January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

Inverse Problems in Science and Engineering

Vol. 00, No. 00, January 2009, 1–14

RESEARCH ARTICLE

A method for identifying a spacewise dependent heat source

under stochastic noise interference

B. Tomas Johanssona∗ and Mihaela Pricopb

a
School of Mathematics, University of Birmingham, Edgbaston, Birmingham B15 2TT,
UK; b Institut für Mathematische Stochastik, Universität Göttingen, Goldschmidtstraße 7,
37073, Göttingen, Germany
(2008)

We consider an inverse problem of determining a spacewise dependent heat source in the time-
dependent heat equation using the usual conditions of the direct problem and information from
a finite amount of supplementary temperature measurements at a given single instant of time.
This spacewise dependent temperature measurement ensures that the inverse problem has a
unique solution, but the solution is unstable and hence the problem is ill-posed. Furthermore,
it is assumed that information about this temperature measurement is available only at a finite
number of points, and moreover, these measurements are blurred by a stochastic noise. The
continuous problem is discretized using a finite element method (FEM) and for the discrete
model an estimate of the heat source is obtained by a Tikhonov regularization. It is proved
that increasing the number of measurements and decreasing the mesh size produce a sequence
of solutions which under appropriate conditions converges to the continuous heat source.

Keywords: Finite element method, heat source, inverse problem; parabolic heat equation;
stochastic noise; Tikhonov regularization

AMS Subject Classification: Primary: 35R25; Secondary: 35K05

1. Introduction

When solving applied engineering problems there is usually a certain amount of

uncertainty involved, for example, the material coefficients are known only up to
some accuracy or through random external influences. Furthermore, the model is
usually of the form that the solution does not depend continuously on the data,
i.e. the problem is ill-posed. In this paper we study one such model having these
properties, i.e. being both ill-posed and stochastic. This model is described below.
We consider the problem of source identification from measured data for the
time-dependent heat equation. There are several important applications of such a
model, for example, in finding a pollution source intensity given measurements of
the pollutant concentrations at a later time and also for designing the final state
in melting and freezing processes.
Assume that we have a non-homogeneous and non-isotropic body, denoted by Ω,
occupying a bounded domain in Rn , where n ≥ 1. For simplicity, assume that the
temperature is zero on the boundary Γ of Ω. Then the temperature u, heat source

∗ Corresponding author. Email: [email protected]

ISSN: 1741-5977 print/ISSN 1741-5985 online

c 2009 Taylor & Francis
DOI: 10.1080/1741597YYxxxxxxxx
http://www.informaworld.com
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

2 Taylor & Francis and I.T. Consultant

f (x) and the initial temperature ϕ(x) satisfy


 ∂t u + Lu = f (x) in Ω × (0, T ),
u=0 on Γ × (0, T ), (1)
u(x, 0) = ϕ(x) for x ∈ Ω,


where T > 0. Here, L is a linear elliptic operator of the second-order with time-
independent smooth coefficients. A possible expression for L is Lu = −∇·(k(x)∇u),
where k > 0 is the thermal conductivity of the body. Using linearity, one can also
consider the case of a non-homogenous Dirichlet boundary condition. Moreover, the
ideas presented can, in principle, be applied to other types of boundary conditions
and to operators L with non-smooth coefficients corresponding to, for example,
layered materials.
In inverse problems it is of importance to specify appropriate data so that the
parameter to be reconstructed is uniquely identifiable. In our model, this will be
assured by using information about the solution obtained from a spacewise temper-
ature measurement at a later instant of time. This type of inverse problem of deter-
mining an unknown inhomogeneous spacewise dependent heat source function in
the parabolic heat equation has mainly been considered in a few theoretical papers
concerned with the existence and uniqueness of the solution, notably in [4, 14, 16].
Recently, see for example [9], an iterative procedure, using these rigorous math-
ematical conditions, was proposed and analysed for finding the source given the
additional information

u(x, T ) = ψ(x), for x ∈ Ω. (2)

Let us mention that the problem of determining ϕ from the spacewise dependent
temperature measurement (2), given that f = 0, i.e. the backward heat conduction
problem, has been extensively studied, see for example [5] and [13].
In most practical applications, however, only a finite amount of data is available.
In this work, we will therefore assume that there are N noisy measurements Ui of
the temperature u at time T > 0 at (random) points Xi described by an inverse
regression model

Ui = ψ(Xi ) + v(Xi )1/2 εi , i = 1, . . . , N. (3)

