Professional Documents
Culture Documents
Math
Math
1. Introduction
Stochastic partial differential equations (SPDE) are used for stochastic
modelling , for instance, in the study of neuronal behviour in neurophysiology
and in building stochastic models of turbulence (cf. Kallianpur and Xiong,
1995). The theory of SPDE is investigated in Ito (1984), Rozovskii (1990)
and De prato and Zabczyk (1992) among others.
Huebner et al. (1993) started the investigation of maximum likelihood
estimation of parameters for a class of SPDE and extended their results to
parabolic SPDE in Huebner and Rozovskii (1995). Bernstein -von Mises the-
orems were developed for such SPDE in Prakasa Rao (1998, 2000b) following
the techniques in Prakasa Rao (1981). Asymptotic properties of Bayes esti-
mators of parameters for SPDE were discussed in Prakasa Rao (1998, 2000b).
Statistical inference for diffusion type processes and semimartingales in gen-
eral is studied in Prakasa Rao (1999a,b).
du (t, x) = (4u (t, x) + u (t, x))dt + θ(t) dWQ (t, x) (2.1)
∂ 2
where 4 = ∂x 2 . Suppose that θ(.) is a positive valued function with θ(t) ∈
C ([0, ∞)) for some m ≥ 1. Further suppose that θ2 (.) ∈ L2 (R) and that
m
the function θ(.) has a compact support contained in the interval [−, T + ]
for some > 0.
Further suppose the initial and the boundary conditions are given by
(
u (0, x) = f (x), f ∈ L2 [0, 1]
(2.2)
u (t, 0) = u (t, 1) = 0, 0 ≤ t ≤ T
stochastic pde with discrete observations 3
and Q is the nuclear covariance operator for the Wiener process WQ (t, x)
taking values in L2 [0, 1] so that
(cf. Rozovskii (1990)). It can be checked that the Fourier coefficient ui (t)
satisfies the stochastic differential equation
dui (t) = (1 − λi )ui (t)dt + √ θ(t)dWi (t), 0 ≤ t ≤ T (2.6)
λi + 1
with the initial condition
Z 1
ui (0) = vi , vi = f (x)ei (x)dx. (2.7)
0
Vj+1 = Vj ⊕ Wj (2.8)
and define
Then (i) for all −∞ < j < ∞, the collection of functions {φj,k , −∞ < k <
∞} is an orthonormal basis of Vj ; (ii) for all −∞ < j < ∞, the collection
of functions {ψj,k , −∞ < k < ∞} is an orthonormal basis of Wj ; and (iii)
the collection of functions {ψj,k , −∞ < j, k < ∞} is an orthonormal basis of
L2 (R).
In view of the earlier assumptions made on the function θ(t), it follows
that the function θ(t) belongs to the Sobolev space H m (R). Let j(n) be an
increasing sequence of positive integers tending to infinity as n → ∞. The
space L2 (R) has the following decomposition:
where Z
µj,k = θ2 (t)φj,k (t)dt (2.13)
R
and Z
νj,k = θ2 (t)ψj,k (t)dt. (2.14)
R
stochastic pde with discrete observations 5
We will now define estimators of the coefficients µj,k based on the observa-
tions {ui (tr ), tr = r2−n , j = 0, 1, . . . , [2n T ]}. Define
−1
(i) λi + 1 MX
µ̂j,k = φj,k (tr )(ui (tr+1 ) − ui (tr ))2 (2.15)
2 r=0
where M = [2n T ].
The subspace Vj is not finite dimensional. However, the functions θ2 and
the functions φ are compactly supported. Hence, for each resolution j, the
set of all k such that µj,k 6= 0 and the set of all k such that µ̂j,k 6= 0 is a
finite set Lj depending only on the constant T and the support of φ and the
cardinality of the set is O(2j ).
Define the estimator of θ2 (t) by
X (i)
θ̂i2 (t) = µ̂j(n),k φj(n),k (t) (2.16)
k∈Lj(n)
X (i)
= µ̂j(n),k φj(n),k (t). (2.17)
−∞<k<∞
Hence
(i) p
µ̂j,k → µj,k as n → ∞. (2.18)
Let h(.) be a continuous function on [0, T ] with compact support contained
0
in (0, T ) and belonging to the Sobolev space H m (R) with m0 > 12 . Let hj
be the projection of h on the space Vj . Further more suppose that
Soblev space to which h belongs to. Applying the Proposition 3.1 of Genon-
Catalot et al. (1992), we obtain that the following representation holds:
Z T
n/2
Jin ≡ 2 h(t)(θ̂i2 (t) − θ2 (t))dt
0
M
X −1 Z tr+1 Z tr+1
n/2 2
= 2 hj(n) (tr )[( θ(s) dWi (s)) − θ2 (s) ds] + Rin
r=0 tr tr
by Theorem 3.1 of Genon-Catalot et al. (1992). Note the estimators {θ̂i (t),
i ≥ 1} are independent estimators of θ(t) for any fixed t since the processes
{Wi , i ≥ 1} are independent Wiener processes.
