You are on page 1of 34

J. Inv. Ill-Posed Problems, Vol. 9, No. 5, pp.

435–468 (2001)
c VSP 2001

Theoretical analysis of inverse extremal problems


of admixture diffusion in viscous fluids
G. V. ALEKSEEV∗ and E. A. ADOMAVICHUS∗ .

Received March 21, 2000. Revised February 26, 2001

Abstract — We consider some inverse extremal problems for the stationary equations
of admixture diffusion in a viscous incompressible fluid under inhomogeneous bound-
ary conditions for the velocity and concentration. We prove theorems on existence
of solutions for the considered inverse problems, derive and analyze the optimality
systems, and also establish conditions of uniqueness of their solutions for concrete ob-
jective functionals. The conditions of solvability of inverse extremal problems for the
stationary Navier – Stokes equations are presented.

INTRODUCTION

Application of mathematical modeling to the problems of forecasting the eco-


logical state of the atmosphere and ocean leads to initial-boundary value prob-
lems for equations describing the diffusion and transformation of an admixture
in a region under consideration. Such boundary-value problems contain some
thermohydrodynamical parameters and functions describing the densities of the
admixture sources. To uniquely solve such boundary-value problems the values
of all input parameters, the boundary and initial functions, and the densities of
the sources are to be given.
At the same time, the parameters of some sources may be unknown. Further-
more, a source may be inaccessible to the direct measurements and the infor-
mation on its parameters may be concealed. Unaccounted emission of harmful
admixtures from such sources may be very dangerous for the environment. Of
especial danger are radioactive admixtures from a sunk object with a nuclear
reactor.
Problems of environment protection against emission of harmful admixtures
∗ Institute of Applied Mathematics, Far Eastern Branch of Russia Academy of Sciences,

Vladivostok 690041, Radio Street, 7.


E-mails: alekseev@iam-mail.febras.ru, eduard@iam-mail.febras.ru
The work was supported by the Russian Foundation for Basic Research (grant 99-01-00214)
436 G. V. Alekseev, E. A. Adomavichus

lead to problems of detection of unknown admixture sources and identification


of their parameters are to be solved by mathematical methods. By their setting,
the above problems fall into a class of inverse problems. In strict mathematical
formulation, they consist in finding the parameters of an unknown admixture
source from the measurements of the concentration field produced by this source
in some region and information about the source as well.
side by side with inverse problems, the extremal problems of admixture diffu-
sion are also very important in the theory and applications. In these problems,
a certain objective functional is introduced and it is required to minimize it,
for example, by appropriate choice of the densities of admixture sources. It
should be noted that solution of inverse problems can be reduced to solution of
extremal problems with appropriate objective functionals.
The main purpose of this paper is to analyze theoretically the inverse ex-
tremal problems for the equations of admixture transfer in an incompressible
viscous fluid considered in a bounded domain with a Lipschitz boundary. Such
problems are formulated as problems to minimize certain objective functionals
on weak solutions of the original boundary value problem. We take an analogue
of the stationary system of the Oberbeck-Boussinesq equations as a mathemat-
ical model describing the admixture diffusion. As the corresponding objective
functionals we take the L2 -deviation of the sought concentration field (or veloc-
ity field) from the given one, the L2 -norm of the concentration (or velocity) field
gradient, etc. The volume density of admixture sources, the concentration dis-
tribution on one part of the boundary and the admixture flux on another part,
and the boundary distribution of the flow velocity are the desired parameters.
We prove the uniqueness and existence theorems for inverse extremal problems,
we obtain and analyze the optimality systems for the general nonlinear and lin-
ear models of admixture diffusion. For the linear model we derive an explicit
formula for finding the minimum of the objective functional in the case where
the functional is linear with respect to the concentration. It generalizes the
known formula obtained in [26] when solving the problem of optimal location
of factories and plants near ecologically important zones.

1. STATEMENTS OF THE BASIC BOUNDARY-VALUE


PROBLEMS
Let Ω be a domain of the space Rd , d = 2, 3, with a Lipschitz boundary Γ
composed of two parts ΓD and ΓN . In this paper we consider inverse extremal
problems for the system of equations
−νΔu + (u · grad)u + grad p = f + βϕG in Ω, div u = 0 in Ω, (1.1)
∂ϕ
−λΔϕ + u · grad ϕ − w0 + κϕ = f in Ω (1.2)
∂z
that describes admixture diffusion in Ω. We assume the following boundary
conditions:
 ∂ϕ 
u = g on Γ, ϕ = ψ on ΓD , λ + αϕ = χ on ΓN . (1.3)
∂n
Admixture diffusion in viscous fluids 437

Here conventional notations are used (see, e.g., 14, 15, 26). Namely, u, p,
and ϕ are the velocity, the pressure, and the admixture concentration (the
desired functions); ν = const > 0 is the coefficient of kinematic viscosity; λ =
const > 0 is the diffusion coefficient; w0 = const ≥ 0 is the vertical speed of
precipitation of admixture particles under the influence of gravity; κ ≥ 0 is the
value characterizing the speed of the chemical reaction; G = −(0, 0, G) is the
free fall acceleration vector (the axis z of the Cartesian coordinate system is
upwards); f is the volume density of external forces; and f , α, ψ, χ, and g are
some functions. The problem (1.1)–(1.3) with given functions f , f , α, ψ, χ,
and g and given β (this parameter is described below) will be called Problem 1.
Note that (1.1), (1.2) is a system of three nonlinear equations for the desired
values u, p, and ϕ connected with each other in the momentum equation by the
term bϕ ≡ βϕG which describes the buoyancy effect due to the influence of the
admixture concentration on the fluid’s dynamic behaviour. In the case when
β = const > 0, κ = 0, and w0 = 0, relations (1.1)–(1.3) are a boundary-value
problem for the system of mass-transfer equations in the Oberbeck – Bussinesq
approximation [14]. If β = const < 0 and the variable ϕ has the sense of temper-
ature, then (1.1),(1.2) for κ = 0 and w0 = 0 are a system of thermal convection
equations. The extremal problems for the latter model were considered by many
authors, see, e.g., [1, 2, 5–7, 12, 13, 21, 23, 29, 31].
As for the original problem (1.1)–(1.3), a theoretical analysis of direct and
extremal problems for the case where (1.1)-(1.2) describes bioconvection can
be found in [16, 24]. In these papers a special case of the problem (1.1)–(1.3)
was considered with κ = 0, the boundary conditions of the form u = 0 on Γ
and λ∂ϕ/∂n + w0 n3 ϕ = 0 on Γ, and w0 ≤ 0. Here n3 = cos(n, z), where n
is the unit vector of the external normal to Γ. In [24] the authors proved the
existence of a solution to the original boundary-value problem and positiveness
of the concentration. In [16] the existence of a solution to the above-mentioned
problem is also proved and the minimization problem for the functional

1 μ
J(ϕ, ϕ∗ ) = (ϕ − ϕd )2 dΩ + ϕ∗ (1.4)
2 Ω 2

is investigated. Here ϕd ∈ L2 (Ω) is a given function that describes the wanted


distribution of concentration in Ω, μ = const ≥ 0, and ϕ∗ is the average value
of concentration (it plays the role of a control).
In another special case when β = 0, i. e. the term bϕ is absent, the prob-
lem (1.1)–(1.3) splits into two problems: the first problem, described by rela-
tions (1.1) for β = 0 and the first condition in (1.3), is the boundary-value
problem for the stationary Navier – Stokes equations. The extremal problems
for the stationary Navier – Stokes system with the Dirichlet boundary condition
in (1.3) and with conditions of Neumann type on Γ were studied, e. g. in [10,
11, 17, 19, 20]. The second problem, described by equation (1.2) with given
velocity vector u and by the rest of the boundary relations in (1.3), is a mixed
boundary-value problem for a linear equation of admixture diffusion. The ex-
tremal problems for a non-stationary analogue of equation (1.2) originate from
438 G. V. Alekseev, E. A. Adomavichus

an important ecological problem of locating plants and factories near ecologi-


cally important zones. Such problems were first formulated by Marchuk in [26]
and then investigated in detail by many authors; see, e. g. [8, 15, 27, 28]. Let
us also mention a recently published article [9] by one of the authors where the
results of investigation of extremal inverse problems for the general model of
heat-and-mass transfer are presented .

2. PROPERTIES OF SOLUTIONS OF THE DIRECT


BOUNDARY-VALUE PROBLEM

2.1. Functional spaces


In investigation of the extremal inverse problems we shall use the weak formu-
lation of the boundary-value problem (1.1)–(1.3) based on application of the
Sobolev functional spaces H s (D), s ∈ R, and Lr (D), 1 ≤ r ≤ ∞, where D
is either the domain Ω, or the boundary Γ, or some its part Γ0 with positive
measure.
We shall denote the corresponding spaces of vector-functions by H s (D)
and Lr (D); the scalar product in L2 (Ω) by (·, ·), in L2 (Γ) by (·, ·)Γ , and in
L2 (Γ0 ) by (·, ·)Γ0 ; the norm in L2 (Ω) by  ·  and in L2 (Γ0 ) by  · Γ0 ; the norm
and semi-norm in H 1 (Ω) or H 1 (Ω) by  · 1 and | · |1 , respectively; in H 1/2 (Γ)
by  · 1/2,Γ ; and the duality relation for a pair X and X ∗ by · , ·
X ∗ ×X or
· , ·
if it causes no confusion. The duality relation for the space H 1/2 (Γ) and
its vector analogue H 1/2 (Γ) will be denoted by · , ·
Γ .
Let the following conditions be fulfilled:
(i) Ω is a bounded finitely connected domain in the space Rd with boundary
Γ ∈ C 0,1 composed of N connected components Γ(i) , i = 1, 2, . . . , N ;
(ii) open parts ΓD and ΓN of the boundary Γ are such that ΓD ∈ C 0,1 ,
ΓD = ∅, ΓN ∈ C 0,1 , ΓD ∩ ΓN = ∅, and Γ = ΓD ∪ ΓN .
It is well-known that if the conditions (i) and (ii) are fulfilled, then there
exist linear continuous trace operators γ : H 1 (Ω) → H 1/2 (Γ), γ|ΓD : H 1 (Ω) →
H 1/2 (ΓD ), and γ|ΓN : H 1 (Ω) → H 1/2 (ΓN ). In particular, the following estimate
is valid with some constant cΓ :
γ|ΓN SΓN ≤ cΓ S1 , ∀S ∈ H 1 (Ω). (2.1)
1 1
Along with the spaces H (Ω) and H (Ω) we shall use their sub-
spaces H 1 (Δ, Ω) = {ϕ ∈ H 1 (Ω) | Δϕ ∈ L2 (Ω)} and H 1 (Δ, Ω) = (H 1 (Δ, Ω))d
with which the following Green formulas [12, 18] hold true:
   ∂ϕ 
− ΔϕS dΩ = ∇ϕ · ∇S dΩ − , γS , ∀ϕ ∈ H 1 (Δ, Ω), S ∈ H 1 (Ω),
Ω Ω ∂n Γ
(2.2)
   ∂u 
− Δu · v dΩ = ∇u : ∇v dΩ − , γv , ∀u ∈ H 1 (Δ, Ω), v ∈ H 1 (Ω).
Ω Ω ∂n Γ
(2.3)
Admixture diffusion in viscous fluids 439

Introduce the spaces


V = {v ∈ H01 (Ω) | div v = 0},
L20 (Ω) = {p ∈ L2 (Ω) | (p, 1) = 0},
T = H 1 (Ω, ΓD ) ≡ {S ∈ H 1 (Ω) | S|ΓD = 0},
L∞
+ (ΓN ) = {α ∈ L∞ (ΓN ) | α ≥ 0 on ΓN },
L2+ (Ω) = {k ∈ L2 (Ω) | k ≥ 0 in Ω},
  
