Limits and Order
For functions of a real variable, the derivative is defined as
f¢(x)=
lim
Dx® 0
f(x+Dx) - f(x)
Dx
,
which means that the difference
f(x+Dx) - f(x)
Dx
- f¢(x)
is small if Dx is small and not 0 (for which the
quotient is not obviously defined).
Multiplying the remainder by Dx, we obtain that
f(x+Dx) - f(x)-f¢(x) Dx= small·Dx,
with the right hand side (RHS) of the equation is "much smaller than
Dx" in the precise sense
lim
Dx® 0
RHS
|Dx|
=0.
Another formal advantage is that the equation is also defined and
true for Dx
= 0.
Definition. An expression (function) f(x) is little o of
x as x ® 0, written f(x) = o(x) [as x ® 0], if
lim
x ® 0
f(x)
|x|
= 0.
If we are not worried about the particular details of f(x),
we write f(x) = o(x) [as x ® 0].
With this convention, the
definition of differentiability and the derivative takes the
convenient form
f(x+Dx) = f(x) +f¢(x)·Dx+ o(Dx).
In a similar way, if limx ® 0 y(x) = 0, we write
y(x) = o(1) with the precise meaning that
lim
x® 0
y(x)
1
= 0.
Definition. Let q(x) be nonzero for x near and not equal 0. Then a function f(x) is little o of q(x), written f(x)=o(q(x)), if
lim
x ® 0
f(x)
|q(x)|
= 0.
Then a function f(x) is big O of q(x), written f(x)=O(q(x)), if
f(x)
|q(x)|
is bounded as x ® 0.
N.B. We are assuming that, for x small
enough and ¹ 0, both f(x) and q(x) are defined and
q(x) ¹ 0.
We could also propose an e-d definition of
little o(q):
Definition. A function f = o(q) as x ® 0 means For every e > 0, there is a d > 0 such that if 0 < |x| < d, then |f(x)| < e|q(x)|
With this convention, continuity of a function f(x) can be
expressed by
f(x+Dx) = f(x) + o(1)
as Dx® 0.
A real valued function of a real variable x is differentiable
at x with derivative f¢(x) if
f(x+Dx) = f(x) +f¢(x) Dx+ o(Dx)
as Dx® 0.
Local boundedness of a function can be expressed as
f(x+Dx)=O(1)
as Dx® 0.
There is a formal calculus for handling sums and products for
functions which are little o or big O of one (or
several) q. Verify that O(1) ·o(Dx) = o(Dx); i.e., the product of a bounded function and a
function which is o(Dx) is o(Dx).
Similarly o(Dx) ±o(Dx) = o(Dx).
Proof of the Chain RuleThe Chain Rule. Let g(z) be differentiable at z, and let
f(w) be differentiable at w = g(z). Then h(z) = f(g(z)) is differentiable at
z and
h¢(z) =
d
dz
f(g(z)) = f¢(g(z))·g¢(z).
Proof: We show that
h(z + Dz)
= f(g(z + Dz))
= f(g(z)) + f¢(g(z)) ·g¢(z)·Dz+ o(Dz).
as Dz® 0.
Let
Dg(z) = g(z + Dz) - g(z) = g¢(z) Dz+ o(Dz).
We are assuming that
g(z+Dz)
= g(z) +g¢(z) ·Dz+ o(Dz),
f(g(z)+Dg(z))
= f(g(z)) +f¢(g(z)) ·Dg(z) +o(Dg(z)).
Since
Dg(z)
= g¢(z) Dz+o(Dz)
= O(Dz),
o(Dg(z))
= o(O(Dz))
= o(Dz),
we have
f(g(z + Dz))
= f(g(z)) +f¢(g(z)) ·g¢(z) ·Dz+ o(Dz).
Remarks
The concepts little o and big O are also useful as the
argument x ® ¥. For example we write
x2=o(ex) as x® ¥ with the precise meaning
lim
x®¥
x2
ex
=0.
File translated from
TEX
by
TTH,
version 3.78. On 06 Nov 2007, 21:32.