Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Z. angew. Math. Phys.

58 (2007) 994–1007
0044-2275/07/060994-14
DOI 10.1007/s00033-006-6114-3 Zeitschrift für angewandte
°c 2007 Birkhäuser Verlag, Basel Mathematik und Physik ZAMP

On the unsteady Poiseuille flow in a pipe


G. P. Galdi1 , K. Pileckas2 and A. L. Silvestre3

Abstract. We show that in an unsteady Poiseuille flow of a Navier–Stokes fluid in an infinite


straight pipe of constant cross-section, σ, the flow rate, F (t), and the axial pressure drop, q(t),
are related, at each time t, by a linear Volterra integral equation of the second type, where the
kernel depends only upon t and σ. One significant consequence of this result is that it allows us
to prove that the inverse parabolic problem of finding a Poiseuille flow corresponding to a given
F (t) is equivalent to the resolution of the classical initial-boundary value problem for the heat
equation.

Mathematics Subject Classification (2000). 35Q30, 76D03, 76D05.

Keywords. Unsteady Poiseuille flow, flow in a pipe, axial pressure drop, Volterra integral equa-
tion.

1. Introduction

The motion of a Navier–Stokes fluid in an infinite straight pipe of constant cross-


section, σ, is one of the primary and most studied problems in fluid dynamics;
see, e.g., [2] [10] . Among all practicable motions, the so-called Poiseuille flow
plays a fundamental role. As is well known, this flow is described by a velocity
field with only one nonzero component, u(x, t), directed along the axis of the pipe
and depending only on the coordinates, x, of points of σ and, possibly, on time t.
The corresponding pressure field does not depend on x and is characterized by a
(nonzero) axial gradient, q(t), that depends on time only.
Typically, there are two ways of determining a Poiseuille flow,R namely, by
prescribing either the axial pressure drop q(t) or the flow rate F (t) := σ u(x, t) dx.
In the first case, the problem reduces to solving an initial-boundary value prob-
lem for the heat equation for u = u(x, t) with a prescribed time-dependent forcing
term q(t) . The unique solvability of this problem is, therefore, quite standard; see
e.g. [9].

1 Partially
supported by the NSF grant DMS–0404834.
2 Supported by EC FP6 MCToK program SPADE2, MTKD–CT–2004–014508
3 Supported by FCT-Project POCI/MAT/61792/2004
Vol. 58 (2007) On the unsteady Poiseuille flow in a pipe 995

In the second case, however, when F = F (t) is prescribed, we have to find


u = u(x, t) and q = q(t), and the problem can be seen as an inverse heat conduction
problem consisting in prescribing an initial distribution of temperature inside a
conducting solid and recovering the evolution of an unknown constant (in space)
source term from the measurament (and, therefore, the prescription) of the average
temperature inside the body. In its general form, it can be formulated as follows,
see [14]. Given u0 (x), f (x, t) and F (t) to find a pair of functions (u(x, t), q(t))
solving the following initial-boundary value problem on the cross-section σ:

 ∂

 u(x, t) − ν∆u(x, t) = q(t) + f (x, t),

 ∂t
u(x, t) |∂σ = 0, u(x, 0) = u0 (x), (1.1)

 Z


 u(x, t) dx = F (t) .
σ
It should be emphasized that the interest in studying the unique solvability of
(1.1) goes well beyond the case of a flow in an infinite straight pipe. In fact,
along with its steady-state and time-periodic counterparts [3], [4],4 problem (1.1)
is particularly significant in a number of situations involving flow in “bent” pipes
or in pipes of varying cross-sections which eventually end up in the shape of semi-
infinite straight pipes. In all these instances, u(x, t) becomes, at each time t, the
spatial asymptotic of the relevant velocity field; see [2], [10] .
Existence, uniqueness and asymptotic in time behavior of solutions to (1.1) in
appropriate function classes were first obtained in [12], [13], [14]. In particular,
in [13], by using a modified Fourier method, the unique solvability of problem
(1.1) was studied in Hölder spaces, while the behavior in time of corresponding
solutions was analyzed in [12]. Analogous results were obtained in [14] within an
L2 -framework, by means of a suitably modified Galerkin method.5
Besides the question of its unique solvability, another important question con-
nected to problem (1.1) is that of determining the relation between F (t) and q(t).
Obviously, this relation is fundamental in any fluid dynamical problem in pipes
where one wants to control the pressure drop in the flow by the corresponding
flow rate, or vice versa. Whereas in its steady counterpart this question admits a
trivial and positive answer,6 in the time-dependent case the answer is not so clear
4 The corresponding steady-state and time periodic problems are obtained from (1.1) when F
and f are time-independent or time-periodic. In both cases the initial condition is, obviously,
disregarded.
5 A problem similar to (1.1), which consists in finding a coefficient p(t) in the source term
S(x, t) := V (x, t) + p(t)G(x, t), with V and G given functions, has been solved by several authors,
see e.g. [15], under a prescribed weighted average of the field u:
Z
ψ(t) = u(x, t)w(x)dx.
σ
However, the methods and the results do not apply to our problem, since the weight function w
must vanish on ∂σ.
6 Actually, F and q are proportional through a constant depending only on σ; see, [6, Exercise
0.1 of Chapter VI], [14].
996 G. P. Galdi, K. Pileckas and A. L. Silvestre ZAMP

