Thin-walled beams: a derivation of

Vlassov theory via Γ-convergence

Lorenzo Freddi ∗ Antonino Morassi†
Roberto Paroni‡

Abstract
This paper deals with the asymptotic analysis of the three-dimen-
sional problem for a linearly elastic cantilever having an open cross-
section which is the union of rectangles with sides of order ε and ε2 ,
as ε goes to zero. Under suitable assumptions on the given loads
and for homogeneous and isotropic material, we show that the three-
dimensional problem Γ-converges to the classical one-dimensional Vlas-
sov model for thin-walled beams.

2001 AMS Mathematics Classification Numbers: 74K20, 74B10, 49J45

Keywords: thin-walled cross-section beams, linear elasticity, Γ-convergence, di-

mension reduction

1 Introduction
In this paper we continue a line of research initiated in [4] which aims at a rigorous
variational deduction of the one-dimensional theory for thin-walled beams from the
three-dimensional linear elasticity.
In [4] we considered a thin-walled cantilever Ωε = ωε × (0, `) of length `, made
of homogeneous linear isotropic material, with a rectangular cross-section ωε of
∗
Dipartimento di Matematica e Informatica, via delle Scienze 206, 33100 Udine, Italy,
email: freddi@dimi.uniud.it
†
Dipartimento di Georisorse e Territorio, via Cotonificio 114, 33100 Udine, Italy, email:
antonino.morassi@uniud.it
‡
Dipartimento di Architettura e Pianificazione, Università degli Studi di Sassari,
Palazzo del Pou Salit, Piazza Duomo, 07041 Alghero, Italy, email: paroni@uniss.it

1
sides ε and ε2 . By working in the framework of Γ-convergence, see, for example,
[3] and [2], we proved that the three-dimensional elasticity problem converges in a
suitable variational sense to a one-dimensional problem as ε goes to zero. The limit
problem is defined by a functional which includes the extensional, the flexural and
the torsional strain energies of the classical thin-walled model of beam, as they
can be deduced from De Saint-Venant’s theory. In particular, the strain energy
density of the limit model can be written as a diagonal homogeneous quadratic
form of the longitudinal strain, the curvatures of the beam axis evaluated with
respect to the principal planes of bending, and the first derivative of the torsional
twist. Therefore, the equations of equilibrium show a full decoupling between
extensional, flexural and torsional effects.
The present paper extends the results obtained in [4] to the case of thin-
walled beams with open (i.e., simply connected) multi-rectangular cross-section.
S (i)
More precisely, we consider a three-dimensional cylinder Ωε = 3i=1 Ωε , where
(i) (i) (i)
Ωε = ωε × (0, `) and ωε is a rectangle having sides of order ε and ε2 . The
(i)
rectangles ωε partially overlap and are joined so as to achieve, for instance, cross-
sections having a >-like or @-like shape.
Under the assumption that a end cross-section of the three-dimensional body is
fixed and the material is homogeneous and isotropic, the theory of Γ-convergence
is used to study the asymptotic behavior of the energy functional as ε goes to zero.
The limit functional strongly depends on the geometry of the cross-section. In
particular, it is a homogeneous quadratic form of the longitudinal strain, of the
two bending curvatures of the beam axis, of the first derivative of the twist and
also of the square of the second derivative of the twist, in the case of a section
composed by at least three not aligned rectangles. This last term is concerned with
the so-called nonuniform torsional effects, which are responsible for the presence
of normal stresses induced by torsional deformations and, as a consequence, for
the coupling between extension, flexure and torsion. It should be remarked that
the above mentioned effects are not included in the classical De Saint-Venant’s
theory and have proved to be important in several engineering fields, especially in
aeronautical applications where the presence of open cross-sections formed by parts
with dimension of different order of magnitude requires more refined mechanical
beam models.
A direct analysis of the limit energy functional shows that the full decoupling
between extensional, flexural and torsional problems can be obtained by choosing
the axes of the reference system as the principal axes of the cross-section centered
in his center of mass, and by determining the sector coordinate function - which
is, roughly speaking, the limit of the warping function of the cross-section - with
respect to the shear center.

2
 The nonuniform torsion theory is well known in the engineering literature of
thin-walled beams since the old paper by Timoshenko [11] and the fundamental
contribution by Vlassov [12]. These approximate theories are usually based on
some a-priori assumptions on the deformation of the body and on the induced
stress field. Little attention has been given to the mathematical justification of
Vlassov’s theory; Rodriguez and Viaño, for example, presented in [10] an extension
of Vlassov’s theory for thin-walled beams as an asymptotic approximation of the
three-dimensional model as the area and the thickness of the cross-section are
assumed to tend to zero independently in a suitable order.
The paper is organized as follows. In Section 2 we introduce the three-di-
mensional problem and in Section 3 we rewrite it in a variational form on a fixed
domain. The proof of some compactness results for a scaled displacement field is
presented in Section 4. Section 5 is devoted to the establishment of the junction
conditions and to use them to characterize the essential kinematic fields of the
cross-section. Γ-convergence results are presented in Sections 6 and 7. The limit
energy functional is investigated in Section 8 and some examples are discussed in
Section 9. The contribution of the external loads and the strong convergence of
minimizers are studied in Section 10.
Notation. Throughout this article, and unless otherwise stated, we use the Ein-
stein summation convention and we index vector and tensor components as follows:
Greek indices α, β and γ take values in the set {1, 2} and Latin indices i, j, h in
the set {1, 2, 3}. Accordingly, α(i) will denote the parity of the index i, that is
the function which takes the value 1 if i is odd and the value 2 if i is even. The
component k of a vector v will be denoted either with (v)k or vk and an analogous
notation will be used to denote tensor components. Eαβ denotes the Ricci’s sym-
bol, that is E11 = E22 = 0, E12 = 1 and E21 = −1. L2 (A; B) and H s (A; B) are the
standard Lebesgue and Sobolev spaces of functions defined on the domain A and
taking values in B, with the usual norms k · kL2 (A;B) and k · kH s (A;B) , respectively.
When B = R or when the right set B is clear from the context, we will simply write
L2 (A) or H s (A), sometimes even in the notation used for norms. Convergence in
the norm, that is the so called strong convergence, will be denoted by → while
weak convergence is denoted with *.
With a little abuse of notation, and because this is a common practice and
does not give rise to any mistake, we use to call “sequences” even those families
indicized by a continuous parameter ε which, throughout the whole paper, will be
assumed to belong to the interval (0, 1].

3
2 The 3-dimensional problem
We consider a three-dimensional body which is at rest in the placement Ωε ⊂ R3 ,
(i)
where Ωε := ωε × (0, `), ωε := ∪3i=1 ωε , and

(i) s(i) (i) s(i)

ωε(i) := (εq1 + εd(i) − εb(i) , εq1 + εd(i) ) × (εq2 − ε2 , εq2 + ε2 ), i = 1, 3
2 2
s(2) s(2) (1) (3)
ωε(2) := (εq1 − ε2 , εq1 + ε2 ) × (εq2 , εq2 ),
2 2
(3) (1)
are three non-empty rectangles. Moreover we set h := q2 − q2 .

(1)
εb x1
(1)
εd
ωε(1) ε q2(1) x
2
2 (1)
ε s

ε q2(3)
ωε(2)
ε 2s (2) εh

ε 2 s (3) ωε(3)
(3)
εb
ε d (3) ε q1

For later convenience we also set

Ω(i) (i)
ε := ωε × (0, `), i = 1, 2, 3,
(1) (2) (3)
and we note that Ωε = Ωε ∪ Ωε ∪ Ωε and that they are not pairwise disjoint.

Remark 2.1 Within our framework we can cover also cross-sections having a >-
like shape by simply setting, for instance, s(3) = b(3) = 0 (of course, in this case
(3)
the rectangle ωε is empty). A section having a @-like shape can be achieved, for
instance, by replacing the quantities d(1) and d(3) in the figure above with other two
(1) (3) (1) (3)
parameters dε and dε satisfying the properties dε /ε → d(1) and dε /ε → d(3) ,
respectively, as ε goes to 0. The @-like shape is reached then by making the choice

4
 (1) (3)
d(1) = d(3) = s(2) /2 (and, for instance, dε = dε = εs(2) /2). Similarly one can
consider also sections with a or -like shape. Our analysis covers also these
more general settings but, for simplicity, we will concentrate ourselves to the case
displayed in the figure above.

In what follows we consider an homogeneous isotropic material, so that the

elasticity tensor C writes as

CA = 2µA + λ(trA) I

for every symmetric matrix A. Above, I denotes the 3 × 3 identity matrix. We

assume µ > 0 and λ ≥ 0 so to have, for every symmetric tensor A,

CA · A ≥ µ|A|2 , (1)

where · denotes the scalar product. We shall consider the spaces

1
© ª
H# (Ωε ; R3 ) := u ∈ H 1 (Ωε ; R3 ) : u = 0 on ωε × {0}

(i)
1 (Ω ; R3 ) defined in a similar way. We further denote by
and H# ε

Du(x) + DuT (x)

Eu(x) := sym(Du(x)) := ,
2
(2)
Du(x) − DuT (x)
Wu(x) := skw(Du(x)) := ,
2
the strain of u : Ωε → R3 and the skew symmetric part of the gradient Du.
We consider the following total energy functionals
Z
Fε (u) := Jε (u) − bε · u dx, (3)
Ωε

where Z
1
Jε (u) := CEu · Eu dx (4)
2 Ωε

are the bulk energies and bε ∈ L2 (Ωε ; R3 ) are the body forces.
Due to the coerciveness inequality (1) and the strict convexity of the integrand,
the total energy functionals Fε admit for every ε a unique minimizer among all
competing displacements u ∈ H# 1 (Ω ; R3 ). As already explained in the introduc-
ε
tion our aim is to study the asymptotic behavior of such minimizers as ε goes to
0, through the theory of Γ-convergence, for an account of it we refer to the books
of Braides [2] and Dal Maso [3].

5
3 The rescaled problem
(i)
To state our results it is convenient to stretch the domains Ωε along the transverse
directions x1 and x2 in suitable ways so that the transformed domains do not
(i)
depend on ε. Hereafter we denote by ω (i) := ω1 and let

p(i)
ε :Ω
(i)
→ Ω(i)
ε ,
(i)
be the (unique) affine invertible transformation between the sets Ω(i) and Ωε . It
turns out to be defined by
¡ α(i) (i) ¢
p(i)
ε (y1 , y2 , y3 ) = ε y1 + (ε − εα(i) )q1 , εα(i+1) y2 + (ε − εα(i+1) )q2 , y3

where α(i) is the parity of i, that is the function which takes the value 1 if i is odd
and the value 2 if i is even.

y1
(1)
b
(1) y2
εb
(1)
d q2(1)
x1
(1)
εd (1) (1)
ε q2 x2 s (1) ω
2 (1)
εs (3)
pε(1) q2
(3)
ε q2
ε 2s (2) εh pε(2) s
(2)

h
(2)
ω
ε 2s (3) pε(3)
(3)
εb s
(3)
(3)
(3) ω
εd ε q1
(3)
b
(3)
d q1

1 (Ω ; R3 ) we define three functions u(i) ∈ H 1 (Ω(i) ; R3 ) by

Given u ∈ H# ε #

u(i) := u ◦ pε(i) .

