Abstract
In this paper, we study the moderate deviations principle (MDP) for slow–fast stochastic dynamical systems where the slow motion is governed by small fractional Brownian motion (fBm) with Hurst parameter \(H\in (1/2,1)\). We derive conditions on the moderate deviations scaling and on the Hurst parameter H under which the MDP holds. In addition, we show that in typical situations the resulting action functional is discontinuous in H at \(H=1/2\), suggesting that the tail behavior of stochastic dynamical systems perturbed by fBm can have different characteristics than the tail behavior of such systems that are perturbed by standard Brownian motion.
We’re sorry, something doesn't seem to be working properly.
Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.
Data Availability
Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
References
Ascione, G., Mishura, Y., Pirozzi, E.: Fractional Ornstein–Uhlenbeck process with stochastic forcing, and its applications. Methodol. Comput. Appl. Probab. 23(1), 53–84 (2021)
Baĭer, D., Freĭdlin, M.I.: Theorems on large deviations, and stability under random perturbations. Dokl. Akad. Nauk SSSR 235(2), 253–256 (1977). MR 0451366
Bayer, C., Friz, P.K., Gulisashvili, A., Horvath, B., Stemper, B.: Short-time near-the-money skew in rough fractional volatility models. Quant. Finance 19(5), 779–798 (2019). MR 3939657
Biagini, F., Yaozhong, H., Øksendal, B., Tusheng, Z.: Stochastic calculus for fractional brownian motion and applications. Springer (2008)
Billingsley, P.: Convergence of Probability Measures. Wiley (2013)
Boué, M., Dupuis, P.: A variational representation for certain functionals of Brownian motion. Ann. Probab. 26(4), 1641–1659 (1998)
Bourguin, S., Gailus, S., Spiliopoulos, K.: Typical dynamics and fluctuation analysis of slow-fast systems driven by fractional Brownian motion. Stoch. Dyn. 21(07), 2150030 (2021)
Budhiraja, A., Song, X.: Large deviation principles for stochastic dynamical systems with a fractional Brownian noise. arXiv preprint arXiv:2006.07683 (2020)
Cheridito, P.: Arbitrage in fractional Brownian motion models. Finance Stoch. 7(4), 533–553 (2003). MR 2014249
Cheridito, P., Kawaguchi, H., Maejima, M.: Fractional Ornstein–Uhlenbeck processes. Electron. J. probab. 8, 1–14 (2003)
Comte, F., Renault, E.: Long memory in continuous-time stochastic volatility models. Math. Finance 8(4), 291–323 (1998). MR 1645101
Comte, F., Renault, E.: Long memory in continuous-time stochastic volatility models. Math. Finance 8(4), 291–323 (1998)
da Silva, J.L., Erraoui, M., El Hassan, E.: Mixed stochastic differential equations: existence and uniqueness result. J. Theor. Probab. 31(2), 1119–1141 (2018)
Dupuis, P., Ellis, R.S.: A Weak Convergence Approach to the Theory of Large Deviations. Wiley (2011)
Dupuis, P., Spiliopoulos, K.: Large deviations for multiscale diffusion via weak convergence methods. Stoch. Process. Appl. 122(4), 1947–1987 (2012)
Forde, M., Zhang, H.: Asymptotics for rough stochastic volatility models. SIAM J. Financ. Math. 8(1), 114–145 (2017). MR 3608743
Forde, M., Zhang, H.: Asymptotics for rough stochastic volatility models. SIAM J. Financ. Math. 8(1), 114–145 (2017)
Fouque, J.-P., Papanicolaou, G., Sircar, R., Sølna, K.: Multiscale Stochastic Volatility for Equity, Interest Rate, and Credit Derivatives. Cambridge University Press (2011)
Freidlin, M.I., Wentzell, A.D.: Random perturbations of dynamical systems, 3rd edn. Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 260. Springer, Heidelberg (2012). Translated from the 1979 Russian original by Joseph Szücs. MR 2953753
Freidlin, M.I.: The averaging principle and theorems on large deviations. Russ. Math. Surv. 33(5), 117 (1978)
Freidlin, M.I., Sowers, R.B.: A comparison of homogenization and large deviations, with applications to wavefront propagation. Stoch. Process. Appl. 82(1), 23–52 (1999)
Friz, P.K., Victoir, N.B.: Multidimensional stochastic processes as rough paths. In: Cambridge Studies in Advanced Mathematics, vol. 120. Cambridge University Press, Cambridge (2010). Theory and applications. MR 2604669
Fukasawa, M.: Short-time at-the-money skew and rough fractional volatility. Quant. Finance 17(2), 189–198 (2017). MR 3592946
Gatheral, J., Jaisson, T., Rosenbaum, M.: Volatility is rough. Quant. Finance 18(6), 933–949 (2018)
Guennoun, H., Jacquier, A., Roome, P., Shi, F.: Asymptotic behavior of the fractional Heston model. SIAM J. Financ. Math. 9(3), 1017–1045 (2018). MR 3836176
Guerra, J., Nualart, D.: Stochastic differential equations driven by fractional Brownian motion and standard Brownian motion. Stoch. Anal. Appl. 26(5), 1053–1075 (2008)
