Theorem 1 p263 (Gagliardo-Nirenberg-Sobolev inequality)

Assume 1 ≤ p < n. There exists a constant C, depending only on p and n,

such that
||u|| n−p
np
(Rn ) ≤ C||Du||Lp (Rn )
L

for all u ∈C01 (Rn ).

Remark: From the proof, we my choose C(n, p) = p(n−1)
n−p . But this may not
be the best constant.
Proof:
1. First, we prove the case p = 1. We want to show that

||u|| n ≤ C||Du||L1 (Rn )

L n−1 (Rn )

for all u ∈ C01 (Rn ). Since u has compact support, we have for each 1 ≤ i ≤ n
and x ∈ Rn
Z xi
u(x) = uxi (x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )dyi
−∞

and
Z xi
|u(x)| ≤ |uxi (x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi
Z−∞
xi
≤ |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi
(1)
Z−∞
∞
≤ |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi .
−∞
| {z }
independent of xi

Consequently,
1
Z ∞ 1
|u(x)| n−1 ≤( |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi ) n−1
−∞

and
n Z ∞
n Y 1
|u(x)| n−1 ≤ ( |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi ) n−1
i=1 | −∞ {z }
independent of xi

1
 2. To illustrate the main ideas, we discuss the case when n = 3. So we
have
Z ∞ Z ∞ Z ∞
3 1 1 1
|u(x)| 2 ≤ ( |Du(y1 , x2 , x3 )|dy1 ) 2 ( |Du(x1 , y2 , x3 )|dy2 ) 2 ( |Du(x1 , x2 , y3 )|dy3 ) 2
−∞ −∞ −∞

and

Z ∞ 3
|u(x)| 2 dx1
−∞
Z ∞ Z ∞ Z ∞ Z ∞
1 1 1
≤ ( |Du(y1 , x2 , x3 )|dy1 ) (
2 |Du(x1 , y2 , x3 )|dy2 ) (
2 |Du(x1 , x2 , y3 )|dy3 ) 2 dx1
−∞
| −∞ {z } −∞ −∞
independent of x1
Z ∞ 1
Z ∞Z ∞
1
Z ∞
1
=( |Du(y1 , x2 , x3 )|dy1 ) 2 ( |Du(x1 , y2 , x3 )|dy2 ) 2 ( |Du(x1 , x2 , y3 )|dy3 ) 2 dx1
−∞ −∞
| −∞ {z−∞ }
R R 1 R 1
apply holder inequality w.r.t. x1 |f g|dx1 ≤( f 2 dx1 ) 2 ( g 2 dx1 ) 2
Z ∞ 1
≤( |Du(y1 , x2 , x3 )|dy1 ) 2
−∞
Z ∞ Z ∞ Z ∞ Z ∞
1 1
( ( |Du(x1 , y2 , x3 )|dy2 )dx1 ) (
2 ( |Du(x1 , x2 , y3 )|dy3 )dx1 ) 2
| −∞ −∞ {z } −∞ −∞
independent of x2
(2)

Z ∞ Z ∞ 3
|u(x)| 2 dx1 dx2
Z−∞ −∞
∞ Z ∞ 1
≤( ( |Du(x1 , y2 , x3 )|dy2 )dx1 ) 2
Z −∞∞ Z ∞
−∞
Z ∞ Z ∞
1 1
( |Du(y1 , x2 , x3 )|dy1 ) 2 ( ( |Du(x1 , x2 , y3 )|dy3 )dx1 ) 2 dx2
−∞
| −∞ −∞ −∞
{z }
apply holder inequality w.r.t. x2
Z ∞
Z ∞
1
≤( ( |Du(x1 , y2 , x3 )|dy2 )dx1 ) 2
Z−∞
∞ Z ∞
−∞
Z ∞ Z ∞ Z ∞
1 1
( ( |Du(y1 , x2 , x3 )|dy1 )dx2 ) 2 ( ( ( |Du(x1 , x2 , y3 )|dy3 )dx1 )dx2 ) 2
−∞ −∞ −∞ −∞ −∞
1
Z ∞ Z ∞ Z ∞ Z ∞
1 1
= ||Du||L1 (R3 ) (
2
( |Du(x1 , y2 , x3 )|dy2 )dx1 ) (
2 ( |Du(y1 , x2 , x3 )|dy1 )dx2 ) 2
−∞ −∞ −∞ −∞
(3)

