IR is continuous in
Chapter 6 DIFFERENTIABLE MAPPINGS In this chapter we will generalize the notion of derivability
known from high school for several variable functions. The notion of differentiability which we will study here will allow us to approximate the value of a function in a certain point with the value of a polynomial of first degree in that point, improving the known results from the continuous functions .
p
We remind that if
f : A IR a , A
→ IR is continuous in then there exists a neighbourhood for which for
U V a f x f (a )
(approximation with a constant). Here we will show that,
x U A
in some conditions, there is a first-degree polynomial of
p variables for which , . T f x T x x U A
1 1
1. PARTIAL DERIVATIVES
p
. A point
Definition 1. Let A IR a A is named interior point
of the set
A if there exists a sphere with the centre in a included in A, i.e.
such that r S a,r . A The set of the interior points of the set
A is named the interior of the set A and is denoted Å or Int A. p
is open if and only if
Remark 1. The set A IR A=Å. Indeed,
if z is open, then for any so that
a A there exists r S a,r , A
hence Å ; from definition
a is an interior point and consequently A
1 results that Å
A, hence A=Å. Conversely, if Å=A, then for any a A
there exists so that
r S a,r , hence A is an open set. A Remark 2. The interior of the set A is the largest open set
included in
A; therefore if G is an open set included in A then G Å.
3
2
2
2 ≤
Example 1. Let A={(x,y,z) IR x,y,z ≥ 0, x +y +z 1}.
3
2
2
2 Then Å ={(x,y,z) IR x,y,z > 0, x +y +z < 1}. p
then Å Proposition 1. If A IR A' .
Proof. Let a Å. According to definition 1 results that there
exists , hence
r>0 so that S a,r ; then A S a,r A\ a Ø a is an
accumulation point of the set A.
then there exists so that Corollary. If a Å x A\ a n
. This affirmation results from proposition 1 and lim x a n n proposition 5 from chapter 5.
p q
be a function and
Definition 2. Let f : A IR IR
respectively the canonical bases of the
B e ,...,e B e ,..., e p 1 p q 1 q p, q
. We say that the function linear spaces IR respectively IR f is
partially derivable according to (where ) in the point
x i i 1 , 2 ,...,p aÅ if exists:
1
q
. (*) lim [f(a te ) f(a)] t i
IR t IR t This limit is named the partial derivative of the function
f
according to (with respect to, or shortly, w.r.t) in the point a and is
x idenoted
f or .
a f a x i
x i
If f has a partial derivative according to every variable we say that
f is partially derivable in
a. If A=Å and f is partially derivable according
to x in any point a A we say that f is partially derivable according to
i (on x A); in this case the function: i
f q
IR f :A , a f (a) x x i i
x i
is named partial derivative of the function
f according to x . If f is i
continuous on in any point
A, partially derivable according to x i a A
1 1
we say that on
f is of class C A and we write f C (A) .
Å , then exists so that
Remark 1. If a r S a,r ; in this A
case for , then the problem of determining the
t r, a te A i partial derivative from (*) has sense.
f
1 , or,
Remark 2. In case p
1 , a lim f a t f a t
x t
denoting x=a+t,
f f x f a
;
a
lim x a
x x a
then in the case of the functions with a single variable the partial derivative coincides with the derivative and we will keep the
df
notations from high-school: .
f a a
dx
2
In case Å : p=2, if a,b IR
1
f a,b f( a,b t x lim 1 , ) f a,b t t
1 f x,b f a,b
lim f a t,b f a,b lim t x a t x a
t x a
and, analogously
f a,y f a,b f a,b . y lim y b y b
, and , then according to
Remark 3. If f f ,...,f a Å 1 q
proposition 1, , and from theorem 7 (chapter 5) it
a a ,...,a A' 1 p
results that
f is partially derivable if and only if the q scalar
components have partial derivatives in
a; also:
f
f f q 1 .
a a ,..., a
x x x i i i
If we denote then if and only if and:
x a te t x a
i i i i
f f a ,...,a ,x ,a ,...,a f a ,...,a j j 1 i 1 i i 1 p j 1 p
.
a lim x a i i
x x a i i i Therefore, the partial derivative of the function in
f a is the derivative j
of the partial function according to , namely the derivative of a
x i
single variable function. Consequently from the partially derivability of the function
f in a results its partial continuity in a. The partially
derivability, when
p
1 , doesn’t imply the continuity, as in the following example.
