Section 1.

15

1.15 Tensor Calculus 2: Tensor Functions

1.15.1 Vector-valued functions of a vector

Consider a vector-valued function of a vector

a  a(b), ai  ai (b j )

This is a function of three independent variables b1 , b2 , b3 , and there are nine partial
derivatives ai / b j . The partial derivative of the vector a with respect to b is defined to
be a second-order tensor with these partial derivatives as its components:

a(b) ai
 ei  e j (1.15.1)
b b j

It follows from this that

1
a  b  a b ai bm
  or  I,   ij (1.15.2)
b  a  b a bm a j

To show this, with ai  ai (b j ), bi  bi (a j ) , note that the differential can be written as

a1 a b j  a b j   b j   b 
da1  db j  1 dai  da1  1   da 2  a1   da3  a1 j 
b j b j ai  b a   b a   b a 
 j 1  j 2  j 3

Since da1 , da 2 , da3 are independent, one may set da 2  da 3  0 , so that

a1 b j
1
b j a1

Similarly, the terms inside the other brackets are zero and, in this way, one finds Eqn.
1.15.2.

1.15.2 Scalar-valued functions of a tensor

Consider a scalar valued function of a (second-order) tensor

   (T), T  Tij e i  e j
.
This is a function of nine independent variables,    (Tij ) , so there are nine different
partial derivatives:

Solid Mechanics Part III 124 Kelly

 Section 1.15

        
, , , , , , , ,
T11 T12 T13 T21 T22 T23 T31 T32 T33

The partial derivative of  with respect to T is defined to be a second-order tensor with

these partial derivatives as its components:

 
 e i  e j Partial Derivative with respect to a Tensor (1.15.3)
T Tij

The quantity  (T) / T is also called the gradient of  with respect to T.

Thus differentiation with respect to a second-order tensor raises the order by 2. This
agrees with the idea of the gradient of a scalar field where differentiation with respect to a
vector raises the order by 1.

Derivatives of the Trace and Invariants

Consider now the trace: the derivative of trA , with respect to A can be evaluated as
follows:

 A A A
trA  11  22  33
A A A A
A A A
 11 e i  e j  22 e i  e j  33 e i  e j (1.15.4)
Aij Aij Aij
 e 1  e1  e 2  e 2  e 3  e 3
I

Similarly, one finds that {▲Problem 1}

 trA 
I

 trA 2 
 2A T
 trA 3    
 3 A2
T

A A A
(1.15.5)

 ( trA ) 2
 2( trA)I

 ( trA) 3

 3( trA) I
2

A A
Derivatives of Trace Functions

From these and 1.11.17, one can evaluate the derivatives of the invariants {▲Problem 2}:

I A
I
A
II A
 IA I  AT Derivatives of the Invariants (1.15.6)
A
III A
A
  2
 A T  I A A T  II A I  III A A T

Solid Mechanics Part III 125 Kelly

 Section 1.15

Derivative of the Determinant

An important relation is


det A   det A A T (1.15.7)
A

which follows directly from 1.15.6c.

Other Relations

The total differential can be written as

  
d  dT11  dT12  dT13  
T11 T12 T13
(1.15.8)

 : dT
T

This total differential gives an approximation to the total increment in  when the
increments of the independent variables T11 , are small.

The second partial derivative is defined similarly:

 2  2
 ei  e j  e p  e q , (1.15.9)
TT Tij Tpq

the result being in this case a fourth-order tensor.

Consider a scalar-valued function of a tensor,  (A ) , but now suppose that the

components of A depend upon some scalar parameter t:    ( A (t )) . By means of the
chain rule of differentiation,

 dAij
  (1.15.10)
Aij dt

which in symbolic notation reads (see Eqn. 1.10.10e)

d  dA    T dA 
 :  tr    (1.15.11)
dt A dt  A  dt 

Identities for Scalar-valued functions of Symmetric Tensor Functions

Let C be a symmetric tensor, C  C T . Then the partial derivative of    C(T)  with

respect to T can be written as {▲Problem 3}

Solid Mechanics Part III 126 Kelly

 Section 1.15

 
(1)  2T for C  T T T
T C

 
(2) 2 C for C  TT T (1.15.12)
T T

    
(3)  2T 2 TT  T for C  TT and symmetric T
T C C C C

Scalar-valued functions of a Symmetric Tensor

Consider the expression

 A   Aij 
B Bij  (1.15.13)
A Aij

If A is a symmetric tensor, there are a number of ways to consider this expression: two
possibilities are that  can be considered to be
(i) a symmetric function of the 9 variables Aij
(ii) a function of 6 independent variables:    A11 , A12 , A13 , A22 , A23 , A33 
where
1
A12   A12  A21   A12  A21
2
1
A13   A13  A31   A13  A31
2
1
A23   A23  A32   A23  A32
2

Looking at (i) and writing    A11 , A12 A12 ,  , A21 A12 ,  , one has, for example,

  A12  A21   
    2 ,
A12 A12 A12 A21 A12 A12 A21 A12

the last equality following from the fact that  is a symmetrical function of the Aij .

