Peano existence theorem

Peano first published the theorem in 1886 with an incorrect proof.[1] In 1890 he published a new correct proof using successive approximations.[2]

Let ${\displaystyle D}$ be an open subset of ${\displaystyle \mathbb {R} \times \mathbb {R} }$ with ${\displaystyle f\colon D\to \mathbb {R} }$ a continuous function and ${\displaystyle y'(x)=f\left(x,y(x)\right)}$ a continuous, explicit first-order differential equation defined on D, then every initial value problem ${\displaystyle y\left(x_{0}\right)=y_{0}}$ for f with ${\displaystyle (x_{0},y_{0})\in D}$ has a local solution ${\displaystyle z\colon I\to \mathbb {R} }$ where ${\displaystyle I}$ is a neighbourhood of ${\displaystyle x_{0}}$ in ${\displaystyle \mathbb {R} }$, such that ${\displaystyle z'(x)=f\left(x,z(x)\right)}$ for all ${\displaystyle x\in I}$.[3]

The solution need not be unique: one and the same initial value ${\displaystyle (x_{0},y_{0})}$ may give rise to many different solutions ${\displaystyle z}$.

By replacing ${\displaystyle y}$ with ${\displaystyle y-y_{0}}$, ${\displaystyle x}$ with ${\displaystyle x-x_{0}}$, we may assume ${\displaystyle x_{0}=y_{0}=0}$. As ${\displaystyle D}$ is open there is a rectangle ${\displaystyle R=[-x_{1},x_{1}]\times [-y_{1},y_{1}]\subset D}$.

Because ${\displaystyle R}$ is compact and ${\displaystyle f}$ is continuous, we have ${\displaystyle \textstyle \sup _{R}|f|\leq C<\infty }$ and by the Stone–Weierstrass theorem there exists a sequence of Lipschitz functions ${\displaystyle f_{k}:R\to \mathbb {R} }$ converging uniformly to ${\displaystyle f}$ in ${\displaystyle R}$. Without loss of generality, we assume ${\displaystyle \textstyle \sup _{R}|f_{k}|\leq 2C}$ for all ${\displaystyle k}$.

We define Picard iterations ${\displaystyle y_{k,n}:I=[-x_{2},x_{2}]\to \mathbb {R} }$ as follows, where ${\displaystyle x_{2}=\min\{x_{1},y_{1}/(2C)\}}$. ${\displaystyle y_{k,0}(x)\equiv 0}$, and ${\displaystyle \textstyle y_{k,n+1}(x)=\int _{0}^{x}f_{k}(x',y_{k,n}(x'))\,\mathrm {d} x'}$. They are well-defined by induction: as

${\displaystyle {\begin{aligned}|y_{k,n+1}(x)|&\leq \textstyle \left|\int _{0}^{x}|f_{k}(x',y_{k,n}(x'))|\,\mathrm {d} x'\right|\\&\leq \textstyle |x|\sup _{R}|f_{k}|\\&\leq x_{2}\cdot 2C\leq y_{1},\end{aligned}}}$

${\displaystyle (x',y_{k,n+1}(x'))}$ is within the domain of ${\displaystyle f_{k}}$.

We have

${\displaystyle {\begin{aligned}|y_{k,n+1}(x)-y_{k,n}(x)|&\leq \textstyle \left|\int _{0}^{x}|f_{k}(x',y_{k,n}(x'))-f_{k}(x',y_{k,n-1}(x'))|\,\mathrm {d} x'\right|\\&\leq \textstyle L_{k}\left|\int _{0}^{x}|y_{k,n}(x')-y_{k,n-1}(x')|\,\mathrm {d} x'\right|,\end{aligned}}}$

where ${\displaystyle L_{k}}$ is the Lipschitz constant of ${\displaystyle f_{k}}$. Thus for maximal difference ${\displaystyle \textstyle M_{k,n}(x)=\sup _{x'\in [0,x]}|y_{k,n+1}(x')-y_{k,n}(x')|}$, we have a bound ${\displaystyle \textstyle M_{k,n}(x)\leq L_{k}\left|\int _{0}^{x}M_{k,n-1}(x')\,\mathrm {d} x'\right|}$, and

${\displaystyle {\begin{aligned}M_{k,0}(x)&\leq \textstyle \left|\int _{0}^{x}|f_{k}(x',0)|\,\mathrm {d} x'\right|\\&\leq |x|\textstyle \sup _{R}|f_{k}|\leq 2C|x|.\end{aligned}}}$

By induction, this implies the bound ${\displaystyle M_{k,n}(x)\leq 2CL_{k}^{n}|x|^{n+1}/(n+1)!}$ which tends to zero as ${\displaystyle n\to \infty }$ for all ${\displaystyle x\in I}$.