H̃ 1 (Ω) = u ∈ H 1 (Ω) | u · n = 0 on ΓN , u · n dΓ = 0, i = 1, . . . , N ,
Γ(i)
1/2 1
H̃ (Γ) = {γu | u ∈ H̃ (Ω)}.
We shall use the following bilinear and trilinear forms:
a : H 1 (Ω) × H 1 (Ω) → R, ã : H 1 (Ω) × H 1 (Ω) → R,
b : H 1 (Ω) × L20 (Ω) → R, b1 : H 1 (Ω) × H 1 (Ω) → R,
c : L4 (Ω) × H 1 (Ω) × H 1 (Ω) → R, c1 : L4 (Ω) × H 1 (Ω) × H 1 (Ω) → R
that act by the formulas
 
a(u, v) = ∇u : ∇v dΩ, ã(ϕ, S) = ∇ϕ · ∇S dΩ,
Ω Ω
 
b(v, q) = − q div v dΩ, b1 (S, v) = bS · v dΩ,
 Ω Ω
c(u, v, w) = [u · ∇v] · w dΩ, c1 (u, ϕ, S) = (u · ∇ϕ)S dΩ,
Ω Ω

where b = βG. It is known (see, e. g. [37]) that these formulas are continuous
and
c(u, v, w) = −c(u, w, v), c(u, v, v) = 0,
1 1
∀u ∈ H̃ (Ω), div u = 0, v, w ∈ H (Ω), v · w = 0 on Γ, (2.4)

c1 (u, ϕ, S) = −c1 (u, S, ϕ), c1 (u, S, S) = 0,


∀u ∈ H̃ 1 (Ω), div u = 0, ϕ ∈ H 1 (Ω), S∈T, (2.5)

c(u, v, w) ≤ c0 uL4 (Ω) v1 w1 , c1 (u, ϕ, S) ≤ c0 uL4 (Ω) ϕ1 S1. (2.6)
Here c0 is a constant. The forms a and ã are coercive on V and T , respectively.
Therefore the following inequalities hold with some constants α0 > 0 and α1 > 0:
a(v, v) ≥ α0 v21 , ∀v ∈ V , ã(S, S) ≥ α1 S21 , ∀S ∈ T . (2.7)
Moreover, on the pair H01 (Ω) × L20 (Ω) the form b satisfies the so-called “inf-sup”
condition [17, p.81]: there exists a constant c̃ > 0 such that
b(v, q)
sup ≥ c̃q0 , ∀q ∈ L20 (Ω). (2.8)
0=v∈H01 (Ω) v1
440 G. V. Alekseev, E. A. Adomavichus

Introduce a bilinear form a1 : H 1 (Ω) × T → R by the formula


 ∂ϕ 
a1 (ϕ, S) = λã(ϕ, S) + λ(αϕ, S)ΓN − w0 , S + (κϕ, S). (2.9)
∂z
We set
|b1 (S, v)| |c(u, v, w)|
β1 = sup , N = sup ,
S∈H 1 (Ω) S1 v1 u,v∈H 1 (Ω) u1 v1 w1
v∈V w∈V
|c1 (u, ϕ, S)|
N1 = sup . (2.10)
u∈H (Ω) 1 u1 ϕ1 S1
ϕ∈H 1 (Ω), S∈T

We divide the input data of Problem 1 into two groups: the group of
fixed data into which we include the invariable coefficients ν, λ, and u0
and functions f , b, κ, and α; and the group of controls into which we in-
clude the functions f , ψ, χ, and g which will play the role of controls. Let
v0 = (ν, λ, w0 , f , b, κ, α) and v = (f, ψ, χ, g). We shall call v a control. Assume
that the control v may vary within some set K and in addition to (i) and (ii)
the following conditions hold:

(iii) ν = const > 0, λ = const > 0, w0 = const, λ∗ ≡ λα1 − |w0 | > 0,


κ ∈ L2+ (Ω), α ∈ L∞
+ (ΓN ), and 0 ≤ β1 < ∞;

(iv) f ∈ H −1 (Ω);

(v) K = K1 × K2 × K3 × K4 , where K1 ⊂ T ∗ , K2 ⊂ H 1/2 (ΓD ), K3 ⊂ L2 (ΓN ),


K4 ⊂ H̃ 1/2 (Γ) are nonempty closed convex sets;

(vi) u0 ∈ H̃ 1 (Ω) and div u0 = 0.

By (2.1) and (2.7), under fulfillment of conditions (i)–(iii) we have

|a1 (ϕ, S)| ≤ (λ + λc2Γ αL∞ (ΓN ) + |w0 | + κ)ϕ1S1 , ∀ϕ, S ∈ H 1 (Ω),
 ∂S 
a1 (S, S) ≥ λã(S, S) − w0 , S ≥ (λα1 − |w0 |)S21 = λ∗ S21 , ∀S ∈ T .
∂z
(2.11)

These relations imply that if conditions (i)–(iii) are fulfilled, then the form a1
introduced by formula (2.9) is continuous on H 1 (Ω) × H 1 (Ω) and T -coercive
with the constant λ∗ .

2.2. Properties of a solution to Problem 1


Multiply the first equation in (1.1) by a function v ∈ H01 (Ω), the second equa-
tion by a function q ∈ L20 (Ω), and equation (1.2) by a function S ∈ T ; then
integrate over the domain Ω; apply the Green formulas (2.2) and (2.3); and
Admixture diffusion in viscous fluids 441

finally use the boundary conditions (1.3). As a result, we obtain the weak for-
mulation of Problem 1: find a triple (u, p, ϕ) ∈ H 1 (Ω) × L20 (Ω) × H 1 (Ω) such
that

νa(u, v) + c(u, u, v) + b(v, p) = f , v


+ b1 (ϕ, v), ∀v ∈ H01 (Ω), (2.12)
a1 (ϕ, S) + c1 (u, ϕ, S) = f, S
+ (χ, S)ΓN , ∀S ∈ T , (2.13)
b(u, q) = 0, ∀q ∈ L20 (Ω), u = g on Γ, ϕ = ψ on ΓD . (2.14)

We shall call such a triple (u, p, ϕ) a weak solution of Problem 1.


Our purpose is to study inverse extremal problems for the model of mass-
transfer. We shall use some results on existence and uniqueness of a weak
solution of Problem 1 that were obtained in [3]. We cite them without proving
in the form of theorems formulated below. We set

Mg ≡ νcν g1/2,Γ + N c2ν g21/2,Γ ,


Mϕ,δ = λ−1
∗ (f T ∗ + cΓ χΓN ) + cN Mδ (ψ1/2,ΓD ), (2.15)

cN = 1 + λ−1 2
∗ (λ + λcΓ αL∞ (ΓN ) + w0 + κ + N1 cν g1/2,Γ ),

1, if ψ1/2,ΓD > 0,
κ= (2.16)
0, if ψ1/2,ΓD = 0.

In (2.15) and (2.16), cν is a constant depending on the coefficient ν and


the “size” of the set K4 ; Mδ : (0, ∞) → (0, ∞) is a family of continuous non-
decreasing functions depending on a parameter δ > 0 and such that Mδ (0) = 0
(see [12]).

Theorem 2.1. If conditions (i)–(v) are fulfilled, then for every element
v ∈ K there exists a weak solution (u, p, ϕ) of Problem 1 and the following
estimates hold:

u1 ≤ Mu (v0 , v), ϕ1 ≤ Mϕ (v0 , v). (2.17)

Here Mu and Mϕ are continuous non-decreasing functions of the norms f T ∗ ,


ψ1/2,ΓD > 0, χΓN , and g1/2,Γ defined by the formulas

2β1 2
Mu (v0 , v) ≡ (1 + κ)Mϕ,δ0 + (1 + κ)(f −1 + Mg ) + cν g1/2,Γ,
α0 ν α0 ν
(2.18)
Mϕ (v0 , v) ≡ (1 + κ)Mϕ,δ0 + (κ/β1 )(f −1 + Mg ), (2.19)

where the constant δ0 depends on Ω, ν, λ∗ , and β1 .

Remark 2.1. In the case where the set K is bounded, the right-hand
sides Mu and Mϕ of the estimates (2.17) are bounded; therefore for every solu-
tion (u, p, ϕ of Problem 1 the inequalities u1 ≤ c < ∞ and ϕ1 ≤ c < ∞
hold true with some constants c and c .
442 G. V. Alekseev, E. A. Adomavichus

If f = 0, ψ = 0, and χ = 0, then (2.15), (2.16), and (2.19) imply Mϕ = 0.


In virtue of (2.17), it means that ϕ = 0 in Ω. Therefore in this case only a
hydrodynamical process described by the equations

−νΔu + (u · grad)u + grad p = f , div u = 0, u = g on Γ (2.20)

takes place in Ω. The problem (2.20) will be called Problem 2; and a pair
(u, p) ∈ H 1 (Ω) × L20 (Ω) such that

b(u, q) = 0, ∀q ∈ L20 (Ω), u = g on Γ, (2.21)


νa(u, v) + c(u, u, v) + b(v, p) = f , v
, ∀v ∈ H01 (Ω), (2.22)

will be called its weak solution The solvability of Problem 2 follows from Theo-
rem 2.1, see the next theorem.

Theorem 2.2. Let conditions (i) and (iv) be fulfilled. Then for every
function g ∈ H̃ 1/2 (Γ) there exists a weak solution (u, p) of Problem 2 and the
following estimate holds:

2
u1 ≤ M̃u (ν, f , g) ≡ (f −1 + Mg ) + cν g1/2,Γ . (2.23)
α0 ν

Introduce the Reynolds number Re and Rayleigh diffusion number Ra by


the formulas
1 1 β1 N1
Re(v0 , v) = N Mu (v0 , v), Ra(v0 , v) = Mϕ (v0 , v). (2.24)
α0 ν α0 ν λ∗

Theorem 2.3. Let in addition to conditions (i)–(v) the following one be


fulfilled:
1 1 β 1 N1
Re(v0 , v) + Ra(v0 , v) ≡ N Mu (v0 , v) + Mϕ (v0 , v) < 1, ∀v ∈ K.
α0 ν α0 ν λ∗
(2.25)
Then a weak solution of Problem 1 is unique for every element v ∈ K.
In the particular case when ϕ = 0 or β1 = 0 (the second case corresponds
to the so-called passive admixture for which the field of concentrations has no
influence on the fluid flow in the domain Ω), condition (2.25) transforms into
the condition N M̄u /(α0 ν) < 1 of uniqueness of the weak solution to Problem 2
for the system of Navier – Stokes equations. It is analogous to the sufficient
condition of uniqueness obtained in [18, 30].
Along with the general non linear Problem 1 we shall consider the weak
formulation of its linear analogue: find a triple (u, p, ϕ) ∈ H 1 (Ω) × L20 (Ω) ×
H 1 (Ω) that satisfies conditions (2.14) and

νa(u, v) + c(u0 , u, v) + b(v, p) = f , v


+ b1 (ϕ, v), ∀v ∈ H01 (Ω), (2.26)
a1 (ϕ, S) + c1 (u0 , ϕ, S) = l, S
, ∀S ∈ T . (2.27)
Admixture diffusion in viscous fluids 443

Here the functional l ∈ T ∗ is defined by the formula


l, S
= f, S
+ (χ, S)ΓN , ∀S ∈ T , (2.28)
and the “velocity” u0 is a given function satisfying condition (vi).
The bilinear continuous forms νa( · , · ) + c(u0 , · , · ) : H 1 (Ω) × H01 (Ω) → R,
a1 ( · , · ) + c1 (u0 , · , · ) : H 1 (Ω) × T → R; b1 : H 1 (Ω) × H01 (Ω) → R, and
b : H01 (Ω) × L20 (Ω) → R that appear in (2.26) and (2.27) may be associated with
continuous operators A : H 1 (Ω) → H −1 (Ω), A1 : H 1 (Ω) → T ∗ , B1 : H 1 (Ω) →
H −1 (Ω), B : H01 (Ω) → (L20 (Ω))∗ ≡ L20 (Ω), and B ∗ : L20 (Ω) → H −1 (Ω) that act
by the formulas
Au, v
H −1 (Ω)×H01 (Ω) = νa(u, v) + c(u0 , u, v),
A1 ϕ, S
T ∗ ×T = a1 (ϕ, S) + c1 (u0 , ϕ, S),
B1 ϕ, v
H −1 (Ω)×H01 (Ω) = b1 (ϕ, v),
Bv, q
L20 (Ω)×L20 (Ω) = b(v, q) = B ∗ q, v
H −1 (Ω)×H01 (Ω) .
Then we can rewrite (2.26) and (2.27) in the form
Au + B ∗ p − B1 ϕ = f , A1 ϕ = l. (2.29)
Setting x = (u, p, ϕ), X0 = H̃ 1 (Ω) × L20 (Ω), Y0 = H −1(Ω) × L20 (Ω) ×
1/2
H̃ (Γ),
X = H̃ 1 (Ω) × L20 (Ω) × H 1 (Ω),
Y = H −1 (Ω) × L20 (Ω) × H̃ 1/2(Γ) × T ∗ × H 1/2 (ΓD ), (2.30)
we introduce an operator Φ = (Φ1 , Φ2 , Φ3 , Φ4 , Φ5 ) : X → Y that acts by the
formulas
Φ1 (x) = Φ̃1 (x) − B1 ϕ, Φ̃1 (x) = Au + B ∗ p, Φ2 (x) = div u, (2.31)

Φ3 (x) = γu, Φ4 (x) = A1 ϕ, Φ5 (x) = γ


ϕ.
ΓD
(2.32)

Then the following theorem holds true [3]:

Theorem 2.4. Let conditions (i)–(iii) and (vi) be fulfilled. Then the oper-
ators
(Φ1 , Φ2 , Φ3 ) : X0 → Y0 , (Φ4 , Φ5 ) : H 1 (Ω) → T ∗ × H 1/2 (ΓD ), Φ:X →Y
defined by formulas (2.31), (2.32) realize linear and continuous isomorphisms of
the corresponding spaces.

