Professional Documents
Culture Documents
Derivable Functions
f,a (x) =
kx ak
0F ,
x U \ {a},
x = a,
xa
So, f is derivable at the point a if there exist f (a) L(E, F ) and f,a : U F such that
(1.2)
and
(1.3)
x U
xa
Notice that due to the uniqueness of the derivative f (a) L(E, F ), we can define a map
0
f : U L(E, F ).
0
Also, notice that the domain of the linear mapping f (a) is the entire space E, despite the fact that
the function f is defined only locally about the point a.
Proposition 1.3 If the function f : U E F is derivable at the point a U , then f is continuous
at the point a.
Proof. Let us suppose that f is derivable at the point a. Therefore, from (1.2) and the continuity
0
of f (a), we obtain
0
Hence,
xa
(f ) (a) = f (a).
Very often we need to deal with elaborate functions, obtained by composing simpler ones. The
next theorem will give us, exactly like in the one-variable calculus, the rule for differentiating composite functions.
Theorem 1.8 (Chain Rule) Let E, F, G be Banach spaces, 6= U E, 6= V F , U, V open. If
f : U V is derivable at the point a U and g : V G is derivable at the point f (a) , then g f
is derivable at the point a and
0
00
The derivative of f at the point a is denoted by f (a) and it is called the second derivative of f at
the point a.
Remark 1.11 If the function f is two-times differentiable on U and the map
f
00
: U L(E, L(E, F ))
is continuous on U , then we shall say that f is twice continuously differentiable on U and we shall
use the notation:
C 2 (U, F ) = {f : U F | f is twice continuously differentiable on U }.
00
C (U, F ) =
C n (U, F ).
nN
A function f
C (U, F )
Particular Case.
Let us briefly consider now the important particular case of one-variable calculus, i.e. the case
in which E = F = R. We shall start by recalling the well-known definition of derivability in the
one-dimensional case.
Definition 1.13 Let 6= U R be an open set and a U . The function f : U R is called
derivable at the point a if
f (x) f (a)
0
lim
= f (a) R.
xa
xa
0
x U .
Indeed, the above definitions are equivalent. It is enough to notice that if f satisfies Definition 1.13,
then the limit
f (x) f (a)
lim
xa
xa
exists. Let
f (x) f (a)
= lim
R
xa
xa
and let
f (x) f (a)
, x U \ {a},
xa
w (x) =
f,a
x = a.
Then, wf,a is continuous at the point a and condition 2) in Definition 1.14 is satisfied.
Conversely, if f satisfies Definition 1.14, then the limit
f (x) f (a)
xa
xa
lim
exists and belongs to R. Hence, the above definitions are equivalent. Notice that
0
x U.
2) Moreover, if f is injective and continuous on U and f (a) 6= 0, then the inverse function f 1
is derivable at the point b = f (a) and
0
(f 1 ) (f (a)) =
1
.
f (a)
0
Proposition 1.16 (Chain Rule) Let U, V R be open nonempty sets and let f : U R and
g : V R such that f (U ) V . If f is derivable at the point a U and g is derivable at the point
b = f (a) V , then g f is derivable at the point a and
0
Also, let us recall the following well-known results from the one-variable calculus:
Definition 1.17 Let f : U R R, U open and nonempty, a U .
a) The function f attains a local minimum at the point a if there exists V V(a) such that
f (x) f (a), for any x V U .
b) The function f attains a local maximum at the point a if there exists V V(a) such that
f (x) f (a), for any x V U .
The point a is called a local extremum (a minimum or a maximum) for the function f . We shall
use the terms global or absolute maximum and global or absolute minimum to refer to the overall
maximum and minimum values of the function on the range under consideration.
Theorem 1.18 (Fermat) Let f : U R R, U open and nonempty, a U . If f is derivable at
0
the point a and a is a local extremum for f , then f (a) = 0.
0
The points a at which f (a) = 0 are called stationary points. If a function f has a local extremum
0
0
at a point a, then either f (a) = 0 or f (a) does not exist. Such points are called critical points.
Theorem 1.19 (Rolle) Let f : [a, b] R. If f is continuous on [a, b], derivable on (a, b) and
0
f (a) = f (b), then there exists c (a, b) such that f (c) = 0.
Theorem 1.20 (Lagranges Mean-Value Theorem) Let f : [a, b] R. If f is continuous on [a, b],
derivable on (a, b), then there exists c (a, b) such that
f (b) f (a)
0
= f (c).
ba
The formula
(1.4)
f (b) f (a)
f (c)
= 0 .
g(b) g(a)
g (c)
Exercises.