Here, (Xi , εi ) are independent and identically distributed random variables; further
requirements on these variables will be imposed below. The function v describes
the conditional variance of the measurement errors, and if v is constant the model
is called homoscedastic, otherwise is heteroscedastic. We point out that for the
method constructed in this paper no a priori information is required on the noise
level, i.e. the method is fully data driven. We also mention that the similar results
hold in the case of a deterministic design model, i.e. when the points Xi are known
and fixed. Moreover, we apply a frequentist’s approach and do not invoke Bayesian
statistics in this work. For different concepts of probability that can be incorpo-
rated, for example, belief and plausibility, see the introduction in [1]. Note also
that we consider the coefficients of the operator L, the domain Ω and the initial
data ϕ as fixed i.e. not stochastical.
An outline of the paper is as follows. In Section 2 we introduce some additional
assumptions and notation, and review some facts about the direct and inverse
source problems. In order to overcome the instability of the solution to (1)–(2),
we consider u(x, T ), where u solves (1), as a control that has to be determined to
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

Inverse Problems in Science and Engineering 3

satisfy (2). For the construction of the function f , we wish to minimize a functional
relating the discrepancy between the known and the calculated values on Ω × {T }.
In Section 3, we review some results for this functional proved in [10], for example,
it is twice Fréchet differentiable. We further show that the functional is weakly
continuous, see Lemma 3.2. For practical applications we discretize the source
problem using a FEM, see Section 4, and prove in Theorem 4.2 that the solution of
the discretized minimization tends to the solution of the continuous one. Then, in
Section 5, we consider the case where instead of (2) we only have finite amount of
data of the form (3). A method for obtaining an approximation of the heat source is
proposed in the beginning of Section 5. In the first step of this method, an estimate
of ψ is obtained from the measurements (3) using a local polynomial estimator, and
then this estimate is used in our discretized model to obtain an approximation to
the heat source using stochastic Tikhonov regularization. Employing recent results
about the stochastical properties of this approach, see [2], we show that increasing
the number of measurements and decreasing the mesh size produce a sequence of
solutions which is a minimizing sequence for the continuous discrepancy functional,
see Theorem 5.2. In Section 6, we present some numerical results based on the
proposed procedure. From these results it can seen that it is possible to obtain an
accurate reconstruction of the heat source. Finally, in Section 7, we present some
conclusions.

2. Solvability for the direct and inverse problems

2.1. Functional spaces

Let Ω be a bounded domain of Rn , where n ≥ 1, with Lipschitz boundary ∂Ω = Γ.

The space L2 (Ω) consists of square integrable functions on the domain Ω with
the usual norm k · kL2 (Ω) and scalar product ( · , · ). As usual, the space H k (Ω),
where k = 1, 2, . . . , denotes the standard Sobolev space on Ω, i.e. the space of
functions with generalized derivatives of order ≤ k in L2 (Ω). Since the boundary
of Ω is Lipschitz, the trace of functions in H 1 (Ω) to the boundary is well-defined
and H01 (Ω) consists of functions in H 1 (Ω) with u|Γ = 0.
Let T > 0 be a fixed number. The space L2 (0, T ; X), where X is a Hilbert space,
denotes the space of measurable functions u(·, t) : (0, T ) → X, such that

Z T
ku(·, t)k2X dt < ∞.
0

By C([0, T ]; X) we mean functions u such that u(·, t) : [0, T ] → X is continuous

(in the usual norm).
Let (ΩP , F, P) be a complete probability space, where ΩP is the set of outcomes,
F is a σ-algebra of events and P is a probability measure (P : F → [0, 1]). Assume
that Y is an Rd -valued random variable in (ΩP , F, P), 1
R which is in L (ΩP ). By
E[Y ] we denote the expected value of Y , i.e. E[Y ] = ΩP Y (ω) dP (ω), and, when
defined, the variance is Var[X] = E[(X − E[X])2 ].
For the random variables in (3) it is assumed that E[εi |Xi ] = 0 and E[ε2i |Xi ] = 1.
Moreover, E[Yi |Xi ] = ψ(Xi ) and for the variance Var [Yi |Xi ] = v(xi ), where the
function v ∈ L2 (ΩP ). The random variables Xi have values in Ω distributed ac-
cording to a common, but unknown design density fΩP .
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

4 Taylor & Francis and I.T. Consultant

2.2. The differential operator L

Let
n
X n
X
Lu = − ∂xi (ai,j (x)∂xj u) + bi (x)∂xi u + c(x)u, (4)
i,j=1 i=1

where ai,j = aj,i and ai,j , bi ∈ C 1 (Ω) and c ∈ C(Ω). We suppose that this operator
L is elliptic, i.e., there exists a constant C such that
n
X
ai,j (x)ξi ξj ≥ C|ξ|2 , for every x ∈ Ω and all ξ ∈ Rn .
i,j=1

