Chapter 10 · Partial Differential Equations

10.1 First-Order Equations and the Method of Characteristics

Linear, semilinear and quasilinear first-order equations $Pp+Qq=R$; characteristic curves, Lagrange's auxiliary equations and the general integral $\varphi(u,v)=0$; the Cauchy problem, the transversality condition, and how characteristics of Burgers' equation cross and break a smooth solution.

A first-order partial differential equation links an unknown function $z(x,y)$ of two variables to its two slopes $p=z_x$ and $q=z_y$. Such equations describe anything that is carried: a pollutant drifting down a river, a density of cars moving along a road, a wavefront advancing through a medium, the level sets of a quantity being rotated by a flow. The single idea that solves them is the method of characteristics. Along certain curves in the plane — the characteristics — the PDE collapses to an ordinary differential equation, so the solution can be computed by solving ODEs and then gluing the answers together. For a linear equation the characteristics are fixed in advance by the coefficients; for a quasilinear equation they depend on the solution itself, and they may cross, which is how a perfectly smooth initial profile can steepen into a shock in finite time.

GATE MA asks about this material in all three formats. MCQs and NATs ask for the solution of a linear equation through given data, its value at a point, the general solution of $a\,z_x+b\,z_y=c\,z$, the characteristics of an equation, or the time at which a solution of Burgers' equation breaks. MSQs ask which data curves give a unique solution (the transversality condition), which functions are solutions, and which statements about general integrals and characteristic curves are true. This section develops all of it with proofs of the key results; the long proofs and the nonlinear theory (Charpit's method) are in the CSIR course linked in the 📚 Go deeper box.

First-Order Equations and Their Solutions

📘 Definition 10.1 — First-order PDE, solution, integral surface

Let $D\subseteq\mathbb{R}^2$ be open. A first-order partial differential equation in two independent variables is a relation

$$F(x,y,z,p,q)=0,\qquad p=\frac{\partial z}{\partial x},\quad q=\frac{\partial z}{\partial y},$$

where $F$ genuinely depends on $(p,q)$. A solution on $D$ is a function $z\in C^1(D)$ with $F\big(x,y,z(x,y),z_x(x,y),z_y(x,y)\big)=0$ for every $(x,y)\in D$. Its graph $\{(x,y,z(x,y))\}$ is called an integral surface.

📘 Definition 10.2 — Linear, semilinear, quasilinear and nonlinear equations

The equation is

  • linear if it has the form $a(x,y)\,p+b(x,y)\,q=c(x,y)\,z+d(x,y)$;
  • semilinear if it has the form $a(x,y)\,p+b(x,y)\,q=c(x,y,z)$: the coefficients of $p,q$ depend on $(x,y)$ only;
  • quasilinear if it has the form $P(x,y,z)\,p+Q(x,y,z)\,q=R(x,y,z)$: it is linear in $p$ and $q$, but the coefficients may depend on $z$;
  • nonlinear (fully nonlinear) otherwise, e.g. $p^2+q^2=1$ or $pq=z$.

Every linear equation is semilinear and every semilinear equation is quasilinear. Throughout this section the coefficients are $C^1$ on an open set, and $(a,b)\ne(0,0)$, respectively $(P,Q)\ne(0,0)$, wherever we use them.

The quasilinear equation has a geometric meaning that drives everything below. Since $(p,q,-1)$ is a normal vector to the surface $z=z(x,y)$, the equation $Pp+Qq=R$ says

$$(P,Q,R)\cdot(p,q,-1)=0 ,$$

that is: at each point of an integral surface the vector $V=(P,Q,R)$ is tangent to the surface. A surface to which a given vector field is everywhere tangent should be made of curves that follow the field. Those curves are the characteristics, introduced in the next two subsections. First, a converse view: equations of this type arise when an arbitrary function is eliminated.

📐 Theorem 10.1 — Eliminating an arbitrary function gives a quasilinear equation

Let $u(x,y,z)$ and $v(x,y,z)$ be $C^1$, let $\varphi$ be a $C^1$ function of two variables with $(\varphi_u,\varphi_v)\ne(0,0)$, and let $z\in C^1(D)$ satisfy $\varphi\big(u(x,y,z),v(x,y,z)\big)=0$ identically on $D$. Then $z$ is a solution of $Pp+Qq=R$ with

$$P=\frac{\partial(u,v)}{\partial(y,z)},\qquad Q=\frac{\partial(u,v)}{\partial(z,x)},\qquad R=\frac{\partial(u,v)}{\partial(x,y)},$$

where $\dfrac{\partial(u,v)}{\partial(y,z)}=u_yv_z-u_zv_y$, and so on. Equivalently $(P,Q,R)=\nabla u\times\nabla v$.

Proof. Differentiate $\varphi(u,v)=0$ with respect to $x$ and to $y$, remembering that $z$ depends on $(x,y)$:

$$\varphi_u\,(u_x+u_zp)+\varphi_v\,(v_x+v_zp)=0,$$ $$\varphi_u\,(u_y+u_zq)+\varphi_v\,(v_y+v_zq)=0 .$$

This is a homogeneous linear system for the nonzero vector $(\varphi_u,\varphi_v)$, so its determinant vanishes:

$$(u_x+u_zp)(v_y+v_zq)-(u_y+u_zq)(v_x+v_zp)=0 .$$

On expanding, the terms in $pq$ cancel and what remains is

$$-p\,(u_yv_z-u_zv_y)-q\,(u_zv_x-u_xv_z)$$ $$+\,(u_xv_y-u_yv_x)=0,$$

which is $Pp+Qq=R$. ∎

By contrast, eliminating two arbitrary constants generally produces a nonlinear equation: the spheres $z^2=1-(x-a)^2-(y-b)^2$ give $z^2(p^2+q^2+1)=1$. GATE's syllabus concentrates on the linear and quasilinear cases; nonlinear equations need Charpit's method (see Go deeper).

✏️ Example 1 — Classify, and eliminate arbitrary functions
(a) Classify $x^2p+e^yq=z\sin x$, $\ yp-xq=z^2$, $\ zp+q=0$ and $p^2+q=0$. (b) Eliminate $f$ from $z=f(x^2-y^2)$. (c) Find the equation whose solutions are given by $\varphi(x+y+z,\ xyz)=0$ with $\varphi$ arbitrary.
(a) $x^2p+e^yq=(\sin x)\,z$ is linear. $yp-xq=z^2$ is semilinear (the coefficients of $p,q$ depend on $(x,y)$, the right side on $z$), not linear. $zp+q=0$ is quasilinear but not semilinear: the coefficient of $p$ is $z$. $p^2+q=0$ is nonlinear.
(b) $p=2xf'$ and $q=-2yf'$, so $yp+xq=2xyf'-2xyf'=0$. The equation is $yp+xq=0$, linear and homogeneous.
(c) Jacobians With $u=x+y+z$, $v=xyz$: $\nabla u=(1,1,1)$, $\nabla v=(yz,zx,xy)$, and $\nabla u\times\nabla v=\big(xy-zx,\ yz-xy,\ zx-yz\big)$.
(c) Equation By Theorem 10.1, $$x(y-z)\,p+y(z-x)\,q=z(x-y).$$
Check For the particular choice $\varphi(u,v)=u-1$, i.e. the plane $z=1-x-y$: $p=q=-1$, and the left side is $-x(y-z)-y(z-x)=z(x-y)$. ✓

Linear Equations: Characteristics Turn the PDE into ODEs

Consider the linear equation

$$a(x,y)\,z_x+b(x,y)\,z_y=c(x,y)\,z+d(x,y)\qquad\text{on an open set }D .$$

The left side is the derivative of $z$ in the direction of the vector field $(a,b)$, multiplied by its length. So along a curve whose velocity is $(a,b)$ the left side is just $\frac{d}{dt}z$ — an ordinary derivative.

📘 Definition 10.3 — Characteristic curves of a linear equation
The characteristic curves (characteristics) of the linear equation are the solution curves $t\mapsto(x(t),y(t))$ of the autonomous system $$\frac{dx}{dt}=a(x,y),\qquad\frac{dy}{dt}=b(x,y);$$ where $a\ne0$ they are the graphs of the solutions of $\dfrac{dy}{dx}=\dfrac{b(x,y)}{a(x,y)}$. A $C^1$ function $\xi(x,y)$ with $\nabla\xi\ne0$ that is constant along every characteristic is a first integral; then $a\,\xi_x+b\,\xi_y=0$, and the characteristics are the level curves $\xi=\text{const}$.