1) Let f : R R,
x sin ,
f (x) =
0,
x 6= 0,
x = 0.
2x
= sgn x,
1 + x2
| x | 1.
Let us consider now briefly another important case, i.e. the case in which E = Rn and F = Rm .
So, let 6= U Rn be an open set, f : U Rm and a U . If
f = (f1 , . . . , fm ),
then f is derivable at the point a if and only if each component fi , with i = 1, m, is derivable at the
point a ( the componentwise nature of differentiability).
Example. The function f : R2 R3 defined by
f (x, y) = (x2 + y 2 , xy, x3 + y 3 )
is obviously differentiable on R2 , since its three components are composed of elementary functions.
Partial Derivatives
n
Y
Ei .
i=1
Let us define
pi : E Ei , pi (x) = xi , x = (x1 , . . . , xn ) E
and
ui : Ei E, ui (xi ) = (0E1 , . . . , 0Ei1 , xi , 0Ei+1 , . . . , 0En ), x = (x1 , . . . , xn ) E.
pi are called the canonical projections and ui the canonical injections.
Properties.
1) It is easy to see that pi and ui are linear.
2) pi and ui are continuous. Indeed,
n
X
kxk k = kxk1 .
k=1
n
X
k=1
kxk k = kxi k.
n
X
ui pi = 1E .
i=1
Indeed,
(pi ui )(xi ) = pi (0, . . . , xi , . . . , 0) = xi = 1Ei (xi ).
and
n
X
n
X
i=1
i=1
(ui pi )(x) =
Remark 2.2 Since pi and ui are linear and continuous, it follows immediately that pi and ui are
derivable.
Definition 2.3 Let =
6 U E, U open and let a U . For any 1 i n, let us define
a,i : Ei E as being:
a,i (xi ) = a + ui (xi ai ), xi Ei ,
i.e.
a,i (xi ) = (a1 , . . . , ai1 , xi , ai+1 , . . . , an ).
The role played by this map is to fix the variable xi in a.
Let us notice that a,i is continuous and 1
a,i (U ) is open in Ei .
Definition 2.4 Let E1 , E2 , . . . , En , F be Banach spaces and E =
n
Q
i=1
Ei . Also, let 6= U E, U
open and f : U F . The function f is called derivable with respect to xi at the point a U (or
partial derivable with respect to xi at the point a) if the map f a,i : 1
a,i (U ) F is derivable
at ai 1
(U
).
In
this
case,
the
derivative
of
the
function
f
at
the
point
ai will be called the
a,i
a,i
f
(a).
partial derivative of f with respect to xi at the point a and we shall denote this derivative by
xi
So,
f
0
(a) = (f a,i ) (ai ).
xi
Theorem 2.5 Let f : U E F , U open and nonempty. If f is derivable at the point a U ,
then f is (partial) derivable with respect to each variable xi , 1 i n, at the point a and
f
0
(a) = f (a) ui ,
xi
n f
P
0
f (a) =
(a) pi .
i=1
xi
Proof. Since
a,i (xi ) = a + ui (xi ai ),
xi Ei ,
0
it follows that a,i is derivable at any point t Ei and (a,i ) (t) = ui . Then, f a,i is derivable and
f
0
0
0
0
(a) = (f a,i ) (ai ) = f (a,i (ai )) (a,i ) (ai ) = f (a) ui .
xi
7
n
X
ui pi ) =
i=1
n
X
f (a) ui pi =
i=1
n
X
f
i=1
xi
(a) pi .
f
: U L(Ei , F )
xi
and
f
(a) L(Ei , F ).
xi
n
Q
i=1
Ei , F =
m
Q
j=1
Fj . Also,
let U E, U open and nonempty and f : U F be derivable at the point a. Then, the derivative
0
f (a) is determined by the matrix
.............................
fm
fm
f1
f1
x (a) x (a)
1
n
x1
(a)
xn
(a)
In particular, for E = Rn and F = Rm , it is not difficult to see that we have the following result:
Proposition 2.8 Let 6= U Rn , U open, f : U Rm derivable at the point a U . The matrix
0
associated to the linear map f (a) : Rn Rm in the canonical bases of Rn and Rm , called the Jacobi
matrix, is
f1
f1
x (a) x (a)
1
n
Mf (a) = ..............................
fm
fm
x1
(a)
xn
(a)
If m = n, the determinant
Jf (a) = det Mf (a)
is called the Jacobian of the function f at the point a.