a priori. Specifically, it is not obvious whether or not the relation between F (t)
and q(t) may depend on the solution u(x, t) itself; see [3], [14]. However, in the
time-periodic case, it was shown in [4] that F (t) and q(t) are, in fact, connected by
an elementary and invertible relationship that depends only on σ . One important
consequence of this fact is that the inverse problem of a prescribed flow rate can
be thus reduced to the elementary problem of a prescribed axial pressure gradient.
The objective of this article is two-fold.
On the one hand, the aim is to prove that the flow rate F (t) and the axial pres-
sure gradient q(t) are related by a linear Volterra integral equation of the second
type with corresponding kernel, K(t), depending only on the cross-section σ and
on time (see Section 2). Furthermore, we show that K(t) is Lebesgue integrable
in (0, ∞) (see Lemma 2.1), and so,by well known results on Volterra’s equations
this implies not only that the integral equation admits a unique solution, but also
that it is invertible, thus obtaining a one-to-one correspondence between F (t) and
q(t) (see Lemma 2.3 and (2.22)) . We also prove that K(t) is positive for all t
and that it admits a simple representation through an infinite series involving the
eigenfunctions and eigenvalues of the Laplace operator on σ (see Lemma 2.2). This
property has the interesting consequence that, for given F (t), the corresponding
solution q(t) can be approximated by solutions to simpler integral equations with
suitable truncated kernels (see (2.20)).
On the other hand, our goal is to furnish results of (global) existence and
uniqueness of solutions to (1.1) in general Sobolev spaces. In fact, the above
mentioned one-to-one relation between F (t) and q(t), independent of the particular
solution, allows us to furnish results of unique solvability in a very elementary
way. Specifically, given F (t) we can evaluate the corresponding q(t) and then
solve (1.1)1,2 with the found q(t). This latter problem, in turn, is just the classical
initial-boundary value problem for the heat equation. Consequently, we may apply
all well known results and prove unique solvability in suitable Sobolev spaces of
arbitrary order, provided, of course, the data are sufficiently smooth and satisfy
the appropriate compatibility conditions (see Theorem 3.2).
The plan of the paper is the following. In Section 2 we derive the linear integral
equation relating F (t) to q(t) and prove some corresponding existence and unique-
ness results. In Section 3 we apply these results to show the unique solvability of
(1.1) in appropriate Sobolev spaces.

2. An integral equation relating axial pressure gradient and flow


rate

The objective of this section is to show that the axial pressure gradient q(t) and
the flow rate F (t) are related by a linear Volterra integral equation. With a view
to the unique solvability of problem (1.1), it will suffice to restrict ourselves to the
Vol. 58 (2007) On the unsteady Poiseuille flow in a pipe 997

case of zero initial data and body force.7


For Ω an open set of Rn , n ≥ 1, q ∈ [1, ∞] and m a non-negative integer, we
shall use the standard notation for Lebesgue spaces (Lq (Ω)) and Sobolev spaces
(Wqm (Ω), W̊21 (Ω), etc.) of functions in Ω; see, e.g. [9].8
The cross-section σ (the relevant region of flow) is assumed to be a bounded
domain of Rn−1 , n ≥ 2, with the boundary ∂σ ∈ C 2 if n > 2.
Let us consider in σ T := σ × (0, T ), T < ∞, the problem (1.1)1,2 with f ≡
u0 ≡ 0, namely, 
 ∂ u(x, t) − ν∆u(x, t) = q(t),