Thus, depending on how the scalar function is presented, one could write

  
(i) B11  , B12  , B13  , etc.
A11 A12 A13
 1  1 
(ii) B11  , B12  , B13  , etc.
A11 2 A12 2 A13

Solid Mechanics Part III 127 Kelly

 Section 1.15

1.15.3 Tensor-valued functions of a tensor

The derivative of a (second-order) tensor A with respect to another tensor B is defined as

A Aij
 ei  e j  e p  e q (1.15.14)
B B pq

and forms therefore a fourth-order tensor. The total differential dA can in this case be
written as

A
dA  : dB (1.15.15)
B

Consider now

A Aij
 ei  e j  e k  el
A Akl

The components of the tensor are independent, so

A11 A11 Amn

 1,  0,  etc.   mp  nq (1.15.16)
A11 A12 A pq

and so

A
 ei  e j  ei  e j  I , (1.15.17)
A

the fourth-order identity tensor of Eqn. 1.12.4.

Example

Consider the scalar-valued function  of the tensor A and vector v (the “dot” can be
omitted from the following and similar expressions),

 A, v   v  Av

The gradient of  with respect to v is

 v v

v v
 Av  v  A
v

 Av  vA  A  A T v 
On the other hand, the gradient of  with respect to A is

 A
 v v  v  Iv  v  v
A A

Solid Mechanics Part III 128 Kelly

 Section 1.15

Consider now the derivative of the inverse, A 1 / A . One can differentiate A 1 A  0

using the product rule to arrive at

A 1 A
A   A 1
A A

One needs to be careful with derivatives because of the position of the indices in 1.15.14);
it looks like a post-operation of both sides with the inverse leads to
A 1 / A   A 1 A / A A 1   Aik1 A jl1e i  e j  e k  e l . However, this is not correct
(unless A is symmetric). Using the index notation (there is no clear symbolic notation),
one has

Aim1 Amj
Amj   Aim1 e  e j  ek  el 
Akl Akl
i

Aim1 Amj 1
 Amj A jn1   Aim1 A jn
Akl Akl
(1.15.18)
Aim1
  mn   Aim1 mk  jl A jn1
Akl
Aij1
   Aik1 Alj1 e  e j  ek  el 
Akl
i

1.15.4 The Directional Derivative

The directional derivative was introduced in §1.6.11. The ideas introduced there can be
extended to tensors. For example, the directional derivative of the trace of a tensor A, in
the direction of a tensor T, is

 A trA [T]  tr A  T   trA  trT   trT

d d
(1.15.19)
d  0 d  0

As a further example, consider the scalar function  ( A)  u  Av , where u and v are

constant vectors. Then

 A A, u, v [T] 
d
u  A  Tv   u  Tv (1.15.20)
d  0

Also, the gradient of  with respect to A is

 
 u  Av   u  v (1.1.5.21)
A A

Solid Mechanics Part III 129 Kelly

 Section 1.15

and it can be seen that this is an example of the more general relation


 A  [T]  :T (1.15.22)
A

which is analogous to 1.6.41. Indeed,


 x [ w ]  w
x

 A  [T]  :T (1.15.23)
A
v
 u v [w ]  w
u

Example (the Directional Derivative of the Determinant)

It was shown in §1.6.11 that the directional derivative of the determinant of the 2  2
matrix A, in the direction of a second matrix T, is

 A det A T  A11T22  A22T11  A12T21  A21T12

 
This can be seen to be equal to det A A  T : T , which will now be proved more generally
for tensors A and T:

 A det A [T]  det A  T 

d
d  0


d
d
 
det A I  A 1T 
 0

 det A
d
d

det I  A 1T 
 0

The last line here follows from (1.10.16a). Now the characteristic equation for a tensor B
is given by (1.11.4, 1.11.5),

1   2   3     0  detB  I 

where i are the three eigenvalues of B. Thus, setting   1 and B  A 1T ,

Solid Mechanics Part III 130 Kelly

 Section 1.15

 A det A [T]  det A

d
d
1   1 A 1T 1   2 A 1T 1   3 A 1T 
 0

 det A
d
d
1    1 A 1T 1    2 A 1T 1    3 A 1T 
 0

 det A 1  A 1T
 2 A 1T
 
3 A 1T

 det A tr A T  1

and, from (1.10.10e),

 A det A [T]  det A A  T : T  (1.15.24)

■

Example (the Directional Derivative of a vector function)

Consider the n homogeneous algebraic equations f x   o :

f 1 x1 , x 2 ,  , x n   0
f 2  x1 , x 2 ,  , x n   0

f n  x1 , x 2 ,  , x n   0

The directional derivative of f in the direction of some vector u is

f z   z  x  u 
d
 x f (x)[u] 
d  0

 f z  dz 
  (1.15.25)
 z d   0
 Ku

where K, called the tangent matrix of the system, is

 f1 / x1 f 1 / x 2  f 1 / x n 
 f 2 / x n 
f f 2 / x1 f 2 / x 2
K  ,  x f [u]  gradf u
x    
 
f n / x1  f n / x n 

which can be compared to (1.15.23c).

