Skip to content

The Cauchy Problem for First-Order PDEs

CSIR NET PDE: Cauchy problem, method of characteristics, transversality condition, existence and uniqueness, characteristic data, Burgers equation, strips.

  • 20 min read
  • Updated
On this page · 14 sections
  1. Prerequisites
  2. The Hook: A Wave That Breaks
  3. Statement of the Cauchy Problem
  4. The Method of Characteristics
  5. The Transversality Condition and the Theorem
  6. Quasilinear Equations: Breaking and Shocks
  7. Nonlinear Equations: Characteristic Strips
  8. Memory Tricks
  9. Don’t Confuse
  10. Common Mistakes
  11. Practice
  12. Quick Revision Table
  13. Key Takeaways
  14. What’s Next

Prerequisites

The Hook: A Wave That Breaks

Statement of the Cauchy Problem

Let

$$a(x, y, z)\,p + b(x, y, z)\,q = c(x, y, z)$$

with \( a, b, c \) continuously differentiable, and let \( \Gamma \) be a curve in \( \mathbb{R}^3 \) given by

$$\Gamma: \quad x = x_0(s), \quad y = y_0(s), \quad z = z_0(s), \qquad s \in I.$$

Cauchy problem: find a solution \( z = u(x, y) \), defined in a neighbourhood of the projection of \( \Gamma \) on the \( xy \)-plane, whose graph contains \( \Gamma \), i.e. \( u(x_0(s), y_0(s)) = z_0(s) \).

The most common case is data on an axis: \( u(x, 0) = f(x) \), i.e. \( x_0 = s \), \( y_0 = 0 \), \( z_0 = f(s) \). In evolution problems \( y \) is time \( t \) and this is an initial-value problem.

The Method of Characteristics

Through each point of \( \Gamma \) send out the characteristic curve of the vector field \( (a, b, c) \). The union of these curves is the integral surface.

$$\boxed{\frac{dx}{dt} = a(x, y, z), \quad \frac{dy}{dt} = b(x, y, z), \quad \frac{dz}{dt} = c(x, y, z)}$$

with initial values \( x(0, s) = x_0(s) \), \( y(0, s) = y_0(s) \), \( z(0, s) = z_0(s) \).

graph TD
    A["Data curve Gamma: (x0(s), y0(s), z0(s))"] --> B["Check transversality
J(s) = a y0' - b x0' at t = 0"] B -->|"J not 0"| C["Solve characteristic ODEs
x(t,s), y(t,s), z(t,s)"] C --> D["Invert (t, s) to (x, y)"] D --> E["u(x, y) = z(t(x,y), s(x,y)):
unique local solution"] B -->|"J = 0 along Gamma"| F{"Is Gamma itself
a characteristic?"} F -->|yes| G["Infinitely many solutions"] F -->|no| H["No solution"]

Steps.

  1. Parametrise the data: \( (x_0(s), y_0(s), z_0(s)) \).
  2. Solve the three ODEs in \( t \) for each fixed \( s \). (For a linear equation the first two do not involve \( z \), so the projected characteristics are known in advance.)
  3. Invert \( x = x(t, s) \), \( y = y(t, s) \) to get \( t, s \) as functions of \( x, y \).
  4. Substitute into \( z(t, s) \).

Shortcut when the general solution is easy. If Lagrange’s method gives \( z = \text{(known)} + f(\text{known}) \), simply impose the data on the general solution to fix \( f \). Both routes give the same answer.

The Transversality Condition and the Theorem

At \( t = 0 \), the Jacobian of \( (t, s) \mapsto (x, y) \) is

$$J(s) = \frac{\partial(x, y)}{\partial(t, s)}\bigg|_{t=0} = \begin{vmatrix} a & b \\ x_0'(s) & y_0'(s) \end{vmatrix} = a\,y_0'(s) - b\,x_0'(s),$$

with \( a, b \) evaluated at \( (x_0(s), y_0(s), z_0(s)) \).