∂t (2.1)


u(x, t) |∂σ = 0, u(x, 0) = 0 .
Denote by Ψ(x) the solution to the following Poisson equation in σ:
(
−ν∆Ψ(x) = 1,
(2.2)
Ψ(x) |∂σ = 0 ,
and let Θ(x, t) be the solution to the following initial boundary value problem in
σ T for the heat equation

 ∂ Θ(x, t) − ν∆Θ(x, t) = 0 ,

∂t (2.3)


Θ(x, t) |∂σ = 0 , Θ(x, 0) = Ψ(x) .
It is well known [8] that, under the stated assumption on σ, problem (2.2) admits
a unique classical solution Ψ(x). For our purpose it is enough to know that Ψ ∈
W22 (σ) ∩ W̊21 (σ) and that
kΨkW22 (σ) ≤ c0 , (2.4)
where c0 depends only on σ. Consequently, we conclude that (2.3) has a unique
solution Θ such that
Z T
∂ 2
k Θ(·, t)kL2 (σT ) + kΘ(·, t)k2W 2 (σ) dt < ∞ ;
∂t 0
2

and that this solution satisfies


Z T

k Θ(·, t)k2L2 (σT ) + kΘ(·, t)k2W 2 (σ) dt ≤ c kΨk2W 1 (σ) ; (2.5)
∂t 0
2 2


see, e.g., [8] . Since Ψ ∈ W22 (σ), we also deduce that ∇ ∂t Θ ∈ L2 (σ T ) and that
∂ ∂
sup k Θ(·, t)kL2 (σ) + k∇ Θ(·, t)kL2 (σT ) ≤ ckΨkW22 (σ) ; (2.6)
t∈[0,T ] ∂t ∂t

7 The generalization to nonzero data is, however, straightforward; see also next section.
8 Of course, the physical interesting cases are n = 2, 3.
998 G. P. Galdi, K. Pileckas and A. L. Silvestre ZAMP

2

see [8]. Finally, for arbitrary δ > 0, we further find ∂t 2 Θ ∈ L2 (σ × (δ, T )) and
∂ 9
∆ ∂t Θ ∈ L2 (σ × (δ, T )).
Now, let u(x, t) be a “sufficiently smooth” solution to (2.1). Multiplying both
sides of equation (2.1) by Θ(x, τ − t) and integrating by parts over σ, we find
Z Z Z

u(x, t)Θ(x, τ − t) dx = ν ∆u(x, t)Θ(x, τ − t) dx + q(t) Θ(x, τ − t) dx,
σ ∂t σ σ
that is Z Z
∂ ∂
u(x, t)Θ(x, τ − t) dx − u(x, t) Θ(x, τ − t) dx
∂t σ σ ∂t
Z Z
= ν u(x, t)∆Θ(x, τ − t) dx + q(t) Θ(x, τ − t) dx.
σ σ
Integrating this last relation with respect to t from 0 to τ and taking into account
that
∂ ∂
ν∆Θ(x, τ − t) = Θ(x, τ − t) = − Θ(x, τ − t),
∂τ ∂t
¯
¯
u(x, 0) = 0, Θ(x, τ − t)¯ = Θ(x, 0) = Ψ(x),
t=τ
we derive Z Z τ
u(x, τ )Ψ(x) dx = H(τ − t)q(t) dt , (2.7)
σ 0
with Z
H(s) = Θ(x, s) dx . (2.8)
σ
We next differentiate both sides of (2.7) with respect to τ to get
Z Z τ

u(x, τ )Ψ(x) dx = H(0)q(τ ) + H ′ (τ − t)q(t) dt.
σ ∂τ 0
Using equation (2.1), (2.2) and integrating twice by parts we deduce
Z Z Z

u(x, τ )Ψ(x) dx = ν ∆u(x, τ )Ψ(x) dx + q(τ ) Ψ(x) dx
σ ∂τ σ σ
Z Z Z
= ν u(x, τ )∆Ψ(x) dx + q(τ ) Ψ(x) dx = − u(x, τ ) dx
σ σ σ
Z Z
+q(τ ) Ψ(x) dx = −F (τ ) + q(τ ) Ψ(x) dx,
σ σ
R R
where
R F (τ ) = σ
u(x, τ ) dx is the flow rate. Since H(0) = σ Θ(x, 0) dx =
σ
Ψ(x) dx, from the two latter displayed equalities, we find that F (t) and q(t)
are related by the following Volterra equation of the first type
Z τ
−F (τ ) = H ′ (τ − t)q(t) dt. (2.9)
0
9 Note that it is not possible to take δ = 0, because the initial datum Ψ(x) does not satisfy the
compatibility condition, i.e., ∆Ψ(x) |∂σ 6= 0.
Vol. 58 (2007) On the unsteady Poiseuille flow in a pipe 999

