Journal of Computational and Applied Mathematics 309 (2017) 28–43

Contents lists available at ScienceDirect

Journal of Computational and Applied

Mathematics
journal homepage: www.elsevier.com/locate/cam

Determination of a term in the right-hand side of

parabolic equations
Dinh Nho Hào a,∗ , Bui Viet Huong b , Nguyen Thi Ngoc Oanh b ,
Phan Xuan Thanh c
a
Hanoi Institute of Mathematics, VAST, 18 Hoang Quoc Viet Road, Hanoi, Viet Nam
b
Thai Nguyen University, College of Science, Thai Nguyen, Viet Nam
c
School of Applied Mathematics and Informatics, Hanoi University of Science and Technology, 1 Dai Co Viet Road, Hanoi, Viet Nam

article info abstract

Article history: The inverse problem of determining a term in the right hand side of parabolic
Received 6 October 2015 equations from integral observations is investigated. The observations can be regarded
as generalized interior point observations which are collected in practice. The problem is
Keywords: then reformulated as a least squares problem in coupling with a Tikhonov regularization
Inverse source problems term. It is proved that the Tikhonov functional is Fréchet differentiable and a formula
Integral observations
for the gradient is derived via an adjoint problem. The variational problem is discretized
Least squares method
Tikhonov regularization
by the finite element method, the convergence of which is proved. The discretized
Finite element method variational problem is numerically solved by the conjugate gradient method. Some
Conjugate gradient method numerical examples are presented for showing the efficiency of the method.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction

The problem of determining a term in the right hand side of parabolic equations attracted great attention of many
researchers during the last 50 years. Despite a lot of results on the existence, uniqueness and stability estimates of a solution
to the problem, its ill-posedness and possible nonlinearity make it not easy and require further investigations. For surveys
on the subject, we refer the reader to the books [1–5] and the recent paper [6]. To be more detailed, let Ω ∈ Rd be a bounded
domain with the boundary Γ . Denote the cylinder Q := Ω ×(0, T ], where T > 0 and the lateral surface area S = Γ ×(0, T ].
Let
aij , i, j ∈ {1, 2, . . . , d}, b ∈ L∞ (Q ), (1.1)
aij = aji , i, j ∈ {1, 2, . . . , d}, (1.2)
d

λ∥ξ ∥2Rd ≤ aij (x, t )ξi ξj ≤ Λ∥ξ ∥2Rd , ∀ξ ∈ Rd , (1.3)
i,j=1

0 ≤ b(x, t ) ≤ µ1 , a.e. in Q , (1.4)

u0 ∈ L (Ω ),
2
ϕ, ψ ∈ L (S ), 2
(1.5)
λ and Λ are positive constants and µ1 ≥ 0. (1.6)

∗
Corresponding author.
E-mail addresses: hao@math.ac.vn (D.N. Hào), buiviethuong84@gmail.com (B.V. Huong), oanhntn.tn@gmail.com (N.T.N. Oanh),
thanh.phanxuan@hust.vn (P.X. Thanh).

http://dx.doi.org/10.1016/j.cam.2016.05.022
0377-0427/© 2016 Elsevier B.V. All rights reserved.
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 29

Consider the initial value problem

∂u  d
∂ ∂u
 
− aij (x, t ) + b(x, t )u = F (x, t ), ( x, t ) ∈ Q , (1.7)
∂ t i ,j = 1 ∂ x i ∂ xj
u|t =0 = u0 (x), x ∈ Ω, (1.8)
with either the Robin boundary condition
∂u
+ σ u|S = ϕ on S (1.9)
∂N
or the Dirichlet boundary condition
u| S = ψ on S . (1.10)
Here,

∂ u  d
 
:= (aij (x, t )uxj ) cos(ν, xi )|S ,
∂ N S i,j=1
ν is the outer normal to S and σ ∈ L∞ (S ) which is supposed to be nonnegative everywhere in S.
The direct problem is that of determining u when the coefficients of Eq. (1.7) and the data u0 , ϕ (or ψ) and F are given
[2,7,8]. The inverse problem is that of identifying the right hand side F when some additional observations of the solution u
are available. Depending on the structure of F and observations of u we have different inverse problems:
• Inverse Problem (IP) 1: F (x, t ) = f (x, t )h(x, t ) + g (x, t ), h and g are given. Find f (x, t ), if u is given in Q [9,10].
• IP2: F (x, t ) = f (x)h(x, t ) + g (x, t ), h and g are given. Find f (x), if u(x, T ) is given, [11–15]. Related inverse problems for
nonlinear equations have been studied in [16–18].
• IP2a: F (x, t ) = f (x)h(x, t ) + g (x, t ), h and g are given. Find f (x), if Ω ω1 (t )u(x, t )dx is given. Here, ω1 is in L∞ (0, T ) and

T
nonnegative. Furthermore, 0 ω1 (t )dt > 0. Such an observation is called integral observation and it is a generalization of
the final observation in IP2, when ω1 is an approximation to the delta function at t = T . The problem of this setting has
been studied in [19–24,6,25].
• IP3: F (x, t ) = f (t )h(x, t ) + g (x, t ), h and g are given. Find f (t ), if u(x0 , t ) is given. Here, x0 is a point in Ω [26–28].
• IP3a: F (x, t ) = f (t )h(x, t ) + g (x, t ), h and g are given. Find f (t ), if Ω ω2 (x)u(x, t )dx is given. Here, ω2 ∈ L∞ (Ω ) with
ω (x)dx > 0. See, e.g. [29,30,22].
Ω 2
• IP4: F (x, t ) = f (x)h(x, t ) + g (x, t ), h and g are given. Find f (x) if an additional boundary observation of u, for example,
in case of the Dirichlet boundary condition (1.9), we require the Neumann condition be given in a subset of S, [31–39,2,
40–42]. A similar problem for identifying f (t ) with F (x, t ) = f (t )h(x, t ) + g (x, t ) has been studied in [11].
• IP5: Find point sources from an additional boundary observation [43–51]. Related inverse problems have been studied
also in [52].
We note that in IP1, IP2, IP2a to identify f (x, t ) or f (x) the solution u should be available in the whole physical domain Ω that
is hardly realized in practice. To overcome this deficiency, we now approach to the source inverse problem from another
point of view: measure the solution u at some interior (or boundary) points x1 , x2 , . . . , xN ∈ Ω (or on ∂ Ω ) and from these
data determine a term in the right hand side of (1.7). As any measurement is an average process, the following data are
collected:

lk u = ωk (x)u(x, t )dx = hk (t ), hk ∈ L2 (0, T ), k = 1, 2, . . . , N , (1.11)
Ω

with ωk ∈ L∞ (Ω ) and ωk (x)dx > 0, k = 1, 2, . . . , N, being weight functions, N the number of measurements. We note

