See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/237150243

An iterative method for solving a complex heat transfer problem

Article  in  Applied Mathematics and Computation · May 2013

DOI: 10.1016/j.amc.2013.03.091

CITATIONS READS

16 867

2 authors:

Andrey Kovtanyuk Alexander Chebotarev

Russian Academy of Sciences Far Eastern Federal University
58 PUBLICATIONS   295 CITATIONS    65 PUBLICATIONS   304 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cerebral Hemodynamics View project

All content following this page was uploaded by Andrey Kovtanyuk on 09 November 2017.

The user has requested enhancement of the downloaded file.

An iterative method for solving a complex heat transfer
problem

Andrey E. Kovtanyuka,b,∗, Alexander Yu. Chebotareva,b

a Far Eastern Federal University, Sukhanova st. 8, 690950, Vladivostok, Russia
b Institute of Applied Mathematics FEB RAS, Radio st. 7, 690041, Vladivostok, Russia

Abstract
Conductive-radiative heat transfer in a scattering and absorbing medium bounded
by two reflecting and radiating plane surfaces is considered. A diffusion model
for describing heat transfer is examined. The theorems of existence and unique-
ness of a solution of this problem are proved. An iterative procedure for solving
radiative-conductive heat transfer problem is proposed.
Keywords: radiative heat transfer, conductive heat transfer, diffusion
approximation

1. Introduction

The study of the radiative-conductive heat transfer [1, 2] is important for

many engineering applications [3–7]. Usually this process is described by a non-
linear system of two differential equations: an equation of the radiative heat
transfer and an equation of the conductive heat exchange. The problem is
characterized by anisotropic scattering of the medium and by specularly and
diffusely reflecting boundaries. In [8] the theorems of existence and uniqueness
of the radiative-conductive heat transfer problem for case of isotropic scattering
and non-reflecting boundaries are proved. Radiative-conductive heat transfer
model is difficult for numerical implementing because of the numerical insta-
bility. In [9] is noted that using the diffusion approximation (also named P1
approximation) allows to describe the behavior of the temperature properly in
computationally difficult cases of high temperatures. Thus, this approximation
can be successfully applied to various heat transfer problems in which high accu-
racy is not required. The aims of the current work are to present a constructive
proof of the solvability of the nonlinear heat transfer problem, to find uniqueness
conditions of the solution, to propose an algorithm for solving, and to show that
the algorithm can be implemented in practice. The proposed approach is related

∗ Corresponding author
Email addresses: [email protected] (Andrey E. Kovtanyuk), [email protected]
(Alexander Yu. Chebotarev)

Preprint submitted to Elsevier April 7, 2013

to finding a fixed point of a nonlinear solution operator A : R2 → R2 constructed
on the base of modified form of the diffusion approximation. Based on the con-
traction property of operator A, the iterative algorithm is proposed. Sufficient
conditions of existence and uniqueness of a fixed point are determined. Under
these conditions the proposed iterative algorithm converges at a geometric rate.

2. Problem formulation
Let us consider a coupled radiative-conductive heat transfer problem formu-
lated as in [1, 2]. The equation of the radiation transfer for a homogenous layer
is written in the normalized form as
Z1
ω
µIτ (τ, µ) + I(τ, µ) = p(µ, µ0 )I(τ, µ0 )dµ0 + (1 − ω)Θ4 (τ ), (1)
2
−1

where I(τ, µ) is the normalized density of the radiation flux at the point τ ∈
[0, τ0 ] in the direction which angle cosine with the positive direction of the axis
τ is µ ∈ [−1, 1]; ω < 1 is the albedo of single scattering; p(µ, µ0 ) is the phase
function; Θ(τ ) is the normalized temperature.
We supply equation (1) with the boundary conditions
Z1
I(0, µ) = ε1 Θ41 + ρs1 I(0, −µ) + 2ρd1 I(0, −µ0 )µ0 dµ0 , µ > 0,
0
(2)
Z1
I(τ0 , µ) = ε2 Θ42 + ρs2 I(τ0 , −µ) + 2ρd2 I(τ0 , µ0 )µ0 dµ0 , µ < 0.
0

