Chapter 3 · First-Order Partial Differential Equations

3.2 Charpit’s Method and the Cauchy Problem

Nonlinear first-order equations $F(x,y,z,p,q)=0$: complete, general and singular integrals, compatible systems, Charpit's method and the standard forms, the Cauchy problem for quasi-linear equations with the transversality condition, characteristic strips, and Jacobi's method.

Lagrange's method works because, for a quasi-linear equation, the direction $(P,Q,R)$ in which an integral surface must extend is known before the surface is: it depends only on the point $(x,y,z)$. For a genuinely nonlinear equation such as $p^2+q^2=1$ this is no longer true. At a given point many tangent planes are allowed (here, every plane making an angle of $45^\circ$ with the horizontal), and which one the solution uses decides in which direction it spreads. The remedy, found by Lagrange and turned into a practical method by Paul Charpit in a memoir of 1784 (preserved because Lacroix published it in his treatise), is to carry the slopes $p,q$ along as unknowns: the characteristics of a nonlinear equation are curves in the five-dimensional space of $(x,y,z,p,q)$ — characteristic strips, little ribbons of tangent plane moving along a curve.

This section develops the three kinds of solution of a nonlinear equation (complete, general, singular), Charpit's method for finding a complete integral, the four standard forms that CSIR NET uses most, and the Cauchy problem: when does a curve in space lie on exactly one integral surface? The answer — the transversality condition — explains every "none, one or infinitely many" question of this chapter, and the breaking of waves in Burgers' equation shows that even for smooth data the solution may exist only locally. Typical exam questions ask for a complete or singular integral, the solution of a Clairaut-type equation, whether two equations are compatible, the solution of a Cauchy problem by characteristics, and the first time at which characteristics cross. Notation: $z=z(x,y)$, $p=z_x$, $q=z_y$; subscripts on $F$ denote partial derivatives.

Complete, General and Singular Integrals

In Example 2 of §3.1 the two-parameter family of unit spheres $(x-a)^2+(y-b)^2+z^2=1$ produced the equation $z^2(1+p^2+q^2)=1$. Reversing the process, a two-parameter family of solutions is the natural "general solution" of a nonlinear first-order equation, and other solutions arise from it by taking envelopes, exactly as singular solutions of ODEs arise from envelopes of one-parameter families (§1.2).

📘 Definition 3.4 — Complete, general, singular and particular integrals

Let $F(x,y,z,p,q)=0$ be a first-order PDE.

  • A complete integral is a two-parameter family of solutions $z=\phi(x,y,a,b)$, with $\phi$ of class $C^2$, such that the matrix $$\begin{pmatrix}\phi_a&\phi_{xa}&\phi_{ya}\\ \phi_b&\phi_{xb}&\phi_{yb}\end{pmatrix}$$ has rank $2$ — the two parameters are essential (they cannot be merged into one).
  • A general integral is obtained by restricting to a one-parameter subfamily $b=h(a)$, $h$ arbitrary, and forming its envelope: eliminate $a$ from $z=\phi(x,y,a,h(a))$ and $\phi_a+\phi_b\,h'(a)=0$.
  • A singular integral is the envelope of the whole two-parameter family: eliminate $a,b$ from $z=\phi$, $\phi_a=0$, $\phi_b=0$ — when this produces a solution.
  • A particular integral is any solution obtained by giving $a,b$ (or $h$) definite values.
📐 Theorem 3.8 — Envelopes of solutions are solutions
Let $z=\psi(x,y,a)$ be a one-parameter family of solutions of $F=0$, with $\psi\in C^2$, and suppose the equation $\psi_a(x,y,a)=0$ defines $a=\alpha(x,y)$ as a $C^1$ function on a domain $D$ (for instance $\psi_{aa}\ne0$). Then the envelope $Z(x,y)=\psi\big(x,y,\alpha(x,y)\big)$ is a solution on $D$. The same holds for two-parameter families: if $\phi_a=\phi_b=0$ defines $a=\alpha(x,y)$, $b=\beta(x,y)$ as $C^1$ functions, then $Z=\phi(x,y,\alpha,\beta)$ is a solution. In particular general and singular integrals are solutions.

Proof. By the chain rule and $\psi_a(x,y,\alpha(x,y))=0$, $$Z_x=\psi_x+\psi_a\,\alpha_x=\psi_x\big(x,y,\alpha(x,y)\big),$$ $$Z_y=\psi_y\big(x,y,\alpha(x,y)\big).$$ So at the point $(x,y)$ the envelope has the same value and the same first derivatives as the member of the family with parameter $a=\alpha(x,y)$: the two surfaces touch there. That member is a solution, so $F\big(x,y,\psi,\psi_x,\psi_y\big)=0$ with parameter $\alpha(x,y)$, which is exactly $F(x,y,Z,Z_x,Z_y)=0$. The two-parameter case is identical, with $\phi_a=\phi_b=0$ killing both extra terms. ∎

A solution of a PDE is determined, at each point, by first-order data $(x,y,z,p,q)$ only; a surface that touches a solution at every point therefore inherits the equation. This is the whole idea behind general and singular integrals.

