Skip to content

Existence and Uniqueness for First-Order IVPs

CSIR NET ODE: Lipschitz condition, Peano and Picard-Lindelof theorems, Picard iterates, interval of existence, non-uniqueness, Gronwall, blow-up.

  • 24 min read
  • Updated
On this page · 15 sections
  1. Prerequisites
  2. The Hook: When Did the Tank Run Dry?
  3. The Setting
  4. The Two Theorems
  5. When Uniqueness Fails
  6. Gronwall’s Inequality and Continuous Dependence
  7. How Far Does the Solution Go?
  8. Solving Toolkit for First-Order Equations
  9. Memory Tricks
  10. Don’t Confuse
  11. Common Mistakes
  12. Practice
  13. Quick Revision Table
  14. Key Takeaways
  15. What’s Next

Prerequisites

The Hook: When Did the Tank Run Dry?

The Setting

An initial value problem (IVP) is

$$y' = f(x, y), \qquad y(x_0) = y_0,$$

with \( f \) defined on a rectangle \( R = \{ (x, y) : \lvert x - x_0 \rvert \le a, \ \lvert y - y_0 \rvert \le b \} \) or on a strip \( [x_0 - a, x_0 + a] \times \mathbb{R} \).

If \( f \) is continuous, a \( C^1 \) function \( y \) solves the IVP on an interval \( I \ni x_0 \) if and only if it is continuous and satisfies the integral equation

$$\boxed{y(x) = y_0 + \int_{x_0}^{x} f(t, y(t))\,dt, \quad x \in I.}$$

Every existence proof works with this integral form, because integrals are easier to control than derivatives.

The Lipschitz condition

\( f \) is Lipschitz in \( y \) on a set \( D \) if there is \( L \ge 0 \) with

$$\lvert f(x, y_1) - f(x, y_2) \rvert \le L \lvert y_1 - y_2 \rvert \quad \text{for all } (x, y_1), (x, y_2) \in D.$$

Only the \( y \)-variable matters. \( f \) may be merely continuous in \( x \).

\( f(x, y) \)DomainLipschitz in \( y \)?Reason
\( x^2 + y^2 \)rectangleyes\( \lvert f_y \rvert = 2\lvert y \rvert \le 2(\lvert y_0 \rvert + b) \)
\( x^2 + y^2 \)strip \( [-a, a] \times \mathbb{R} \)no\( f_y \) unbounded
\( \lvert y \rvert \)anywhereyes, \( L = 1 \)\( \big\lvert \lvert y_1 \rvert - \lvert y_2 \rvert \big\rvert \le \lvert y_1 - y_2 \rvert \); not differentiable at 0
\( \sin y + x \)\( \mathbb{R}^2 \)yes, \( L = 1 \)\( \lvert \cos y \rvert \le 1 \)
\( y^{2/3} \), \( \sqrt{\lvert y \rvert} \)any set meeting \( y = 0 \)no\( \frac{f(y) - f(0)}{y} \to \infty \) as \( y \to 0 \)
\( y^{2/3} \)\( \lvert y - 1 \rvert \le \frac{1}{2} \)yesaway from 0, \( f_y \) is bounded
\( x \lvert y \rvert^{1/2} \)rectangle around \( (0, 0) \)nofails at \( y = 0 \) for every \( x \ne 0 \)

The Two Theorems

Let \( f \) be continuous on \( R \), put \( M = \max_R \lvert f \rvert \), and

$$\boxed{h = \min\left(a, \frac{b}{M}\right).}$$

Why \( b / M \)? A solution starts at \( y_0 \) with slope at most \( M \) in absolute value, so in time \( \lvert x - x_0 \rvert \) it moves at most \( M \lvert x - x_0 \rvert \). Keeping \( M \lvert x - x_0 \rvert \le b \) keeps the graph inside \( R \), where the bounds hold. The theorem cannot promise more without information outside \( R \).

