ADE (G1156) Spring 2006 Handout 3: Lipschitz condition and Lipschitz continuity7

3 Lipschitz condition and Lipschitz continuity

3.1 Denition (Lipschitz-continuous function) Lipschitz-continuity comes in three dierent
avours. Let f : R
m
R
m
.
(a) Given an open set B R
m
, we say that f is Lipschitz-continuous on the open subset B if
there exists a constant R
+
0
(called the Lipschitz constant of f on B) such that
f(x) f(y) x y , x, y B. (15)
(b) The function f is called locally Lipschitz-continuous, if for each z R
n
there exists an
L > 0 such that f is Lipschitz-continuous on the open ball of center z and radius L
B
L
(z) := {y R
m
: y z < L} . (16)
(c) If f is Lipschitz continuous on all of the space R
m
(i.e. B = R
m
in (15)), then f is called
globally Lipschitz-continuous.
3.2 Remark (local vs. global) Notice the fundamental dierence between the local and global
versions of the Lipschitz-continuity. Whereas in the local version the Lipschitz constant and
the open set B depend on each point z R
m
, in the global version the constant is xed and
B = R
m
. In particular, a globally Lipschitz-continuous function is locally Lipschitz-continuous,
but the viceversa is not true.
3.3 Remark (norms and Lipschitz constants) In (15) the norm can be any norm. However,
once a norm has been chosen one should stick to that single norm, as the Lipschitz constant
depends on the particular choice of this norm. Unless otherwise stated, we use the Euclidean
norm in all our analysis.
3.4 Interpretation
To see what these denitions mean, let us consider the situation in one dimension. Suppose f
is a Lipschitz function on a neighborhood B of x R. This implies that, y B,
|f(x) f(y)| |x y|

f(x) f(y)
x y

f(x +h) f(x)

h

by putting y = x +h.
If we were to let h 0 and if the function f were dierentiable, then the result would mean
|f

(x)| ; that is that the derivative is bounded by the Lipschitz constant.

However, there is nothing in the dention of Lipschitz-continuity that implies that f is dif-
ferentiable. So in general we cant proceed to this limit, since we dont know if f is dierentiable
at x. But this tells us all we need to know: being Lipschitz just means f cant be too steep, the
bound on the dierence quotient being .
ADE (G1156) Spring 2006 Handout 3: Lipschitz condition and Lipschitz continuity8
3.5 Examples and Counterexamples
The functions below are pictured in gure 1; we will examine them in turn, with B = [1, 1].
3.6 Example Consider f
1
(x) = x
2
. We will show that f
1
is locally Lipschitz-continuous but
not globally so.
This function is continuously dierentiable. Pick up any point x R, we observe that
sup
y(x1,x+1)

1
(y)

= sup
y(x1,x+1)
|2y| 2 |x| + 1 (17)
being |y| |x| + 1 for y (x 1, x + 1) by the triangle inequality. Now picking up two points,
y, z (x 1, x +1) it follows by the mean value theorem that for some between y and z, that
|f
1
(z) f
1
(y)| =

1
()(z y)

sup
(x1,x+1)

1
()

|z y| .
(18)
Using (17), we conclude that
|f
1
(z) f
1
(y)| (2 |x| + 1) |z y| , z, y (x 1, x + 1). (19)
Hence the Lipschitz constant of f
1
on (x 1, x + 1) is = 2 |x| + 1 and the function is locally
Lipschitz-continuous on all of R.
Observe that the Lipschitz constant depends on x and its neighbourhood. In particular, if
x then . This is an indication that the function may not be globally Lipschitz
continuous. Indeed, for any y = 0
|f
1
(y) f
1
(0)|
|y 0|
= |y| , as y , (20)
which means that there is no that can satisfy the global Lipschitz property for y R.
3.7 Example f
2
(x) =
_
x
2
sin(1/x
2
) if x = 0,
0 if x = 0.
This function is dierentiable everywhere:
f

2
(x) =
_
2xsin
_
1
x
2
_

2
x
cos
_
1
x
2
_
if x = 0,
0 if x = 0.
(One has to use the denition of derivative to obtain f