3. STATEMENT AND SOLVABILITY OF INVERSE


EXTREMAL PROBLEMS
Problem 1 considered above contains a number of physical parameters and func-
tions describing the densities of thermohydrodynamical sources. To find its so-
lution we need to know these parameters and the source densities. At the same
444 G. V. Alekseev, E. A. Adomavichus

time the true information about the sources may be unknown. In such cases it
is necessary to solve inverse problems for the model (1.1)–(1.3), which consist
in recovering the parameters of unknown sources using the information about
the state of the environment. For example, we may choose the values ϕd (x)
of the concentration ϕ measured at the points of a set Q ⊂ Ω as information
about the solution. In this case the unknown sources are determined from the
condition ϕ(x) = ϕd (x), x ∈ Q. We may also consider a more general situ-
ation when the unknown sources are determined from a condition of the form
Λϕ(x) = ϕd (x), x ∈ Q, where Λ is some operator. For example, choosing a
surface S from Ω as Q and the derivative ∂ϕ/∂n along the external normal to S
as Λϕ, we obtain an inverse problem of finding the source densities from the
condition ∂ϕ/∂n = χS (x), x ∈ S. Here χS is a given (measured) function on S.
Another type of information is used in extremal inverse problems of ad-
mixture diffusion theory. In these problems a certain objective functional J is
introduced and it is required to minimize it, for example, by a proper choice
of the admixture sources. It should be noted that solution of inverse problems
can be reduced to solution of extremal problems under appropriate choice of
objective functionals. With this in view, we shall study in detail the extremal
problems for the model (1.1)–(1.3), where as possible controls we shall consider
the densities f , ψ, and χ of the volume or boundary admixture sources and also
the function g in the boundary condition for the velocity in (1.3). As was noted
in Section 2, we assume that v ≡ (f, ψ, χ, g) ∈ K ≡ K1 × K2 × K3 × K4 , where
the sets Ki , i = 1, . . . , 4, are defined in (v).
Let J˜ : X ≡ H̃ 1 (Ω) × L20 (Ω) × H 1 (Ω) → R be a weakly lower semicontinuous
functional; and let μi ≥ 0, i = 1, . . . , 4, be some constants. Introduce an
auxiliary functional J : X × K → R by the formula
μ1 μ2 μ3 μ4
J(x, v) = J(x) ˜ + f 2T ∗ + ψ21/2,ΓD + χ2ΓN + g21/2,Γ . (3.1)
2 2 2 2
In addition to condition (v) we suppose that the following conditions hold: (ai ,
i = 1, . . . , 4) μi > 0 and J˜ is bounded below, or μi ≥ 0 and Ki is a bounded
set.
Considering the functional J on weak solutions of Problem 1, we write
the corresponding constraint, which has the form of the weak formulation
(2.12)–(2.14) of Problem 1, as follows:
F (x, v) ≡ F (u, p, ϕ, f, ψ, χ, g) = 0. (3.2)
Here F ≡ (F1 , F2 , F3 , F4 , F5 ) : X ×K → Y is the operator acting by the formulas
F1 (x, v), v
≡ νa(u, v) + c(u, u, v) + b(v, p) − b1 (ϕ, v) − f , v
, ∀v ∈ H01 (Ω),
F2 (x, v), q
≡ b(u, q), ∀q ∈ L20 (Ω), F3 (x, v) ≡ γu − g,
F4 (x, v), S
≡ a1 (ϕ, S) + c1 (u, ϕ, S) − f, S
− (χ, S)ΓN , ∀S ∈ T ,

F5 (x, v) ≡ γ
ϕ − ψ.
ΓD
(3.3)
Let us study the following problem of conditional minimization:
J(x, v) ≡ J(u, p, ϕ, f, ψ, χ, g) → inf, F (x, v) = 0, (x, v) ∈ X × K. (3.4)
Admixture diffusion in viscous fluids 445

The following examples of the functional J are of the greatest interest:


 
1 2 ν
J0 (x) = |∇u| dΩ, J1 (x) = |grad u + (grad u) |2 dΩ,
2 Ω 2 Ω

1 1
J2 (x) = r|u − ud |2 dΩ, J3 (x) =  rot u − ζd 2L2 (Ω) ,
2 Ω 2
 
1 1
J4 (x) = |∇ϕ|2 dΩ, J5 (x) = r|ϕ − ϕd |2 dΩ,
2 Ω 2 Ω

J6 (x) = rϕ dΩ,
Ω

where r ∈ L2+ (Ω) is a non-negative weight function. For the physical sence of
the functionals Jk , k = 0, . . . , 5, we refer to [3, 5, 11, 19]. The functional J6
with r = ρQ , where ρQ is the characteristic function of a subdomain Q ⊆ Ω,
describes the amount of the admixture that enters the domain Q in unit time.
This functional is widely applied for solving the problem of optimal location
of plants and factories in the neighborhood of ecologically significant zones.
Detailed statements of the corresponding problems and an effective method of
their solution for linear non-stationary equations of admixture diffusion can be
found in Marchuk’s book [26] and also in [15, 27, 28].
We note that each of the functionals Jk , k = 0, . . . , 5, is non-negative,
bounded below, weakly lower semicontinuous and independent of p. Thus, all
the statements, that will be proved for a weakly lower semicontinuous func-
tional J˜ that is bounded and independent of p , will hold true for each of
J0 , . . . , J5 . Let us introduce a set Zad = {(x, v) ∈ X × K | F (x, v) = 0,
J(x, v) < ∞} of admissible pairs for the problem (3.4).

Theorem 3.1. Let J˜ : X → R be a weakly lower semicontinuous functional


independent of p, (i)–(v) and (a1 )–(a4 ) hold, and the set Zad be nonempty.
Then there exists at least one solution of the problem (3.4)
Proof. Let (xm , vm ) = (um , pm , ϕm , fm , ψm , χm , gm ) ∈ Zad be a minimiz-
ing sequence for which

lim J(xm , vm ) = inf J(xm , vm ) ≡ J ∗ .


m→∞ (xm ,vm )∈Zad

Conditions (a1 )–(a4 ) and the form of the functional J imply that all the con-
trols vm are bounded by a constant independent of m. Hence, by Theorem 2.1
and Remark 2.1, the following estimates hold:

fm T ∗ ≤ c1 , ψm 1/2,ΓD ≤ c2 , χm ΓN ≤ c3 ,


gm 1/2,Γ ≤ c4 , um 1 ≤ c5 , ϕm 1 ≤ c6 . (3.5)

Here ci , ..., c6 , i = 1, . . . , 6, are some constants independent of m. It follows


from (3.6) that there exist subsequences {fmk }, {ψmk }, {χmk }, {gmk }, {umk },
and {ϕmk } of the sequences {fm }, {ψm }, {χm }, {gm }, {um }, and {ϕm } such
446 G. V. Alekseev, E. A. Adomavichus

that

fmk → f ∗ ∈ T ∗ weakly in T ∗ ,
ψmk → ψ∗ ∈ H 1/2 (ΓD ) weakly in H 1/2 (ΓD ),
χmk → χ∗ ∈ L2 (ΓN ) weakly in L2 (ΓN ),
gmk → g ∗ ∈ H̃ 1/2 (Γ) weakly in H 1/2 (Γ),
umk → u∗ ∈ H 1 (Ω) weakly in H 1 (Ω) and strongly in L4 (Ω),
ϕmk → ϕ∗ ∈ H 1 (Ω) weakly in H 1 (Ω) and strongly in L4 (Ω). (3.6)

Since Ki , i = 1, . . . , 4, are convex closed sets in Hilbert spaces, therefore, they


are closed in weak topology as well. Therefore, f ∗ ∈ K1 , ψ∗ ∈ K2 , χ∗ ∈ K3 ,
and g ∗ ∈ K4 . Moreover, u ∈ H̃ 1 (Ω).
It remains to prove the existence of a function p∗ ∈ L20 (Ω) such that the
element x∗ = (u∗ , p∗ , ϕ∗ ) together with the control v ∗ = (f ∗ , ψ ∗ , χ∗ , g ∗ ) satisfies
the relation
F (x∗ , v ∗ ) = 0. (3.7)
To prove (3.8) we note that for each mk the functions umk and ϕmk satisfy the
identities

νa(umk , v) + c(umk , umk , v) = f , v


+ b1 (ϕmk , v), ∀v ∈ V , (3.8)
a1 (ϕmk , S) + c1 (umk , ϕmk , S) = fmk , S
+ (χmk , S)ΓN , ∀S ∈ T , (3.9)

2
b(umk , q) = 0, ∀q ∈ L0 (Ω), γumk = gmk ,

γ Γ ϕmk = ψmk . (3.10)


D

Pass the the limit in (3.11) for k → ∞. Using the continuity of the form b and
trace operators γ and γ|ΓD , we obtain

b(u∗ , q) = 0, ∀q ∈ L20 (Ω), γu∗ = g ∗ , γ


Γ ϕ∗ = ψ ∗ . (3.11)
D


The first condition in (3.12) means that div u = 0. Taking this into account
and also using (2.4), (2.5), and (3.7), we easily conclude that for k → ∞

c(umk , umk , v) = −c(umk , v, umk ) → −c(u∗ , v, u∗ ) = c(u∗ , u∗ , v),


c1 (umk , ϕmk , S) = −c1 (umk , S, ϕmk ) → −c1 (u∗ , S, ϕ∗ ) = c1 (u∗ , ϕ∗ , S).

Therefore, pass in (3.9), (3.10) to the limit for k → ∞, we obtain

νa(u∗ , v) + c(u∗ , u∗ , v) = f , v
+ b1 (ϕ∗ , v), ∀v ∈ V , (3.12)
∗ ∗ ∗ ∗ ∗
λa1 (ϕ , S) + c1 (u , ϕ , S) = f , S
+ (χ , S)ΓN , ∀S ∈ T . (3.13)

In virtue of the results in [18, p.25], identity (3.13) is valid if and only if there
exists a unique function p∗ ∈ L20 (Ω) such that

νa(u∗ , v) + c(u∗ , u∗ , v) + b(v, p∗ ) = f , v


+ b1 (ϕ∗ , v), ∀v ∈ H01 (Ω). (3.14)
Admixture diffusion in viscous fluids 447

Equations (3.12), (3.14), (3.15) verify equality (3.8). Finally, since the func-
tional J is weakly lower semicontinuous on the set X × K, we have

J ∗ = lim J(xmk , vmk ) = lim J(xmk , vmk ) ≥ J(x∗ , v ∗ ) ≥ J ∗ .


k→∞ k→∞

The theorem is proved.

Corollary 3.1. Under the assumptions of Theorem 3.1 there exists at least
one solution of the extremal problem (3.4) for J˜ = Jk , k = 0, . . . , 5.

It is more difficult to investigate the extremal problem in the case of the


functional J6 because it is unbounded below. But J6 is certainly bounded
below in the case where the component ϕ of a solution to Problem 1 is non-
negative for all elements v ∈ K. Sufficient conditions for the nonnegativity
of the concentration were presented in, e. g., [16, 23]. Moreover, in virtue of
Remark 2.1, the functional J6 may be considered bounded below in the case
where all the sets Ki in (v) are bounded. Thus, we obtain the following result:

Corollary 3.2. 1) Let conditions (i)–(v), (a1 )–(a4 ), and the condition r ∈
L2+ (Ω) be fulfilled; and let the component ϕ of a solution (u, p, ϕ) to Problem 1
be nonnegative on the set K. Then the problem (3.4) for J˜ = J6 has a solution
(x, v) ∈ X × K. 2) Let conditions (i)–(v) be fulfilled; let Ki be bounded sets
and μi ≥ 0, i = 1, . . . , 4; and let r ∈ L2+ (Ω). Then there exists at least one
solution (x, v) ∈ X × K to the problem (3.4) for J˜ = J6 .

Parallel with the general multiparameter extremal problem (3.4), “partic-


ular” problems containing a fewer number of controls also play an important
role. For example, consider the case when the boundary function g for the ve-
locity u on the boundary Γ is the sole control. This case may be considered as
a special case of the problem (3.4) when K1 = {f0 }, K2 = {ψ0 }, K3 = {χ0 },
and μ1 = μ2 = μ3 = 0. Here f0 ∈ T ∗ , ψ0 ∈ H 1/2 (ΓD ), and χ0 ∈ L2 (ΓN ) are
fixed elements. The next two statements follow from Theorem 3.1.

