Some Important Inequalities Math 354, Winter 2008

Triangle Inequality: For all real a and b, |a + b| |a| +|b|.

Inverse Triangle Inequality: For all real a and b, |a b| ||a| |b||.
Proof. By the triangle inequality,
|a| = |a b + b| |a b| +|b| |a b| |a| |b|.
Symmetrically,
|b| = |b a + a| |b a| +|a| |b a| |b| |a|.
But |b a| = |a b|, so |a b| is greater than or equal to both |a| |b| and |b| |a|.
Convexity of e
x
: For all real a and b and all t [0, 1],
e
ta+(1t)b
te
a
+ (1 t)e
b
.
Proof Sketch. Let
f(x) =
e
b
e
a
b a
(x a) + e
a
be the line through (a, e
a
) and (b, e
b
). Notice that ta + (1 t)b is in [a, b], and that
e
x
f(x) on [a, b], since y = e
x
is convex (meaning concave-up). Thus
e
ta+(1t)b
f(ta + (1 t)b) = te
a
+ (1 t)e
b
.
Youngs Inequality: Let p, q > 1 be real such that
1
p
+
1
q
= 1. Then for all real a and b,
ab
1
p
a
p
+
1
q
b
q
.
Proof. Let t =
1
p
, so that 1 t =
1
q
. Then, using the convexity of e
x
,
ab = e
log a+log b
= e
1
p
log a
p
+
1
q
log b
q

1
p
e
log a
p
+
1
q
e
log b
q
=
1
p
a
p
+
1
q
b
q
.
Cauchys Inequality: For all real a and b,
ab
a
2
2
+
b
2
2
.
Proof. This is the special case of Youngs Inequality with p = q = 2. It is also easy
to prove directly: 0 (a b)
2
= a
2
2ab + b
2
2ab a
2
+ b
2
.
1
Some Important Inequalities Math 354, Winter 2008
The following inequalities concern the
p
-norms of vectors a = (a
1
, . . . , a
n
) R
n
. In partic-
ular, Minkowskis inequality is the triangle inequality for the
p
-norm. All sums

are taken
over an index i running from 1 to n.
Discrete Holders Inequality: Let p, q > 1 be real such that
1
p
+
1
q
= 1. Then for all a, b R
n
,

|a
i
b
i
|
_

|a
i
|
p
_
1/p
_

|b
i
|
q
_
1/q
.
Proof. Let A = (

|a
i
|
p
)
1/p
and B = (

|b
i
|
q
)
1/q
. Then, using Youngs inequality
and the triangle inequality,

|a
i
b
i
|
AB
=

a
i
A
b
i
B

1
p
_
a
i
A
_
p
+
1
q
_
b
i
B
_
q

1
p
_
a
i
A
_
p

1
q
_
b
i
B
_
q

.
This simplies to
1
pA
p

|a
i
|
p
+
1
qB
q

|b
i
|
q
=
1
p

|a
i
|
p
A
p
+
1
q

|b
i
|
q
B
q
=
1
p
+
1
q
= 1.
Thus

|a
i
b
i
| = AB, as desired.
Discrete Minkowskis Inequality: Let p > 1 be real. Then for all a, b R
n
,
_

|a
i
+ b
i
|
p
_
1/p

|a
i
|
p
_
1/p
+
_

|b
i
|
p
_
1/p
.
Proof. By the triangle inequality,

|a
i
+ b
i
|
p

(|a
i
| +|b
i
|) |a
i
+ b
i
|
p1
=

|a
i
||a
i
+ b
i
|
p1
+

|b
i
||a
i
+ b
i
|
p1
.
Applying the discrete Holders inequality with q =
p
p1
(so that
1
p
+
1
q
= 1) to the
rst term on the right-hand side rewrites it as
_

|a
i
|
p
_
1/p
_

_
|a
i
+ b
i
|
p1
_
p
p1
_
p1
p
=
_

|a
i
|
p
_
1/p
_

|a
i
+ b
i
|
p
_p1
p
.
Do the same to the other term and combine the results. Then the inequality is

|a
i
+ b
i
|
p

_
_

|a
i
|
p
_
1/p
+
_

|b
i
|
p
_
1/p
_
_

|a
i
+ b
i
|
p
_p1
p
.
Divide through by (

|a
i
+ b
i
|
p
)
p1
p
to obtain the result.
These results also hold for p = 1 and q = , by ad hoc arguments. They also hold if
we replace the vectors a, b with (appropriate) functions f, g and the sums

with integrals
_
d
c
. . . dx. They are then statements about L
p
-norms on vector spaces of functions, which are
studied in courses in functional analysis.
2