Since we can always change u into eλt u, the coefficients are assumed chosen such
that there exists a constant C > 0 with

a(u, u) + λkuk2L2 (Ω) ≥ Ckuk2H01 (Ω) (5)

for all u ∈ H01 (Ω) and λ ≥ 0, where

n Z
X n Z
X Z
a(u, v) = ai,j (x)(∂xi u)(∂xj v) dx + bi (x)(∂xi u)v dx + c(x)uv dx.
i,j=1 Ω i=1 Ω Ω

Given these conditions the operator −L generates a C0 (continuous) contraction

semigroup {G(t)}t∈R+ in L2 (Ω) with D(L) = {u : u ∈ H01 (Ω), Lu ∈ L2 (Ω)} a
dense set of its domain, see for example [7]. Moreover, −L is strictly dissipative
and hence

kG(T )k < 1. (6)

The space H01 (Ω) can be compactly embedded into L2 (Ω), which implies that
G(t) is a compact operator for every t > 0. We point out that the above results
hold under weaker assumptions on the coefficients, for example, the coefficients can
be piecewise smooth, corresponding to a layered material.
As a direct consequence of the fact that −L generates a semigroup we have the
following solvability result due to [12, Chapter 3].
Lemma 2.1: Suppose that the initial temperature ϕ and the heat source f
are given in L2 (Ω). Then (1) has a unique solution u ∈ L2 (0, T ; H01 (Ω)) ∩
C([0, T ]; H01 (Ω)) in the distributional sense and

kukL2 (0,T ;H01 (Ω)) ≤ C(kf kL2 (Ω) + kϕkL2 (Ω) ). (7)

Note that the homogeneous Dirichlet boundary condition in (1) is satisfied since
the solution belongs to the space L2 (0, T ; H01 (Ω)).

2.3. Solvability of the inverse problem (1)–(2)

We only consider the case of strong solutions and therefore assume that ψ ∈
H01 (Ω) ∩ H 2 (Ω) and that the heat source f is sought in L2 (Ω). The set of ad-
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

Inverse Problems in Science and Engineering 5

missable temperatures consists of functions u with

Z T  
kut (·, t)k2L2 (Ω) + ku(·, t)k2H 2 (Ω) dt < ∞. (8)
0

Exploring the fact that the operator −L generates a contraction semigroup and
condition (6), we have the following result due to [15, Theorem 1].
Theorem 2.2 : Let ϕ, ψ ∈ H01 (Ω)∩H 2 (Ω), then the inverse problem (1)–(2) has a
unique solution among heat sources f ∈ L2 (Ω) and temperatures u that satisfy (8).
This result can be extendend to include weaker assumptions on the data but one
then has to consider an appropriate weak formulation of (1)–(2). In Lemma 3.1
in [10] it was shown that there exists a solution for all sources belonging to a dense
set of L2 (Ω). As was pointed out in [10, 15] the assumption that both kϕkH 1 (Ω) ,
kψkH 1 (Ω) → 0 is not sufficient to guarantee the stability of the heat source f .

3. A discrepancy functional and some of its properties

Let the operator A : L2 (Ω) → L2 (Ω) be defined by

Ag = u(x, T ), (9)

where u is the unique (weak) solution to (1) with heat source f (x) = g(x), guar-
anteed by Lemma 2.1. This operator is well-defined and bounded according to
Lemma 2.1 but not linear (unless ϕ = 0) but affine linear. From [10, Section 3.1]
the equation (9) has a unique solution and the range of A is dense in L2 (Ω). To
construct a solution to (1)–(2), we shall minimize the (quadratic) functional

1
J(g) = kAg − ψk2L2 (Ω) , (10)
2

with respect to sources g ∈ L2 (Ω). The following result is the content of Theo-
rem 3.2 in [10].
Theorem 3.1 : The functional J given by (10) is twice Fréchet differentiable
and strictly convex. The gradient of J is given by

J ′ (g) = −v(x, 0), (11)

where v is the weak solution to


 ∂t v − L∗ v = g in Ω × (0, T ),
v=0 on Γ × (0, T ), (12)
v(x, T ) = ζ for x ∈ Ω,


with ζ = 0 and g = u(x, T ) − ψT , and

n
X n
X
∗
L v=− ∂xi (ai,j (x)∂xj v) − ∂xi (bi (x)v) + c(x)v.
i,j=1 i=1
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

6 Taylor & Francis and I.T. Consultant

In the proof of this Theorem given in [10], the following estimate is shown

(J ′′ (f )h, h)L2 (Ω) ≥ kuh (·, T )k2L2 (Ω) > 0, (13)

where uh is a solution to (1) with f = h and ϕ = 0.

