Mathematics Department Stanford University

Math 51H – Contraction mapping theorem and ODEs

The contraction mapping theorem concerns maps f : X → X, (X, d) a metric space, and their fixed
points. A point x is a fixed point of f if f (x) = x, i.e. f fixes x. A contraction mapping is a map
f : X → X such that there is θ ∈ (0, 1) such that d(f (x), f (y)) ≤ θd(x, y) for all x, y ∈ X.

Theorem 1 Suppose X is a complete metric space, and f : X → X is a contraction mapping. Then

f has a unique fixed point x.

Remark 1 Note that if (Z, d) is a complete metric space and X is a closed subset, then X is complete
(with the relative metric). Indeed, if {xn }∞
n=1 is a Cauchy sequence in X, then it is such in Z, thus it
converges to some x ∈ Z. But then x is a limit point of X, and X is closed, so x ∈ X, proving the
desired completeness (every Cauchy sequence in X converges to a point in X).

Remark 2 Note that any contraction mapping is (uniformly!) continuous; one can take δ = ε in the
definition of continuity.

Proof of the Theorem: First suppose x, x 0 are fixed points of f . Then f (x) = x, f (x 0) = x 0, so
d(x, x 0) = d(f (x), f (x 0)) ≤ θd(x, x 0),
so (1 − θ)d(x, x 0) ≤ 0, so d(x, x 0) ≤ 0, so d(x, x 0) = 0, so x = x 0, showing uniqueness.
We now turn to existence. To do so, take an arbitrary x0 ∈ X, and define xn inductively: xn+1 = f (xn ),
n ≥ 0. We claim that {xn }∞n=0 is Cauchy. To see this, notice first that for all n ≥ 1,
d(xn+1 , xn ) = d(f (xn ), f (xn−1 )) ≤ θd(xn , xn−1 ),
so by induction,
d(xn+1 , xn ) ≤ θn d(x1 , x0 ).
Thus, for n > m, using the triangle inequality,
d(xn , xm ) ≤ d(xn , xn−1 ) + . . . + d(xm+1 , xm ) ≤ (θn−1 + θn−2 + . . . + θm )d(x1 , x0 ).
Summing the finite geometric series,
1 − θn−m d(x1 , x0 )
d(xn , xm ) ≤ θm (1 + θ + . . . + θn−m−1 )d(x1 , x0 ) = θm d(x1 , x0 ) ≤ θm .
1−θ 1−θ
Since limm→∞ θm = 0, we have that given ε > 0 there exists N such that m ≥ N implies θm <
1−θ
ε 1+d(x 1 ,x0 )
, and thus for n, m ≥ N , d(xn , xm ) < ε, proving the Cauchy claim.
But X is complete, so x = limn→∞ xn exists. Since f is continuous and lim xn = x, sequen-
tial continuity shows that lim f (xn ) = f (x). But f (xn ) = xn+1 , so limn→∞ xn+1 = f (x). Since
limn→∞ xn = limn→∞ xn+1 , we deduce that f (x) = x, so x is a fixed point of f as claimed. 
We now use this in the simplest ODE setting. An ODE (ordinary differential equation) is an equation
of the form
x 0(t) = F (t, x(t)), x(t0 ) = x0 (1)
where F : R × Rn → Rn is a given function, t0 ∈ R, x0 ∈ Rn , and one is looking for a solution x at
least on an interval near t0 , say [t0 − δ, t0 + δ], δ > 0, so x ∈ C 1 ([t0 − δ, t0 + δ]). We impose the minimal
requirement that F be continuous for the following discussion.
Rather than considering the ODE directly, we rewrite it as an integral equation using the fundamental
theorem of calculus: Z t Z t
x(t) = x(t0 ) + x 0 = x0 + F (τ, x(τ )) dτ.
t0 t0
Note that any solution of the original ODE solves the integral equation
Z t
x(t) = x0 + F (τ, x(τ )) dτ, (2)
t0

and conversely if x ∈ C 0 ([t0 − δ, t0 + δ]) merely solving (2), then in fact x is C 1 by the fundamental
theorem of calculus (since F is continuous, being the composite of continuous functions!), and solves
the ODE.
Thus, from now on we consider (2). We consider the equation as a fixed point claim for a map
T : X → X, X = C 0 ([t0 − δ, t0 + δ]). Concretely, let
Z t
T x(t) = x0 + F (τ, x(τ )) dτ ;
t0

so certainly T : X → X. Now, as d(x, y) = sup{kx(t) − y(t)k : t ∈ [t0 − δ, t0 + δ]}, we have

n Z t Z t o
d(T x, T y) = sup (x0 + F (τ, x(τ )) dτ ) − (x0 + F (τ, y(τ )) dτ ) : t ∈ [t0 − δ, t0 + δ]
t0 t0
n Z t o
= sup (F (τ, x(τ )) − F (τ, y(τ ))) dτ ) : t ∈ [t0 − δ, t0 + δ] .
t0

So let us suppose that F is globally Lipschitz in the second variable, i.e. for some δ0 > 0,

kF (t, x) − F (t, y)k ≤ M kx − yk, x, y ∈ Rn , t ∈ [t0 − δ0 , t0 + δ0 ]. (3)