2
Integrating w.r.t x3 , we have
Z ∞Z ∞Z ∞
3
|u(x)| 2 dx1 dx2 dx3
−∞ −∞ −∞
1
Z ∞ Z ∞ Z ∞ Z ∞ Z ∞
1 1
≤ ||Du||L2 1 (R3 ) ( ( |Du(x1 , y2 , x3 )|dy2 )dx1 ) 2 ( ( |Du(y1 , x2 , x3 )|dy1 )dx2 ) 2 dx3
1
Z−∞ −∞ −∞
∞ Z ∞ Z ∞
−∞ −∞
1
≤ ||Du||L2 1 (R3 ) ( ( ( |Du(x1 , y2 , x3 )|dy2 )dx1 )dx3 ) 2
Z ∞ Z ∞ Z −∞ ∞
−∞ −∞
1
( ( ( |Du(y1 , x2 , x3 )|dy1 )dx2 )dx3 ) 2
−∞ −∞ −∞
3
= ||Du||L1 (R3 ) .
2

(4)

Thus we have 3 3
||u|| 2 3 ≤ ||Du||L2 1 (R3 ) .
L 2 (R3 )

This implies
||u|| 3 ≤ ||Du||L1 (R3 ) .
L 2 (R3 )

3. For the general case, we start with

n Z ∞
n Y 1
|u(x)| n−1 ≤ ( |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi ) n−1 .
i=1 | −∞ {z }
independent of xi
Z ∞ n
|u(x)| n−1 dx1
−∞
Z ∞ Z ∞ n Z ∞
1 Y 1
≤ ( |Du(y1 , x2 , · · · xn )|dy1 ) n−1 ( |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi ) n−1 dx1
−∞
| −∞ {z } i=2 | −∞ {z }
independent of x1 independent of xi
Z ∞ Z ∞ n Z ∞
1 Y 1
=( |Du(y1 , x2 , · · · xn )|dy1 ) n−1 ( |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi ) n−1 dx1
−∞ −∞ −∞
|i=2 {z }
R n−1 1
|f2 ···fn |dx1 ≤ n
R Q
apply holder inequality w.r.t. x1 i=2 ( fi dx1 ) n−1
Z ∞ n Z ∞ Z ∞
1 Y 1
≤( |Du(y1 , x2 , · · · xn )|dy1 ) n−1 ( |Du(x1 , · · · , xi−1 , yi , xi+1 , · · · , xn )|dyi dx1 ) n−1
−∞ i=2 −∞ −∞
(5)

3
Repeating this process, we get
Z Z
n n
|u(x)| n−1 dx ≤ ( |Du|dx) n−1 .
Rn Rn

Thus n n
||u|| n−1n ≤ ||Du||Ln−1
1 (Rn ) .
L n−1 (Rn )

This implies
k|u|| n ≤ ||Du||L1 (Rn ) . (6)
L n−1 (Rn )
γ
4. Now we consider the case 1 < p < n. Let v := |u|γ = (u2 ) 2 where
γ γ
γ > 1 is to be determined. Then vxi = γ2 (u2 ) 2 −1 (2uuxi ) = γ(u2 ) 2 −1 (uuxi )
and |Dv| = γ|u|γ−1 |Du|. We apply estimate 6 to v := |u|γ to get
Z Z
n n−1
( |v(x)| n−1 dx) n ≤ ( |Dv|dx).
Rn Rn

Thus we have
Z
γn n−1
( |u(x)| n−1 dx) n
n
ZR
≤γ |u|γ−1 |Du|dx (7)
R n
Z Z
p p−1 1
(γ−1) p−1
≤ γ( |u| dx) p ( |Du|p dx) p
Rn Rn

γn p 1 n(p−1) p(n−1)
Choose γ so that n−1 = (γ − 1) p−1 , i.e γ = 1− p(n−1) and γ = n−p >1
p(n−1)
γn p np
n−p
n
(since 1 < p < n). Now we have n−1 = (γ −1) p−1 = n−1 = n−p . Hence 7
np n−1 np p−1 R 1
can be rewritten as ( Rn |u(x)| n−p dx) n ≤ p(n−1) p
R R
n−p ( Rn |u| Rn |Du| dx) .
n−p dx) p ( p

np n−p 1
p−1 n−p p(n−1) R
Note that n−1 p
R
n − p = np . Thus we have ( Rn |u(x)|
n−p dx) np ≤
n−p ( Rn |Du| dx)
p

which is
p(n − 1)
||u|| n−p
np
(Rn ) ≤ ||Du||Lp (Rn ) .
L n−p