Theorem 3.2. Let conditions (i)–(iv), (v) for K4 , and (a4 ) hold; let J˜ :
X → R be a weakly lower semicontinuous functional independent of p; and let
the set Zad be nonempty. Then the extremal problem

μ4
˜
J(x, v) ≡ J(x) + g21/2,Γ → inf, F (x, v) = 0, x ∈ X, (3.15)
2

where v = (f0 , ψ0 , χ0 , g) and g ∈ K4 , has at least one solution.


448 G. V. Alekseev, E. A. Adomavichus

Corollary 3.3. Let conditions (i)–(iv), (v) for K4 , and (a4 ) hold. Then the
problem (3.16) with J˜ = Jk , k = 0, . . . , 5 has at least one solution.
Let us now suppose in addition that

ψ0 = 0 on ΓD , χ0 = 0 on ΓN , f0 = 0 in Ω. (3.16)

Then it follows from Theorem 2.1 that ϕ = 0 and, therefore, only proceeds in
the domain Ω a hydrodynamical process described by Problem 2. In this case
the problem (3.4) becomes a purely hydrodynamical extremal problem
˜ p) + (μ4 /2)g2 F 0 (x, g) = 0,
J(x, g) ≡ J(u, 1/2,Γ → inf,

(x, g) ≡ (u, p, g) ∈ X0 × K4 . (3.17)

Here x ≡ (u, p); F 0 = (F10 , F2 , F3 ) : X0 → Y0 is the operator correspond-


ing to Problem 2; X0 = H̃ 1 (Ω) × L20 (Ω); Y0 = H −1 (Ω) × L20 (Ω) × H̃ 1/2 (Γ);
F10 (x, v), v
= νa(u, v) + c(u, u, v) + b(v, p) − f , v
; and the operators F2
0
and F3 are defined in (3.3). Let Zad = {(x, g) ∈ X0 × K4 | F 0 (x, g) = 0,
J(x, g) < ∞}. Theorem 3.2. implies the following two assertions:

Theorem 3.3. Let conditions (i), (iv), (v) for K4 , and (a4 ) hold; let J˜ :
X → R be a weakly lower semicontinuous functional independent of p; and let
0
the set Zad be nonempty. Then the extremal problem (3.18) has at least one
solution.

Corollary 3.4. Let conditions (i), (iv), (v) for K4 , and (a4 ) hold. Then for
J˜ = Jk , k = 0, . . . , 3, the extremal problem (3.18) has at least one solution.
We have proved the existence of solutions of the extremal inverse problems.
Our next aim is to study the question of uniqueness of their solutions. The
issue is sufficiently complicated and requires additional techniques that will
be developed in Section 4. We shall justify the application of the Lagrange
multiplier principle to the solution of the general extremal problem (3.4) and
derive the corresponding optimality systems both for an arbitrary differentiable
objective functional and for concrete functionals. Derivation of the optimality
system pursues a double aim. On the one hand, analysis of the optimality
systems allows to reveal new important properties of solutions to the inverse
problem. On the other hand, the uniqueness of a solution to the extremal
inverse problem for a concrete objective functional (for J = J0 ) will be proved
in Section 5 exactly on the base of the properties of the optimality system.

4. DERIVATION OF THE OPTIMALITY SYSTEM

4.1. Justification of the Lagrange principle


In this section we derive the optimality system for the problem (3.4). We
shall use the extremal principle in smoothly convex problems of conditional
minimization [22]. For this purpose we calculate the Fréchet derivatives of
Admixture diffusion in viscous fluids 449

the operator F with respect to x. A trivial analysis shows that the Fréchet
derivative of the operator F : X × K → Y with respect to x at each
point (x̂, v̂) ≡ (û, p̂, ϕ̂, fˆ, ψ̂, χ̂, ĝ) ∈ X × K is a linear continuous operator
Fx (x̂, v̂) ≡ (F1x
 
(x̂, v̂), F2x 
(x̂, v̂), F3x 
(x̂, v̂), F4x 
(x̂, v̂), F5x (x̂, v̂)) : X → Y de-
fined on elements x ≡ (w, r, τ ) ∈ X by the relations

F1x (x̂, v̂), v
= νa(w, v) + c(û, w, v) + c(w, û, v) + b(v, r) − b1 (τ, v),
∀v ∈ H01 (Ω),

F2x (x̂, v̂)x, q
= b(w, q), ∀q ∈ L20 (Ω), 
F3x (x̂, v̂)x = γw,

F4x (x̂, v̂)x, S
= a1 (τ, S) + c1 (û, τ, S) + c1 (w, ϕ̂, S), ∀S ∈ T ,

F5x
(x̂, v̂)x = γ
τ.
ΓD
(4.1)

Denote by F (x̂, K) the image, of the set K, in the space Y under action of
the operator F (x̂, · ) . By (3.3), the operator F is linearly dependent on the
controls, which implies that the set F (x̂, K) is convex in Y for each fixed x̂ ∈ X.
Let y ∗ = (ξ, σ, ζ, η, ζ1 ) ∈ Y ∗ = H01 (Ω) × L20 (Ω) × (H̃ 1/2 (Γ))∗ × T ×
(H 1/2 (ΓD ))∗ . For the functional J we introduce the Lagrangian L : X × K ×
R × Y ∗ → R by the formula

L(x, v, λ0 , y ∗ ) = λ0 J(x, v) + y ∗ , F (x, v)

≡ λ0 J(x, v) + F1 (x, v), ξ


H −1 (Ω)×H01 (Ω) + (F2 (x, v), σ)
+ ζ, F3 (x, v)
Γ + F4 (x, v), η
T ∗ ×T + ζ1 , F5 (x, v)
ΓD . (4.2)

Here and subsequently, ζ, w


Γ = ζ, γw
(H̃ 1/2 (Γ))∗ ×H̃ 1/2 (Γ) , ζ1 , τ
ΓD =
ζ1 , τ
(H 1/2 (ΓD ))∗ ×H 1/2 (ΓD ) . Let R+ = [0, ∞). Let us formulate the main re-
sult of this section.

Theorem 4.1. Let the conditions (i)–(iv) hold, let K1 ⊂ T ∗ , K2 ⊂


H (ΓD ), K3 ⊂ L2 (ΓN ), and K4 ⊂ H̃ 1/2 (Γ) be nonempty convex sets; let
1/2

(x̂, v̂) ∈ X × K be an element on which a local minimum in the problem (3.4)


be reached; and let the functional J(x, · ) : K → R be convex for each point
x ∈ X. Moreover, we assume that the function x → Jx (x, v) with values in X ∗
belongs to the class C 0 at the point x̂ for each element v ∈ K. Then there exists
a nonzero Lagrange multiplier (λ0 , y ∗ ) = (λ0 , ξ, σ, ζ, η, ζ1 ) ∈ R+ × Y ∗ such that
the following Euler – Lagrange equation is valid:

λ0 Jx (x̂, v̂), (w, r, τ )


X ∗ ×X + y ∗ , Fx (x̂, v̂)(w, r, τ )
Y ∗ ×Y = 0,
∀(w, r, τ ) ∈ X, (4.3)

and the minimum principle holds:

L(x̂, v̂, λ0 , y ∗ ) ≤ L(x̂, v, λ0 , y ∗ ), ∀v ∈ K. (4.4)

Proof. The proof is based on a theorem from [22, p. 79]. By that theorem,
to prove the existence of the Lagrange multiplier (λ0 , y ∗ ) it suffices to prove
450 G. V. Alekseev, E. A. Adomavichus

the Fredholm property of the operator Fx (x̂, v̂) : X → Y , taking into account
the convexity of the sets K and F (x̂, K) ⊂ Y . Comparing relations (4.1) for
the operator F (x̂, v̂) with relations (2.31) and (2.32) that define the operator
Φ = (Φ1 , Φ2 , Φ3 , Φ4 , Φ5 ) : X → Y , we can see that the following representation
is valid:
Fx (x̂, v̂) = Φ + Φ̂ ≡ Φ + (Φ̂1 , 0, 0, Φ̂4, 0).
Here the operators Φ̂1 : H̃ 1 (Ω) → H −1 (Ω) and Φ̂4 : H̃ 1 (Ω) → T ∗ are defined
by the relations
Φ̂1 (w, r, τ ), v
= c(w, û, v), ∀v ∈ H01 (Ω),
Φ̂4 (w, r, τ ), S
= c1 (w, ϕ̂, S), ∀S ∈ T .
By Theorem 2.4, the operator Φ : X → Y is an isomorphism; and by the esti-
mates (2.6), the operator (Φ̂1 , Φ̂4 ) : L4 (Ω) → H −1 (Ω) × T ∗ is continuous. Since
the space H̃(Ω) is compactly embedded in L4 (Ω) in virtue of the embedding
theorem, therefore, the operator (Φ̂1 , Φ̂4 ) acting from H̃ 1 (Ω) into H −1(Ω) × T ∗
is compact. In this case the operator Fx (x̂, v̂) is a Fredholm operator as a sum
of an isomorphism and a compact operator [32].
Let us consider the minimum principle (4.4). Taking into account the char-
acter of the dependencies of the operator F and functional J on (f, ψ, χ, g), we
rewrite (4.4) in the form of the variational inequality

ζ, g − ĝ
Γ + ζ1 , ψ − ψ̂
ΓD + (χ − χ̂, η)ΓN + f − fˆ, η
T ∗ ×T
≤ λ0 [J(x̂, v) − J(x̂, v̂)], ∀v ∈ K. (4.5)
Taking into account (4.1) and (4.2), we rewrite equation (4.3) in the form

λ0 Jx (x̂, v̂), (w, r, τ )


X ∗ ×X + νa(w, ξ) + c(û, w, ξ) + c(w, û, ξ)
+ b(ξ, r) − b1 (τ, ξ) + b(w, σ) + a1 (τ, η) + c1 (û, τ, η) + c1 (w, ϕ̂, η)
+ ζ, w
Γ + ζ1 , τ
ΓD = 0, ∀(w, r, τ ) ∈ X. (4.6)
Suppose that J does not depend on p. Then we have

Jx (x̂, v̂), (w, r, τ )


X ∗ ×X
= Ju (x̂, v̂), w
(H̃ 1 (Ω))∗ ×H̃ 1 (Ω) + Jϕ (x̂, v̂), τ
(H 1 (Ω))∗ ×H 1 (Ω) .

Setting in (4.6) at first τ = 0, r = 0, then r = 0, w = 0, and finally w = 0,


τ = 0, we obtain the following relations for determination of the Lagrange
multipliers ξ, σ, ζ, η, and ζ1 :

νa(w, ξ) + c(û, w, ξ) + c(w, û, ξ) + c1 (w, ϕ̂, η) + b(w, σ) + ζ, w


Γ
+ λ0 Ju (x̂, v̂), w
(H̃ 1 (Ω))∗ ×H̃ 1 (Ω) = 0, ∀w ∈ H̃ 1 (Ω), (4.7)

a1 (τ, η) + c1 (û, τ, η) − b1 (τ, ξ) + ζ1 , τ


ΓD
+ λ0 Jϕ (x̂, v̂), τ
(H 1 (Ω))∗ ×H 1 (Ω) = 0, ∀τ ∈ H 1 (Ω), (4.8)
Admixture diffusion in viscous fluids 451

b(ξ, r) = 0, ∀r ∈ L20 (Ω). (4.9)


Relations (4.7)–(4.9), the minimum principle (4.4) which is equivalent to the
variational inequality (4.5), and the operator constraint (3.2) which is equivalent
to (2.12)–(2.14) constitute the optimality system. It consists of three parts. The
first one has the form of the weak formulation (2.12)–(2.14) of Problem 1; the
second consists of identities (4.7)–(4.9) for the multipliers ξ, σ, ζ, η, and ζ1 ; and
the third is the variational inequality (4.5) for the control v.