Picard iteration

Start with \( y_0(x) \equiv y_0 \) and define

$$y_{n+1}(x) = y_0 + \int_{x_0}^{x} f(t, y_n(t))\,dt.$$

On \( \lvert x - x_0 \rvert \le h \) every iterate stays in \( R \), and induction with the Lipschitz condition gives

$$\lvert y_{n+1}(x) - y_n(x) \rvert \le \frac{M L^n \lvert x - x_0 \rvert^{n+1}}{(n+1)!}.$$

The right side sums to \( \frac{M}{L}\left(e^{Lh} - 1\right) \), so the series \( y_0 + \sum (y_{n+1} - y_n) \) converges uniformly by the M-test. The limit is continuous, and passing to the limit under the integral (uniform convergence) shows it solves the integral equation.

The same argument in modern form: the map \( (Ty)(x) = y_0 + \int_{x_0}^x f(t, y(t))\,dt \) is a contraction on a closed subset of \( C[x_0 - h, x_0 + h] \) (sup metric, or a weighted sup metric if \( Lh \ge 1 \)), and the Banach fixed point theorem gives one fixed point.

Decision path for a Part B or C question

graph TD
    A["IVP y' = f(x,y), y(x0) = y0"] --> B{"f continuous near (x0, y0)?"}
    B -- no --> X["No general guarantee"]
    B -- yes --> C["A solution exists (Peano)"]
    C --> D{"f Lipschitz in y near (x0, y0)?"}
    D -- yes --> E["Unique local solution (Picard)"]
    D -- no --> F{"f(x0, y0) = 0 on a non-Lipschitz line?"}
    F -- yes --> G["Test for a second solution leaving that line"]
    F -- no --> H["Often still unique: check directly"]
    E --> I{"Lipschitz on the whole strip, or linear growth?"}
    I -- yes --> J["Solution exists on the whole interval"]
    I -- no --> K["Possible blow-up: solve or compare"]

When Uniqueness Fails

Continuity without Lipschitz gives existence (Peano) but not uniqueness.

IVPSolutions
\( y' = 3y^{2/3} \), \( y(0) = 0 \)\( y \equiv 0 \); \( y = x^3 \); and for any \( \alpha \le 0 \le \beta \): \( y = (x - \alpha)^3 \) for \( x < \alpha \), 0 on \( [\alpha, \beta] \), \( (x - \beta)^3 \) for \( x > \beta \)
\( y' = \sqrt{\lvert y \rvert} \), \( y(0) = 0 \)\( y \equiv 0 \); \( y = \frac{(x - c)^2}{4} \) for \( x \ge c \) and 0 before, for any \( c \ge 0 \)
\( y' = 2\sqrt{y} \) (\( y \ge 0 \)), \( y(0) = 0 \)\( y \equiv 0 \); \( y = (x - c)^2 \) for \( x \ge c \), 0 before
\( xy' = 2y \), \( y(0) = 0 \)\( y = cx^2 \) for every \( c \) (here \( f = \frac{2y}{x} \) is not even defined at \( x = 0 \))
\( xy' = 2y \), \( y(0) = 1 \)no solution at all

The last two rows are a reminder that Picard’s theorem needs the equation in the form \( y' = f(x, y) \) with \( f \) continuous at \( (x_0, y_0) \). Dividing by a coefficient that vanishes at \( x_0 \) breaks the hypothesis, and anything can happen.

Gronwall’s Inequality and Continuous Dependence

Gronwall’s inequality. If \( u \) is continuous and non-negative on \( [x_0, x_1] \), \( C \ge 0 \), \( L \ge 0 \), and

$$u(x) \le C + L \int_{x_0}^{x} u(t)\,dt \quad \text{for } x \in [x_0, x_1],$$

then \( u(x) \le C e^{L(x - x_0)} \) there.