Ω
that if we take

 1
, if x ∈ Ωk ,
ωk (x) = |Ωk | (1.12)
0, otherwise,

with |Ωk | being the volume of Ωk , a neighbourhood of xk . Then lk u shows the result of the measurement at xk and can be
understood as an average of u(xk , t ) if it exists. If we let |Ωk | tend to zero it will converge to u(xk , t ) if the last exists. However,
since in general the solution u is understood in the weak sense, u(xk , t ) does not always make sense. Thus, the assumption
that lk u are available is meaningful in practice. Further, it is clear that if only lk u are available, the uniqueness will not be
guaranteed except for the case of determining f (t ) in IP3, IP3a (see [26,27]). Hence, to avoid this ambiguity, assume that an
a-priori information f ∗ of f is available which is reasonable in practice. In short, our inverse problem setting is as follows:
Suppose that lk u = hk (t ), k = 1, 2, . . . , N, are available with some noise and an a-priori information f ∗ of f is
available. Identify f .
30 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43

This inverse problem will be investigated by the least squares method [53,54]: minimize the functional
N
1 γ
Jγ (f ) = ∥lk u − hk ∥2L2 (0,T ) + ∥f − f ∗ ∥2∗ (1.13)
2 k=1 2

with γ being a regularization parameter, ∥ · ∥∗ an appropriate norm.

We will prove that this functional is Fréchet differentiable in the next section and will discretize the variational problem
by the finite element method (FEM) and prove its convergence in Section 3. Finally, we will test our method for some concrete
problems in Section 4. We note that the problem of determining f (t ) has been previously studied in [30] by the splitting
method.

2. Variational problem

To introduce the concept of weak solution, we use the standard Sobolev spaces H 1 (Ω ), H01 (Ω ), H 1,0 (Q ) and H 1,1 (Q )
[55,7,8]. Further, for a Banach space B, we define

L2 (0, T ; B) = {u : u(t ) ∈ B a.e. t ∈ (0, T ) and ∥u∥L2 (0,T ;B) < ∞},

with the norm

 T
∥ u∥ 2
L2 (0,T ;B)
= ∥u(t )∥2B dt .
0

In the sequel, we shall use the space W (0, T ) defined as

W (0, T ) = {u : u ∈ L2 (0, T ; H 1 (Ω )), ut ∈ L2 (0, T ; (H 1 (Ω ))′ )},

equipped with the norm

∥u∥2W (0,T ) = ∥u∥2L2 (0,T ;H 1 (Ω )) + ∥ut ∥2L2 (0,T ;(H 1 (Ω ))′ ) .

For simplicity we consider the Robin problem (1.7)–(1.9), only. The case of the Dirichlet (1.7), (1.8) and (1.10) with
the homogeneous boundary condition (1.10) is similar. The solution of the Robin problem (1.7)–(1.8) is understood in
the weak sense as follows: Suppose that F ∈ L2 (Q ), a weak solution in W (0, T ) of the problem (1.7)–(1.9) is a function
u(x, t ) ∈ W (0, T ) satisfying the identity
T d
∂ u ∂η
    
(ut , η)(H 1 (Ω ))′ ,H 1 (Ω ) dt + aij (x, t ) + b(x, t )uη dxdt + σ uηdξ dt
0 Q i,j=1
∂ xi ∂ xj S
 
= F ηdxdt + ϕηdξ dt , ∀η ∈ L2 (0, T ; H 1 (Ω )) (2.1)
Q S

and

u(x, 0) = u0 (x), x ∈ Ω. (2.2)

Following [8, Chapter IV] and [7, p. 141–152] we can prove that there exists a unique solution in W (0, T ) of the problem
(1.7)–(1.9). Furthermore, there is a positive constant cd independent of aij , b, F , σ and u0 such that

∥u∥W (0,T ) ≤ cd ∥F ∥L2 (Q ) + ∥u0 ∥L2 (Ω ) + ∥ϕ∥L2 (S ) .

 
(2.3)

Suppose that F has the form F (x, t ) = f (x, t )h(x, t ) + g (x, t ) with f ∈ L2 (Q ), h ∈ L∞ (Q ) and g ∈ L2 (Q ). We wish to
recover f from the observations (1.11). Since the solution u(x, t ) of (1.7)–(1.9) depends on f (x, t ), we denote it by u(x, t ; f )
or u(f ) to emphasize its dependence on f . To identify f , we minimize the functional

N
 1
J0 ( f ) = ∥lk u(f ) − hk ∥2L2 (0,T ) (2.4)
k=1
2

over L2 (Q ). However, this minimization problem is unstable and there might be many minimizers to it. Therefore, we
minimize the Tikhonov functional instead
N
 1 γ
Jγ (f ) = ∥lk u(f ) − hk ∥2L2 (0,T ) + ∥f − f ∗ ∥2L2 (Q ) (2.5)
k=1
2 2
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 31

with γ > 0 being Tikhonov regularization parameter, f ∗ ∈ L2 (Q ) an a priori estimation of f . It is easily seen that, if γ > 0,
there exists a unique solution to this minimization problem. Next, we prove that Jγ is Fréchet differentiable and derive a
formula for its gradient. In doing so, we introduce the adjoint problem

∂p  ∂ ∂p
 d   N

( , ) ( , ) ωk (x) (lk u − hk ) , (x, t ) ∈ Q ,

 − − a ij x t + b x t p =
 ∂ t i,j=1 ∂ xj ∂ xi



k=1

∂p (2.6)
 + σ (x, t )p = 0, (x, t ) ∈ S ,
∂

 N



p(x, T ) = 0, x ∈ Ω.

As ωk ∈ L (Ω ), lk u − hk ∈ L (0, T ), the right hand side of the first equation in (2.6) belongs to L2 (Q ). By changing the time
2 2

direction, we get a Robin problem for parabolic equations, and it can be seen that there exists a unique solution in W (0, T )
to this problem.

Theorem 2.1. The functional Jγ is Fréchet differentiable and its gradient ∇ Jγ at f has the form

∇ Jγ (f ) = h(x, t )p(x, t ) + γ (f (x, t ) − f ∗ (x, t )), (2.7)

where p(x, t ) is the solution to the adjoint problem (2.6).