Of course, in the regions where the domains overlap the following two “junction
conditions” must be satisfied
−1 −1
u(i) ◦ pε(i) = u(2) ◦ p(2)
ε in Ω(i) (2)
ε ∩ Ωε , i = 1, 3. (5)

Let us consider the following 3 × 3 matrix valued differential operators

µ ¶
(i) (i) −1 D1 u D2 u
Hε u := Du ◦ Dpε = , , D3 u
εα(i) εα(i+1)

6
where Di u denotes the column vector of the partial derivatives of u with respect
to yi . We also set

Eε(i) u := sym( H(i)

ε u), W(i) (i)
ε u := skw( Hε u). (6)

Let us split the bulk energy Jε , defined in (4), into the sum of three energies each
one defined on a rectangular component Ω(i) ; more precisely
3
X Z
1
Jε (u) = Jε(i) (u), where Jε(i) (u) := χε (x) CEu · Eu dx,
2 Ωε
(i)
i=1

and χε : Ωε → {1/2, 1} is defined by

(
(1) (3) (2)
1/2 if x ∈ (Ωε ∪ Ωε ) ∩ Ωε
χε (x) :=
1 otherwise.
© ª
Let Aε := (u(1) , u(2) , u(3) ) ∈ ×3i=1 H#
1
(Ω(i) ; R3 ) : conditions (5) are satisfied ;
then we can consider the rescaled bulk energies Iε : Aε → [0, +∞) obtained by
(i)
rescaling each term Iε of the sum on the corresponding domain Ω(i) with the
suitable change of variable, that is
3
X
(1) (2) (3)
Iε (u ,u ,u ) := Iε(i) (u(i) ) (7)
i=1

where
Z
1 −1 1
Iε(i) (u(i) ) := 3 Jε(i) (u(i) ◦ p(i)
ε )= χ(i) (i) (i)
ε (y) CEε u · E(i) (i)
ε u dy, (8)
ε 2 Ω(i)

and
χ(i) (i)
ε := χε ◦ pε .

Note that
1
Iε (u(1) , u(2) , u(3) ) = Jε (u).
ε3

4 Compactness lemmata
In this section we establish the compactness of appropriately rescaled sequences of
displacements and prove that the limit functions are displacements of Bernoulli-
Navier type. The proof of the next lemma follows immediately from (1).

7
 (1) (2) (3)
Lemma 4.1 Let (uε , uε , uε ) be a sequence in Aε . If
1
sup I (u(1) , u(2)
4 ε ε
(3)
ε , uε ) < +∞, (9)
ε∈(0,1] ε

then there exists a constant C > 0 such that

3
X
kE(i) (i)
ε uε kL2 (Ω(i) ;R3×3 ) ≤ Cε
2
(10)
i=1

for every ε ∈ (0, 1].

To prove the compactness of the displacements we need the following scaled Korn
inequalities (already obtained in [4] in the particular case i = 2).

Theorem 4.2 Let i ∈ {1, 2, 3}. There exists a constant C > 0 such that
Z ³ ´ Z
u1 u2 u3 2 (i) 2 C
|( α(i+1)−1 , α(i)−1 , 2 )| + |Hε u| dy ≤ 4 |E(i) 2
ε u| dy (11)
Ω(i) ε ε ε ε Ω (i)

1 (Ω(i) ; R3 ) and every ε ∈ (0, 1].

for every u ∈ H#

(i)
Proof. Let us divide the section ωε in squares of size ε2 and apply Korn’s
inequality (the one obtained by Anzellotti, Baldo and Percivale in [1]; see also [9]
and Kondrat’ev and Oleinik [6], Theorem 2) to each beam of length ` and with
section a square with side proportional to ε2 . Then, summing over all the obtained
inequalities we have that there exists a constant C > 0 such that
Z Z
¡ 2 ¢ C
|u| + |Du|2 dx ≤ 4 |Eu|2 dx (12)
(i)
Ωε ε Ωε
(i)

(i)
1 (Ω ; R3 ).
for every u ∈ H# ε
R (i) R (i)
The inequality Ω(i) |Hε u|2 dy ≤ εC4 Ω(i) |Eε u|2 dy is simply obtained by
rescaling inequality (12). To show that
Z Z
u1 u2 u3 C
|( α(i+1)−1 , α(i)−1 , 2 )|2 dy ≤ 4 |E(i) 2
ε u| dy,
Ω (i) ε ε ε ε Ω(i)

¡ u1 u2 u3 ¢
it suffices to set vε := α(i+1)−1
,
α(i)−1
, 2 , notice that |Eε u| ≥ ε2 |Evε | and
ε ε ε
apply the standard Korn inequality to vε on the domain Ω (see for instance [8],
Theorem 2.7). 2

8
 Let us consider the usual space of Bernoulli-Navier displacements on Ω(i)
n o
1
HBN (Ω(i) ; R3 ) := v ∈ H# (Ω(i) ; R3 ) : (Ev)jα = 0, j = 1, 2, 3, α = 1, 2 (13)

and set
HBN := ×3i=1 HBN (Ω(i) ; R3 ). (14)
Analogously we denote by

L2 := ×3i=1 L2 (Ω(i) ) and H 1 := ×3i=1 H 1 (Ω(i) ; R3 ).

Inequality (11) motivates the introduction of the following scaling operators

³ u1 u2 u3 ´
Sε(i) u := α(i+1)−1 , α(i)−1 , 2 . (15)
ε ε ε
(1) (2) (3)
Lemma 4.3 Let (uε , uε , uε ) be a sequence in Aε which satisfies (10). Then,
for any sequence of positive numbers εn converging to 0 there exist a subsequence
(not relabeled) and functions (v(1) , v(2) , v(3) ) ∈ HBN and (ϑ(1) , ϑ(2) , ϑ(3) ) ∈ L2
such that, as n goes to ∞,

S(i) (i)
εn uεn * v
(i)
in H 1 (Ω(i) ; R3 ), (16)

(W(i) (i)
εn uεn )12 * −ϑ
(i)
in L2 (Ω(i) ). (17)

Proof. It is convenient to set v(i) (i) (i)

ε := Sε uε . It is easily checked that, for ε ≤ 1,
(i) (i) (i) (i)
|Eε uε | ≥ ε2 |Evε |, hence, by (10), Evε is uniformly bounded in L2 (Ω(i) ; R3×3 ),
(i)
and by Korn’s inequality vε is uniformly bounded in H 1 (Ω(i) ; R3 ). It then exists a
v(i) ∈ H# 1 (Ω(i) ; R3 ), and a subsequence (not relabeled) of ε such that v(i) * v(i)
n εn
(i) (i) (i)
in H 1 (Ω(i) ; R3 ). Again, it is easy to check that |(Eε uε )jα | ≥ ε|(Evε )jα | for
(i)
j = 1, 2, 3 and α = 1, 2, thus, using (10), we deduce that Cε ≥ k(Evε )jα kL2 (Ω(i) )
and consequently (Ev(i) )jα = 0. Hence v ∈ HBN (Ω(i) ; R3 ).
Using assumption (10) together with Theorem 4.2 we obtain that the sequence
(i) (i)
Hεn uεn is bounded in L2 so that, up to subsequences, it weakly converges in
(i) (i)
L2 (Ω(i) ; R3×3 ) to some H(i) ∈ L2 (Ω(i) ; R3×3 ). Since, from (10), Eεn uεn → 0 in
(i) (i)
L2 (Ω(i) ; R3×3 ) we have Wεn uεn → H(i) weakly in L2 (Ω(i) ; R3×3 ). In particular,
H(i) is, almost everywhere, a skew-symmetric matrix. Denoting (H(i) )12 = −ϑ(i)
we obtain (17). 2

9
Remark 4.4 Using the same notation as in the proof of Lemma 4.3, and since
(i) (i) (i) (i) (i) (i) (i) (i)
(Hε uε )13 = D3 uε1 = εα(i+1)−1 D3 vε1 , and (Hε uε )23 = D3 uε2 = εα(i)−1 D3 vε2
we note that we have also
¡ ¢ (i) ¡ ¢ (i)
(H(i) )13 = α(i) − 1 D3 v1 and (H(i) )23 = α(i + 1) − 1 D3 v2 .

By Lemma 4.3, the displacements v(i) are of Bernoulli-Navier type and there-
fore they can be written as
(i) (i) 0
vα(i) = ξα(i) (y3 ), α = 1, 2, v3 = ξ3 (y3 ) − yα ξα(i) (y3 ), (18)

where
ξα(i) ∈ H#
2
(0, `) := {ξ ∈ H 2 (0, `) : ξ(0) = ξ 0 (0) = 0}
and
(i) 1
ξ3 ∈ H# (0, `).
The notation established in this section, in particular in Lemma 4.3 and in (18),
will be used throughout the rest of the paper.

5 Junction conditions
The present section is devoted to establish the relationship existing between the
limit fields (v(1) , v(2) , v(3) ) and (ϑ(1) , ϑ(2) , ϑ(3) ) introduced in Lemma 4.3. Hence,
(1) (2) (3)
during the whole section, we assume that (uε , uε , uε ) be a sequence in Aε
which satisfies (10), that εn be a sequence of positive numbers converging to 0 and
that a subsequence (not relabeled) and triples of functions (v(1) , v(2) , v(3) ) ∈ HBN
and (ϑ(1) , ϑ(2) , ϑ(3) ) ∈ L2 have been chosen in order to satisfy (16) and (17). Let
us moreover use notation (18).
To find relations between the limit fields we study the junction conditions, that
is the system (5), by adapting some inspiring ideas of Le Dret [7] and [5]. Since

−1 ¡ x1 + (εα(i) − ε)q1 x2 + (εα(i+1) − ε)q2(i) ¢

p(i)
ε (x1 , x2 , x3 ) = α(i)
, α(i+1)
, x3 ,
ε ε
the two junction conditions can be equivalently written as
¡ ¢ ¡ (i) (i) ¢
u(i) ε(z1 − q1 ) + q1 , z2 , z3 = u(2) z1 , ε(z2 − q2 ) + q2 , z3 ,
(19)
z ∈ Ω(i) ∩ Ω(2) , i = 1, 3,

10
where

s(2) s(2) (1) (1) s(1)

Ω(1) ∩ Ω(2) = (q1 − , q1 + ) × (q2 , q2 + ) × (0, `),
2 2 2

s(2) s(2) (3) s(3) (3)

Ω(3) ∩ Ω(2) = (q1 −, q1 + ) × (q2 − , q2 ) × (0, `).
2 2 2
It is worth notice that in this way also the two junction regions, which originally
depend on ε, have been trasformed into the fixed domains Ω(i) ∩ Ω(2) , i = 1, 3.
The following lemma is stated for Ω(1) ∩ Ω(2) but, with straightforward adap-
tations, it holds also for Ω(3) ∩ Ω(2) .

Lemma 5.1 Let w ∈ H 1 (Ω(1) ∩ Ω(2) ) and wε ∈ H 1 (Ω(1) ∩ Ω(2) ) be a sequence such
that
wε * w in H 1 (Ω(1) ∩ Ω(2) ).
Then the sequence of functions
Z q1 +s(2) /2
¡ ¢
(z2 , z3 ) 7→ − wε ε(z1 − q1 ) + q1 , z2 , z3 dz1
q1 −s(2) /2

(1) (1)
converges in the norm of L2 ((q2 , q2 + s(1) /2) × (0, `)) to the trace of the function
(1) (1)
w on {q1 } × (q2 , q2 + s(1) /2) × (0, `). We will denote such a trace simply by
w(q1 , z2 , z3 ).