Because $a,b\in C^1$, the Picard–Lindelöf theorem (§6.1, in its form for systems) gives exactly one characteristic through each point of $D$; in particular two characteristics of a linear equation never cross.

📐 Theorem 10.2 — Along a characteristic the equation is an ODE

Let $z\in C^1(D)$ solve the linear equation and let $(x(t),y(t))$ be a characteristic in $D$. Then $Z(t)=z\big(x(t),y(t)\big)$ satisfies the linear ODE

$$\frac{dZ}{dt}=c\big(x(t),y(t)\big)\,Z+d\big(x(t),y(t)\big).$$

In particular, solutions of the homogeneous equation $a\,z_x+b\,z_y=0$ are constant along characteristics, and if $\xi$ is a first integral then $z=f(\xi(x,y))$ is a solution for every $C^1$ function $f$.

Proof. By the chain rule, $Z'(t)=z_x\,x'(t)+z_y\,y'(t)$, which is $a\,z_x+b\,z_y=cZ+d$, all evaluated at $(x(t),y(t))$. For $z=f(\xi)$: $a\,z_x+b\,z_y=f'(\xi)\,(a\,\xi_x+b\,\xi_y)=0$. ∎

The cleanest way to use this is to take the characteristics as one family of coordinate lines.

📐 Theorem 10.3 — Characteristic coordinates; constant coefficients

(1) Suppose $a\ne0$ on $D$ and $\xi(x,y)$ is a first integral with $\xi_y\ne0$ on $D$. In the coordinates $\xi=\xi(x,y)$, $\eta=x$ the linear equation becomes, for $z(x,y)=w(\xi,\eta)$,

$$a\,w_\eta=c\,w+d\qquad(\text{coefficients expressed in }\xi,\eta),$$

a linear first-order ODE in $\eta$ in which $\xi$ is only a parameter. Its general solution contains an arbitrary function of $\xi$.

(2) For constants $a\ne0$, $b$, $c$, the general solution of $a\,z_x+b\,z_y=c\,z$ is $$z=e^{cx/a}\,f(bx-ay),\qquad f\in C^1\text{ arbitrary.}$$ In particular $a\,z_x+b\,z_y=0$ has general solution $f(bx-ay)$, and the transport equation $z_t+c\,z_x=0$ has general solution $z=f(x-ct)$: the initial profile moves to the right with speed $c$ without changing shape.

Proof. (1) The Jacobian of $(x,y)\mapsto(\xi,\eta)$ is $\xi_x\cdot0-\xi_y\cdot1=-\xi_y\ne0$, so this is a local change of variables (inverse function theorem, §1.1). By the chain rule $z_x=w_\xi\,\xi_x+w_\eta$ and $z_y=w_\xi\,\xi_y$, hence

$$a\,z_x+b\,z_y=w_\xi\,(a\,\xi_x+b\,\xi_y)+a\,w_\eta=a\,w_\eta ,$$

because $\xi$ is a first integral. (2) Here $dy/dx=b/a$ and $\xi=bx-ay$ is a first integral with $\xi_y=-a\ne0$. The equation becomes $a\,w_\eta=c\,w$, whose solutions are $w=e^{c\eta/a}f(\xi)$ with $f$ depending on $\xi$ only (the "constant" of integration may depend on the parameter $\xi$). Returning to $x,y$ gives the formula. Conversely, every function of this form is a solution, by direct substitution. ∎

✏️ Example 2 — Constant coefficients and the transport equation
(a) Solve $2z_x+3z_y=z$ with $z(x,0)=\sin x$. (b) A contaminant in a channel obeys $z_t+2z_x=-z$ ($x$ in km, $t$ in hours): it moves at $2$ km/h and decays at rate $1$ per hour. If $z(x,0)=\dfrac{1}{1+x^2}$, find $z(x,t)$ and $z(3,1)$.
(a) General Theorem 10.3(2) with $a=2$, $b=3$, $c=1$: $z=e^{x/2}f(3x-2y)$.
(a) Data $z(x,0)=e^{x/2}f(3x)=\sin x$, so $f(s)=e^{-s/6}\sin\tfrac s3$. Substituting $s=3x-2y$: $$z=e^{x/2}\,e^{-(3x-2y)/6}\sin\!\left(x-\tfrac{2y}{3}\right)=e^{y/3}\sin\!\left(x-\tfrac{2y}3\right).$$
(a) Check $z_x=e^{y/3}\cos(\cdot)$ and $z_y=\tfrac13e^{y/3}\sin(\cdot)-\tfrac23e^{y/3}\cos(\cdot)$, so $2z_x+3z_y=e^{y/3}\sin(\cdot)=z$; and $z(x,0)=\sin x$. ✓
(b) Characteristics $x-2t=\text{const}$; along each, $\frac{dZ}{dt}=-Z$, so $Z$ is multiplied by $e^{-t}$. Hence $$z(x,t)=\frac{e^{-t}}{1+(x-2t)^2}.$$ Then $z(3,1)=\dfrac{e^{-1}}{2}\approx0.1839$: the peak, which started at $x=0$, is at $x=2$ after one hour, and has decayed to $e^{-1}$ of its size.
✏️ Example 3 — Variable coefficients
(a) Solve $x\,z_x-y\,z_y=z$ on $x\gt0$ with $z(1,y)=y^2$, and find $z\left(2,\tfrac12\right)$. (b) Show that every $C^1$ solution of $y\,z_x-x\,z_y=0$ on $\mathbb{R}^2$ is constant on each circle centred at the origin. Is there a $C^1$ solution on $\mathbb{R}^2$ with $z(x,0)=x$ for all $x\in\mathbb{R}$? Near the point $(1,0)$?
(a) Characteristics $\dfrac{dy}{dx}=-\dfrac yx$ gives $xy=\text{const}$: the first integral is $\xi=xy$ (hyperbolas).
(a) Along them With $\eta=x$, Theorem 10.3(1) gives $x\,w_x=w$, so $w=x\,g(\xi)$: the general solution is $z=x\,g(xy)$.
(a) Data $z(1,y)=g(y)=y^2$, so $z=x(xy)^2=x^3y^2$, and $z\left(2,\tfrac12\right)=8\cdot\tfrac14=2$. Check: $x\cdot3x^2y^2-y\cdot2x^3y=x^3y^2=z$. ✓
(b) Circles The characteristics of $(a,b)=(y,-x)$ solve $x'=y$, $y'=-x$: they are the circles $x^2+y^2=\text{const}$, traversed clockwise. By Theorem 10.2, $z$ is constant on each one.
(b) Data The circle of radius $1$ meets the $x$-axis at $(1,0)$ and $(-1,0)$, where the data are $1$ and $-1$. A global solution would have to take both values on one circle: impossible. Near $(1,0)$, however, $z=\sqrt{x^2+y^2}$ is a $C^1$ solution with $z(x,0)=x$ for $x\gt0$. The data line meets each characteristic twice — the transversality question of the last subsection.

Quasilinear Equations: Lagrange's Method

For $Pp+Qq=R$ with coefficients depending on $z$ we cannot draw the characteristics in the plane before knowing $z$. Instead we draw them in $(x,y,z)$-space, where the vector field $V=(P,Q,R)$ is known.

📘 Definition 10.4 — Characteristics, Lagrange's auxiliary equations, first integrals

Let $P,Q,R$ be $C^1$ on an open set $\Omega\subseteq\mathbb{R}^3$, not all zero. The characteristic curves of $Pp+Qq=R$ are the solutions of

$$\frac{dx}{dt}=P(x,y,z),\qquad\frac{dy}{dt}=Q(x,y,z),\qquad\frac{dz}{dt}=R(x,y,z),$$

written without the parameter as Lagrange's auxiliary (subsidiary) equations

$$\frac{dx}{P}=\frac{dy}{Q}=\frac{dz}{R}.$$

Their projections onto the $xy$-plane are the base (projected) characteristics. A $C^1$ function $u(x,y,z)$ with $\nabla u\ne0$ that is constant along every characteristic is a first integral; equivalently $Pu_x+Qu_y+Ru_z=0$ on $\Omega$. Two first integrals $u,v$ are independent if $\nabla u\times\nabla v\ne0$.