If f : U Rn Rn , U open and nonempty and if f = (f1 , . . . , fn ), we shall denote the Jacobian
of the vector valued function f at the point a by
Jf (a) =
(f1 , ..., fn )
(a).
(x1 , . . . , xn )
So, if a function is differentiable at a point, its derivative is given by the Jacobian matrix.
Example. If we consider the function f : R2 R2 defined by
f (x, y) = (x2 + y 2 , 2xy),
8
Mf (1, 1) =
n
Q
i=1
Ei . Also, let 6= U E,
1 i n, are
f (a) L(Rn , R)
and
f
(a) L(R, R) ' R.
xi
Therefore,
we get
f
(a) is a real number and, hence, in Theorem 2.5, becomes the usual product. So,
xi
0
(2.1)
f (a) =
n
X
f
i=1
xi
(a) dxi ,
where
f (a1 , . . . , ai1 , xi , ai+1 , . . . , an ) f (a1 , . . . , ai1 , ai , ai+1 , . . . , an )
f
(a) = lim
.
xi ai
xi
xi ai
In fact, {p1 , . . . , pn } is the canonical basis for L(Rn , R) and, so, (2.1) is just the unique representation
f
0
of the linear map f (a) in this basis. The partial derivatives
(a) are exactly the coefficients of
xi
0
f (a) in the canonical basis.
0
Since f (a) L(Rn , R) ' Rn , the derivative f (a) can be also regarded as being the vector
0
f (a) ' (
f
f
(a), . . . ,
(a)).
x1
xn
9
f (a, b) =
f
f
(a, b)dx +
(a, b)dy,
x
y
f
f (a, y) f (a, b)
(a, b) = lim
.
yb
y
yb
0
Sometimes, the partial derivatives can be denoted by fx (a, b) and, respectively, fy (a, b). In fact,
the partial derivative of the function f with respect to x at the point (a, b) is the usual derivative
at the point a of the one-dimensional function t f (t, b). Of course, the partial derivative of
the function f with respect to y at the point (a, b) is the usual derivative at the point b of the
one-dimensional function t f (a, t).
Let us see now which are the interpretations for partial derivatives.
0
On one hand, it is not difficult to see that fx represents the rate of change of the function f as
0
we change x, holding y fixed, while fy represents the rate of change of f as we change y, holding x
fixed.
0
On the other hand, we know from one-variable calculus that f (a) represents the slope of the
0
tangent line to y = f (x) at x = a. The partial derivatives fx (a, b) and fy (a, b) also represent the
slopes of some tangent lines. More precisely, the partial derivative fx (a, b) is the slope of the trace
0
of f (x, y) for the plane y = b at the point (a, b). Also, the partial derivative fy (a, b) is the slope of
the trace of f for the plane x = a at the point (a, b).
Partial derivatives are very easy to compute: first, we have to fix all but one of the variables and
then we have to take the one-variable derivative with respect to the variable that remains.
Examples.
1) For the function f : R3 R defined by
f (x, y, z) = x2 ey + z,
we get
f
d 2 b
(a, b, c) =
(x e + c) |x=a = 2aeb ,
x
dx
f
d 2 y
(a, b, c) =
(a e + c) |y=b = a2 eb
y
dy
10
and
f
d 2 b
(a, b, c) =
(a e + z) |z=c = 1.
z
dz
2) Let f : R2 R be defined by
q
f (x, y) =
x2 + y 2 .
f
f (0, y) f (0, 0)
|y|
(0, 0) = lim
= lim
.
y0
y0 y
y
y0
But these limits dont exist and, hence, f is not differentiable at the point (0, 0).
3) Define f : R2 R as follows:
xy
x2 + y 2 ,
f (x, y) =
0,
(x,y)(0,0)
f
f
differentiable at the point (0, 0). Still, the partial derivatives
(0, 0) and
(0, 0) exist and are
x
x
equal to zero.
4) Test the following function for differentiability:
f (x, y) =
x2 y
,
x2 + y 2
0,
f (a, b) = (
f
f
2ab3
a2 (a2 b2 )
(a, b),
(a, b)) = ( 2
,
),
x
y
(a + b2 )2 (a2 + b2 )2
f
(0, 0) = 0. Therefore, if the function f would be differentiable at the
y
point (0, 0), then the derivative of f at this point should be the zero map and
and, in a similar manner,
x2 y
x2 + y 2
p
lim
= 0.
x2 + y 2
(x, y) (0, 0)
But this limit doesnt exist and, hence, f is not differentiable at the origin.
