Abstract
The aim of this paper is to study the asymptotic properties of the maximum likelihood estimator (MLE) of the drift coefficient for fractional stochastic heat equation driven by an additive space-time noise. We consider the traditional for stochastic partial differential equations statistical experiment when the measurements are performed in the spectral domain, and in contrast to the existing literature, we study the asymptotic properties of the maximum likelihood (type) estimators (MLE) when both, the number of Fourier modes and the time go to infinity. In the first part of the paper we consider the usual setup of continuous time observations of the Fourier coefficients of the solutions, and show that the MLE is consistent, asymptotically normal and optimal in the mean-square sense. In the second part of the paper we investigate the natural time discretization of the MLE, by assuming that the first N Fourier modes are measured at M time grid points, uniformly spaced over the time interval [0, T]. We provide a rigorous asymptotic analysis of the proposed estimators when \(N\rightarrow \infty \) and/or \(T,M\rightarrow \infty \). We establish sufficient conditions on the growth rates of N, M and T, that guarantee consistency and asymptotic normality of these estimators.
Similar content being viewed by others
Notes
Whenever convenient, we will also use the notation ‘\(\overset{d}{\longrightarrow }\)’ to denote the convergence in distribution of random variables.
Throughout the text we will use the notation \({\mathcal {N}}(\mu _0,\sigma _0^2)\) to denote a Gaussian random variable with mean \(\mu _0\) and variance \(\sigma _0^2\).
We will denote by C with subindexes generic constants that may change from line to line.
Throughout we will use the following version of the Chebyshev inequality: \({\mathbb {P}}(|X|>a)\le {\mathbb {E}}(X^2)/a^2\), for \(a>0\).
References
Bishwal, J.P.N.: Parameter Estimation in Stochastic Differential Equations. Lecture Notes in Mathematics, vol. 1923. Springer, Berlin (2008)
Bibinger, M., Trabs, M.: On central limit theorems for power variations of the solution to the stochastic heat equation. In: Proceedings in Mathematics and Statistics. Springer (Forthcoming) (2019)
Bibinger, M., Trabs, M.: Volatility estimation for stochastic PDEs using high-frequency observations. In: Stochastic Processes and their Applications (2019). https://doi.org/10.1016/j.spa.2019.09.002
Cheng, Z., Cialenco, I., Gong, R.: Bayesian estimations for diagonalizable bilinear SPDEs. In: Stochastic Processes and their Applications. (Forthcoming) (2019). https://doi.org/10.1016/j.spa.2019.03.020
Cialenco, I., Glatt-Holtz, N.: Parameter estimation for the stochastically perturbed Navier–Stokes equations. Stoch. Process. Appl. 121(4), 701–724 (2011)
Cialenco, I., Gong, R., Huang, Y.: Trajectory fitting estimators for SPDEs driven by additive noise. Stat. Inference Stoch. Process. 21(1), 1–19 (2018)
Cialenco, I., Huang, Y.: A note on parameter estimation for discretely sampled SPDEs. In: Stochastics and Dynamics (2019). https://doi.org/10.1142/S0219493720500161
Chow, P.: Stochastic Partial Differential Equations. Applied Mathematics and Nonlinear Science Series. Chapman & Hall, Boca Raton (2007)
Chong, C.: High-frequency analysis of parabolic stochastic PDEs. arXiv:1806.06959 (2019)
Cialenco, I.: Statistical inference for SPDEs: an overview. Stat. Inference Stoch. Process. 21(2), 309–329 (2018)
Cialenco, I., Lototsky, S.V., Pospíšil, J.: Asymptotic properties of the maximum likelihood estimator for stochastic parabolic equations with additive fractional Brownian motion. Stoch. Dyn. 9(2), 169–185 (2009)
Cialenco, I., Xu, L.: Hypothesis testing for stochastic PDEs driven by additive noise. Stoch. Process. Appl. 125(3), 819–866 (2015)
Lototsky, S.V.: Statistical inference for stochastic parabolic equations: a spectral approach. Publ. Mat. 53(1), 3–45 (2009)
Lototsky, S.V., Rozovsky, B.L.: Stochastic Partial Differential Equations. Springer, New York (2017)