Here, Θ1 and Θ2 are the normalized temperatures on the boundaries (Θ1,2 >
0), for definiteness, we assume that Θ1 ≤ Θ2 ; ρsi and ρdi are the coefficients
of specular and diffuse reflections, respectively, and εi = 1 − ρsi − ρdi are the
emissivity coefficients for the boundary surfaces. It is assumed that ε1 , ε2 > 0.
The equation of the conductive heat transfer is written as
 1 0
Z
1 
Θ00 (τ ) = I(τ, µ)µdµ , (3)
2Nc
−1

and Nc is the conduction-to-radiation parameter [1, 2]. For equation (3), we set
the following boundary conditions:
Θ(0) = Θ1 , Θ(τ0 ) = Θ2 . (4)
Now we consider an approach based on the diffusion approximation (also
named P1 approximation). We represent the function I(τ, µ) by the two first
terms in the Fourier expansion in Legendre polynomials:
I(τ, µ) ' Φ0 (τ ) + µΦ1 (τ ).

2
In the framework of the diffusion approximation we use the following expansion
for the phase function (see [10]):

p = 1 + Aµµ0 ,

where A ∈ [−1, 1] is an anisotropy parameter.

Denote x = τ /τ0 , and φ0 (x) = Φ0 (xτ0 ), θ(x) = Θ(xτ0 ). It gives us the
following P1 approximation of equation (1):

−φ000 (x) + αφ0 (x) = αθ4 (x), (5)

where α = τ02 (1−ω)(3−Aω). There are different approaches to the derivation of

the boundary conditions for the diffusion approximation. In the present work,
we choose the Marshak boundary conditions [11] which give

φ0 (0) − β1 φ00 (0) = Θ41 , (6)

φ0 (1) + β2 φ00 (1) = Θ42 , (7)

where
2 (2 − εi )
βi = , i = 1, 2.
τ0 εi (3 − Aω)
With new notations, equation (3) and boundary conditions (4) are rewritten as
follows:
1
θ00 (x) = −σφ000 (x), σ = , (8)
3Nc
θ(0) = Θ1 , θ(1) = Θ2 . (9)
The system (5)-(9) is a diffusion approximation of the radiative-conductive heat
transfer problem (1)-(4). Note that the function φ0 (x) is interpreted here as the
function I(xτ0 , µ) averaged over all directions µ.
From equation (8), we obtain

θ(x) = −σφ0 (x) + (κ1 − κ2 )x + κ2 . (10)

By use of (10) the problem (5)-(9) is rewritten as follows:

−θ00 (x) + αθ(x) + ασθ4 (x) = α((κ1 − κ2 )x + κ2 ), x ∈ (0, 1), (11)

θ(0) = Θ1 , θ(1) = Θ2 , (12)

0
θ(0) − β1 θ (0) = −σΘ41 − β1 (κ1 − κ2 ) + κ2 , (13)
θ(1) + β2 θ0 (1) = −σΘ42 + (β2 + 1)(κ1 − κ2 ) + κ2 . (14)
The system (11)-(14) is another form of the diffusion approximation (5)-(9).
Our goal is to prove the theorems of existence and uniqueness of the solution of
the boundary-value problem (11)-(14).

3
3. Reduction boundary-value problem to operator equation

Denote ϕ(θ) = σθ4 + θ, θ ∈ R. The following statement can be proved

similarly as in [8] (Lemma 3.1, Theorem 3.2):
Lemma 1 If inequality ϕ(Θ1 ) ≤ (κ1 − κ2 )x + κ2 ≤ ϕ(Θ2 ) holds for any x ∈
[0, 1], then the problem (11)-(12) has unique solution θ ∈ C 2 [0, 1], Θ1 ≤ θ(x) ≤
Θ2 , x ∈ (0, 1).

