1: \begin{abstract}

2: Stochastic processes are often used to model complex scientific

3: problems in

4: fields ranging from biology and finance to engineering and physical science.

5: This paper investigates rate-optimal estimation of the volatility

6: matrix of a high-dimensional It\^{o} process

7: observed with measurement errors at discrete time points.

8: The minimax rate of convergence is established for estimating sparse

9: volatility matrices. By combining the multi-scale

10: and threshold approaches we construct a volatility matrix estimator to

11: achieve the optimal convergence rate.

12: The minimax lower bound is derived by considering a subclass of It\^{o}

13: processes for which the minimax

14: lower bound is obtained through a novel equivalent model of covariance

15: matrix estimation for independent but

16: nonidentically distributed observations and through a delicate

17: construction of the least favorable

18: parameters. In addition, a simulation study was conducted to test the

19: finite sample performance of the

20: optimal estimator, and the simulation results were found to support the

21: established asymptotic theory. %our theoretical findings.

22: 

23: %a $p$-dimensional It\^{o} process is observed with measurement errors

24: %at $n$ distinct time points, and our goal is to

25: %estimate the volatility matrix of the process. We establish the

26: %minimax theory for estimating large sparse volatility matrices under

27: %matrix

28: %norm as both $n$ and $p$ go to infinity. The theory shows that the

29: %optimal convergence rate depends on $n$ and $p$

30: %through $n^{-1/4} \sqrt{\log p}$ and a volatility matrix estimator is

31: %explicitly constructed to achieve the optimal convergence rate.

32: %diffusion processes play an important role in many modern scientific

33: %studies.

34: \end{abstract}