Proof. Taking a small variation δ f ∈ L2 (Q ) of f , we have

N N
 1  1
J0 (f + δ f ) − J0 (f ) = ∥lk u(f + δ f ) − hk ∥2L2 (0,T ) − ∥lk u(f ) − hk ∥2L2 (0,T )
k=1
2 k=1
2
N N
 1 
= ∥lk u(f + δ f ) − lk u(f )∥2L2 (0,T ) + ⟨lk u(f + δ f ) − lk u(f ), lk u(f ) − hk ⟩L2 (0,T )
k=1
2 k=1
N N
 1 
= ∥lk δ u(f )∥2L2 (0,T ) + ⟨lk δ u(f ), lk u(f ) − hk ⟩L2 (0,T ) ,
k=1
2 k =1

where δ u(f ) is the solution to the problem

∂δ u  ∂ ∂δ u
 d  
aij (x, t ) + b(x, t )δ u = δ f (x, t )h(x, t ), (x, t ) ∈ Q ,

 −
 ∂t ∂ xj ∂ xi


i,j=1

∂δ u (2.8)
 + σ (x, t )δ u(x, t ) = 0, (x, t ) ∈ S ,
 ∂N




δ u(x, 0) = 0, x ∈ Ω.

Due to the a priori estimate (2.3) for the direct problem, for k = 1, 2, . . . , N, we have

∥lk δ u(f )∥2L2 (0,T ) = o ∥δ f ∥L2 (Q ) when ∥δ f ∥L2 (Q ) → 0.

 

Hence
N

J0 ( f + δ f ) − J0 ( f ) = ⟨lk δ u(f ), lk u(f ) − hk ⟩L2 (0,T ) + o ∥δ f ∥L2 (Q )
 
k=1
N 
 T
 
ωk (x)δ u(x, t )dx (lk u(f ) − hk ) dt + o ∥δ f ∥L2 (Q )
 
=
k=1 0 Ω
N 
 T
 
ωk (x)δ u(x, t )(lk u(f ) − hk )dx dt + o ∥δ f ∥L2 (Q )
 
=
k=1 0 Ω
N 
 T 
δ u ωk (x)(lk u − hk )dxdt + o ∥δ f ∥L2 (Q ) .
 
=
k=1 0 Ω

Applying the Green formula (see, e.g. [7, Theorem 3.18, p. 158]) to (2.6) and (2.8) we obtain
N 
 T   T 
δ u ωk (x)(lk u − hk )dxdt = δ f (x, t )h(x, t )p(x, t ) dxdt .
k=1 0 Ω 0 Ω
32 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43

Hence
 T 
J0 ( f + δ f ) − J0 ( f ) = δ f (x, t )h(x, t )p(x, t ) dxdt + o ∥δ f ∥L2 (Q )
 
0 Ω
= ⟨ph, δ f ⟩L2 (Q ) + o ∥δ f ∥L2 (Q ) .
 
(2.9)

Thus, J0 is Fréchet differentiable and its gradient has the form

∇ J0 (f ) = h(x, t )p(x, t ).
Hence the Tikhonov functional Jγ is also Fréchet differentiable and its gradient is represented by (2.7). The theorem is
proved.

Remark 2.2. In this theorem we write the Tikhonov functional for F (x, t ) = f (x, t )h(x, t ) + g (x, t ). When F has another
structure, the penalty term should be modified.

• F (x, t ) = f (t )h(x, t ) + g (x, t ): the penalty functional is ∥f − f ∗ ∥L2 (0,T ) and


∇ J0 ( f ) = h(x, t )p(x, t )dx.
Ω

• F (x, t ) = f (x)h(x, t ) + g (x, t ): the penalty functional is ∥f − f ∗ ∥L2 (Ω ) and

 T
∇ J0 ( f ) = h(x, t )p(x, t )dt .
0

To find the minimizer of (2.5), we use the conjugate gradient method (CG). It proceeds as follows: Assume that at the kth
iteration we have f k . Then the next iteration is

f k+1 = f k + αk dk ,

with

−∇ Jγ (f k ) if k = 0,

dk =
−∇ Jγ (f k ) + βk dk−1 if k > 0,
∥∇ Jγ (f k )∥2L2 (Q )
βk = ,
∥∇ Jγ (f k−1 )∥2L2 (Q )

and

αk = argminα≥0 Jγ (f k + α dk ).
To evaluate αk we denote by u(u0 , ϕ) (the solution u depends also on g) the solution to the problem

∂u  ∂ ∂u
 d  
aij (x, t ) + b(x, t )u = g (x, t ), (x, t ) ∈ Q ,

 −
 ∂ t i,j=1 ∂ xj ∂ xi



∂u
 + σ (x, t )u(x, t ) = ϕ(x, t ), (x, t ) ∈ S ,
∂

 N



u(x, 0) = u0 (x), x∈Ω

with u[f ] being the solution to the linear problem

∂u  ∂ ∂u
 d  
( , ) + b(x, t )u = f (x, t )h(x, t ), (x, t ) ∈ Q ,

 − a ij x t
 ∂ t i,j=1 ∂ xj ∂ xi



∂u
 + σ (x, t )u(x, t ) = 0, ( x, t ) ∈ S ,
∂

 N



u(x, 0) = 0, x ∈ Ω.

In this case, the observation operators have the form li u(f ) = li u[f ] + li u(u0 , ϕ) = Ai f + li u(u0 , ϕ) with Ai being bounded
linear operators from L2 (Q ) into L2 (0, T ), i = 1, . . . , N.
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 33

We have
N
 1 γ
Jγ (f k + α dk ) = ∥li u(f k + α dk ) − hi ∥2L2 (0,T ) + ∥f k + α dk − f ∗ ∥2L2 (Q )
i=1
2 2
N
 1 γ
= ∥α Ai dk + Ai f k + li u(u0 , ϕ) − hi ∥2L2 (0,T ) + ∥α dk + f k − f ∗ ∥2L2 (Q )
i=1
2 2
N
 1 γ
= ∥α Ai dk + li u(f k ) − hi ∥2L2 (0,T ) + ∥α dk + f k − f ∗ ∥2L2 (Q ) .
i=1
2 2

Differentiating Jγ (f k + α dk ) with respect to α , we get

∂ J (f k + α dk )  N N
=α ∥Ai dk ∥2L2 (0,T ) + Ai dk , li u(f k ) − hi L2 (0,T ) + γ α∥dk ∥2L2 (Q ) + γ dk , f k − f ∗ L2 (Q ) .
   