Note that Qin is the integrated mean square error of the estimator θ̂i2 (t)
of the function θ2 (t) corresponding to the weight function γ(t). It can be
written in the form
2
Qin = Bin + Vin (2.23)
where Z T
2
Bin = γ(t)(E θ̂i2 (t) − θ2 (t))2 dt (2.24)
0
is the integrated square of the bias term with the weight function γ(t) and
Z T
Vin = E{ γ(t)(θ̂i2 (t) − E θ̂i2 (t))2 dt} (2.25)
0
is the integrated square of the variance term with the weight function γ(t).
Let Z T
Din = E{ (θ̂i2 (t) − E θ̂i2 (t))2 dt} (2.26)
0
and suppose that sup{γ(t) : t ∈ [0, T ]} ≤ K. Further suppose that j(n)− n2 →
−∞. Then it follows, by Theorem 4.1 of Genon-Catalot et al. (1992), that
stochastic pde with discrete observations 7
and Z T
Din = 2j(n)−n 2 θ4 (t)dt + o(2j(n)−n ). (2.28)
0
Further more
Vin ≤ KDin . (2.29)
Let
N
2 1 X
θ̃N (t) = θ̂2 (t). (2.30)
N i=1 i
It is obvious that, for any function h satisfying the conditions stated above,
and for any fixed integer N ≥ 1,
Z T
n/2 2
2 h(t)(θ̃N (t) − θ2 (t))dt
0
N
−1
X
= N Jin
i=1
N M −1 Z tr+1
−1
X X
n/2
= N {2 hj(n) (tr )[( θ(s) dWi (s))2
i=1 r=0 tr
Z tr+1 N
θ2 (s) ds]} + N −1
X
− Rin
tr i=1
M −1 N Z tr+1
−1
X X
n/2
= 2 hj(n) (tr ){N [( θ(s) dWi (s))2
r=0 i=1 tr
Z tr+1 N
θ2 (s) ds]} + N −1
X
− Rin .
tr i=1
Theorem 2.1. Under the conditions stated above , the estimator θ̃N 2 (t)
of θ2 (t)
satisfies the following property for any function h(t) as defined earlier:
Z T Z T
L
2n/2 2
h(t)(θ̃N (t) − θ2 (t))dt → N (0, 2N −1 h2 (t)θ4 (t) dt) as n → ∞.
0 0
(2.32)
Let γ(t) be a nonnegative continuous function with support contained in the
interval [0, T ]. Define
Z T
2
Qn = E{ γ(t)(θ̃N (t) − θ2 (t))2 dt}. (2.33)
0
Note that Qn is the integrated mean square error of the estimator θ̃N 2 (t)
2
of the function θ (t) corresponding to the weight function γ(t). It can be
written in the form
Qn = Bn2 + Vn (2.34)
where Z T
Bn2 = 2
γ(t)(E θ̃N (t) − θ2 (t))2 dt (2.35)
0
is the integrated square of the bias term with the weight function γ(t) and
Z T
2 2
Vn = E{ γ(t)(θ̃N (t) − E θ̃N (t))2 dt} (2.36)
0
is the integrated square of the variance term with the weight function γ(t).
Let Z T
2 2
Dn = E{ (θ̃N (t) − E θ̃N (t))2 dt}. (2.37)
0
We have the following theorem from the estimates on {Bin , 1 ≤ i ≤ N } and
on {Din , 1 ≤ i ≤ N } given above.
and Z T
Dn = N −1 2j(n)−n 2 θ4 (t)dt + o(N −1 2j(n)−n ). (2.39)
0
Further more
Vn ≤ KDn (2.40)
where K = sup{γ(t) : 0 ≤ t ≤ T }.
stochastic pde with discrete observations 9
(cf. Rozovskii, 1990). Following the arguments given in the Section 2, it can
be checked that the Fourier coefficient ui (t) satisfies the stochastic differen-
tial equation
dui (t) = −λi ui (t)dt + √ θ(t)dWi (t), 0 ≤ t ≤ T (3.4)
λi + 1
with the initial condition
Z 1
ui (0) = vi , vi = f (x)ei (x)dx. (3.5)
0
where Z
µj,k = θ2 (t)φj,k (t)dt (3.8)
R
and Z
νj,k = θ2 (t)ψj,k (t)dt. (3.9)
R
We will now define estimators of the coefficients µj,k based on the observa-
tions {ui (tr ), tr = r2−n , j = 0, 1, . . . , [2n T ]}. Define
−1
(i) λi + 1 MX
µ̂j,k = φj,k (tr )(ui (tr+1 ) − ui (tr ))2 (3.10)
2 r=0
where M = [2n T ].