Lototsky, S.V., Rozovsky, B.L.: Stochastic Evolution Systems. Linear Theory and Applications to Non-linear Filtering. Probability Theory and Stochastic Modelling, vol. 89, 2nd edn. Springer, New York (2018)
Markussen, B.: Likelihood inference for a discretely observed stochastic partial differential equation. Bernoulli 9(5), 745–762 (2003)
Nourdin, I., Peccati, G.: Normal Approximations with Malliavin Calculus, from Stein’s Method to Universality. Cambridge Tracts in Mathematics, vol. 192. Cambridge University Press, Cambridge (2012)
Nualart, D.: The Malliavin Calculus and Related Topics, Probability and Its Applications (New York), 2nd edn. Springer, Berlin (2006)
Piterbarg, L.I., Rozovskii, B.L.: On asymptotic problems of parameter estimation in stochastic PDE’s: discrete time sampling. Math. Methods Stat. 6(2), 200–223 (1997)
Prakasa Rao, B.L.S.: Nonparametric inference for a class of stochastic partial differential equations based on discrete observations. Sankhyā Ser. A 64(1), 1–15 (2002)
Prakasa Rao, B.L.S.: Estimation for some stochastic partial differential equations based on discrete observations. II. Calcutta Stat. Assoc. Bull. 54(215–216), 129–141 (2003)
Pasemann, G., Stannat, W.: Drift estimation for stochastic reaction–diffusion systems. Preprint arXiv:1904.04774v1 (2019)
Pospíšil, J., Tribe, R.: Parameter estimates and exact variations for stochastic heat equations driven by space-time white noise. Stoch. Anal. Appl. 25(3), 593–611 (2007)
Shiryaev, A.N.: Probability. Graduate Texts in Mathematics, vol. 2, 2nd edn. Springer, New York (1996)
Shubin, M.A.: Pseudodifferential Operators and Spectral Theory, 2nd edn. Springer, Berlin (2001)
Acknowledgements
The second author would like to thank David Nualart for fruitful discussions and advice that improved an early version of this paper. The authors are also grateful to the editors and the anonymous referees for their helpful comments and suggestions which helped to improve greatly the paper.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that there is 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
Proofs of technical Lemmas
Proof (Proof of Lemma 1)
We start by computing the Malliavin derivative of \(\widehat{F}_{N,T}\). If \(r\le t\) and for \(1\le k\le N\), then
Moreover, one has that \(D_{r,j}u_k(t)=0\) if \(j\ne k\) or \(r>t\). Therefore, for \(r\le T\) and \(1\le j\le N\), we have by (45),
We continue by setting
and in view of (33), we obtain
It is easy to see that for any process \(\varPhi =\{\varPhi (s), s\in [0,t]\}\) such that \(\sqrt{{\mathbb {V}}\mathrm {ar}(\varPhi (s))}\) is integrable on [0, t], it holds that \(\sqrt{ {\mathbb {V}}\mathrm {ar}\left( \int _0^t \varPhi _s d \!s \right) } \le \int _0^t \sqrt{{\mathbb {V}}\mathrm {ar}(\varPhi _s)} d \!s.\) Therefore, we have
where
For \(B_1\), by the independence of \(\{u_k\}_{k\ge 1}\) and (34),
For \(B_2\), we note that \(u_k\) and \(W_l\) are independent if \(k\ne l\). Therefore, we rewrite \(B_2\) as
By straightforward calculations, we have that
Therefore, we get
Let us now consider \(B_3\). Since \(w_k\) and \(w_j\) are independents for \(k\ne j\), we have
Finally, combining (35), (36) and (37), we have that for every \(\varepsilon >0\), there exist two independent constants \(N_0,T_0>0\) such that for all \(N\ge N_0\) and \(T\ge T_0\),
This completes the proof. \(\square \)
Proof (Proof of Lemma 2)
Using (9), (21) follows by direct evaluations. As far as (22) and (23), since \(u_k(t)-u_k(s)\) is a Gaussian random variable, it is enough to prove (22) and (23) for \(l=1\). We note that for \(t<s\), using (21), we deduce that
for some \(C>0\), and where in the last inequality we used the fact that \(e^{-x}\) is Lipschitz continuous on \([0,\infty )\). Hence, (22) is proved. Similarly, we have
for some \(C>0\), which implies (23). The proof is complete. \(\square \)
Proof (Proof of Lemma 3)
Since \(u_k, \ k \ge 1\), are independent, and taking into account that
we have that
and hence, by (22), there exist constants \(C_1,C_2>0\), such that
Hence, (24) follows at once.