5) Define f : R2 R as follows:
f (x, y) =
x = y 6= 0,
otherwise.
f
f
Show that f is not continuous at the point (0, 0), but both partial derivatives
(0, 0) and
(0, 0)
x
x
exist.
It is easy to see that if we let (x, y) to approach the origin along the line x = y, then
1 6= f (0, 0) and this implies the fact that f is not continuous at the origin.
However, both partial derivatives exist at the origin. Indeed,
f
f (x, 0) f (0, 0)
0
(0, 0) = lim
= lim = 0
x0
x0 x
x
x0
and, in a similar manner,
f
(0, 0) = 0.
y
Exercises.
1) Let f : R2 R be defined by
f (x, y) = ln(1 + 2x2 + 3y 2 ).
0
xy
px2 + y 2 ,
f (x, y) =
0,
f (x, y) =
1
2
2
,
(x + y ) sin 2
x + y2
0,
lim
(x,y)(0,0)
f (x, y) =
f (x, y) =
x2 y 3
,
x2 + y 2
0,
n
Q
i=1
Ei . Also, let 6= U E,
U open and f : U F . The function f is said to be two-times partial derivable at the point a U
with respect to the argument xi and then with respect to the argument xj if there exists U0 open
such that:
1) a U0 U ;
2) f is derivable with respect to xi on U0 ;
f
: U0 L(Ei , F ) is derivable with respect to xj at the point a.
3) the map
xi
f
2f
The partial derivative of the map
with respect to xj at the point a is denoted by
(a).
xi
xj xi
Let us notice that
2f
(a) L(Ej , L(Ei , F )) ' L2 (Ej , Ei ; F ).
xj xi
n
Q
i=1
Ei . Also, let 6= U E, U
00
(f (a)x)y = (f (a)y)x,
x, y E.
lim
= 0.
kx ak
xa
x 6= a
Particular Case.
Let Ei = F = R. Therefore, E = Rn . Let f : U Rn R. In this particular case,
00
2f
2f
(a)
=
(a), i, j = 1, n,
xj xi
xi xj
(3.1)
n
2f
00
f
(a)
=
(a) pi pj .
xi xj
i,j=1
In (3.1),
f
f
(a1 , . . . , xi , . . . , an )
(a1 , . . . , ai , . . . , an )
2f
xj
xj
(a) = lim
xi ai
xi xj
xi ai
2f
f
(a)
(a)
x2
x1 xn
Hf (a) = .................................
2f
2f
xn x1
(a)
x2n
(a)
Hf (x, y, x) = 2yz 3
3y 2 z 2
2yz 3
3y 2 z 2
2xz 3
6xyz 2
6xyz 2
6xy 2 z
i, j = 1, n.
Remark 3.7 The existence of the mixed partial derivatives of the second order for a given function
doesnt imply their symmetry.
As a counterexample, it is not difficult to see that for the function f : R2 R, defined by:
x2 y 2
one has
2f
2f
(0, 0) 6=
(0, 0).
xy
yx
2f
(0, 0). We get:
xy
f
f
(x, 0)
(0, 0)
2f
y
y
(0, 0) = lim
.
x0
xy
x0
But
and
So,
f (x, y) f (x, 0)
f
(x, 0) = lim
=x
y0
y
y0
f
f (0, y) f (0, 0)
(0, 0) = lim
= 0.
y0
y
y0
2f
x0
(0, 0) = lim
= 1.
x0 x 0
xy
2f
(0, 0), we obtain:
yx
f
f
(0, y)
(0, 0)
2f
x
x
(0, 0) = lim
.
y0
yx
y0
But
and
So,
Hence,
f
f (x, y) f (0, y)
(0, y) = lim
= y
x0
x
x0
f
f (x, 0) f (0, 0)
(0, 0) = lim
= 0.
x0
x
x0
2f
y 0
(0, 0) = lim
= 1.
y0 y 0
yx
2f
2f
(0, 0) 6=
(0, 0).
xy
yx
Also, we shall usually omit to write explicitly the symbol . So, we have:
00
f (a) =
n
X
2f
(a) dxi dxj .
xi xj
i,j=1
f (a, b) =
2f
2f
2f
2f
2
(a,
b)
dx
+
(a,
b)
dx
dy
+
(a,
b)
dy
dx
+
(a, b) dy 2 .
x2
xy
yx
y 2
n
X
i1 ,...,im
mf
(a) pi1 pim .
xi1 xim
=1
Examples.