4.2. Differential equations and boundary conditions


for the Lagrange multipliers
Let us derive differential equations and boundary conditions which are satisfied
(at least formally) by the Lagrange multipliers. For this purpose we take the
contraction of relations (4.7) and (4.8) on the spaces H01 (Ω) ⊂ H̃ 1 (Ω) and
H01 (Ω) ⊂ H 1 (Ω), respectively. Using (2.4), (2.5), (2.9), and the formulas of
integration by parts, we can easily show that

d 
∂ ûj
c(w, û, ξ) ≡ wi ξj dΩ
i,j=1 Ω ∂xi
d   
∂ ûj 
= ξj wi dΩ = (∇û · ξ) · w dΩ,
Ω ∂xi
i,j=1 Ω

νa(w, ξ) + c(û, w, ξ) + c(w, û, ξ) + c1 (w, ϕ̂, η) + b(w, σ)



= [ν∇w : ∇ξ − (û · ∇)(ξ · w) + (∇û · ξ) · w + (w · ∇ϕ̂)η − div wσ] dΩ
Ω
= −νΔξ − (û · ∇)ξ + ∇û · ξ + η∇ϕ̂ + ∇σ, w
H −1 (Ω)×H01 (Ω) , ∀w ∈ H01 (Ω),
(4.10)

a1 (τ, η) + c1 (û, τ, η) − b1 (τ, ξ)


  ∂τ 
=λ ∇τ · ∇η dΩ − w0 , η + (κτ, η) − (û · ∇η, τ ) − (bτ, ξ)
Ω ∂z
 ∂η 
= − λΔη + w0 + κη − û · ∇η − b · ξ, τ ,
∂z H −1 (Ω)×H01 (Ω)

∀H01 (Ω). (4.11)

Here wi , uj , and ξj are the Cartesian components of the vectors w, u, and ξ,


respectively. Obviously,

ζ, w
Γ = 0, ∀w ∈ H01 (Ω), ζ1 , τ
ΓD = 0, ∀τ ∈ H01 (Ω).

Introduce linear continuous operators SH : (H̃ 1 (Ω))∗ → H −1 (Ω), SH :


452 G. V. Alekseev, E. A. Adomavichus

(H 1 (Ω))∗ → H −1 (Ω) that act by the formulas

SH l, w
H −1 (Ω)×H01 (Ω) = l, w
(H̃ 1 (Ω))∗ ×H̃ 1 (Ω) ,
∀w ∈ H01 (Ω) ⊂ H̃ 1 (Ω), l ∈ (H̃ 1 (Ω))∗ ,
SH l, S
H −1 (Ω)×H01 (Ω) = l, S
(H 1 (Ω))∗ ×H 1 (Ω) ,
∀S ∈ H01 (Ω) ⊂ H 1 (Ω), l ∈ (H 1 (Ω))∗ .

Using the above operators, we consider the contractions of relations (4.7) and
(4.8) on the spaces H01 (Ω) and H01 (Ω), respectively. Taking into account (4.10),
and (4.11), and the fact that equality (4.9) equivalent to the equation div ξ = 0
in Ω, we conclude that the elements ξ, σ, and η satisfy the equations

−νΔξ − (û · ∇)ξ + ∇û · ξ + ∇σ + η∇ϕ̂ = −λ0 SH Ju (û, v̂) in H −1 (Ω),
div ξ = 0 in Ω, (4.12)
∂η
−λΔη − û · ∇η + w0 + κη − b · ξ = −λ0 SH Jϕ (x̂, v̂) in H −1(Ω). (4.13)
∂z

Let us formulate the obtained result.

Theorem 4.2. Let the assumptions of Theorem 4.1 hold and the func-
tional J be independent of p. Then there exist functions ξ ∈ H01 (Ω), σ ∈ L20 (Ω),
ζ ∈ (H̃ 1/2 (Γ))∗ , η ∈ T , and ζ1 ∈ (H 1/2 (ΓD ))∗ and a constant λ0 ≥ 0 which
together with the solution (x̂, v̂) = (û, p̂, ϕ̂, fˆ, ψ̂, χ̂, ĝ) of the extremal prob-
lem (3.4) satisfy equations (4.12) and (4.13), the integral identities (4.7)–(4.9),
and the minimum principle (4.4).
If we assume that ξ, σ, and η have higher smoothness properties, for example,

ξ ∈ H 1 (Δ, Ω) ∩ H01 (Ω), η ∈ H 1 (Δ, Ω) ∩ T , σ ∈ H 1 (Ω), (4.14)

then from (4.7), (4.8) and (4.12), (4.13) we can obtain “pointwise” boundary
relations for the Lagrange multipliers. Indeed, if conditions (4.14) hold, then
the left-hand sides of (4.12) and (4.13) and, hence, the right-hand sides belong
to the spaces L6/5 (Ω) and L6/5(Ω), respectively. Thus, we can multiply (4.12)
and (4.13) by the functions w ∈ H̃ 1 (Ω) ⊂ L6 (Ω) and τ ∈ H 1 (Ω) ⊂ L6 (Ω),
respectively, integrate them over the domain Ω, and apply the Green formulas
(2.3) and (2.2). Subtracting the obtained relations from (4.7) and (4.8) and
taking into account (2.9), we get

∂ξ
ζ, w
Γ − σn · w dΓ + ν ,w Γ
Γ ∂n

= λ0 SH Ju (x̂, v̂) · w dΩ − λ0 Ju (x̂, v̂), w
(H̃ 1 (Ω))∗ ×H̃ 1 (Ω) ,
Ω
∀w ∈ H̃ 1 (Ω), (4.15)
Admixture diffusion in viscous fluids 453


∂η
ζ1 , τ
ΓD + λ(ατ, η)ΓN − w0 n3 τ η dΓ + λ ,τ Γ
ΓN ∂n

= λ0 SH Jϕ (x̂, v̂)τ dΩ − λ0 Jϕ (x̂, v̂), τ
(H 1 (Ω))∗ ×H 1 (Ω) ,
Ω
∀τ ∈ H 1 (Ω). (4.16)

Further derivation of the boundary relations depends on the form of the func-
tional J.
At first we consider the case when J = J0 , where J0 is defined in (3.5). In
 
this case J0ϕ = 0 and the derivative J0u (x̂, v̂) : H̃ 1 (Ω) → R is given by the
formula


J0u (x̂, v̂), w
(H̃ 1 (Ω))∗ ×H̃ 1 (Ω) = ∇û : ∇w dΩ = a(û, w), ∀w ∈ H̃ 1 (Ω).
Ω
(4.17)
Assuming in addition to (4.14) that û ∈ H 1 (Δ, Ω) and ϕ̂ ∈ H 1 (Δ, Ω) and using
the Green formula (2.3), from (4.17) we derive


SH J0u (x̂, v̂), w
(H̃ 1 (Ω))∗ ×H 1 (Ω) = − Δû · w dΩ, ∀w ∈ H01 (Ω),
Ω


which yields SH J0u (x̂, v̂) = −Δû ∈ L2 (Ω). Using the last fact and applying
once more the Green formula (2.3), we come to the relation
 

∂ û
SH J0u (x̂, v̂)·w dΩ ≡− Δû·w dΩ = a(û, w)− , w Γ, ∀w ∈ H̃ 1 (Ω).
Ω Ω ∂n

As a result, relations (4.12), (4.13), (4.15), and (4.16) take the form

−νΔξ − (û · ∇)ξ + ∇û · ξ + ∇σ + η∇ϕ̂ = λ0 Δû in Ω,


div ξ = 0 in Ω, (4.18)
∂η
−λΔη − û · ∇η + w0 + κη − b · ξ = 0 in Ω, (4.19)
∂z
∂ξ ∂ û
ζ − σn + ν + λ0 , w Γ = 0, ∀w ∈ H̃ 1 (Ω), (4.20)
∂n ∂n
∂η
λ , τ Γ + ζ1 , τ
ΓD + (λαη − w0 n3 η, τ )ΓN = 0, ∀τ ∈ H 1 (Ω). (4.21)
∂n

When the test function w runs through the space H̃ 1 (Ω) its trace γw runs
through the whole of the space H̃ 1/2 (Γ). Taking this into account, from (4.20)
we obtain the following expression for the Lagrange multiplier ζ:

∂ξ ∂ û
ζ = σn − ν − λ0 in (H̃ 1/2 (Γ))∗ . (4.22)
∂n ∂n

Analogously, from (4.21) in virtue of the arbitrariness of the function τ ∈ H 1 (Ω)


454 G. V. Alekseev, E. A. Adomavichus

we obtain the following boundary relations for η and ζ1 :


 ∂η


λ
+ αη − w0 n3 η = 0 in H −1/2 (ΓN ),
∂n ΓN
∂η

ζ1 = −λ
in H −1/2 (ΓD ). (4.23)
∂n ΓD
For given functions û and ϕ̂, relations (4.18), (4.19), (4.22), and (4.23) together
with the conditions

ξ = 0 on Γ, η=0 on ΓD (4.24)

constitiute a closed system of equations (the second part of the optimality sys-
tem for the functional J0 ) with respect to the Lagrange multipliers ξ, σ, ζ,
η, ζ1 .
Analogous arguments for the functional J4 result in the relations

−νΔξ − (û · ∇)ξ + ∇û · ξ + ∇σ + η∇ϕ̂ = 0 in Ω,


div ξ = 0 in Ω, (4.25)
∂η
−λΔη − û · ∇η + w0 + κη − b · ξ = λ0 Δϕ̂ in Ω, (4.26)
∂z
 ∂η
 ∂ ϕ̂

λ
+ αη − w0 n3 η = −λ0
in H −1/2 (ΓN ),
∂n ΓN ∂n ΓN
 ∂η ∂ ϕ̂ 

ζ1 = − λ + λ0
in H −1/2 (ΓD ), (4.27)
∂n ∂n ΓD
∂ξ
ζ = σn − ν in (H̃ 1/2 (Γ))∗ , (4.28)
∂n
which together with (4.24) constitute the second part of the optimality system.
At last, for the functional J5 or J6 the second part of the optimality system is
determined by relations (4.25), (4.23), (4.28), (4.24), and

∂η
−λΔη − û · ∇η + w0 + κη − b · ξ = −λ0 r(ϕ̂ − ϕd ) in Ω (4.29)
∂z
or
∂η
−λΔη − û · ∇η + w0 + κη − b · ξ = −λ0 r in Ω. (4.30)
∂z
The optimality systems for the rest of the functionals in (3.5) can be found
in [3].

4.3. The optimality system in the case of linear Problem 1


Analogously we can derive the optimality system for the case when the condi-
tional constraint for the extremal problem (3.4) has the form of the weak for-
mulation (2.26), (2.27), and (2.30) of the boundary value problem for a linear
system of equations of admixture diffusion. To derive the system it is suffi-
cient to write these relations in the form (3.2), where F = (F1 , F2 , F3 , F4 , F5 ) :
Admixture diffusion in viscous fluids 455

X × K → R is the operator corresponding to the linear problem, to compose the


Lagrangian L by formula (4.2), and then to apply Theorem 4.1. The optimality
system obtained will consist of the equations

νa(û, v) + c(u0 , û, v) + b(v, p̂) − b1 (ϕ̂, v) − f , v


= 0, ∀v ∈ H01 (Ω), (4.31)

a1 (ϕ̂, S) + c1 (u0 , ϕ̂, S) − fˆ, S


− (χ̂, S)ΓN = 0, ∀S ∈ T , γ
ΓD ϕ̂ = ψ̂,
(4.32)
b(û, q) = 0, ∀q ∈ L20 (Ω), γ û = ĝ, (4.33)
the minimum principle (4.4), identity (4.9), and the identities

νa(w, ξ) + c(u0 , w, ξ) + b(w, σ) + ζ, w


Γ + λ0 Ju (x̂, v̂), w
= 0,
∀w ∈ H̃ 1 (Ω), (4.34)

a1 (τ, η) + c1 (u0 , τ, η) − b1 (τ, ξ) + ζ1 , τ


ΓD + λ0 Jϕ (x̂, v̂), τ
= 0,
∀τ ∈ H 1 (Ω) (4.35)

for (û, p̂, ϕ̂, fˆ, ψ̂, χ̂, ĝ) and the Lagrange multiplier (ξ, σ, ζ, η, ζ1 ). The last iden-
tities are linear analogues of identities (4.7) and (4.8). Further transformation
of the system (4.34), (4.35) by the above scheme leads to the linear differential
equations

−νΔξ − (u0 · ∇)ξ + ∇σ = −λ0 SH Ju (x̂, v̂) in H −1 (Ω),


div ξ = 0 in Ω, (4.36)
∂η
−λΔη − u0 · ∇η + w0 + κη − b · ξ = −λ0 SH Jϕ (x̂, v̂) in H −1 (Ω) (4.37)
∂z
with respect to ξ, σ, and η and also to the boundary relations for the functions ξ,
σ, ζ, η, and ζ1 which exactly coincide with (4.15) and (4.16).
In the particular case where J does not depend on u, the multipliers ξ, σ,
and ζ are equal to zero

ξ = 0 in Ω, σ=0 in L20 (Ω), ζ = 0 in (H̃ 1/2 (Γ))∗ . (4.38)