Proof. As
Z `Z q2 +s(1) /2 ¯Z q1 +s(2) /2
(1)
¯2
¯ ¡ ¢ ¯
(1)
¯− wε ε(z1 − q1 ) + q1 , z2 , z3 dz1 − wε (q1 , z2 , z3 )¯ dz2 dz3 =
0 q2 q1 −s(2) /2
Z `Z q2 +s(1) /2 ¯Z q1 +εs(2) /2
(1)
¯2
¯ ¯
= ¯− wε (z1 , z2 , z3 ) − wε (q1 , z2 , z3 ) dz1 ¯ dz2 dz3
(1)
0 q2 q1 −εs(2) /2
Z `Z (1)
q2 +s(1) /2 ¯Z q1 +εs(2) /2 Z z1 ¯2
¯ ¯
= ¯− D1 wε (t, z2 , z3 ) dt dz1 ¯ dz2 dz3
(1)
0 q2 q1 −εs(2) /2 q1
Z `Z q2 +s(1) /2 Z q1 +εs(2) /2
(1)

≤ s(2) ε |D1 wε (t, z2 , z3 )|2 dtdz2 dz3

(1)
0 q2 q1 −εs(2) /2
≤ s(2) εkD1 wε k2L2 (Ω(1) ∩Ω(2) ) ≤ Cε,

the claim follows by continuity of the trace. 2

11
Lemma 5.2 The following equalities hold for almost every y3 ∈ (0, `).
(i)
1. ξ2 (y3 ) = 0, i = 1, 3;
(2)
2. ξ1 (y3 ) = 0;
(i) (i) 0 (2) (i) (2) 0
3. ξ3 (y3 ) − q1 ξ1 (y3 ) = ξ3 (y3 ) − q2 ξ2 (y3 ), i = 1, 3.

(1) (2)
Proof. As the pair (uεn , uεn ) satisfies (19) with i = 1, averaging the second
components with respect to z1 we have
Z q1 +s(2) /2
(1) ¡ ¢
− uεn 2 εn (z1 − q1 ) + q1 , z2 , z3 dz1 =
q1 −s(2) /2
Z q1 +s(2) /2
(2) ¡ (1) (1) ¢
=− uεn 2 z1 , εn (z2 − q2 ) + q2 , z3 dz1 .
q1 −s(2) /2

(1)
Applying Lemma 5.1 to the sequence uεn 2 and using (16), we deduce that the
(1)
left hand side of the equality above converges to (z2 , z3 ) 7→ v2 (q1 , z2 , z3 ) in
(1) (1)
L2 ((q2 , q2 + s(1) /2) × (0, `)). On the other hand the right hand side converges
to zero in the same space, indeed
Z `Z q2 +s(1) /2 ¯Z q1 +s(2) /2
(1)
¯2
¯ (2) ¡ (1) (1) ¢ ¯
(1)
¯− uεn 2 z1 , εn (z2 − q2 ) + q2 , z3 dz1 ¯ dz2 dz3 ≤
0 q2 q1 −s(2) /2
Z ` Z q1 +s(2) /2 Z (1)
q2 +s(1) /2
(2) ¡ (1) (1) ¢
≤ − |uεn 2 z1 , εn (z2 − q2 ) + q2 , z3 |2 dz2 dz1 dz3
(1)
0 q1 −s(2) /2 q2
Z ` Z q1 +s(2) /2 Z q(1) +εn s(1) /2
1 2
(2)
≤ − |uεn 2 (z1 , z2 , z3 )|2 dz2 dz1 dz3
(2)
0 q1 −s /2 ε n q2 (1)

Z ¯ (2) ¯
εn ¯ uεn 2 ¯2
≤ (2) ¯ ¯ dz
s Ω(2) εn

and the last term of the chain tends to zero due to (16). Taking into account (18),
this proves 1 for i = 1. The case i = 3, follows by the same argument applied to
(2)
the other junction condition. The item 2 of the statement, concerning ξ1 , can be
proved similarly by considering the first component of (19) with i = 1.
In order to prove 3 we consider the following scaled average of the third com-

12
ponent of (19) with i = 1
Z q(1) +s(1) /2 Z q1 +s(2) /2 (1) ¡ ¢
2 uεn 3 εn (z1 − q1 ) + q1 , z2 , z3
− − dz1 dz2 =
(1)
q2 q1 −s(2) /2 ε2n
Z q1 +s(2) /2 Z q(1) +s(1) /2 (2) ¡ (1) (1) ¢
2 uεn 3 z1 , εn (z2 − q2 ) + q2 , z3
=− − dz2 dz1
q1 −s(2) /2 q2
(1) ε2n
and by applying twice Lemma 5.1 and (16) we deduce
Z q(1) +s(1) /2 Z q1 +s(2) /2
2
(1) (2) (1)
− v3 (q1 , z2 , z3 ) dz2 = − v3 (z1 , q2 , z3 ) dz1 .
(1)
q2 q1 −s(2) /2

Taking into account (18) and the statements 1 and 2 of the present lemma we
deduce item 3 for i = 1. The case i = 3 follows similarly by considering the second
junction condition. 2

To deduce further “limit” junction conditions we need the following two-dimen-

sional Korn’s inequality (see [8], Theorem 2.5)
kw − ℘wkH 1 (ω;R2 ) ≤ C kEwkL2 (ω;R2×2 ) , (20)
which holds for all w ∈ H 1 (ω; R2 ) and where ω is any Lipschitz bounded subset of
R2 . If we denote with (y1 (G), y2 (G)) the center of mass of ω, we have that the α-
component of the projection of w on the space of “two-dimensional” infinitesimal
rigid displacements, see [4], is (with the summation convention)
℘wα = tα (w) + Eβα (yβ − yβ (G))ϑ(w) (21)
where
Z
1
ϑ(w) = Eγδ (yγ − yγ (G))wδ dy1 dy2 ,
IG (ω) ω
Z
1
tα (w) = wα dy1 dy2 .
|ω| ω
Above IG (ω) denotes the polar moment of inertia of the section ω with respect to
the center of mass,
Z
IG (ω) := (y1 − y1 (G))2 + (y2 − y2 (G))2 dy1 dy2 .
ω

Analogously, IG (ω (i) )
will denote the polar moment of inertia with respect to the
center of mass (y1 (G ), y2 (G(i) ) of the section ω (i) . For later convenience we set
(i)

(i) (i)
Sε uε
w(i)
ε := (22)
ε

13
and Z
1 (i)
ϑ(i)
ε := Eγδ (yγ − yγ (G(i) ))wεδ dy1 dy2 ,
IG (ω (i) ) ω (i)
so that
(℘w(i) (i) (i) (i)
ε )α = tα (wε ) + Eβα (yβ − yβ (G ))ϑε . (23)

Lemma 5.3 There exists a constant C > 0 such that

kwε(i) − ℘w(i)
ε kL2 (0,`;H 1 (ω (i) ;R2 )) ≤ Cε,

for every ε ∈ (0, 1].

i i+1
Proof. Since (Ewε(i) )11 = ε(−1) (Eε(i) u(i) (i)
ε )11 , (Ewε )22 = ε
(−1)
(E(i) (i)
ε uε )22 , and
(Ew(i) (i) (i)
ε )12 = (Eε uε )12 , we have

1
k(Ew(i) (i) (i)
ε )αβ kL2 (Ω(i) ) ≤ k(Eε uε )αβ kL2 (Ω(i) ) . (24)
ε
Hence, taking into account (20), (24) and (10), we have
Z ` Z `X
kw(i)
ε − ℘wε(i) k2H 1 (ω(i) ;R2 ) dy3 ≤C k(Ew(i) 2
ε )αβ kL2 (ω (i) ) dy3
0 0 αβ
C X
≤ 2 k(E(i) (i) 2 2
ε uε )αβ kL2 (Ω(i) ) ≤ Cε ,
ε
αβ

which concludes the proof. 2

Lemma 5.4 We have

1. ϑ(i)
εn * ϑ
(i)
in L2 (Ω(i) ). Therefore, ϑ(i) does not depend on y1 and y2 ;

2. ϑ(i) ∈ H#
1 (0, `);

3. ϑ(1) (y3 ) = ϑ(2) (y3 ) = ϑ(3) (y3 ) =: ϑ(y3 ) for a.e. y3 in (0, `).

Proof. Let’s prove 1. From Lemma 5.3 we have that

kDα (w(i) (i)

ε − ℘wε )kL2 (Ω(i) ) ≤ Cε. (25)

(i) (i) (i) (i) (i)

Since (W℘wε )12 = −ϑε and (Wwε )12 = (Wε uε )12 , from the identity

ϑ(i) (i) (i) (i) (i) (i)

ε = −(W℘wε )12 = −(Wε uε )12 + (W(wε − ℘wε ))12 , (26)

14
and using (25), we get the following estimate
(i) (i) (i) (i) (i)
kϑε + (Wε uε )12 kL2 (Ω(i) ) = k(W(wε − ℘wε ))12 kL2 (Ω(i) ) ≤ Cε. (27)

The claim 1 follows then by taking into account (17) and from the fact that, by
(i)
definition, the functions ϑεn do not depend on y1 and y2 .
Part 2 of the statement will be proven by following the same argument of
Lemma 4.6 of [4]. Let ξ ∈ C0∞ (ω (i) ) be such that
Z
IG (ω (i) )
ξ dy1 dy2 = − .
ω (i) 2

Then, taking into account (23), we have

Z Z
IG (ω (i) )ϑ(i)
ε = −2ϑ (i)
ε ξ dy 1 dy 2 = −ϑ(i)
ε ξDα yα dy1 dy2
ω (i) ω (i)
Z Z
(i) (i)
= ϑε yα Dα ξ dy1 dy2 = ϑε Eαγ Eβγ yβ Dα ξ dy1 dy2
Z ω (i) ω (i)

= Eαγ (Eβγ yβ ϑ(i) ε )Dα ξ dy1 dy2

(i)
Zω ³ Z ´
(i) 1 (i)
= Eαγ (℘wε )γ − (i) wεγ dy1 dy2 Dα ξ dy1 dy2
(i) |ω | ω(i)
Zω
= Eαγ (℘w(i) ε )γ Dα ξ dy1 dy2
ω (i)
Z Z
(i)
= Eαγ wεγ Dα ξ dy1 dy2 − Eαγ (w(i) (i)
ε − ℘wε )γ Dα ξ dy1 dy2 .
ω (i) ω (i)

Hence, denoting by
Z
1
ϑ̃(i)
ε :=
(i)
Eαγ wεγ Dα ξ dy1 dy2 ,
IG (ω (i) ) ω (i)

and recalling (25), we find

ϑε(i) − ϑ̃(i) 2 (i)

ε → 0 in L (Ω ). (28)
(i)
We now show that D3 ϑ̃ε is bounded in L2 . Since Eαγ Dα Dγ ξ = 0 in ω (i) and

15
Dα ξ = 0 on ∂ω (i) , we have
Z
(i) (i) (i)
IG (ω )D3 ϑ̃ε = Eαγ Dα ξ D3 wεγ dy1 dy2
ω (i)
Z Z
ε (i)
=2 Eαγ Dα ξ (Ew )γ3 dy1 dy2 − Eαγ Dα ξ Dγ wε3 dy1 dy2
(i) (i)
Zω Zω
(i) (i)
=2 Eαγ Dα ξ (Ewε )γ3 dy1 dy2 − Dγ (Eαγ Dα ξ wε3 ) dy1 dy2
ω (i) (i)
Zω
(i)
+ Eαγ Dα Dγ ξwε3 dy1 dy2
ω (i)
Z
(i)
=2 Eαγ Dα ξ (Ewε )γ3 dy1 dy2 ,
ω (i)

but (Ewε )13 = (Eε uε )13 /εα(i+1) and (Ewε )23 = (Eε uε )23 /εα(i) , and therefore
(i)
D3 ϑ̃ε is bounded in L2 (0, `). Thus, from (28) and Lemma 5.4 we conclude that

ϑ̃(i)
ε *ϑ
(i)
in H 1 (Ω(i) ).
(i)
Therefore, since ϑ̃ε (0) = 0, we conclude that ϑ(i) ∈ H#1 (0, `), that is 2.