The next lemma is needed for the convergence of the regularizing method that
we develop to reconstruct the heat source.
Lemma 3.2: The functional J defined in (10) is weakly continuous.
Proof : Let fk be a sequence in L2 (Ω) that converges weakly in L2 (Ω) to a function
f ∈ L2 (Ω). Denote by uk the solution of the boundary value problem (1) with the
source data f = fk . Then from the definition of a weak solution in L2 (0, T ; H01 (Ω))
we have
Z T Z Z T Z
(u − uk )∂t v dx dt + a(u − uk , v) = − (f − fk )v dx dt,
0 Ω 0 Ω

for every v ∈ L2 (0, T ; H01 (Ω)). Since fk converges to f weakly, the right-hand side
tends to zero when k → ∞. Therefore,
Z T Z
(u − uk ) ∂t v dx dt + a(u − uk , v) → 0 when k → ∞.
0 Ω

This and the inequality (5) imply that uk (·, t) converges weakly to u(·, t) in the
H01 (Ω)) norm for t ∈ (0, T ]. Combining this with the compact imbedding of H01 (Ω)
into L2 (Ω) give in particular that uk (·, T ) converges strongly to u(·, T ) in L2 (Ω).
Now, using the definition of the functional J in (10) implies that J(gk ) converges
to J(g).

Since the domain of A is closed and convex and by the Lemma 3.2 A is weakly
continuous, we conclude that:
Lemma 3.3: The operator A in (9) is weakly closed.

4. A discretized version of the minimization problem

In this section we introduce a discretization of the functional J in (10) using a FEM.

We prove that the minimum to this discretized problem tends to the minimum of
the continuous one as the mesh size tends to zero.

4.1. Some further assumptions and FEM spaces

For simplicity, to avoid singularities in the solution, we assume that Ω is a convex

polyhedral domain. There are several different FEM’s for parabolic equations, for
an overview see, for example, [17]. In principle, any FEM can be applied for the
discretization in the next section; for the sake of completeness we outline one
possible FEM in this section.
Let Xhx ⊂ H01 (Ω) be a family of finite element approximation spaces consisting
of piecewise linear continuous functions on conforming (non-degenerate) triangu-
lations τhx of the domain Ω with a maximum mesh spacing parameter hx > 0.
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

Inverse Problems in Science and Engineering 7

Standard approximation estimates then hold for these finite element spaces, e.g.

min ku − uhx kH01 (Ω) ≤ Chx kukH 2 (Ω) , for every u ∈ H01 (Ω) ∩ H 2 (Ω), (14)
uhx ∈Xhx

where C is independent of both the function u and the mesh size hx . By Phx we
denote the projection of L2 (Ω) onto Xhx , defined (on H 1 (Ω)) by

a(Phx u, v) = (f, v)L2 (Ω) , for every v ∈ Xh .

Note that Phx is not necessarily symmetric but due to (5) it is positive definite
on the space Xhx . Then let {tm }M m=0 be a finite sequence of numbers with t0 = 0
and tM = T , and maximum time-step size ht . We consider the finite element
space Tht of (discontinuous) piecewise polynomials with degree ≤ 1 on each interval
(tm , tm+1 ), m = 0, 1, . . . , M − 1. This space has the similar approximation property
as (14), namely

min kv − vht kL2 ((0,T )) ≤ Ch2t k∂t2 vkL2 ((0,T )) , for every v ∈ H 2 ((0, T )). (15)
vht ∈Tht

We define the tensor space Vhx ,ht = Xhx ⊗ Tht . Let u be a solution to (1) and
ϕhx = Phx ϕ. The finite element approximation uhx ,ht ∈ Vhx ,ht of u can then be
defined as the element in Vhx ,ht with uhx ,ht (x, 0) = ϕhx (x) and
Z tm+1 h i
(uhx ,ht , ∂t v)L2 (Ω) + a(uhx ,ht , v) dt + (u+ +
hx ,ht (·, tm ), v (·, tm ))L2 (Ω)
tm
(16)
Z tm+1
−(u− −
hx ,ht (·, tm+1 ), v (·, tm+1 ))L2 (Ω) = (f, v)L2 (Ω) dt
tm

for every v ∈ Vhx ,ht and m = 0, 1, . . . , M − 1. Here, v +(−) (·, tm ) =

lims→0+(−) v(·, tm + s). It is a standard fact that uhx ,ht exists and is unique, see
for example [17].