∂α i =1 i=1

∂ Jγ (f k +α dk )
Putting ∂α
= 0, we obtain
N 
Ai dk , li u(f k ) − hi + γ dk , f k − f ∗ L2 (Q )
   
L2 (0,T )
i =1
αk = − N
∥Ai dk ∥2L2 (0,T ) + γ ∥dk ∥2L2 (Q )

i=1
N 
dk , (Ai )∗ li u(f k ) − hi + γ dk , f k − f ∗ L2 (Q )
    
L2 (Q )
i =1
=− N
∥Ai dk ∥2L2 (0,T ) + γ ∥dk ∥2L2 (Q )

i=1
N 
dk , (Ai )∗ li u(f k ) − hi + γ (f k − f ∗ )
   
L2 (Q )
i =1
=− N
∥Ai dk ∥2L2 (0,T ) + γ ∥dk ∥2L2 (Q )

i=1

dk , ∇ Jγ (f k ) L2 (Q )
 
=− N .
∥Ai dk ∥2L2 (0,T ) + γ ∥dk ∥2L2 (Q )

i =1

Since dk = −∇ Jγ (f k ) + βk dk−1 , rk = −∇ Jγ (f k ) and ⟨rk , dk−1 ⟩L2 (Q ) = 0, we have

∥rk ∥2L2 (Q )
αk = N
.
Ai dk 2L2 (0,T ) + γ∥ ∥ dk 2L2 (Q )

∥ ∥
i =1

Thus, the CG has the form

Step 1: Set k = 0, initiate f 0 .
Step 2: Calculate r0 = −∇ Jγ (f 0 ) and set d0 = r0 .
Step 3: Evaluate
∥r0 ∥2L2 (Q )
α0 = N
.
Ai d0 2L2 (0,T ) + γ∥ ∥ d0 2L2 (Q )

∥ ∥
i=1

Set f 1 = f 0 + α0 d0 .
Step 4: For k = 1, 2, . . . . Calculate
rk = −∇ Jγ (f k ), dk = rk + βk dk−1
with
∥rk ∥2L2 (Q )
βk = .
∥rk−1 ∥2L2 (Q )
34 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43

Step 5: Calculate
∥rk ∥2L2 (Q )
αk = N
.
Ai dk 2L2 (0,T ) + γ∥ ∥ dk 2L2 (Q )

∥ ∥
i=1

Update
f k+1 = f k + αk dk .

3. Finite element method

We first rewrite the observation operators in the form lk u(f ) = lk u[f ] + lk u(u0 , ϕ) = Ak f + lk u(u0 , ϕ), where
Ak : L2 (Q ) → L2 (0, T ) are the bounded linear operators, k = 1, . . . , N. Then the functional has the following form
N
 1 γ
Jγ (f ) = ∥lk u[f ] + lk u(u0 , ϕ) − hk ∥2L2 (0,T ) + ∥f − f ∗ ∥2L2 (Q )
k=1
2 2
N
 1 γ
= ∥Ak f + lk u(u0 , ϕ) − hk ∥2L2 (0,T ) + ∥f − f ∗ ∥2L2 (Q )
k=1
2 2
N
 1 γ
= hk ∥2L2 (0,T ) +
∥A k f −  ∥f − f ∗ ∥2L2 (Q ) .
k=1
2 2

The solution f γ of the minimization problem (2.5) is characterized by the first-order optimality condition
N
A∗k (Ak f γ − 
hk ) + γ (f γ − f ∗ ) = 0.

(3.1)
k=1

Here A∗k : L2 (0, T ) → L2 (Q ) is the adjoint operator of Ak defined by A∗k q = pk , where pk is the solution of the adjoint problem

∂ pk  ∂ ∂ pk
 d  
aij (x, t ) + b(x, t )pk = ωk (x)q(t ), (x, t ) ∈ Q

 − −
 ∂t ∂ xj ∂ xi


i,j=1

∂ pk (3.2)
 + σ (x, t )pk = 0, ( x, t ) ∈ S
 ∂N




pk (x, T ) = 0, x ∈ Ω.
We note that here we split the adjoint problem (2.6) into N independent problems (3.2). By linear superposition, the adjoint
N
state p is k=1 pk . We will approximate (3.1) by the finite element method (FEM). In fact, we will approximate Ak and A∗k as
follows.

3.1. Finite element approximation of Ak , A∗k , k = 1, . . . , N

We present now a fully discrete FE approximation for the above variational problem. Supposing that Ω is a polyhedral
domain, we triangulate Ω into a shape regular quasi-uniform mesh Th and define the piecewise linear finite element space
Vh ⊂ H 1 (Ω ) by

Vh = {vh : vh ∈ C (Ω ), vh |K ∈ P1 (K ), ∀K ∈ Th }. (3.3)
Here, P1 (K ) is the space of linear polynomials on the element K . For full discretization we introduce a uniform partition of
the integral [0, T ]:
0 = t0 < t1 < · · · < tM , where tn = n1t, n = 0, 1, . . . , M with the time step size 1t = T /M .
Let
d
∂v ∂w
   
an (v, w) := anij (x) dx + bn (x)v(x)w(x)dx + σ n (ξ )v(ξ )w(ξ )dξ ,
Ω i ,j = 1 ∂ xj ∂ xi Ω Γ

for v, w ∈ H 1 (Ω ) and for a function φ(x, t ), we define φ n (x) := φ(x, tn ). Then an (·, ·) : H 1 (Ω )× H 1 (Ω ) → R is the bounded
bilinear form and H 1 (Ω )-elliptic, i.e.,
an (v, v) ≥ C1a ∥v∥2H 1 (Ω ) ∀v ∈ H 1 (Ω ).
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 35

We now define the fully discrete FE approximation for the variational problem (2.1) by the backward Euler–Galerkin method
as follows: Find unh ∈ Vh for n = 1, 2, . . . , M such that

dt unh , χ + an (unh , χ ) = F n , χ L2 (Ω ) + ϕ n , χ L2 (Γ ) ,
     
L2 (Ω )
∀χ ∈ Vh (3.4)

and

⟨u0h , χ⟩L2 (Ω ) = ⟨u0 , χ⟩L2 (Ω ) , ∀χ ∈ Vh , (3.5)

n−1
un −u
where dt unh := h 1th , n = 1, 2, . . . , M.
The discrete variational problem (3.4) admits a unique solution unh ∈ Vh . Let uh (x, t ) be the linear interpolation of unh with
respect to t. Hence the discrete version of the optimal control problem (1.13) reads

N
 1 γ
J γ ,h ( f ) = hk,h ∥2L2 (0,T ) +
∥Ak,h f −  ∥f − f ∗ ∥2L2 (Q ) → min . (3.6)
k=1
2 2

Here the computational observations lk uh (f ) = lk uh [f ] + lk uh (u0 , ϕ) = Ak,h f + lk uh (u0 , ϕ) and 

hk,h = lk uh (u0 , ϕ) − hk . The
solution of the optimal problem (3.6) is characterized by the variational equation

N

A∗k,h (Ak,h f − 
hk,h ) + γ (f − f ∗ ) = 0, (3.7)
k=1

where A∗k,h is the adjoint operator of the linear operator Ak,h , k = 1, . . . , N.