Two consequences, both proved in one line:

  1. Uniqueness. If \( y \) and \( z \) solve the same IVP, \( u = \lvert y - z \rvert \) satisfies the inequality with \( C = 0 \), so \( u \equiv 0 \).
  2. Continuous dependence on initial data. If \( y, z \) solve \( y' = f(x, y) \) with \( f \) Lipschitz (constant \( L \)), then
$$\boxed{\lvert y(x) - z(x) \rvert \le \lvert y(x_0) - z(x_0) \rvert \, e^{L \lvert x - x_0 \rvert}.}$$

A corollary used constantly in Part C: two different solutions of a Lipschitz equation never meet. Their graphs cannot cross or touch, so a solution that starts above an equilibrium stays above it. For example, every solution of \( y' = y(1 - y) \) with \( 0 < y(0) < 1 \) stays in \( (0, 1) \) forever.

How Far Does the Solution Go?

Global existence on a strip

If \( f \) is continuous on the strip \( S = [x_0 - a, x_0 + a] \times \mathbb{R} \) and Lipschitz in \( y \) on all of \( S \), the IVP has a unique solution on the whole interval \( [x_0 - a, x_0 + a] \). No \( b / M \) restriction appears, because there is no top or bottom of \( R \) to leave.

Main instances:

  • Linear equations \( y' = p(x) y + q(x) \) with \( p, q \) continuous on an interval \( I \): the solution exists on all of \( I \).
  • \( f \) continuous and bounded on \( I \times \mathbb{R} \), or of linear growth \( \lvert f(x, y) \rvert \le A + B \lvert y \rvert \): solutions exist on all of \( I \) (uniqueness still needs Lipschitz).

Maximal interval and blow-up

If \( f \) is continuous and locally Lipschitz on an open set \( D \), every solution extends to a maximal interval \( (\alpha, \omega) \), and as \( x \to \omega \) the point \( (x, y(x)) \) leaves every compact subset of \( D \). When \( D = \mathbb{R}^2 \) and \( \omega < \infty \), this means \( \lvert y(x) \rvert \to \infty \) as \( x \to \omega^- \): the solution blows up.

IVPSolutionMaximal interval
\( y' = y^2 \), \( y(0) = y_0 > 0 \)\( \frac{y_0}{1 - y_0 x} \)\( \left(-\infty, \frac{1}{y_0}\right) \)
\( y' = y^2 \), \( y(0) = y_0 < 0 \)same formula\( \left(\frac{1}{y_0}, \infty\right) \)
\( y' = 1 + y^2 \), \( y(0) = 0 \)\( \tan x \)\( \left(-\frac{\pi}{2}, \frac{\pi}{2}\right) \)
\( y' = y^3 \), \( y(0) = 1 \)\( (1 - 2x)^{-1/2} \)\( \left(-\infty, \frac{1}{2}\right) \)
\( y' = -y^3 \), \( y(0) = 1 \)\( (1 + 2x)^{-1/2} \)\( \left(-\frac{1}{2}, \infty\right) \)
\( y' = \cos(xy) \), any \( y(0) \)no closed form; right side bounded by 1\( \mathbb{R} \)

The pattern: superlinear growth (\( y^2 \), \( y^3 \), \( e^y \)) can blow up in finite time; linear growth cannot. Note the direction: \( y' = -y^3 \) is perfectly tame forward in time and blows up backward.

Solving Toolkit for First-Order Equations

Most Part B items need the explicit solution before you can read off its interval. The standard methods, compressed:

TypeFormMethod
Separable\( y' = g(x)h(y) \)\( \int \frac{dy}{h(y)} = \int g\,dx \); check zeros of \( h \) separately (they are constant solutions)
Linear\( y' + P(x) y = Q(x) \)integrating factor \( \mu = e^{\int P} \): \( (\mu y)' = \mu Q \)
Bernoulli\( y' + P y = Q y^n \), \( n \ne 0, 1 \)\( v = y^{1-n} \) makes it linear
Homogeneous\( y' = F\left(\frac{y}{x}\right) \)\( y = vx \) makes it separable
Exact\( M\,dx + N\,dy = 0 \), \( M_y = N_x \)find \( \phi \) with \( \phi_x = M \), \( \phi_y = N \); solution \( \phi = c \)
Integrating factor\( \frac{M_y - N_x}{N} = g(x) \)\( \mu = e^{\int g(x)\,dx} \)
Integrating factor\( \frac{N_x - M_y}{M} = k(y) \)\( \mu = e^{\int k(y)\,dy} \)
Riccati\( y' = P + Q y + R y^2 \)with one known solution \( y_1 \), \( y = y_1 + \frac{1}{v} \) makes it linear in \( v \)

Equations not solved for \( y' \) (Clairaut, Lagrange) are on the next page, Singular Solutions and Systems.