4.2. The discretized functional J and some of its properties

Let ϕhx = Phx ϕ and Phx ψ = ψhx , where Phx is the operator defined in the previous
section mapping into the space Xhx .
Let u ∈ L2 (0, T ; H01 (Ω)) ∩ C([0, T ]; H01 (Ω)) be the weak solution to (1) guaran-
teed by Lemma 2.1. Let Shx be the image of L2 (Ω) under the projection Phx and
introduce the operator Ahx ,ht : Shx → Vhx ,ht by

Ahx ,ht f˜ = uhx ,ht (x, T ), (17)

where uhx ,ht ∈ Vhx ,ht is the (unique) Galerkin approximation of the weak solution u
with initial data ϕhx , constructed in the previous section. From (16) with m = M −1
and with v time-independent, it follows that Ahx ,ht = Phx A, where A is defined
in (9) and Phx is the operator mapping into the Galerkin space Xhx . Thus, from
the properties shown for the operator A in Section 3 we note, in particular, that
Ahx ,ht is weakly closed and that the equation

Ahx ,ht f˜ = ψhx , (18)

January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

8 Taylor & Francis and I.T. Consultant

has a solution in Shx . To construct a solution let

1
Jhx ,ht (f ) = kAhx ,ht f − ψhx (·)k2L2 (Ω) , (19)
2

where f ∈ Shx . From Theorem 3.1 it follows that also Jhx ,ht is twice Fréchet
differentiable and one can find an expression for its derivative that will involve the
corresponding Galerkin approximation of (12).
Denote by J ∗ and Jh∗x ,ht the minimum of the functionals J and Jhx ,ht , respec-
tively.

Lemma 4.1: Given a tolerance δ > 0 there is a number hδ > 0 such that for
max{hx , ht } < hδ and f ∈ L2 (Ω)

|J(f ) − Jhx ,ht (Phx f )| < δ. (20)

Proof : From the definition of the functionals J and Jhx ,ht we have

1  2 2

|J(f ) − Jhx ,ht (Phx f )| =  ku(·, T ) − ψ(·)kL2 (Ω) − kuhx ,ht (·, T ) − ψhx (·)kL2 (Ω) 

2
1 
= ku(·, T ) − ψ(·)kL2 (Ω) + kuhx ,ht (·, T ) − ψhx (·)kL2 (Ω) × (21)
2
 
 ku(·, T ) − ψ(·)kL2 (Ω) − kuhx ,ht (·, T ) − ψhx (·)kL2 (Ω) .
 

Applying the triangle inequality to the expression on the right-hand side of (21),
 
|J(f )−Jhx ,ht (Phx f )| ≤ C ku(·, T )−uhx ,ht (·, T )kL2 (Ω) +kψ(·)−ψhx (·)kL2 (Ω) . (22)

Due to the homogeneous boundary condition and the smoothness of ψ it follows

that u is sufficiently smooth near t = T , so standard Galerkin error estimates can
be applied showing that the right-hand side of (22) tends to zero as max{hx , ht }
tends to zero.

We can construct a sequence fhxj ,htj , j = 1, 2, . . ., such that {hxj , htj } tends to
zero as j tends to infinity and

Jh∗xj ,htj ≤ Jh∗xj ,htj (fhxj ,htj ) + εhxj ,htj , (23)

with εhxj ,htj → 0 as j → ∞. We can then show, following [6, Theorem 6]:

Theorem 4.2 : For a given tolerance δ > 0, by chosing max{hx , ht } sufficiently

small, we have

|J ∗ − Jh∗x ,ht | < δ.

Proof : Given a δ > 0 there exists an element fδ such that

δ
J ∗ ≤ J(fδ ) ≤ J ∗ + . (24)
2
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

Inverse Problems in Science and Engineering 9

According to Lemma 4.1 we can choose h so small that

δ
|J(fδ ) − Jhx ,ht (Phx fδ )| < .
2

Combining this with (24) we conclude that for max{hx , ht } sufficiently small

Jh∗x ,ht < J ∗ + δ. (25)

Conversely, we also find that for max{hx , ht } sufficiently small

J ∗ < Jh∗x ,ht + δ. (26)

Then, employing the sequence in (23), we obtain

J(fhxj ,htj ) − J ∗ ≤ |J(fhxj ,htj ) − Jhxj ,htj (fhxj ,htj )| + |Jhxj ,htj (fhxj ,htj ) − Jh∗xj ,htj |
(27)
+|J ∗ − Jh∗xj ,htj |.