📐 Theorem 3.9 — The singular integral satisfies $F_p=F_q=0$
Let $z=\phi(x,y,a,b)$ be a complete integral of $F=0$, with $F\in C^1$, and let $Z=\phi(x,y,\alpha,\beta)$ be a singular integral as in Theorem 3.8. Then along it $$F=0,\qquad F_p=0,\qquad F_q=0 .$$ Consequently every singular integral is contained in the surface obtained by eliminating $p,q$ from these three equations. The converse fails: the eliminant must be checked (it may not be a solution).

Proof. For all $(x,y,a,b)$ we have $F\big(x,y,\phi,\phi_x,\phi_y\big)=0$. Differentiate with respect to $a$ and to $b$: $$F_z\phi_a+F_p\phi_{xa}+F_q\phi_{ya}=0,$$ $$F_z\phi_b+F_p\phi_{xb}+F_q\phi_{yb}=0 .$$ On the singular integral $\phi_a=\phi_b=0$, so $F_p\phi_{xa}+F_q\phi_{ya}=0$ and $F_p\phi_{xb}+F_q\phi_{yb}=0$. Since the first column of the rank-2 matrix of Definition 3.4 vanishes there, its remaining $2\times2$ block $\begin{pmatrix}\phi_{xa}&\phi_{ya}\\ \phi_{xb}&\phi_{yb}\end{pmatrix}$ is invertible, and the homogeneous system forces $F_p=F_q=0$, evaluated at the first-order data of the member $(\alpha,\beta)$ — which are the data of $Z$ (proof of Theorem 3.8). ∎

✏️ Example 1 — Spheres, cylinders and two planes
For $z^2\left(1+p^2+q^2\right)=1$ with the complete integral $(x-a)^2+(y-b)^2+z^2=1$: (a) find the singular integral, both from the envelope and from Theorem 3.9; (b) find the general integrals corresponding to $b=0$ and to $b=a$.
(a) Envelope Differentiating $(x-a)^2+(y-b)^2+z^2=1$ with respect to $a$ and $b$: $x-a=0$, $y-b=0$. Substituting, $z^2=1$: the planes $z=1$ and $z=-1$, which every unit sphere centred on $z=0$ touches.
(a) Discriminant $F=z^2(1+p^2+q^2)-1$: $F_p=2z^2p=0$, $F_q=2z^2q=0$. Since $z\ne0$ on any solution, $p=q=0$, and $F=0$ gives $z^2=1$. Both planes do satisfy the PDE ($p=q=0$, $z^2=1$): they are the singular integrals.
(b) $b=0$ Spheres centred on the $x$-axis: $(x-a)^2+y^2+z^2=1$; $\partial_a$: $x=a$. Envelope: $$y^2+z^2=1,$$ the unit cylinder about the $x$-axis — the tube swept by a ball rolling along the axis.
(b) $b=a$ $(x-a)^2+(y-a)^2+z^2=1$; $\partial_a$: $(x-a)+(y-a)=0$, $a=\tfrac{x+y}2$. Then $x-a=\tfrac{x-y}2=-(y-a)$, and the envelope is $$\frac{(x-y)^2}{2}+z^2=1,$$ the unit cylinder about the line $x=y$, $z=0$ (the distance of $(x,y)$ from that line is $|x-y|/\sqrt2$).
Check For $z=\sqrt{1-y^2}$: $p=0$, $q=-y/z$, so $z^2(1+q^2)=z^2+y^2=1$ ✓.

Compatible Systems and Charpit's Method

Charpit's idea: to solve $F=0$, look for a second equation $G(x,y,z,p,q)=a$ such that the two together can be solved for $p$ and $q$, $p=\varphi(x,y,z,a)$, $q=\psi(x,y,z,a)$, and such that the Pfaffian equation $dz=\varphi\,dx+\psi\,dy$ can be integrated. Its solutions contain a second constant $b$, giving a complete integral. Everything rests on knowing when $dz=\varphi\,dx+\psi\,dy$ is integrable.

📘 Definition 3.5 — Compatible equations
Two first-order equations $F(x,y,z,p,q)=0$ and $G(x,y,z,p,q)=0$ are compatible on a region if every point $(x_0,y_0,z_0)$ of it lies on a solution $z=z(x,y)$ common to both equations. We assume throughout that $J=\dfrac{\partial(F,G)}{\partial(p,q)}=F_pG_q-F_qG_p\ne0$, so that the two equations can be solved for $p$ and $q$.
📐 Theorem 3.10 — Integrability of $dz=\varphi\,dx+\psi\,dy$
Let $\varphi,\psi$ be $C^1$ on an open set $\Omega\subseteq\mathbb{R}^3$. Through every point of $\Omega$ there passes a $C^1$ surface $z=z(x,y)$ with $z_x=\varphi(x,y,z)$, $z_y=\psi(x,y,z)$ if and only if $$\varphi_y+\psi\,\varphi_z=\psi_x+\varphi\,\psi_z\quad\text{in }\Omega .$$

Proof. Necessity. If such a surface passes through $(x_0,y_0,z_0)$, then $z_x=\varphi(x,y,z)$ is $C^1$ along it and $z_{xy}=\varphi_y+\varphi_z z_y=\varphi_y+\varphi_z\psi$; similarly $z_{yx}=\psi_x+\psi_z\varphi$. Equality of mixed partials gives the condition at $(x_0,y_0,z_0)$.

