← all posts

Picard-Lindelöf Theorem

2025-04-07·7 min read·

We discuss the Picard-Lindelöf theorem on local existence and uniqueness of solutions to first-order ODEs, building up from the completeness of C(K,E) under the supremum norm, the Banach fixed point theorem, and Grönwall's inequality.

Approach

Lemma

Let KK be a compact metric space and (E,)(E,\|\cdot\|) be a Banach space. Then the space C(K,E)C(K,E) equipped with a supremum norm f=supxKf(x)\|f\|_{\infty} = \sup_{x \in K}\|f(x)\| is a Banach space.

Proof.

Let ε>0\varepsilon > 0 be arbitrary and (fn)C(K,E)(f_n) \subseteq C(K,E) be a Cauchy sequence, then there exists NNN \in \mathbb{N} such that fnfm<ε\|f_n - f_m\|_{\infty} < \varepsilon whenever n,m>Nn,m > N. Fix xKx \in K, then we have

fn(x)fm(x)Efnfm<ε.\|f_n(x) - f_m(x)\|_E \leq \|f_n - f_m\|_{\infty} < \varepsilon.

Then (fn(x))(f_n(x)) is Cauchy in EE. Since EE is complete, there exists f(x)Ef(x) \in E such that fn(x)f(x)f_n(x) \to f(x). We claim that fnff_n \to f uniformly. Fix n>Nn > N and xKx \in K, then we have

fn(x)f(x)E=limm+fn(x)fm(x)limm+fnfmε.\|f_n(x) - f(x)\|_{E} = \lim_{m \to +\infty}\|f_n(x) - f_m(x)\| \leq \lim_{m \to +\infty}\|f_n - f_m\|_{\infty} \leq \varepsilon.

Then fnff_n \to f in EE. In addition, taking mm \to \infty we obtain fnfε\|f_n - f\|_\infty \leq \varepsilon for all nNn \geq N. Thus fnff_n \to f in the sup norm. To prove fC(K,E)f \in C(K,E), fix x0Kx_0 \in K, we choose NNN \in \mathbb{N} large enough such that

ffn<ε3for all n>N.\|f - f_n\|_{\infty} < \frac{\varepsilon}{3} \quad \text{for all } n > N.

Since fnf_n is pointwise continuous, then there exists δ>0\delta > 0 such that fn(x)fn(x0)<ε3\|f_n(x) - f_n(x_0)\| < \frac{\varepsilon}{3} whenever d(x,x0)<δd(x,x_0) < \delta. Then we obtain

f(x)f(x0)f(x)fn(x)+fn(x)fn(x0)+fn(x0)f(x0)2ffn+fn(x)fn(x0)<ε.\begin{aligned} \|f(x) - f(x_0)\| &\leq \|f(x) - f_n(x)\| + \|f_n(x) - f_n(x_0)\| + \|f_n(x_0) - f(x_0)\| \\ &\leq 2\|f - f_n\|_\infty + \|f_n(x) - f_n(x_0)\| \\ &< \varepsilon. \end{aligned}

Since x0Kx_0 \in K was arbitrary, then fC(K,E)f \in C(K,E), which implies that C(K,E)C(K,E) with the supremum norm is a Banach space.


Banach Fixed Point Theorem

Theorem (Banach Fixed Point Theorem)

Let (X,d)(X,d) be a complete metric space, q[0,1)q \in [0,1) and T:XXT: X \to X be a mapping on XX satisfying

d(T(x),T(y))qd(x,y)for all x,y.d(T(x),T(y)) \leq q\,d(x,y) \quad \text{for all } x,y.

Then TT admits a unique fixed point xx^* in XX.

Proof.

Let x0Xx_0 \in X be arbitrary and let (xn)(x_n) be the sequence defined by xn=T(xn1)x_n = T(x_{n-1}) for all n1n \geq 1. We first establish the estimate

d(xn+1,xn)qnd(x1,x0),for all nN.d(x_{n+1}, x_n) \leq q^n d(x_1, x_0), \quad \text{for all } n \in \mathbb{N}.

This follows by induction: the base case n=1n = 1 is immediate. Assuming it holds for nn, then

d(xn+2,xn+1)=d(T(xn+1),T(xn))qd(xn+1,xn)qn+1d(x1,x0).d(x_{n+2}, x_{n+1}) = d(T(x_{n+1}), T(x_n)) \leq q\,d(x_{n+1}, x_n) \leq q^{n+1}d(x_1,x_0).

