Abstract:We prove a large deviation principle for stochastic differential equations driven by semimartingales, with additive controls. Conditions are given in terms of characteristics of driven semimartingales, so that if the noise-control pairs satisfy a large deviation principle with some good rate function, so do the solution processes. There is no joint exponential tightness assumption for noise-control-solution triplets and no uniform exponential tightness assumption for noise.
What problem does this paper attempt to address?
### What problem does this paper attempt to solve?
This paper aims to explore the large deviation principle (LDP) of the solution process of stochastic differential equations (SDEs) driven by semimartingales. Specifically, the authors are concerned with whether the solution process \(Y^{\epsilon}\) also satisfies the large deviation principle given that the noise - control pair \((X^{\epsilon}, U^{\epsilon})\) satisfies the large deviation principle. This problem can be formalized as:
If the noise - control pair \(\{(X^{\epsilon}, U^{\epsilon})\}_{\epsilon>0}\) satisfies the large deviation principle with a good rate function \(I'\), does the solution process \(\{Y^{\epsilon}\}_{\epsilon>0}\) also satisfy the large deviation principle? If so, what is its rate function?
#### Main contributions
1. **Without the joint exponential tightness assumption**: Previous studies usually require the joint exponential tightness assumption of the noise - control - solution triple \((X^{\epsilon}, U^{\epsilon}, Y^{\epsilon})\), which may be difficult to verify in practical applications. In this paper, by introducing the Slomiński - type criterion (Proposition 2.2), this assumption is avoided, making the results more in line with the principle of causality.
2. **Giving sufficient conditions**: The authors provide sufficient conditions on the characteristics of semimartingales to ensure that the solution process \(Y^{\epsilon}\) satisfies the large deviation principle. These conditions only involve the characteristics of the driving noise \(X^{\epsilon}\) and are easy to check in practice.
3. **Generalized results**: The results of this paper are not only applicable to the finite - dimensional case but can also be generalized to the infinite - dimensional case and cover the large deviation problems of SDEs driven by exponentially integrable additive processes.
#### Summary of mathematical formulas
- **System (1.1)**:
\[
Y^{\epsilon}_t=U^{\epsilon}_t+\int_{0}^{t}F(Y^{\epsilon}_{s -})dX^{\epsilon}_s
\]
where \(F:\mathbb{R}^n\rightarrow\mathbb{R}^{n\times d}\) is a globally Lipschitz function.
- **Large deviation principle (Theorem 1.1)**:
Suppose there exist \(b > 0\) and \(0 < r<\frac{1}{4c}\), for each \(t > 0\), the following three real - valued families are all exponentially tight:
\[
\left\{V(B^{\epsilon}(h_b))_t\right\}_{\epsilon>0},\quad\left\{\frac{1}{\epsilon}C^{\epsilon}_t\right\}_{\epsilon>0},\quad\left\{\left(\epsilon\int_{\mathbb{R}^d\setminus\{0\}}\exp\left(\frac{|x|}{\epsilon^r\vee1}\right)\nu^{\epsilon}(dx, [0, t])\right)\right\}_{\epsilon>0}
\]
If the noise - control pair \(\{(X^{\epsilon}, U^{\epsilon})\}_{\epsilon>0}\) satisfies the large deviation principle with a good rate function \(I'\), then the solution process \(\{(X^{\epsilon}, U^{\epsilon}, Y^{\epsilon})\}_{\epsilon>0}\) also satisfies the large deviation principle, and its rate function is:
\[
I(x, u, y)=\begin{cases}
I'(x, u),&\text{if }y = u+F(y)\cdot x
\end{cases}
\]