📐 Theorem 10.4 — Integral surfaces are unions of characteristics
Let $P,Q,R\in C^1(\Omega)$ and let $z\in C^1(D)$ have its graph $S$ in $\Omega$.
  1. If $z$ solves $Pp+Qq=R$, then the characteristic through any point of $S$ stays in $S$ (for as long as its projection stays in $D$).
  2. Conversely, if through every point of $S$ there passes a characteristic lying in $S$, then $z$ solves $Pp+Qq=R$.

Proof. (1) Let $(x_0,y_0,z_0)\in S$. The functions $\tilde P(x,y)=P(x,y,z(x,y))$ and $\tilde Q(x,y)=Q(x,y,z(x,y))$ are continuous (and $C^1$ if $z$ is $C^2$). Solve $x'=\tilde P$, $y'=\tilde Q$ with $(x(0),y(0))=(x_0,y_0)$ — a solution exists by Peano's theorem (§6.1), and by Picard–Lindelöf if $z\in C^2$ — and put $Z(t)=z(x(t),y(t))$. Then

$$Z'=z_x\,x'+z_y\,y'=pP+qQ=R\big(x(t),y(t),Z(t)\big),$$

so $(x(t),y(t),Z(t))$ solves the characteristic system with the initial point $(x_0,y_0,z_0)$. Since $P,Q,R\in C^1$ that system has only one such solution (Picard–Lindelöf), so this curve is the characteristic, and it lies on $S$ by construction. (2) If the characteristic $\gamma$ through a point lies in $S$, its velocity $(P,Q,R)$ is tangent to $S$, hence orthogonal to the normal $(p,q,-1)$: $Pp+Qq-R=0$ at that point. ∎

📐 Theorem 10.5 — Lagrange's theorem: the general integral $\varphi(u,v)=0$

Let $u=c_1$ and $v=c_2$ be two independent first integrals of $\dfrac{dx}{P}=\dfrac{dy}{Q}=\dfrac{dz}{R}$ on $\Omega$. If $\varphi$ is $C^1$ with $(\varphi_u,\varphi_v)\ne(0,0)$ and $z\in C^1(D)$ satisfies $\varphi\big(u,v\big)=0$ (with $u,v$ evaluated at $(x,y,z(x,y))$), then $z$ solves $Pp+Qq=R$.

Conversely, near each point of an open dense subset of its domain, every $C^1$ solution satisfies such a relation (in fact $v=g(u)$ or $u=g(v)$). The relation $\varphi(u,v)=0$ with $\varphi$ arbitrary is the general integral (general solution).

Proof of the first part. Since $u$ and $v$ are first integrals, the vector $(P,Q,R)$ is orthogonal to both $\nabla u$ and $\nabla v$. These two gradients are independent, so the vectors orthogonal to both form the line spanned by $\nabla u\times\nabla v\ne0$; hence $(P,Q,R)=\mu\,\nabla u\times\nabla v$ for some function $\mu\ne0$. By Theorem 10.1, $z$ satisfies $(\nabla u\times\nabla v)\cdot(p,q,-1)=0$; multiplying by $\mu$ gives $Pp+Qq-R=0$. Converse (idea). Along a solution, $U(x,y)=u(x,y,z(x,y))$ and $W(x,y)=v(x,y,z(x,y))$ satisfy $\frac{\partial(U,W)}{\partial(x,y)}=-\mu^{-1}(Pp+Qq-R)=0$ by the same computation; two functions with vanishing Jacobian are locally functionally dependent where one of them has a nonzero gradient. ∎

🧭 Problem-solving strategy — finding two independent first integrals
  1. Separate a pair. If one of the three ratios, say $\frac{dx}{P}=\frac{dy}{Q}$, involves only two variables after cancelling, integrate it as an ODE.
  2. Use the first integral to eliminate. Substitute $u=c_1$ into another pair to reduce it to an ODE in two variables; integrate; then put $c_1=u$ back.
  3. Multipliers. For functions $l,m,n$, every ratio equals $\dfrac{l\,dx+m\,dy+n\,dz}{lP+mQ+nR}$. If $lP+mQ+nR=0$, then $l\,dx+m\,dy+n\,dz=0$ along characteristics; when this is an exact differential $du$, $u$ is a first integral. Try $(1,1,1)$, $(x,y,z)$, $\left(\frac1x,\frac1y,\frac1z\right)$ and combinations suggested by the coefficients.
  4. Write $\varphi(u,v)=0$, or $v=g(u)$, and check that $u$ and $v$ are independent.
✏️ Example 4 — Two general integrals
Find the general integral of (a) $x\,p+y\,q=z+xy$ and (b) $x(2y-3z)\,p+y(3z-x)\,q=z(x-2y)$.
(a) First $\dfrac{dx}{x}=\dfrac{dy}{y}$ gives $u=\dfrac yx=c_1$.
(a) Second From $\dfrac{dx}{x}=\dfrac{dz}{z+xy}$ with $y=c_1x$: $\dfrac{dz}{dx}-\dfrac zx=c_1x$, i.e. $\left(\dfrac zx\right)'=c_1$, so $\dfrac zx=c_1x+c_2$. With $c_1x=y$: $v=\dfrac zx-y=c_2$.
(a) Answer $\varphi\!\left(\dfrac yx,\ \dfrac{z-xy}{x}\right)=0$, or explicitly $z=xy+x\,f\!\left(\dfrac yx\right)$. Check: $xz_x+yz_y=2xy+xf=z+xy$ (the $f'$ terms cancel). ✓
(b) Multipliers $(1,2,3)$: $x(2y-3z)+2y(3z-x)+3z(x-2y)=0$, so $dx+2\,dy+3\,dz=0$ and $u=x+2y+3z$.
(b) Second $\left(\frac1x,\frac1y,\frac1z\right)$: $(2y-3z)+(3z-x)+(x-2y)=0$, so $\frac{dx}x+\frac{dy}y+\frac{dz}z=0$ and $v=xyz$.
(b) Answer $\varphi(x+2y+3z,\ xyz)=0$. Independence: $\nabla u\times\nabla v=\big(x(2y-3z),\,y(3z-x),\,z(x-2y)\big)$, which is exactly $(P,Q,R)$, so $u$ and $v$ are independent wherever $(P,Q,R)\ne0$. (Compare Example 1(c): the same mechanism with $u=x+y+z$.)

The Cauchy Problem and Transversality

The general integral contains an arbitrary function; a single solution is picked out by prescribing $z$ along a curve. Which curves are allowed?

📘 Definition 10.5 — Cauchy problem; transversality condition

Let $\Gamma:\ s\mapsto\big(x_0(s),y_0(s),z_0(s)\big)$, $s\in I$, be a $C^1$ curve whose projection $\Gamma_0:\ s\mapsto(x_0(s),y_0(s))$ has $(x_0',y_0')\ne(0,0)$. The Cauchy problem for $Pp+Qq=R$ asks for a solution $z(x,y)$ near $\Gamma_0$ with

$$z\big(x_0(s),y_0(s)\big)=z_0(s)\qquad(s\in I).$$

The transversality condition at $s$ is

$$J(s)=x_0'(s)\,Q\big(\Gamma(s)\big)-y_0'(s)\,P\big(\Gamma(s)\big)\ne0 ,$$

i.e. the data curve $\Gamma_0$ is not tangent to the base characteristic direction $(P,Q)$ at that point. If $J\ne0$ on $I$, $\Gamma$ is non-characteristic.

📐 Theorem 10.6 — Local existence and uniqueness for the Cauchy problem
Let $P,Q,R\in C^1(\Omega)$, let $\Gamma\subseteq\Omega$ be as in Definition 10.5, and suppose $J(s^*)\ne0$. Then there is a neighbourhood $N$ of $\big(x_0(s^*),y_0(s^*)\big)$ and a solution $z\in C^1(N)$ of $Pp+Qq=R$ with $z(x_0(s),y_0(s))=z_0(s)$ for all $s$ near $s^*$, and any other $C^1$ solution with these data coincides with $z$ near that point.

Proof. Step 1 — emit characteristics from the data. For each $s$ let $t\mapsto\big(x(s,t),y(s,t),z(s,t)\big)$ solve $x_t=P$, $y_t=Q$, $z_t=R$ with value $\Gamma(s)$ at $t=0$. By the existence–uniqueness theorem and the differentiable dependence of solutions on initial values, these functions are $C^1$ near $(s^*,0)$.