1) Verify that the function f : R2 R, defined by f (x, y) = ex cos y is a solution of Laplaces
equation, i.e.
2f
2f
+
= 0.
x2
y 2
Indeed, it is easy to see that f C 2 (R2 ). It follows immediately that
2f
= ex cos y
x2
and
2f
= ex cos y.
y 2
f (x, y) = ex
Obviously, f C 2 (R2 ) and it is not difficult to compute its second order partial derivatives at the
point (1, 1). As a result, we get
00
f (1, 1) = 6 e2 dx2 + 4 e2 dx dy + 4 e2 dy dx + 6 e2 dy 2 .
Exercises.
x2
1) Compute the first and the second differential for the function f : R2 R,
+ y 2 + ln (1 + x2 + y 2 ) at the point (1, 1).
16
f (x, y) =
2) Compute the second derivative at the point (2, 2) of the function f : R2 R2 , defined by
f (x, y) = (x2 + y, x + y 2 ).
3) For the function f : R2 R, defined by:
f (x, y) =
prove that
x2 y 2
,
xy sin 2
2
x +y
0,
2f
2f
(0, 0) 6=
(0, 0).
xy
yx
4) Compute the first and the second differential at the point (1, 1) for the function f : R2 R,
defined by
f (x, y) = sin (x2 + y 2 ).
Directional Derivatives
We have already seen one possible interpretation for partial derivatives. Indeed, we saw in the
0
previous section that fx represents the rate of change of the function as we change x, holding y fixed,
0
while fy represents the rate of change of f as we change y, holding x fixed. Still, sometimes, we are
interested in finding the rate of change of a given function f in a particular direction, given by a
vector u. This rate will be expressed by the so-called directional derivative of f .
6 U E, U open, a U . Also, let f : U F
Definition 4.1 Let E and F be Banach spaces, =
and u E, u 6= 0. If there exists
f (a, u) = lim
t0
f (a + tu) f (a)
F,
t
then f (a, u) is called the derivative of the function f , at the point a, in the direction u.
Sometimes, we shall use similar notation, i.e.
f (a, u) =
or
f
(a)
u
0
f (a, u) = fu (a).
Usually, a direction in Rn is specified by a unit vector, i.e. a vector u Rn with kuk = 1. So,
often, we shall consider that the vectors u are taken to be normalized, although the definition above
works for arbitrary (even zero) vectors.
Example. If f : R2 R is defined by
f (x, y) = 4 2x2 y 2
17
and
1
u = (1, 1),
2
then
f
(1, 1) = 3 2.
u
Properties.
1) If f is derivable at the point a in the direction u, then f is derivable at a in the direction u
and
f
f
(a) = (a).
(u)
u
More generally, if f is derivable at the point a in the direction u, then f is derivable at a in the
direction u, K and
f
f
(a) = (a).
(u)
u
2) If f is derivable at the point a in the direction u, then f is continuous at a in the direction u.
Theorem 4.2 Let E and F be Banach spaces, 6= U E, U open, a U . Also, let f : U F
and u E, u 6= 0. If f is derivable at the point a, then f is derivable at the point a in any direction
u 6= 0 and
0
f (a, u) = f (a)(u), u 6= 0.
0
Proof. Since f is derivable at the point a, it follows that there exists f (a) L(E, F ) such that
0
(4.1)
lim
xa
x 6= a
f (a + hu) f (a)
0
= f (a)(u),
h
for any u 6= 0, which ends the proof of the theorem.
The converse implication in Theorem 4.2 is not true. As a counterexample, let us consider the
function f : R2 R, defined by:
lim
h0
f (x, y) =
x2 y
,
x2 + y 2
0,
It is not difficult to see that this function is not differentiable at the point (0, 0), despite the fact
that it possesses directional derivatives at this point with respect to any direction u 6= 0.
18
We have previously used in some contexts the notation
u for denoting a vector from Rn , instead
of the usual notation u. This was done only to explicitly mark the fact that we are dealing with a
x2 + y 2
f (x, y) =
is continuous at the point (0, 0), but it is not Gateaux differentiable at this point. On the other
hand, if we consider the function f : R2 R, defined by
x2
f (x, y) =
0,
y 6= 0,
y = 0,
it is not difficult to see that f is Gateaux differentiable at the point (0, 0), but it is not continuous
at this point.