■

Properties of the Directional Derivative

The directional derivative is a linear operator and so one can apply the usual product rule.
For example, consider the directional derivative of A 1 in the direction of T:

Solid Mechanics Part III 131 Kelly

 Section 1.15

 
 A A 1 [T] 
d
d
A  T1
 0

 
To evaluate this, note that  A A 1 A [T]   A I [T]  0 , since I is independent of A. The
 
product rule then gives  A A 1 [T] A   A 1 A A [T] , so that

 
 A A 1 [T]   A 1 A A[T]A 1   A 1TA 1 (1.15.26)

Another important property of the directional derivative is the chain rule, which can be
applied when the function is of the form f (x)  fˆ Bx  . To derive this rule, consider (see
§1.6.11)

f (x  u)  f (x)   x f [u] , (1.15.27)

where terms of order o(u) have been neglected, i.e.

o(u)
lim u 0  0.
u

The left-hand side of the previous expression can also be written as

fˆ Bx  u   fˆ B(x)   x B[u]

 fˆ B(x)    B fˆ B [ x B[u]]

Comparing these expressions, one arrives at the chain rule,

 x f [u]   B fˆ B [ x B[u]] Chain Rule (1.15.28)

As an application of this rule, consider the directional derivative of det A 1 in the

direction T; here, f is det A 1 and fˆ  fˆ B( A )  . Let B  A 1 and fˆ  det B . Then, from
Eqns. 1.15.24, 1.15.25, 1.10.3h, f,

 
 A det A 1 [T]   B det B [ A A 1 [T]]
 
 det B  B T :  A 1TA 1  (1.15.29)
 
  det A 1 A T : A 1TA 1 
  det A 1 A : T
T

1.15.5 Formal Treatment of Tensor Calculus

Following on from §1.6.12 and §1.14.6, a scalar function f : V 2  R is differentiable at

A  V 2 if there exists a second order tensor Df A   V 2 such that

Solid Mechanics Part III 132 Kelly

 Section 1.15

f A  H   f A   Df A  : H  o H  for all H  V 2 (1.15.30)

In that case, the tensor Df A  is called the derivative of f at A. It follows from this that
Df A  is that tensor for which

 A f B  Df A  : B  f A  B 
d
for all B  V 2 (1.15.31)
d  0

For example, from 1.15.24,

  
 A det A [T]  det A A  T : T  det A A  T : T  (1.15.32)

from which it follows, from 1.15.31, that


det A  det A A T (1.15.33)
A

which is 1.15.7.

Similarly, a tensor-valued function T : V 2  V 2 is differentiable at A  V 2 if there

exists a fourth order tensor DTA   V 4 such that

TA  H   TA   DTA H  o H  for all H  V 2 (1.15.34)

In that case, the tensor DTA  is called the derivative of T at A. It follows from this that
DTA  is that tensor for which

 A TB  DTA  : B  TA  B 

d
for all B  V 2 (1.15.35)
d  0

1.15.6 Problems
1. Evaluate the derivatives (use the chain rule for the last two of these)
 trA 2   trA 3   ( trA) 2   ( trA) 2 
, , ,
A A A A
2. Derive the derivatives of the invariants, Eqn. 1.15.5. [Hint: use the Cayley-Hamilton
theorem, Eqn. 1.11.15, to express the derivative of the third invariant in terms of the
third invariant.]
3. (a) Consider the scalar valued function    CF  , where C  F T F . Use the chain
rule
  C mn
 ei  e j
F C mn Fij
to show that

Solid Mechanics Part III 133 Kelly

 Section 1.15

   
 2F ,  2 Fik
F C Fij C kj
(b) Show also that
  
 2U 2 U
U C C
for C  UU with U symmetric.
[Hint: for (a), use the index notation: first evaluate C mn / Fij using the product rule,
then evaluate  / Fij using the fact that C is symmetric.]
4. Show that
A 1 A 1
(a) : B   A 1BA 1 , (b) : A  A 1   A 1  A 1
A A
5. Show that
A T
: B  BT
A
6. By writing the norm of a tensor A , 1.10.14, where A is symmetric, in terms of the
trace (see 1.10.10), show that
A A

A A
7. Evaluate
(i)  
 A A 2 [ T]
(ii) A trA [T] (see 1.10.10e)
2

8. Derive 1.15.29 by using the definition of the directional derivative and the relation
1.15.7,  det A  / A  det A A  T .

Solid Mechanics Part III 134 Kelly