Guillin, A.: Averaging principle of SDE with small diffusion: moderate deviations. Ann. Probab. 31(1), 413–443 (2003)
Hairer, M., Li, X.-M.: Averaging dynamics driven by fractional Brownian motion. Ann. Probab. 48(4), 1826–1860 (2020)
Horvath, B., Jacquier, A., Lacombe, C.: Asymptotic behaviour of randomised fractional volatility models. J. Appl. Probab. 56(2), 496–523 (2019). MR 3986948
Kilbas, A.A., Srivastava, H.M., Trujillo, J.J.: Theory and applications of fractional differential equations. In: North-Holland Mathematics Studies, vol. 204, p. 2218073. Elsevier Science B.V, Amsterdam (2006)
Klebaner, F.C., Liptser, R.: Moderate deviations for randomly perturbed dynamical systems. Stoch. Process. Appl. 80(2), 157–176 (1999)
Li, X.-M., Sieber, J.: Slow-fast systems with fractional environment and dynamics. Ann. Appl. Probab. 32(5), 3964–4003 (2022). MR 4498200
Liptser, R., Stoyanov, J.: Stochastic version of the averaging principle for diffusion type processes. Stoch.: Int. J. Probab. Stoch. Process. 32(3-4), 145–163 (1990)
Mishura, Y.S., Shevchenko, G.M.: Existence and uniqueness of the solution of stochastic differential equation involving wiener process and fractional Brownian motion with Hurst index \(h > 1/2\). Commun. Stat.-Theory Methods 40(19–20), 3492–3508 (2011)
Morse, M.R., Spiliopoulos, K.: Moderate deviations for systems of slow-fast diffusions. Asympt. Anal. 105(3–4), 97–135 (2017)
Nualart, D.: The Malliavin Calculus and Related Topics, vol. 1995. Springer (2006)
Pardoux, E., Veretennikov, Y.: On the Poisson equation and diffusion approximation. I. Ann. Probab. 29(3), 1061–1085 (2001)
Pardoux, E., Veretennikov, A.Y.: On the Poisson equation and diffusion approximation 3. Ann. Probab. 33(3), 1111–1133 (2005)
Pavliotis, G.A., Stuart, A.M.: Parameter estimation for multiscale diffusions. J. Stat. Phys. 127(4), 741–781 (2007)
Pei, B., Inahama, Y., Xu, Y.: Averaging principles for mixed fast-slow systems driven by fractional Brownian motion. arXiv preprint arXiv:2001.06945 (2020)
Samko, S.G., Kilbas, A.A., Marichev, O.I., et al.: Fractional integrals and derivatives. Gordon and Breach Science Publishers, Yverdon Yverdon-les-Bains (1993)
Shiryaev, A.N.: Essentials of Stochastic Finance: Facts, Models, Theory, vol. 3. World Scientific (1999)
Sottinen, T., Valkeila, E.: On arbitrage and replication in the fractional Black–Scholes pricing model. Stat. Dec. 21(2), 93–107 (2003). MR 2000665
Sottinen, T., Valkeila, E.: On arbitrage and replication in the fractional Black–Scholes pricing model. Stat. Decs. 21(2), 93–108 (2003)
Spiliopoulos, K.: Large deviations and importance sampling for systems of slow-fast motion. Appl. Math. Optim. 67(1), 123–161 (2013)
Spiliopoulos, K.: Fluctuation analysis and short time asymptotics for multiple scales diffusion processes. Stoch. Dyn. 14(03), 1350026 (2014)
Spiliopoulos, K., Morse, M.R.: Importance sampling for slow-fast diffusions based on moderate deviations. Multiscale Model. Simul. 18(1), 315–350 (2020)
Wenqing, H., Salins, M., Spiliopoulos, K.: Large deviations and averaging for systems of slow-fast stochastic reaction–diffusion equations. Stoch. Part. Differ. Equ.: Anal. Comput. 7(4), 808–874 (2019)
Young, L.C.: An inequality of the Hölder type, connected with Stieltjes integration. Acta Math. 67(1), 251–282 (1936). MR 1555421
Zähle, M.: On the link between fractional and stochastic calculus. Stoch. Dyn. 305–325 (1999)
Zähle, M.: Integration with respect to fractal functions and stochastic calculus. I. Probab. Theory Related Fields 111(3), 333–374 (1998)
Zhang, X.: A variational representation for random functionals on abstract wiener spaces. J. Math. Kyoto Univer. 49(3), 475–490 (2009)
Funding
Solesne Bourguin was partially supported by the Simons Foundation Award 635136. Konstantinos Spiliopoulos was partially supported by the National Science Foundation (DMS 1550918, DMS 2107856) and Simons Foundation Award 672441.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix
Appendix A. Fractional Brownian Motion and Pathwise Stochastic Integration
1.1 A.1 Fractional Brownian Motion: Definition and Main Properties
A one-dimensional fractional Brownian motion (fBm) is a centered Gaussian process \(W^H = \left\{ W^H_t :t \in [0,1] \right\} \subset L^2(\Omega )\), characterized by its covariance function
It is straightforward to verify that increments of fBm are stationary. The parameter \(H \in (0,1)\) is usually referred to as the Hurst exponent, Hurst parameter, or Hurst index.