If we differentiate (2.9) with respect to τ , we obtain


Z τ
−F ′ (τ ) = H ′ (0)q(τ ) + H ′′ (τ − t)q(t) dt , (2.10)
0

and since, by (2.2), (2.3), we have


Z ¯ Z
∂ ¯
H ′ (0) = Θ(x, τ )¯ dx = ν ∆Ψ(x) dx = −|σ|
σ ∂τ τ =0 σ

we conclude that (2.10) takes the following form


Z τ
q(τ ) = K(τ − t)q(t) dt + Φ(τ ) , (2.11)
0

where Φ(τ ) := F ′ (τ )/|σ| and


Z
1 d2
K(s) := Θ(x, s) dx . (2.12)
|σ| ds2 σ

Notice that (2.11) is a Volterra integral equation of the second type with respect
to q(τ ).
Lemma 2.1. The kernel K(s), defined by (2.12), belongs to the space L1 (0, ∞).
Proof. We begin to show integrability of K in (0, T ), for all T > 0. From (2.2) we
get
Z Z
1 ∂ ν ∂ ∂
K(s) = ν∆ Θ(x, s) dx = Θ(x, s) dS s ∈ (0, T ).
|σ| σ ∂s |σ| ∂σ ∂n ∂s
By Cauchy–Schwarz inequality and by well know trace inequality (see [7]) we have
Z ¯ ¯2
2 ν|∂σ| ¯ ∂ ∂ ¯
|K(s)| ≤ ¯ Θ(x, s)¯ dS
|σ| ∂σ ∂n ∂s
µ ¶ (2.13)
∂ ∂ ∂
≤ c k∇ Θ(·, s)k2L2 (σ) + k∇ Θ(·, s)kL2 (σ) k∆ Θ(·, s)kL2 (σ) .
∂s ∂s ∂s

Here we have also used the inequality


X
kDα wkL2 (σ) ≤ ck∆wkL2 (σ)
|α|=2

which is valid for functions w vanishing on ∂σ and bounded domains σ with C 2


boundary (see, e.g., [7]). Let us now show that

k∆ Θ(·, t)kL2 (σ) ≤ c t−1 t ∈ (0, T ) . (2.14)
∂t

Actually, setting W (x, t) = ∂t Θ(x, t), we find

W (x, t) = ν∆W (x, t).
∂t
1000 G. P. Galdi, K. Pileckas and A. L. Silvestre ZAMP

Therefore, recalling that Ψ ∈ W22 (σ), by classical estimates of the heat semigroup
(see, e.g., [5]), we have

kν∆W (·, t)kL2 (σ) = k W (·, t)kL2 (σ)
∂t
≤ c t−1 kW (·, 0)kL2 (σ) = c t−1 k∆ΨkL2 (σ) = e c t−1
which proves (2.14). Thus, from (2.13), (2.14), by the Hölder inequality and by
(2.6) it follows that
Z t ³Z t° ∂ °
|K(s)| ds ≤ c °∇ Θ(·, s)° ds
∂s L2 (σ)
0 0
Z t ´
° ∂ ° ° °
+ °∇ Θ(·, s)°1/2 °∆ ∂ Θ(·, s)°1/2 ds
∂s L2 (σ) ∂s L2 (σ)
0
³ Z t
° ∂ ° ° ∂ °1/2 ¡ ° ∂ ° ¢ ´
≤ c t1/2 °∇ Θ°L (σ×(0,t)) + °∇ Θ°L (σ×(0,t)) °∆ Θ(·, s)°2/3 ds 3/4
∂s 2 ∂s 2 ∂s L2 (σ)
0
³ Z
¡ t −2/3 ¢3/4 ´
≤ c(T ) 1 + s ds ≤ C(T )
0
which proves that K ∈ L1 (0, T ), for any finite T > 0. From (2.12) we also have
∂2
Θ(·, t)kL2 (σ) .
|K(t)| ≤ ck (2.15)
∂t2
However, for any t > t0 and any t0 > 0, again by classical estimates for the heat
semigroup [5] we get
∂2 −1 ∂
k 2 Θ(·, t)kL (σ) ≤ c1 (t − t0 )
2 k Θ(·, t0 )kL2 (σ) ≤ c2 (t − t0 )−1 t−1
0 kΘ(·, 0)kL2 (σ) .
∂t ∂t
By taking t0 = t/2, from this relation and (2.15) it follows that K ∈ L1 (a, ∞) for
all a > 0 and the lemma is proved.
Our next objective is to show that the kernel K(t) admits a useful represen-
tation in terms of a suitable infinite series. To this end, let ψk ∈ W̊21 (σ) ∩ W22 (σ)
and λk > 0 be eigenfunctions and eigenvalues of the Laplace operator, namely,