Example 2. The function: 2 2 xy
, x y
2 2
2 f : IR IR, f x,y x y
, x y
is discontinuous in (0,0) (see example 26, chapter 5) but is partial derivable, because:
f x, f ,
f , lim x x x
and 2 2 2 2 2
y x y
2 x y y y x
f x,y , for x,y , .
x 2 2 2 2 2 2 x y x y
Then 2 2 2 2
y y x
if x y
f 2 2 2
,
x,y
x y
x
if
x y
and analogously 2 2 2 2
x x y
,x y
f 2 2 2
x,y
x y
y
,x y .
2 1
2 Therefore ) and consequently ). f C (IR f C (IR
Example 3. Let
2 ð , ñ, ñ . f:A , ,
2
IR f cos ,ñ sin Certainly . Because ñ, and
f C A f cos , sin ñ
ñ, ñ it results that . Therefore
f sin ,ñ cos f , f C A 1 ñ . f C A p q
and . If
Definition 3. Let f : A IR IR a Å f is partially derivable according to on sphere (i.e. x i S a,r A ,
and f x , x S a,r f is partially derivable according to x j in a x x i i we say
and f is partially derivable two times according to x i x j , hence
exists: 1 f
q lim a te f a .
IR
j
t
t x i
We denote this limit (if exists) with 2
f
or
f a a x x i j
x x i j
and we call it the partial derivative of second order (derivative II) of 2
f
the function and in and
f according x i x ja. If i j then a
2
x x i j
f
are called mixed derivatives according to and
a x i x j in a.
x x j i
If and the partial
i=j we call the derivative II according to x i x j
derivative of second order (or derivative II) according to x in a and
iwe denote 2
f , or . 2 a f x i 2 a
x i
Analogously we define the derivatives of higher order. If
A=Å,
f is continuous and admits partial derivatives of any order h , n
ncontinuous on on n A we say f is of class C A and we write .
f C A n
If for any on
f C A nIN we say f is of class C A and we write . f C A
The partial derivative of order (if exists) is
n according to x i
denoted n
f n
, or n , n f x i
x i
and the partial derivative of order
m according to x (if exists) for the j
n
function we denote:
f n x i n m
f
n m or . n m f n m x x i j
x x i j
Remark. For the single variable function derivable n times it
is known, from high school, the derivation formula of the product (Leibniz Formula): n
n k n k k
,
u v C u v n k
(0) where, from derivative of order 0, i.e. u , of the function u (or v) we
understand function
u (respectively v). As the partial derivatives are,
actually, the derivatives of the partial functions, therefore derivatives of the single variable functions, results that the Leibniz formula is valid for products of scalar functions of several variables as well. 13
f for the function
Example 4. Let us compute 7 6
x y x y .
f x,y x y e 7 Calculate first f 7 7 x using Leibniz’ formula: 7
k
f k x y 7 k k
x,y C x y 7 k e 7 x 7 x
x 7 x y k 6 x y x y .
C x y e C e e x y 7 7 7
Then: 13 6 7 6
6 k
f f k k x y 6 k
x,y x,y C e x y
7 k 7 6 6 7 y 6 y
x y y x k x y 1 x y x y .
C e x y 6 7 C e 6 1 e x y 1 Example 5. Let us calculate the mixed derivatives of second
order for the function: 3 2 2
x y ,x y
2 2
2
. f : IR IR ,f x,y x y
,x y
For we have:
x,y 0, 2 2 2 4 4 2 3
3 x y x y 2 x y x y 3 x y
f x,y x 2 2 2 2 2 2 x y x y and / 4 2 2 2 2 2 2 2 4 2 3
x
9 x y x y 4 y x y x y 3 x y
f x,y f x,y xy x y 2 2 4
x y 6 4 2 2 4 x
6 x y 3 x y ;
2 2 3
x y
analogously: ' x x y 3 2 2 3 2 5 3 2 2 x y x x y
f x,y y 2 2 2 2 2 2 x y x y
and 4 2 2 2 2 2 2 2 5 3 2 5 x
3 x y x y 4 x x y x x y
.
f x,y yx 2
2
4 f x,y xy
x y
For :
x,y 0, f x, f ,
f , lim x x x
and
f ,y f , x x
,
f , xy lim y y f , y f , f , lim y y y
and
f x , f , y x
.
f , yx lim lim x x
1
x x
Then: 2 2 f f .
x y y x
We remark that, if we pass to polar coordinates , then: x , y ,
x cos , y sin 2
and ; hence , therefore
f x , y lim f x , y f , x , y
2
); moreover:
f C ( IR '
2
, hence ) and ( IR
f x , y x x 4 f C
2
, hence f C ).
