ODE existence & uniqueness — Picard-Lindelöf

Picard-Lindelöf Theorem: If

f (t, y)

is continuous in

t

and Lipschitz continuous in

y

near a point

(t_{0}, y_{0})

, then the initial value problem

y^{'} = f (t, y)

y (t_{0}) = y_{0}

has a unique solution on some interval around

t_{0}

. Also called the Cauchy-Lipschitz theorem. It is the foundational guarantee that "nice" differential equations have exactly one solution. Proved via the Banach Fixed-Point Theorem (Picard iteration); due to Émile Picard (1890) and Ernst Lindelöf (1894).

The Picard-Lindelöf theorem answers the two most basic questions one can ask about an ordinary differential equation: does a solution exist, and is it unique? For the initial value problem $y^{'} = f (t, y)$ with $y (t_{0}) = y_{0}$ , the theorem provides sufficient conditions — continuity in $t$ and a Lipschitz condition in $y$ — under which a unique solution exists on a neighborhood of $t_{0}$ . It matters because virtually all of applied mathematics, physics, and engineering models phenomena with ODEs, and without uniqueness, prediction (and the very notion of a deterministic law) collapses. The surprising fact is that mere continuity of $f$ is not enough to guarantee uniqueness — one genuinely needs the stronger Lipschitz condition, as the equation $y^{'} = ∣ y ∣$ vividly demonstrates by admitting infinitely many solutions through the origin.

The Core Insight

Solving the ODE is equivalent to solving an integral equation, and that integral equation can be written as a fixed-point problem $y = T (y)$ for an operator $T$ . When $f$ is Lipschitz, $T$ becomes a contraction on a suitable complete metric space of functions. The Banach fixed-point theorem then delivers a unique fixed point — the unique solution — and even tells us how to construct it: iterate $T$ starting from any guess.

Formal Statement

Lipschitz Condition

A function $f (t, y)$ satisfies a Lipschitz condition in $y$ on a region $R$ if there exists a constant $L \geq 0$ (the Lipschitz constant) such that

∣ f (t, y_{1}) - f (t, y_{2}) ∣ \leq L ∣ y_{1} - y_{2} ∣ for all (t, y_{1}), (t, y_{2}) \in R .

⟳This is a uniform bound on how fast

f

can change in the

y

-direction; it is weaker than differentiability but stronger than mere continuity.

A sufficient (and commonly used) condition: if $\partial f / \partial y$ exists and is bounded on a convex region $R$ , then by the mean value theorem $f$ is Lipschitz with $L = sup_{R} ∣ \partial f / \partial y ∣$ .

The Initial Value Problem

We study the initial value problem (IVP)

y^{'} (t) = f (t, y (t)), y (t_{0}) = y_{0},

⟳

y : I \to R

(or

R^{n}

) is the unknown;

f

is the given vector field;

(t_{0}, y_{0})

is the prescribed initial condition.

The symbols: $t$ is the independent variable (often time); $y$ is the dependent variable (state); $f$ encodes the law of evolution; $t_{0}$ is the initial time; $y_{0}$ is the initial state. A solution is a differentiable function satisfying both the equation and the initial condition.

Picard-Lindelöf

Let $R = {(t, y) : ∣ t - t_{0} ∣ \leq a, ∣ y - y_{0} ∣ \leq b}$ be a closed rectangle. Suppose:

$f$ is continuous on $R$ (so $∣ f ∣ \leq M$ for some $M$ ), and
$f$ satisfies a Lipschitz condition in $y$ on $R$ with constant $L$ .

Then the IVP [2] has a unique solution $y (t)$ on the interval $∣ t - t_{0} ∣ \leq h$ , where

h = min (a, \frac{b}{M}) .

⟳

b / M

is the time before the solution could escape the rectangle vertically — speed

\leq M

, vertical room

b

— so we shrink the interval to keep the iterates inside

R

Why It Holds / Derivation

Existence and Uniqueness via Picard Iteration

Step 1 — Reformulate as an integral equation. A continuous function

y

solves [2] if and only if it satisfies

y (t) = y_{0} + \int_{t_{0}}^{t} f (s, y (s)) d s .

⟳Integrate

y^{'} = f (t, y)

from

t_{0}

t

and apply the Fundamental Theorem of Calculus; the initial condition supplies the constant

y_{0}

Step 2 — Define the Picard operator. On the space

X = C ([t_{0} - h, t_{0} + h])

of continuous functions (with the sup norm), define

(T y) (t) = y_{0} + \int_{t_{0}}^{t} f (s, y (s)) d s .