Really, setting in (4.34) w = v ∈ H01 (Ω) and attaching relations (4.9), in virtue
of the assumption Ju (x̂, v̂) = 0 we obtain the following problem with respect
to ξ and σ:

νa(ξ, v) + c(u0 , v, ξ) + b(v, σ) = 0, ∀v ∈ H01 (Ω),


b(ξ, r) = 0, ∀r ∈ L20 (Ω), ξ=0 on Γ. (4.39)

Applying Theorem 2.4 to the homogeneous problem (4.39), we conclude that


ξ = 0 and σ = 0. Substituting ξ = 0, σ = 0, Ju (x̂, v̂) = 0 into (4.34), we obtain
ζ = 0.
456 G. V. Alekseev, E. A. Adomavichus

Application of this scheme to the case of the functional J0 leads to the same
boundary relations (4.22)–(4.24) as in the nonlinear case and to the following
linear differential equations with respect to ξ, σ, and η:

−νΔξ − (u0 · ∇)ξ + ∇σ = λ0 Δû in Ω, div ξ = 0 in Ω, (4.40)


∂η
−λΔη − u0 · ∇η + w0 + κη − b · ξ = 0 in Ω. (4.41)
∂z

However, for the functional J5 (it is independent of u) the second part of the
optimality system is determined by the relations (4.38) and

∂η
−λΔη − u0 · ∇η + w0 + κη = −λ0 r(ϕ̂ − ϕd ) in Ω,
∂z
η = 0 on ΓD , (4.42)
 ∂η


λ
+ αη − w0 n3 η = 0 in H −1/2 (ΓN ),
∂n ΓN
∂η

ζ1 = −λ
in H −1/2 (ΓD ), (4.43)
∂n ΓD

whereas for the functional J6 it is determined by the relations (4.38), (4.43),


and
∂η
−λΔη − u0 · ∇η + w0 + κη = −λ0 r in Ω. (4.44)
∂z
Let us compare the optimality systems that were constructed in this sub-
section for the functionals J0 , J5 , and J6 . We shall denote them by (S0 ), (S5 ),
and (S6 ), respectively. The analysis of the system (S0 ) shows that it is a con-
nected (through the coefficients and boundary conditions) system of ten linear
differential equations together with the boundary relations with respect to the
initial (û, p̂, ϕ̂) and conjugate (ξ, σ, ζ, η, ζ1 ) states and the variational inequali-
ties for finding the desired controls fˆ, ψ̂, χ̂, and ĝ. As concerns the systems (S5 )
and (S6 ), they are separated in some sense; these are systems with respect to
the basic state (û, p̂, ϕ̂), the Lagrange multipliers η, ζ1 , and the controls fˆ, ψ̂, χ̂,
ĝ. This is connected with the fact that the corresponding problems for determi-
nation of a pair of conjugate variables (ϕ̂, η) do not contain the hydrodynamical
variables. So we can solve them independently of these values. Moreover, in the
case of the system (S6 ), in order to find the optimal value J6min one does not
need to determine the solution ϕ̂ of the direct problem (4.32). It is sufficient
only to find the conjugate state η by solving the elliptic boundary value prob-
lem for η in (4.44), (4.43), to calculate the multiplier ζ1 from the last relation
in (4.43), and to determine the controls fˆ, ψ̂, and χ̂ from (4.5). To make sure
of this we set S = η in relation (4.32), J = J6 and τ = ϕ̂ in (4.35) and then

subtract them. Since J6ϕ (x̂, ϕ̂), τ
= (r, τ ), we obtain the relation

−λ0 rϕ̂ dΩ = fˆ, η
+ (χ̂, η)ΓN + ζ1 , ψ̂
ΓD .
Ω
Admixture diffusion in viscous fluids 457

Assuming for clearness that fˆ ∈ L2 (Ω), ζ1 ∈ L2 (ΓD ), and λ0 = 1 (see Sec-


tion 5.1), we get
   
min ˆ
J6 ≡ rϕ̂ dΩ = − f η dΩ − χ̂η dΓ − ζ1 ψ̂ dΓ. (4.45)
Ω Ω ΓN ΓD

Formula (4.45) gives an explicit representation of the minimum of the func-


tional J6 in terms of the values of the Lagrange multipliers η, ζ1 and the con-
trols fˆ, ψ̂, and χ̂ satisfying (4.5) for ζ = 0. It is analogous to the formula
obtained in [26] when solving the problem of optimal location of plants near
ecologically significant zones. In [26] an economical approach to solving that
problem was developed. That approach is based exactly on minimization of the
functional J6 .
We note, however, that this approach can be applied only to the func-
tional J6 , which is linear with respect to ϕ, and only if the admixture diffusion
is described by linear equations. If the objective functional is nonlinear, then
to find its minimum we have to solve the complete optimality system. It is
a very difficult problem even when the linear model of admixture diffusion is
considered.

4.4. The optimality system for the Navier-Stokes equations


According to the results of Sections 2 and 3, the exremal problems for the
hydrodynamical Problem 2 may be considered as a special case of Problem 1
when conditions (3.17) hold. Using the notations of Section 2.2, we set y∗ =
(ξ, σ, ζ) ∈ Y0∗ ≡ H01 (Ω) × L20 (Ω) × (H̃ 1/2 (Γ))∗ and introduce the Lagrangian
L : X0 × K4 × R × Y0∗ → R for the problem (3.18) by the formula

L(x, g, λ0 , y ∗ ) = λ0 J(x, g) + y ∗ , F (x, g)


Y0∗ ×Y0 .

The following analogues of Theorems 4.1 and 4.2 are valid:

Theorem 4.3. Let conditions (i) and (iv) hold; let K4 ⊂ H̃ 1/2(Γ) be
a convex nonempty set; let (x̂, ĝ) ∈ X0 × K4 be an element on which the
local minimum in the problem (3.18) is reached, and let J(x, · ) : K4 → R
be a convex functional for each point x ∈ X0 . moreover, we assume that
the function x → Jx (x, g) with values in X0∗ belongs to the class C 0 at the
point x̂ for every element g ∈ K4 . Then there exists a nonzero multiplier
(λ0 , y ∗ ) = (λ0 , ξ, σ, ζ) ∈ R+ × Y0∗ such that the following Euler – Lagrange
equation is valid:

λ0 Jx (x̂, ĝ), (w, r)


X0∗ ×X0 + y ∗ , Fx (x̂, ĝ)(w, r)
Y0∗ ×Y0 = 0,
∀(w, r) ∈ X0 .
(4.46)
which in the case where J does not depend on p is equivalent to the identities

νa(w, ξ) + c(û, w, ξ) + c(w, û, ξ) + b(w, σ) + ζ, w


Γ
+ λ0 Ju (û, p̂, ĝ), w
(H̃ 1 (Ω))∗ ×H̃ 1 (Ω) = 0, ∀w ∈ H̃ 1 (Ω),
458 G. V. Alekseev, E. A. Adomavichus

b(ξ, r) = 0, ∀r ∈ L20 (Ω), (4.47)


and the following minimum principle (variational inequality) holds:
ζ, g − ĝ
Γ ≤ λ0 [J(x̂, g) − J(x̂, ĝ)], ∀g ∈ K4 . (4.48)
Theorem 4.4. Let the assumptions of Theorem 4.3 hold and the func-
tional J do not depend on p. Then there exist functions ξ ∈ H01 (Ω), σ ∈ L20 (Ω),
and ζ ∈ (H̃ 1/2 (Γ))∗ and a constant λ0 ≥ 0 which together with the solution
(û, p̂, ĝ) of the problem (3.18) satisfy identities (4.47), the minimum princi-
ple (4.48), the boundary relation (4.15), and the equations
−νΔξ − (û · ∇)ξ + ∇û · ξ + ∇σ = −λ0 SH Ju (û, ĝ) in H −1 (Ω),
div ξ = 0 in Ω. (4.49)
Using Theorems 4.3 and 4.4 and arguing as in Section 4.2, we can easily
derive the second part of the optimality system for each of the hydrodynamical
functionals J0 , . . . , J3 in (3.5). For example, for the functional J0 it has the
form
−νΔξ − (û · ∇)ξ + ∇û · ξ + ∇σ = λ0 Δû in H −1(Ω),
div ξ = 0 in Ω, (4.50)
∂ξ ∂ û
ξ=0 on Γ, ζ = σn − ν − λ0 in (H̃ 1/2 (Γ))∗ . (4.51)
∂n ∂n
Note that system (4.50) is obtained from (4.18) if we set ϕ̂ = 0 there.

5. UNIQUENESS OF SOLUTIONS OF THE EXTREMAL


INVERSE PROBLEMS
Establishment of sufficient conditions which guarantee the uniqueness of so-
lutions to the considered extremal inverse problems is a very interesting and
complex question. We shall investigate it for some special cases. At first we
consider the question whether the Lagrange multiplier is regular.

5.1. The positiveness of the Lagrange multiplier λ0


We can distinguish two different cases in Theorem 4.1: the regular case, when all
nontrivial Lagrange multipliers satisfying (4.3) and (4.4) are regular, i. e., have
the form (λ0 , y∗ ) with λ0 > 0; and the irregular case, when there exists at least
one Lagrange multiplier (0, y ∗ ), where y ∗ = 0, satisfying (4.3) and (4.4). In the
first case, replacing in (4.3) and (4.4) the Lagrange multiplier y ∗ by y ∗ /λ0 , we
may think that λ0 = 1. In the second case the Euler – Lagrange equation takes
the form of the relation
y ∗ , Fx (x̂, v̂)(w, r, τ )
Y ∗ ×Y = 0, ∀(w, r, τ ) ∈ X,
which is equivalent to the following equation homogeneous with respect to
y∗ ∈ Y ∗:
[Fx (x̂, v̂)]∗ y ∗ = 0. (5.1)
Admixture diffusion in viscous fluids 459

Here [Fx (x̂, v̂)]∗ : Y ∗ → X ∗ is the operator conjugate to the operator Fx (x̂, v̂) :
X → Y in (4.1).
We shall prove below the regularity of the Lagrange multiplier under the
assumption that a fixed element v0 = (ν, λ, w0 , f , b, κ, α) and the controls v ∈ K
satisfy the condition
1 1 β
N Mu (v0 , v) + N1 Mϕ (v0 , v) < 1, ∀v ∈ K, (5.2)
α0 ν α0 ν λ∗
which provides the uniqueness of the solution to Problem 1 in virtue of The-
orem 2.3. Here the constants Mu (v0 , v) and Mϕ (v0 , v) are defined by rela-
tions (2.18), (2.19). To prove the regularity it is sufficient to prove that equa-
tion (5.1) or, which is the same, the system (4.7)–(4.9) has only the trivial solu-
tion y ∗ = (ξ, σ, ζ, η, ζ1 ) for λ0 = 0. Assume the contrary, i. e., that there exists
a nontrivial solution (ξ, σ, ζ, η, ζ1 ) of the system (4.7)–(4.9) for λ0 = 0, where
the elements x̂ = (û, p̂, ϕ̂) and v̂ are connected by the relation F (x̂, v̂) = 0.
Then we set in this system w = ξ, r = σ, and τ = η and use the conditions
ξ = 0 on Γ and η = 0 on ΓD and the relations c(û, ξ, ξ) = 0, c1 (û, η, η) = 0,
and b(ξ, σ) = 0, which follow from (2.4), (2.5), and (4.9). As a result, we arrive
at the equalities

νa(ξ, ξ) + c(ξ, û, ξ) + c1 (ξ, ϕ̂, η) = 0, a1 (η, η) − b1 (η, ξ) = 0. (5.3)

Using the coerciveness inequalities (2.7) and (2.11) and relations (2.10),
from (5.3) we obtain the inequalities

α0 νξ21 ≤ N û1 ξ21 + N1 ξ1 ϕ̂1 η1 , λ∗ η21 ≤ β1 η1 ξ1 . (5.4)

From the second inequality in (5.4) we get η1 ≤ (β1 /λ∗ )ξ1 . Then the first
inequality yeilds
 β1 
α0 νξ21 ≤ N û1 + N1 ϕ̂1 ξ21 . (5.5)
λ∗
Since the elements û, p̂, and ϕ̂ are connected by the relation F (x̂, v̂) = 0,
therefore, the estimates (2.17) are valid for û and ϕ̂. These estimates and (5.2)
imply
β1 β1
N û1 + N1 ϕ̂1 ≤ N Mu (v0 , v) + N1 Mϕ (v0 , v) < α0 ν. (5.6)
λ∗ λ∗
From (5.4)–(5.6) we obtain ξ = 0 and η = 0, which together with (4.7) and (4.8)
for λ0 = 0 yeilds

b(w, σ) + ζ, w
Γ = 0, ∀w ∈ H̃ 1 (Ω), ζ1 , τ
ΓD = 0, ∀τ ∈ H 1 (Ω).