By Kolmogorov’s continuity criterion, such a process admits a modification with continuous sample paths, and we always choose to work with such. In this case, one may show in fact that almost every sample path is locally Hölder continuous of any order strictly less than H. It is this sense in which it is often said that the value of H determines the regularity of the sample paths.
Note that when \(H = \frac{1}{2}\), the covariance function is \(R_{\frac{1}{2}}(t,s) = t \wedge s\). Thus, one sees that \(W^{\frac{1}{2}}\) is a standard Brownian motion, and in particular that its disjoint increments are independent. In contrast to this, when \(H \ne \frac{1}{2}\), nontrivial increments are not independent. In particular, when \(H > \frac{1}{2}\), the process exhibits long-range dependence.
Note moreover that when \(H \ne \frac{1}{2}\), the fractional Brownian motion is not a semimartingale, and the usual Itô calculus therefore does not apply.
Another noteworthy property of fractional Brownian motion is that it is self-similar in the sense that, for any constant \(a >0\), the processes \(\left\{ W^H_t :t \in [0,1]\right\} \) and \(\left\{ a^{-H} W^H_{at}:t \in [0,1]\right\} \) have the same distribution.
Finally, an n-dimensional fractional Brownian motion is a random vector where the components are independent one-dimensional fractional Brownian motions with the same Hurst parameter \(H\in (0,1)\).
The self-similarity and long-memory properties of the fractional Brownian motion make it an interesting and suitable input noise in many models in various fields such as analysis of financial time series, hydrology, and telecommunications. However, in order to develop interesting models based on fractional Brownian motion, one needs an integration theory with respect to it, which we present in the next subsection.
1.2 A.2 Pathwise Stochastic Integration with Respect to Fractional Brownian Motion
Stochastic integrals with respect to fractional Brownian motion can be understood, when \(H \ge 1/2\), as generalized Stieltjes integral as introduced in the work of Zähle [51]. Let \(f \in L^1([a,b])\) and \(\alpha > 0\). The left-sided and right-sided fractional Riemann–Liouville integrals of f of order \(\alpha \) are defined for almost all \(x \in [a,b]\) by
and
respectively, where \(\Gamma (\alpha )\) is the Euler gamma function. This naturally leads to the definition of the function spaces
and
The following integration by parts formula holds
for \(f\in L^p([a,b]),g\in L^q([a,b])\) such that \(1/p+1/q\le 1+\alpha \).
For \(0< \alpha <1\), we can define the fractional derivatives
and
as long as the right-hand sides are well-defined. Furthermore, if \(f \in I_{a^+}^{\alpha }\left( L^p([a,b])\right) \) (respectively \(f \in I_{b^-}^{\alpha }(L^p([a,b]))\) and \(0< \alpha <1\), then the previous fractional derivatives admit the Weyl representation
and
respectively, for almost all \(x \in [a,b]\). There is also the integration by parts formula
for \(f \in I_{a^+}^{\alpha }(L^p([a,b])),g \in I_{b^-}^{\alpha }(L^q([a,b])) \) such that \(1/p+1/q\le 1+\alpha \).
The upcoming lemma contains a useful technical result in [41].
Lemma 5
Let \(p\ge 1\) and \(b>0\). Then, the operator \(t^\beta I_{0^+}^\alpha t^\gamma \) is bounded in \(L^p([0,b])\) if \(\alpha >0, \alpha +\beta +\gamma =0\) and \((\gamma +1)p>1\). Meanwhile, the operator \(t^\beta I_{1^-}^\alpha t^\gamma \) is bounded in \(L^p([0,b])\) if \(\alpha >0, \alpha +\beta +\gamma =0\) and \((\alpha +\gamma )p<1\).
Proof
This is a consequence of Samko et al. [41, (5.45’) and (5.46’)]. \(\square \)
We refer to Samko et al. [41] for more detailed properties of fractional operators.
Next, let \(f(a+) = \lim _{x \searrow 0} f(a + x)\) and \(g(b-) = \lim _{x \searrow 0} g(b-x)\) and define
We recall from Zähle et al. [51] the definition of generalized Stieltjes fractional integrals with respect to irregular functions (in the sense of which we view the stochastic integrals with respect to fractional Brownian motion appearing in this paper).