Memory Tricks

  • “Continuity gives a solution, Lipschitz gives the solution.” Peano: at least one. Picard: exactly one.
  • \( h = \min(a, b/M) \): “time = distance / speed”. The solution may travel vertical distance \( b \) at speed at most \( M \).
  • Fractional power at zero = alarm. \( y^{2/3} \), \( \sqrt{\lvert y \rvert} \), \( y^{1/3} \) at \( y_0 = 0 \): look for a second solution that sits at 0 and then leaves.
  • Linear growth stays, superlinear may explode. \( \lvert f \rvert \le A + B \lvert y \rvert \): global. \( y^2 \): check for blow-up.
  • Lipschitz solutions never meet. Use it to trap solutions between equilibria.

Don’t Confuse

ThisvsThat
Peano: \( f \) continuous \( \Rightarrow \) existencePicard: + Lipschitz in \( y \) \( \Rightarrow \) existence and uniqueness
Lipschitz in \( y \) (needed)Lipschitz or differentiable in \( x \) (not needed)
\( f_y \) bounded \( \Rightarrow \) LipschitzLipschitz \( \Rightarrow \) \( f_y \) exists (false: \( \lvert y \rvert \))
Guaranteed interval \( \lvert x - x_0 \rvert \le h \)Actual maximal interval (often larger)
Lipschitz on a rectangle (local)Lipschitz on the strip (global existence)
Non-Lipschitz \( \Rightarrow \) may be non-uniqueNon-Lipschitz \( \Rightarrow \) non-unique (false: \( 1 + y^{2/3} \))

Common Mistakes

Practice

All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.

Level 1: Foundation

  1. Is \( f(x, y) = \lvert y \rvert \) Lipschitz in \( y \) on \( \mathbb{R}^2 \)? Is it differentiable in \( y \)?
  2. Write the first two Picard iterates for \( y' = 2xy \), \( y(0) = 1 \).
  3. Does \( y' = y^{1/3} \), \( y(0) = 0 \) have a unique solution?
Answers
  1. Lipschitz, yes, with \( L = 1 \), by the reverse triangle inequality. Not differentiable in \( y \) at \( y = 0 \). Lipschitz does not need a derivative.
  2. \( y_1 = 1 + \int_0^x 2t\,dt = 1 + x^2 \); \( y_2 = 1 + \int_0^x 2t(1 + t^2)\,dt = 1 + x^2 + \frac{x^4}{2} \). These are the partial sums of \( e^{x^2} \), the exact solution.
  3. No. \( y \equiv 0 \) and \( y = \left(\frac{2x}{3}\right)^{3/2} \) for \( x \ge 0 \) (0 for \( x < 0 \)) both work: \( y' = \left(\frac{2x}{3}\right)^{1/2} = y^{1/3} \). \( y^{1/3} \) is not Lipschitz at 0.

Level 2: Part B Style

Practice Q1. For \( y' = 1 + y^2 \), \( y(0) = 0 \), on the rectangle \( \lvert x \rvert \le 1 \), \( \lvert y \rvert \le 1 \), Picard’s theorem guarantees a unique solution on \( \lvert x \rvert \le h \) with \( h \) equal to:

(a) 1 (b) \( \frac{1}{2} \) (c) \( \frac{\pi}{4} \) (d) \( \frac{\pi}{2} \)

Answer: (b)

\( M = \max(1 + y^2) = 2 \) on the rectangle, so \( h = \min\left(1, \frac{1}{2}\right) = \frac{1}{2} \). The actual solution \( \tan x \) exists on \( \left(-\frac{\pi}{2}, \frac{\pi}{2}\right) \), which is option (d), but that is not what the theorem guarantees. (Even optimising over all \( b \), \( \frac{b}{1 + b^2} \le \frac{1}{2} \).)