Then, if δ ≤ δ0 , t ≥ t0 (with a similar calculation if t < t0 )

Z t Z t Z t
(F (τ, x(τ )) − F (τ, y(τ ))) dτ ) ≤ k(F (τ, x(τ )) − F (τ, y(τ ))k dτ ≤ M kx(τ ) − y(τ )k dτ
t0 t0 t0
Z t
≤ M sup{kx(s) − y(s)k : s ∈ [t0 − δ, t0 + δ]} dτ (4)
t0
= M (t − t0 ) sup{kx(s) − y(s)k : s ∈ [t0 − δ, t0 + δ]}
≤ M δ sup{kx(s) − y(s)k : s ∈ [t0 − δ, t0 + δ]} = M δd(x, y).

Thus,
d(T x, T y) ≤ M δd(x, y),
1
so it is a contraction mapping provided δ < M.
So take δ < 1/M , δ ≤ δ0 . Then T is a contraction mapping on the complete metric space C 0 ([t0 −
δ, t0 + δ]; Rn ) (continuous maps from [t0 − δ, t0 + δ] to Rn ), with completeness shown in Problem set
9, Problem 7. Thus, it has a unique fixed point x ∈ C 0 ([t0 − δ, t0 + δ]; Rn ), which as explained means
that the ODE has a unique solution on [t0 − δ, t0 + δ]:

Theorem 2 Suppose F is globally Lipschitz on [t0 − δ0 , t0 + δ0 ] × Rn in the sense of (3). Then there
is δ > 0, δ ≤ δ0 such that the ODE (1) has a unique C 1 solution on [t0 − δ, t0 + δ].

This is the simplest local existence and uniqueness theorem for ODE. The only unsatisfactory assump-
tion in it is (3). Notice that the estimate of (3) is a very reasonable assumption locally, namely if it
is replaced by

kF (t, x) − F (t, y)k ≤ M kx − yk, x, y ∈ B R (0), t ∈ [t0 − δ0 , t0 + δ0 ], (5)

where B R (0) is the closed ball of radius R in Rn : B R (0) = {x ∈ Rn : kxk ≤ R}. Indeed, in this
case it follows from F being C 1 , since by the fundamental theorem of calculus, just as in our proof of
Taylor’s theorem,
n
X Z 1 Z 1
F (t, x) − F (t, y) = (xj − yj ) Dj F (t, y + s(x − y)) ds = DF (t, y + s(x − y))(x − y) ds,
j=1 0 0

where Dj , D denote derivatives in the second slot. Thus,

Z 1
kF (t, x) − F (t, y)k ≤ DF (t, y + s(x − y))(x − y) ds
0
Z 1 Z 1
≤ DF (t, y + s(x − y)) kx − yk ds = kx − yk DF (t, y + s(x − y)) ds,
0 0

so if F is C 1 , so Dj F are bounded on the compact set [t0 − δ0 , t0 + δ0 ] × B R (0),

kF (t, x) − F (t, y)k ≤ Ckx − yk, C = sup{kDF (z)k : z ∈ [t0 − δ0 , t0 + δ0 ] × B R (0)}.

The more natural ODE theorem is then:

Theorem 3 Suppose F is locally Lipschitz on [t0 − δ0 , t0 + δ0 ] × B R (0), R > 0, in the sense of (5).
Then there is δ > 0, δ ≤ δ0 such that for x0 ∈ B R/2 (0) the ODE (1) has a unique C 1 solution on
[t0 − δ, t0 + δ] with kx(t)k ≤ R on [t0 − δ, t0 + δ].

Proof: Let C0 = sup{kF (z)k : ∈ [t0 − δ0 , t0 + δ0 ] × B R (0)}. Then for kx0 k ≤ R/2, δ ≤ δ0 , x ∈
C 0 ([t0 − δ, t0 + δ]; Rn ) with kx(t)k ≤ R for t ∈ [t0 − δ, t0 + δ], and for t ≥ t0 (with a similar formula
for t < t0 ),
Z t
k(T x)(t)k ≤ kx0 k + kF (τ, x(τ )k dτ ≤ R/2 + C0 (t − t0 ) ≤ R/2 + C0 δ.
t0

Thus, if δ < R/(2C0 + 1), then

k(T x)(t)k ≤ R, t ∈ [t0 − δ, t0 + δ].

Thus, if we let X = {x ∈ C 0 ([t0 − δ, t0 + δ]; Rn ) : sup kxk ≤ R}, then T : X → X provided

δ < R/(2C0 + 1) (and δ ≤ δ0 ).
Further, for x ∈ X, we have by (5)

kF (t, x(t)) − F (t, y(t))k ≤ M kx(t) − y(t)k, t ∈ [t0 − δ, t0 + δ],

so the calculation of (4) applies. This gives that if in addition δ < M −1 then T is a contraction
mapping, and thus the contraction mapping theorem gives a unique fixed point. This gives a unique
solution to the ODE with the property that kx(t)k ≤ R for t ∈ [t0 − δ, t0 + δ], and proves the theorem.