Definition 2
(Generalized Stieltjes integral) Suppose that f and g are functions such that \(f(a+)\), \(g(a+)\) and \(g(b-)\) exist, \(f_{a^+} \in I_{a^+}^{\alpha }(L^p([a,b]))\) and \(g_{b^-} \in I_{b^-}^{1-\alpha }(L^p([a,b]))\) for some \(p,q \ge 1\), \(1/p + 1/q \le 1\), \(0< \alpha <1\). Then, the integral of f with respect to g is defined by
Remark 11
If \(\alpha p <1\), under the assumptions of the preceding definition, we have that \(f \in I_{a^+}^{\alpha }(L^p([a,b]))\) and we can write
In [51], it was further shown that if f and g are, respectively, \(\lambda \) and \(\mu \)-Hölder continuous such that \(\lambda +\mu >1\), then the conditions for the generalized Stieltjes integral \(\int _a^b f dg\) are satisfied for \(p=q=\infty \) and \(\alpha<\lambda ,1-\alpha <\mu \). In particular, this class of generalized Stieltjes integrals with Hölder continuous f, g coincides with the class of Riemann–Stieltjes integrals studied in [49] by Young. We note here that any Young’s integrals appearing in this paper are constructed from Hölder continuous paths of a fractional Brownian motion. Further details are given in [51, Sect. 5.1].
1.3 A.3 The Cameron–Martin Space of Fractional Brownian Motion
Consider the deterministic kernel
for which \(c_H=\left( H(2H-1)/\beta (2-2H,H-1/2)\right) ^{1/2}\). Slightly abusing notation, we also write \(K_H\) for the integral operator
For \(H \ge 1/2\), the operator \(K_H\) can be represented as
Additionally, we denote by \({\dot{K}}_H\) the “derivation” of the operator \(K_H\), i.e.,
The Cameron–Martin space \({\mathcal {H}}_H\) associated with \(W^H\) is
equipped with the inner product \(\left\langle g,f \right\rangle _{{\mathcal {H}}_H}=\left\langle {\hat{g}},{\hat{f}} \right\rangle _{L^2\left( [0,1];{\mathbb {R}}^m\right) }\). Note that later on, we will alternate between \({\hat{g}}\) and \(K_H^{-1}g\), which are equivalent ways of writing the same quantity.
In this paper, the noise process we consider in the slow–fast systems we study is of the form
where B is a m-dimensional standard Brownian motion, \(W^H\) is a p-dimensional fractional Brownian motion of Hurst parameter H and they are independent. We will hence need to work with the Cameron–Martin space associated with the process \((W^H,B)\), which, based on the previous description, is defined to be the space \({\mathcal {S}}\) given by
As a Cameron–Martin space, \({\mathcal {S}}\) is a Hilbert space equipped with the inner product given by
Let us now state an important fact regarding the differentiability of elements in \({\mathcal {H}}_H\) when \(H> 1/2\) which we will need throughout the paper.
Lemma 6
If \(H> 1/2\) and \(u\in {\mathcal {H}}_H\) such that \(u=K_H {\hat{u}},{\hat{u}}\in L^2([0,1];{\mathbb {R}}^n)\), then we have
such that \(c_H=\left( H(2H-1)/\beta (2-2H,H-1/2)\right) ^{1/2}\).
Meanwhile, if \(H=1/2\) and \(u\in {\mathcal {H}}_{1/2}\), then \({\dot{u}}_t={\hat{u}}_t\).
Proof
This is a direct consequence of formula (41). \(\square \)
The following is another important property of the operator \({\dot{K}}_H\).
Proposition 10
The map \({\dot{K}}_H\) as described in Lemma 6 is a bounded operator in \(L^2([0,1];{\mathbb {R}}^n)\).
Proof
The assumptions of Lemma 5 are satisfied for \(p=2,\alpha =H-1/2,\beta =0\) and \(\gamma =1/2-H\); hence, the operator \(I^{H-1/2}_{0^+}t^{1/2-H}\) is bounded in \(L^2([0,1];{\mathbb {R}}^n)\). Since \({\dot{K}}_H=t^{H-1/2}I^{H-1/2}_{0^+}t^{1/2-H}\) based on Lemma 6, this implies
\(\square \)
For more details about fractional Brownian motion, we refer the reader to the monographs [4, 36].
1.4 A.4 Results Related to Young’s Integrals
The two results presented here provide us with a way of bounding Young’s integrals and with a version of change of variable formula for differential equations that contain Young’s integrals, respectively.
Lemma 7
(Young–Loéve’s inequality) Let f and g be, respectively, \(\alpha \) and \(\beta \)-Hölder continuous, such that \(\alpha +\beta >1\). Then, a.s one has
Moreover, assume f is bounded, then
Proof
Refer to Friz and Victoir [22, Proposition 6.4]. \(\square \)
Theorem 3
For \(i=1,\ldots ,m\), let \(0<\alpha _i<1/2\), \(f^i\in I^{\alpha _i}_{0+}(L^2([0,b]))\) be bounded and \(g^i_{b-}\in I^{1-\alpha _i}_{b-}(L^2([0,b]))\), where the function \(g^i_{b-}\) is defined in Lemma 5. Moreover, assume \(h=(h^1,\ldots ,h^m)\) such that
Then, for any \({\mathcal {C}}^1\) mapping \(F:{\mathbb {R}}^m\times {\mathbb {R}}\rightarrow {\mathbb {R}}^n\) such that \(\frac{\partial F}{\partial x_i}\in {\mathcal {C}}^1,i=1,\ldots ,m\) and \(r\le t\le T\), it holds that
In particular, this change of variable formula applies to the special case when \(f_i\) and \(g_i\) are, respectively, \(\lambda _i\) and \(\mu _i\)-Hölder continuous such that \(\lambda _i+\mu _i>1,i=1,\ldots ,m\).