The first term on the right-hand side of (27) tends to zero according to Lemma 4.1,
the second term tends to zero from the construction of fhxj ,htj , and the last term
tends to zero from the first part of the proof. The proof is thus complete.

Remark. Using the estimate (22) and standard Galerkin approximation estimates
it is possible to explicitly establish how to choose hx and ht so that the first and
third term on the right-hand side in (27) each are less than δ/3. For the second
term on the right-hand side in (27) one can undertake an analysis based on the
singular value decomposition to control its size.

5. Minimization of the discrete functional (19)

From now on, we assume that data is chosen so that there exists a source solution
f ∈ C 2 (Ω) to the inverse problem (1)–(2), and that the design density (for Xi )
is in C 1 (Ω). Then, to minimize, in a stable way, the non-linear functional Jhx ,ht
defined in (19), we can apply a recent Tikhonov regularization method described
in [2]. We assume that we only have access to a finite number N of measurements
of the function ψ of the form (3). The method we propose and investigate is the
following.
1. Obtain an estimate ψ̂ (N ) of ψ.
(N )
2. Construct the Galerkin approximation ψ̂hx ∈ Xhx of ψ̂ (N ) .
(N )
3. Substitute ψ̂ (N ) for ψ̂hx in (19) and minimize Jhx ,ht (f )+αN kf −f0 k2L2 (Ω) , where
αN > 0 is the regularization parameter and f0 is a guess of the actual source.
We shall show that we can obtain a minimizing sequence to (10) by using the
above method where the number of measurements is increasing and the mesh size
decreasing. For this we choose a specific estimate ψ̂ (N ) of ψ such that

E[kψ − ψ̂ (N ) k2L2 (Ω) ] = O(βN

2
), (28)

where βN is a sequence of real numbers tending to zero when the number of ob-
servations N tends to infinity. To construct this estimate from (3) one can, for
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

10 Taylor & Francis and I.T. Consultant

example, use a local linear polynomial estimator, see [2, Section 4].
(N )
Let ψ̂hx ∈ Xhx be the Galerkin approximation of ψ̂ (N ) . We note that (28) implies
that also

(N )
E[kψhx − ψ̂hx k2L2 (Ω) ] = O(βN
2
). (29)

(N )
Now, let fˆhx ,ht be a minimizer of

Mhx ,ht = kAhx ,ht (f˜) − ψhNx kL2 (Ω) + αN kf˜ − f˜0 k2L2 (Ω) , (30)

where f˜ is in Sh and f˜0 = Phx f0 and let fh∗x ,ht be a minimizer of (19). Since the
operator Ahx ,ht is weakly closed the following result is a consequence of [2, Theorem
2],
Lemma 5.1: There exists a sequence of regularizing parameters α = αN , N =
1, 2, . . . , such that

(N )
E[kfh∗x ,ht − fˆhx ,ht k2L2 (Ω) ] → 0, (31)

as N → ∞.
Remark The regularizing parameters in Lemma 5.1 can be estimated by a-priori
methods, see [2, Section 3].
We can then prove the following result:
Theorem 5.2 : Given a tolerance δ > 0 one can find hx , ht and an integer N
such that
h i
(N )
E |J ∗ − J(fˆhx ,ht )|2 < δ. (32)

Proof : Note that by definition Jh∗x ,ht = Jhx ,ht (fh∗x ,ht ). From (27) we find

h i h i
(N ) (N ) (N )
E |J(fˆhx ,ht ) − J ∗ |2 ≤ E |J(fˆhx ,ht ) − Jhx ,ht (fˆhx ,ht )|2
h i h (33)
i
(N )
+E |Jhx ,ht (fˆhx ,ht ) − Jhx ,ht (fh∗x ,ht )|2 + E |J ∗ − Jh∗x ,ht |2 .

We can choose the mesh size so that the first and third term on the left-hand side
each is less than δ/3 according to Lemma 4.1 and Theorem 4.2 respectively, and
by the previous lemma we can then choose N so that the second term is also less
than δ/3.
(N (δ)) (N (δ))
We can then construct a sequence fhx(δ) ,ht(δ) with J(fhx(δ) ,ht(δ) ) → J ∗ as δ → 0.
Since, according to Lemma 3.2, the functional J is weakly continuous we can
(N (δ))
conclude that also fhx(δ) ,ht(δ) converges weakly.