 −ν∆ψk (x) = λk ψk (x),
¯ (2.16)
 ¯
ψk (x)¯ = 0.
∂σ
Note that {λk } → ∞ as k → ∞. We take the eigenfunction ψk (x) to be orthonor-
mal in L2 (σ). Thus,
Z Z
ν |∇ψk (x)|2 dx = λk , ∇ψk (x) · ∇ψl (x) dx = 0, k 6= l.
σ σ
Moreover, {ψk (x)} form a basis in L2 (σ).
The following result holds.
Vol. 58 (2007) On the unsteady Poiseuille flow in a pipe 1001

Lemma 2.2. The kernel K(t) admits the representation



1 X 2
K(t) = βk λk exp(−λk t) , (2.17)
|σ|
k=1
R
where βk = ψk (x)dx, k = 1, 2, . . . , . Thus, K(t) > 0 for all t > 0 and, moreover,
σ
Z ∞
K(t) dt = 1 . (2.18)
0

Proof. Since the constant function 1 belongs to L2 (σ), we may write it as a Fourier
series:
X∞
1= βk ψk (x),
k=1

P
where βk2 = |σ|. It is easy to see that the solution Ψ(x) to problem (2.2) can
k=1
be represented as

X βk
Ψ(x) = ψk (x) (2.19)
λk
k=1

and that the series (2.19) converges in W22 (σ). Now, solving problem (2.3) by the
classical Fourier method, we find that Θ(x, t) has the representation

X βk
Θ(x, t) = exp(−λk t)ψk (x).
λk
k=1

Hence,
Z ∞ ³Z ´
1 d2 1 X
K(t) = Θ(x, t) dx = βk λk exp(−λk t) ψk (x) dx
|σ| dt2 σ |σ| σ
k=1

X
1
= βk2 λk exp(−λk t),
|σ|
k=1

and (2.17) follows. It remains to prove (2.18). Actually,


Z ∞ ∞ Z ∞ ∞
1 X 2 1 X 2
K(t) dt = βk λk exp(−λk t) dt = βk = 1.
0 |σ| 0 |σ|
k=1 k=1

The lemma is therefore proved.


Remark 2.1. One remarkable consequence of Lemma 2.2 is that the solution
q(t) to (2.11) can be approximated by the sequence of solutions to the following
problem Z t
q (N ) (t) = K (N ) (t − s)q (N ) (s) ds + Φ(t) , (2.20)
0
1002 G. P. Galdi, K. Pileckas and A. L. Silvestre ZAMP

N = 1, 2, . . . , , where
¡ K (N ) is¢ the N − th partial sum of the series (2.17). In [13],
[12] the solution u(x, t), q(t) to the inverse problem (1.1) is constructed by using
¡ ¢
Galerkin approximations u(N ) (x, t), q (N ) (t) with q (N ) (t) satisfying (2.20) .

Once the summability property of the kernel K is proved, it is easy to es-


tablish existence of solutions to the integral equation (2.11) and the validity of
corresponding estimates. In fact, the following result holds.
Lemma 2.3. Let Φ ∈ Lr (0, T ), T < ∞, 1 ≤ r < ∞. Then equation (2.11) has
one and only one solution q ∈ Lr (0, T ), and the following inequality holds
c(T ) ′
kqkLr (0,T ) ≤ kF kLr (0,T ) . (2.21)
|σ|
Furthermore, q(t) ≥ 0 (respectively, q(t) ≤ 0) in [0, T ], if F ′ (t) ≥ 0 (respectively,
F ′ (t) ≤ 0) in [0, T ] .