Proof
For the change of variable formula in the general case, we refer to Zähle [50, Theorem 5.2].
Now, let us consider the special case and assume there is a constant C such that \(\left|f_i \right|_{\lambda _i}\),\(\left|g_i \right|_{\mu _i}<C\) and \(\lambda _i+\mu _i>1\) for \(1\le i\le m\). Then, one can choose \(\alpha _i\) in the interval (0, 1/2) such that \(\lambda _i>\alpha _i\) and \(\mu _i>1-\alpha _i\) for \(1\le i\le m\). Based on the previous fact, Lemmas 13.2 and 13.2’ in [41] imply, respectively, that
Moreover, \(f^i\) as Hölder continuous functions on [0, b] are necessarily bounded. Consequently, the general change of variable formula covers this particular case. \(\square \)
Appendix B. Regularity Results and Other Technical Lemmas
This appendix gathers results related to Poisson equations as well as the technical lemmas required for the analysis of the control problems.
1.1 B.1 Results Related to Poisson Equations
The following theorem is a consequence of Pardoux and Veretennikov [37, Theorem 2] and [38, Theorem 3] for solutions of Poisson equations. Let \({\mathcal {L}}\) be the infinitesimal generator defined in (5).
Theorem 4
Recall \({\mathcal {C}}^{2,\zeta }({\mathbb {R}}^n\times {\mathcal {Y}})\) for some \(\zeta >0\) is the function space defined at the beginning of Sect. 2. Let \(h \in {\mathcal {C}}^{2,\zeta }({\mathbb {R}}^n\times {\mathcal {Y}})\) such that
and that for some positive constants K and \(D_h\),
uniformly with respect to x.
Then, there is a unique solution to
Moreover, \(u(\cdot ,y)\in {\mathcal {C}}^2\), \(\nabla ^2_x u\in {\mathcal {C}}({\mathbb {R}}^n\times {\mathcal {Y}})\) and there exists a positive constant M such that
Remark 12
Consider the Poisson equation in (6). Under Conditions H1 and H2-A, Theorem 4 states that there exists a positive constant C such that, uniformly,
On the other hand, under Conditions H1 and H2-B, Theorem 4 states that there exists a positive constant C such that, uniformly with respect to x,
1.2 B.2 Ancillary Results Related to the Control Problems
This subsection gathers all technical results related to the study of the control problems appearing throughout the paper.
Lemma 8
(Lemma 5.2 in [48]) Let \(H,H'\) be Hilbert spaces and \(a:H \rightarrow H'\) be a bounded linear operator. Moreover, let \(q=aa^*\) and \(q^{-1}\) be the inverse of q. Then, for any \(u\in H\),
Lemma 9
Assume that for all x and nonzero \(z\in {\mathbb {R}}^n\),
Then, the operator \(Q^{H}_x\) defined in (36) is invertible and its inverse \((Q^{H}_x)^{-1}\) is a bounded in \(L^2\left( [0,1];{\mathbb {R}}^n\right) \).
Proof
Using the operators \(\pi ,\pi ^*,\rho ,\rho ^*\) defined in Section 5.5, we have \(Q^{H}_xh=\left( \pi \pi ^*+\rho \rho ^*\right) h\) with
Furthermore, \(\pi \pi ^*,\rho \rho ^*\) are positive and self-adjoint operators, which means that \(Q^{H}_x\) is also positive and self-adjoint. In addition, the fact that \(Q^{H}_{x}\ge \rho \rho ^*\) and Condition H2-A or H2-B imply that \((Q^{H}_{x})^2\ge \left( \rho \rho ^*\right) ^2>0\). This leads to
so that \(Q^{H}_x\) is bounded from below and \(\ker Q^{H}_x=\{0\}\). This combined with self-adjointness implies
It follows that \(Q^{H}_{{\bar{X}}}\) is bijective. The operator \(Q^{H}_{{\bar{X}}}\) is also bounded in \(L^2\left( [0,1];{\mathbb {R}}^n\right) \) via Proposition 10, so we can conclude it has a bounded inverse by the inverse mapping theorem. \(\square \)
Lemma 10
It can be assumed that there exists a finite constant N such that, almost surely, the control process \(w^{\epsilon }\) appearing in the variational representation (12) satisfies
Proof
This is an immediate consequence of Zhang [52, Theorem 3.2]. \(\square \)
Lemma 11
Assume \(w^\epsilon \in {\mathcal {S}}\) is a control such that
for some finite constant N. Then, under Condition H1, it holds that for \(\epsilon _0>0\) small enough,
for some constant \(C>0\), which further implies that
Proof
The first estimate was proven in [47, Lemma 3.1]. For the second estimate, the dissipative property of the drift coefficient of \(Y^{\epsilon ,w^\epsilon }\) and Itô’s formula yield
We then apply the Burkhölder–Davis–Gundy inequality to the Itô integral term and Hölder’s inequality to the Riemann integral terms to get
Since \(\sigma (y)\sigma ^T(y)\) is bounded and \(\zeta (y)\) is sublinear, the first estimate of this lemma can be applied to the expression \({\mathbb {E}}{\left[ \int _0^1 \left|\zeta (Y^{\epsilon ,w^\epsilon }) \right|^2 ds \right] }\). Then, the simple fact that \(\int _r^t e^{-\frac{2}{\epsilon }\Gamma (t-s)}ds\le \epsilon \int _0^\infty e^{-2\Gamma s}ds=\frac{\epsilon }{2\Gamma }\) implies that
\(\square \)
Lemma 12
Assume \(w^\epsilon \in {\mathcal {S}}\) is a control such that
for some finite constant N.