⟳A solution of [4] is exactly a fixed point

T y = y

; we now show

T

maps a complete space to itself and is a contraction.

Step 3 — $T$ maps the closed ball into itself. Restrict to functions with

∣ y (t) - y_{0} ∣ \leq b

. Then

∣ (T y) (t) - y_{0} ∣ = \int_{t_{0}}^{t} f (s, y (s)) d s \leq M ∣ t - t_{0} ∣ \leq M h \leq b,

⟳Use

∣ f ∣ \leq M

and

h \leq b / M

from [3]; this guarantees iterates never leave the rectangle

R

Step 4 — $T$ is a contraction (Lipschitz key step). For two functions

y_{1}, y_{2}

∣ (T y_{1}) (t) - (T y_{2}) (t) ∣ \leq \int_{t_{0}}^{t} L ∣ y_{1} (s) - y_{2} (s) ∣ d s \leq L h ∥ y_{1} - y_{2} ∥_{\infty} .

⟳Apply the Lipschitz condition [1] inside the integral; the worst case replaces

∣ y_{1} - y_{2} ∣

by its sup norm.

Taking the supremum over $t$ gives $∥ T y_{1} - T y_{2} ∥_{\infty} \leq L h ∥ y_{1} - y_{2} ∥_{\infty}$ . If we choose $h$ small enough that $L h < 1$ , then $T$ is a contraction.

Step 5 — Apply Banach. Since

X

(the closed ball in

C

) is a complete metric space and

T

is a contraction, the Banach Fixed-Point Theorem yields a unique fixed point

y^{*}

, which is the unique solution of [4], hence of [2].

■

Removing $L h < 1$ : One need not require

L h < 1

. A sharper estimate by induction shows the

n

-th iterate satisfies

∥ T^{n} y_{1} - T^{n} y_{2} ∥_{\infty} \leq \frac{( L h ) ^{n}}{n !} ∥ y_{1} - y_{2} ∥_{\infty}

, and since

(L h)^{n} / n! \to 0

, some power

T^{n}

is a contraction. This recovers uniqueness on the full interval

∣ t - t_{0} ∣ \leq h

from [3], independent of

L

Think of Picard iteration as successive approximation. Start with the crude guess $y_{0} (t) \equiv y_{0}$ (the solution stands still). Feed it into the integral [5] to get a better guess $y_{1}$ , which incorporates the slope $f$ at the starting point. Feed $y_{1}$ back in to get $y_{2}$ , accounting for how the slope changes as $y$ moves, and so on. Each pass corrects the previous approximation. The Lipschitz condition controls how much error can be amplified per pass — the factor $L h$ — and because integration shrinks errors faster than $L$ grows them (the $1/ n!$ ), the corrections shrink to zero and the sequence locks onto the one true solution. The contraction is the mathematical statement that "errors die out."

Picard iterates converging to the solution of $y'=y$, $y(0)=1$

<svg viewBox="0 0 460 300" xmlns="http://www.w3.org/2000/svg">
<rect width="460" height="300" fill="#0a0e17"/>
<line x1="40" y1="260" x2="440" y2="260" stroke="#94a3b8" stroke-width="1.5"/>
<line x1="40" y1="260" x2="40" y2="20" stroke="#94a3b8" stroke-width="1.5"/>
<text x="430" y="278" fill="#e2e8f0" font-size="12">t</text>
<text x="14" y="30" fill="#e2e8f0" font-size="12">y</text>
<text x="30" y="264" fill="#e2e8f0" font-size="11">0</text>
<text x="20" y="232" fill="#e2e8f0" font-size="11">1</text>
<line x1="36" y1="228" x2="44" y2="228" stroke="#94a3b8"/>
<!-- y0 = 1 constant -->
<line x1="40" y1="228" x2="440" y2="228" stroke="#64748b" stroke-width="1" stroke-dasharray="4 3"/>
<text x="345" y="222" fill="#64748b" font-size="10">y₀=1</text>
<!-- y1 = 1+t : line -->
<line x1="40" y1="228" x2="440" y2="68" stroke="#7a8aa0" stroke-width="1.2" stroke-dasharray="2 2"/>
<text x="350" y="100" fill="#7a8aa0" font-size="10">y₁=1+t</text>
<!-- y2 = 1+t+t^2/2 parabola approx -->
<path d="M40,228 Q240,168 440,40" stroke="#a8b4c8" stroke-width="1.2" fill="none"/>
<text x="200" y="160" fill="#a8b4c8" font-size="10">y₂</text>
<!-- true solution e^t -->
<path d="M40,228 C140,215 250,150 440,28" stroke="#d4a853" stroke-width="2.4" fill="none"/>
<text x="370" y="50" fill="#d4a853" font-size="11">y = eᵗ</text>
<circle cx="40" cy="228" r="3.5" fill="#d4a853"/>
<text x="46" y="248" fill="#e2e8f0" font-size="10">(0,1)</text>
</svg>