In order to prove 3, let us prove the equality ϑ(i) (y3 ) = ϑ(2) (y3 ) for i = 1, 3.
By differentiating (5) we find
(i) (i) −1 (2) (2) −1
εn D1 uεn ◦ pεn = D1 uεn ◦ pεn
in Ω(i) (2)
εn ∩ Ωεn
(i) (i) −1 (2) (2) −1
D2 uεn ◦ pεn = εn D2 uεn ◦ pεn

which immediately lead to

¡ (i) (i) ¢ −1 ¡ ¢ (2) −1
Wεn uεn 12 ◦ p(i)
εn = W(2) (2)
εn uεn 12 ◦ pεn in Ω(i) (2)
εn ∩ Ωεn .

Then, from the equality

Z h¡ i
(2) (i)
¢ (2) −1
ϑεn (z3 ) − ϑεn (z3 ) = − W(2) u(2)
εn εn 12 ◦ pεn + ϑ(2)
εn (z3 ) dx1 dx2 +
(i) (2)
ωεn ∩ωεn
Z h¡ i
¢ (i) −1
−− W(i) u(i)
εn εn 12 ◦ pεn + ϑ(i)
εn 3 dx1 dx2
(z )
(i) (2)
ωεn ∩ωεn
Z h¡ i
ε3n ¢
= (i) (2)
W(2) u(2)
εn εn 12 + ϑ(2)
εn (z3 dy1 dy2 +
)
(2) −1 (i)
|ωεn ∩ ωεn | pεn (ωεn )∩ω (2)
Z h¡ i
ε3n ¢
− (i) (2)
W(i) (i) (i)
εn uεn 12 + ϑεn (z3 ) dy1 dy2 ,
(i) −1 (2)
|ωεn ∩ ωεn | ω (i) ∩pεn (ωεn )

16
 (i) −1 (2) (2) −1 (i)
where, with a small abuse of notation, pεn (ωεn ) and pεn (ωεn ) denote the
inverse of the restriction to ω (i) and ω (2) , respectively, of the projection on the
(i) (2)
first two factors of pεn and pεn .
(i) (2)
As |ωεn ∩ ωεn | = s(i) s(2) ε4n , and using Hölder’s inequality, we have
Z `
|ϑ(i) (2)
εn − ϑεn | dy3 ≤
0
Z `Z ¯¡ ¯
1 ¯ (2) (2)
¢ (2) ¯
≤ (i) (2) −1
¯ W u
εn εn 12 + ϑεn (y3 )¯ dy1 dy2 dy3 +
s s εn 0 pεn (ωεn )∩ω(2)
(2) (i)
Z `Z ¯¡ ¯
1 ¯ (i) (i)
¢ (i) ¯
+ (i) (2) −1
¯ Wεn uεn 12 + ϑεn (y3 )¯ dy1 dy2 dy3
s s εn 0 ω(i) ∩pεn (ωεn )
(i) (2)

εn ³ ¡ (2) (2) ¢ ´
1/2
¡ (i) (i) ¢
≤C k Wεn uεn 12 + ϑ(2) (i)
εn kL2 (Ω(2) ) + k Wεn uεn 12 + ϑε kL2 (Ω(i) )
εn
≤ Cε1/2
n ,

where the estimates (27) have been used to conclude the computation. Hence the
right hand side goes to zero as εn → 0 while the liminf of the left hand side is
greater then the L1 -norm of ϑ(i) − ϑ(2) . Thus, ϑ(i) = ϑ(2) . 2

5.1 The case without any junction

It is the simplest case of the rectangular cross-section which has been considered
in detail in [4]. It can be obtained as a particular case of the present setting by
taking b(i) = s(i) = d(i) = 0 for i = 1, 3.

5.2 The case with only one junction

This case arises, for instance, when b(3) = s(3) = d(3) = 0 and turns out to be sim-
pler than the general case. Indeed the displacement fields v(1) and v(2) , by (18),
(i)
can be written in terms of six fields ξj , i = 1, 2, j = 1, 2, 3, which depend only
on the coordinate y3 . These six fields can be reduced, by using Lemma 5.2 to
only three fields, which together with the rotation angle ϑ will fully describe the
kinematics of the beam.

Lemma 5.5 We have that

ϑ = ϑ(1) = ϑ(2) ∈ H#
1
(0, `),

17
 2 (0, `) and η ∈ H 1 (0, `) such that
and there exist η1 , η2 ∈ H# 3 #
(1) (1) (1) (1)
v1 = η1 (y3 ), v2 = 0, v3 = η3 (y3 ) − y1 η10 (y3 ) − q2 η20 (y3 ),
(29)
(2) (2) (2)
v1 = 0, v2 = η2 (y3 ), v3 = η3 (y3 ) − q1 η10 (y3 ) − y2 η20 (y3 ).
for almost every y3 ∈ (0, `).
Proof. The first part of the statement has been already proven in Lemma 5.4. It
(2) (1)
remains to show that equalities (29) hold. That v1 = v2 = 0 follows from (18)
(1) (2)
and Lemma 5.2, and that v1 = η1 (y3 ) and v2 = η2 (y3 ) follows after setting
(1) (2)
η1 = ξ1 and η2 = ξ2 . From 3 of Lemma 5.2 follows that
(1) (2) (1)
ξ3 − q1 η10 = ξ3 − q2 η20 ,
thus setting
(2) (1) (2)
η3 := ξ3 + q1 η10 = ξ3 + q2 η20 ,
(1) (2)
we deduce the desired expressions of v3 and v3 . 2

5.3 The case with two junctions

In the case with a beam section composed by three rectangles the displacement of
(i)
the beam is described by the rotation angle ϑ and by nine fields ξj , i, j = 1, 2, 3,
all depending only on the coordinate y3 . Lemma 5.2 gives us five conditions to
which these fields have to satisfy. Thus, using these five conditions, we can reduce
the ten original fields to only five fields and not to four as in the case of only one
junction. This fact points out that we are still missing a junction condition.
The next lemma holds only if there are two distinct junctions.
Lemma 5.6 We have that
(1) (3)
ξ1 (y3 ) − ξ1 (y3 ) = hϑ(y3 )
2 (0, `).
for almost every y3 ∈ (0, `). Moreover ϑ ∈ H#
Proof. From the first component of (19), rescaling, integrating and making a
change of variables, we obtain for a.e. z3 ∈ (0, `)
Z q1 +εn s(2) /2 Z q(1) +s(1) /2 (1) Z q1 +s(2) /2 Z q(1) +εn s(1) /2 (2)
2 uεn 1 2 uεn 1
− − dz2 dz1 = − − dz2 dz1 ,
q1 −εn s(2) /2 q2
(1) εn q1 −s(2) /2 q2
(1) εn
Z q1 +εn s(2) /2 Z q(3) (3) Z q1 +s(2) /2 Z q(3) (2)
2 uεn 1 2 uεn 1
− − dz2 dz1 = − − dz2 dz1 .
q1 −εn s(2) /2 q2 −s(3) /2 εn q1 −s(2) /2 q2 −εn s(3) /2 εn
(3) (3)

(30)

18
Recalling (22), the difference of the right hand sides of the two equations above
can be rewritten as
Z q1 +s(2) /2 Z q(1) +εn s(1) /2 Z q(3)
¡ 2 (2) 2
(2) ¢
− − wεn 1 dz2 − − wεn 1 dz2 dz1 =
(1) (3)
q1 −s(2) /2 q2 q2 −εn s(3) /2
Z q1 +s(2) /2 ³Z q(1) +εn s(1) /2
2
(2)
=− − wεn 1 − (℘w(2)
εn )1 dz2 +
(1)
q1 −s(2) /2 q2
Z q(3) ´
2
(2)
−− wεn 1 − (℘w(2)
εn 1) dz2 dz1 +
(3)
q2 −εn s(3) /2
Z q1 +s(2) /2 Z q(1) +εn s(1) /2 Z q(3)
¡ 2 (2)
2 ¢
+− − (℘wεn )1 dz2 − − (℘w(2)
εn )1 dz2 dz1 ,
(1) (3)
q1 −s(2) /2 q2 q2 −εn s(3) /2

Z
¡ q2(1) + q2(3) ¢ (2) (2)
but (℘w(2)
εn )1 (z) = − z2 ϑεn + − wεn 1 dz1 dz2 and therefore
2 ω (2)

Z q1 +s(2) /2 Z q(1) +εn s(1) /2 Z q(3)

¡ 2 (2)
2 ¢
− − (℘wεn )1 dz2 − − (℘w(2)
εn )1 dz2 dz1 =
(1) (3)
q1 −s(2) /2 q2 q2 −εn s(3) /2
s(1) + s(3) (2)
= h ϑ(2)
εn − εn ϑεn .
2
Taking into account the identities above and using Lemma 5.3, from the difference
of (30) we deduce
Z ` ¯Z q1 +εn s(2) /2 ³Z q(1) +s(1) /2 (1) Z q(3) (3) ´
¯ 2 uεn 1 2 uεn 1
lim ¯− − dz2 − − dz2 dz1 +
n→∞ 0 q1 −εn s(2) /2
(1)
q2 εn q2 −s(3) /2 εn
(3)
¯
¯
−h ϑ(2)
εn ¯ dz3 = 0

and, applying Lemma 5.1, we get

Z ` ¯Z q(1) +s(1) /2 Z q(3) ¯
¯ 2 (1) 2
(3) ¯
¯−(1) v1 (q1 , z2 , z3 ) dz2 − −
(3)
v1 (q1 , z2 , z3 ) dz2 − hϑ¯ dy3 = 0.
0 q2 q2 −s̄/2

The statement of the lemma follows from (18). 2

The next lemma, which summarizes the results for a beam with a section
having two junctions, states that the transverse displacement of the cross section
is described by a rigid translation, of components η1 and η2 , and a rotation ϑ
around a point c.