We remark that instead of using a Tikhonov type regularization for minimiz-

ing (19) one can apply, for example, a Landweber-Fridman type iterative procedure,
see [3]. Invoking such an iterative procedure is deferred to future work.
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

Inverse Problems in Science and Engineering 11

6. Numerical Simulations

1
spline order 1
spline order 2
0.8

Exact heat source

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
x

Figure 1. The exact heat source f as a spline of order one and two

In this section we illustrate the method described in Section 5 by giving nu-

merical results. Let the domain Ω be the interval (0, 1) and the final time T = 1.
We choose the operator L as the negative Laplace operator, i.e. L = −∆, the ini-
tial condition u(x, 0) = 2 sin(πx) and homogeneous Dirichlet boundary conditions
u(0, t) = u(1, t) = 0, where t ∈ (0, 1), in the equation (1). The direct problem is
described by

 ∂t u − ∆u = f (x) in (0, 1) × (0, 1),
u(0, t) = u(1, t) = 0 on t ∈ (0, 1), (34)
u(x, 0) = 2 sin(πx) for x ∈ (0, 1).


We consider the heat source given with the help of B-splines of order one and two,
which reflects different smoothness of the solution u(·, ·) of the direct problem. We
recall that the inverse problem we study here is to estimate the heat source f by
the algorithm in section (19) given noisy measurements of u(·, 1). We illustrate in
Fig. 1 the exact heat source f as a spline of order one and two.
We use a piecewise linear ansatz as finite elements for the semi-discretization
with respect to the spatial variable with M = 50 uniformly distributed nodes and
the same uniform grid length is used for the time discretization. From classical re-
sults for the finite element method the order of convergence for our approximation
space is O(h2x + h2t ), see [11]. Since we aim to illustrate the stability of the problem,
we consider noisy data with √ stochastic variance σ > 0, which corresponds to the
deterministic noise level σ. Hence, we choose the mesh size M for the discretiza-
1
tion of the problem to be smaller than σ34 , according to Section 4.1. For details of
the FEM for the heat equation see Section 4.1, and further details can be found,
for example, in [11] and [17].

To simulate our data, we assume an equally spaced deterministic design for our
regression problem and we generate the data (Yi )i=1,...,N as

Yi = u(xi , T ) + σǫi

where xi = Ni , i = 1, . . . , N , and ǫi are independent N (0, 1) random variables. To

avoid an “inverse crime” we choose N = 100.
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

12 Taylor & Francis and I.T. Consultant

0.06
data
exact u
0.05 estimator

0.04

Final temperature u(.,T)

0.03

0.02

0.01

−0.01
0 0.2 0.4 0.6 0.8 1
x

Figure 2. The exact data, noisy data and the local polynomial estimator for σ = 0.12

In the first step of our algorithm we estimate u(x, T ) by a local polynomial

2
1
estimator of order one, ψ̂ (N ) , with Gaussian kernel Kh (x) = √2πh 2
exp( −x
2h2 ). Since
our algorithm is data driven, we choose the bandwidth h by cross validation, which
is a consistent adaptive choice of the smoothing parameter h (see e.g. [18]). In this
case we approximate the estimated squared error of ψ (N ) by a simple estimator

N
X
β2 ∼ |Yi − ψ (N ) (xi , h)|2 .
i=1

In practical applications we can also approximate the unknown variance σ by a

consistent estimator (see e.g. [8]), making the method totally data-driven. Fig. 3
shows the noisy data for a sample of size N = 100 and the local polynomial
estimator of the exact data u(x, T ) for the two cases we considered. Since ψ̂ (N ) is
piecewise linear, it belongs to our ansatz space and it coincides with its Galerkin
approximation.

In the next step of our algorithm, we solve the minimization problem for the
Tikhonov functional (30). We have at our disposal the Fréchet derivative of the
operator J, but not the one of the operator A. Hence, we choose the steepest
descent algorithm for the minimization of the Tikhonov functional. Even if this
method is very slow, it converges to the global minimum of our functional. This
minimum approximates well the exact solution f for an a-priori choice of the
regularization parameter α ∼ β, as we noticed from the results in Section 5. Fig. 3

1.2 1.2

1 1

0.8 0.8
Heat source f(x)

Heat source f(x)

0.6 0.6

0.4 0.4

0.2 0.2

0 0

−0.2 −0.2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x

a) Heat source f a spline of order one b) Heat source f a spline of order two

Figure 3. Estimation of the exact heat source f (-) for σ = 0.12 (−⊲), 0.052 (−+), 0.012 (−−)
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

REFERENCES 13

illustrates the accuracy of the reconstruction of the spline of order one, respectively
two, for various amounts of noise. Since in this framework convergence results for
statistical inverse problems in a continuous setting (when the variance σ → 0) are
asymptotically equivalent to similar results in a discrete setting (as the sample size
N → ∞), we consider the variance of the noise σ equal to 0.12 , 0.052 , 0.012 and
we illustrate the results of Lemma 5.1, where the sequence of the regularization
parameter is α ∼ β. We conclude that an accurate reconstruction of the heat source
can be obtained with the proposed method given in Section 5 and, as expected,
the accuracy increases when the variance decreases.

