution of a linear integral equation with respect to time. p. 3 The continuous path convergence is also shown by demonstrating that the perturbation term tends to zero in probability uniformly as the mesh size increases. p. 2
Jump Activity Analysis
The paper demonstrates how simultaneous vector jumps produce a nonzero conditional covariance between the limiting Euler error coordinates through a common uniform mark. The use of the discretized processes Y[n] and U bar[n] is essential when Y has jumps, as the exact solution and its Euler approximation have the same jumps. p. 3 The vector marking is a mathematical construction that identifies the continuous Gaussian and marked jump components jointly. p. 5
Sobolev Regularity Verification
The verification of condition (B) in Theorem 3.3 involves showing that the occupation estimate for XR, restricted to [0, τR], makes their Hessian integrals agree, independently of the representatives. p. 5 This cancellation between the two corrections in (C. p. 09) and (C.10) yields the stopped Itô formula for f(X). p. 9 The final step involves showing that the jump-remainder sum is absolutely convergent after stopping. p. 6
Gaussian Smoothing Bounds
The bounds (3.17) can be checked without a density theorem for the drifted equation, as shown for Brownian motion and stable noise. p. 3 p. 3 The Gaussian bounds, Minkowski’s inequality, and Tt d= t 2/αT1 give the required norm estimates. p. 4 p. 4 The bounds are integrable in time when α > 1 and p > max(1, d/α), as asserted in Subsection 3. p. 93.
Common Uniform Marks
The analysis of common uniform marks in a linear equation shows that the product in the sum is nonzero precisely for a mark (1, 1). This is a statement about the limiting error, not convergence of finite-mesh covariances. p. 0 p. 0 The linear equation can also be solved exactly when all event times are available. p. 6
Convergence in the Continuous Case
Since Y is continuous, X is continuous and Kf = 0, the forcing in (3.11) and the solution of (3. p. 012) are therefore continuous. p. 4 The applicable result, Theorem 3. p. 22 or Theorem 3.3, gives (Y[n], U bar[n]) ⇒st (Y, U) in C([0, 1], R(q+d)) with the uniform topology. p. 2 The J1-topology on C([0, 1], R(q+d)) is the uniform topology by [JP12, (2. p. 3 p. 32.11)].
The C 2 Case
In the C 2 case, substitution into (4.17) yields a representation for Z defined in (3. p. 211) with I(Y) as in (4.25). The jump integral is absolutely convergent by the local quadratic bound and the finite large-jump sum. p. 9 p. 9 The coefficient 1/4 is the product of the factor 1/2 in (3. p. 4 p. 411) and the factor 1/2 in Itô’s formula. p. 1 p. 1
Consistency and Compact Containment
Lemma A.3 establishes that Hn → H in probability in L 1([0, T], R(d×d)) (A. p. 66) for every T ≤ 1. p. 7 This stability is further extended to the joint convergence with the mean-value coefficient, leading to (Y[n], Z[n], H[n]) ⇒st (Y, Z, H) in D([0, 1], R(q+d))×L 1([0, 1], R(d×d)). p. 1
Uniform Stability of Linear Equations
Lemma A.4 proves that if the joint convergence (Γn, Jn, Hn) ⇒ (Γ, J, H) holds stably with respect to F and HnL inf([0,1]) = OP(1), then the discretized linear equation converges stably to the continuous-time solution U. p. 0 The stability is also stable with respect to F. p. 7
Removal of Regularization
The proof of Theorem 2.1 involves regularizing the characteristics so that [JP12, Theorem 6. p. 41.2] can be applied and then removing the regularization through estimates uniform in the mesh parameter. p. 7 This argument identifies the continuous Gaussian and marked jump components jointly. p. 5 The final step justifies the limiting stochastic integrals on [0, 1] by showing that the localization along this countable sequence proves (4. p. 58).