19
Lemma 5.7 For every c = (c1 , c2 ) ∈ R2 there exist η1 , η2 ∈ H#
2 (0, `) and η ∈
3
1
H# (0, `) such that

(i) (i) (i)

v1 = η1 (y3 ) − (q2 − c2 )ϑ(y3 ), v2 = 0, i = 1, 3,
(2) (2)
v1 = 0, v2 = η2 (y3 ) + (q1 − c1 )ϑ(y3 ),

(i) (i)
v3 = η3 (y3 ) − y1 η10 (y3 ) − q2 η20 (y3 ) + ψ (i) (y1 )ϑ0 (y3 ), i = 1, 3,
(2)
v3 = η3 (y3 ) − q1 η10 (y3 ) − y2 η20 (y3 ) + ψ (2) (y2 )ϑ0 (y3 ),
(31)
where the so-called “sector coordinates”
(i) (i) (i)
ψ (i) (y1 ) := y1 (q2 − c2 ) − q1 (q2 − c2 ) − (q1 − c1 )q2 + K, i = 1, 3,

ψ (2) (y2 ) := −(q1 − c1 )y2 + K,

(1) (3)
were y1 ∈ (q1 +d(i) −b(i) , q1 +d(i) ) and y2 ∈ (q2 , q2 ), are defined up to an additive
constant K.
(2) (1) (3)
Proof. The equalities v1 = v2 = v2 = 0 follow from (18) and Lemma
(1) (1)
5.2. From (18) and Lemma 5.6, setting η1 := ξ1 + (q2 − c2 )ϑ, we deduce
(1) (1) (3) (1) (3)
v1 = η1 − (q2 − c2 )ϑ and ξ1 = η1 − (q2 + h − c2 )ϑ = η1 − (q2 − c2 )ϑ. Thus
(3) (3) (2) (2)
v1 = η1 −(q2 −c2 )ϑ. Setting η2 := ξ2 −(q1 −c1 )ϑ, we have v2 = η2 +(q1 −c1 )ϑ.
2 (0, `).
Moreover, by definition, η1 , η2 ∈ H#
Let K be any constant. Setting
(2)
η3 := ξ3 + q1 η10 − Kϑ0 ,

1 (0, `).(2)
from (18) we immediately obtain the desired expression for v3 and η3 ∈ H#
From statement 3 of Lemma 5.2 we deduce
(i) (i) ¡ (i) (i) ¢
ξ3 = η3 − q2 η20 + − q1 (q2 − c2 ) − (q1 − c1 )q2 + K ϑ0 ,
(i)
from which follow the expressions for v3 , i = 1, 3. 2

We finally note that the sector coordinates depend on the coordinates of the
point c and an additive constant K. For simplicity of notation, in the sequel we
do not explicitly stress this dependence.

20
Remark 5.8 The main difference between Lemma 5.5 and 5.7, which respectively
hold for one and two junction conditions, is that in the former ϑ is only once
differentiable while in the latter it is twice differentiable. This higher regularity
will be used to construct the recovery sequence in Section 7. On the other hand,
the same procedure can be applied also in the case of only one junction by simply
assuming ϑ smooth at the starting step (38).
We also note that formally we can deduce the displacements for a beam with
(1)
only one junction from Lemma 5.7 by setting c1 = q1 , c2 = q2 and K = 0.
Remark 5.9 The same technique can be used to treat hollow cross-sections. In
this case it can be proved that the twist angle ϑ vanishes, which suggests that
a different scaling of the rotation of the section, at level ε, is needed. This will
require a further study.

6 A liminf inequality
In this section we deduce a lower bound for the limiting energy. We start by
deducing the convergence of some of the components of the rescaled strain. The
results of this section hold, with straightforward adaptations, also in the case of
only one junction.
Lemma 6.1 We have, up to subsequences,
(i) (i)
(Eεn uεn )33 (i) (i)
* D3 v3 = η30 − y1 η100 − q2 η200 + ψ (i) (y1 )ϑ00 in L2 (Ω(i) ), i = 1, 3,
ε2n
(2) (2)
(Eεn uεn )33 (2)
* D3 v3 = η30 − q1 η100 − y2 η200 + ψ (2) (y2 )ϑ00 in L2 (Ω(2) ),
ε2n
(i) (i)
(Eεn uεn )13 (i)
* −(y2 − q2 )ϑ0 + η (i) in L2 (Ω(i) ), i = 1, 3,
ε2n
(2) (2)
(Eεn uεn )23
* +(y1 − q1 )ϑ0 + η (2) in L2 (Ω(2) ),
ε2n
(32)
where η (i)
∈ L2 (Ω(i) ), i = 1, 3, are independent of y2 , while η (2) ∈ L2 (Ω(2) )is
independent of y1 .
Proof. To prove the first and the second of the (32) it suffices to notice that
(i)
(E(i) (i) 2
ε uε )33 = D3 uε3 , divide by ε and apply (16).
The statements concerning the quantities defined on Ω(2) are proven in [4],
Lemma 4.7. The similar statements on the quantities defined on Ω(i) , i = 1, 3, can
be obtained by carefully exchanging the exponents of ε.

21
 (i) (i)
(Eε uε )α3 (i)
From (10) we deduce that, up to subsequences, 2
* Eα3 in L2 (Ω(i) ).
ε
(i)
To characterize Eα3 ∈ L2 (Ω(i) ), note that
µ (i) (i)
¶
(i) (i) D2 uε1 D1 uε2
2D3 (Wε uε )12 = D3 εα(i+1) − εα(i)
à ! à !
(i) (i) (i) (i)
D3 uε1 D1 uε3 D2 uε3 D3 uε2
= D2 + − D1 + α(i)
εα(i+1) ε3 ε3 ε
(i) (i) (i) (i)
(Eε uε )13 (Eε uε )23
= 2D2 α(i+1)
− 2D1 ,
ε εα(i)
in the sense of distributions. Hence for ψ ∈ C0∞ (Ω(i) ) we have
Z Z (i) (i) Z (i) (i)
(Eε uε )13 (Eε uε )23
(W(i)
ε u(i)
)
ε 12 3D ψ dy = D2 ψ dy − D1 ψ dy.
Ω(i) Ω(i) εα(i+1) Ω(i) εα(i)
On the other hand, using (10) we have that
(i) (i)
(Eε uε )13
* 0 in L2 (Ω(i) ), if i = 2
εα(i+1)
(i) (i)
(Eε uε )23
* 0 in L2 (Ω(i) ), if i = 1, 3.
εα(i)
Hence, passing to the limit in the previous equality we find
Z Z
(i)
−ϑD3 ψ dy = − E23 D1 ψ dy, if i = 2,
(i) (i)
ZΩ Z Ω
(i)
−ϑD3 ψ dy = E13 D2 ψ dy, if i = 1, 3.
Ω(i) Ω(i)

Thus
(i)
D1 E23 = D3 ϑ, if i = 2,
(i)
D2 E13 = −D3 ϑ, if i = 1, 3.
in the sense of distributions, and therefore, taking into account that ϑ does not
depend on y1 and y2 we have that
(i)
E23 = y1 D3 ϑ + γ (i) , if i = 2,
(i)
E13 = −y2 D3 ϑ + γ (i) , if i = 1, 3.

with γ (i) independent of y2 if i = 1, 3 and of y1 if i = 2. The conclusion is obtained

by setting η (i) := γ (i) − q2 D3 ϑ if i = 1, 3 and η (2) := γ (2) + q1 D3 ϑ. 2

22
Remark 6.2 In order to shorten notation, let us observe that the statement of
Lemma 6.1 can be summarized as follows
(i) (i)
(Eεn uεn )α(i)3 (i)
2
* Eβα(i) (yβ − qβ )ϑ0 + η (i) in L2 (Ω(i) ), i = 1, 2, 3,
εn
(i)
where we agree that q1 := q1 for any i, and α(i) is the parity of the index i.

As previously said, we consider homogeneous isotropic materials and denote

by
1 λ
f (A) := CA · A = µ|A|2 + |trA|2 , (33)
2 2
the stored energy density. Let
f0 (α, β) : = min{f (A) : A ∈ Sym, A23 = α, A33 = β}
= min{f (A) : A ∈ Sym, A13 = α, A33 = β}
Let us remark that, by isotropy, in the definition of f0 , A23 can be replaced with
A13 . A simple computation shows that
1
f0 (α, β) = 2µα2 + Eβ 2 (34)
2
where E is the Young modulus, which is given by
µ(2µ + 3λ)
E= .
µ+λ
Theorem 6.3 For every sequence of positive numbers εn converging to 0 and for
(1) (2) (3)
every sequence (uεn , uεn , uεn ) ∈ Aεn which satisfies (32), we have
3X Z
1 ¡ (i) (i) ¢
lim inf 4 Iεn (u(1)
ε , u(2)
ε , u(3)
ε )≥ f0 Eβα(i) (yβ − qβ )ϑ0 , D3 v3 dy,
n→+∞ εn n n n
Ω(i)
i=1

(i)
where q1 := q1 .

Proof. Taking into account the decomposition given in (7), it suffices to show
that Z
1 (i) (i) ¡ (i) (i) ¢
lim inf 4 Iεn (uεn ) ≥ f0 Eβα(i) (yβ − qβ )ϑ0 , D3 v3 dy.
n→+∞ εn Ω(i)
(i)
Looking at the expression (8) of the functional Iε and observing that, by the
definitions of f and f0 given in (33) and (34),
1
CA · A ≥ f0 (Aα(i)3 , A33 ),
2

23
we have
Z (i) (i) (3) (3)
1 (i) (i) (i) (Eεn uεn )α(i)3 (Eεn uεn )33
I (u ) ≥ χεn f0 ( , ) dy.
ε4n εn εn Ω(i) ε2n ε2n

Using the convexity of f0 ,

Z (3) (i) (i) (i)
(i) (Eεn uεn )α(i)3 (Eεn uεn )33
χεn f0 ( , ) dy ≥
Ω(i) ε2n ε2n
Z
£ (E(i) (i)
εn uεn )α(i)3 (i) 0 ¤
≥ χ(i)
εn 4µ − Eβα(i) (y β − qβ )ϑ − η (i)
·
Ω(i) ε2n
£ (i) ¤
· Eβα(i) (yβ − qβ )ϑ0 + η (i) dy
Z
£ (E(i) (i)
εn uεn )33 (i) ¤ (i)
+ χ(i)
εn E 2
− D3 v3 D3 v3 dy
(i) εn
ZΩ
¡ (i) 0 (i) ¢
+ χ(i) (i)
εn f0 Eβα(i) (yβ − qβ )ϑ + η , D3 v3 dy
Ω(i)

and by Lemma 6.1 we find

Z
1 ¡ (i) (i) ¢
lim inf 4 Iε(i) (u(i)
εn ) ≥ f0 Eβα(i) (yβ − qβ )ϑ0 + η (i) , D3 v3 dy.
k→+∞ εn n Ω(i)

On the other hand, by (34),

Z
¡ (i) (i) ¢
f0 Eβα(i) (yβ − qβ )ϑ0 + η (i) , D3 v3 dy =
Ω(i) Z
(i) (i)
= f0 (Eβα(i) (yβ − qβ )ϑ0 , D3 v3 )+
Ω (i)
Z Z
(i) 0 (i) 2
+4µ Eβα(i) (yβ − qβ )ϑ η dy + 2µ η (i) dy.
Ω(i) Ω(i)

and the proof is concluded by observing that

Z
(i)
Eβα(i) (yβ − qβ )ϑ0 η (i) dy = 0,
Ω(i)

because η (i) is independent of y3−α(i) , by Lemma 6.1. 2

24
7 The limit energy
The content of this section refers to the case of two junctions, but it can be adapted
with straightforward modifications to the case of a single junction.
Let
n
A := (v(1) , v(2) , v(3) , ϑ) ∈ HBN × H# 2
(0, `) :
o (35)
2 1
: ∃ η1 , η2 ∈ H# (0, `), and η3 ∈ H# (0, `) satisfying (31) .