Using this and the triangle inequality, for all m>nm > n we obtain

d(xm,xn)k=nm1d(xk+1,xk)k=nm1qkd(x1,x0)qn1qd(x1,x0).d(x_m, x_n) \leq \sum_{k=n}^{m-1} d(x_{k+1}, x_k) \leq \sum_{k=n}^{m-1} q^k d(x_1,x_0) \leq \frac{q^n}{1-q}d(x_1,x_0).

Let ε>0\varepsilon > 0 be arbitrary, then one can choose NNN \in \mathbb{N} large enough such that qn1qd(x1,x0)<ε\frac{q^n}{1-q}d(x_1,x_0) < \varepsilon. Thus (xn)(x_n) is a Cauchy sequence, and since XX is complete, then xnx_n converges to some xXx^* \in X and we obtain

x=limn+xn=limn+T(xn1)=T(x),x^* = \lim_{n\to+\infty} x_n = \lim_{n\to+\infty} T(x_{n-1}) = T(x^*),

where the last equality follows from continuity of TT. Thus xx^* is a fixed point in XX. To prove the uniqueness, suppose p1p2p_1 \neq p_2 are two fixed points in XX, then we have

d(p1,p2)=d(T(p1),T(p2))qd(p1,p2)<d(p1,p2),d(p_1,p_2) = d(T(p_1),T(p_2)) \leq q\,d(p_1,p_2) < d(p_1,p_2),

which is a contradiction, as desired.


Grönwall Inequality

Theorem (Grönwall Inequality)

Let u:[t0,T][0,)u: [t_0,T] \to [0,\infty) be a continuous function, C,L0C, L \geq 0 and we have

u(t)C+Lt0tu(s)dsfor all t[t0,T].u(t) \leq C + L\int_{t_0}^t u(s)\,ds \quad \text{for all } t \in [t_0,T].

Then we have the estimate

u(t)CeL(tt0)for all t[t0,T].u(t) \leq Ce^{L(t-t_0)} \quad \text{for all } t \in [t_0,T].
Proof.

Let v(t)=C+Lt0tu(s)dsv(t) = C + L\int_{t_0}^t u(s)\,ds, then we have u(t)v(t)u(t) \leq v(t) and v(t)=Lu(t)Lv(t)v'(t) = Lu(t) \leq Lv(t). Consider w(t)=eL(tt0)v(t)w(t) = e^{-L(t-t_0)}v(t), differentiating ww yields

w(t)=eL(tt0)(v(t)Lv(t))0.w'(t) = e^{-L(t-t_0)}(v'(t) - Lv(t)) \leq 0.

Thus w(t)w(t0)=v(t0)=Cw(t) \leq w(t_0) = v(t_0) = C, we obtain

u(t)v(t)CeL(tt0).u(t) \leq v(t) \leq Ce^{L(t-t_0)}.

Hence we are done.


Picard–Lindelöf Theorem

Theorem (Picard–Lindelöf)

Consider the Cauchy problem

{x(t0)=x0,x(t)=F(t,x(t)).(1)\begin{cases} x(t_0) = x_0, \\ x'(t) = F(t, x(t)). \end{cases} \tag{1}

Let UR×RnU \subseteq \mathbb{R} \times \mathbb{R}^n be a closed set and F:URnF: U \to \mathbb{R}^n be a continuous function in tt, locally Lipschitz in xx, i.e. for every compact set KUK \subseteq U, there exists a local constant L>0L > 0 such that

F(t,x)F(t,y)L(t,x)(t,y)\|F(t,x) - F(t,y)\| \leq L\|(t,x) - (t,y)\|

for all (t,x),(t,y)K(t,x),(t,y) \in K. Then there exists an interval II containing t0t_0 and a unique solution xC1(I,Rn)x \in C^1(I, \mathbb{R}^n).

Proof.

Let a,rRa, r \in \mathbb{R} such that r>0r > 0 and W=[t0a,t0+a]×B(x0,r)UW = [t_0 - a, t_0 + a] \times \overline{B(x_0,r)} \subseteq U. Since WW is closed and bounded, it is compact. Then FF is bounded on WW.

Let M=sup(t,x)WF(t,x)M = \sup_{(t,x) \in W}\|F(t,x)\|, b=rb = r, h=min{bM,12L,a}h = \min\left\{\frac{b}{M}, \frac{1}{2L}, a\right\} and I=[t0h,t0+h]I = [t_0 - h, t_0 + h]. We define the metric space (X,)(X, \|\cdot\|_{\infty})

X={xC(I,Rn)xx0b}.X = \{x \in C(I, \mathbb{R}^n) \mid \|x - x_0\|_\infty \leq b\}.