The functions ${\displaystyle y_{k,n}}$ are equicontinuous as for ${\displaystyle -x_{2}\leq x<x'\leq x_{2}}$ we have

${\displaystyle {\begin{aligned}|y_{k,n+1}(x')-y_{k,n+1}(x)|&\leq \textstyle \int _{x}^{x'}|f_{k}(x'',y_{k,n}(x''))|\,\mathrm {d} x''\\&\textstyle \leq |x'-x|\sup _{R}|f_{k}|\leq 2C|x'-x|,\end{aligned}}}$

so by the Arzelà–Ascoli theorem they are relatively compact. In particular, for each ${\displaystyle k}$ there is a subsequence ${\displaystyle (y_{k,\varphi _{k}(n)})_{n\in \mathbb {N} }}$ converging uniformly to a continuous function ${\displaystyle y_{k}:I\to \mathbb {R} }$. Taking limit ${\displaystyle n\to \infty }$ in

${\displaystyle {\begin{aligned}\textstyle \left|y_{k,\varphi _{k}(n)}(x)-\int _{0}^{x}f_{k}(x',y_{k,\varphi _{k}(n)}(x'))\,\mathrm {d} x'\right|&=|y_{k,\varphi _{k}(n)}(x)-y_{k,\varphi _{k}(n)+1}(x)|\\&\leq M_{k,\varphi _{k}(n)}(x_{2})\end{aligned}}}$

we conclude that ${\displaystyle \textstyle y_{k}(x)=\int _{0}^{x}f_{k}(x',y_{k}(x'))\,\mathrm {d} x'}$. The functions ${\displaystyle y_{k}}$ are in the closure of a relatively compact set, so they are themselves relatively compact. Thus there is a subsequence ${\displaystyle y_{\psi (k)}}$ converging uniformly to a continuous function ${\displaystyle z:I\to \mathbb {R} }$. Taking limit ${\displaystyle k\to \infty }$ in ${\displaystyle \textstyle y_{\psi (k)}(x)=\int _{0}^{x}f_{\psi (k)}(x',y_{\psi (k)}(x'))\,\mathrm {d} x'}$ we conclude that ${\displaystyle \textstyle z(x)=\int _{0}^{x}f(x',z(x'))\,\mathrm {d} x'}$, using the fact that ${\displaystyle f_{\psi (k)}}$ are equicontinuous by the Arzelà–Ascoli theorem. By the fundamental theorem of calculus, ${\displaystyle z'(x)=f(x,z(x))}$ in ${\displaystyle I}$.

The Peano theorem can be compared with another existence result in the same context, the Picard–Lindelöf theorem. The Picard–Lindelöf theorem both assumes more and concludes more. It requires Lipschitz continuity, while the Peano theorem requires only continuity; but it proves both existence and uniqueness where the Peano theorem proves only the existence of solutions. To illustrate, consider the ordinary differential equation

${\displaystyle y'=\left\vert y\right\vert ^{\frac {1}{2}}}$ on the domain ${\displaystyle \left[0,1\right].}$

According to the Peano theorem, this equation has solutions, but the Picard–Lindelöf theorem does not apply since the right hand side is not Lipschitz continuous in any neighbourhood containing 0. Thus we can conclude existence but not uniqueness. It turns out that this ordinary differential equation has two kinds of solutions when starting at ${\displaystyle y(0)=0}$, either ${\displaystyle y(x)=0}$ or ${\displaystyle y(x)=x^{2}/4}$. The transition between ${\displaystyle y=0}$ and ${\displaystyle y=(x-C)^{2}/4}$ can happen at any ${\displaystyle C}$.

The Carathéodory existence theorem is a generalization of the Peano existence theorem with weaker conditions than continuity.

• Osgood, W. F. (1898). "Beweis der Existenz einer Lösung der Differentialgleichung dy/dx = f(x, y) ohne Hinzunahme der Cauchy-Lipschitzchen Bedingung". Monatshefte für Mathematik. 9: 331–345. doi:10.1007/BF01707876. S2CID 122312261.
• Coddington, Earl A.; Levinson, Norman (1955). Theory of Ordinary Differential Equations. New York: McGraw-Hill.
• Murray, Francis J.; Miller, Kenneth S. (1976) [1954]. Existence Theorems for Ordinary Differential Equations (Reprint ed.). New York: Krieger.
• Teschl, Gerald (2012). Ordinary Differential Equations and Dynamical Systems. Providence: American Mathematical Society. ISBN 978-0-8218-8328-0.