A∗k,h q = pk,h of A∗k q. Moreover, if instead of the observations
For the FE approximation of (3.2) we define an approximation 
δ
hk we get only the perturbations hkk satisfying

δ
∥hkk − hk ∥L2 (0,T ) ≤ δk for k = 1, . . . , N . (3.8)

Therefore we arrive at the variational problem

N
γ δ γ

A∗k,h (Ak,h fh − 
 hk,kh ) + γ (fh − f ∗ ) = 0, (3.9)
k =1

δ δ
hk,kh = lk uh (u0 , ϕ) − hkk , k = 1, . . . , N.
where 

3.2. Convergence results

Let
d
∂ u ∂v
   
a(u, v) := aij (x, t ) dx + b(x, t )u(x, t )v(x)dx + σ (ξ , t )u(ξ , t )v(ξ )dξ ,
Ω i ,j = 1 ∂ xj ∂ xi Ω Γ

for u ∈ W (0, T ), v ∈ H 1 (Ω ).
We define u(x, t ) ∈ W (0, T ) the weak solution of (1.7)–(1.9) satisfying the variational formulation

⟨ut , v⟩L2 (Ω ) + a(u, v) = ⟨F , v⟩L2 (Ω ) + ⟨ϕ, v⟩L2 (Γ ) , ∀v ∈ H 1 (Ω ), t ∈ (0, T ), (3.10)

and

u(x, 0) = u0 (x), x ∈ Ω. (3.11)

For φ ∈ H (Ω ) we define the elliptic projection Rh : H (Ω ) → Vh as the unique solution of the variational problem
1 1

a(Rh φ, vh ) = a(φ, vh ) ∀vh ∈ Vh . (3.12)

There holds the error estimate, see [56],

∥φ − Rh φ∥L2 (Ω ) ≤ C h2 ∥φ∥H 2 (Ω ) ∀φ ∈ H 2 (Ω ). (3.13)

36 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43

Lemma 3.1. Let u be the unique solution of variational problem (3.10)–(3.11) and unh ∈ Vh for n = 1, 2, . . . , M be the solution
of (3.4)–(3.5). Then there holds the error estimate

|||uh − Rh u|||ℓ2 (0,T ;H 1 (Ω )) ≤ C h2 ∥ut ∥L2 (0,T ;H 2 (Ω )) + 1t ∥utt ∥L2 (0,T ;L2 (Ω )) + h2 ∥u0 ∥H 2 (Ω ) ,
 
(3.14)

where
 1/2
M

|||w|||ℓ2 (0,T ;H 1 (Ω )) := 1t n 2
∥w ∥ H 1 (Ω )
.
n =1

Proof. The idea throughout the error analysis is to write the error as a sum of two terms, see [56],

uh − u = (uh − Rh u) + (Rh u − u) =: θ + ρ.

Subtracting the variational formulation (3.10) for v = χ ∈ Vh ⊂ H 1 (Ω ), t = tn from the Galerkin variational problem (3.4)
gives

⟨unt − dt unh , χ⟩L2 (Ω ) + an (un − unh , χ ) = 0 ∀χ ∈ Vh ,

or

⟨unt − dt unh , χ⟩L2 (Ω ) + an (Rh un − unh , χ ) = 0 ∀χ ∈ Vh . (3.15)

By definition we note that

⟨dt θ n , χ⟩L2 (Ω ) + an (θ n , χ ) = ⟨dt unh − dt Rh un , χ⟩L2 (Ω ) + an (unh − Rh un , χ )

= ⟨unt − dt Rh un , χ⟩L2 (Ω )
= ⟨unt − dt un , χ⟩L2 (Ω ) + ⟨dt un − dt Rh un , χ⟩L2 (Ω )
= ⟨unt − dt un , χ⟩L2 (Ω ) − ⟨dt ρ n , χ⟩L2 (Ω ) ∀χ ∈ Vh .
Taking χ = θ n ∈ Vh we find that

⟨dt θ n , θ n ⟩L2 (Ω ) + an (θ n , θ n ) = ⟨unt − dt un , θ n ⟩L2 (Ω ) − ⟨dt ρ n , θ n ⟩L2 (Ω ) . (3.16)

For the right hand side of (3.16) we have

⟨unt − dt un , θ n ⟩L2 (Ω ) = (1t )−1 ⟨1t unt − (un − un−1 ), θ n ⟩L2 (Ω )

≤ (1t )−1 ∥1t unt − (un − un−1 )∥L2 (Ω ) ∥θ n ∥L2 (Ω )
 
 tn 
= (1t ) 
−1 
(τ − tn−1 )utt (·, τ )dτ  ∥θ n ∥L2 (Ω )

 tn−1 2
L (Ω )
 
tn tn
≤ (1t )−1 (τ − tn−1 )2 dτ ∥utt (·, τ )∥2L2 (Ω ) dτ ∥θ n ∥L2 (Ω )
tn−1 tn−1

1
= √ (1t )1/2 ∥utt ∥L2 (tn−1 ,tn ;L2 (Ω )) ∥θ n ∥L2 (Ω ) ,
3
and

−⟨dt ρ n , θ n ⟩L2 (Ω ) = ⟨dt un − Rh dt un , θ n ⟩L2 (Ω )

≤ ∥dt un − Rh dt un ∥L2 (Ω ) ∥θ n ∥L2 (Ω )
≤ C h2 ∥dt un ∥H 2 (Ω ) ∥θ n ∥L2 (Ω )
= C h2 (1t )−1 ∥u(·, tn ) − u(·, tn−1 )∥H 2 (Ω ) ∥θ n ∥L2 (Ω )
 
 tn 
= C h (1t ) 
2 −1 
ut (·, τ )dτ  ∥θ n ∥L2 (Ω )

 tn−1  2
H (Ω )
 tn
≤ C h2 (1t )−1 ∥ut (·, τ )∥H 2 (Ω ) dτ ∥θ n ∥L2 (Ω )
tn−1
−1/2
≤ C h (1t )
2
∥ut ∥L2 (tn−1 ,tn ;H 2 (Ω )) ∥θ n ∥L2 (Ω ) .
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 37

The left hand side of (3.16) can be estimated as follows

1  
∥θ n ∥2L2 (Ω ) − ⟨θ n−1 , θ n ⟩L2 (Ω ) + an (θ n , θ n )
⟨dt θ n , θ n ⟩L2 (Ω ) + an (θ n , θ n ) =
1t
1  n 2 
≥ ∥θ ∥L2 (Ω ) − ∥θ n−1 ∥2L2 (Ω ) + C1a ∥θ n ∥2H 1 (Ω ) .
21t
To sum up we can conclude that for n = 1, 2, . . . , M,