The subspace Vj is not finite dimensional. However, the functions θ2 and
the functions φ are compactly supported. Hence, for each resolution j, the
set of all k such that µj,k 6= 0 and the set of all k such that µ̂j,k 6= 0 is a
finite set Lj depending only on the constant T and the support of φ and the
cardinality of the set is O(2j ).
Define the estimator of θ2 (t) by
X (i)
θ̂i2 (t) = µ̂j(n),k φj(n),k (t) (3.11)
k∈Lj(n)
X (i)
= µ̂j(n),k φj(n),k (t). (3.12)
−∞<k<∞
stochastic pde with discrete observations 11
Hence
(i) p
µ̂j,k → µj,k as n → ∞. (3.13)
Let h(.) be a continuous function on [0, T ] with compact support contained
0
in (0, T ) and belonging to the Sobolev space H m (R) with m0 > 12 . Let hj
be the projection of h on the space Vj . Further more suppose that
by Theorem 3.1 of Genon-Catalot et al. (1992). Note the estimators {θ̂i (t),
i ≥ 1} are independent estimators of θ(t) for any fixed t since the processes
{Wi , i ≥ 1} are independent Wiener processes.
Z T
Q̃in = E{ γ(t)(θ̂i2 (t) − θ2 (t))2 dt}. (3.17)
0
Note that Q̃in is the integrated mean square error of the estimator θ̂i2 (t)
of the function θ2 (t) corresponding to the weight function γ(t). It can be
written in the form
2
Q̃in = B̃in + Ṽin (3.18)
where Z T
2
B̃in = γ(t)(E θ̂i2 (t) − θ2 (t))2 dt (3.19)
0
is the integrated square of the bias term with the weight function γ(t) and
Z T
Ṽin = E{ γ(t)(θ̂i2 (t) − E θ̂i2 (t))2 dt} (3.20)
0
is the integrated square of the variance term with the weight function γ(t).
Let Z T
D̃in = E{ (θ̂i2 (t) − E θ̂i2 (t))2 dt} (3.21)
0
and suppose that sup{γ(t) : t ∈ [0, T ]} ≤ K. Further suppose that j(n)− n2 →
−∞. Then it follows, by Theorem 4.1 of Genon-Catalot et al. (1992), that
there exists a constant C̃i depending on , λi and the functions φ, γ and θ2
such that
2
B̃in ≤ C̃i (24j(n)−2n + 2−2j(n)(m∧r) + 2−n ) (3.22)
and Z T
j(n)−n
D̃in = 2 2 θ4 (t)dt + o(2j(n)−n ). (3.23)
0
Further more
Ṽin ≤ K D̃in . (3.24)
Let
N
2 1 X
θ̃N (t) = θ̂2 (t). (3.25)
N i=1 i
It is obvious that, for any function h satisfying the conditions stated above,
and for any fixed integer N ≥ 1,
Z T
n/2 2
2 h(t)(θ̃N (t) − θ2 (t))dt
0
N
= N −1 J˜in
X
i=1
stochastic pde with discrete observations 13
N M −1 Z tr+1
−1
X X
n/2
= N {2 hj(n) (tr )[( θ(s) dWi (s))2
i=1 r=0 tr
Z tr+1 N
θ2 (s) ds]} + N −1
X
− R̃in
tr i=1
M −1 N Z tr+1
−1
X X
n/2
= 2 hj(n) (tr ){N [( θ(s) dWi (s))2
r=0 i=1 tr
Z tr+1 N
θ2 (s) ds]} + N −1
X
− R̃in .
tr i=1
Theorem 3.1. Under the conditions stated above , the estimator θ̃N 2 (t) of
θ2 (t) satisfies the following property for any function h(t) as defined earlier:
Z T Z T
n/2 2 2 L −1
2 h(t)(θ̃N (t) − θ (t))dt → N (0, 2N h2 (t)θ4 (t) dt) as n → ∞.
0 0
(3.27)
Let γ(t) be a nonnegative continuous function with support contained in the
interval [0, T ]. Define
Z T
2
Q̃n = E{ γ(t)(θ̃N (t) − θ2 (t))2 dt}. (3.28)
0
Note that Q̃n is the integrated mean square error of the estimator θ̃N 2 (t)
2
of the function θ (t) corresponding to the weight function γ(t). It can be
written in the form
Q̃n = B̃n2 + Ṽn (3.29)
where Z T
B̃n2 = 2
γ(t)(E θ̃N (t) − θ2 (t))2 dt (3.30)
0
is the integrated square of the bias term with the weight function γ(t) and
Z T
2 2
Ṽn = E{ γ(t)(θ̃N (t) − E θ̃N (t))2 dt} (3.31)
0
14 b.l.s. prakasa rao
is the integrated square of the variance term with the weight function γ(t).