Jump Activity and Small-Jump Bounds
Condition (A) of Theorem 3.3 concerns jump integrability, where the continuous local martingale characteristic is absolutely continuous and a specific integral bound holds. p. 2 This condition implies that the small-jump sums converge in total variation by the preceding summable bound. p. 3 The paper shows that for a rotationally symmetric α-stable driver, C 1 suffices for α α suffices for 1 ≤ α < 2. p. 1
Stable and Gaussian Smoothing
The bounds (3.17) can be checked without a density theorem for the drifted equation, as shown for Brownian motion and stable noise. The Gaussian bounds, Minkowski’s inequality, and Tt d= t 2/αT1 give the required norm estimates. Convolution with the law of an independent Jt does not increase either norm, by Young’s inequality. p. 7
Multidimensional Sobolev Drift
The paper provides an example of a multidimensional Sobolev drift where the drift is globally Lipschitz and C 1, but is not locally C 1,1 at the origin. This example separates the two sufficient conditions without invoking a one-dimensional occupation-time argument. p. 9
Common Uniform Marks in Linear Equations
The bound R 1 0 λt < ∞ and the finite mark set imply that condition (A) of Theorem 3.3 applies to every C 1 drift with a nonexplosive solution. p. 0 The same uniform mark is used for all coordinates of a vector jump. p. 3 This is a statement about the limiting error, not convergence of finite-mesh covariances.
The C 1,1 Case
For continuous drivers satisfying local C 1,1 regularity and nonexplosion, the corresponding additive Euler limit can already be obtained by combining classical results. For continuous drivers satisfying either Theorem 3.2 or Theorem 3.3, Corollary 3. p. 44 records stable convergence of the continuous-time Euler error in the uniform topology. p. 4 The J1-topology on C([0, 1], R(q+d)) is the uniform topology by [JP12, (2.2.11)].
For the process Rf defined in (4.14), Itô’s formula identifies R f i t = 1/2 Z t 0 f i j k(X s-) σ j α c a b s σ k β ds, and the residual identity becomes f i(X t) - f i(x0) = Z t 0 f i j k(X s-) dX s + 1/2 Z t 0 Q j k ∂jkv(X s-) ds + X 0<s≤t Θf (Xs-, ∆Ys). p. 1 The jump integral is absolutely convergent by the local quadratic bound and the finite large-jump sum. The coefficient 1/4 is the product of the factor 1/2 in (3.11) and the factor 1/2 in Itô’s formula.
The C 0 Case
The drift process is B f(X) t = Z t 0 Df(Xs-) f(Xs-) ds + Z t 0 Df(Xs-) σ b s ds + R f t + Z t 0 Z R q rs(z) F(dz) ds. The process B f(X) is continuous and adapted, hence predictable. p. 0 Its paths are absolutely continuous, so [JS03, Proposition I. p. 83.13], applied componentwise, yields a predictable process β such that B f(X) t = Z t 0 βs ds. p. 4
The C 1,γ Case
For a rotationally symmetric α-stable driver, 0 α. Thus C 1 suffices for α α suffices for 1 ≤ α < 2. p. 1
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Adaptive Numerical Scheme for SDEs: The system can implement a numerical scheme whose error convergence rate is determined by the local regularity of drift functions, specifically using assumptions like
small-jump integrability adapted to the modulus of Df
orLévy semigroup smoothing.
-
Joint Error Identification in High-Dimensional Systems: The AI can jointly identify continuous Gaussian terms and marked jump contributions in the limiting error process using a vector-valued Skorokhod J1 topology, even when martingale representation properties are not assumed.
-
Robust Euler Approximation for Jump-Driven Models: The system can generate an Euler approximation for SDEs with additive semimartingale noise where the limiting error solves a linear integral equation explicitly involving the driver and an auxiliary Brownian motion and independent uniform marks attached to jumps.
-
Continuous-Time Error Analysis: For continuous drivers satisfying local C1,1 regularity, the system can establish stable convergence of the continuous-time Euler error in the uniform topology.
-
Regularity-Based Drift Estimation: The system can utilize
Sobolev Itô formula
techniques to bound discretization errors when drift functions are not globally Lipschitz but satisfy certain Sobolev regularity conditions (like W2,p loc), allowing for more general noise structures. -
Stability of Discretized Linear Systems: The AI can analyze the stability of discretized linear equations where the error process is defined by a linear integral equation, guaranteeing convergence to the unique solution under specific conditions on the discretization and coefficient convergence.
Sources
Related papers
- Rank-One HCIZ Integrals and Multiplicative Convolutions
- Repeated averaging on random d-regular graphs
- Precise cover times for branching random walks on Hamming graphs: (iterated) logarithmic corrections
- Markov Chain CLTs: Resolving Open Problems
- Marked stable L'evy processes in Brownian local time
- A Sharp Refinement of Samuels' Inequality via Cap Moves