2
(0).)
However, since f

2
(x) 0 as x 0, the derivative is not continuous. Is f
2
Lipschitz-
continuous?
Dene x
n
= (2n + /2)
1/2
and y
n
= (2n)
1/2
for n = 1, 2, . . . , and suppose that f
2
is
Lipschitz-continuous. Since x
n
, y
n
B for all n, there must exist such that

f
2
(x
n
) f
2
(y
n
)
x
n
y
n

for all n
=
(2n +

2
)
1
(2n)
1/2
(2n +

2
)
1/2
= 4n
_
(2n)
1/2
+
_
2n +

2
_
1/2
_
as n .
ADE (G1156) Spring 2006 Handout 3: Lipschitz condition and Lipschitz continuity9
Plainly no such can exist. Hence f
2
is not Lipschitz-continuous.
In the rst example we used the theorem continuously dierentiable =locally Lipschitz.
Here we saw an example where a dierentiable function whose derivative is not continuous turned
out not to be Lipschitz. However, this is not true of all such functions. For example, consider
the function (shown in the handout 2.2 Continuity and Dierentiabilitysome facts of life)
f(x) =
_
x
2
sin(1/x) if x = 0,
0 if x = 0,
which is dierentiable, but f

is not continuous. It is left as an exercise to show that f is globally

Lipschitz continuous.
3.8 Example This example shows that dierntiability is a stronger concept than Lipschitz
continuity. f
3
(x) = |x| This function is not dierentiable at x = 0, but
f

3
(x) =
_
1 if x < 0,
1 if x > 0.
Since
|f
3
(x) f
3
(y)| = ||x| |y|| |x y|, for all x, y R,
we see that f
3
is globally Lipschitz continuous, with Lipschitz constant = 1.
3.9 Example f
4
(x) = |x|
1/2
This function is not dierentiable at x = 0, but
f

4
(x) =
_

1
2
(x)
1/2
if x < 0,
1
2
x
1/2
if x > 0.
Is f
4
Lipschitz-continuous?
Dene x
n
= 1/n
2
and y
n
= 0 for n = 1, 2, . . . , and suppose that f
4
is Lipschitz-continuous.
Since x
n
, y
n
B for all n, there must exist such that

f
4
(x
n
) f
4
(y
n
)
x
n
y
n

for all n
= n as n .
Plainly no such can exist. Thus f
4
is not Lipschitz-continuous.
3.10 Relationships between Lipschitz continuity, continuity and dierentia-
bility
3.11 Theorem (Lipschitz C
0
) Every locally Lipschitz-continuous function is continuous.
The proof of this Theorem is left as an exercise.
3.12 Theorem (C
1
Lipschitz) Every continuously dierentiable function is locally Lipschitz.
ADE (G1156) Spring 2006 Handout 3: Lipschitz condition and Lipschitz continuity 10
Proof Let f : R
m
R
m
be continuously dierentiable. Fix any two points z, y R
m
and
dene the function f : [0, 1] R as f() := f(z + (y z)). It is clear that
f(0) = f(z) and f(1) = f(y)). (21)
Furthermore, by the chain rule, we know that f is dierentiable and that
d
d
f() = Df(z + (y z)) (y z). (22)
Here Df(w) is the Jacobian matrix of f at w:
Df(w) =
_
_
_
f
1
w
1
(w)
f
1
w
m
(w)
.
.
.
.
.
.
.
.
.
f
m
w
1
(w)
f
m
w
m
(w)
_
_
_
(23)
where
f
i
w
j
(w) is the partial derivative of the i-th component of f with respect to the j-th
coordinate.
By the denition of f and then by the fundamental theorem of calculus, we have
f(y) f(z)) = f(1) f(0) =
_
1
0
f

() d
=
__
1
0
Df
_
z + (y z)
_
d
_
(y z);
where the integral in the last line is a matrix whose i, j-th component is given by
_
1
0
f
i
w
j
_
z + (y z)
_
d.
It follows that
f(y) f(z)
_
1
0
Df
_
z + (y z)
_
d y z,
where we use the notation for both the vector norm and its associated matrix norm.
Notice that, by the triangle inequality for integrals, we have

_
1
0
Df
_
z + (v z)
_
d
_
1
0
Df
_
z + (v z)
_
d
sup
[0,1]
Df(z + (v z))
_
1
0
d
= sup
[0,1]
Df(z + (v z)).
ADE (G1156) Spring 2006 Handout 3: Lipschitz condition and Lipschitz continuity 11
Furthermore by the equivalence of matrix norms we have
c
0
> 0 : A c
0
A

, A R
mm
, (24)
where c
0
depends only on m and

is the maximum row sum norm from last handout.