Émile Picard | 1856–1941

Charles Émile Picard was a towering figure of French mathematics, working across complex analysis, algebraic geometry, and differential equations. His method of successive approximations (1890) made the existence-uniqueness problem constructive: rather than merely asserting a solution exists, Picard built it as the limit of an explicitly computable sequence. He is also famous for Picard's theorem in complex analysis (an entire non-constant function omits at most one value). His iteration scheme directly anticipated the abstract fixed-point methods that Banach later formalized, making Picard a bridge between classical analysis and modern functional analysis. Ernst Leonard Lindelöf (1870–1946), the Finnish mathematician, refined the conditions and sharpened the interval of existence in 1894 — hence the joint name.

Worked Examples

Linear ODE — verifying the hypotheses and iterating

Given:

y^{'} = 2 y

y (0) = 1

. Here

f (t, y) = 2 y

t_{0} = 0

y_{0} = 1

. Step 1 — Check Lipschitz.

\partial f / \partial y = 2

, bounded everywhere, so

f

is Lipschitz with

L = 2

. Continuity in

t

is trivial. Hypotheses hold globally. Step 2 — Picard iterates. Start

y_{0} (t) = 1

y_{1} (t) = 1 + \int_{0}^{t} 2 \cdot 1 d s = 1 + 2 t .

⟳Substitute

y_{0} \equiv 1

into the operator [5] with

f = 2 y

y_{2} (t) = 1 + \int_{0}^{t} 2 (1 + 2 s) d s = 1 + 2 t + 2 t^{2} .

⟳Substitute

y_{1} = 1 + 2 s

and integrate term by term.

Continuing, $y_{n} (t) = \sum_{k = 0}^{n} \frac{( 2 t ) ^{k}}{k !}$ , the partial sums of $e^{2 t}$ .

Step 3 — Limit.

y (t) = e^{2 t}

The Picard iterates literally generate the Taylor series of the exact solution. Existence interval: all of $R$ .

Nonlinear ODE — finite existence interval

Given:

y^{'} = y^{2}

y (0) = 1

. Here

f (t, y) = y^{2}

. Step 1 — Local Lipschitz, not global.

\partial f / \partial y = 2 y

, bounded on any rectangle

∣ y - 1∣ \leq b

L = 2 (1 + b)

. So Picard-Lindelöf applies locally. Step 2 — Estimate the interval. Take

a = \infty

b = 1

∣ y ∣ \leq 2

, giving

M = sup y^{2} = 4

. Then

h = min (a, \frac{b}{M}) = min (\infty, \frac{1}{4}) = \frac{1}{4} .

⟳Apply [3] with

b = 1

M = 4

; the theorem guarantees existence at least on

∣ t ∣ \leq \frac{1}{4}

Step 3 — Exact solution. Separating variables,

- 1/ y = t - 1

, so

y (t) = \frac{1}{1 - t}, valid for t < 1.

The solution blows up at $t = 1$ (finite-time blow-up). The theorem's guaranteed interval ( $h = \frac{1}{4}$ ) is conservative but correctly local — no choice of rectangle extends the guarantee past $t = 1$ .

Common Misconceptions

Continuity alone does NOT give uniqueness. A common error is to assume that if

f

is continuous, the IVP has a unique solution. Continuity (Peano's theorem) guarantees existence but not uniqueness. The full Lipschitz condition is what forces uniqueness. Conversely, the Lipschitz condition is sufficient, not necessary — some uniqueness can hold without it (e.g., via Osgood's criterion), so failure of Lipschitz does not automatically mean non-uniqueness. Also: the theorem is local. It guarantees a solution only on a (possibly tiny) interval around

t_{0}

. Global existence requires additional structure (e.g., a global Lipschitz/linear-growth bound).

$y'=\sqrt{|y|}$ — non-uniqueness from a failed Lipschitz condition

Consider $y^{'} = ∣ y ∣$ , $y (0) = 0$ . Here $f (t, y) = ∣ y ∣$ is continuous but not Lipschitz at $y = 0$ , since $\partial f / \partial y = \frac{1}{2 y} \to \infty$ as $y \to 0^{+}$ .

