16.21 Techniques of Structural Analysis and
Design
Spring 2003
Unit #9 - Calculus of Variations
Let u be the actual con?guration of a structure or mechanical system. u
?
satis?es the displacement boundary conditions: u = u on S
u
. De?ne:
ˉu = u + αv, where:
α : scalar
v : arbitrary function such that v = 0 on S
u
We are going to de?ne αv as δu, the ?rst variation of u:
δu = αv (1)
Schematically:
u1
u2u
a b
u(a)
u(b)
v
1
?
As a ?rst property of the ?rst variation:
dˉu du dv
= + α
dx dx
????
dx
so we can identify α
dv
with the ?rst variation of the derivative of u:
dx
?
du
dx
= α
dv
dx
δ
But:
dv
dx
αdv
dx
d
dx
(δu)α = =
We conclude that:
?
du
?
d
δ =
dx dx
(δu)
Consider a function of the following form:
F = F (x,u(x),u
?
(x))
It depends on an independent variable x, another function of x (u(x)) and its
derivative (u
?
(x)). Consider the change in F , when u (therefore u
?
) changes:
ΔF = F (x,u + δu,u
?
+ δu
?
) ? F (x,u,u
?
)
= F (x,u + αv,u
?
+ αv
?
) ? F (x,u,u
?
)
expanding in Taylor series:
?F ?F 1 ?
2
F 1 ?
2
F
ΔF = F + αv +
?u
?
αv
?
+
?u
2
(αv)
2
+
?u?u
?
(αv)(αv
?
) + ···? F
?u 2! 2!
?F ?F
= αv +
?u
?
αv
?
+ h.o.t.
?u
First total variation of F:
?
ΔF
?
δF = α lim
α→0
α
?
F (x,u + αv,u
?
+ αv
?
) ? F (x,u,u
?
)
?
= α lim
α→0
α
? ?F
αv +
?F
= α lim
?u ?u
?
αv
? ?
=
?F
αv +
?F
?
?u
?
αv
α→0
α ?u
δF =
?F
?u
δu +
?F
?u
?
δu
?
2
?
?
?
?
Note that:
dF (x,u + αv,u
?
+ αv
?
)
?
?
δF = α ?
dα α=0
since:
dF (x,u + αv,u
?
+ αv
?
) ?F (x,u + αv,u
?
+ αv
?
) ?F (x,u + αv,u
?
+ αv
?
)
?
= v +
dα ?u ?u
?
evaluated at α = 0
dF (x,u + αv,u
?
+ αv
?
)
?
?
?F (x,u,u
?
) ?F (x,u,u
?
)
?
? = v + v
dα α=0 ?u ?u
?
Note analogy with di?erential calculus.
δ(aF
1
+ bF
2
) = aδF
1
+ bδF
2
linearity
δ(F
1
F
2
) = δF
1
F
2
+ F
1
δF
2
etc
The conclusions for F (x,u,u
?
) can be generalized to functions of several
?u
i
independent variables x
i
and functions u
i
,
?x
j
:
?
?u
i
?
F x
i
,u
i
,
?x
j
We will be making intensive use of these properties of the variational operator
δ:
d d dv
?
du
?
= δ
?
dx
(δu)
?
=
dx
(αv) =
?
α
dx dx
δudx = αvdx = α vdx = δ udx
Concept of a functional
b
I(u) = F (x,u(x),u
?
(x))dx
a
3
v
?
?
?
?
? ?
?
?
?
First variation of a functional:
δI = δ F (x,u(x),u
?
(x))dx
= δ F (x,u(x),u
?
(x)) dx
?
?
?F ?F
?
?
δI = δu + dx
?u ?u
?
δu
Extremum of a functional
“u
0
” is the minumum of a functional if:
I(u) ≥ I(u
0
)?u
A necessary condition for a functional to attain an extremum at “u
0
” is:
dI
δI(u
0
) = 0, or
dα
(u
0
+ αv,u
?
0
+ αv
?
)
?
? = 0
α=0
Note analogy with di?erential calculus. Also di?erence since here we require
dF
= 0 at α = 0.
dα
b?
?F ?F
?
?
δI = δu +
?u
?
δu dx
?u
a
Integrate by parts the second term to get rid of δu
?
.
b?
?F d
?
?F
?
d
?
?F
??
δI = δu +
?u
?
δu ? δu
dx ?u
?
dx
?u dx
a
?
b?
?F d
?
?F
??
?F
?
?
b
= ? δudx + ?
?u dx ?u
?
?u
?
δu
a
a
Require δu to satisfy homogeneous displacement boundary conditions:
δu(b) = δu(a) = 0
Then:
?
b?
?F d
?
?F
??
δI = ? δudx = 0,
?u dx ?u
?
a
4
?
?
?
?δu that satisfy the appropriate di?erentiability conditions and the homoge-
neous essential boundary conditions. Then:
?F
?u
?
d
dx
?
?F
?u
?
?
= 0
These are the Euler-Lagrange equations corresponding to the variational
problem of ?nding an extremum of the functional I.
Natural and essential boundary conditions A weaker condition on
δu also allows to obtain the Euler equations, we just need:
?F
?
b
? = 0
?u
?
δu
a
which is satis?ed if:
? δu(a) = 0 and δu(b) = 0 as before
?F
? δu(b) = 0 and
?u
?
(b) = 0
?F
?
?u
?
(a) = 0 and δu(b) = 0
?F ?F
?
?u
?
(a) = 0 and
?u
?
(b) = 0
Essential boundary conditions: δu
?
S
u
= 0, or u = u
0
on S
u
?F
Natural boundary conditions:
?u
?
= 0 on S.
Example: Derive Euler’s equation corresponding to the total po-
tential energy functional Π = U + V of an elastic bar of length L, Young’s
modulus E, area of cross section A ?xed at one end and subject to a load P
at the other end.
?
L
EA
?
du
?
2
Π(u) = dx ? Pu(L)
0
2 dx
Compute the ?rst variation:
EA du
?
du
?
δΠ =
? 2
? 2
dx
δ
dx
dx ? Pδu(L)
5
? ? ? ? ?
? ? ?
?
?
?
?
?
?
?
? ? ?
?
??
?
Integrate by parts
?
d
δΠ =
dx
?
L
EA ?
d
dx
EA δu dx ? Pδu(L)
du
dx
δu
du
dx
EA dx + EA
L
0
d
dx
du
dx
du
dx
= ? ? Pδu(L) δu δu
0
Setting δΠ = 0, ? δu / δu(0) = 0:
d
dx
?
EA
du
dx
?
= 0
du
?
P = EA ?
dx L
Extension to more dimensions
I = F (x
i
,u
i
,u
i,j
)dV
V
?
?F
V
dV
?F
δu
i,j
?u
i,j
?
?F
δI = δu
i
+
?u
i
=
V
?
?F
δu
i
+
?u
i
?
?F
δu
i
?u
i,j
δu
i
dV
?
?x
j
?
?x
j
?
?u
i,j
Using divergence theorem:
δI =
V
?
?F
?u
i
?
S
?
?F
?u
i,j
δu
i
dV +
?
?x
j
?F
δu
i
n
j
dS
?u
i,j
Extremum of functional I is obtained when δI = 0, or when:
?F
?
?F
?
?
?x
j
?
?u
i,j
= 0 , and
?u
i
δu
i
= 0 on S
u
?F
n
j
onS ? S
u
= S
t
?u
i,j
The boxed expressions constitute the Euler-Lagrange equations correspond-
ing to the variational problem of ?nding an extremum of the functional I.
6