Properties.
1) If U 6= , U open and fi : U Rn R, i = 1, n, are derivable at the point a U in the
direction u, then
n
X
i=1
n
n
X
X
fi
( ci fi )(a) =
ci
(a).
u i=1
u
i=1
f
g
(f g)(a) =
(a)g(a) + f (a) (a).
u
u
u
3) If U 6= , U open and f : U Rn Rm , f = (f1 , . . . , fm ), then f is derivable at the point
a U in the direction u if and only if f1 , . . . , fm are derivable at the point a in the direction u.
Let us take now f : U Rn R, U 6= , U open, a = (a1 , . . . , an ) U . We shall see in
what follows another possible way for defining the partial derivatives of f , in terms of directional
derivatives. We know that
f
f (a1 , . . . , xj , ..., an ) f (a1 , ..., aj , . . . , an )
(a) = lim
.
xj aj
xj
xj aj
19
If we take
t = xj aj ,
then
f (a + tej ) f (a)
f
f
(a) = lim
=
(a),
t0
ej
t
xj
where
ej = (0, . . . , 1, . . . , 0).
Hence, partial derivatives can be seen as particular directional derivatives along the standard basis
vector directions.
Let us suppose that f : Rn R is derivable at a given point a. Then,
0
u 6= 0.
So, the slopes in all directions are known from slopes in n directions (given by the partial derivatives).
Exercises.
1) Compute the directional derivative of the function f : R2 R2 , defined by
f (x, y) = (x2 + y, x + y 2 )
at the point a = (2, 3), in the direction of the vector u = (1, 1).
3) Let f : R2 R, defined by:
f (x, y) =
a)
b)
c)
u with
d)
xy 2
,
+ y2
x2
0,
Composite Functions
f1
f1
x (a) x (a)
1
n
fm
fm
x1
(a)
xn
20
(a)
If m = n, the determinant
Jf (a) = det Mf (a)
was called the Jacobian of the function f at the point a and it was also denoted by
Jf (a) =
(f1 , . . . , fn )
(a).
(x1 , . . . , xn )
dg u
F
0 u
=
=g
,
x
du x
x
F
dg u
0 u
=
=g
.
du y
dg u
dg
F
0
=
=
2x = g 2x,
x
du x
du
dg u
dg
F
0
=
=
2y = g 2y.
C1
du y
du
2) Let U and V be nonempty open sets in R2 and let f = (u, v), f : U V and g : V R be
functions. If
F (x, y) = g(u(x, y), v(x, y)),
21
F
g u g v
=
+
,
x
u x v x
F
g u g v
=
+
.
u y
v y
Example. Let
F (x, y) = xy + g(x + y, x3 + y 3 ),
where g : R2 R is an arbitrary given function of class C 1 . If we denote by
u(x, y) = x + y
and by
v(x, y) = x3 + y 3 ,
then
F
g u g v
=y+
+
,
x
u x v x
i.e.
F
g
g
=y+
+ 3x2 .
x
u
v
F
g
g
=x+
+ 3y 2 .
y
u
v
3) Let U and V be nonempty open sets in R3 and let f = (u, v, w), f : U V and g : V R
be C 1 functions. If
F (x, y, z) = g(u(x, y, z), v(x, y, z), w(x, y, z)),
then F is of class C 1 on U and
g u g v
g w
F
=
+
+
,
x
u x v x w x
F
g u g v
g w
=
+
+
,
y
u y v y w y
F
g u g v
g w
=
+
+
.
u z
v z
w z
Example. Let
F (x, y, z) = xyz + g(x + y + z, x2 + y 2 + z 2 , xyz),
where g : R3 R is an arbitrary given function of class C 1 . If we denote
u(x, y, z) = x + y + z,
v(x, y, z) = x2 + y 2 + z 2
and
w(x, y, z) = xyz,
then
F
g u g v
g w
= yz +
+
+
,
x
u x v x w x
22
i.e.
