OUTLINE FOR THE LECTURES

ON
ADVANCED COMPUTATIONAL TECHNIQUES
SOME USEFUL INEQUALITIES

by

ANTONIO ORLANDO

This handout presents a collection of elementary and fundamental inequalities that must be memo-
rized.

Algebraic inequalities
• For a, b, x ≥ 0, if x2 ≤ ax + b then there holds

x2 ≤ a2 + 2b.
√
(a+ a2 +4b)2
Proof. Let x∗ be the positive root of the equation x2∗ − ax∗ − b = 0, that is, x2∗ = 4 , then
we have
√
2 2 (a + a2 + 4b)2
x ≤ ax + b ≤ ax∗ + b = x∗ =
√ √ 4
(a + a2 + 4b)2 + (a − a2 + 4b)2
≤ = a2 + 2b .
4
The inequality ax + b ≤ ax∗ + b appears clear by depicting the function x2 and ax + b and from the
meaning of the root x∗ .

• Let a, b, c ≥ 0, if a2 ≤ b2 + c2 then there holds

a ≤ b + c.
√ √
Proof. Taking the square root of both sides, we have a ≤ b2 + c2 ≤ b2 + c2 + 2bc = b + c.

• Let α > 0 and β, x, y ≥ 0 then there holds

α 2 2 2 2
αx2 − 2βxy ≥ x − β y .
2 α
Proof. This follows at once by noting that
r r 2
α 2 α 2 α 2
2
αx − 2βxy − x2 + β 2 y 2 = x2 − 2βxy + β 2 y 2 = x− βy ≥ 0.
2 α 2 α 2 α

• Cauchy–Young inequality:
1
∀  > 0 2ab ≤ a2 + b2 , ∀ a, b ∈ R

√ 1
Proof. It follows from 0 ≤ ( a − √

b)2 = −2ab + a2 + 1 b2

1 December 11, 2008

• Young inequality (with ):

b2
∀ > 0, ab ≤ a2 + , ∀ a, b ∈ R
4
Proof. Apply Cauchy–Young inequality with  = 2 and then divide both sides for two.

1 1
• Young inequality: For p > 1, q < ∞, p + q = 1, a, b > 0, it is

ap bq
ab ≤ +
p q
p q p
1 1 1
+ 1q logbq
Proof. Consider ab = e p loga e q logb = e p loga . Since the function ex is convex, then it follows
eλ1 x1 +λ2 x2 ≤ λ1 ex1 + λ2 ex2 , that is,
1 p
+ q1 logbq 1 logap 1 logbq ap bq
ab = e p loga ≤ e + e = + .
p q p q

• Given p ≥ 1, for a, b ≥ 0, there holds

(a + b)p ≤ 2p (ap + bp )

Proof. If 0 ≤ a ≤ b then (a + b)p ≤ (b + b)p = 2p bb ≤ 2p (ap + bp );

If 0 ≤ b ≤ a then (a + b)p ≤ (a + a)p = 2p ab ≤ 2p (ap + bp ) .

• The previous inequality can be sharpened for p = 2. For a, b ≥ 0, there holds

(a + b)2 ≤ 2(a2 + b2 )

Proof. (a + b)2 = a2 + b2 + 2ab ≤ a2 + b2 + a2 + b2 = 2(a2 + b2 ).

• Let α, β > 0, with α + β = 1.

α β
∀ u, v > 0 ⇒ |u {zv } ≤ αu + βv .
| {z }
geometric mean weighted arithmetic mean

Proof. For α ≤ 1, the function f = tα is concave downwards for t > 0. Indeed, its first derivative
is f 0 = αtα−1 whereas its second derivative is f 00 = α(α − 1)tα−2 , thus for α > 0, α − 1 < 0,
hence f 00 < 0 for t > 0. Observe that f 0 (1) = α, thus the tangent at (1, f (1)) has equation
y(t) = f 0 (1)(t − 1) + f (1) = αt − α + 1 = αt + β, since β = 1 − α. For the concavity of the
function it follows tα ≤ αt + β, ∀t > 0, ∀α, β with α > 0, α + β = 1. It follows that for u, v > 0, if
we let t = u/v, we have ( uv )α ≤ α uv + β ⇒ uα v −α v ≤ αu + βv ⇒ uα v β ≤ αu + βv.

• Let a, b ≥ 0 and α , β > 0. Then there holds

αβ
(a + b) ≤ αa + βb.
α+β

Proof. Observe that for α , β > 0

α
≤1
α+β
hence
αβ
≤β
α+β
and
β
≤1
α+β

2 December 11, 2008

hence
αβ
≤ α.
α+β
Therefore it is also
αβ
a ≤ aα
α+β
αβ
b ≤ bβ
α+β
that is, by summing both sides,
αβ
(a + b) ≤ aα + bβ .
α+β
Note that this inequality is sharper than min{α, β}, since

αβ
α≥
α+β αβ
=⇒ min{α, β} ≥ .
αβ α+β
β≥
α+β

• Let ak , bk ≥ 0 for k = 1, 2, . . . , n then there holds

n
X Xn Xn
ak b k ≤ ( a2k )1/2 ( b2k )1/2
k=1 k=1 k=1

Proof. We give the proof only for the case n = 3 since the same argument applies for any n ∈ N.
Since it is
2a1 b1 a2 b2 ≤ a21 b22 + a22 b21
2a1 b1 a3 b3 ≤ a21 b23 + a23 b21
2a2 b2 a3 b3 ≤ a22 b23 + a23 b22

then it is also

2a1 b1 a2 b2 + 2a1 b1 a3 b3 + 2a2 b2 a3 b3 ≤ a21 b22 + a22 b21 + a21 b23 + a23 b21 + a22 b23 + a23 b22 .

