Section 2.

13

Variation and Linearisation of Kinematic Tensors

2.13.1 The Variation of Kinematic Tensors

The Variation

In this section is reviewed the concept of the variation, introduced in Part I, §5.5.

The variation is defined as follows: consider a function u(x) , with u* (x) a second function
which is at most infinitesimally different from u(x) at every point x, Fig. 2.13.1

δu(x)
u * (x)

du

u(x) dx

Figure 2.13.1: the variation

Then define

δ u = u* ( x ) − u ( x ) The Variation (2.13.1)

The operator δ is called the variation symbol and δ u is called the variation of u(x) .

The variation of u(x) is understood to represent an infinitesimal change in the function at x.

Note from the figure that a variation δu of a function u is different to a differential du . The
ordinary differentiation gives a measure of the change of a function resulting from a specified
change in the independent variable (in this case x). Also, note that the independent variable
does not participate in the variation process; the variation operator imparts an infinitesimal
change to the function u at some fixed x – formally, one can write this as δ x = 0 .

The Commutative Properties of the variation operator

d du
(1) δu =δ (2.13.2)
dx dx

Solid Mechanics Part III 308 Kelly

 Section 2.13

Proof:

du ⎛ du ⎞ * du du * du d (u * −u) d
δ =⎜ ⎟ − = − = = (δ u(x) )
dx ⎝ dx ⎠ dx dx dx dx dx

x2 x2

(2) δ ∫ u(x)dx = ∫ δ u(x)dx (2.13.3)

x1 x1

Proof:

x2 x2 x2 x2 x2

δ ∫ u(x)dx = ∫ u * (x)dx − ∫ u(x)dx =

x1 x1 x1
∫ [u * (x) − u(x)]d x = ∫ δ u(x)dx
x1 x1

Variation of a Function

Consider A, a scalar-, vector-, or tensor-valued function of u . The value of A at u + δu ,

where δu is a variation of u is, as in, for example, 1.15.27,

A(u + δu) ≈ A(u) + ∂ u A[δu] (2.13.4)

The directional derivative in this context is also denoted by δA(u, δu ) and is called the
variation of A:

δA(u, δu ) ≡ ∂ u A[δu] = A (u + εδu )

d
(2.13.5)
dε ε =0

The variation of A is thus the directional derivative of A in the direction of the variation δu .

For example, consider the scalar function φ = P : E , where P and E are second order tensors.
Then

δφ (E, δE ) ≡ ∂ Eφ[δE] = P : (E + εδE ) = P : δE

d
(2.13.6)
dε ε =0

The second variation is defined as

δ 2 A = δ (δA ) = ∂ uδA[δu] = δA(u + εδu )

d
(2.13.7)
dε ε =0

For example, for a scalar function φ (u ) of a vector u,

Solid Mechanics Part III 309 Kelly

 Section 2.13

∂φ
δφ = ⋅ δu
∂u
(2.13.8)
∂δφ ⎛ ∂φ ⎞ ⎛ ∂ 2φ ⎞ ∂ 2φ
δ 2φ = ⋅ δu = ⎜ δ ⎟ ⋅ δu = ⎜⎜ δu ⎟⎟ ⋅ δu = δu δu
∂u ⎝ ∂u ⎠ ⎝ ∂u∂u ⎠ ∂u∂u

Variation of Functions of the Displacement

In what follows is discussed the change (variation) in functions A(u) when the displacement
(or velocity) fields undergo a variation. These ideas are useful in formulating variational
prionciples of mechanics (see, for example, §3.8).

Shown in Fig. 2.13.2 is the current configuration frozen at some instant in time. The
displacement field is then allowed to undergo a variation δu . This change to the
displacement field evidently changes kinematic tensors, and these changes are now
investigated. Note that this variation to the displacement induces a variation to x, δx , but X
remains unchanged, δX = 0 .

reference
configuration current
configuration
u(x)

X δu
x

Figure 2.13.2: a variation of the displacement

To evaluate the variation of the deformation gradient F, δF(u, δu ) , where u is the

displacement field, note that u = x − X and Eqn. 2.2.43, F (u ) = Gradu + I . One has, from
2.13.5,

δF(u, δu ) = ∂ u F[δu] = F(u + εδu )

d
dε ε =0

F(u ) + εGrad(δu )
d
= (2.13.9)
dε ε =0

= Grad(δu )

Noting the first commutative property of the variation, 2.13.2, this can also be expressed as

Solid Mechanics Part III 310 Kelly

 Section 2.13

δF (u, δu ) = δGradu (2.13.10)

Note that δu is completely independent of the function u.

Here are some other examples, involving the inverse deformation gradient, the Green-
Lagrange strain, the inverse right Cauchy-Green strain and the spatial line element:
{▲Problem 1-3}

δF −1 = −F −1gradδu
δE = F T δεF (2.13.11)
δC −1
= −2F εF −1 −T

where ε is the small strain tensor, Eqn. 2.2.48.