Since II is compact and Rn\mathbb{R}^n is complete, the above lemma gives that C(I,Rn)C(I,\mathbb{R}^n) is Banach. Since XC(I,Rn)X \subseteq C(I,\mathbb{R}^n) is closed, then XX is also a Banach space. Integrating both sides of the second equation of (1)(1) yields

x(t)=x0+t0tF(s,x(s))ds.x(t) = x_0 + \int_{t_0}^t F(s, x(s))\,ds.

Let T:XC(I,Rn)T: X \to C(I, \mathbb{R}^n) be the operator defined by T[x](t)=x0+t0tF(s,x(s))dsT[x](t) = x_0 + \int_{t_0}^t F(s, x(s))\,ds. We claim that TT is invariant. Indeed, we have

T[x]x0=suptIt0tF(s,x(s))dst0t0+hF(s,x(s))dsMhb.\|T[x] - x_0\|_\infty = \sup_{t \in I}\left\|\int_{t_0}^t F(s,x(s))\,ds\right\| \leq \int_{t_0}^{t_0+h}\|F(s,x(s))\|\,ds \leq Mh \leq b.

Thus T[x]XT[x] \in X. Furthermore, we aim to prove that TT is a contraction. Since WW is compact, then there exists L>0L > 0 such that FF is Lipschitz with constant LL on WW. Consider the following estimate

(Tx)(t)(Ty)(t)=t0tF(s,x(s))F(s,y(s))dst0tF(s,x(s))F(s,y(s))dsLt0tx(s)y(s)dsLhxy12xy.\begin{aligned} \|(Tx)(t) - (Ty)(t)\| &= \left\|\int_{t_0}^t F(s,x(s)) - F(s,y(s))\,ds\right\| \\ &\leq \int_{t_0}^t \|F(s,x(s)) - F(s,y(s))\|\,ds \\ &\leq L\int_{t_0}^t \|x(s) - y(s)\|\,ds \\ &\leq Lh\|x - y\|_\infty \\ &\leq \frac{1}{2}\|x - y\|_\infty. \end{aligned}

Taking the supremum over tIt \in I yields TxTy12xy\|Tx - Ty\|_\infty \leq \frac{1}{2}\|x - y\|_\infty. Thus TT is a contraction. Applying the Banach Fixed Point theorem, then there exists a unique xXx^* \in X such that T[x]=xT[x^*] = x^*, or equivalently,

x(t)=x0+t0tF(s,x(s))ds.x^*(t) = x_0 + \int_{t_0}^t F(s, x^*(s))\,ds.

Then F(s,x(s))F(s, x^*(s)) is continuous, the fundamental theorem of calculus implies that

ddtx(t)=F(t,x(t)).\frac{d}{dt}x^*(t) = F(t, x^*(t)).

Hence xx^* is a solution in C1(I,Rn)C^1(I,\mathbb{R}^n) of (1)(1). To prove the uniqueness, suppose x1,x2C1(I,Rn)x_1, x_2 \in C^1(I,\mathbb{R}^n) are solutions of (1)(1). Since both satisfy the integral equation, using the similar estimate we obtain

x1(t)x2(t)=(Tx1)(t)(Tx2)(t)Lt0tx1(s)x2(s)ds.\|x_1(t) - x_2(t)\| = \|(Tx_1)(t) - (Tx_2)(t)\| \leq L\int_{t_0}^t \|x_1(s) - x_2(s)\|\,ds.

If t<t0t < t_0, the integral t0t=tt0\int_{t_0}^t = -\int_t^{t_0} is nonpositive while the left side is nonnegative, so both sides must vanish, which implies x1(t)x2(t)=0\|x_1(t) - x_2(t)\| = 0, i.e. x1(t)=x2(t)x_1(t) = x_2(t). If tt0t \geq t_0, apply the Grönwall inequality for the function u(t)=x1(t)x2(t)u(t) = \|x_1(t) - x_2(t)\| with C=0C = 0, we obtain u(t)0u(t) \leq 0 for all tt0t \geq t_0, thus u(t)=0u(t) = 0. Therefore x1(t)=x2(t)x_1(t) = x_2(t) for all tIt \in I, it follows that the solution xx^* in C1(I,Rn)C^1(I,\mathbb{R}^n) is unique.


References

[1] H. Brezis, Functional Analysis, Sobolev Spaces and Partial Differential Equations, Springer (2011).

[2] G. Teschl, Ordinary Differential Equations and Dynamical Systems, Graduate Studies in Mathematics, Vol. 140, American Mathematical Society (2012).

2025-04-07

comments

loading...

leave a note