f x , y y y 2 ( IR
But for mixed derivatives 2 4 1 6 m 3 m lim f x , y lim f x , mx lim f x , y , xy xy yx 3 x x x 2 y mx y mx
1 m hence doesn’t exist limit in origin, namely (0,0) is a point of 2 2 discontinuity of the mixed derivatives; consequently .
f C (
IR )
2 As following, we will show that the functions of class have C
the mixed derivatives equal. For simplifying the notations we will
show this property for and in general hypothesis.
p
2 q
1
2
. If
Theorem 1. (Schwarz) Let f : A IR IR and
a, b Å
there exists an open neighbourhood
V V , there exists the partial (a,b)
derivatives for any , and
f x , y , f x , y , f x , y x , y x y xy xy '' V f is
continuous in , then there exists and
a, b f a , b yx . f a , b f a , b xy yx
- so that and
Proof. Let
h, k IR a , h b k
V
(1)
F h , k f a h , b k f a h , b f a , b k f a , b
If we denote g x , y f x , y k f x , y , then:
.
F h , k g a h , b g a , b
With respect to the definition of function g it results that: (2)
F h , k hk f a h , b k xy
But
f is continuous in (a,b), hence the function xy (3)
h , k f a h , b k f a , b xy xy
has, in the origin, the limit:
. (4) h , k lim h , k
Let . (5)
h lim h , k k
Certainly, from (4) and (5) results that:
. (6)
lim h h
1 Multiplying the equality (1) with , replacing F , h k from (2) and
k
form (3) we obtain:
f a h , b k xy
1
h f a , b h h , k f a h , b k f a h , b xy k
1 (7)
f a , b k f a , b k
Because on
V there exists the partial derivative f , passing to the y
limit when in (7) and considering (5) we have:
k
,
h f a , b h h f a h , b f a , b xy y y
or:
1 . (8)
f a h , b f a , h f a , b h
y y xy h
Passing at the limit in (8) for , from (6) it results that
h f a , b yx exists, and . f a , b f a , b yx xy p 2
be an open set. If , then the
Corollary. Let A IR f C A
mixed derivatives of second order are equal on 2 A (their number being C for p2). p
2. DIRECTIONAL DERIVATIVES
We have noticed that the partial derivative of a function w.r.t. the variable assumes the existence of the limit (*) (definition
x i
2). In this context the vector
p y
tends to when .
a te a IR t i
2 (a,b)
For example, in , for the
IR
A
derivative in w.r.t the variable
a, b
x, a t , b tends (when t ) to x along a parallel line with Ox.
a, b
Let there be . Then the ( u , v ) , point of coordinates
a , tu b tv (a,b)
tends to the point of coordinates (when ) along the line
a, b t (u,v) x
parallel with the vector . We
u i v j
will generalize the notion of partial derivative replacing the vector
e i
from (*) with an arbitrary vector, introducing the concept of
(directional) derivative along a vector, concept that plays an
important role in the study of electromagnetic phenomena. p qa vectorial
Definition 4. Let there be f : A IR IR p
function, \{0} a vector. If there exists
a and A v IR
1 q
I R (**) lim f a tv f a
t t t I R
we say that
f is derivable at a along the vector v. The limit (**), if
f
exists, is denoted and is called the directional derivative of
a
v function f along the vector (with respect to (w.r.t.)) v at the point a.
The directional derivative at a of the function
f along the versor of
1 the vector ) is also called the (directional)
v (i.e. with respect to v
v
derivative of f in the direction v.
f f then .
Remark 1. If v e B a a i p
e x i i
or 1 , we
Remark 2. For p=q=1, denoting a tv x t x a
vremark that (**) becomes:
f f x f a
,
a lim v f a x a
v x a
so, in this particular case, the derivability along a vector implies the derivability.
- Conversely, if , taking we
f is derivable in a, and v IR vt x a
have:
f x f a
1
1 1 f
,
f a lim lim f a tv f a a
x a t
x a t v v v t I R so, derivability implies derivability along any non-zero vector.
Example 6. The derivative along any non-zero vector v of a
linear application is exactly the respective application calculated in
p q p p
), and \{0} then
v. Truly if fL(IR , IR a IR v IR
f
, for any ,
f a tv f a tf v t IR. From (**) results a f v
v p for any . a IR p IR .
Definition 5. Let there be x = (x 1 ,…, x p ), y = (y 1 ,…, y p )
The real number p
x , y x y x y ... x y k k 1
1 p p
k 1is named the scalar product of the vectors x and y.