Proof. For r = 1 the lemma is proved in [11]. Moreover, it is shown in [17] that
the unique solution q is represented by the formula
Z t
q(t) = R(t − s)Φ(s) ds + Φ(t), t ∈ [0, T ], (2.22)
0

where R(t) is the resolvent associated with a given kernel K(t). Recall that R(t)
is defined as the unique L1 -solution of the equation
Z t
R(t) = K(t − s)R(s) ds + K(t),
0

and that, moreover, it satisfies the following estimate [11]


kRkL1 (0,T ) ≤ C(T )kKkL1 (0,T ) . (2.23)
Applying Young’s inequality for convolutions (see, e.g., [1]), from (2.22), (2.23) we
conclude q ∈ Lr (0, T ) and that
kqkLr (0,T ) ≤ kRkL1 (0,T ) kΦkLr (0,T ) + kΦkLr (0,T )

C1 ′
≤ C kKkL1 (0,T ) kΦkLr (0,T ) + kΦkLr (0,T ) ≤ C1 kΦkLr (0,T ) = kF kLr (0,T ) ,
|σ|
where C1 = C1 (T ) . The final part of the lemma follows by classical results on
Volterra’s equations with non-negative kernels [11], by the fact that, by Lemma
2.2, K(t) > 0 for all t > 0 and by recalling that Φ(t) = F ′ (t)/|σ| .

Let us now go back to the integral equation (2.9). From (2.6) it follows that
H ′ ∈ L∞ (0, T ). So, if q ∈ L1 (0, T ), we conclude from (2.9) the necessary compat-
ibility condition:
F (0) = 0 . (2.24)
Vol. 58 (2007) On the unsteady Poiseuille flow in a pipe 1003

Assume, now, that Φ′ (t) := |σ|−1 F ′′ (t) is integrable in (0, T ). Then, from [11,
Theorem 1] we obtain that q ′ ∈ L1 (0, T ) and that, moreover, differentiating (2.11),
q ′ (t) satisfies the following equation
Z t
q ′ (t) = K(t)q(0) + K(t − s)q ′ (s) ds + Φ′ (t) . (2.25)
0
Thus, if
q(0) = 0 , (2.26)
we find that Z t
q ′ (t) = K(t − s)q ′ (s) ds + Φ′ (t) . (2.27)
0
Note that, by (2.11), condition (2.26) requires
F ′ (0) = |σ|Φ(0) = 0 . (2.28)
Now, (2.27) is formally identical to (2.11), so that we can apply the results of
Lemma 2.3, provided F (t) obeys the compatibility condition (2.28) . Obviously,
this argument can be iterated.
From the above considerations and from Lemma 2.3 we thus deduce the fol-
lowing result.
Lemma 2.4. Let F ∈ Wrl+1 (0, T ), 1 ≤ r < ∞, l non-negative integer, T > 0,
and assume the compatibility conditions
dF dl F
F (0) = 0, (0) = 0, . . . , (0) = 0 . (2.29)
dt dtl
Then integral equation (2.11) admits a unique solution q ∈ Wrl (0, T ),
kqkWrl (0, T ) ≤ c(T )kF kWrl+1 (0, T ) . (2.30)
Moreover, for l ≥ 1,
dq dl−1 q
q(0) = 0, (0) = 0, . . . , (0) = 0. (2.31)
dt dtl−1

3. Unique solvability of problem (1.1)

In this section we prove results regarding the unique solvability of problem (1.1).
This will be accomplished at once, by combining Lemma 2.4 along with classical
results on the heat equation.
In order to state our results, we need to recall the definition of appropriate
function spaces. For any real numbers s > 0 and q ≥ 1, we denote by Wqs (σ) the
Sobolev–Slobodetskii space with the norm
µ X Z Z ¯¯ α ¯q ¶1/q
q Dx u(x) − Dyα u(y)¯
kukWqs (σ) = kuk [s] + dx dy ,
Wq (σ)
σ σ |x − y|n+q(s−[s])
|α|=[s]
1004 G. P. Galdi, K. Pileckas and A. L. Silvestre ZAMP

∂ |α|
where Dxα = α
∂x1 1 ...∂xα n , |α| = α1 + . . . + αn , and [s] denotes the integer part of
n

s. Furthermore, by Wr2l, l (σ T ), l a non-negative integer, r ≥ 1, we indicate the


space of functions having (generalized) derivatives Dtk Dxα with k and α obeying
the condition 2k + |α| ≤ 2l. The norm in Wr2l, l (σ T ) is defined by the formula
µX
2l X Z T Z ¶1/r
kukWr2l, l (σT ) = |Dtk Dxα u(x, t)|r dxdt .
j=0 2k+|α|=j 0 σ