Geometric meaning. \( J \ne 0 \) says the projected characteristic direction \( (a, b) \) is not tangent to the projected data curve \( (x_0', y_0') \). The characteristics leave the curve, sweep out a strip, and the inverse function theorem gives local coordinates \( (t, s) \).

When \( J \equiv 0 \) along \( \Gamma \): the projected data curve is itself a projected characteristic.

SituationOutcome
\( J \ne 0 \) along \( \Gamma \)unique local solution
\( J \equiv 0 \), and \( \Gamma \) is a characteristic curve (\( (x_0', y_0', z_0') \parallel (a, b, c) \))infinitely many solutions
\( J \equiv 0 \), and \( \Gamma \) is not a characteristic curveno solution

The reason: along a characteristic, the PDE fixes how \( z \) changes. If the data obey that rule, they say nothing new (many surfaces contain the curve); if they disobey it, no surface can.

Quasilinear Equations: Breaking and Shocks

For \( u_t + c(u)\,u_x = 0 \), \( u(x, 0) = f(x) \), the characteristic equations are \( \frac{dx}{dt} = c(u) \), \( \frac{du}{dt} = 0 \). So \( u \) is constant along each characteristic, and the characteristics are straight lines:

$$x = s + c(f(s))\,t, \qquad u = f(s), \qquad \text{i.e. } u = f\big(x - c(u)\,t\big) \text{ implicitly.}$$

For the inviscid Burgers equation \( c(u) = u \):

$$u_t + u\,u_x = 0, \qquad u = f(x - u\,t), \qquad x = s + f(s)\,t.$$

Breaking time. \( \frac{\partial x}{\partial s} = 1 + f'(s)\,t \). The map \( s \mapsto x \) stops being invertible when this first vanishes. If \( f' < 0 \) somewhere,

$$\boxed{t^* = \frac{1}{\max_s \big(-f'(s)\big)} = -\frac{1}{\min_s f'(s)}}$$

and if \( f' \ge 0 \) everywhere the smooth solution exists for all \( t > 0 \) (the characteristics fan out).

Data \( f(x) \)SolutionBreaking
\( x \)\( u = \frac{x}{1 + t} \)never for \( t > 0 \) (rarefaction)
\( -x \)\( u = \frac{x}{t - 1} \)\( t^* = 1 \); all characteristics meet at \( (0, 1) \)
\( \frac{1}{1 + x^2} \)implicit\( t^* = \frac{8}{3\sqrt{3}} = \frac{8\sqrt{3}}{9} \approx 1.54 \)
\( -\tanh x \)implicit\( t^* = 1 \) (\( \min f' = -1 \) at \( x = 0 \))

Nonlinear Equations: Characteristic Strips

For \( F(x, y, z, p, q) = 0 \), the characteristics carry \( p \) and \( q \) as well. The characteristic strip equations are Charpit’s equations written with a parameter:

$$\dot x = F_p, \quad \dot y = F_q, \quad \dot z = pF_p + qF_q, \quad \dot p = -(F_x + pF_z), \quad \dot q = -(F_y + qF_z).$$

Initial strip. On the data curve you need \( p_0(s), q_0(s) \) as well as \( x_0, y_0, z_0 \). They are fixed by two conditions:

  1. Strip condition: \( z_0'(s) = p_0(s)\,x_0'(s) + q_0(s)\,y_0'(s) \) (the surface contains the curve).
  2. The PDE itself: \( F(x_0, y_0, z_0, p_0, q_0) = 0 \).

These may have several solutions \( (p_0, q_0) \): a nonlinear Cauchy problem can have more than one solution, one for each consistent choice. The transversality condition becomes \( F_q\,x_0' - F_p\,y_0' \ne 0 \) on the initial strip.