-
(i)
Under Conditions H1 and H2-A, there exist constants C that change from line to line such that
$$\begin{aligned}&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left| \int _r^t \nabla _x\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) g\left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s \right) ds \right|\right] }\le C\rho ,\\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left|\int _r^t\nabla _y\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) \sigma \left( Y^{\epsilon ,w^\epsilon }_s\right) dB_s \right|^2\right] }\le C\rho \\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left| \int _r^t \nabla _y\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) \sigma \left( Y^{\epsilon ,w^\epsilon }_s\right) {\dot{v}}^\epsilon _sds \right|^2\right] }\le C\rho \\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left| \int _r^t \nabla _x\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) f\left( Y^{\epsilon ,w^\epsilon }_s\right) {\dot{u}}^\epsilon _sds \right|^2\right] }\le C\rho ,\\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}}\left|\int _0^t f\left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) {\dot{u}}^\epsilon _sds \right|^2\right] }\le C\rho . \end{aligned}$$ -
(ii)
Under Conditions H1 and H2-B, there exist constants C that change from line to line such that for any q in \(\Big (1,\frac{1}{D_f+D_g}\Big ]\), we have
$$\begin{aligned}&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left| \int _r^t \nabla _x\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) g\left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s \right) ds \right|^q\right] }\le C\rho ^{q-1},\\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left|\int _r^t\nabla _y\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) \sigma \left( Y^{\epsilon ,w^\epsilon }_s\right) dB_s \right|^{2q}\right] }\le C\rho ^{q-1}\\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left| \int _r^t \nabla _y\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) \sigma \left( Y^{\epsilon ,w^\epsilon }_s\right) {\dot{v}}^\epsilon _sds \right|^{2q}\right] }\le C\rho ^{q-1},\\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}} \left| \int _r^t \nabla _x\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) f\left( Y^{\epsilon ,w^\epsilon }_s\right) {\dot{u}}^\epsilon _sds \right|^{2q}\right] }\le C\rho ^{q-1},\\&{\mathbb {E}}{\left[ \sup _{\begin{array}{c} 0\le r,t\le 1 \\ \left|r-t \right|<\rho \end{array}}\left|\int _0^t f\left( Y^{\epsilon ,w^\epsilon }_s\right) {\dot{u}}^\epsilon _sds \right|^{2q}\right] }\le C\rho ^{q-1}. \end{aligned}$$
Proof
We start with part (i). The first estimate is straightforward due to the boundedness of \(\nabla _x\phi (x,y)\) stated in (43) and the boundedness of g(x, y) guaranteed by Condition H2-A. For the second estimate, we assume that \(0\le r\le t\le 1\) and apply the Burkhölder–Davis–Gundy inequality to obtain
For the third estimate, we can write
The last inequality in part (i) is a consequence of the boundedness of \(\sigma (y)\sigma ^\top (y)\) in Condition H1 and the boundedness of \(\nabla _x\phi (x,y)\) stated in (43) (requiring Condition H2-A). Finally, the two remaining estimates of part (i) are derived similarly to the previous one.