Step 2 — invert. At $t=0$ we have $x_s=x_0'$, $y_s=y_0'$, $x_t=P$, $y_t=Q$, so

$$\frac{\partial(x,y)}{\partial(s,t)}\Big|_{(s^*,0)}=x_0'Q-y_0'P=J(s^*)\ne0 .$$

By the inverse function theorem (§1.1) the map $(s,t)\mapsto(x,y)$ is a $C^1$ diffeomorphism from a neighbourhood $U$ of $(s^*,0)$ onto a neighbourhood $N$; call its inverse $(S(x,y),T(x,y))$ and set $Z(x,y)=z\big(S(x,y),T(x,y)\big)$.

Step 3 — it is a solution. On $U$, $z(s,t)=Z(x(s,t),y(s,t))$. Differentiating in $t$ gives $R=z_t=Z_x\,P+Z_y\,Q$ at every point of the graph, and at $t=0$ the data are reproduced.

Step 4 — uniqueness. Another solution $\tilde Z$ with the same data contains, by Theorem 10.4(1), the characteristic through each data point $\Gamma(s)$ for small $|t|$. These are exactly the curves that make up the graph of $Z$, and their projections fill $N$; so $\tilde Z=Z$ near the point. ∎

📐 Theorem 10.7 — Characteristic data: no solution or infinitely many
In the setting of Definition 10.5 suppose $J\equiv0$ on $I$ and $(P,Q)\ne(0,0)$ along $\Gamma$.
  1. If the Cauchy problem has a $C^1$ solution near $\Gamma_0$, then $\Gamma$ itself is a characteristic curve.
  2. If $\Gamma$ is a characteristic curve, the Cauchy problem has infinitely many solutions near each point of $\Gamma_0$.

Proof. (1) $J=0$ means $(x_0',y_0')=\lambda(s)(P,Q)$ with $\lambda\ne0$. Differentiating $z(x_0(s),y_0(s))=z_0(s)$ gives $z_0'=p\,x_0'+q\,y_0'=\lambda(pP+qQ)=\lambda R$. So $(x_0',y_0',z_0')=\lambda(P,Q,R)$: the curve $\Gamma$ is everywhere tangent to the characteristic field, hence (by ODE uniqueness) a reparametrised characteristic. (2) Through a point of $\Gamma$ draw a short curve $\tilde\Gamma$ whose projection is transversal to $(P,Q)$, carrying arbitrary data that agree with $\Gamma$ at the common point. Theorem 10.6 gives a solution containing $\tilde\Gamma$; by Theorem 10.4 it contains the characteristic through the common point, which is $\Gamma$. Different data on $\tilde\Gamma$ give different solutions. ∎

🧭 Problem-solving strategy — solving a Cauchy problem by characteristics
  1. Parametrise the data: $x=x_0(s)$, $y=y_0(s)$, $z=z_0(s)$, and compute $J(s)=x_0'Q-y_0'P$. Where $J=0$ expect trouble.
  2. Solve $x_t=P$, $y_t=Q$, $z_t=R$ with these values at $t=0$ to get $x(s,t)$, $y(s,t)$, $z(s,t)$.
  3. Solve $x=x(s,t)$, $y=y(s,t)$ for $(s,t)$ — often $t$ is $y$ or $\log$ of something — and substitute into $z(s,t)$.
  4. Check the PDE and the data, and find where the inversion is valid: the solution lives where $\partial(x,y)/\partial(s,t)\ne0$. Where this Jacobian vanishes, base characteristics touch or cross and the solution can break down.
  5. Shortcut for linear equations: find the general solution by Theorem 10.3, then fit the arbitrary function to the data.
✏️ Example 5 — A quasilinear Cauchy problem, and where its solution lives
Solve $z\,z_x+z_y=1$ with $z(x,0)=x$. On which region is the solution defined?
Transversality $\Gamma:(s,0,s)$, so $x_0'=1$, $y_0'=0$ and $J=1\cdot Q-0\cdot P=1\ne0$ everywhere.
Characteristics $x_t=z$, $y_t=1$, $z_t=1$ with $(x,y,z)=(s,0,s)$ at $t=0$: $y=t$, $z=s+t$, and $x_t=s+t$ gives $x=s+st+\tfrac{t^2}2$.
Invert $t=y$ and $s(1+y)=x-\tfrac{y^2}2$, so $$z=s+y=\frac{x-\frac{y^2}{2}}{1+y}+y=\frac{x+y+\frac{y^2}{2}}{1+y}.$$
Check $z(x,0)=x$. ✓ Also $z_x=\dfrac1{1+y}$ and $z_y=\dfrac{1+y-z}{1+y}$, so $zz_x+z_y=\dfrac{z+1+y-z}{1+y}=1$. ✓
Region $\dfrac{\partial(x,y)}{\partial(s,t)}=\det\begin{pmatrix}1+t&s+t\\0&1\end{pmatrix}=1+t$, which vanishes at $t=-1$. The formula is valid on the half-plane $y\gt-1$: there every point lies on exactly one characteristic from the data line. As $y\to-1^+$ all the base characteristics $x=s(1+y)+\tfrac{y^2}{2}$ squeeze together, $z_x=\frac1{1+y}$ blows up, and the solution cannot be continued.

The breakdown in Example 5 is typical of quasilinear equations: the base characteristics depend on the solution and can collide. The model case is Burgers' equation, in which $y$ plays the role of time.

📐 Theorem 10.8 — Burgers' equation: implicit solution and breaking time

Let $g\in C^1(\mathbb{R})$ with $g$ and $g'$ bounded, and consider $z_y+z\,z_x=0$, $z(x,0)=g(x)$, for $y\ge0$.

  1. The characteristics are the straight lines $x=s+g(s)\,y$, on which $z=g(s)$ is constant; a classical solution satisfies the implicit equation $z=g(x-z\,y)$.
  2. If $g'\ge0$ everywhere, there is a unique $C^1$ solution for all $y\ge0$.
  3. If $g'$ takes negative values, the $C^1$ solution exists exactly for $0\le y\lt y_b$, where the breaking time is $$y_b=-\frac{1}{\inf_{s}g'(s)} ,$$ and $z_x$ becomes unbounded as $y\to y_b^-$.

More generally, for $z_y+c(z)\,z_x=0$ the lines are $x=s+c(g(s))\,y$ and $y_b=-1/\inf_s\frac{d}{ds}c(g(s))$.

Proof. (1) Here $P=z$, $Q=1$, $R=0$, so $z_t=0$, $y=t$ and $x=s+g(s)\,t$: $z$ keeps its initial value $g(s)$ and the base characteristic is a straight line of slope $1/g(s)$ in the $(x,y)$-plane. Since $s=x-zy$ on that line, $z=g(x-zy)$. (2), (3) Fix $y\ge0$ and consider $X(s)=s+g(s)y$. Then $X'(s)=1+g'(s)y$. If $y\lt y_b$ (with $y_b=+\infty$ when $g'\ge0$), then $X'(s)\ge1+y\inf g'=1-\frac{y}{y_b}\gt0$ for all $s$, so $X$ is strictly increasing with derivative bounded below by a positive constant; hence it is a $C^1$ bijection of $\mathbb{R}$ onto $\mathbb{R}$ with $C^1$ inverse. So each point $(x,y)$ lies on exactly one characteristic, $z(x,y)=g\big(X^{-1}(x)\big)$ is $C^1$, and $z_x=\dfrac{g'(s)}{1+g'(s)y}$. It is a solution by Theorem 10.4(2), and it is the only one, because by Theorem 10.4(1) every $C^1$ solution contains these lines. If $y_b\lt\infty$, choose $s_k$ with $g'(s_k)\to\inf g'$; then $1+g'(s_k)y\to1-\frac{y}{y_b}$, which tends to $0$ as $y\to y_b$, so $z_x$ is unbounded near $y_b$. For $y\gt y_b$ some $X'(s)\lt0$, so $X$ is not injective: two characteristics carrying different values of $z$ reach the same point, and no continuous solution exists. ∎