Practice Q2. For \( y' = x + y^2 \), \( y(0) = 0 \), the second Picard iterate \( y_2(x) \) (with \( y_0 \equiv 0 \)) is:

(a) \( \frac{x^2}{2} \) (b) \( \frac{x^2}{2} + \frac{x^5}{20} \) (c) \( \frac{x^2}{2} + \frac{x^4}{8} \) (d) \( x + \frac{x^3}{3} \)

Answer: (b)

\( y_1 = \int_0^x t\,dt = \frac{x^2}{2} \). Then \( y_2 = \int_0^x \left(t + \frac{t^4}{4}\right) dt = \frac{x^2}{2} + \frac{x^5}{20} \).

Practice Q3. The maximal interval of existence of the solution of \( y' = y^2 \), \( y(1) = -1 \) is:

(a) \( (-\infty, 0) \) (b) \( (0, \infty) \) (c) \( (1, \infty) \) (d) \( \mathbb{R} \)

Answer: (b)

Separating, \( -\frac{1}{y} = x + c \); \( y(1) = -1 \) gives \( c = 0 \), so \( y = -\frac{1}{x} \). The interval must contain \( x = 1 \) and avoid the singularity at 0: \( (0, \infty) \).

Practice Q4. Which IVP has more than one solution?

(a) \( y' = \sin y \), \( y(0) = 0 \) (b) \( y' = \lvert y \rvert \), \( y(0) = 0 \) (c) \( y' = y^{4/5} \), \( y(0) = 0 \) (d) \( y' = 1 + y^{2/3} \), \( y(0) = 0 \)

Answer: (c)

(a) and (b) have Lipschitz right sides: unique (\( y \equiv 0 \)). (c): \( y \equiv 0 \) and \( y = \left(\frac{x}{5}\right)^5 \) both work, since \( y' = \left(\frac{x}{5}\right)^4 = y^{4/5} \). (d) is unique by the inversion argument: every solution is strictly increasing.

Level 3: Part C Style

Practice Q5. Consider \( y' = \sqrt{\lvert y \rvert} \). Which statements are true?

(a) The IVP with \( y(0) = 1 \) has a unique solution on some open interval containing 0. (b) The IVP with \( y(0) = 0 \) has infinitely many solutions on \( [0, \infty) \). (c) \( f(y) = \sqrt{\lvert y \rvert} \) is Lipschitz on \( [-1, 1] \). (d) The IVP with \( y(0) = 1 \) has a unique solution on all of \( \mathbb{R} \).

Answer: (a) and (b)
  • (a) True. Near \( y = 1 \), \( f'(y) = \frac{1}{2\sqrt{y}} \) is bounded, so Picard applies. The solution is \( y = \left(\frac{x}{2} + 1\right)^2 \) for \( x \ge -2 \).
  • (b) True. For each \( c \ge 0 \), \( y = 0 \) on \( [0, c] \) and \( \frac{(x - c)^2}{4} \) after \( c \) is a solution.
  • (c) False. \( \frac{f(y) - f(0)}{y - 0} = \frac{1}{\sqrt{y}} \to \infty \) as \( y \to 0^+ \).
  • (d) False. The solution reaches \( y = 0 \) at \( x = -2 \). Continuing backward, it may stay at 0 on \( [c, -2] \) for any \( c \le -2 \) and then follow \( -\frac{(x - c)^2}{4} \) for \( x < c \) (check: \( y' = -\frac{x - c}{2} = \frac{c - x}{2} = \sqrt{\lvert y \rvert} \)). Solutions exist on all of \( \mathbb{R} \) (linear growth), but they are not unique there.

Practice Q6. Let \( y_1, y_2 \) solve \( y' = \sin y + x \) with \( y_1(0) = 0 \), \( y_2(0) = 0.01 \). Which statements are true?

