On The Differentiability of Multivariable Functions - Pradeep Kumar Pandey
On The Differentiability of Multivariable Functions - Pradeep Kumar Pandey
On The Differentiability of Multivariable Functions - Pradeep Kumar Pandey
Abstract : Objective of this article is to review the differentiability of functions of several real variables.
Keywords: Differentiability, Linear Transformation, Multivariable Function.
I.
INTRODUCTION
Calculus is unarguably one of the most fascinating and useful subject of the modern science and
engineering curriculum. Calculus of functions of several variables, also called multivariable calculus. The
to n-space
In the study of calculus the differentiability of functions is of prime importance since derivatives appear almost
everywhere in the field of science and engineering. Applications of derivative in many physical problems e.g.
approximation of functions about a point, error analysis, obtaining extremum, population growth etc. are of
great practical interest.
While teaching calculus to undergraduate students I observed that students grasp the idea of
differentiability in one variable case quite easily, nevertheless they were not at the same ease with the
differentiability concept of multivariable functions. In this article an attempt has been made to exposit
differentiability of multivariable functions in an elegant manner to address pedagogical problems. A reader
having knowledge of basic calculus and linear algebra will find this article fairly accessible.
II.
DIFFERENTIABILITY OF SINGLE VARIABLE FUNCTIONS
The idea of the notion of the derivative originated from a problem in geometry the problem of finding
the tangent at a point of a curve. Though derivative was originally formulated to study the problem of tangents,
sooner it was observed that it also provides a way to calculate velocity and, more generally the rate of change of
a function [1].
To start with the basic definition of derivative first we wish to fix our notation. Most of the texts of
dy
dy
as another notation for f ( x) where y f ( x) , but the fact that
and f ( x) are
dx
dx
dy
not interchangeable is evident when you consider that one does not write
for f (4) . The obvious
d4
calculus describe
objection for this is that it does not make any sense to differentiate with respect to a constant i.e. we cannot have
a rate of change with respect to something that is not changing at all. Though, if there is no risk of confusion one
can use the notations
dy
and f ( x) freely and interchangeably [2].
dx
Now we recall the definition of differentiability of a real valued function of a real variable [3].
Definition: Let A be a subset of R containing a neighborhood of the point a , that is a is an interior point of
A. The derivative of a function f : A R at a , defined by
f (a) lim
h 0
f ( a h) f ( a )
h
provided the limit exists. In this case, we say that f is differentiable at point a .
The following facts are immediate consequence of the above definition:
(i) Differentiable functions are continuous.
At this juncture it is worthwhile to mention that above consequence merely tells us that if a function is
differentiable at a point of the domain then it must be continuous there but fails to provide further information
about the continuity of the derivative and existence of subsequent derivatives. For example the
1
0, x 0 and f (0) 0
x
function f ( x) x sin
2
www.iosrjournals.org
has
81 | Page
1
1
cos , x 0 and f (0) 0 ,
x
x
derivative f ( x) 2 x sin
here
1
lim 2 x sin 0 but
x 0
x
1
lim cos does not have a definite value, so f is not continuous at 0.
x 0
x
One can check that the function g ( x) x x is differentiable for all x in R , with g ( x) x but derivative
g ( x) x is no more differentiable at 0.
Remark: Regarding the existence of subsequent derivatives in a neighborhood of the domain, behaviour of
complex valued functions is pretty stronger than real ones. In case of the complex-valued function of a complex
variable, existence of first derivative in a neighborhood of the domain guarantees not only the continuity of
function but the existence of derivatives of all orders.
(ii) Composites of differentiable functions are differentiable. (Chain rule)
D R n containing a
neighborhood of point a (i.e. a (a1 , . . . , an ) is an interior point of D ) and a function f : D R . One
may tempt to define the differentiability of a multivariable function just by replacing a and h in definition of
n
the derivative of single variable function by points of R but this does not make any sense, as division of a real
n
number by a point in R has not been defined if n 1 .
III.
x12 x2
, ( x1 , x2 ) 0 with f (0) 0 .
x14 x22
It can be easily checked that all directional derivatives of f exist at 0 (0,0) .
2
Let 0 u (u1 , u2 ) R . A straightforward calculation yields
Example: Let
f : R 2 R , defined by f ( x1 , x2 )
u12
if u2 0
u
2
Du f (0)
0 if u2 0
i.e. function has all directional derivatives at 0 . It can be verified that as we approach to origin (e.g. along the
2
curve x2 kx1 , k 0 ) limit of the given function does not exist; hence we infer that function is not
continuous at origin however it has all directional derivatives at origin.
It appears that the directional derivative is an appropriate generalization of the notion of derivative
but this is not true, because differentiability implies continuity and from above example it is clear that function
possesses all directional derivatives at origin but not continuous at origin. Obviously directional derivative is
very useful definition but it restricts us to study the change of f in one direction at a time and in particular, if
we choose
the direction of
82 | Page
f
(a) . For instance in above example taking {e1 (1,0), e2 (0,1)} as standard basis of R 2 gives
xi
D1 f (0) 0 and D2 f (0) 0 , respectively the first partial derivatives of f at origin, namely
f
f
(0) and
(0) .
x1
x2
Now we seek something stronger to define the concept of differentiability of multivariable functions.