Theorem 7.1 Let I : H 1 × H 2 (0, `) → [0, +∞) be defined by

 3
 X (i) (i)

(1) (2) (3) I (v , ϑ) if (v(1) , v(2) , v(3) , ϑ) ∈ A,
I(v , v , v , ϑ) := (36)

 i=1
+∞ otherwise,
where Z
(i) (i)
¡ (i) (i) ¢
I (v , ϑ) = f0 Eβα(i) (yβ − qβ )ϑ0 , D3 v3 dy.
Ω(i)
1
As ε → 0+ , the sequence of functionals 4 Iε Γ-converges to the functional I, in
ε
the following sense:
1. [liminf inequality] for every sequence of positive numbers εn converging to 0
(1) (2) (3)
and for every sequence (uεn , uεn , uεn ) ∈ Aεn such that
S(i) (i)
εn uεn * v
(i)
in H 1 (Ω(i) ; R3 ),

(Wε(i) u(i) ) * −ϑ in L2 (Ω(i) ),

n εn 12

we have
1
lim inf Iε (u(1) , u(2) , u(3) ) ≥ I(v(1) , v(2) , v(3) , ϑ);
n→+∞ ε4n n εn εn εn
2. [recovery sequence] for every sequence of positive numbers εn converging to 0
(1) (2) (3)
and for every (v(1) , v(2) , v(3) , ϑ) ∈ A there exists a sequence (un , un , un ) ∈
Aεn such that
S(i) (i)
εn un * v
(i)
in H 1 (Ω(i) ; R3 ),

(Wε(i)
n n
u(i) )12 * −ϑ in L2 (Ω(i) ),
and
1
lim sup Iε (u(1) , u(2) (3) (1) (2) (3)
n , un ) ≤ I(v , v , v , ϑ).
n→+∞ ε4n n n

25
Proof. We start by proving the liminf inequality. Without loss of generality we
may suppose that
1
lim inf 4 Iεn (u(1) (2) (3)
εn , uεn , uεn ) < +∞,
n→+∞ εn

and therefore that

1
sup Iε (u(1) , u(2) , u(3) ) < +∞.
n ε4n n εn εn εn
Thus the assumptions of Lemma 4.1 are satisfied and therefore Lemma 4.3, Lemma
5.7 and Lemma 6.1 hold. The liminf inequality then follows from Theorem 6.3.
We now find a recovery sequence. Let us first note that

f0 (α, β) = f (Λ(i) (α, β)),

where Λ(i) are symmetric matrices with

(i) (i) (i) (i)
Λ11 (α, β) = Λ22 (α, β) = −νβ, Λ12 (α, β) = 0, Λ33 (α, β) = β, i = 1, 2, 3,
(i) (2) (i) (2)
Λ23 (α, β) = Λ13 (α, β) = 0, Λ13 (α, β) = Λ23 (α, β) = α, i = 1, 3,
(37)
and ν denotes the Poisson’s coefficient
λ
ν= .
2(λ + µ)

Let us assume that I(v(1) , v(2) , v(3) , ϑ) < +∞, otherwise there is nothing to prove.
Then (v(1) , v(2) , v(3) , ϑ) ∈ A and therefore there exist η1 , η2 ∈ H#
2 (0, `), and η ∈
3
1
H# (0, `) such that (31) hold.
To start, we further assume η1 , η2 , η3 and ϑ to be equal to zero in a neigh-
borhood of y3 = 0. Let u0,ε : Ωε → R3 , be the sequence of functions defined
by
u0,ε := uf,ε + ut,ε (38)
with ¡ ¢
ε
(uf,ε )1 := εη1 − ν ε2 x1 η30 + (x22 − x21 )η100 − εx1 x2 η200 ,
2
¡ 2 ε ¢
(uf,ε )2 := εη2 − ν ε x2 η30 − εx1 x2 η100 + (x21 − x22 )η200 , (39)
2
(uf,ε )3 := ε2 η3 − εx1 η10 − εx2 η20 ,
and
(ut,ε )1 := −(x2 − εc2 )ϑ − νr1ε ϑ00 ,

(ut,ε )2 := (x1 − εc1 )ϑ − νr2ε ϑ00 , (40)

(ut,ε )3 := ψ ε ϑ0 ,

26
where
 Z y1
 3 (1) −1 (1)

 ε ( ψ (1) (s) ds κ(1)
ε ) ◦ pε in ωε ,

 q1

(2) −1 (2)
ε
r1 (x1 , x2 ) := ε (ψ (2) (y2 )y1 κ̌(2)
4 (2)
ε κ̂ε ) ◦ pε in ωε ,

 Z y1


 3 (3) −1 (3)
 ε ( ψ (3) (s) ds κ(3)
ε ) ◦ pε in ωε ,
q1


(1) −1 (1)

 ε4 (ψ (1) (y1 )y2 κ(1)
ε ) ◦ pε in ωε ,


 Z (2) (2)
ε ¡ y2 (2) κ̂ε + κ̌ε ¢ (2) −1 (2)
r2 (x1 , x2 ) := ε 3
ψ (s) ds ◦ pε in ωε ,

 (1) 2

 q2
 4 (3) (3) −1 (3)
ε (ψ (y1 )y2 κ(3) ε ) ◦ pε in ωε ,

and 
 2 (1) (1) −1 (1)

 ε ψε ◦ pε in ωε ,
ψ ε (x1 , x2 ) := (2) (2) −1 (2)
 ε2 ψε ◦ pε in ωε ,

 2 (3) (3) −1 (3)
ε ψε ◦ pε in ωε ,
where
(1) (1) (1) (1)
ψε := −ε(y1 − q1 )(y2 − q2 )κε + ψ (1) − ε(q1 − c1 )(y2 − q2 ),
(2) (2) (2)
ψε := κ̌ε κ̂ε (ψ (2) + ε(y1 − q1 )(y2 − c2 ))+
(3)
+(1 − κ̌(2)
ε )(ψ
(2)
+ ε(y1 − q1 )(q2 − c2 ))+
(1)
+ (1 − κ̂(2)
ε )(ψ
(2)
+ ε(y1 − q1 )(q2 − c2 )),
(3) (3) (3) (3)
ψε := −ε(y1 − q1 )(y2 − q2 )κε + ψ (3) − ε(q1 − c1 )(y2 − q2 ).
(1) (3)
In the definitions above the cut-off functions κε : ω (1) → [0, 1], κε : ω (3) → [0, 1],
(2) (2)
and κ̌ε , κ̂ε : ω (2) → [0, 1], have the following properties

D2 κ(1) (3) (2) (2)

ε = D2 κε = D1 κ̌ε = D1 κ̂ε = 0,

2
|D1 κ(1) (3) (2) (2)
ε |, |D1 κε |, |D2 κ̌ε |, |D2 κ̂ε | ≤ ,
ε
s(2) s(2)
κ(3) (1)
ε (y1 , y2 ) = κε (y1 , y2 ) = 0 for q1 − ε ≤ y1 ≤ q1 + ε ,
2 2
s(2) s(2)
κ(3) (1)
ε (y1 , y2 ) = κε (y1 , y2 ) = 1 for y1 ≥ q1 + ε + ε and y1 ≤ q1 − ε − ε,
2 2

27
 s(1)
(1) (1) s(1)
κ̂(2)
ε (y1 , y2 ) = 0 for y2 ≤ q2 + ε , κ̂(2)
ε (y ,
1 2 y ) = 1 for y 2 ≥ q 2 + ε + ε,
2 2
(3) s(3) (3) s(3)
κ̌(2)
ε (y 1 , y2 ) = 0 for y 2 ≥ q2 − ε , κ̌(2)
ε (y1 , y2 ) = 1 for y 2 ≤ q 2 − ε − ε.
2 2
It can be easily checked that ψ ε is well defined.
(i) (i) (i)
From the definitions above one can easily compute u0,ε := u0,ε ◦ pε and verify
(i)
that u0,ε satisfies the following estimates

(i) (i)
Eε u0,ε (i) (i)
k − Λ(i) (Eβα(i) (yβ − qβ )ϑ0 , D3 v3 )kL2 (Ω(i) ) ≤ εC(v(i) , ϑ),
ε2
k(W(i)
(i) (i) (41)
ε u0,ε )12 + ϑkL2 (Ω(i) ) ≤ εC(v , ϑ),
(i)
kS(i) (i) (i)
ε u0,ε − v kH 1 (Ω(i) ) ≤ εC(v , ϑ),

where C(v(i) , ϑ) depends only on v(i) and ϑ. Hence in this case (u0,εn ) is a recovery
sequence.
In the general case, i.e., (v(1) , v(2) , v(3) , ϑ) ∈ A, we can find, by convolution,
(1) (2) (3)
functions (vk , vk , vk , ϑk ) ∈ A which are smooth, equal to zero near by y3 = 0
and such that
(i) (i) (i) (i)
kΛ(i) (Eβα(i) (yβ − qβ )ϑ0k , D3 v3k ) − Λ(i) (Eβα(i) (yβ − qβ )ϑ0 , D3 v3 )kL2 (Ω(i) ) ≤ 1/k,
kϑk − ϑkL2 (Ω(i) ) ≤ 1/k,
(i)
kvk − v(i) kH 1 (Ω(i) ) ≤ 1/k,

for every k. Denoting by uk,ε the sequence defined as u0,ε in (38) but with η (i) ’s
(i) (1) (3) (2)
and ϑ replaced by the ηk ’s and ϑk used in the definition of (vk , vk , vk , ϑk ) ∈
A, given a sequence εn converging to zero, we can find an increasing sequence
(i) (i)
of integers (kn ) and therefore a diagonal un := ukn ,εn such that the sequence
(1) (2) (3)
un = (un , un , un ) satisfies the required recovery sequence conditions. 2

8 The elastic energy and the shear center

The limit energy functional I(v(1) , v(2) , v(3) , ϑ) can be written in a more explicit
form by using Lemma 5.7 and the fact that η1 , η2 , η3 and ϑ depend only on y3 .

28
Indeed, the limit strain energy rewrites as
    0 
A −S2 −S1 Sψ η30 η3
Z `
µ 
E  I22 I12 −Iψ2   η100   η100 
I(η, ϑ) = Jt ϑ02 +     dy3
2 2  I11 −Iψ1   η200  ·  η200 
0
sym Iψψ ϑ00 ϑ00
(42)
where (denoting with da = dy1 dy2 )
Z Z Z
A= da + da + da = b(1) s(1) + b(3) s(3) + hs(2)
ω (1) ω (3) ω (2)

is the total area,

Z Z Z
(1) (3)
S1 = q2 da + q2 da + y2 da,
(1) ω (3) ω (2)
Zω Z Z
S2 = y1 da + y1 da + q1 da
ω (1) ω (3) ω (2)

are the static moments,

³Z Z Z ´
(1) 2 (3)
Jt = 4 (y2 − q2 ) da + (y2 − q2 )2 da + (y1 − q1 )2 da
ω (1) ω (3) ω (2)
1 3 3 (2) 3
= (b(1) s(1) + b(3) s(3) + hs )
3
is the “torsional” moment of inertia,
Z Z Z
(1) 2 (3) 2
I11 = q2 da + q2 da + y22 da,
(1) (3) (2)
Zω Z ω Z ω

I22 = y12 da + y12 da + q12 da,

ω (1) ω (3) ω (2)
Z Z Z
(1) (3)
I12 = y1 q2 da + y1 q2 da + q1 y2 da,
ω (1) ω (3) ω (2)

are the moments of inertia, and

Z Z Z
Sψ = ψ (1) (y1 ) da + ψ (3) (y1 ) da + ψ (2) (y2 ) da,
(1) (3) (2)
Zω ω Z ω Z
(1) (1) (3) (3)
Iψ1 = q2 ψ (y1 ) da + q2 ψ (y1 )da + y2 ψ (2) (y2 ) da,
(1) (3) (2)
Zω Z ω Z ω
(1) (3)
Iψ2 = y1 ψ (y1 ) da + y1 ψ (y1 ) da + q1 ψ (2) (y2 ) da,
(1) (3) (2)
Zω Z ω Z ω
Iψψ = ψ (1) (y1 )2 da + ψ (3) (y1 )2 da + ψ (2) (y2 )2 da,
ω (1) ω (3) ω (2)

29
are the sectorial statical moment, the sectorial products of inertia and the sectorial
moment of inertia, respectively.
We note that, as it stands, the extensional, flexural and torsional problems are
all coupled together. If the axes are centered in the center of mass, i.e. S1 = S2 = 0,
and are principal axes, i.e. I12 = 0, then there is coupling only between extension
and torsion through the coupling term Sψ η30 ϑ00 , and flexure and torsion through
the coupling terms Iψα ηα00 ϑ00 . We will have full decoupling only in the case in which
also Sψ = Iψ1 = Iψ2 = 0. We recall (see Lemma 5.7) that the sector coordinates
depend on three parameters, the coordinates of the point c = (c1 , c2 ) and the
additive constant K. Hence we can solve the system

 Sψ = 0
I =0 (43)
 ψ1
Iψ2 = 0
of three equations in the three unknowns c1 , c2 and K. Roughly speaking we can
say that we choose the constant K so that the sector coordinate will have null
mean, i.e., Sψ = 0, and then the point c so to make the sectorial products of
inertia Iψ1 , Iψ2 equal to zero. This latter point is usually called, in the engineering
literature, the shear center and the resulting Iψψ is the warping stiffness.