F
g
g
g
= yz +
+ 2x
+ yz
.
x
u
v
w
and
F
g
g
g
= xz +
+ 2y
+ xz
.
y
u
v
w
F
g
g
g
= xy +
+ 2z
+ xy
.
z
u
v
w
fl
() =
()
.
xj
fl xj
l=1
For second order partial derivatives, if assume the needed regularity, we get:
m X
m
m
X
hk
2 gk fr fl X
gk 2 fl
(
)=
+
.
xi xj
f
f
x
x
f
x
x
r
i
j
i
j
l
l
r=1
l=1
l=1
Example. Let
F (x, y) = xy + g(x2 + y 2 , xy),
where g : R2 R is an arbitrary given function of class C 2 . If we denote by
u(x, y) = x2 + y 2
and by
v(x, y) = xy,
then
and
2
2
2F
2g
g
2 g
2 g
=
4x
+
4xy
+
y
+2 ,
2
2
2
x
u
uv
v
u
2
2
2
F
g
g
g
g
= 1 + 4xy 2 + (2x2 + 2y 2 )
+ xy 2 +
xy
u
uv
v
u
2
2
2g
g
2F
2 g
2 g
=
4y
+
4xy
+
x
+2 .
2
2
2
y
u
uv
v
u
Exercises.
1) Compute the first and the second differential for the function
F (x, y) = (x2 + y 2 )u(x + y),
where u : R R is an arbitrary function of class C 2 .
2) Let
F (x, y) = xyu(x + y, xy),
23
where u : R2 R is a given function of class C 2 . Compute the first and the second differential of F .
3) Let F (x, y, z) = (x2 + y 2 + z 2 )u(x + y + z, xyz), where u C 2 . Compute the first and the
second differential of F .
4) Let F (x, y) = u(x + y, xy, x2 + y 2 ), where u C 2 . Compute
F =
2F
2F
+
.
x2
y 2
Differential Operators
f
f
(a), . . . ,
(a)).
x1
xn
Usually, it is convenient to work with scalar fields of class C 1 . So, we obtain the following definition:
Definition 6.2 Let 6= U Rn , U open. Also, let f : U Rn R be a scalar field, f C 1 (U )
and a U . The gradient of the function f at the point a is defined as being
grad f : U Rn , (grad f )(a) = grada f.
24
f
f
(a), . . . ,
(a)).
x1
xn
f
f
(a), ,
(a))
x1
xn
or
f (a) =
n
X
f
i=1
xi
(a)ei .
,,
)
x1
xn
is called the Hamiltonian operator. So, in the n-dimensional Euclidian space Rn , this operator can
be written as
n
X
=
ei
xi
i=1
or, using the Einstein summation notation, as
=
ei
xi
The name of the symbol comes from the Greek word for a Hebrew harp, which has a similar
shape. The symbol was used for the first time in 1837 by William Rowan Hamilton (1805-1865).
Another name for this symbol is atled (delta spelled backwards), because this symbol nabla looks
like an inverted delta. Also, the word del is used for the Hamiltonian operator.
Let us see now which is the connection between the gradient and the directional derivative. Let
(a) = hf (a),
u i = |f (a)| cos ,
in the direction
u depends on
u and varies from |f (a)| (when = , i.e.
u points in a direction
opposite to f (a)) to |f (a)| (when = 0, i.e. u points in then same direction as f (a)). The
vector f (a) points in the direction of the greatest rate of increase of the function f at the point a
and the greatest rate of change is exactly the magnitude of this vector, i.e. |f (a)|.
For instance, let us consider a room in which, at each point (x1 , x2 , x3 ), the temperature is given
by the scalar field T = T (x1 , x2 , x3 ). If we assume that the temperature does not change in time,
then, at each point in the room, the gradient at that point will show the direction in which the
temperature rises most quickly. The magnitude of the gradient will tell us how fast the temperature
rises in that direction.
A generalization of the gradient, for functions which have vectorial values, is the Jacobian matrix.
The gradient f is orthogonal to the level surface f (x) = C. Indeed, since
f d
r = df
25
f d
r = 0.
Examples. 1) The gradient of the scalar field f : R3 R, defined by
f (x, y, z) = 2x + y 2 cos z,
is the vector:
f = (2, 2y, sin z).
r = k
r k 6= 0.
1
Computing grad ( ), we get
r
r
1
grad ( ) = 3 .
r
r
f 0 (r)
r,
r
for any f C 1 (U ), U R+ .
We shall define now the so-called Laplace operator (Pierre-Simon Laplace, 1749-1827) or the
Laplacian, denoted by or by 2 . This operator is a second order linear differential operator,
arising in the modeling of various physical phenomena, such as wave propagation and heat flow
in homogeneous bodies. Also, this operator appears in electrostatics, in dispersion theory, in the
so-called Laplaces and Poissons equations, in quantum mechanics, etc.