Next we will prove (25). We note that
Consequently, letting \(U_i(t):=u_k^2(t_{i-1})-u_k^2(t), \ k\ge 1\), we continue
Note that by Cauchy-Schwartz inequality,
Again by Cauchy-Schwartz inequality and (38), we have that
Turning to \(I_2\), we first notice that
By the Wick’s Lemma [1, Lemma 3.1], we continue
For \(J_2\), we have
By (21), for \(i<j\) and \(t<s\),
for some \(c_2>0\). By similar arguments, we also obtain
for some \(c_3>0\). Thus,
for some \(c_4>0\). By analogy, one can treat \(J_1\) and \(J_3\), and derive the following upper bounds:
for some \(c_5,c_6>0\). Finally, combining the above, we have
Thus, using the estimates for \(I_1,I_2\), and the fact that \(\lambda _k\sim k^{1/d}\), we conclude that
and hence (25) is proved. The estimate (26) is proved by similar arguments, and we omit the details here. This completes the proof. \(\square \)
1.1 Auxiliary results
For reader’s convenience, we present here some simple, or well-known, results from probability. Let X, Y, Z be random variables, and assume that \(Z>0\) a.s.. For any \(\varepsilon >0\) and \(\delta \in (0,\varepsilon /2)\), the following inequalities hold true.
where \(\varPhi \) denotes the probability function of a standard Gaussian random variable.
Elements of Malliavin calculus
In this section, we recall some facts from Malliavin calculus associated with a Gaussian process, that we use in Sect. 3.1. For more details, we refer to [18]. Toward this end, let \(T>0\) be given. We consider the space \({\mathcal {H}}=L^2\left( [0,T]\times {\mathcal {M}}\right) \), where \({\mathcal {M}}\) is the counting measure on \({\mathbb {N}}\), namely, for \(v\in {\mathcal {H}}\),
We endow \({\mathcal {H}}\) with the inner product and the norm
We fix an isonormal Gaussian process \(W=\{W(h)\}_{h\in {\mathcal {H}}}\) on \({\mathcal {H}}\), defined on a suitable probability space \((\varOmega , {\mathscr {F}},{\mathbb {P}})\), such that \({\mathscr {F}}=\sigma (W)\) is the \(\sigma \)-algebra generated by W. Denote by \(C_p^{\infty }({\mathbb {R}}^n)\), the space of all smooth functions on \({\mathbb {R}}^n\) with at most polynomial growth partial derivatives. Let \({\mathcal {S}}\) be the space of simple functionals of the form
As usual, we define the Malliavin derivative D on \({\mathcal {S}}\) by
We note that the derivative operator D is a closable operator from \(L^p(\varOmega )\) into \(L^p(\varOmega ; {\mathcal {H}})\), for any \(p \ge 1\). Let \({\mathbb {D}}^{1,p}\), \(p \ge 1\), be the completion of \({\mathcal {S}}\) with respect to the norm
Also, for F of the form
we define the Malliavin derivative of F at the point t as
where \(\mathbb {1}_{A}\) denotes the indicator function of set A. For simplicity, from now on, we define \(W(t):=W\left( \mathbb {1}_{[0,t]}\right) \), \(t\in [0,T]\), to represent a standard Brownian motion on [0, T]. If \({\mathscr {F}}\) is generated by a collection of independent standard Brownian motions \(\{W_k,\ k\ge 1\}\) on [0, T], we define the Malliavin derivative of F at the point t by