We are now in a position to prove the unique solvability of (1.1). Consider first
the case u0 (x) = 0, f (x, t) = 0.
Theorem 3.1. Let ∂σ ∈ C 2l+2 , F ∈ Wrl+1 (0, T ), l ≥ 0, 1 < r < ∞, r 6= 3/2, T <
∞, u0 (x) ≡ f (x, t) ≡ 0, and let F (t)
¡ satisfy
¢ the compatibility conditions (2.29).
Then there exists a unique solution u, q ∈ Wr2l+2,l+1 (σ T ) × Wrl (0, T ) of problem
(1.1) and the following estimate holds
kukWr2l+2,l+1 (σT ) + kqkWrl (0,T ) ≤ c kF kWrl+1 (0,T ) . (3.1)

Proof. By Lemma 2.4, the integral equation (2.11) has a unique solution q ∈
Wrl (0, T ), which satisfies the estimate (2.30) and the conditions (2.31). Consider
now problem (2.1) with the right-hand side equal to that q(t). According to well
known results [9], [16], this problem has a unique solution u ∈ Wr2l+2,l+2 (σ T ),
r 6= 3/2, and
kukWr2l+2,l+1 (σT ) ≤ ckqkWr2l,l (σT ) ≤ ckqkWrl (0, T ) . (3.2)

Obviously, u(x, t) satisfies the flow rate condition (1.1)3 . Estimate (3.1) follows
from (3.2) and from (2.30).

For the general case of problem (1.1), we have the following result.
2l+2−2/r
Theorem 3.2. Let ∂σ ∈ C 2l+2 , F ∈ Wrl+1 (0, T ), u0 ∈ Wr (σ), f ∈
Wr2l,l (σ T ), l ≥ 0, 1 < r < ∞, r 6= 3/2, T < ∞. Suppose that following compatibi-
lity conditions hold
Z
ds F (s)
(0) = u0 (x) dx, s = 0, 1, . . . , l,
dts σ (3.3)
(s)
u0 (x)|∂σ = 0, s = 0, 1, . . . , [l + 1 − 3/2r],
where
(0)
u0 (x) := u0 (x),
(s) (s−1) (s−1)
u0 (x) := ν∆u0 (x) + f0 (x), s = 1, 2, . . . , l,
(3.4)

(s) ∂sf
f0 (x) := (x, 0), s = 0, 1, . . . , l .
∂ts
Vol. 58 (2007) On the unsteady Poiseuille flow in a pipe 1005
¡ ¢
Then problem (1.1) admits a unique solution u, q ∈ Wr2l+2,l+1 (σ T ) × Wrl (0, T )
and the following estimate holds
kukWr2l+2,l+1 (σT ) + kqkWrl (0,T )
µ ¶
≤ c kF kWrl+1 (0,T ) + ku0 kW 2l+2−2/r (σ) + kf kWr2l,l (σT ) . (3.5)
r

¡ ¢
Proof. We look for the solution u(x, t), q(t) of problem (1.1) in the form
¡ ¢ ¡ ¢
(u(x, t), q(t)) = u(1) (x, t), 0 + u(2) (x, t), q(t) ,
where u(1) (x, t) is the solution of the heat equation

 ∂ (1)
 u (x, t) − ν∆u(1) (x, t) = f (x, t),
∂t (3.6)

 (1) ¯
u (x, t)¯∂σ = 0, u(1) (x, 0) = u0 (x)
and (u(2) (x, t), q(t)) is the solution of the inverse problem

 ∂ (2)

 u (x, t) − ν∆u(2) (x, t) = q(t),

 ∂t


u(2) (x, t) |∂σ = 0 , u(2) (x, 0) = 0, (3.7)



 Z


 u(2) (x, t) dx = Fe(t)
σ
R
with Fe(t) := F (t) − σ u(1) (x, t) dx. The solvability of problem (3.6) in the ap-
propriate spaces is well known; see, e.g., [9], [16]). Specifically, since ∂σ ∈ C 2l+2
(s)
and u0 (x)|∂σ = 0, s = 0, 1, . . . , [l + 1 − 3/2r], there exists a unique solution
u(1) ∈ Wr2l+2,l+1 (σ T ). This solution satisfies
µ ¶
ku(1) kWr2l+2,l+1 (σT ) ≤ c ku0 kW 2l+2−2/r (σ) + kf kWr2l,l (σT ) . (3.8)
r

The function Fe(t) in (3.7) satisfies the estimate


 
l+1
X j

kFekWrl+1 (0,T ) ≤ c kF kWrl+1 (0,T ) + k j u(1) kLr (σT ) 
∂t
j=0 (3.9)
³ ´
≤ c kF kWrl+1 (0,T ) + ku0 kW 2l+2−2/r (σ) + kf kWr2l,l (σT ) .
r