7. Conclusion

We have proposed and investigated a stable numerical method for the problem of
determining a spacewise dependent heat source in the time-dependent heat equa-
tion, where only a finite amount of supplementary temperature measurements are
given at a fixed time point, and these data are blurred by stochastic noise. The pro-
posed method involves minimizing a certain functional using stochastic Tikhonov
regularization, and in this method no a priori information is required on the noise
level, i.e. the method is fully data driven. Numerical investigations were under-
taken based on local polynomials, B-splines and cross-validation techniques. These
numerical results show that accurate reconstructions of the heat source can be
obtained without too many data points needed.

References

[1] Babuška, I., Tempone, R. and Zouraris, G. E., Solving elliptic boundary value problems with
uncertain coefficients by the finite element method: the stochastic formulation, Comput. Meth. Appl.
Mech. Engrg. 194 (2005), 1251–1294.
[2] Bissantz, N., Hohage, T. and Munk, A., Consistency and rates of convergence of nonlinear Tikhonov
regularization with random noise, Inverse Problems 20 (2004), 1773–1789.
[3] Bissantz, N., Hohage, T., Munk, A. and Ruymgaart, F., Convergence rates of general regular-
ization methods for statistical inverse problems and applications, SIAM J. Numer. Anal. 45 (2007),
2610–2636.
[4] Cannon, J. R., Determination of an unknown heat source from overspecified boundary data, SIAM
J. Numer. Anal. 5 (1968), 275–286.
[5] Colton, D., The approximation of solutions to the backwards heat equation in a nonhomogeneous
medium J. Math. Anal. Appl. 72 (1979), 418–429.
[6] Dinh Nho Hào, Johansson, T., Lesnic, D. and Pham Minh Hien, A variational method and ap-
proximations of a Cauchy problem for elliptic equations, submitted to M2An Math. Model. Numer.
Anal.
[7] Goldstein, J. A., Semigroups of Linear Operators and Applications, Oxford Mathematical Mono-
graphs, Oxford, 1985.
[8] Hall, P. and Marron, J. S., On variance estimation in nonparametric regression, Biometrika 77
(1990), 415–419.
[9] Johansson, T. and Lesnic, D., Determination of a spacewise dependent heat source, J. Comput.
Appl. Math. 209 (2007), 66–80.
[10] Johansson, B. T. and Lesnic, D., A variational method for identifying a spacewise dependent heat
source, IMA J. Appl. Math. 72 (2007), 748–760.
[11] Knabner, P. and Angermann, L., Numerical methods for elliptic and parabolic partial differential
equations, Texts in Applied Mathematics, Vol. 44, Springer-Verlag, New York, 2003.
[12] Lions, J.-L. and Magenes, E., Non-homogeneous Boundary Value Problems and Applications, Vol.
I, Springer-Verlag, Berlin, 1972.
[13] Mera, N. S., Elliott, L., Ingham, D. B. and Lesnic, D., An iterative boundary element method
for solving the one-dimensional backward heat conduction problem, Int. J. Heat Mass Transfer 44
(2001), 1937–1946.
[14] Prilepko, A. I. and Solov’ev, V. V., Solvability theorems and Rothe’s method for inverse problems
for a parabolic equation, Differential Equations 23 (1988), 1341–1349.
[15] Rundell, W., Determination of an unknown non-homogeneous term in a linear partial differential
equation from overspecified boundary data, Appl. Anal. 10 (1980), 231–242.
[16] Solov’ev, V. V., Solvability of the inverse problems of finding a source, using overdetermination on
the upper base for a parabolic equation, Differential Equations 25 (1990), 1114–1119.
January 30, 2009 15:50 Inverse Problems in Science and Engineering Joh˙Pri˙source

14 REFERENCES

[17] Thomée, V., Galerkin Finite Element Methods for Parabolic Problems, Second Edition, Springer-
Verlag, Berlin, 2006.
[18] Xia, Y. and Li, W. K., Asymptotic behavior of bandwidth selected by the cross-validation method
for local polynomial fitting, J. Multivariate Anal. 83 (2002), 265–287.