∥θ n ∥2L2 (Ω ) − ∥θ n−1 ∥2L2 (Ω ) + 2C1a 1t ∥θ n ∥2H 1 (Ω )

2
≤ √ (1t )3/2 ∥utt ∥L2 (tn−1 ,tn ;L2 (Ω )) ∥θ n ∥L2 (Ω ) + 2C h2 (1t )1/2 ∥ut ∥L2 (tn−1 ,tn ;H 2 (Ω )) ∥θ n ∥L2 (Ω ) .
3
Then by Young’s inequality this implies
M

∥θ M ∥2L2 (Ω ) − ∥θ 0 ∥2L2 (Ω ) + 2C1a 1t ∥θ n ∥2H 1 (Ω )
n=1
M M
2 
(1t )3/2 ∥utt ∥L2 (tn−1 ,tn ;L2 (Ω )) ∥θ n ∥L2 (Ω ) + 2C h2 (1t )1/2 ∥ut ∥L2 (tn−1 ,tn ;H 2 (Ω )) ∥θ n ∥L2 (Ω )

≤ √
3 n =1 n =1
M
 M

≤ C (δ)(1t )2 ∥utt ∥2L2 (t +δ 1t ∥θ n ∥2L2 (Ω )
n−1 ,tn ;L (Ω ))
2
n=1 n=1
M
 M

+ C (δ)h4 ∥ut ∥2L2 (t +δ 1t ∥θ n ∥2L2 (Ω ) .
n−1 ,tn ;H
2 (Ω ))
n =1 n=1

Hence
M

1t ∥θ n ∥2H 1 (Ω ) ≤ C (1t )2 ∥utt ∥2L2 (0,T ;L2 (Ω )) + C h4 ∥ut ∥2L2 (0,T ;H 2 (Ω )) + ∥θ 0 ∥2L2 (Ω )
n =1

≤ C (1t )2 ∥utt ∥2L2 (0,T ;L2 (Ω )) + C h4 ∥ut ∥2L2 (0,T ;H 2 (Ω )) + Ch4 ∥u0 ∥2H 2 (Ω ) .
The assertion follows. 

Lemma 3.2. Let uh (x, t ) and (Rh u)(x, t ) be the linear interpolations of unh and Rh un with respect to t, respectively. Then there
holds the error estimate

∥uh − Rh u∥L2 (0,T ;H 1 (Ω )) = O (h2 + 1t ). (3.17)

Proof. By definition we have, for x ∈ Ω , t ∈ (tn−1 , tn ), n = 1, 2, . . . , M

t − tn−1 tn − t n−1
uh (x, t ) = unh (x) + u (x),
1t 1t h
t − tn−1 tn − t
(Rh u)(x, t ) = Rh un (x) + Rh un−1 (x).
1t 1t
This implies
tn  t − tn−1 n tn − t n−1 2
tn
  
∥uh − Rh u∥2H 1 (Ω ) dt = 
 1t θ + θ  1 dt , θ := uh − Rh u ,
 n n n

tn−1 tn−1 1t H (Ω )
 tn  2  tn  2
t − tn−1 tn − t
≤2 n 2
dt ∥θ ∥H 1 (Ω ) + 2 dt ∥θ n−1 ∥2H 1 (Ω ) ,
tn−1 1t tn−1 1t
2 2
= 1t ∥θ n ∥2H 1 (Ω ) + 1t ∥θ n−1 ∥2H 1 (Ω ) .
3 3
Hence
 T M
4 2
∥uh − Rh u∥2H 1 (Ω ) dt ≤ 1t ∥θ n ∥2H 1 (Ω ) + 1t ∥θ 0 ∥2H 1 (Ω ) = O (h4 + (1t )2 ).
0 3 n =1 3

The assertion follows. 

38 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43

In addition, by a standard approximation we have

∥Rh u − u∥L2 (Q ) = O (h2 + (1t )2 ). (3.18)

Hence by using the triangle inequality, we conclude
∥uh − u∥L2 (Q ) = O (h2 + 1t ).
Then, we can estimate the computational observations as follows
 T  T
 2
∥lk uh (f ) − lk u(f )∥2L2 (0,T ) = [lk uh (f ) − lk u(f )]2 dt = ωk (x)[uh (x, t ) − u(x, t )]dx dt
0 0 Ω
T
   
≤ ωk2 (x)dx [uh (x, t ) − u(x, t )]2 dx dt
0 Ω Ω

= ∥ω ∥ 2
k L2 (Ω ) ∥uh − u∥2L2 (Q ) ,
or
∥lk uh (f ) − lk u(f )∥L2 (0,T ) ≤ ∥ωk ∥L2 (Ω ) ∥uh − u∥L2 (Q ) ≤ C (h2 + 1t ).
Therefore we can conclude the convergence results

∥(Ak,h − Ak )f ∥L2 (0,T ) = O (h2 + 1t ) and ∥(

A∗k,h − A∗k )q∥L2 (Q ) = O (h2 + 1t ), (3.19)

for all f ∈ L (Q ), q ∈ L (0, T ). By the same technique as in the proof of [57] we can prove for γ > 0 that
2 2


γ
∥fh − f γ ∥L2 (Q ) = O (h2 + 1t + δ), δ= δ12 + δ22 + · · · + δN2 . (3.20)

4. Numerical examples

In all examples in this section we choose the domain Ω = (0, 1) × (0, 1), T = 1 and
aij (x, t ) = δij , b(x, t ) = 1, σ (x, t ) = 1.
For the temperature we take the exact solution be given by
u(x, t ) = et (x1 − x21 ) sin π x2 .
We shall test several structure of F , i.e.,
• Example 1: F (x, t ) = f (t )h(x, t ) + g (x, t ),
• Example 2: F (x, t ) = f (x)h(x, t ) + g (x, t ),
• Example 3: F (x, t ) = f (x, t ) + g (x, t ),
with integral observations (1.11) or point observations.
For the backward Euler–Galerkin method we use a uniform decomposition of the domain Ω into 4096 finite elements
and the time step size τ = T /M = 1/M with M = 64.
For the first example we will test for one observation N = 1: integral observation with ω(x) = x21 + x22 + 1, or point
observation at x0 = (0.48; 0.48). We take h(x, t ) = x1 x2 + t + 1, initial guess f ∗ = 0, γ = 10−5 . Then prescribing f we can
take the given data g (x, t ).
We would like reconstruct the following functions

if 0 ≤ t ≤ 0.5,

2t
f (t ) = for Example 1.1 (4.1)
2(1 − t ) if 0.5 ≤ t ≤ 1,
if 0.25 ≤ t ≤ 0.75,

1
f (t ) = for Example 1.2 (4.2)
0 otherwise,

from the perturbation observations of several noise level δ = 1%, 3%, 5%. (See Figs. 1 and 2.)
In the second example we would like to reconstruct the following functions
f (x) = x31 + x22 for Example 2.1 (4.3)
1 for x = (0.5; 0.5),

f (x) = 0 for x ∈ {(0; 0), (0; 1), (1; 1), (1; 0)}, for Example 2.2, (4.4)
linear otherwise,
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 39

Fig. 1. The exact and numerical solutions of Example 1.1: integral observation (left) and point observation (right).