Next, we denote by \(\varvec{\delta }\), the adjoint of the Malliavin derivative D (as defined in (43)) given by the duality formula
for \(F \in {\mathbb {D}}^{1,2}\) and \(v\in {\mathcal {D}}(\varvec{\delta })\), where \({\mathcal {D}}(\varvec{\delta })\) is the domain of \(\varvec{\delta }\). If \(v\in L^2(\varOmega ;{\mathcal {H}})\cap {\mathcal {D}}(\varvec{\delta })\) is a square integrable process, then the adjoint \(\varvec{\delta }(v)\) is called the Skorokhod integral of the process v (cf. [18]), and it can be written as
Proposition 1
[18, Theorem 1.3.8] Suppose that \(v \in L^2(\varOmega ;{\mathcal {H}})\) is a square integrable process such that \(v(t)\in {\mathbb {D}}^{1,2}\) for almost all \(t\in [0,T]\). Assume that the two parameter process \(\{D_tv(s)\}\) is square integrable in \(L^2\left( [0,T]\times \varOmega ;{\mathcal {H}}\right) \). Then, \(\varvec{\delta }(v) \in {\mathbb {D}}^{1,2}\) and
Next, we present a connection between Malliavin calculus and Stein’s method. For symmetric functions \(f\in L^2\big ([0,T]^q\big )\), \(q\ge 1\), let us define the following multiple integral of order q
with \(0<t_1<t_2<\cdots<t_q<T\). Note that \({\mathbb {I}}_q(f)\) is also called the q-th Wiener chaos [17, Theorem 2.7.7]. Denote by \(d_{TV}(F,G)\), the total variation of two random variables F and G.
Theorem 4
[17, Corollary 5.2.8] Let \(F_N={\mathbb {I}}_q(f_N)\), \(N\ge 1\), be a sequence of random variables for some fixed integer \(q\ge 2\). Assume that \({\mathbb {E}}\left( F_N^2\right) \rightarrow \sigma ^2>0\), as \(N\rightarrow \infty \). Then, as \(N \rightarrow \infty \), the following assertions are equivalent:
-
1.
\(F_N\overset{d}{\longrightarrow }{\mathcal {N}}:={\mathcal {N}}(0,\sigma ^2)\);
-
2.
\(d_{TV}\left( F_N,{\mathcal {N}}\right) \longrightarrow 0\).
We conclude this section with a result about an upper bound for the total variation of a q-th multiple integral and a Gaussian random variable.
Proposition 2
[17, Theorem 5.2.6] Let \(q\ge 2\) be an integer, and let \(F={\mathbb {I}}_q(f)\) be a multiple integral of order q such that \({\mathbb {E}}(F^2)=\sigma ^2>0\). Then, for \({\mathcal {N}}={\mathcal {N}}(0,\sigma ^2)\),
Rights and permissions
About this article
Cite this article
Cialenco, I., Delgado-Vences, F. & Kim, HJ. Drift estimation for discretely sampled SPDEs. Stoch PDE: Anal Comp 8, 895–920 (2020). https://doi.org/10.1007/s40072-019-00164-4
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40072-019-00164-4
Keywords
- Fractional stochastic heat equation
- Parabolic SPDE
- Stochastic evolution equations
- Statistical inference for SPDEs
- Drift estimation
- Discrete sampling
- High-frequency sampling