Remark. The scalar product has the following properties: 2
p
(P1) and
x , x x , x
IR x , x x
p
(P2) < IR
x,y>=<y,x>, x,y p
(P3)
I R
x , y x , y , , x , y
IR
p
(P4)
x y , z x , z y , z , x , y , z
IR
p
(P5) x , y x y , x , y (the Cauchy-Schwarz
IR inequality).
p Theorem 2 (Riesz). Let fL(IR , IR). Then there exists a p
unique such that:
a IR p
(a) , for any
f x a x , x IR
(b) , where A a A is the matrix of the linear application f.
p
. Then:
Proof. Let x x ,..., x
IR p p 1 p
,
f x f x e x f e x , a i i i i
i 1 i 1
p p
where . If exists such
a f e ,..., f e
IR b b ,..., b
IR
1 p 1 pp
that: , for any , then , for any
f x b x , x IR e , b e , a i i
, so hence
i
1 , p b a , i i i 1 , p b and the point (a) is proved. a
Of course A =a, so (b) is, also, proved. We will give a set of sufficient conditions for derivability w.r.t. a vector and the formula of the derivative w.r.t. a vector.
Theorem 3 (Sufficient conditions for derivability). p q
Let and If exists a neighbourhood f : A IR IR a Å.
V
such that
V f has partial derivatives on V, continuous in a, then f
ap
is derivable w.r.t. any vector \{0} and: p v IR f f
a a v , i
v x i 1 i where
v = (v , … , v ).
1 p
- such that . Then:
Proof. Let t IR a tv
V
f a tv f a f a tv , a tv ,..., a tv f a , a tv ,..., a tv
1 1 2 2 p p 1 2 2 p p
f a , a tv , a tv ,..., a tv f a , a , a tv ,..., a tv ...
1 2 2 3 3 p p 1 2 3 3 p p
(1)
f a , a ,..., a , a tv f a , a ,..., a , a 1 2 p 1 p p 1 2 p 1 p
Knowing that f has partial derivatives on V, applying the theorem of Lagrange on the intervals , for the
a , a tv , k
1 , p p differences
k k k
from (1) it results the existence of the elements , such that:
c t a , a tv , k k k k k 1 , p
f
f a tv f a tv c t , a tv ,..., a tv 1
1
2 2 p p
x 1
f f
tv a tv , c t , a tv ,..., a tv ... tv a , a , a ,..., a , c t (2)
2 1 1 2 3 3 p p p 1 2 3 p 1 p x 2 x 2 From the hypothesis of the theorem we know that the partial
derivatives are continuous in
a; multiplying both members from (2)
1 with and passing to limit with , knowing that
t t
, it follows that:
c t a , k k k 1 , p f
1 f f a lim f a tv f a a v ... a v ,
1 p t v t x x 1 p so
f is derivable w.r.t. the vector v, and the formula of the directional derivative is proved.
p
Definition 6. Let f : A IR IR be a scalar function, partial
derivable in
aÅ. The vector: p
f f f
a ,..., a a e k
x x x 1 p k 1 k
is named the gradient of function , or
f at a and is denoted grad f a
f a (to be read “nabla”, or del applied to the function f in the point
a”). If A=Å and f is partial derivable on A, the vectorial function: p
f p
grad f f e , f : A
IR , f (a ) k a→
k 1 x k is called the gradient of function f, or, the nabla operator applied to f.
f f f f f If , , and if , .
p
2 f i j p 3 f i j k x y x y z
p
Corollary 1. If f : A IR IR verifies the hypothesis of
f theorem 3, then .
a f ( a ), v
v
p
1and then
Corollary 2. If A = Å
IR f C ( A ) f is derivable
f w.r.t. any non-zero vector . v from A and f , v
v
2 2 x z x z
, . Let us
Example 7. Let f : IR f x , y xe , ye
→ IR calculate the derivative of
f in (0,0) w.r.t. a versor v=(v 1 ,v 2 ) which
forms an angle of 30 with the axis Ox.
3
1 Then and .
v cos 1 30 v sin 2
30
2
2
3
1 Therefore . Moreover:
v ,
2
2 x y x y x y x y
,
f x , y
1 x e , ye f x , y xe , 1 y e x y and, according to theorem 3:
f f f
3
1 , , v , v 1 , , 1 v .
1 2
v x y
2
2 Remark. The derivability of a function in a point w.r.t. any non-zero vector implies the continuity of it along any line which passes through the respective point. Nevertheless it does not imply the continuity for .
p
1
2 Example 8. Let f : IR
2 → IR,
, y x .
f x , y y
, y 2