One solution is the trivial $y (t) \equiv 0$ . But another is

y (t) = {\frac{1}{4} t^{2} 0 t \geq 0 t < 0,

⟳For

y \geq 0

, separate variables:

\int y^{- 1/2} d y = \int d t \Rightarrow 2 y = t \Rightarrow y = t^{2} /4

In fact, for any $c \geq 0$ the function that stays at $0$ until $t = c$ then takes off as $\frac{1}{4} (t - c)^{2}$ is also a solution — infinitely many solutions through the origin. Peano gives existence; the absent Lipschitz condition destroys uniqueness.

Connections

Picard-Lindelöf sits at the crossroads of several areas. Its proof is the showcase application of the Banach Fixed-Point Theorem — the abstract contraction principle from Metric Spaces. The weaker Peano Existence Theorem drops the Lipschitz hypothesis and uses compactness (Arzelà-Ascoli) to get existence without uniqueness. For systems and higher-order ODEs, the same theorem applies verbatim with $y \in R^{n}$ and a vector norm, underpinning the whole theory of Linear Systems of ODEs. The dependence of solutions on initial data — continuous and even smooth dependence — extends Picard-Lindelöf and is essential to Dynamical Systems and stability theory. The integral reformulation [4] is the gateway to Volterra Integral Equations.

Real-World Applications

Newtonian mechanics: $m \overset{x}{¨} = F (x, \overset{x}{˙}, t)$ with smooth force fields has unique trajectories — the mathematical basis of deterministic classical physics (Laplace's demon).
Population dynamics: the logistic equation $\dot{P} = r P (1 - P / K)$ has a unique solution for each initial population, used in ecology and epidemiology.
Circuit analysis (SPICE): transient simulation of RLC circuits solves IVPs whose well-posedness rests on Lipschitz vector fields.
Orbital mechanics & spacecraft navigation: numerical integrators (Runge-Kutta) presuppose Picard-Lindelöf to know the trajectory being approximated actually exists and is unique.
Chemical kinetics: reaction-rate ODEs (mass-action laws) are polynomial, hence locally Lipschitz, guaranteeing unique concentration profiles.
Control theory: state-space models $\overset{x}{˙} = f (x, u)$ require uniqueness for the notions of controllability and feedback stabilization to make sense.
Epidemiological SIR/SEIR models: uniqueness of solutions justifies parameter-fitting and forecast reproducibility during outbreaks.

Historical Development

1768–1820 — Euler, Lagrange, Cauchy develop methods for solving ODEs; Cauchy gives the first rigorous existence proof (the Cauchy-Euler polygon method) in his 1820s lectures.
1868 — Rudolf Lipschitz introduces the condition now bearing his name, sharpening Cauchy's existence-uniqueness arguments.
1876 — Hermann Amandus Schwarz / Cauchy-Kovalevskaya lineage strengthens analytic-data results for PDEs and ODEs.
1890 — Émile Picard publishes the method of successive approximations, making existence constructive and iterative.
1894 — Ernst Lindelöf refines Picard's conditions and the interval of existence; the joint name Picard-Lindelöf dates from here.
1890 — Giuseppe Peano proves existence under mere continuity (no uniqueness), delineating exactly what Lipschitz buys.
1898 — William Osgood gives a uniqueness criterion weaker than Lipschitz (the Osgood condition), mapping the boundary of uniqueness.
1922 — Stefan Banach proves the fixed-point theorem, retroactively recasting Picard's iteration as a contraction argument — the modern textbook proof.
1920s–30s — Carathéodory extends existence to $f$ merely measurable in $t$ (Carathéodory conditions), broadening applicability to discontinuous forcing.

Subtleties & Limits

The Picard-Lindelöf framework is essentially closed as a theorem, but its boundary raises subtle, still-active questions. (1) Sharp uniqueness criteria: exactly which moduli of continuity (between Lipschitz and bare continuity) force uniqueness? Osgood's condition is sufficient but the complete characterization is delicate. (2) Maximal interval of existence: the theorem gives a local interval; determining the exact blow-up time for nonlinear systems is generally undecidable in closed form and connects to deep open problems (e.g., global regularity for the 3D Navier-Stokes equations is the PDE analogue of "does the solution exist for all time?"). (3) Infinite-dimensional generalizations: in Banach spaces, Lipschitz vector fields still give well-posedness, but for the weaker compactness-based existence (Peano) the analogue can fail — Godunov showed Peano's theorem is false in general infinite-dimensional Banach spaces. (4) Numerical realizability: how well Picard iteration (vs. Runge-Kutta) approximates stiff or chaotic systems remains a practical research area in numerical analysis.