Using the arbitrariness of the functions w ∈ H̃ 1 (Ω) and τ ∈ H 1 (Ω) and the
inf-sup property (2.8) of the form b, we conclude that σ = 0, ζ = 0, and ζ1 = 0,
which contradicts the nontriviality of the element (ξ, σ, ζ, η, ζ1 ).
460 G. V. Alekseev, E. A. Adomavichus

The uniqueness of the regular multiplier (1, y ∗ ) satisfying the system (4.7)–
(4.9) for λ0 = 1, under the natural condition F (x̂, v̂) = 0 on the elements x̂
and v̂, follows from the absence of non-trivial solutions of the homogeneous for
(λ0 = 0) problem (4.7)–(4.9) and its Fredholm property (see Theorem 4.1). Let
us formulate the results obtained.

Theorem 5.1. Let the assumptions of Theorem 4.1 and condition (5.2)
hold. Then: 1) the homogeneous equation (5.1) has only trivial solution; 2) ev-
ery nontrivial Lagrange multiplier satisfying (4.3) is regular, i. e., has the form
(1, y ∗ ); 3) equation (4.3) with λ0 = 1 has unique solution y ∗ .

Corollary 5.1. Under the assumptions of Theorem 5.1 the Lagrange mul-
tiplier for the problem (3.4) which satisfies relations (4.3) and (4.4) exists, is
regular, and is determined uniquely.
Another corollary of Theorem 5.1 is the following result for the prob-
lem (3.18).

Theorem 5.2. Let the assumptions of Theorem 4.4 hold and


1
N M̃u (ν, f , g) < 1, ∀g ∈ K4 .
α0 ν

Then: 1) the equation [Fx0 (û, p̂, ĝ)]∗ y ∗ = 0 with x = (u, p), which is an equation
for the element y ∗ = (ξ, σ, ζ) ∈ Y0∗ and which is equivalent to the identities

νa(w, ξ) + c(û, w, ξ) + c(w, û, ξ) + b(w, σ) + ζ, w


Γ = 0, ∀w ∈ H̃ 1 (Ω),
b(ξ, r) = 0, ∀r ∈ L20 (Ω),

has only trivial solution; 2) every nontrivial Lagrange multiplier which satis-
fies (4.46) is regular, i. e., has the form (1, y ∗ ); 3) equation (4.46) with λ0 = 1
has unique solution y ∗ = (ξ, σ, ζ).

5.2. Uniqueness of solution of the extremal problem


for the functional J0
In this subsection we investigate the uniqueness of solution of the problem (3.4)
in the case when J = J0 and the function g entering in the boundary condition
for the velocity play the role of a control. The last is equivalent to choosing
fixed elements f0 ∈ T ∗ , ψ0 ∈ H 1/2 (ΓD ), and χ0 ∈ L2 (ΓN ) as K1 , K2 , and K3 .
Besides, we choose as K4 the following subset:

K4 = {g ∈ H̃ 1/2 (Γ) | g = g0 on Γ0 , g1/2,Γ ≤ c7 = const < ∞}. (5.7)

Here Γ0 is a part of the boundary Γ with meas Γ0 > 0, and g0 ∈ γ|Γ0 (H̃ 1 (Γ)) is
a given function. In virtue of the Friedrichs – Poincare inequality the following
estimate is valid:

u21 ≤ (1/α∗0 )|u|21 , ∀u ∈ H̃ 1 (Ω, Γ0 ), α∗0 = const > 0, (5.8)


Admixture diffusion in viscous fluids 461

where H̃ 1 (Ω, Γ0 ) = {v ∈ H̃ 1 (Ω) | v = 0 on Γ0 }.


Taking into account (5.7), we choose K = {f0 } × {ψ0 } × {χ0 } × K4 and
set μ1 = μ2 = μ3 = μ4 = 0 in (3.4). Thus, we consider the following extremal
problem

1 1
J(x) = |∇u|2 dΩ ≡ |u|21 → inf,
2 Ω 2
F (x, f0 , ψ0 , χ0 , g) = 0, x ∈ X, g ∈ K4 . (5.9)

Let us assume that the sufficient conditions of existence of a solution to the


problem (5.9) (see Corollary 3.3) hold as well as the assumptions of Theorem 4.1
and the condition
1 1 β1
N Mu0 + N1 Mϕ0 < 1, (5.10)
α0 ν α0 ν λ∗
which is an analogue of condition (5.2) for the above-mentioned choice of K.
Here

Mu0 ≡ sup Mu (v0 , f0 , ψ0 , χ0 , g), Mϕ0 ≡ sup Mϕ (v0 , f0 , ψ0 , χ0 , g). (5.11)


g∈K4 g∈K4

Let (x, g) ≡ (u, p, ϕ, g) be a solution of the problem (5.9). In virtue of


Theorems 4.1 and 5.1, for this solution there exists a unique Lagrange multi-
plier (1, ξ, σ, ζ, η, ζ1 ) such that identities (4.7)–(4.9), which in view of (4.17) take
the form

νa(w, ξ) + c(u, w, ξ) + c(w, u, ξ) + c1 (w, ϕ, η)


+ b(w, σ) + ζ, w
Γ + a(u, w) = 0, ∀w ∈ H̃ 1 (Ω),
b(ξ, r) = 0, ∀r ∈ L20 (Ω), (5.12)
1
a1 (τ, η) + c1 (u, τ, η) − b1 (τ, ξ) + ζ1 , τ
ΓD = 0, ∀τ ∈ H (Ω) (5.13)
and the variational inequality (4.5), which in view of the choice of K takes the
form
ζ, g̃ − g
Γ ≤ 0, ∀g̃ ∈ K4 (5.14)
hold.
Now suppose that there exist two solutions (x1 , g1 ) ≡ (u1 , p1 , ϕ1 , g1 ) and
(x2 , g2 ) ≡ (u2 , p2 , ϕ2 , g2 ) of the problem (5.9). In virtue of Theorem 2.1, the
a priori estimates (2.17)–(2.19) are valid. Taking into account (5.11), let us
rewrite them in the form

ϕi 1 ≤ Mϕ0 , ui 1 ≤ Mu0 , i = 1, 2. (5.15)

Let us denote by (1, yi∗ ) ≡ (1, ξi , σi , ζi , ηi , ζ1i ), i = 1, 2, the regular Lagrange


multipliers that correspond to the solutions (xi , gi ). We set

g = g1 − g2 , u = u 1 − u2 , p = p 1 − p2 , ϕ = ϕ1 − ϕ2 , ξ = ξ1 − ξ2 ,
σ = σ1 − σ2 , ζ = ζ1 − ζ2 , η = η1 − η2 , ζ1 = ζ11 − ζ12 (5.16)
462 G. V. Alekseev, E. A. Adomavichus

and subtract equations (2.12)–(2.14) written for u2 , p2 , ϕ2 , g2 from the same


equations written for u1 , p1 , ϕ1 , g1 . Then we obtain

νa(u, v) + c(u1 , u, v) + c(u, u2 , v) − b1 (ϕ, v) + b(v, p) = 0, ∀v ∈ H01 (Ω),


(5.17)
a1 (ϕ, S) + c1 (u1 , ϕ, S) + c1 (u, ϕ2 , S) = 0, ∀S ∈ T , (5.18)
b(u, q) = 0, ∀q ∈ L20 (Ω), u = g on Γ, ϕ = 0 on ΓD . (5.19)

Since g1 |Γ0 = g2 |Γ0 = g0 , therefore g = 0 on Γ0 . In virtue of the condition u = g


on Γ in (5.19) this means that u ∈ H̃ 1 (Ω, Γ0 ) and, hence, the estimate (5.8) is
valid for u.
We set v = ξ, q = σ, and S = η in (5.17)–(5.19). Taking into account the
second equation in (5.12), we get

νa(u, ξ) + c(u1 , u, ξ) + c(u, u2 , ξ) − b1 (ϕ, ξ) = 0, b(u, σ) = 0, (5.20)


a1 (ϕ, η) + c1 (u1 , ϕ, η) + c1 (u, ϕ2 , η) = 0. (5.21)

Let us subtract equations (5.12) and (5.13) written for the multiplier (1, y1∗ )
ons written for (1, y2∗ ). Setting w = u, r = p, and τ = ϕ and using the
notations (5.16), we get

νa(u, ξ) + c(u1 , u, ξ) + 2c(u, u, ξ2 ) + c(u, u1 , ξ)


+ c1 (u, ϕ1 , η) + c1 (u, ϕ, η2 ) + ζ, g
Γ = −|u|21 ,
b(ξ, p) = 0, (5.22)
a1 (ϕ, η) + c1 (u1 , ϕ, η) + c1 (u, ϕ, η2 ) − b1 (ϕ, ξ) = 0. (5.23)

Setting g̃ = g2 in inequality (5.14) written for g1 and g̃ = g1 in the same


inequality written for g2 and summing the obtained inequalities, we obtain

− ζ, g
Γ ≡ − ζ, g1 − g2
Γ ≤ 0. (5.24)

Subtracting (5.20) from (5.22) and (5.23) from (5.21) and then subtracting
the obtained results one from the other, in virtue of inequality (5.24) we get the
relation

c(u, u, ξ1 + ξ2 ) + c1 (u, ϕ, η1 + η2 ) + |u|21 = − ζ, g


Γ ≤ 0. (5.25)

Suppose that the following condition holds:

|c(u, u, ξ1 + ξ2 ) + c1 (u, ϕ, η1 + η2 )| < |u|21 . (5.26)

Then (5.25) and the estimate (5.8) imply u0 = 0. Setting u = 0 in (5.17)–(5.19)


and recalling Theorem 2.4 and the inf-sup property of the form b, we easily get
ϕ = 0, p = 0, and g = 0. This means that the solution to the problem (5.9) is
unique.
Admixture diffusion in viscous fluids 463

Write the conjugated system (5.12), (5.13) for the values ξi , σi , ζi , ηi , ζ1i
for u = ui , ϕ = ϕi , i = 1, 2, and then set w = ξi , r = σi , and τ = ηi in it.
Reasoning as in derivation of (5.3), we obtain

νa(ξi , ξi ) + c(ξi , ui , ξi ) + c1 (ξi , ϕi , ηi ) = −a(ui , ξi ), b(ξi , σi ) = 0, (5.27)


a1 (ηi , ηi ) − b1 (ηi , ξi ) = 0, i = 1, 2. (5.28)

Using the coerciveness inequalities (2.7) and (2.11), relations (2.10), and the
estimate |a(ui , ξi )| ≤ ui 1 ξi 1 , we deduce from (5.27) and (5.28) that

α0 νξi 21 ≤ N ξi 21 ui 1 + N1 ξi 1 ϕi 1 ηi 1 + ui 1 ξi 1 , (5.29)
λ∗ ηi 21 ≤ β1 ηi 1 ξi 1 , i = 1, 2. (5.30)

It follows from (5.30) that

ηi 1 ≤ (β1 /λ∗ )ξi 1 , i = 1, 2. (5.31)

Using (5.31), we conclude from (5.29) that


  
α0 ν − N ui 1 + N1 (β1 /λ∗ )ϕi 1 ξi 21 ≤ ui1 ξi 1 , i = 1, 2. (5.32)

Taking into account the a priori estimates (5.15) for ui and ϕi , from (5.32) we
obtain the inequality
  
α0 ν − N Mu0 + (β1 /λ∗ )N1 Mϕ0 ξi 21 ≤ Mu0 ξi 1 , i = 1, 2. (5.33)

The expression in the square brackets on the left-hand side of (5.33) is positive
in virtue of (5.10). Dividing by it, we obtain the following estimate for ξi :

Mu0
ξi 1 ≤ , i = 1, 2. (5.34)
α0 ν − [N Mu0 + (β1 /λ∗ )N1 Mϕ0 ]

Identity (5.18) where we should set S = ϕ ∈ T together with (2.5), (2.11),


and (5.15) implies

a1 (ϕ, ϕ) + c1 (u, ϕ2 , ϕ) = 0,
ϕ21 ≤ (N1 /λ∗ )u1 ϕ2 1 ϕ1 ≤ (N1 /λ∗ )Mϕ0 u1 ϕ1
or
ϕ1 ≤ (N1 /λ∗ )Mϕ0 u1 . (5.35)

Using (2.10), the estimates (5.31), (5.34), (5.35), and the estimates (5.8) for
u ∈ H̃ 1 (Ω, Γ0 ), we get

|c(u,u, ξ1 + ξ2 ) + c1 (u, ϕ, η1 + η2 )|
≤ N u21 (ξ1 1 + ξ2 1 ) + N1 u1ϕ1 (η1 1 + η2 1 )
 
≤ N u1 + N1 (β1 /λ∗ )ϕ1 u1(ξ1 1 + ξ2 1 )
1   N 2
1
 2Mu0
≤ ∗ N + β1 Mϕ0 |u|21 . (5.36)
α0 λ∗ α0 ν − [N Mu0 + (β1 /λ∗ )N1 Mϕ0 ]
464 G. V. Alekseev, E. A. Adomavichus

It follows from (5.36) that (5.26) is necessarily valid if the initial data of the
problem (5.9) and, in particular, the elements f0 , ψ0 , and χ0 and the set K4
are such that the following condition holds:

1   N 2
1 0
 2Mu0
∗ N + β 1 M < 1,
α0 λ∗ ϕ
α0 ν − [N Mu + (β1 /λ∗ )N1 Mϕ0 ]
0

and under the fulfillment (5.10) is equal to the condition

1 1 β1 2   N 2 
1
N Mu0 + N1 Mϕ0 + ∗ Mu0 N + β1 Mϕ0 < 1. (5.37)
α0 ν α0 ν λ∗ α0 α0 ν λ∗

Obviously, if condition (5.37) holds, then the initial data of the problem (5.9)
will a fortiori satisfy condition (5.10). This means that if condition (5.37)
is fulfilled and hence condition (5.10) is so, then the extremal problem (5.9)
has unique solution (u, p, ϕ, g), to which by Theorem 5.1, there corresponds a
unique solution (ξ, σ, ζ, η, ζ1 ) of the problem (5.12), (5.13). The last together
with Theorem 2.3 means the uniqueness of a solution to the optimality system
for the problem (5.9), which consists of relations (2.12)–(2.14), the conjugate
system (5.12), (5.13), and the variational inequality (5.14). Let us formulate
the obtained results.