We continue with part (ii). For the first inequality, the sublinear growth of \(\nabla _x\phi (x,y)\) in y stated at (44) (requiring Condition H2-B) and the sublinear growth of g(x, y) in y from Condition H2-B imply for any q in \(\Big (1,\frac{1}{D_g}\Big ]\),
where the last inequality is due to Lemma 11. For the second estimate, assume that \(0\le r\le t\le 1\). Then, the Burkhölder–Davis–Gundy inequality combined with the sublinear growth of \(\nabla _y\phi (x,y)\) in y (requiring Condition H2-B) and the boundedness of \(\sigma (y)\sigma ^\top (y)\) in Condition H1 imply that for any q in \(\Big (1,\frac{1}{D_g}\Big ]\),
The arguments for the three remaining estimates of part (ii) are similar, so we will handle one case only. The sublinear growth of \(\nabla _x\phi (x,y)\) in y stated at (44) (requiring Condition H2-B) and sublinear growth of f(y) in y in Condition H2-B imply that for any q in \(\Big (1,\frac{1}{D_f+D_g}\Big ]\),
where the last inequality is once again obtained using Lemma 11. \(\square \)
Lemma 13
Assume \(w^\epsilon \in {\mathcal {S}}\) is a control such that
for some finite constant N. Under Condition H1, for \(0<\alpha \le 1/2\), we have the almost sure Hölder estimate
Proof
Without loss of generality, let us assume \(t>r\). The dissipative property of the drift coefficient of \(Y^{\epsilon ,w^\epsilon }\) and Itô’s formula yield
Now, by subtracting \(Y^{\epsilon ,w^\epsilon }_r\) from both sides and applying Hölder’s inequality along with the Burkhölder–Davis–Gundy inequality, we get
To bound the first term on the right-hand side, we combine the second estimate in Lemma 11 and the fact that \(e^{-\frac{1}{\epsilon }\Gamma (t-r)}-1=\frac{1}{\epsilon }\int _r^t e^{-\frac{1}{\epsilon }\Gamma (t-s)}ds\le \left|t-r \right|\). For the second term, note that \(\int _r^t e^{-\frac{2}{\epsilon }\Gamma (t-s)}ds=C\epsilon \left|t-r \right|.\) Moreover, the sublinearity of \(\zeta (y)\) and the first estimate in Lemma 11 yield a finite bound on the expression \({\mathbb {E}}{\left[ \sqrt{\int _0^1 \left|\zeta (Y^{\epsilon ,w^\epsilon }) \right|^2 ds }\right] }.\) The third term on the right-hand side of (45) can be treated similarly with the help of Lemma 10. Regarding the last term, recall that \(\sigma (y)\sigma ^T(y)\) is bounded in Condition H1. Thus, we have
The Kolmogorov continuity theorem then yields the almost sure Hölder continuity of \(Y^{\epsilon ,w^\epsilon }\). \(\square \)
Lemma 14
Assume \(w^\epsilon \in {\mathcal {S}}\) is a control such that
for some finite constant N. Under Conditions H1 and H2-A or H2-B, there exist a constant C and \(\epsilon _0\) small enough such that for \(0<\beta \le 1/2\),
Proof
We begin by proving the result under Conditions H1 and H2-A. According to Condition H2-A, f(x, y) is Lipschitz-continuous and bounded, so that f(x, y) is also \(\gamma \)-Hölder continuous for \(0<\gamma \le 1\). This further implies
and hence that for \(0<\gamma \le 1\),
This last estimate, together with the Young–Loéve inequality in Lemma 7, implies that for \(1-H<\beta \le 1\),
Meanwhile, a similar estimate to the one stated in part (i) of Lemma 12 states that
Moreover, boundedness of g(x, y) in Condition H2-A yields
Thus, using the estimate \({\mathbb {E}}{\left[ \left|Y^{\epsilon ,w^\epsilon } \right|_{\alpha }\right] }\le C\frac{1}{\sqrt{\epsilon }},\alpha \le \frac{1}{2}\) in Lemma 13 (which requires Condition H1), we can deduce that, for \(1-H<\beta \le 1/2\),
and consequently,
Now, by choosing \(\epsilon _0\) small enough, we get \({\mathbb {E}}{\left[ \left|X^{\epsilon ,w^\epsilon } \right|_\beta \right] }\le C\) for some constant C. Since for \(0<\beta _1\le \beta _2\le 1\), \(\beta _2\)-Hölder continuity of \(X^{\epsilon ,w^\epsilon }\) implies \(\beta _1\)-Hölder continuity, the conclusion follows.
We now present a proof of the claim under Conditions H1 and H2-B. Under Condition H2-B, f(y) is \(M_f\)-Hölder continuous, while \(Y^{\epsilon ,w^\epsilon }\) is \(\frac{1}{2}\)-Hölder continuous by Lemma 13, so that
or equivalently
Then, the Young–Loéve inequality in Lemma 7 implies that, for \(1-\frac{M_f}{2}<K\le H\),
where the first inequality is obtained by Condition H2-B and the last inequality is a consequence of the estimate \({\mathbb {E}}{\left[ \left|Y^{\epsilon ,w^\epsilon } \right|_{\alpha }\right] }\le C\frac{1}{\sqrt{\epsilon }},\alpha \le \frac{1}{2}\) in Lemma 13 (which requires Condition H1). Moreover, similar calculations to those performed in the proof of part (ii) of Lemma 12 yield that, for any q in \(\bigg (1,\frac{1}{D_f}\bigg ]\),
as well as
Consequently, we have
By choosing \(\epsilon _0\) small enough and noting that \(K>1-\frac{M_f}{2}\ge \frac{1}{2}\), we arrive at
for \(0\le \beta \le \frac{1}{2}\). \(\square \)
Lemma 15
Assume \(w^\epsilon \in {\mathcal {S}}\) is a control such that
for some finite constant N. Then, the following two assertions hold.