Moreover, from (3.3)1 , we conclude that


ds Fe
(0) = 0, s = 0, 1, . . . , l .
dts
¡ ¢
Therefore, by Theorem 3.1, problem (3.7) admits a unique solution u(2) , q ∈
Wr2l+2,l+1 (σ T ) × Wrl (0, T ) and estimate (3.1) holds. Estimate (3.5) follows now
from inequalities (3.8), (3.9) and (3.1).
1006 G. P. Galdi, K. Pileckas and A. L. Silvestre ZAMP

Remark 3.1. Note that all obtained results are true for arbitrary finite T > 0. In
[14] the solvability of problem (1.1) is proved for the infinite time interval [0, ∞]
in spaces W22l+2,l+1 (σ × (0, ∞)) × W l (0, ∞) by using Galerkin approximations. In
order to get a priori estimates for Galerkin approximations on the infinite interval
[0, ∞], in [14] the whole system (1.1) was used.

References

[1] Adams, R. A., Sobolev Spaces, Academic Press, New York, San Francisco, London 1975.
[2] Batchelor, G. K., An Introduction to Fluid Dynamics, Cambridge University Press, Cam-
bridge 2002.
[3] Beirão da Veiga, H., On time-periodic solutions of the Navier–Stokes equations in an un-
bounded cylindrical domain. Leray’s problem for periodic flows, Arch. Rational Mech. Anal.
178 (2005), 301–325.
[4] Galdi, G. P., Robertson, A. M., The relation between flow rate and axial pressure gradient
for time-periodic Poiseuille flow in a pipe, J. Math. Fluid Mech. 7 Supplement No. 2 (2004),
215–223.
[5] Friedman, A., Partial Differential Equations, Holt, Rinehart & Winston Inc., New York
1969.
[6] Galdi, G. P., An Introduction to the Mathematical Theory of the Navier–Stokes Equations:
Linearized Steady Problems, Springer Tracts in Natural Philosophy, Vol. 38, Springer-Verlag,
New York, Revised Edition 1998.
[7] Ladyzhenskaya, O. A., The Mathematical Theory of Viscous Incompressible Flow, Gordon
and Breach, New York, London, Paris 1969.
[8] Ladyzhenskaya, O. A., Boundary Value Problems of Mathematical Physics, Springer-Verlag
1985.
[9] Ladyzhenskaya, O. A., Solonnikov, V. A., Ural’ceva, N. N., Linear and Quasilinear Equations
of Parabolic Type, American Mathematical Society, Providence, R.I. 1967.
[10] Landau, L. D., Lifschitz, E. M., Hydromechanik, Akademie-Verlag, Berlin 1971.
[11] Miller, R. K., Feldstein, A., Smoothness of solutions of Volterra integral equations with
weakly singular kernels, SIAM J. Math. Anal. 2 (1971), 242–258.
[12] Pileckas, K., On the behavior of the nonstationary Poiseuille solution as t → ∞, Siberian
Math. J. 46 (2005), 890–900.
[13] Pileckas, K., Keblikas, V., On the existence of nonstationary Poiseuille solution, Siberian
Math. J. 46 (2005), 514–526.
[14] Pileckas, K., Existence of solutions with the prescribed flux of the Navier–Stokes system in
an infinite cylindet, J. Math. Fluid. Mech. 8 (2006), 542–563.
[15] Prilepko, A. I., Orlovsky, D. G., Vasin, I. A., Methods for Solving Inverse Problems in
Mathematical Physics, Marcel Dekker, New York, Basel 1999.
[16] Solonnikov, V. A., On boundary value problems for linear parabolic systems of general form,
Trudy Mat. Inst. Steklov, 83 (1965), 1–162. English Transl.: Proc. Math. Inst. Steklov 83
(1965).
[17] Tricomi, F. G., Integral Equations, Intersience, New York 1957.
Vol. 58 (2007) On the unsteady Poiseuille flow in a pipe 1007

G. P. Galdi
Department of Mechanical Engineering
University of Pittsburgh
PA, USA

K. Pileckas
Institute of Mathematics and Informatics
Faculty of Mathematics and Informatics
Vilnius University
Vilnius
Lithuania

A. L. Silvestre
Centro de Matemática e Aplicações
Department of Mathematics
Instituto Superior Técnico
Lisbon
Portugal

(Received: September 12, 2006)

Published Online First: April 10, 2007

To access this journal online:


www.birkhauser.ch/zamp

You might also like