Let Z T
2 2
D̃n = E{ (θ̃N (t) − E θ̃N (t))2 dt}. (3.32)
0
Theorem 3.2. Suppose that j(n) − n2 → −∞. Then there exists a con-
stant C̃N depending on N, φ, γ, θ2 such that
and Z T
D̃n = N −1 2j(n)−n 2 θ4 (t)dt + o(N −1 2j(n)−n ). (3.34)
0
Further more
Ṽn ≤ K D̃n (3.35)
where K = sup{γ(t) : 0 ≤ t ≤ T }.
Remarks. It can be seen, from the Theorems 2.1 and 2.2 and from the
Theorems 3.1 and 3.2, that the limiting behaviour of the estimator θ̃N 2 (t)
of θ2 (t) does not depend on the “trend” terms in the SPDE’s discussed in
both the examples as long as the “trend” terms in the SDE’s satisfied by the
Fourier coefficients do not depend on the function θ(t) or any other unknown
functions. This has also been pointed out by Genon-Catalot et al. (1992) in
their work on the estimation of the diffusion coefficient for SDE’s.
References
Da Prato, G. and Zabczyk, J. (1992). Stochastic Equations in Infinite Dimensions,
Cambridge University Press.
Genon-Catalot, V., Laredo, C. and Picard, D. (1992). Non-parametric estimation
of the diffusion coefficient by wavelets methods, Scand. J. Statist., 19, 317-335.
Huebner, M., Khasminskii, R. and Rozovskii. B.L. (1993). Two examples of param-
eter estimation for stochastic partial differential equations, Stochastic Processes :
A Festschrift in Honour of Gopinath Kallianpur, Springer, New York, 149-160.
Huebner, M. and Rozovskii, B.L. (1995). On asymptotic properties of maximum
likelihood estimators for parabolic stochastic SPDE’s. Probab. Theory Related
Fields, 103, 143-163.
Ito, K. (1984). Foundations of Stochastic Differential Equations in Infinite Dimensional
Spaces, Vol. 47, CBMS Notes, SIAM, Baton Rouge.
Kallianpur, G. and Xiong, J. (1995). Stochastic Differential Equations in Infinite
Dimensions, Vol. 26, IMS Lecture Notes, Hayward, California.
stochastic pde with discrete observations 15
Kutoyants, Yu. (1994). Identification of Dynamical Systems with small noise, Kluwer
Academic Publishers, Dordrecht.
Prakasa Rao, B.L.S. (1981). The Bernstein-von Mises theorem for a class of diffusion
processes, Teor. Sluch. Proc., 9, 95-101 (In Russian).
Prakasa Rao, B.L.S. (1998). Bayes estimation for parabolic stochastic partial differ-
ential equations, Preprint, Indian Statistical Institute, New Delhi.
Prakasa Rao, B.L.S. (1999a). Statistical Inference for Diffusion type Processes, Arnold,
London and Oxford university Press, New York.
Prakasa Rao, B.L.S. (1999b). Semimartingales and their Statistical Inference, CRC
Press, Boca Raton, Florida and Chapman and Hall, London.
Prakasa Rao, B.L.S. (2000a). Nonparametric inference for a class of stochastic partial
differential equations, Tech. Report No. 293, Dept. of statistics and Actuarial
Science, University of Iowa.
Prakasa Rao, B.L.S. (2000b). Bayes estimation for some stochastic partial differential
equations. J. Statist. Plann. Inference, 91, 511-524.
Prakasa Rao, B.L.S. (2000c). Estimation for some stochastic partial differential equa-
tions based on discrete observations, Calcutta Statist. Assoc. Bull., 50, 193-206.
Prakasa Rao, B.L.S. (2001a). Nonparametric inference for a class of stochastic partial
differential equations II, Stat. Inference Stoch. Process., 4, 41-52.
Prakasa Rao, B.L.S. (2001b). Estimation for some stochastic partial differential equa-
tions based on discrete observations II, Preprint, Indian Statistical Institute, New
Delhi.
Prakasa Rao, B.L.S. (2001c). Statistical inference for stochastic partial differential
equations, In Selected Proc. Symp. Inference for Stoch. Proc., Ed. I.V.Basawa,
C.C.Heyde and R.L.Taylor, IMS Lecture Notes, Hayward, California, 37, 47-70.
Rozovskii, B.L. (1990). Stochastic Evolution Systems, Kluwer, Dordrecht.
B.L.S.Prakasa Rao
Indian Statistical Institute
7, S.J.S.Sansnwal Marg
New Delhi 110 016
INDIA
E-mail: blsp@isid.ac.in