For this purpose we revisit the differentiability definition of one variable case and take a closer look
(geometrically). It can be observed that if a function is differentiable at an interior point of the domain then at
that point there is a (local) tangent line associated with it i.e. differentiable functions are locally linear, or in
other words differentiable functions enjoy reasonably good local approximation by linear functions. Intuitively
this means, if a function f is differentiable at a point a then after sufficiently zooming-in around the point,
graph of f locally looks like part of the straight line that is at the point of differentiability there does not exist
any sharp corner (in addition it may be noted that at the point of differentiability local tangent line must not be
vertical). For illustration if we observe the graph of function f ( x) x (Fig. 1), at the origin no matter up to
which magnitude we zoom-in around the point 0, locally it never looks like part of the straight line and a sharp
corner at point 0 always remains there (inferring that function is not locally linear at the origin), hence function
f ( x) x is not differentiable at 0.
lim
h 0
f (a h) f (a) L(h)
h
h is a point of R n and f (a h) f (a) L(h) a point of R , so the norm (in 1-dimension mod) signs
are necessary. The linear transformation L , which is unique, is called the derivative of f at a .
Since
www.iosrjournals.org
83 | Page
f (a h, b k ) f (a, b) L(h, k )
lim
( h , k ) (0,0)
f : R 2 R is differentiable at (a, b) D R 2 , if
0 ,
(h, k )
Analogous to differentiability of real valued functions of several real variables the differentiability of functions
lim
h 0
f (a h) f (a) L(h)
h
1 2 matrix, say 1 2 .
R m , ( x1 , . . . , xn ) F( F1, . . ., Fm ) , is an m n
x j , 1 j n , given by
DF
F
x
i
mn
www.iosrjournals.org
84 | Page
f : R 2 R be defined by
1
f ( x1 , x2 ) x12 x22 sin
2
x x2
2
1
, ( x1 , x2 ) 0 and f ( x1 , x2 ) 0 if ( x1 , x 2 ) 0 .
1
h 2 sin 0
f
f (0 h, 0) f (0, 0)
h
We have,
(0, 0) lim
lim
0 1 and
h 0
h 0
x1
h
h
f
Similarly
(0,0) 0 2
x2
( h, k) R 2 , then
f (0 h, 0 k ) f (0, 0) L(h, k )
f (h, k ) f (0, 0) (0 h 0 k )
lim
lim
( h , k )(0,0)
( h , k ) (0,0)
(h, k )
h2 k 2
Let (0,0)
lim
( h , k ) (0,0)
lim
( h , k ) (0,0
Since function
lim
( h , k ) (0,0
1
(h 2 k 2 )sin
2
2
h k
h2 k 2
1
h2 k 2 sin
2
2
h k
1
sine is bounded by both -1 and 1, therefore sin
1 , hence above limit becomes
2
2
h k
We have already seen that both first partial derivatives of the function at the origin are 0. Now other than the
origin
x1
f
1
1
( x1 , x2 ) 2 x1 sin
cos
2
2
x2 x2
x2 x2
x1
x
x
2
1
2
2
1
1
Consider above partial derivative along x1 -axis (that is along x2 0) , then
1 x
1
f
( x1 , 0) 2 x1 sin 1 cos
x x
x
x1
1
1
1
1
1
x
0 but the second term 1 cos oscillates between -1
Now if x1 0 , the first term 2 x1 sin
x
x
x1
1
1
x1
1 or 1 depending upon the sign of x1 , and cosine function is bounded by -1
and 1 (since for x1 0,
x1
and 1), hence this limit does not exist thereby showing that
f
(0, 0) exists and is equal to 0).
x1
www.iosrjournals.org
f
is not continuous at the origin (though
x1
85 | Page
f
f
not continuous at the origin (though
(0, 0) exists and is equal to 0).
x2
x2
IV.
CONCLUSION
The concept of differentiability of functions starting from single real variable function to several real
variables functions has been reviewed. It is hoped that article will benefit the readers to develop a better
understanding of the differentiability concept in higher dimensions.
Acknowledgements
Author thanks Dr. Shri Ram, librarian JUIT Waknaghat, Solan for arranging a copy of beautiful book
Calculus on Manifolds by Prof. Spivak.
REFERENCES
[1]
Tom M. Apostol, Calculus volume 1: one-variable calculus, with an introduction to linear algebra (USA: John Wiley & Sons
Inc., 1967).
[2]
[3]
[4]
James R. Munkres, Analysis on manifolds (Redwood City, CA 94065: Addison-Wesley Publishing Company, 1991).
Sudhir R. Ghorpade and Balmohan V. Limaye, A course in multivariable calculus and analysis (New Delhi: Springer India
Private Limited, 2012).
Michael Spivak, Calculus on manifolds: a modern approach to classical theorems of advanced calculus (USA: Addison-Wesley
Publishing Company, 1995).
[5]
dy
dx
www.iosrjournals.org
86 | Page