Fig. 2. The exact and numerical solutions of Example 1.2: integral observation (left) and point observation (right).

from measurements at N = 9 observation points as presented in Fig. 3, where

h(x, t ) = t 2 + 2, γ = 10−5 , δ = 1%.

In the third example we would like to reconstruct the following function

f (x, t ) = (x31 + x32 )(t 2 + 1), Example 3.1, (4.5)

from measurements at 9 points as above. (See Figs. 4–9.)

5. Conclusions

In this paper we propose a new approach to the problem of identifying a term in the right hand side of parabolic equations.
The sought term can solely depend only on time variable, or solely on space variable, or simultaneously on both space and
time variables. The observations are of integral form which can be seen as generalization of pointwise observations taken
inside the space domain. These kinds of observations have the direct physical meaning that we need to observe the process
at some points in the space domain to identify to source rather than to observe it in the whole space domain that can hardly
realized in practice. We then reformulate the problem in a least squares form and prove the functional to be minimized
is Fréchet differentiable as well as derive a formula for its gradient. The problem is then discretized by the finite element
method and solved by the conjugate gradient method. Numerical results show that the approach is promising.
40 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43

Fig. 3. Observation points.

Fig. 4. The exact f (x) and the numerical solutions of Example 2.1 for noise level 1%.

Fig. 5. The exacts f (0.5, x2 ), f (x1 , 0.5) and the numerical solutions of Example 2.1 for noise level 1%.
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 41

Fig. 6. The exact f (x) and the numerical solutions of Example 2.2 for noise level 1%.

Fig. 7. The exacts f (0.5, x2 ), f (x1 , 0.5) and the numerical solutions of Example 2.2 for noise level 1%.

Fig. 8. The exact f (x0 , t ), x0 = (0.5; 0.4) and the numerical solutions of Example 3.1 for noise level 0.1%.
42 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43

Fig. 9. The exact f (x, 0.5) and the numerical solutions of Example 3.1 for noise level 0.1%.

Acknowledgement

This research by Phan Xuan Thanh was supported by Vietnam National Foundation for Science and Technology
Development (NAFOSTED) under grant number 101.01-2014.36.

References

[1] J.R. Cannon, The one-dimensional heat equation, in: Advanced Book Program, Addison-Wesley Publishing Company, Reading, MA, 1984.
[2] D.N. Hào, Methods for Inverse Heat Conduction Problems, Peter Lang Verlag, Frankfurt/Main, Bern, New York, Paris, 1998.
[3] V. Isakov, Inverse Source Problems, Amer. Math. Soc., Providence, RI, 1990.
[4] V. Isakov, Inverse Problems for Partial Differential Equations, Second ed., Springer, New York, 2006.
[5] A.I. Prilepko, D.G. Orlovsky, I.A. Vasin, Methods for Solving Inverse Problems in Mathematical Physics, Marcel Dekker, Inc., New York, 2000.
[6] A.I. Prilepko, D.S. Tkachenko, Inverse problem for a parabolic equation with integral overdetermination, J. Inverse Ill-Posed Probl. 11 (2003) 191–218.
[7] F. Tröltzsh, Optimal Control of Partial Differential Equations: Theory, Methods and Applications, Amer. Math. Soc., Providence, Rhode Island, 2010.
[8] J. Wloka, Partial Differential Equations, Cambridge Univ. Press, Cambridge, 1987.
[9] M.M. Lavrent’ev, V.I. Maksimov, On the reconstruction of the right-hand side of a parabolic equation, Comput. Math. Math. Phys. 48 (2008) 641–647.
[10] P.N. Vabishchevich, Numerical solution of the problem of the identification of the right-hand side of a parabolic equation, Russian Math. (Iz. VUZ) 47
(1) (2003) 27–35.
[11] A. Hasanov, B. Pektaş, A unified approach to identifying an unknown spacewise dependent source in a variable coefficient parabolic equation from
final and integral overdeterminations, Appl. Numer. Math. 78 (2014) 49–67.
[12] A.D. Iskenderov, Some inverse problems on determining the right-hand sides of differential equations, Izv. Akad. Nauk Azerbaijan. SSR Ser. Fiz.-Tehn.
Mat. Nauk (2) (1976) 58–63 (in Russian).
[13] A.D. Iskenderov, R.G. Tagiev, An inverse problem on determinating the right hand side of evolution equations in Banach spaces, in: Questions of
Applied Mathematics and Cybernetics, 1979, pp. 51–56. A collection of Scientific Papers, Azerbaijan State University, No. 1.
[14] V.L. Kamynin, On the unique solvability of an inverse problem for parabolic equations with a final overdetermination condition, Math. Notes 73 (2003)
202–211.
[15] W. Rundell, Determination of an unknown nonhomogeneous term in a linear partial differential equation from overspecified boundary data, Appl.
Anal. 10 (3) (1980) 231–242.
[16] N.L. Gol’dman, Inverse problem with final observation for quasilinear parabolic equations with unknown right hand side, Vychisl. Metody Program.
4 (2003) 155–166.
[17] N.L. Gol’dman, Determination of the right-hand side in a quasilinear parabolic equation with final observation, Differ. Equ. 41 (2005) 384–392.
[18] N.L. Gol’dman, Finding the right-hand side in multidimensional parabolic equations with terminal observation, Differ. Equ. 43 (2007) 1101–1110.
[19] A. Erdem, D. Lesnic, A. Hasanov, Identification of a spacewise dependent heat source, Appl. Math. Model. 37 (2013) 10231–10244.
[20] V.L. Kamynin, On an inverse problem of determining the right-hand side of a parabolic equation with the integral overdetermination condition, Math.
Notes 77 (2005) 482–493.
[21] D.G. Orlovskii, Determination of parameter evolution in an abstract quasilinear parabolic equation, Mat. Zametki 50 (2) (1991) 111–119 (in Russian).
[22] D.G. Orlovskii, Solvability of an inverse problem for a parabolic equation in the Hölder class, Mat. Zametki 50 (3) (1991) 107–112 (in Russian).
[23] A.I. Prilepko, D.S. Tkachenko, Properties of solutions of a parabolic equation and the uniqueness of the solution of the inverse source problem with
integral overdetermination, Comput. Math. Math. Phys. 43 (2003) 537–546.
[24] A.I. Prilepko, D.S. Tkachenko, The Fredholm property and the well-posedness of the inverse source problem with integral overdetermination, Comput.
Math. Math. Phys. 43 (2003) 1338–1347.
[25] D.S. Tkachenko, On an inverse problem for a parabolic equation, Math. Notes 75 (2004) 676–689.
[26] V.T. Borukhov, P.N. Vabishchevich, Numerical solution of an inverse problem of source reconstructions in a parabolic equation, Math. Model. 10 (11)
(1998) 93–100 (in Russian).
[27] A.I. Prilepko, V.V. Solov’ev, Solvability theorems and the Rothe method in inverse problems for an equation of parabolic type. I, Differ. Uravn. 23 (1987)
1791–1799 (in Russian).
[28] A.I. Prilepko, V.V. Solov’ev, Solvability theorems and the Rothe method in inverse problems for an equation of parabolic type. II, Differ. Uravn. 23
(1987) 1971–1980 (in Russian).
[29] Yu.A. Kriksin, S.N. Plyushchev, E.A. Samarskaya, V.F. Tishkin, The inverse problem of source reconstruction for a convective diffusion equation, Math.
Model. 7 (11) (1995) 95–108 (in Russian).
[30] N.T.N. Oanh, B.V. Huong, Determination of a time-dependent term in the right hand side of linear parabolic equations, Acta Math. Vietnam. 41 (2016)
313–335.
 D.N. Hào et al. / Journal of Computational and Applied Mathematics 309 (2017) 28–43 43