Summing a21 b21 + a22 b22 + a23 b23 to both sides, we have

a21 b21 + a22 b22 + a23 b23 + 2a1 b1 a2 b2 + 2a2 b2 a3 b3 + 2a3 b3 a1 b1 ≤

a21 b22 + a22 b21 + a21 b23 + a23 b21 + a22 b23 + a23 b22 + a21 b21 + a22 b22 + a23 b23 =
a21 (b21 + b22 + b23 ) + a22 (b21 + b22 + b23 ) + a23 (b21 + b22 + b23 ) =
(a21 + a22 + a23 )(b21 + b22 + b23 )

that is,
(a1 b1 + a2 b2 + a3 b3 )2 ≤ (a21 + a22 + a23 )(b21 + b22 + b23 ) ,
and by taking the square root on both sides we obtain the statement.
• Let ak , bk ≥ 0 for k = 1, 2, . . . , n and cj , bj ≥ 0 for j = 1, 2, . . . , m, then there holds
   1/2  1/2
n
X m
X n
X m
X n
X m
X
 ak b k + cj dj  ≤  a2k + c2j   b2k + d2j  .
k=1 j=1 k=1 j=1 k=1 j=1

3 December 11, 2008

Integral inequalities
• Poincaré-Friedrich inequality. Let Ω be an open bounded set in Rd . Then there exists a constant
C = C(Ω) > 0 such that
Z
∇s u : ∇s u dx ≥ Ckuk2L2(Ω;Rd ) for all u ∈ H01 (Ω; Rd ) .
Ω

Such inequality holds also for functions v ∈ H01 (Ω; R) in the form

|u|2H 1 ≥ Ckuk2L2 (Ω;R) for all u ∈ H01 (Ω; R) ,

where X
|u|2H 1 = kDα uk2L2 (Ω;R) .
|α|=1

Note that Z
X
kDα uk2L2 (Ω;R) = k∇uk2L2 (Ω;Rd ) = ∇u · ∇u dx .
|α|=1 Ω

• Korn’s first inequality. Let Ω be an open bounded set in Rd with piecewise smooth boundary.
Then there exists a constant C = C(Ω) > 0 such that
Z Z
∇s u : ∇s u dx ≥ C ∇u : ∇u dx
Ω Ω

where
1
∇s u = (∇u + ∇uT )
2
and for A, B ∈ Rd×d
A : B = tr(B T A) .

• Korn’s second inequality. Let Ω be an open bounded set in Rd with piecewise smooth boundary.
Then there exists a constant C = C(Ω) > 0 such that
Z
∇s u : ∇s u dx ≥ Ckuk2H 1 (Ω;Rd )
Ω

where
kuk2H 1 (Ω;Rd ) = kuk2L2(Ω;Rd ) + k∇uk2L2(Ω;Rd×d ) ,
with Z
k∇uk2L2(Ω;Rd×d ) = ∇u : ∇u dx .
Ω

Minimum and maximum of functions

• Let f, g be non negative real valued functions.
If f (t) + g(t) ≤ a then there holds
max{supf, supg} ≤ a .


Proof. f (t) + g(t) ≤ a ⇒ sup f (t) + g(t) ≤ a

Since g ≥ 0, f (t) ≤ f (t) + g(t) ⇒ supf (t) ≤ sup f (t) + g(t)
≤ a
Since f ≥ 0, g(t) ≤f (t) + g(t) ⇒ supg(t) ≤ sup f (t) + g(t)

≤a
thus, it is also max supf (t), supg(t) ≤ sup f (t) + g(t) ≤ a.

4 December 11, 2008

• Let f, g be real valued functions. Then there holds:
1


min(f, g) = f (x) + g(x) − |f (x) − g(x)|
2
1


max(f, g) = f (x) + g(x) + |f (x) − g(x)| .
2
Proof. If f (x) ≤ g(x) then min(f (x), g(x)) = f (x) and max(f (x),
g(x)) = g(x), indeed f (x)−g(x) ≤ 0
thus |f (x)−g(x)| = g(x)−f (x), and 12 f (x)+g(x)−|f (x)−g(x)| = 12
f (x)+g(x)+f (x)−g(x) = f (x)
whereas 21 f (x) + g(x) + |f (x) − g(x)| = 12 f (x) + g(x) − f (x) + g(x) = g(x).

Viceversa, if g(x) − f (x) ≤ 0
it is min(f (x), g(x)) = g(x) and max(f (x), g(x)) = f (x) and it follows
1 1 1
2 f (x) + g(x) − |f (x) − g(x)| = 2 f
(x) + g(x) − f (x) + g(x) = g(x) whereas 2 f (x) + g(x) + |f (x) −
1
g(x)| = 2 f (x) + g(x) + f (x) − g(x) = f (x).

• Let f ≥ 0, then there holds

supf 2 = (supf )2 .

Proof. 0 ≤ f ≤ supf ⇒ f 2 ≤ (supf )2 ⇒ supf 2 ≤ (supf )2

f 2 ≤ supf 2 ⇒ f ≤ (supf 2 )1/2 ⇒ supf ≤ (supf 2 )1/2 ⇒ (supf )2 ≤ supf 2

5 December 11, 2008