Memory Tricks

Don’t Confuse

ThisvsThat
general solution (arbitrary function)Cauchy problem solution (function fixed by the data)
\( J = 0 \) at isolated points\( J \equiv 0 \) along the whole data curve
local existence near \( \Gamma \)global existence (can fail: closed characteristics, breaking)
linear: projected characteristics independent of \( u \)quasilinear: characteristic speed depends on \( u \)
breaking time \( -\frac{1}{\min f'} \)rarefaction when \( f' \ge 0 \): no breaking
quasilinear Cauchy problem: unique when transversalnonlinear: one solution per admissible initial strip \( (p_0, q_0) \)

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 of the Mathematical Sciences paper, one or more options may be correct.

Level 1: Foundation

  1. Write the transversality condition for \( a p + b q = c \) with data on \( (x_0(s), y_0(s), z_0(s)) \).
  2. Solve \( u_t + 3u_x = 0 \), \( u(x, 0) = \sin x \).
  3. Along which curves is the solution of \( u_t + u u_x = 0 \) constant?
Answers
  1. \( a\,y_0'(s) - b\,x_0'(s) \ne 0 \) along the curve, with \( a, b \) evaluated on the data.
  2. \( u = \sin(x - 3t) \).
  3. Along the characteristic lines \( x = s + f(s)\,t \); on each, \( u = f(s) \).

Level 2: Part B Style

Practice Q1. The solution of \( u_x + 2u_y = 0 \) with \( u(x, 0) = e^{x} \) is:

(a) \( e^{x - 2y} \) (b) \( e^{x - y/2} \) (c) \( e^{2x - y} \) (d) \( e^{x + y/2} \)

Answer: (b)

General solution \( u = F(2x - y) \). At \( y = 0 \): \( F(2x) = e^x \), so \( F(w) = e^{w/2} \) and \( u = e^{(2x - y)/2} = e^{x - y/2} \). Check: \( u_x = u \), \( u_y = -\frac{u}{2} \), \( u_x + 2u_y = 0 \).

Practice Q2. For \( u_t + u u_x = 0 \), \( u(x, 0) = \frac{1}{1 + x^2} \), the time at which the classical solution first breaks down is:

(a) \( 1 \) (b) \( \frac{8\sqrt{3}}{9} \) (c) \( \frac{3\sqrt{3}}{8} \) (d) it never breaks

Answer: (b)

\( f'(x) = -\frac{2x}{(1 + x^2)^2} \). Maximise \( g(x) = \frac{2x}{(1 + x^2)^2} \): \( g'(x) = 0 \) gives \( 1 + x^2 = 4x^2 \), \( x = \frac{1}{\sqrt{3}} \), where \( g = \frac{2/\sqrt{3}}{16/9} = \frac{3\sqrt{3}}{8} \). So \( t^* = \frac{8}{3\sqrt{3}} = \frac{8\sqrt{3}}{9} \approx 1.54 \). Option (c) is \( \max(-f') \) itself, not its reciprocal.

Practice Q3. The Cauchy problem \( u_x + u_y = 1 \), \( u(s, s) = s + 2 \) for all \( s \in \mathbb{R} \), has:

(a) a unique solution (b) exactly two solutions (c) infinitely many solutions (d) no solution

Answer: (c)

General solution \( u = x + F(x - y) \). On \( y = x \): \( u = x + F(0) \), so the data require \( F(0) = 2 \). Every \( F \) with \( F(0) = 2 \) works. The data curve is a characteristic and the data are consistent with \( \frac{du}{dt} = 1 \) along it.

Practice Q4. The solution of \( p + q = z \) with \( z(x, 0) = \cos x \) has \( z(\pi, \pi) \) equal to:

(a) \( -1 \) (b) \( e^{\pi} \) (c) \( -e^{\pi} \) (d) \( 1 \)

Answer: (b)

\( z = e^{y}\cos(x - y) \); at \( (\pi, \pi) \): \( e^{\pi}\cos 0 = e^{\pi} \approx 23.14 \).

Level 3: Part C Style

Practice Q5. Consider \( y\,u_x - x\,u_y = 0 \) on \( \mathbb{R}^2 \). Which statements are true?