✏️ Example 6 — Two Burgers problems
For $z_y+z\,z_x=0$ find the solution and the breaking time when (a) $z(x,0)=-x$; (b) $z(x,0)=\dfrac{1}{1+x^2}$ (find only $y_b$, and the place where the profile first becomes vertical).
(a) Solve $z=g(x-zy)=-(x-zy)$, so $z(1-y)=-x$ and $$z=\frac{x}{y-1}.$$ Check: $z_y=-\dfrac{x}{(y-1)^2}$ and $zz_x=\dfrac{x}{(y-1)^2}$. ✓
(a) Breaking $g'\equiv-1$, so $y_b=1$. Indeed all the lines $x=s(1-y)$ pass through the single point $(0,1)$: the profile is a line that rotates until it is vertical at $y=1$. ($g$ is unbounded here, but the computation of Theorem 10.8 goes through.)
(b) Steepest descent $g'(s)=-\dfrac{2s}{(1+s^2)^2}$ and $g''(s)=\dfrac{2(3s^2-1)}{(1+s^2)^3}$, so $g'$ is smallest at $s=\tfrac1{\sqrt3}$, where $g'=-\dfrac{2/\sqrt3}{16/9}=-\dfrac{3\sqrt3}{8}$.
(b) Answer $$y_b=\frac{8}{3\sqrt3}\approx1.5396 .$$ The profile first becomes vertical on the characteristic from $s=\tfrac1{\sqrt3}$, at $x=s+g(s)y_b=\tfrac{1}{\sqrt3}+\tfrac34\cdot\tfrac{8}{3\sqrt3}=\sqrt3\approx1.732$: the front (right) side of the bump, which is where the fast high values overtake the slow low ones. Interactive 1 shows exactly this.
✏️ Example 7 — Prove or disprove: Cauchy problems
(a) The Cauchy problem $z_x+z_y=0$, $z(x,x)=x$ has a solution near $(0,0)$. (b) The Cauchy problem $z_x+z_y=0$, $z(x,x)=5$ has infinitely many solutions. (c) The problem $y\,z_x+x\,z_y=z$, $z(x,0)=x^3$ has a unique solution near $(1,0)$; find it. (d) If $g$ is smooth and bounded, the solution of $z_y+zz_x=0$, $z(x,0)=g(x)$ is smooth for all $y\ge0$.
(a) False The data line $y=x$ has direction $(1,1)=(P,Q)$, so $J=1\cdot1-1\cdot1=0$: it is a base characteristic, along which every solution is constant (Theorem 10.2). The data $z=x$ are not constant, so there is no solution — Theorem 10.7(1).
(b) True Now the data are constant along the characteristic, and $z=5+h(x-y)$ is a solution for every $C^1$ function $h$ with $h(0)=0$ — Theorem 10.7(2).
(c) True On $\Gamma_0=(s,0)$: $J=1\cdot Q-0\cdot P=x_0(s)=s$, nonzero at $s=1$. Characteristics: $x_t=y$, $y_t=x$, $z_t=z$ from $(s,0,s^3)$ give $x=s\cosh t$, $y=s\sinh t$, $z=s^3e^t$. Since $x+y=se^t$ and $x^2-y^2=s^2$, $$z=s^2\cdot se^t=(x^2-y^2)(x+y).$$ One checks $yz_x+xz_y=z$ directly; by Theorem 10.6 this is the only solution near $(1,0)$. Note what the data do not decide: the polynomial solves the equation on all of $\mathbb{R}^2$, but the characteristics from the data line fill only $|y|\lt|x|$. In $|y|\gt|x|$ the hyperbolas $x^2-y^2=c\lt0$ never meet the $x$-axis, and there the solution can be changed: $(x+y)\big(x^2-y^2+\phi(x^2-y^2)\big)$ is another solution for any $C^1$ function $\phi$ vanishing on $[0,\infty)$.
(d) False $g=\dfrac1{1+x^2}$ is smooth and bounded, but by Example 6(b) the solution breaks at $y_b\approx1.54$. The statement is true exactly when $g'\ge0$ (Theorem 10.8).
⚠️ Counterexamples and pitfalls
  1. Quasilinear is not linear. $zz_x+z_y=0$ has the solution $z=\frac{x}{1+y}$, but $2z$ is not a solution: superposition holds only for linear homogeneous equations.
  2. Semilinear is not linear either. $z_x+z_y=z^2$ has base characteristics fixed in advance (lines $x-y=c$), yet its solutions $z=\frac{1}{h(x-y)-y}$ can blow up along a characteristic.
  3. "$dx/P=dy/Q$" first, always? Only if that pair involves just $x$ and $y$; for $zp+q=1$ the pair $\frac{dx}{z}=\frac{dy}1$ contains $z$, and one must first find $z-y=c_1$ from $\frac{dy}{1}=\frac{dz}{1}$.
  4. Two first integrals must be independent. $u=\frac yx$ and $u^2$ are both first integrals of $xp+yq=z$, but $\varphi\big(\frac yx,\frac{y^2}{x^2}\big)=0$ says nothing about $z$.
  5. Data on a characteristic. For $z_x+z_y=0$ the data $z(x,x)=x$ admit no solution and $z(x,x)=5$ admit infinitely many (Example 7). Check $J=x_0'Q-y_0'P$ first.
  6. Transversality is local. For $yz_x-xz_y=0$ the $x$-axis is transversal at every point except the origin, yet there is no global solution with $z(x,0)=x$: each circle meets the axis twice (Example 3).
  7. A solution through every point of a curve need not extend. In Example 5 the solution exists only on $y\gt-1$, although the data and the coefficients are smooth everywhere.
  8. Smooth data can break. Burgers' equation with $g=\frac{1}{1+x^2}$ breaks at $y=\frac{8}{3\sqrt3}$; with $g=\tan^{-1}x$ (increasing) it never breaks.
  9. Only negative slopes cause breaking. $g=\tan^{-1}x$ has slope $1$ at the origin and never breaks; $g=-\tan^{-1}x$ has the same steepness but breaks at $y_b=-1/\min g'=1$. For a bump such as $e^{-x^2}$ only the front side, where $g'\lt0$, steepens.
  10. Sign of the transversality determinant. $J=x_0'Q-y_0'P$ compares the data direction with $(P,Q)$, not with $(P,Q,R)$: for $z_x+z_y=z$ the line $y=x$ is characteristic whatever $R$ is.
📚 Go deeper — full proofs in the CSIR NET course

Two Labs

Interactive 1 — Characteristics Carry the Data

Left: the base characteristics in the $(x,y)$-plane, starting from the data line $y=0$ (here $y$ plays the role of time). Right: the profile $z(x,y)$ at the current level $y$ (dashed: the data $z(x,0)=e^{-x^2}$). The dots ride along the characteristics and carry their values of $z$ to the right panel. Press Play or drag the slider. For Burgers' equation, watch the lines cross and the profile fold over at the breaking time.

level y—
characteristics—
min ∂x/∂s—
breaking level—
Interactive 2 — Transversality: Where Can the Data Live?

The equation is $\cos\alpha\,z_x+\sin\alpha\,z_y=0$: its characteristics are the straight lines in direction $(\cos\alpha,\sin\alpha)$, and $z$ is constant on each. Data $z=x$ are given on the thick curve $\Gamma_0$. Each line through a data point is coloured by the value it carries (blue negative, red positive). Amber marks points where $J=0$ (the curve is tangent to the lines) and dashed amber lines meet $\Gamma_0$ twice with conflicting data. Change $\alpha$ or press Play to rotate the direction.