(a) Both solutions exist on all of \( \mathbb{R} \). (b) \( \lvert y_1(1) - y_2(1) \rvert \le 0.01\,e \). (c) The graphs of \( y_1 \) and \( y_2 \) cross at some \( x > 0 \). (d) \( f(x, y) = \sin y + x \) is Lipschitz in \( y \) on \( \mathbb{R}^2 \) with constant 1.

Answer: (a), (b) and (d)
  • (d) True. \( \lvert f_y \rvert = \lvert \cos y \rvert \le 1 \) on all of \( \mathbb{R}^2 \).
  • (a) True. Lipschitz on every strip \( [-a, a] \times \mathbb{R} \) gives existence on \( [-a, a] \) for every \( a \).
  • (b) True. Gronwall: \( \lvert y_1(x) - y_2(x) \rvert \le 0.01\,e^{1 \cdot \lvert x \rvert} \); at \( x = 1 \) this is \( 0.01e \).
  • (c) False. If they met at some point, both would solve the same IVP there, and uniqueness would make them equal everywhere, contradicting \( y_1(0) \ne y_2(0) \).

Practice Q7. Consider \( y' = x^2 + y^2 \), \( y(0) = 0 \), and the rectangle \( \lvert x \rvert \le 1 \), \( \lvert y \rvert \le 1 \). Which statements are true?

(a) Picard’s theorem guarantees a unique solution on \( \lvert x \rvert \le \frac{1}{2} \). (b) The second Picard iterate (from \( y_0 \equiv 0 \)) is \( \frac{x^3}{3} + \frac{x^7}{63} \). (c) The solution is an odd function on its interval of existence. (d) The solution exists on all of \( \mathbb{R} \).

Answer: (a), (b) and (c)
  • (a) True. \( M = \max(x^2 + y^2) = 2 \), so \( h = \min\left(1, \frac{1}{2}\right) = \frac{1}{2} \).
  • (b) True. \( y_1 = \int_0^x t^2\,dt = \frac{x^3}{3} \); \( y_2 = \int_0^x \left(t^2 + \frac{t^6}{9}\right) dt = \frac{x^3}{3} + \frac{x^7}{63} \).
  • (c) True. Let \( z(x) = -y(-x) \). Then \( z'(x) = y'(-x) = x^2 + y(-x)^2 = x^2 + z(x)^2 \) and \( z(0) = 0 \). By uniqueness \( z = y \), i.e. \( y(-x) = -y(x) \).
  • (d) False. The solution is increasing for \( x > 0 \) and \( y(1) > 0 \). For \( x \ge 1 \), \( y' \ge 1 + y^2 \), so \( \arctan y(x) - \arctan y(1) \ge x - 1 \). The right side passes \( \frac{\pi}{2} \) before \( x = 1 + \frac{\pi}{2} \), so \( y \) must blow up before then.

Quick Revision Table

IdeaFact
Integral form\( y = y_0 + \int_{x_0}^x f(t, y(t))\,dt \)
Lipschitz test\( \lvert f_y \rvert \le L \) on a rectangle or strip
Peanocontinuous \( \Rightarrow \) at least one solution on \( \lvert x - x_0 \rvert \le h \)
Picard–Lindelöf+ Lipschitz in \( y \) \( \Rightarrow \) exactly one
Interval\( h = \min(a, b/M) \), \( M = \max_R \lvert f \rvert \)
Iterates\( y_{n+1} = y_0 + \int_{x_0}^x f(t, y_n)\,dt \); error terms \( \frac{ML^n \lvert x - x_0 \rvert^{n+1}}{(n+1)!} \)
Non-uniqueness\( y^{2/3} \), \( \sqrt{\lvert y \rvert} \), \( y^{1/3} \) at \( y_0 = 0 \)
Gronwall\( \lvert y - z \rvert \le \lvert y(x_0) - z(x_0) \rvert e^{L \lvert x - x_0 \rvert} \)
Global existenceLipschitz on the strip, or linear growth; linear equations: whole interval
Blow-up\( y' = y^2 \): \( \frac{y_0}{1 - y_0 x} \); \( y' = 1 + y^2 \): \( \tan x \)

Key Takeaways

What’s Next