Thus, to establish Lipschitz continuity, we x an arbitraty point x R
m
and we establish a
bound for
_
1
0
Df(z + (y z)) d in an appropriate neighbourhood B of x.
Let B = B
L
(x), with L arbitrary. Since f is continuously dierentiable on B, which is
closed and bounded, there exists
0
such that
sup
wB
_
_
_
_
f
i
w
j
(w)
_
_
_
_

0
, ij [1 : m] .
Here we have applied the Weierstrass theorem which says that each continous function (
w
j
f
i
)
is bounded on a closed and bounded set (B).
Now, given z, y B, it follows that z +(z y) B for all [0, 1] because B is a ball, so
sup
[0,1]
_
_
Df
_
z + (y z)
__
_
c
0
m
0
=: .
It follows that
f(y) f(z) y z , y, z B;
which is to say that f is Lipschitz-continuous on B
L
(x). Since x is arbitrary this means that f
is locally Lipschitz-continuous on R
m
.
3.13 Remark Notice that in the proof above
0
, and therefore , might depend on the point
x and the constant L. However, if we can nd a bound that is independent of x and L(i.e., a
uniform bound), then it means that the function f is actually globally Lipschitz-continuous.
3.14 Corollary If Df(w) is uniformly bounded as w , then f is globally Lipschitz.
3.15 Lipschitz-continuity with respect to some arguments
As you may have noticed in the course ntoes we used a slightly dierent version of Lipschitz
continuity: the Lipschitz continuity with respect to the rst argument of a function
f : R
m
R R. (25)
We give now an adaptation of our earlier denition. The main point is that the only variable
that counts is the rst one. It is useful though to have the function continuous with respect to
both variables.
3.16 Denition A function f such as in (25) is locally Lipschitz continuous with respect to its
rst argument if is continuous and if for each (x, t) R
m
R there exists a L > 0 and a > 0
such that
f(y, s) f(z, s) z y (26)
ADE (G1156) Spring 2006 Handout 3: Lipschitz condition and Lipschitz continuity 12
for all z, y B
L
(x) and s (t L, t + L). (Notice how the right-hand side is independent of
the second argument s.)
The function f is called globally Lipschitz-continuous with respect to its rst argument if it
is continuous and if there is a > 0 such that (26) is satised for all z, y in R
m
and s R.
(Notice again the secondary role played by the time variable.)
A fact that we have used in the proof of uniqueness is the following:
3.17 Lemma (Characterization of local Lipschitz-continuity) A continuous function f, such as
in (25), is locally Lipschitz continuous with respect to its rst argument if and only if for each
closed and bounded subset K of R
m
R, there exists an open set A R
m
such that K A
and a constant
K
> 0 such that
f(x, t) f(y, t)
K
x y , (x, t), (y, t) A. (27)
Proof (For the bold and the knowledgeable.) The if part is easy and is left as an ex-
ercise. The only if part can be proved by contradiction. Suppose that the conclusion is
false. This means is, that there exists a closed and bounded subset K
0
R
m
R such
that any n Z
+
there exist x
n
, y
n
R
m
, t
n
R, such that (x
n
, t
n
), (y
n
, t
n
) A
n
, with
A
n
= {(z, s) R
m
R : (y, t) K
0
: z y + |s t| 1/n} and for which
f(x
n
, t
n
) f(y
n
, t
n
) > nx
n
y
n
. (28)
Notice that A
n
A
1
which is closed and bounded. It follows that the sequences ((x
n
, t
n
))
nZ
+
and ((y
n
, t
n
)
nZ
+ are bounded. There exists thus a family of integers X Z
+
and a point
(x

, t

) A (the closure of A) such that

lim
n
(x
n
, t
n
) (x

, t

), n X. (29)
Likewise, there exists (y

, t

) A and Y X such that

lim
n
(y
n
, t
n
) (y

, t

), n Y. (30)
Notice that (x

, t

), (y

, t

nY
A
n
= K
0
. We show next that x

= y

. Indeed, by the
continuity of f, it has to be bounded by a constant M over the closed and bounded set K
0
.
So (28) implies that x
n
y
n
2M/n, for n Y , which means, by passing to the limit, as
n , that
x

= 0, i.e., x

= y

. (31)
To conclude the proof it is sucient to observe that the function f fails to be Lipschitz-
continuous in any neighbourhood of (x

, t

). Indeed for any > 0 and for any open ball

B (x

, t

), for n Y big enough, we have that (x

n
, t
n
) (y
n
, t
n
) B, and
f(x
n
, t
n
) f(y
n
, t
n
) > nx
n
y
n
x
n
y
n
, (32)
in contrast with the local Lipschitz-continuity with respect to the rst argument.
ADE (G1156) Spring 2006 Handout 3: Lipschitz condition and Lipschitz continuity 13
1 0 1
1
0
1
x
f
1
(
x
)
1 0 1
1
0
1
x
f
2
(
x
)
1 0 1
1
0
1
x
f
3
(
x
)
1 0 1
1
0
1
x
f
4
(
x
)
Figure 1: Example functions