Let us transform the boundary conditions (13)-(14). Using integration by

parts, we have:
Z1
θ00 (x)(1 − x)dx = Θ2 − Θ1 − θ0 (0), (15)
0

Z1
θ00 (x)xdx = Θ1 − Θ2 + θ0 (1). (16)
0

Using (11), we obtain from (15) and (16) boundary values for θ0 :

Z1
1 1
θ0 (0) = Θ2 − Θ1 − α (1 − x)ϕ(θ(x))dx + α(κ1 − κ2 ) + ακ2 , (17)
6 2
0

Z1
0 1 1
θ (1) = Θ2 − Θ1 + α xϕ(θ(x))dx − α(κ1 − κ2 ) − ακ2 , (18)
3 2
0

Substituting the expressions (17), (18) into (13), (14), we obtain the following
system for κ1 , κ2 :
³α ´ µ ¶
β1 α
β1 − 1 κ1 + + β 1 + 1 κ2 =
6 3

Z1
ϕ(Θ1 ) − β1 (Θ2 − Θ1 ) + αβ1 (1 − x)ϕ(θ(x))dx, (19)
0
µ ¶ ³α ´
αβ2
1 + β2 + κ1 + β2 − 1 κ2 =
3 6
Z1
ϕ(Θ2 ) + β2 (Θ2 − Θ1 ) + αβ2 xϕ(θ(x))dx. (20)
0

From (19), (20) we derive expressions for κ1 , κ2 . For simplicity, consider the
case where β1 = β2 = β. Corresponding to this case, the expressions for κ1 , κ2
are written:
κ1 = γ1 (γ2 ϕ(Θ1 ) + γ3 ϕ(Θ2 ) + γ4 ) +

4
 Z1 Z1
γ1 γ5 (1 − x)ϕ(θ(x))dx + γ1 γ6 xϕ(θ(x))dx, (21)
0 0

κ2 = γ1 (γ2 ϕ(Θ2 ) + γ3 ϕ(Θ1 ) − γ4 ) +

Z1 Z1
γ1 γ5 xϕ(θ(x))dx + γ1 γ6 (1 − x)ϕ(θ(x))dx, (22)
0 0

where
³ ³ α´ ³ α ´´−1 ³ α´
γ1 = 1 + 2β 1 + + β2α 1 + , γ2 = 1 − β,
3 12 6
³ µ ¶
α´ αβ
γ3 = 1 + β 1 + , γ4 = β(Θ2 − Θ1 ) 1 + ,
3 2
³ α´ ³ ³ α ´´
γ5 = αβ 2 1 − , γ6 = αβ 1 + β 1 + .
6 3
In the case of different parameters β1 and β2 , there is no principal difficulty
in solving the linear system for determination of κ1 and κ2 . This case is not
considered here due to the complexity of the resulting formulas.
Let
D = {κ = (κ1 , κ2 ) ⊂ R2 : ϕ(Θ1 ) ≤ κ1 , κ2 ≤ ϕ(Θ2 )},
and define the operators T : D → C[0, 1], A : D → R2 as follows: T [κ] = θ,
where θ is a solution of the problem (11)-(12); A[κ] is defined by the right-hand
sides of equations (21) and (22), where θ = T [κ]. Thus, the following statement
is true.

Lemma 2 The boundary-value problem (11)-(14) is equivalent to operator equa-

tion κ = A[κ].

4. Solvability of the boundary-value problem

First, we prove the auxiliary statements.

Lemma 3 The operator A : D → R2 is continuous.

Proof. By the definition of the operator, it is enough to prove the continuity

of the following maps: B1 : D → R, B2 : D → R,

Z1 Z1
B1 [κ] = xϕ(θ(x))dx, B2 [κ] = (1 − x)ϕ(θ(x))dx,
0 0

where θ = T [κ].

5
 Let θ̂ = T [κ̂], where κ̂ ∈ D. The following inequality holds:
¯ 1 ¯
¯Z ¯
¯ ¯ ¡ ¢
¯ x(ϕ(θ̂(x)) − ϕ(θ(x))dx¯ ≤ 1 1 + 4σΘ32 max |θ̂(x) − θ(x)|. (23)
¯ ¯ 2 [0,1]
¯ ¯
0

Further consider the function η(x) = θ̂(x) − θ(x), x ∈ [0, 1]. It satisfies the
following conditions:

−η 00 (x) + α(ϕ(θ̂(x)) − ϕ(θ(x))) = α(fˆ(x) − f (x)), x ∈ (0, 1), (24)

η(0) = η(1) = 0,
where f (x) = (κ1 − κ2 )x + κ2 and fˆ(x) = (κ̂1 − κ̂2 )x + κ̂2 , x ∈ [0, 1].
Note
Z1
kf − f k = (fˆ(x) − f (x))2 dx → 0, if |κ̂ − κ| → 0.
ˆ 2