direction (a, b)—
J(s) range—
points with J = 0—
conflicting lines—

Exercises

I
Types of equations; eliminating arbitrary functions.
  • 1 Classify as linear, semilinear, quasilinear or nonlinear: (a) $e^xp+xyq=z+\cos y$; (b) $p+q=\sin z$; (c) $(x+z)p+yq=0$; (d) $pq=x$.
  • 2 Eliminate $f$ from $z=f(x+y^2)$.
  • 3 Eliminate the constants $a,b$ from $z=(x-a)^2+(y-b)^2$. Is the resulting equation quasilinear?
  • 4 Eliminate $f$ from $z=e^{2y}f(x-3y)$.
  • 5 Use Theorem 10.1 to find the equation whose general integral is $\varphi\big(x^2+y^2+z^2,\ x+y\big)=0$.
  • 6 Show that eliminating $f$ from $z=f(xy)$ gives $xp-yq=0$, and check that $z=\sin(xy)+(xy)^3$ satisfies it.
  • 7 Prove that the solutions of a linear homogeneous equation $ap+bq=cz$ form a vector space, and show by an example that the solutions of $zp+q=0$ do not.
1 (a) Linear. (b) Semilinear (not linear: $\sin z$). (c) Quasilinear (coefficient $x+z$ of $p$ depends on $z$). (d) Nonlinear.
2 $p=f'$ and $q=2yf'$, so $q=2yp$, i.e. $2yp-q=0$.
3 $p=2(x-a)$, $q=2(y-b)$, so $(x-a)^2+(y-b)^2=\frac{p^2+q^2}{4}$ and $p^2+q^2=4z$. It is nonlinear — eliminating constants usually is.
4 $p=e^{2y}f'$ and $q=2e^{2y}f-3e^{2y}f'=2z-3p$. So $3p+q=2z$ (linear, constant coefficients; compare Theorem 10.3(2)).
5 $\nabla u=(2x,2y,2z)$, $\nabla v=(1,1,0)$: $\nabla u\times\nabla v=(-2z,\ 2z,\ 2x-2y)$. So $-2zp+2zq=2(x-y)$, i.e. $z(q-p)=x-y$.
6 $p=yf'$, $q=xf'$, so $xp-yq=0$. For $z=\sin(xy)+(xy)^3=f(xy)$ with $f(w)=\sin w+w^3$ it holds automatically; directly, $xp=xy\left(\cos(xy)+3x^2y^2\right)=yq$. ✓
7 The map $z\mapsto az_x+bz_y-cz$ is linear, so its kernel is a vector space. For $zp+q=0$: $z_1=\frac{x}{1+y}$ is a solution ($z_1p+q=\frac{x}{(1+y)^2}-\frac{x}{(1+y)^2}=0$), but $2z_1$ gives $\frac{4x}{(1+y)^2}-\frac{2x}{(1+y)^2}\ne0$.
II
Linear equations by characteristics.
  • 1 Find the general solution of $4z_x-3z_y=0$ and the solution with $z(x,0)=x^2$.
  • 2 Solve $z_x+z_y=2z$ with $z(x,0)=\cos x$.
  • 3 Solve $z_t+3z_x=0$ with $z(x,0)=e^{-x^2}$. When is the peak at $x=12$? Find $z(10,3)$.
  • 4 Find the general solution of $xz_x+yz_y=3z$ on $x\gt0$, and the solution with $z(1,y)=1+y^2$.
  • 5 Find the characteristics of $yz_x+xz_y=0$ and the solution in the region $|y|\lt x$ with $z(x,0)=e^x$ for $x\gt0$.
  • 6 Find the general solution of $z_x+2x\,z_y=y$.
  • 7 Modelling. A dissolved pollutant in a river moves with the water at $0.5$ m/s and decays at rate $0.01$ per second: $c_t+0.5\,c_x=-0.01\,c$. A slug with peak concentration $10$ mg/L is released at $x=0$, $t=0$. How far downstream is the peak when it has dropped to $5$ mg/L?
  • 8 Let $a,b$ be constants, not both $0$. Prove that $az_x+bz_y=0$ with $z$ given on the line $bx-ay=0$ has no solution unless the data are constant, and infinitely many if they are.
1 Theorem 10.3(2) with $a=4$, $b=-3$: $z=f(-3x-4y)$, equivalently $z=F(3x+4y)$. $F(3x)=x^2$ gives $F(w)=\frac{w^2}9$, so $z=\frac{(3x+4y)^2}{9}$.
2 $z=e^{2x}f(x-y)$; $e^{2x}f(x)=\cos x$ gives $f(w)=e^{-2w}\cos w$ and $z=e^{2x}e^{-2(x-y)}\cos(x-y)=e^{2y}\cos(x-y)$.
3 $z=e^{-(x-3t)^2}$. The peak $x=3t$ is at $12$ when $t=4$; $z(10,3)=e^{-1}\approx0.3679$.
4 Characteristics $\frac yx=c$; along them $xw_x=3w$, so $z=x^3f\!\left(\frac yx\right)$ (the homogeneous functions of degree 3). $f(y)=1+y^2$ gives $z=x^3+xy^2$.
5 $\frac{dy}{dx}=\frac xy$: $y^2-x^2=c$ (hyperbolas, and the lines $y=\pm x$). $z=f(x^2-y^2)$; $f(x^2)=e^x$ for $x\gt0$, so $z=e^{\sqrt{x^2-y^2}}$ for $|y|\lt x$.
6 Characteristics $y-x^2=c$. Along one, $\frac{dz}{dx}=y=x^2+c$, so $z=\frac{x^3}3+cx+g(c)$. With $c=y-x^2$: $$z=xy-\tfrac23x^3+g(y-x^2).$$ Check: $z_x+2xz_y=(y-2x^2-2xg')+2x(x+g')=y$. ✓
7 $c(x,t)=e^{-0.01t}c_0(x-0.5t)$. The peak value is $10e^{-0.01t}=5$ at $t=100\log2\approx69.3$ s, when it is at $x=0.5t\approx34.7$ m.
8 The line $bx-ay=0$ has direction $(a,b)$, so it is itself a characteristic, and every solution is constant on it (Theorem 10.2): non-constant data are impossible. If the data equal $k$, then $z=k+h(bx-ay)$ works for every $C^1$ function $h$ with $h(0)=0$.
III
Lagrange's method.
  • 1 Find the general integral of $x^2p+y^2q=z^2$.
  • 2 Find the general integral of $yz\,p+xz\,q=xy$.
  • 3 Solve $xz\,p+yz\,q+x^2+y^2=0$.
  • 4 Solve $x(y-z)p+y(z-x)q=z(x-y)$ by multipliers, and compare with Example 1(c).
  • 5 Find the integral surface of $xp+yq=z$ through the curve $x+y=1$, $z=1+x^2$.
  • 6 Find the general integral of $p+q=z^2$, and write the solutions explicitly.
  • 7 Show that the characteristics of $yp-xq=0$ in $(x,y,z)$-space are horizontal circles centred on the $z$-axis, and deduce from Theorem 10.4 that every integral surface is a surface of revolution about the $z$-axis.
  • 8 Prove: if $u$ is a first integral of the characteristic system of $Pp+Qq=R$ with $u_z\ne0$, then the function $z(x,y)$ defined implicitly by $u(x,y,z)=c$ is a solution.
1 $\frac{dx}{x^2}=\frac{dy}{y^2}$ gives $\frac1y-\frac1x=c_1$; $\frac{dy}{y^2}=\frac{dz}{z^2}$ gives $\frac1z-\frac1y=c_2$. General integral $\varphi\left(\frac1x-\frac1y,\ \frac1y-\frac1z\right)=0$.
2 $\frac{dx}{yz}=\frac{dy}{xz}$ gives $x\,dx=y\,dy$: $x^2-y^2=c_1$; $\frac{dx}{yz}=\frac{dz}{xy}$ gives $x\,dx=z\,dz$: $x^2-z^2=c_2$. So $\varphi(x^2-y^2,\ x^2-z^2)=0$.
3 $P=xz$, $Q=yz$, $R=-(x^2+y^2)$. $\frac{dx}{xz}=\frac{dy}{yz}$: $\frac yx=c_1$. Multipliers $(x,y,z)$: $x^2z+y^2z-z(x^2+y^2)=0$, so $x^2+y^2+z^2=c_2$. Answer $\varphi\left(\frac yx,\ x^2+y^2+z^2\right)=0$: spheres sliced by planes through the $z$-axis.
4 $(1,1,1)$: $x(y-z)+y(z-x)+z(x-y)=0$, so $x+y+z=c_1$. $\left(\frac1x,\frac1y,\frac1z\right)$: $(y-z)+(z-x)+(x-y)=0$, so $xyz=c_2$. $\varphi(x+y+z,xyz)=0$ — Example 1(c) read backwards.
5 General solution $z=xf\!\left(\frac yx\right)$. On $x+y=1$ put $w=\frac yx=\frac{1-x}{x}$, so $x=\frac1{1+w}$, and $1+x^2=xf(w)$ gives $f(w)=\frac1x+x=(1+w)+\frac1{1+w}$. Hence $$z=x+y+\frac{x^2}{x+y}.$$ On $x+y=1$ this is $1+x^2$. ✓
6 $dx=dy$: $x-y=c_1$; $dy=\frac{dz}{z^2}$: $y+\frac1z=c_2$. General integral $\varphi\left(x-y,\ y+\frac1z\right)=0$, i.e. $y+\frac1z=h(x-y)$ and $z=\dfrac{1}{h(x-y)-y}$ (together with $z\equiv0$).
7 $x_t=y$, $y_t=-x$, $z_t=0$: $x^2+y^2$ and $z$ are constant, so the characteristics are circles $x^2+y^2=r^2$, $z=$ const. An integral surface is a union of such circles (Theorem 10.4), hence invariant under rotation about the $z$-axis: $z=f(x^2+y^2)$ where it is a graph.
8 By the implicit function theorem $p=-\frac{u_x}{u_z}$, $q=-\frac{u_y}{u_z}$. So $Pp+Qq=-\frac{Pu_x+Qu_y}{u_z}=-\frac{-Ru_z}{u_z}=R$, using $Pu_x+Qu_y+Ru_z=0$.
IV
Cauchy problems, transversality and breaking.
  • 1 For $z_x+2z_y=0$ compute $J$ for data on (a) $y=0$; (b) $y=2x$; (c) $y=x^2$. Where does transversality fail?
  • 2 Solve $xz_x+yz_y=z$ with $z=1$ on the circle $x^2+y^2=1$, and check transversality.
  • 3 Solve $z_y+zz_x=0$ with $z(x,0)=3-2x$, and find the breaking time.
  • 4 Find the breaking time of $z_y+zz_x=0$ for $z(x,0)=\sin x$, and for $z(x,0)=\tan^{-1}x$.
  • 5 Solve $z_y+zz_x=0$, $z(x,0)=x$, and find $z(4,1)$.
  • 6 Solve the Cauchy problem $z_x+z_y=z^2$, $z(x,0)=1$, and find where the solution exists.
  • 7 Prove that two base characteristics of a linear equation cannot cross, and show that two base characteristics of $z_y+zz_x=0$ can.
  • 8 Modelling: a traffic ramp. For $z_y+zz_x=0$ let $z(x,0)=1$ for $x\lt0$, $1-x$ for $0\le x\le1$ and $0$ for $x\gt1$. Show that all characteristics from $[0,1]$ pass through one point, and find it.
1 $J=x_0'\cdot2-y_0'\cdot1$. (a) $(s,0)$: $J=2$. (b) $(s,2s)$: $J=2-2=0$ everywhere — the line is characteristic. (c) $(s,s^2)$: $J=2-2s$, zero at $s=1$, i.e. at $(1,1)$, where the parabola has slope $2$.
2 On $(\cos s,\sin s)$: $J=-\sin s\cdot\sin s-\cos s\cdot\cos s=-1\ne0$. General solution $z=xf\!\left(\frac yx\right)$, homogeneous of degree $1$, and $z=1$ on the unit circle forces $z=\sqrt{x^2+y^2}$ on $\mathbb{R}^2\setminus\{0\}$.
3 $z=3-2(x-zy)$, so $z(1-2y)=3-2x$ and $z=\dfrac{3-2x}{1-2y}$. $g'\equiv-2$: $y_b=\tfrac12$.
4 $\min\cos=-1$: $y_b=1$. For $\tan^{-1}x$, $g'=\frac1{1+x^2}\gt0$: no breaking, the solution exists for all $y\ge0$.
5 $z=x-zy$ gives $z=\frac{x}{1+y}$ (the lines spread out: a rarefaction). $z(4,1)=2$.
6 Characteristics $x=s+t$, $y=t$, $z_t=z^2$, $z(0)=1$: $z=\frac1{1-t}$. So $z=\dfrac{1}{1-y}$, defined for $y\lt1$, independent of $x$.
7 For a linear equation the base characteristics solve the planar system $x'=a(x,y)$, $y'=b(x,y)$, which has a unique solution through each point (Picard–Lindelöf, $a,b\in C^1$): two crossing curves would be two solutions through the crossing point. For Burgers, the lines $x=s$ (from $g(s)=0$) and $x=s'+y$ (from $g(s')=1$) with $s'\lt s$ meet at $y=s-s'$.
8 For $s\in[0,1]$ the line is $x=s+(1-s)y$, and at $y=1$ every one of them gives $x=1$. So they all meet at $(1,1)$: the ramp steepens to a vertical jump at $y_b=1$, in agreement with $y_b=-1/\min g'=1$. (After that a shock moves on; that needs the theory of weak solutions.)
🎯 GATE corner GATE MA · MCQGATE MA · MSQGATE MA · NAT