One also has, using the chain rule for the directional derivative, Eqn. 1.15.28, the directional
derivative for the determinant, Eqn. 1.15.32, the trace relation 1.10.10e, Eqn. 2.2.8b,

δ J (u, δu ) = δ det F(u, δu )

= ∂ u det F[δu ]
= ∂ F det F[∂ u F[δu ]]
= ∂ F det F[Grad(δu )]
(2.13.12)
[
= det F F −T : Grad(δu ) ]
(
= Jtr Grad(δu )F −1 )
= Jtr (grad(δu ))
= J div(δu )

The Lie Variation

The Lie-variation is defined for spatial vectors and tensors as a variation holding the
deformed basis constant. For example, analogous to 2.12.33a,

δ Lb a = δaij g i ⊗ g j (2.13.13)

The object is first pulled-back, the variation is then taken and finally a push-forward is
carried out. For example, analogous to 2.12.40,

δ L a(u, δu ) ≡ χ * (∂ u (χ *−1 (a ))[δu]) (2.13.14)

For example, consider the Lie-variation of the Euler-Almansi strain e. First, from 2.12.24,
( )
χ −*1 (e )b = E . Then 2.13.11b gives ∂ u χ −*1 (e )b [δu ] = δE = F T δεF . From 2.12.9a,

Solid Mechanics Part III 311 Kelly

 Section 2.13

( ( ) )
δ L e(u, δu ) = χ * ∂ u χ −*1 (e )b [δu ] = χ * (F T δεF ) = δε
b b
(2.13.15)

2.13.2 Linearisation of Kinematic Functions

Linearisation of a Function

As for the variation, consider A, a scalar-, vector-, or tensor-valued function of u . If u

undergoes an increment Δu , then, analogous to 2.13.4,

A(u + Δu ) ≈ A(u ) + ∂ u A[Δu] (1.13.16)

The directional derivative ∂ u A[Δu] in this context is also denoted by ΔA(u, Δu ) . The
linearization of A with respect to u is defined to be

L A(u, Δu ) = A(u ) + ΔA(u, Δu ) (1.13.17)

Using exactly the same method of calculation as was used for the variations above, the
linearization of F and E, for example, are

L F(u, Δu ) = F(u ) + ∂ u F[Δu ] = F + GradΔu

(2.13.18)
L E(u, Δu ) = E(u ) + ∂ u E[Δu ] = E + F T ΔεF

where Δε = 1
2
((gradΔu )T
)
+ (gradΔu ) is the linearised small strain tensor ε .

Linearisation of Variations of a Function

One can also linearise the variation of a function. For example,

L δA(u, Δu ) = δA(u, δu ) + ΔδA (u, Δu ) (2.13.19)

The second term here is the directional derivative

ΔδA[u, Δu] = ∂ u δA[Δu]

(2.13.20)
δA(u + εΔu )
d
=
dε ε =0

This leads to an expression similar to δ 2 A . For example, for a scalar function φ (u ) of a

vector u,

Solid Mechanics Part III 312 Kelly

 Section 2.13

∂δφ ∂ 2φ
Δδφ = ⋅ Δu = Δu δu (2.13.21)
∂u ∂u∂u

Consider now the virtual Green-Lagrange strain, 2.13.11b, δE = F T δεF . To carry out the
linearization of δE , it is convenient to first write it in the form

δE = F T δεF
[
= 12 F T (gradδu ) + gradδu F
T
] (2.13.22)
= 1
2
[(Gradδu) F + F Gradδu]
T T

Then

ΔδE = ∂ u δE[Δu] = ∂ u { [(Gradδu) F + F

1
2
T T
]}
Gradδu [Δu] (2.13.23)

Recall that the variation δu is independent of u; this equation is being linearised with respect
to u, and δu is unaffected by the linearization (see Fig. 2.13.3 below). However, the motion,
and in particular F, are affected by the increment in u. Thus {▲Problem 4}

(
ΔδE = sym (GradΔu ) Gradδu
T
) (2.13.24)

δu
u δu
Δu

reference current
configuration configuration

Figure 2.13.3: linearisation

As with the variational operator, one can define the linearization of a spatial tensor as
involving a pull back, followed by the directional derivative, and finally the push forward
operation. Thus

( (
Δa(u, Δu ) ≡ χ * ∂ u χ *−1 (a ) [Δu] ) ) (2.13.25)

Solid Mechanics Part III 313 Kelly

 Section 2.13

2.13.3 Problems

1. Use Eqn. 2.2.22, E = 1

2
(F T
)
F − I , Eqn. 2.13.9, δF(u, δu ) = Grad(δu ) , and Eqn. 2.2.8b,
gradv = (Gradv )F −1 , to show that δE = F T δεF , where ε is the small strain tensor, Eqn.
2.2.48.
2. Use 2.13.9 to show that the variation of the inverse deformation gradient F −1 is
δF −1 = −F −1gradδu . [Hint: differente the relation F −1F = I by the product rule and then
use the relation gradv = (Gradv )F −1 for vector v.]
3. Use the definition C = F T F to show that δC −1 = −2F −1εF − T .
4. (
Use the relation symA = 12 A T + A to show that )
{ [(Gradδu) F + F Gradδu]}[Δu] = sym((GradΔu) Gradδu )
ΔδE = ∂ u 1
2
T T T

Use δe = δε = [(gradδu ) + gradδu ] and Eqn. 2.7.21 to show that the

T
5. 1
2

Δδe = χ (∂ (χ (δe ))[Δu]) = χ sym((GradΔu ) Gradδu )

−1 T
* u * *

= sym[(gradΔu ) ⋅ gradδu ]
T

Solid Mechanics Part III 314 Kelly