(a) \( u = (x^2 + y^2)^2 \) is a solution. (b) With \( u(x, 0) = x \) for \( x > 0 \), the solution on \( \mathbb{R}^2 \setminus \{(0, 0)\} \) is \( u = \sqrt{x^2 + y^2} \). (c) There is a \( C^1 \) solution on \( \mathbb{R}^2 \) with \( u(x, 0) = x \) for every real \( x \). (d) For data on the \( x \)-axis, the transversality condition fails only at the origin.

Answer: (a), (b) and (d)
  • (a) True. Any \( F(x^2 + y^2) \) works: \( u_x = 2xF' \), \( u_y = 2yF' \), \( yu_x - xu_y = 0 \).
  • (b) True. \( u = F(x^2 + y^2) \) with \( F(x^2) = x \) for \( x > 0 \) gives \( F(r^2) = r \).
  • (c) False. \( u \) is constant on the circle through \( (1, 0) \) and \( (-1, 0) \), but the data give 1 and \( -1 \).
  • (d) True. \( J = a y_0' - b x_0' = 0 \cdot 0 - (-s) \cdot 1 = s \), zero only at \( s = 0 \).

Practice Q6. Consider \( u_t + u u_x = 0 \), \( u(x, 0) = -x \). Which statements are true?

(a) \( u(x, t) = \frac{x}{t - 1} \) for \( 0 \le t < 1 \). (b) All characteristics pass through the point \( (x, t) = (0, 1) \). (c) \( u \) is constant along each characteristic. (d) The characteristic through \( (s, 0) \) is \( x = s(1 + t) \).

Answer: (a), (b) and (c)
  • (a) True. \( u = f(x - ut) = -(x - ut) \) gives \( u = \frac{x}{t - 1} \); at \( t = 0 \) it is \( -x \).
  • (b) True. The characteristics are \( x = s + f(s)t = s(1 - t) \), which vanish at \( t = 1 \) for every \( s \).
  • (c) True. \( \frac{du}{dt} = 0 \) along characteristics of any equation \( u_t + c(u)u_x = 0 \).
  • (d) False. The speed is \( f(s) = -s \), so \( x = s(1 - t) \).

Practice Q7. Consider the nonlinear Cauchy problem \( pq = z \), \( z(0, y) = y^2 \), solved by characteristic strips. Which statements are true?

(a) On the data curve, \( q_0 = 2s \) and \( p_0 = \frac{s}{2} \) (for \( s \ne 0 \)). (b) The solution is \( z = \frac{(x + 4y)^2}{16} \). (c) The solution satisfies \( z \ge 0 \) everywhere. (d) \( z(4, -1) = 1 \).

Answer: (a), (b) and (c)
  • (a) True. Strip condition \( z_0' = q_0 y_0' \) gives \( q_0 = 2s \); the PDE gives \( p_0 = \frac{s^2}{2s} \).
  • (b) True. Derived in worked example 8 and checked by substitution.
  • (c) True. It is a square divided by 16.
  • (d) False. \( x + 4y = 4 - 4 = 0 \), so \( z(4, -1) = 0 \).

Quick Revision Table

IdeaFact
Cauchy problemintegral surface of \( ap + bq = c \) through a given curve \( \Gamma \)
Characteristic ODEs\( \dot x = a \), \( \dot y = b \), \( \dot z = c \), start on \( \Gamma \)
Transversality\( J = a y_0' - b x_0' \ne 0 \Rightarrow \) unique local solution
\( J \equiv 0 \), data characteristicinfinitely many solutions
\( J \equiv 0 \), data not characteristicno solution
Transport\( u_t + cu_x = 0 \Rightarrow u = f(x - ct) \)
Burgers\( u = f(x - ut) \), \( x = s + f(s)t \)
Breaking time\( t^* = -\frac{1}{\min f'} \) if \( f' < 0 \) somewhere
Nonlinearstrip equations + strip condition \( z_0' = p_0x_0' + q_0y_0' \) + \( F = 0 \)

Key Takeaways

What’s Next