Definition 6.3 Let 6= U Rn , U open and f : U Rn R be a scalar field such that f C 2 (U ).
The Laplacian of the function f is defined as being:
f =
2f
2f
.
+
+
x2n
x21
So, the Laplace operator is a second order differential operator defined as the sum of all the unmixed
second partial derivatives:
n
X
2
=
.
x2i
i=1
Remark 6.4 A function f is said to be harmonic if f = 0.
Example. Let f : R2 \ {(0, 0)} R be defined by
f (x, y) = ln(x2 + y 2 ).
Then, f satisfies Laplaces equation, i.e.
f = 0.
Indeed, it is not difficult to see that
2(y 2 x2 )
2f
=
.
x2
(x2 + y 2 )2
26
In a similar manner,
2f
2(x2 y 2 )
=
.
y 2
(x2 + y 2 )2
Hence,
2f
2f
+
= 0.
x2
y 2
f =
This proves that our function f is harmonic.
Let us define now the divergence operator, acting on vector fields, and measuring the magnitude
of a given vector fields source or sink at a given point. In other words, the divergence of a vector
at a point a will be a signed scalar which measures the spreading of the given vector field at that
point, i.e. the extent to which the vector field flow behaves like a source or a sink at the point a.
div
v =
v1
vn
+ +
.
x1
xn
The divergence of
v can be interpreted as the symbolic scalar product of the Hamiltonian
div
v =
v.
Notice that the divergence of a vector field is a scalar field.
1) Let
r = (x, y, z) R3 ,
r =
6 0 and
r = k
r k,
i.e.
r=
x2 + y 2 + z 2 .
q
E = 3
r
In a similar manner,
and
Therefore,
Ex
r3 3x2 r
=q
.
x
r6
Ey
r3 3y 2 r
=q
y
r6
Ez
r3 3z 2 r
=q
.
z
r6
div E = 0.
div
r = 3.
Let us define now the so-called curl operator, acting on vector fields in the Euclidian space R3 .
The curl of a vector field at a point will measure the amount of swirling or spinning. So, the curl
will be the measure of the rotational strength at a point.
6 U R3 , U open and
v : U R3 be a vector field,
v C 1 (U ). The
Definition 6.7 Let =
differential operator
v3
v2 v1
v3 v2
v1
rot
v =(
)
x2 x3 x3 x1 x1 x2
The curl of
v can be interpreted as the symbolic vector product of the Hamiltonian operator
curl
v =
v,
i.e.
curl v =
e1
e2
e3
x1 x2 x3
v1
v2
v3
The most important physical examples of irrotational vectors are the gravitational and, respectively, the electrostatic forces.
Exercises.
1 Prove that
curl
r =0
28
and
curl (f (r)
r ) = 0,
for any f C 1 (U ), U R+ .
2) Let V = V0 +
r be the velocity field in a rigid solid. It is not difficult to prove that
curl V = 2
.
and
div V = 0.
Remark 6.9 If
v is an irrotational field in a simply connected neighborhood U of a point x, then
in this neighborhood,
v is a potential field, i.e.
v is given in terms of the gradient of a scalar field
:
v = .
We shall list now some properties of the above introduced differential operators and some relationships between them. The proof of these properties is very simple and it is left as an exercise to
the reader.
Properties.
Let U R3 be a nonempty open set. Prove that:
1) grad (cf ) = c grad f,
c R, f : U R, f C 1 (U ).
f, g : U R, f, g C 1 (U ).
f, g : U R, f, g C 1 (U ).
4) grad (f /g) =
ggrad f f grad g
,
g2
f, g : U R, f, g C 1 (U ),
g(x) 6= 0, x U .
0
5) grad F () = F () grad ,
F, C 1 .
6) div (
u +
v ) = div
u + div
v,
7) div (
v ) = div
v + (grad )
v,
, R,
u,
v C 1.
,
v C 1.
8) div (
u
v)=
v rot
u
u rot
v ,
u,
v C 1 . As a consequence, if
u and
v are irrotatational vector fields, then
u
v is solenoidal.
9) div (grad ) = 4 ,
10) rot (grad ) = 0,
C 2 .
C 2 .
11) rot (
u +
v ) = rot
u + rot
v,
, R,
u,
v C 1.
12) rot (
v ) = rot
v
v grad ,
,
v C 1.
u C 2.
v C 2.
1
r
r
= 3 , grad ln r = 2 ,
r = (x, y, z) R3 , r =|
r |6= 0.
r
r
r
29