Remark 8.1 When only one junction is present, which happens for instance if
b(3) = s(3) = d(3) = 0 (see Section 5.2) then, from Remark 5.8 follows that the
sector coordinates can be chosen equal to zero and the limit strain energy functional
takes the form
  0   0 
Z ` A −S2 −S1 η3 η3
µ E 
I(η, ϑ) = Jt ϑ02 + I22 I12   η100  ·  η100  dy3 .
0 2 2
sym I11 η200 η200
It is worth notice that this De Saint-Venant limit energy is formally the same
which holds in the case when there are no junctions (see [4], Section 7).

9 Some examples
9.1 Single-junction sections
By Lemma 5.5, Lemma 5.7 and Remark 5.8 (see also Remark 8.1) follows that
when the cross-section consists of straight segments connected by a unique common
junction, that is when the one dimensional limit structure is star-shaped, then the
warping stiffness is zero, the shear center is the common point of intersection. The
figure below shows three examples of this kind of sections.

30
9.2 Channel-section
Consider the so-called Channel-section shown in the figure below.

b
s

h=2b G
x1=y1 c

x2 y2

The thickness s is much smaller then the height h. Let us follow the scheme of
Section 8. The area is
A = 4bs.
The point G with coordinates x1 (G) = −b/4 and x2 (G) = 0 is the center of mass
of the cross-section. Choosing a reference system Gy1 y2 with origin in G we have
(1) (1)
that q1 = b/4, q2 = b and q2 = −b. Hence, from the integration formulae of
Section 8, we easily get
4 8 5
S1 = 0, S2 = 0, Jt = bs3 , I11 = b3 s, I22 = b3 s, I12 = 0,
3 3 12
sb3 5
Sψ = 4Kbs, Iψ1 = − (5b − 8c1 ), Iψ2 = − sb3 c2 .
3 12
Solving the system (43) we obtain that K = 0 and the shear center c has coordi-
nates c1 = 85 b and c2 = 0, in the reference Gy1 y2 , as shown in the figure above.
7 5
The resulting warping stiffness is Iψψ = 24 b s.

31
9.3 Symmetric double-T section
Consider the double-T section shown in the figure below.

(1)
b
(1)
s
G h/2
y1 c
x1 (2)
s h/2
(3)
s
(3)
b
x2=y 2

The thicknesses s(3) , s(2) and s(1) are much smaller then the height h. Let us
follow the scheme of Section 8. It is useful to set A(3) := b(3) s(3) , A(1) := b(1) s(1)
and A(2) := hs(2) . Then the total area is

A = A(1) + A(2) + A(3) .

The center of mass G of the cross-section has the following coordinates

h (A(3) − A(1) )
x1 (G) = 0, x2 (G) = .
2 A
Choosing a reference system Gy1 y2 we have that

(1) h A(3) − A(1) (1) h A(3) − A(1)

q1 = 0, q2 = (1 − ), q2 = − (1 + ).
2 A 2 A
Hence, from the integration formulae of Section 8, and setting also
2 2
(1) A(1) b(1) (3) A(3) b(3)
I2 := , I2 := ,
12 12
which are the axial moments of inertia of ω (1) and ω (3) with respect to the axis y2 ,
after some simple calculations we get
1 3 3 3
S1 = 0, S2 = 0, I12 = 0, Jt = (b(1) s(1) + b(3) s(3) + hs(2) ),
3
h2 £ A(3) − A(1) (1) A(2) ¤ (1) (3)
I11 = (A − A(3) + A(2) ) + A(1) + A(3) + , I22 = I2 + I2 ,
4 A 3

32
 h (3) ³ h A(3) − A(1) ´ (3)
(1) (1)
Sψ = AK, Iψ1 = c1 I11 , Iψ2 = (I2 − I2 ) − c2 + (I2 + I2 ).
2 2 A
Solving the system (43) we obtain that K = 0 and the shear center c has coordi-
nates
h ³ I2 − I2 A(3) − A(1) ´
(3) (1)
c1 = 0, c2 = − ,
2 I (3) + I (1) A
2 2
in the reference Gy1 y2 , as shown in the figure above. The resulting warping stiffness
is
(3) (1)
I I
Iψψ = h2 (3)2 2 (1) .
I2 + I2
Let us remark that from the expressions above one can deduce, in particular, that
(3) (1)
if A(3) = A(1) and I2 = I2 , then c = (0, 0), that is c ≡ G.

10 The loaded beam

As stated in (3), the total energy Fε is equal to the elastic energy Jε minus the
work done by the external loads:
Z
Fε (u) = Jε (u) − bε · u dx.
Ωε

Let us denote by
Z 3 Z
X
Lε (u) := bε · u dx = χ(i) ε
ε b · u dx
(i)
Ωε i=1 Ωε

=: L(1)
ε (u) + (2)
Lε (u) + L(3)
ε (u)

(i)
where χε is defined in Section 3. Then we can write

Fε (u) = Fε(1) (u) + Fε(3) (u) + Fε(2) (u)

(1) (1) (3) (3) (2) (2)
:= Jε (u) − Lε (u) + Jε (u) − Lε (u) + Jε (u) − Lε (u).

With change of variables, as done in section 3, we deduce and set

1 (i) (i) (i) −1
Fε(i) (u(i) ) : = F (u ◦ pε )
ε3 ε
Z (44)
= Iε(i) (u(i) ) − b(i) (i) (i) (i) (i) (i)
ε · u dy =: Iε (u ) − Lε (u ),
Ω(i)

33
where
b(i) (i) ε (i)
ε = χε b ◦ pε .

Let us assume that

³ m(i) (y3 ) ´
(i) (i)
bε1 = εα(i) ε2 b1 (y) − ε (y2 − y 2 (G (i)
)) ,
IG (ω (i) )
³ m(i) (y3 ) ´ (45)
(i) (i)
bε2 = εα(i+1) ε2 b2 (y) + ε (y 1 − y 1 (G (i)
)) ,
IG (ω (i) )
(i) (i)
bε3 = ε2 b3 (y),

(i) (i) (i)

with b(i) = (b1 , b2 , b3 ) ∈ L2 (Ω(i) ; R3 ) and m(i) ∈ L2 (0, `), while IG (ω (i) ) is
always the moment of inertia with respect to the center of mass (y1 (G(i) ), y2 (G(i) )
of the section ω (i) . With this choice, the work of the external loads can be rewritten
as
3
X
Lε (u(1) , u(2) , u(3) ) = L(i) (i)
ε (u ) (46)
i=1

where
Z Z
L(i) (i)
ε (u ) =ε 4
b (i)
· S(i) (i)
ε u dy +ε 4
m(i) ϑ(i) (i)
ε (u ) dy.
Ω(i) Ω(i)

The sequence of functionals Lε /ε4 is continuously convergent with respect to the

(1) (2) (3)
convergence used in Theorem 7.1, that is, for any sequence (uε , uε , uε ) ∈ Aε
such that

Sε(i) u(i)
ε *v
(i)
in H 1 (Ω(i) ; R3 ),

(W(i) (i) 2 (i)

ε u )12 * −ϑ in L (Ω ),

we have
Z Z `
1 (i) (i)
L (uε ) → L(i) (v(i) , θ) := b (i)
·v (i)
dy + m(i) ϑ dy3 ,
ε4 ε Ω(i) 0

so that
1
Lε (u(1) (2) (3) (1) (2) (3)
ε , uε , uε ) → L(v , v , v , θ) (47)
ε4
where
3
X 3 ³Z
X Z ` ´
(1) (2) (3) (i) (i) (i) (i)
L(v ,v ,v , θ) := L (v , θ) = b ·v dy + m(i) ϑ dy3 .
i=1 i=1 Ω(i) 0

34
In terms of the variables (η, ϑ) the limit load can be written in the form
Z `
L(η, ϑ) = l1 η1 + l2 η2 + l3 η3 + mϑ − mα ηα0 + bϑ0 dy3 (48)
0
where
(1) (3) (2) (1) (2) (3)
l1 = b1 + b1 , l2 = b2 , l3 = b3 + b3 + b3 ,
3
X Z Z
(i) (1) (1) (3) (3)
m= m − (q2 − c2 )b1 da − (q2 − c2 )b1 da+
ω (1) ω (3)
i=1 Z
(2)
+ (q1 − c1 )b2 da,
Z Z Z ω (2)
(1) (3) (2)
m1 = y1 b3 da + y1 b3 da + q1 b3 da,
(1) (3) ω (2)
Zω ω
Z Z
(1) (1) (3) (3) (2)
m2 = q2 b3 da + q2 b3 da + y2 b3 da,
Zω (1) ω (3)
Z ω (2)
Z
(1) (3) (2)
b= b3 ψ (1) da + b3 ψ (3) da + b3 ψ (2) da,
ω (1) ω (3) ω (2)
and da = dy1 dy2 .
We note that the contribution to the moments m of the loads b(i) are moments
evaluated with respect to the shear center and not with respect the center of mass
of the section.
The theorem below follows from the stability of Γ-convergence with respect
to continuously convergent, real valued, perturbations (see Dal Maso [3], Proposi-
tion 6.20).
Theorem 10.1 Let the rescaled total energy Fε be defined as
3
X
(1) (2) (3)
Fε (u ,u ,u )= Fε(i) (u(i) ). (49)
i=1

1
As ε → 0+ , the sequence of functionals 4 Fε Γ-converges to the limit functional
ε
F := I − L in the sense specified in Theorem 7.1.
For every ε ∈ (0, 1] let us denote by ūε the solution of the following minimiza-
tion problem
min{Fε (u) : u ∈ H 1 (Ωε ; R3 ), u = 0 on Sε (0)}. (50)
The existence of a solution can be proved by the direct method of the Calculus
of Variations and the uniqueness follows by the strict convexity of the functionals
Fε . The next theorem describes the behaviour of the sequence of minimizers ūε .