E1. GATE MA · MCQ Let $z(x,y)$ solve $x\,z_x+y\,z_y=2z$ for $y\gt0$ with $z(x,1)=1+x$. Then $z(2,3)$ equals (A) $5$; (B) $15$; (C) $45$; (D) $\tfrac53$.

Idea Characteristics are rays $\frac xy=c$, and along them $z$ grows like the square of the distance: the general solution is $z=y^2f\!\left(\frac xy\right)$ (homogeneous of degree $2$).
Data $z(x,1)=f(x)=1+x$, so $z=y^2\left(1+\frac xy\right)=y^2+xy$. Check: $x\cdot y+y(2y+x)=2(y^2+xy)$. ✓
Evaluate $z(2,3)=9+6=15$. The distractors come from the wrong degree of homogeneity: degree $1$ gives $y+x=5$ (A), degree $3$ gives $y^3+xy^2=45$ (C), and degree $0$ (data carried along the rays without growth) gives $1+\frac xy=\frac53$ (D).
Answer (B).

E2. GATE MA · MCQ The solution of $z_y+z\,z_x=0$, $z(x,0)=e^{-x^2}$, remains $C^1$ exactly for $0\le y\lt y_b$, where $y_b$ equals (A) $\sqrt{e/2}$; (B) $\sqrt{2/e}$; (C) $\tfrac12$; (D) $\sqrt e$.

Idea Theorem 10.8: $y_b=-1/\min g'$.
Compute $g'(x)=-2xe^{-x^2}$ and $g''(x)=(4x^2-2)e^{-x^2}$, so the minimum of $g'$ is at $x=\frac1{\sqrt2}$: $g'=-\sqrt2\,e^{-1/2}$.
Options $y_b=\dfrac{e^{1/2}}{\sqrt2}=\sqrt{e/2}\approx1.166$. (B) is $\min g'$ in absolute value; (C) and (D) are slips.
Answer (A).

E3. GATE MA · MSQ CSIR Part C Consider $z_x+z_y=z$ with the data $z=1$ prescribed on a curve $\Gamma_0$. For which of the following choices does the Cauchy problem have a unique $C^1$ solution in some neighbourhood of the indicated point? (One or more options may be correct.) (A) $\Gamma_0$: $y=0$, near $(0,0)$. (B) $\Gamma_0$: $y=x$, near $(0,0)$. (C) $\Gamma_0$: $x+y=0$, near $(0,0)$. (D) $\Gamma_0$: $y=x^2$, near $\left(\tfrac12,\tfrac14\right)$.

Setup $(P,Q)=(1,1)$, so $J=x_0'-y_0'$. Characteristics are the lines $x-y=c$, along which $z=Ce^{x}$.
(A) Yes $(s,0)$: $J=1\ne0$. Theorem 10.6; explicitly $z=e^{y}$.
(B) No $J=1-1=0$: the line is a characteristic, along which $z=Ce^x$ cannot be constant. No solution (Theorem 10.7).
(C) Yes $(s,-s)$: $J=1+1=2\ne0$; the solution is $z=e^{(x+y)/2}$.
(D) No $(s,s^2)$: $J=1-2s=0$ at $s=\tfrac12$. The lines $x-y=c$ with $c\lt\tfrac14$ meet the parabola twice, at $s$ and $1-s$, and $z=Ce^x$ equals $1$ at both only if $s=\tfrac12$; lines with $c\gt\tfrac14$ miss it. So no solution exists in any neighbourhood of $\left(\tfrac12,\tfrac14\right)$.
Answer (A) and (C).

E4. GATE MA · MSQ For $z_y+z\,z_x=0$ with $z(x,0)=g(x)$, which of the following statements is/are true? (One or more options may be correct.) (A) If $g(x)=x$, then $z=\dfrac{x}{1+y}$ for all $y\ge0$. (B) If $g(x)=-x$, the classical solution exists for $0\le y\lt1$ and not beyond. (C) If $g(x)=\sin x$, the solution is $C^1$ for all $y\ge0$. (D) The solution is constant along each line $x=s+g(s)\,y$, as long as it is $C^1$.