-
(i)
Under Conditions H1 and H2-A, there exists a constant C such that for any \(\beta \) in \((1-H,1]\),
$$\begin{aligned} {\mathbb {E}}{\left[ \sup _{0\le t\le 1}\left|\int _0^t f\left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) dW^H_s \right|\right] } \le C{\epsilon ^{-\frac{\beta }{2}}} \end{aligned}$$and
$$\begin{aligned} {\mathbb {E}}{\left[ \left|\sup _{t\in [0,1]} \int _0^t\nabla _x\phi \left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) f\left( X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s\right) d{W}^H_s \right|\right] }&\le C\epsilon ^{-\frac{1}{2}}. \end{aligned}$$ -
(ii)
Under Conditions H1 and H2-B, there exists a constant C such that
$$\begin{aligned} {\mathbb {E}}{\left[ \sup _{t\in [0,1]}\left|\int _0^t f\left( Y^{\epsilon ,w^\epsilon }_s\right) dW^H_s \right|\right] } \le C{\epsilon ^{-\frac{M_f}{2}}} \end{aligned}$$and
$$\begin{aligned} {\mathbb {E}}{\left[ \sup _{t\in [0,1]}\left|\int _0^t \nabla _x\phi (X^{\epsilon ,w^\epsilon }_s,Y^{\epsilon ,w^\epsilon }_s)f(Y^{\epsilon ,w^\epsilon }_s)dW^H_s \right|\right] }\le C\epsilon ^{-\frac{M_k}{2}}. \end{aligned}$$
Proof
We begin by proving part (i). The first estimate is immediate based on Lemma 14 and the estimate at (46). Regarding the second estimate, the inequality at (43) and Condition H2-A imply that \(\nabla _x\phi (x,y)f(x,y)\) is Lipschitz continuous. Hence, by the Young–Loéve inequality for \(1-H<\beta \le 1\), we have
where the last inequality is a consequence of Lemmas 13 and 14.
We now proceed to the proof of part (ii). For the first estimate, we perform a similar calculation to the one that was done at (47) (this requires Conditions H1 and H2-B) and get
Next, under Conditions H1 and H2-B, the \(M_k\)-Hölder continuity of \(\nabla _x\phi (x,y)f(x)\) together with the estimates in Lemmas 13 and 14 yields
so that
Therefore, as \(\frac{M_k}{2}+H>1\) in Condition H2-B, we can apply the Young–Loéve inequality to obtain
Lemma 16
Assume \(w^\epsilon \in {\mathcal {S}}\) is a control such that
for some finite constant N. Under Conditions H1 and either H2-A or H2-B, there exists a constant C such that
Furthermore, this implies for any \(\rho >0\),
Proof
Under Condition H2-A or H2-B, \(\nabla _x{\bar{g}}(x)\) is bounded. This fact, combined with equation (27) and the fact that \(X^{\epsilon ,w^\epsilon }\) converges to \({\bar{X}}\) in probability, implies that there exists some constant C such that
In addition, based on equation (25), we have
with
We will estimate the terms on the right-hand side of (49), starting with those which contain Young’s integrals. Condition H2-A guarantees that there exists some \(\beta \) in [0, 1] such that \(\beta +H>1\) and \(h(\epsilon )^{-1}\epsilon ^{-\frac{\beta }{2}}\rightarrow 0\) as \(\epsilon \rightarrow 0\), so that part (i) of Lemma 15 (which requires Conditions H1 and H2-A) yields
Part (i) of Lemma 15 also implies that
Meanwhile, under Condition H2-B, we use part (ii) of Lemma 15 to get, as \(\epsilon \rightarrow 0\),
and
The remaining terms on the right-hand side of (49), except the term
are bounded by using Lemmas 11 and 12 (which require Conditions H1 and H2-B). Thus, it follows from the estimates at (48) and (49) that
An application of Gronwall’s inequality then yields the first claim of our statement, which is
For the second claim, we proceed similarly to the derivation of the estimate at (48). Then, for \(\rho >0\),
\(\square \)
Lemma 17
Let \(R_2^\epsilon \) be the remainder term that appears in equation (27). Under Conditions H1 and either H2-A or H2-B , it holds that \(R_2^\epsilon \rightarrow 0\) in \(C([0,1];{\mathbb {R}}^{n})\) in probability along a subsequence.
Proof
For the purpose of identifying the limit of \(R_2^\epsilon \), we invoke the Skorokhod representation theorem and assume that \(X^{\epsilon ,w^\epsilon }\rightarrow {\bar{X}}\) a.s. in \(C([0,1];{\mathbb {R}}^{n})\) as \(\epsilon \rightarrow 0\). As \({\bar{X}}\) is bounded under Condition H2-A or H2-B, the dominated convergence theorem implies that
Now, we employ the bound (28) and get
In particular, the second inequality is due to the boundedness of \(\nabla ^2_x{\bar{g}}\) implied by either Condition H2-A or H2-B. The last inequality is a consequence of Lemma 16 (which requires Conditions H1 and either H2-A or H2-B). (50) then gives us the desired limit. \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Bourguin, S., Dang, T. & Spiliopoulos, K. Moderate Deviation Principle for Multiscale Systems Driven by Fractional Brownian Motion. J Theor Probab 37, 352–408 (2024). https://doi.org/10.1007/s10959-023-01235-y
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10959-023-01235-y