35
Theorem 10.2 The following minimization problem for the Γ-limit functional
F =I −L
2 1 2
min{F (η, ϑ) : η1 , η2 ∈ H# (0, `), η3 ∈ H# (0, `) and ϑ ∈ H# (0, `)} (51)

admits a unique solution (η̄, ϑ̄). Moreover, denoting with (v̄(1) , v̄(2) , v̄(3) ) a field
related to (η̄, ϑ̄) throughout equations (31) we have that, as ε → 0+ ,

1. S(i) (i)
ε ūε → v̄
(i)
strongly in H i (Ω(i) ; R3 ),
(i) (i)
2. (Wε ūε )12 → −ϑ̄ strongly in L2 (Ω(i) ),
1
3. Fε (ū(1) (2) (3)
ε , ūε , ūε ) converges to F (η̄, ϑ̄).
ε4
Proof. From the Γ-convergence Theorem 10.1 and from well known properties
of Γ-limits and in particular by putting together Propositions 6.8 and 8.16 (lower
semicontinuity of sequential Γ-limits), Theorem 7.8 (coercivity of the Γ-limit) and
Corollary 7.24 (convergence of minima and minimizers) of Dal Maso [3], follows
that

1 0 . S(i) (i)
ε ūε * v̄
(i)
in H i (Ω(i) ; R3 ),
(i) (i)
2 0 . (Wε ūε )12 * − ϑ̄ in L2 (Ω(i) ),
1
3 0. Fε (ū(1) (2) (3)
ε , ūε , ūε ) converges to F (η̄, ϑ̄).
ε4
It remains to prove that in 1 0 and 2 0 the convergence is, in fact, strong.
Let us denote by aε the approximate minimizers defined as the sequence u0,ε
in (38), (39), (40) with (η, ϑ) replaced by (η̄, ϑ̄). Even if we cannot say that
it is a recovery sequence, for it does not satisfy the boundary conditions, the
estimates (41) with (v, ϑ) replaced by (v̄, ϑ̄) hold and therefore
1
lim Fε (a(1) (2) (3)
ε , aε , aε ) = F (η̄, ϑ̄),
ε→0+ ε4
(52)
1
lim Lε (a(1) (2) (3)
ε , aε , aε ) = L(η̄, ϑ̄).
ε→0+ ε4

In particular we have that

(i) (i) (i) (i) (i) (i) (i)
Fε (ūε ) − Fε (aε ) Lε (ūε − aε )
lim = 0, lim = 0. (53)
ε→0+ ε4 ε→0+ ε4

36
A key point in the proof of strong convergence of 1 0 consists in showing that
° E(i) (ū(i) − aε ) °
° ε ε °
lim ° ° 2 (i) = 0. (54)
ε→0+ ε2 L (Ω )

To start we observe that the quadratic form f (A) defined in (33) satisfies the
identity
f (U) = f (A) + CA · (U − A) + f (U − A)
for every pair of 3 × 3 matrices A and U. By (1) we thus obtain the inequality
f (U) ≥ f (A) + CA · (U − A) + µ|U − A|2 ,
(i)
which can be used in the expression of the integral functional Fε to obtain that
Z
Fε(i) (ū(i)
ε ) ≥ F (i) (i)
ε (aε ) + χ(i) (i) (i) (i) (i) (i)
ε CEε aε · Eε (ūε − aε ) dy +
Ω(i)
Z
+µ (i)
χε |E(i) (i) (i) (i) 2 (i) (i)
ε ūε − Eε aε | dy − Lε (ūε − aε ).
(i)
Ω(i)
(i)
Hence, taking also into account that χε ≥ 1/2, we have
(i) (i) (i) (i) Z (i) (i) (i) (i) (i)
Fε (ūε ) − Fε (aε ) 1 CEε aε · Eε (ūε − aε )
≥ dy +
ε4 2 Ω(i) ε4
Z (i) (i) (i) (i)
(55)
µ |Eε ūε − Eε aε |2
+ dy − L(i) (i) (i)
ε (ūε − aε ).
2 Ω(i) ε4
Let us then prove that
Z (i) (i) (i) (i) (i)
CEε aε · Eε (ūε − aε )
lim dy = 0 (56)
ε→0+ Ω(i) ε4
and (54) is obtained by passing to the upper limit as ε → 0+ in (55). The proof
of (56) proceeds along the same lines of Theorem 8.1 of [4] but we give here full
details for convenience of the reader.
In order to prove (56) we observe that for every pair of matrices A and B
CA · B = 2µA · B + λ tr(A) tr(B).
Then
Z (i) (i) (i) (i) (i) Z (i) (i) (i) (i) (i)
CEε aε · Eε (ūε − aε ) Eε aε · Eε (ūε − aε )
dy = 2µ dy +
Ω(i) ε4 Ω(i) ε4
Z (i) (i) (i) (i) (i)
tr(Eε aε ) tr(Eε (ūε − aε ))
+λ dy.
Ω(i) ε4
(57)

37
In order to perform the computation, it is convenient to shorten the notation by
setting
(i) (i) (i) (i)
(Eε aε )hj (Eε ūε )hj
Aεhj := , U ε
hj := ,
ε2 ε2
(i) (i) (i)
(Eε (ūε − aε ))hj
∆εhj
:= − ε
Uhj
= Aεhj ,
ε2
so that the integrand in (57) takes the following expression

2µ(Aεhj ∆εhj ) + λAεhh ∆εjj .

By (41) we have
(i) (i)
Aε → Λ(i) (Eβα(i) (yβ − qβ )ϑ̄0 , D3 v̄3 ) in L2 (Ω(i) ), (58)

where Λ(i) is defined in (37), and by the equation above and Lemma 4.1 it follows
that ∆ε is bounded in L2 (Ω(i) ). Thus, from (58) we immediately deduce that
Z
lim Aε12 ∆ε12 dy = 0,
ε→0+ Ω(i)

and Z
lim Aεα(i+1)3 ∆εα(i+1)3 dy = 0. (59)
ε→0+ Ω(i)

From (58) and 1 0 follows that ∆ε33 * 0 in L2 (Ω(i) ), and therefore

Z
lim Aεhj ∆ε33 dy = 0, h, j = 1, 2, 3.
ε→0+ Ω(i)

From 1 0 , 2 0 , 3 0 and Lemma 6.1, follows that, up to subsequences, Uα(i)3

ε weakly
(i)
converges in L2 (Ω(i) ) to Eβα(i) (yβ − qβ )ϑ̄0 + η̄ (i) , for some η̄ (i) as specified in the
(i)
statement of Lemma 6.1. By (58) we have that Aεα(i)3 → Eβα(i) (yβ −qβ )ϑ̄0 strongly
in L2 (Ω(i) ) and hence, up to subsequences, ∆εα(i)3 → η̄ (i) weakly in L2 (Ω(i) ). Thus
Z Z
(i)
lim Aεα(i)3 ∆εα(i)3 dy = Eβα(i) (yβ − qβ )ϑ̄0 η̄ (i) dy = 0.
ε→0+ Ω(i) Ω(i)

Putting together with (59) we have that

Z
lim Aεh3 ∆εh3 dy = 0, h, i = 1, 2, 3.
ε→0+ Ω(i)

38
Let ∆11 and ∆22 be, up to subsequences, the weak limits in L2 (Ω(i) ) of ∆ε11 and
∆ε22 , respectively. Summarizing and taking the limit as ε → 0+ in (57) we obtain
(even for the whole sequence)
Z (i) (i) (i) (i) (i)
CEε aε · Eε (ūε − aε )
lim dy =
ε→0+ Ω(i) ε4
Z
= lim 2µ(Aε11 ∆ε11 + Aε22 ∆ε22 ) + λAεhh ∆εαα dy
ε→0+ Ω(i)
Z
(i)
= D3 v̄3 (∆11 + ∆22 )[−2ν(µ + λ) + λ] dy = 0
Ω(i)

because −2ν(µ + λ) + λ = −λ + λ = 0, and (56) and hence (54) are proven.

(i) (i) (i) (i)
Setting zε = Sε (ūε − aε ), we have
(i) (i) (i)
Eε (ūε − aε )
kEz(i)
ε kL2 (Ω(i) ) ≤k kL2 (Ω(i) )
ε2
(i)
and by the application of the standard Korn inequality to zε , and using (54), we
obtain that
kz(i)
ε kH 1 (Ω(i) ) → 0.
From this fact and the third equation of (41) applied to the sequence aε we gain
the strong convergence in 1 0 . This proves 1.
On the other hand, by (54) and inequality (11), we have that
Z
lim |H(i) (i) (i) 2
ε (ūε − aε )| dy = 0
ε→0+ Ω(i)
(i)
and since the definition of Hε we deduce from here that
(i) (i) (i) (i)
D1 (ūε2 − aε2 ) D2 (ūε1 − aε1 )
α(i)
→ 0, → 0, in L2 (Ω(i) ).
ε εα(i+1)
Thus, from the second equation of (41) applied to the sequence aε follows that
(i) (i) (i) (i)
D2 (ūε1 − aε1 ) D1 (ūε2 − aε2 )
(Wε(i) ū(i)
ε )12 = − + (W(i) (i) 2 (i)
ε aε )12 → −ϑ̄ in L (Ω ),
2εα(i+1) 2εα(i)
which proves 2 . 2

Acknowledgements
The work of L.F. and R.P. has been partially supported by Progetto Cofinan-
ziato 2005 “Modellazione matematica di strutture e materiali complessi”, while
A.M. has been partially supported by MURST, grant no. 2003082352.

39
References
[1] G. Anzellotti, S. Baldo, and D. Percivale, Dimension reduction in variational
problems, asymptotic development in Γ-convergence and thin structures in
elasticity, Asymptot. Anal. 9(1) (1994), 61–100.

[2] A. Braides, Γ-convergence for beginners, Oxford Lecture Series in Mathemat-

ics and its Applications, 22. Oxford University Press, Oxford, 2002.

[3] G. Dal Maso, An introduction to Γ-convergence, Birkhäuser, Boston, 1993.

[4] L. Freddi, A. Morassi and R. Paroni, Thin-walled beams: the case of the
rectangular cross-section, J. Elasticity 76 (2004), 45–66.

[5] A. Gaudiello, R. Monneau, J. Mossino, F. Murat, A. Sili, On the junction

of elastic plates and beams, C. R. Math. Acad. Sci. Paris 335 n.8 (2002),
717–722.

[6] V.A. Kondrat’ev and O.A. Oleinik, On the dependence of the constant in
Korn’s inequality on parameters characterizing the geometry of the region,
Russian Math. Surveys 44 n.6 (1989), 187–195.

[7] H. Le Dret, Problémes variationnels dans le multi-domaines. Modélisation des

jonctions et applications., RMA 19, Masson Springer-Verlag, Paris, 1991.

[8] O.A. Oleinik, A.S. Shamaev, and G.A. Yosifian, Mathematical problems in
elasticity and homogenization, North-Holland, Amsterdam, 1992.

[9] D. Percivale, Thin elastic beams: the variational approach to St. Venant’s
problem, Asymptot. Anal. 20 (1999), 39–59.

[10] J.M. Rodrı́guez, and J.M. Viaño, Asymptotic derivation of a general linear
model for thin-walled elastic rods, Comput. Methods Appl. Mech. Engrg. 147
(1997), 287–321.

[11] S.P. Timoshenko, De la stabilité à la flexion plane d’une poutre en double

té, Nouvelles de l’Institut Polytechnique de Saint-Pétesbourg T. IV-V (1905-
1906).

[12] B.Z. Vlassov, Pièces longues en voiles minces, Éditions Eyrolles, Paris, 1962.

40