0
p
Hereinafter by the sign k · k we denote the norm in L2 (0, 1), and |κ| = κ21 + κ22
is the length of the vector κ. Since the function ϕ(θ) increases monotonically
over θ ≥ 0, then
(ϕ(θ̂) − ϕ(θ))(θ̂ − θ) ≥ 0.
After multiplying (24) by η and integrating over (0,1), we obtain
° x °
Z1 °Z °
2 ° °
kη 0 k2 = η 0 (x)dx ≤ αkfˆ − f kkηk = αkfˆ − f k °
° η 0
(s)ds° ≤ αkfˆ − f kkη 0 k.
°
° °
0 0

Thus, ¯ x ¯
¯Z ¯
¯ ¯
max |θ̂(x) − θ(x)| = max ¯¯ η 0 (s)ds¯¯ ≤ kη 0 k ≤ αkfˆ − f k,
[0,1] [0,1] ¯ ¯
0

and continuity B1 follows from (23).

The integral
Z1 ³ ´
(1 − x) ϕ(θ̂(x)) − ϕ(θ(x)) dx
0

is estimated similarly. The lemma is proved. ¥

Lemma 4 If the following inequality holds

α(1 + 3β) ¡ ¢
0< ≤ σ(Θ1 + Θ2 ) Θ21 + Θ22 , (25)
6−α
then A[D] ⊂ D.

6
Proof. Let
l = (l1 , l2 ) = A[κ] = (A1 [κ], A2 [κ]) , κ ∈ D,
where A1 , A2 are defined by the right-hand sides of equalities (21), (22), respec-
tively.
Let us consider the case l1 ≥ l2 , and prove that l1 ≤ ϕ(Θ2 ). For α < 6 the
following inequality holds:
 
Z1 Z1
l1 = γ1 γ2 ϕ(Θ1 ) + γ3 ϕ(Θ2 ) + γ4 + γ5 (1 − x)ϕ(θ)dx + γ6 xϕ(θ)dx ≤
0 0

ϕ(Θ2 ) + γ1 (γ2 (ϕ (Θ1 ) − ϕ (Θ2 )) + γ4 ) .

If inequality (25) holds, then

(γ2 (ϕ (Θ1 ) − ϕ (Θ2 )) + γ4 ) ≤ 0.

Hence l1 ≤ ϕ(Θ2 ).
Similarly, we can prove the inequality l2 ≥ ϕ(Θ1 ) and also that l1 ≥ ϕ(Θ1 ),
l2 ≤ ϕ(Θ2 ) for case l1 < l2 . Thus, A[κ] ⊂ D for all κ ∈ D. ¥

Using Brouwer’s fixed point theorem and Lemmas 3,4, we can formulate the
following statement.
Theorem 1 Let β1 = β2 = β and inequality (25) holds. Then there exists at
least one solution θ(x) of the problem (11)-(14), such that

Θ1 ≤ θ(x) ≤ Θ2 , x ∈ [0, 1].

5. Uniqueness of the boundary-value problem

Determine the sufficient conditions under which the operator A is a con-

traction. Let κ, κ̂ ∈ D, θ = T [κ], θ̂ = T [κ̂]. Further, we obtain a more ac-
curate estimate of function η(x) = θ̂(x) − θ(x) than in Lemma 3. For this
goal, we multiply (24) by
¡ η and integrate
¢ over (0,1). Taking into account that
(ϕ(θ̂) − ϕ(θ))(θ̂ − θ) ≥ 1 + 4Θ31 σ η 2 , we obtain
¡ ¢
kηx k2 + α 1 + 4Θ31 σ kηk2 ≤ αkfˆ − f kkηk.