[31] J.R. Cannon, Determination of an unknown heat source from overspecified boundary data, SIAM J. Numer. Anal. 5 (1968) 275–286.
[32] J.R. Cannon, P. DuChateau, Structural identification of an unknown source term in a heat equation, Inverse Problems 14 (1998) 535–551.
[33] J.R. Cannon, R.E. Ewing, Determination of a source term in a linear parabolic partial differential equation, Z. Angew. Math. Phys. 27 (1976) 393–401.
[34] J.R. Cannon, Y.-P. Lin, Determination of a source term in a linear parabolic differential equation with mixed boundary conditions, in: Inverse Problems
(Oberwolfach, 1986), in: Internat. Schriftenreihe Numer. Math., vol. 77, Birkhäuser, Basel, 1986, pp. 31–49.
[35] M. Choulli, M. Yamamoto, Conditional stability in determining a heat source, J. Inverse Ill-Posed Probl. 12 (2004) 233–243.
[36] M. Choulli, M. Yamamoto, Some stability estimates in determining sources and coefficients, J. Inverse Ill-Posed Probl. 14 (2006) 355–373.
[37] H.W. Engl, O. Scherzer, M. Yamamoto, Uniqueness and stable determination of forcing terms in linear partial differential equations with overspecified
boundary data, Inverse Problems 10 (1994) 1253–1276.
[38] D.N. Hào, A Noncharacteristic Cauchy Problem for Linear Parabolic Equations and Related Inverse Problems II: A Variational Method, in: Pitman Res.
Notes in Maths, vol. 263, 1992, pp. 43–56.
[39] D.N. Hào, A noncharacteristic Cauchy problem for linear parabolic equations and related inverse problems I: solvability, Inverse Problems 10 (1994)
295–315.
[40] D.D. Trong, A. Pham Ngoc Dinh, P.T. Nam, Determine the special term of a two-dimensional heat source, Appl. Anal. 88 (2009) 457–474.
[41] M. Yamamoto, Conditional stability in determination of force terms of heat equations in a rectangle, Math. Comput. Modelling 18 (1993) 79–88.
[42] M. Yamamoto, Conditional stability in determination of densities of heat sources in a bounded domain, in: Control and Estimation of Distributed
Parameter Systems: Nonlinear Phenomena (Vorau, 1993), in: Internat. Ser. Numer. Math., vol. 118, Birkhäuser, Basel, 1994, pp. 359–370.
[43] M. Andrle, F. Ben Belgacem, A. El Badia, Identification of moving pointwise sources in an advection–dispersion-reaction equation, Inverse Problems
27 (2011) 025007.
[44] M. Andrle, A. El Badia, On an inverse source problem for the heat equation. Application to a pollution detection problem II, Inverse Probl. Sci. Eng. 23
(2015) 389–412.
[45] A. El Badia, T. Ha-Duong, On an inverse source problem for the heat equation. Application to a pollution detection problem, J. Inverse Ill-Posed Probl.
10 (2002) 585–599.
[46] A. El Badia, T. Ha-Duong, A. Hamdi, Identification of a point source in a linear advection–dispersion-reaction equation: application to a pollution source
problem, Inverse Problems 21 (2005) 1121–1136.
[47] A. El Badia, A. Hamdi, Inverse source problem in an advection–dispersion-reaction system: application to water pollution, Inverse Problems 23 (2007)
2103–2120.
[48] A. Hamdi, Identification of point sources in two-dimensional advection-diffusion-reaction equation: application to pollution sources in a river.
Stationary case, Inverse Probl. Sci. Eng. 15 (2007) 855–870.
[49] A. Hamdi, Identification of a time-varying point source in a system of two coupled linear diffusion-advection-reaction equations: application to surface
water pollution, Inverse Problems 25 (2009) 115009.
[50] L.V. Ling, T. Takeuchi, Point sources identification problems for heat equations, Commun. Comput. Phys. 5 (2009) 897–913.
[51] L.V. Ling, M. Yamamoto, Y.C. Hon, T. Takeuchi, Identification of source locations in two-dimensional heat equations, Inverse Problems 22 (2006)
1289–1305.
[52] F. Hettlich, W. Rundell, Identification of a discontinuous source in the heat equation, Inverse Problems 17 (2001) 1465–1482.
[53] D.N. Hào, A noncharacteristic Cauchy problem for linear parabolic equations II: A variational method, Numer. Funct. Anal. Optim. 13 (1992) 541–564.
[54] D.N. Hào, A noncharacteristic Cauchy problem for linear parabolic equations III: A variational method and its approximation schemes, Numer. Funct.
Anal. Optim. 13 (1992) 565–583.
[55] O.A. Ladyzhenskaya, The Boundary Value Problems of Mathematical Physics, Springer-Verlag, New York, 1985.
[56] V. Thomée, Galerkin Finite Element Methods for Parabolic Problems, Second ed., Springer-Verlag, Berlin, 2006.
[57] D.N. Hào, P.X. Thanh, D. Lesnic, B.T. Johansson, A boundary element method for a multi-dimensional inverse heat conduction problem, Int. J. Comput.
Math. 89 (2012) 1540–1554.