Theorem 5.3. Let conditions (i)–(iv)hold; let f0 ∈ T ∗ , χ0 ∈ L2 (ΓN ), and


ψ0 ∈ H 1/2 (ΓD ) be fixed elements; let the set K4 be defined by formula (5.7);
and let condition (5.37) hold, where Mu0 and Mϕ0 are defined by relations (5.11).
Then the problem (5.9) has unique solution (u, p, ϕ, g) ∈ H̃ 1 (Ω) × L20 (Ω) ×
H 1 (Ω) × K4 .

Corollary 5.2. Under assumptions of Theorem 5.3 the optimality system


(2.12)–(2.14), (5.12)–(5.14) for the extremal problem (5.9) has unique solution
(u, p, ϕ, g, ξ, σ, ζ, η, ζ1 ) ∈ X × K4 × Y ∗ .
Let f0 = 0, ψ0 = 0, and χ0 = 0. Then ϕ = 0 by Theorem 2.1, and from
Theorem 5.3 we obtain the following result for the hydrodynamical extremal
problem:

1
J(x) = |∇u|2 dΩ → inf,
2 Ω
F 0 (x, g) = 0, x ≡ (u, p) ∈ X0 ≡ H̃ 1 (Ω) × L20 (Ω), g ∈ K4 . (5.38)

Theorem 5.4. Let conditions (i) and (iv)hold; the set K4 be defined by
formula (5.7); and the following condition hold:
 1 2 
+ ∗ N M̃u0 < 1, M̃u0 = sup M̃u (ν, f , g).
α0 ν α0 α0 ν g∈K4

Then the problem (5.38) has unique solution (u, p, g) ∈ H̃ 1 (Ω) × L20 (Ω) × K4 .
Admixture diffusion in viscous fluids 465

CONCLUSION

In the present paper we have investigated the inverse extremal problems for
the differential equations of mass-transfer which describe admixture diffusion
within the framework of the Oberbeck – Boussinesq model with inhomogeneous
boundary conditions for the velocity and concentration. Assuming the fulfill-
ment of certain conditions on the input data, we have proved the solvability
of the inverse problems in question, substantiated the application of the La-
grange multiplier principle, and derived the optimality system. The optimality
system was derived at first for an arbitrary differentiable objective functional
and then was specified for concrete objective functionals. In the general case it
contains ten nonlinear differential equations connected with each other by the
coefficients, boundary conditions, and variational inequalities for determination
of the densities of unknown admixture sources. Basing on the properties of
the optimality system, we have obtained sufficient conditions on the input data
which provide the regularity of the Lagrange multiplier and the uniqueness of
solution of the extremal inverse problem for ones objective functionals. Using
the constructed theory, we have proved the solvability and uniqueness of the
extremal inverse problems for the stationary Navier – Stokes equations.

In this paper we have also derived the optimality system for the linearized
model of admixture transfer. The system contains ten linear differential equa-
tions with respect to the initial and conjugate states together with the corre-
sponding boundary relations and variational inequalities for controls. In the
case when an objective functional does not depend on the velocity, the opti-
mality system can be simplified to a system of two scalar differential equations
for the concentration and the correspond Lagrange multiplier η. Moreover, if
the objective functional depends linearly on the concentration, as, for example,
in the problem of location of plants near ecologically significant zones, which
was considered in [15, 26], then for finding a solution to the original extremal
problem it is sufficient to solve the conjugate (elliptic boundary-value) prob-
lem for η and to determine the desired controls from the variational inequality.
Then the desired minimum is determined by an explicit formula analogous to
the corresponding formula in [26]. Thus, in this case it is not necessary to solve
the original direct problem, which essentially simplifies solution of the extremal
problem.

In the general case the optimality system is nonlinear, and its numerical
investigation involves considerable difficulties and requires development of ef-
fective numerical algorithms. The authors intend to consider this question in a
separate paper.
466 G. V. Alekseev, E. A. Adomavichus

Acknowledgement
The authors express their gratitude to N. N. Frolova and D. A. Tereshko for
useful discussions of our work.

REFERENCES

1. F. Abergel and E. Casas Renteria, Some optimal control problems of mul-


tistate equations appearing in fluid mechanics. RAIRO Modél. Math. Anal.
Numér. (1993) 27, No. 2, 223–247.

2. E. A. Adomavichus, On solvability of some extremal problems for station-


ary equation of heat convection. In: Far Easten Mathematical Reports.
Issue 5. Dal’nauka, Vladivostok, 1998, 74–85 (in Russian).

3. E. A. Adomavichus and G. V. Alekseev, Theoretical Analysis of Inverse


Extremal Problems for Stationary Mass-Transfer Equations. Preprint
No. 7. Institute of Applied Mathematics, Far Eastern Branch of Russia
Acad. Sci. Dal’nauka, Vladivostok, 1999 (in Russian).

4. E. A. Adomavichus and G. V. Alekseev, Theoretical Analysis of Inverse


Extremal Problems for Stationary Mass-Transfer Equations. II. Preprint
No. 18. Institute of Applied Mathematics, Far Eastern Branch of Russia
Acad. Sci. Dal’nauka, Vladivostok, 1999 (in Russian).

5. G. V. Alekseev, Theoretical Analysis of Boundary Control Problems for the


Boussinesq Equations. Preprint No. 16. Institute of Applied Mathematics,
Far Eastern Branch of Russia Acad. Sci. Dal’nauka, Vladivostok, 1996 (in
Russian).

6. G. V. Alekseev, Stationary problems of boundary control for heat convec-


tion equations. Dokl. Math. (1998) 58, No. 2, 314–317.

7. G. V. Alekseev, Solvability of stationary boundary control problems for


heat convection equations. Siberian Math. J. (1998) 39, No. 5, 844–858
(in Russian).

8. G. V. Alekseev, Inverse Problems of Detecting the Pollutant Sources in


Viscous Fluids. Preprint No. 8. Institute of Applied Mathematics, Far
Eastern Branch of Russia Acad. Sci. Dal’nauka, Vladivostok, 1999 (in
Russian).

9. G. V. Alekseev, Inverse extremal problems for stationary equations of heat


and mass transfer. Dokl. Math. (2000) 62, No. 3, 420–424.

10. G. V. Alekseev and V. V. Malikin, Numerical analysis of extremal prob-


lems for the two-dimensional equations of viscous fluids. Vychisl. Tekhnol.
(1993) 2, No. 5, 5–16 (in Russian).
Admixture diffusion in viscous fluids 467

11. G. V. Alekseev and V. V. Malikin, Numerical analysis of optimal boundary


control problems for stationary Navier – Stokes equations. Comput. Fluid
Dynamics J. (1994) 3, No. 1, 1–26.

12. G. V. Alekseev and D. A. Tereshko, Stationary optimal control problems


for equations of a viscous heat-conducting fluid. Sib. Zh. Industr. Matem.
(1998) 1, No. 2, 24–44 (in Russian).

13. G. V. Alekseev and D. A. Tereshko, On solvability of inverse extremal prob-


lems for stationary equations of viscous heat conducting fluid. J. Inv. Ill-
Posed Problems (1998) 6, No. 6, 521–562.

14. V. K. Andreev, O. V. Kaptsov, V. V. Pukhnachev, and A. A. Rodi-


onov, Application of Group-Theoretic Methods in Hydrodynamics. Nauka,
Novosibirsk, 1994 (in Russian).

15. V. M. Belolipetsky and Yu. I. Shokin, Mathematical Modelling in Prob-


lems of Environmental Protection. INFOLIO-Press, Novosibirsk, 1997 (in
Russian).

16. A. Capatina and R. Stavre, A control problem in biconvective flow.


J. Math. Kyoto Univ. (1997) 37, No. 4, 585–595.

17. M. Desai and K. Ito, Optimal controls of Navier – Stokes equations. SIAM
J. Control Optim. (1994) 32, No. 5, 1428–1446.

18. V. Girault and P.-A. Raviart, Finite Element Methods for Navier – Stokes
Equations Theory and Algorithms. Springer-Verlag, Berlin – New York,
1986.

19. M. D. Gunzburger, L. Hou, and T. P. Svobodny, Analysis and finite


element approximation of optimal control problems for the stationary
Navier – Stokes equations with distributed and Neumann controls. Math.
Comp. (1991) 57, No. 195, 123–151.

20. M. D. Gunzburger, L. Hou, and T. P. Svobodny, Analysis and finite


element approximation of optimal control problems for the stationary
Navier – Stokes equations with Dirichlet controls. RAIRO Modél Math.
Anal. Numér (1991) 25, No. 6, 711–748.

21. M. D. Gunzburger, L. Hou, and T. P. Svobodny, The approximation of


boundary control problems for fluid flows with an application to control
by heating and cooling. Comput. Fluids (1993) 22, No. 2, 3, 239–251.

22. A. D. Ioffe and V. M. Tikhomirov, The Theory of Extremal Problems.


Nauka, Moscow, 1974 (in Russian).

23. K. Ito and S. S. Ravindran, Optimal control of thermally convected fluid


flows. SIAM J. Sci. Comput. (1998) 19, No. 6, 1847–1869.
468 G. V. Alekseev, E. A. Adomavichus

24. Y. Kan-On, K. Narukawa, and Y. Teramoto, On the equations of biocon-


vective flow. J. Math. Kyoto Univ. (1992) 32, No. 1, 135–153.

25. H.-C. Lee and O. Yu. Imanuvilov, Analysis of optimal control problems
for the 2-D stationary Boussinesq equiations. J. Math. Anal. Appl. (2000)
242, No. 2, 191–211.

26. G. I. Marchuk, Mathematical Modelling in the Problem of the Environ-


ment. Nauka, Moscow, 1982 (in Russian).
27. G. I. Marchuk, V. I. Agoshkov, and V. P. Shutyaev, Conjugat Equations
and Perturbation Algorithms in Nonlinear Problems of Mathematical
Physics. Nauka, Moscow, 1993 (in Russian).

28. V. V. Penenko and A. E. Aloyan, Models and Methods for Problems of


Environmental Protection. Nauka, Novosibirsk, 1985 (in Russian).
29. A. B. Smyshlyaev, Extremal boundary control problems for the linear
equations of the heat convection. In: Far Eastern Mathematical Reports.
Issue 5. Dal’nauka, Vladivostok, 1998, 86–101 (in Russian).
30. R. Temam, Navier – Stokes Equations. Theory and Numerical analysis.
North-Holland Publ. Co., Amsterdam – New York – Oxford, 1979.
31. D. A. Tereshko, Analysis of inverse extremum problems for the stationary
Boussinesq equations. In: Far Eastern Mathematical Reports. Issue 4.
Dal’nauka, Vladivostok, 1997, 75–85 (in Russian).
32. V. A. Trenogin, Functional Analysis. Nauka, Moscow, 1993 (in Russian).

You might also like