(A) True $z=x-zy$ gives $z=\frac{x}{1+y}$; $g'=1\gt0$, no breaking.
(B) True $z=\frac{x}{y-1}$ (Example 6(a)); it blows up at $y=1=y_b$.
(C) False $\min g'=-1$, so $y_b=1$: the profile becomes vertical at $y=1$, at $x=\pi$ (mod $2\pi$).
(D) True Theorem 10.8(1): $z_t=R=0$ along characteristics, which are these lines.
Answer (A), (B) and (D).

E5. GATE MA · NAT Let $z(x,y)$ solve $z_x+x\,z_y=z$ with $z(0,y)=\cos y$. The value of $z(1,1)$ is ______ (correct to two decimal places).

Characteristics $\frac{dy}{dx}=x$: $y-\frac{x^2}2=c$. With $\eta=x$, Theorem 10.3(1) gives $w_\eta=w$, so $z=e^xf\!\left(y-\frac{x^2}2\right)$.
Data $z(0,y)=f(y)=\cos y$ (the line $x=0$ is transversal: $J=0\cdot Q-1\cdot P=-1$). So $z=e^x\cos\!\left(y-\frac{x^2}{2}\right)$.
Answer $z(1,1)=e\cos\tfrac12\approx2.3855$. Answer: $2.39$ (accept $2.38$ to $2.40$).

E6. GATE MA · NAT Let $z(x,y)$ solve $z_x+z_y=z^2$ with $z(x,0)=\dfrac{1}{1+x^2}$. The value of $z\left(1,\tfrac12\right)$ is ______ (correct to two decimal places).

Characteristics $x=s+t$, $y=t$, and $z_t=z^2$ with $z(0)=z_0=\frac1{1+s^2}$ gives $z=\dfrac{z_0}{1-z_0t}=\dfrac{1}{1+s^2-t}$.
Invert $s=x-y$, $t=y$: $z=\dfrac{1}{1+(x-y)^2-y}$. Check: $z_x+z_y=z^2\left(-2(x-y)+2(x-y)+1\right)=z^2$. ✓
Answer $z\left(1,\tfrac12\right)=\dfrac{1}{1+\frac14-\frac12}=\dfrac43\approx1.3333$. Answer: $1.33$ (accept $1.32$ to $1.34$).
🧑‍🏫 For the teacher

Suggested plan — 2 lectures × 60 min.

  1. Linear equations and Lagrange's method. Definitions 10.1–10.2 with Example 1(a); the dot-product picture $(P,Q,R)\cdot(p,q,-1)=0$; Theorem 10.1 and Example 1(c). Then Definition 10.3 and Theorems 10.2–10.3 with Examples 2 and 3; project Interactive 1 in "transport" and "stretching" modes. Finish with Definition 10.4, Theorems 10.4–10.5, the strategy box, and Example 4 on the board.
  2. The Cauchy problem. Definition 10.5, Theorem 10.6 (do Steps 1–3 carefully; the Jacobian $J$ is the whole point), Theorem 10.7 with Example 7(a)(b). Project Interactive 2: rotate $\alpha$ for the parabola and ask the class to predict where the amber point will appear before releasing the slider; then choose the line and find the angle at which every line is characteristic. Example 5, Theorem 10.8, Example 6, and Interactive 1 in Burgers mode. GATE corner E3 and E4 live.

Discussion questions. (1) Why does the general solution of a first-order PDE contain an arbitrary function, whereas that of a first-order ODE contains an arbitrary constant? (2) In Example 5 the solution dies on $y=-1$ although nothing is singular there in the equation or the data. What exactly happens to the characteristics? (3) In Burgers' equation the bump breaks on its front side. Could a different equation $z_y+c(z)z_x=0$ make it break on the back side?

Misconceptions to watch for. Calling $zp+q=0$ linear; using $\frac{dx}P=\frac{dy}Q$ when $P,Q$ contain $z$; taking two dependent first integrals; forgetting to check $J$ before solving a Cauchy problem; believing that smooth data give a global solution; computing a breaking time from the maximum of $|g'|$ instead of the most negative $g'$.

Suggested homework. I(2, 4, 5, 7), II(1, 2, 4, 6, 8), III(1, 3, 5, 6, 8), IV(1, 2, 4, 6, 7, 8); GATE corner E1, E3, E5.

🔬 Research–Teaching NexusResearch-tutored · optional, about 60 minutes

Research question. On a motorway, a smooth patch of dense traffic can turn into a sharp "stop wave" with no accident and no bottleneck. Can a first-order PDE predict when and where such a jam front forms?

Why it matters. The Lighthill–Whitham–Richards model treats traffic as a fluid with density $\rho(x,t)$ (cars per km) and speed $v(\rho)=v_{\max}\left(1-\frac{\rho}{\rho_{\max}}\right)$. Conservation of cars gives $\rho_t+\big(\rho v(\rho)\big)_x=0$, i.e. $\rho_t+c(\rho)\rho_x=0$ with $c(\rho)=v_{\max}\left(1-\frac{2\rho}{\rho_{\max}}\right)$ — a quasilinear equation of exactly the Burgers type. Variants of this model are used by traffic engineers in Europe, the US and Asia to design ramp metering and variable speed limits.

What to do.

  1. Derive $c(\rho)$ from the flux $\rho v(\rho)$ and explain why information travels backwards ($c\lt0$) when $\rho\gt\rho_{\max}/2$, although every car drives forwards.
  2. Take $\rho(x,0)=\rho_{\max}\left(\tfrac14+\tfrac12e^{-x^2}\right)$ ($x$ in km) and $v_{\max}=100$ km/h. Using the general form of Theorem 10.8, find the time (in seconds) and the place at which the density profile first becomes vertical.
  3. Draw the characteristics $x=s+c(\rho_0(s))\,t$ for $s\in[-3,3]$ on graph paper or with a short script, and mark where they first cross.
  4. Compare with Interactive 1 (Burgers mode): which side of the bump breaks here, and why is it the opposite side to Burgers?

Tools: pencil and paper for 1–2; Python, a spreadsheet or graph paper for 3.

Hand in: the derivation of step 1, the breaking time and position with working, a plot of characteristics, and a paragraph answering the research question.

Going further. After breaking, the model needs a shock: a moving discontinuity whose speed is fixed by conservation (the Rankine–Hugoniot condition $\dot x_s=\frac{[\rho v]}{[\rho]}$). Look it up and compute the speed of a jam front between $\rho=0.2\rho_{\max}$ and $\rho=0.8\rho_{\max}$.

Instructor note: $c(\rho_0(x))=v_{\max}\left(1-\tfrac12-e^{-x^2}\right)=v_{\max}\left(\tfrac12-e^{-x^2}\right)$, whose $x$-derivative $2v_{\max}xe^{-x^2}$ is most negative at $x=-\tfrac1{\sqrt2}$, with value $-\sqrt2\,v_{\max}e^{-1/2}$. So $t_b=\dfrac{e^{1/2}}{\sqrt2\,v_{\max}}\approx\dfrac{1.1658}{100}$ h $\approx42$ s, on the characteristic from $s=-\tfrac1{\sqrt2}\approx-0.707$ km, which carries $c=v_{\max}\left(\tfrac12-e^{-1/2}\right)\approx-10.65$ km/h; it is at $x\approx-0.707-0.124\approx-0.83$ km at $t_b$. The back of the dense patch steepens (cars run into the queue), the mirror image of Burgers, because $c$ decreases with $\rho$. Shock speed: $\dot x_s=v_{\max}\left(1-\frac{\rho_1+\rho_2}{\rho_{\max}}\right)=0$ for $0.2$ and $0.8$ — a stationary jam front.

🧪 Self-check quiz

0/5 correct

Q1. The general solution of $3z_x-2z_y=0$ is

Q2. Lagrange's auxiliary equations for $x^2p+y^2q=z^2$ are

Q3. For $z_x+2z_y=0$, the Cauchy problem with data on which line has no unique solution?

Q4. The equation $z\,z_x+z_y=0$ is

Q5. For $z_y+zz_x=0$ with $z(x,0)=2\sin x$, the breaking time is