From the last estimate and well-known inequality πkηk ≤ kηx k, it follows
α
kηk ≤ kfˆ − f k. (26)
π2 + α (1 + 4Θ31 σ)

Taking into account (26), we obtain the estimate:

¯ 1 ¯
¯Z ³ ´ ¯¯ Z1 ¡
¯ ¢
¯ x ϕ(θ̂(x)) − ϕ(θ(x)) dx¯ ≤ x 1 + 4σΘ32 |η(x)|dx ≤
¯ ¯
¯ ¯
0 0

7
 ¡ ¢
1 ¡ ¢ 1 α 1 + 4σΘ32
3
√ 1 + 4σΘ2 kηk ≤ √ kfˆ − f k.
3 3 (π 2 + α (1 + 4Θ31 σ))
√
Since kfˆ − f k ≤ |κ̂ − κ|/ 2, then
¯ 1 ¯
¯Z ³ ´ ¯¯
¯
¯ x ϕ(θ̂(x)) − ϕ(θ(x)) dx¯ ≤ q|κ̂ − κ|, (27)
¯ ¯
¯ ¯
0

and similarly,
¯ 1 ¯
¯Z ³ ´ ¯¯
¯
¯ (1 − x) ϕ(θ̂(x)) − ϕ(θ(x)) dx¯ ≤ q|κ̂ − κ|. (28)
¯ ¯
¯ ¯
0

Here ¡ ¢
1 α 1 + 4Θ32 σ
q=√ .
6 (π 2 + α (1 + 4Θ31 σ))
Note that µ ¶3
1 Θ2
0<q< √ ,
6 Θ1
and q = O(α) as α → +0 for fixed σ.
Let us consider l = A[κ], ˆl = A[κ̂] and estimate the |ˆl − l|2 = (ˆl1 − l1 )2 +
ˆ
(l2 − l2 )2 . From (27), (28) it follows
αβ 2 (1 − α/6) + αβ (1 + β (1 + α/3))
|ˆl1 − l1 | ≤ q|κ̂ − κ| =
1 + 2β (1 + α/3) + αβ 2 (1 + α/12)
2αβ
q|κ̂ − κ|.
2 + αβ
The expression |ˆl2 − l2 | is estimated by the same term. Hence,
√
2 2αβ
|A[κ̂] − A[κ]| ≤ q|κ̂ − κ|.
2 + αβ
Thus, the following statement is true.
Theorem 2 Let inequality (25) and
¡ ¢
2 1 + 4Θ32 σ βα2
√ <1 (29)
3 (π 2 + (1 + 4Θ31 σ) α) (2 + αβ)
hold, then there exists a unique solution θ of the boundary-value problem (11)-
(14), Θ1 ≤ θ(x) ≤ Θ2 , x ∈ (0, 1).
Inequalities (25) and (29) are complex restrictions on the parameters of the
problem. However, for example, we can see that these restrictions are fulfilled
when ω is close enough to 1 (i.e. α is close enough to 0). This corresponds
to the case of dominant scattering, which is favorable for the use of the P1
approximation.

8
6. An iterative procedure

The proved contraction property of the operator A is the basis for the fol-
lowing iterative procedure. For finding a solution of the problem (11)-(14), we
will use a simple iterative method:

κ(0) ∈ D, κ(n) = A[κ(n−1) ], n = 1, 2, ... (30)

Initial approximation κ(0) can be calculated using equalities (21) and (22), where
θ(x) = (1 − x)Θ1 + xΘ2 , x ∈ [0, 1]. If the sequence κ(n) converges in R2 , then
its limit κ is a fixed point of the operator A, and θ = A[κ] is a solution of the
problem (11)-(14). Notice that condition (25) ensures convergence at least of a
subsequence κ(np ) , and condition (29) ensures convergence of κ(n) at a geometric
rate.
Thus, the contraction property of the operator A provides fast convergence of
the proposed iterative procedure. In the following section, we will demonstrate
this by numerical experiments.

7. Numerical experiments

For a numerical experiment we set the following values of problem data:

τ0 = 1, σ = 200, ω = 0.9, ε = 0.7, A = 1, Θ1 = 0.5, Θ2 = 1, which satisfy
conditions (25) and (29). Our choice of the parameters for the problem is based
on [12]. Nevertheless, we took higher value of σ, as it corresponds to higher
temperature. This case is a more interesting from a computational point of
view.
The normalized temperatures calculated at the 20th step of the proposed
iterative procedure (solid plot) and by the modified Monte Carlo method [9]
(dots) are shown in Fig. 1. To demonstrate the fast convergence of the proposed
algorithm, we compare it with a simple iterative procedure which we apply to
the problem (5)-(9) in the following way: for a given temperature θ, we find the
solution ϕ0 of the problem (5)-(7), further, the temperature is calculated from
(8) and (9) for given ϕ0 , and so on. The numerical results are compared with
the data obtained by the modified Monte Carlo method [9]. In Fig. 2, the plots
of mean-square deviations between temperatures obtained by diffusion approx-
imations and by the modified Monte Carlo method are shown. The dashed plot
corresponds to simple iterative procedure based on system (5)-(9). The solid
plot corresponds to the proposed iterative procedure. It can be seen that the
proposed algorithm provides a good accuracy after 2 iterations. Whereas the
simple iterative procedure based on system (5)-(9) needs about 20 iterations
for obtaining the same accuracy. Thus, the proposed algorithm demonstrates
fast convergence of iterative process. This does not mean that the proposed
algorithm is more efficient in terms of computation time. Due to nonlinearity
of problem (11)-(12), the implementation of a single step of the procedure (30)
takes more computational time than the implementation of a single step of the

9
simple iterative procedure. The advantage of the proposed algorithm is its guar-
anteed convergence. Also, there is an opportunity for finding an a priory error
estimate.
The calculations for the diffusion models were performed by use of mathe-
matical software Maple 9.5.
1

0.95

0.9

0.85
Normalized temperature (θ)

0.8

0.75

0.7

0.65

0.6

0.55

0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Point of the layer (x)

Fig. 1. The temperature profiles calculated by the proposed iterative

procedure (solid plot) and by the modified Monte Carlo method [9] (dots).
0.12

0.1
Mean−square deviation

0.08

0.06

0.04

0.02

0
0 2 4 6 8 10 12 14 16 18 20
Number of iterations

10
View publication stats

Fig. 2. Mean-square deviations between temperatures obtained by diffu-

sion approximations and by the modified Monte Carlo method [9]. The
dashed plot corresponds to iterative procedure based on system (5)-(9).
The solid plot corresponds to proposed iterative procedure.

References

[1] M.N. Ozisik, Radiative Transfer and Interaction with Conduction and Con-
vection, John Wiley, New York, 1973.

[2] M.F. Modest, Radiative Heat Transfer, McGraw-Hill, New York, 1993.
[3] S. Andre, A. Degiovanni, A theoretical study of the transient coupled con-
duction and radiation heat transfer in glass: phonic diffusivity measurements
by the Nash technique, Int. J. Heat Mass Transfer, 38 (18) (1995) 3401–3412.
[4] S. Andre, A. Degiovanni, A new way of solving transient radiative-conductive
heat transfer problems, J. Heat Transfer, 120 (4) (1998) 943–955.
[5] J.M. Banoczi, C.T. Kelley, A fast multilevel algorithm for the solution of
nonlinear systems of conductive-radiative heat transfer equations, SIAM J.
Sci. Comp., 19 (1) (1998) 266–279.
[6] A. Klar, N. Siedow, Boundary layers and domain decomposition for radiative
heat transfer and diffusion equations: applications to glass manufacturing
process, Eur. J. Appl. Math., 9 (4) (1998) 351–372.

[7] O. Tse, R. Pinnau, N. Siedow, Identification of Temperature Dependent

Parameters in Laser-Interstitial Thermo Therapy, Math. Models Methods
Appl. Sci., 22 (9) (2012)
[8] C.T. Kelley, Existence and uniqueness of solutions of nonlinear systems of
conductive-radiative heat transfer equations, Transport Theory Statist. Phys.,
25 (2) (1996) 249–260.
[9] A.E. Kovtanyuk, N.D. Botkin, K.-H. Hoffmann, Numerical simulations of
coupled radiative-conductive heat transfer model using a modified Monte
Carlo method, Int. J. Heat and Mass Transfer, 55 (2012) 649–654.
[10] T.A. Germogenova, The regular components of asymptotic approximations
of solution to the transport equation for optically thick media, Zh. Vychisl.
Mat. Mat. Fiz., 37 (4) (1997) 464–482
[11] R. Marshak, Note on the spherical harmonic method as applied to the
Milne problem for sphere, Phys. Rev., 71 (7) (1947) 443–446.

[12] C.E. Siewert, J.R. Thomas, A computational method for solving a class
of coupled conductive-radiative heat-transfer problems, J. Quant. Spectrosc.
Radiat. Transfer, 45 (5) (1991) 273–281.

11