Sufficiency (sketch with the key step). Given $(x_0,y_0,z_0)$, first solve the ODE $\zeta'(x)=\varphi(x,y_0,\zeta)$, $\zeta(x_0)=z_0$, and then, for each fixed $x$, solve $\tfrac{\partial z}{\partial y}=\psi(x,y,z)$ with $z(x,y_0)=\zeta(x)$. By differentiable dependence on initial values, $z(x,y)$ is $C^1$ and $z_y=\psi$ by construction. Put $w=z_x-\varphi(x,y,z)$; then $w(x,y_0)=0$, and using $z_{xy}=z_{yx}$ and the hypothesis, $$w_y=\big(\psi(x,y,z)\big)_x-\varphi_y-\varphi_z z_y$$ $$=\psi_x+\psi_z(w+\varphi)-\varphi_y-\varphi_z\psi=\psi_z\,w .$$ For fixed $x$ this is a linear ODE in $y$ with zero initial value, so $w\equiv0$: $z_x=\varphi$ as well. ∎

📐 Theorem 3.11 — The compatibility condition
Let $F,G\in C^1$ with $J=\dfrac{\partial(F,G)}{\partial(p,q)}\ne0$. Then $F=0$ and $G=0$ are compatible if and only if, at every point where $F=G=0$, $$[F,G]:=\frac{\partial(F,G)}{\partial(x,p)}+p\,\frac{\partial(F,G)}{\partial(z,p)}$$ $$+\frac{\partial(F,G)}{\partial(y,q)}+q\,\frac{\partial(F,G)}{\partial(z,q)}=0,$$ where $\dfrac{\partial(F,G)}{\partial(x,p)}=F_xG_p-F_pG_x$, and so on.

Proof. Since $J\ne0$, the implicit function theorem solves $F=G=0$ locally as $p=\varphi(x,y,z)$, $q=\psi(x,y,z)$, and compatibility means exactly that $dz=\varphi\,dx+\psi\,dy$ is integrable. Differentiate the identities $F(x,y,z,\varphi,\psi)=0$, $G(x,y,z,\varphi,\psi)=0$ with respect to $x$ and solve by Cramer's rule; then do the same with $y$ and with $z$: $$\psi_x=\frac1J\frac{\partial(F,G)}{\partial(x,p)},\quad \psi_z=\frac1J\frac{\partial(F,G)}{\partial(z,p)},$$ $$\varphi_y=-\frac1J\frac{\partial(F,G)}{\partial(y,q)},\quad \varphi_z=-\frac1J\frac{\partial(F,G)}{\partial(z,q)} .$$ (For instance, the $x$-derivatives give $F_p\varphi_x+F_q\psi_x=-F_x$, $G_p\varphi_x+G_q\psi_x=-G_x$, and Cramer's rule yields $\psi_x=(F_xG_p-F_pG_x)/J$.) Substituting into the condition $\varphi_y+\psi\varphi_z=\psi_x+\varphi\psi_z$ of Theorem 3.10, with $\varphi=p$, $\psi=q$, and multiplying by $-J$, we get exactly $[F,G]=0$. ∎

✏️ Example 2 — Compatible or not?
(a) Show that $x\,p+y\,q=3z$ and $q=x^2$ are compatible, and find their common solutions. (b) Show that $p=y$ and $q=2x$ are not compatible.
(a) Bracket $F=xp+yq-3z$, $G=q-x^2$. Then $F_x=p$, $F_y=q$, $F_z=-3$, $F_p=x$, $F_q=y$; $G_x=-2x$, $G_q=1$, the rest $0$. $J=F_pG_q-F_qG_p=x\ne0$ for $x\ne0$. $$\tfrac{\partial(F,G)}{\partial(x,p)}=0-x(-2x)=2x^2,\quad \tfrac{\partial(F,G)}{\partial(z,p)}=0,$$ $$\tfrac{\partial(F,G)}{\partial(y,q)}=q,\quad \tfrac{\partial(F,G)}{\partial(z,q)}=-3 .$$ So $[F,G]=2x^2+q-3q=2(x^2-q)$, which vanishes wherever $G=0$: compatible.
(a) Common solutions $q=x^2$ gives $z=x^2y+h(x)$. Then $xp+yq-3z=x(2xy+h')+x^2y-3x^2y-3h=xh'-3h$, which is $0$ iff $h=cx^3$. Common solutions: $$z=x^2y+cx^3,$$ a one-parameter family, one through each point with $x\ne0$ ($c=(z_0-x_0^2y_0)/x_0^3$).
(b) $F=p-y$, $G=q-2x$: $\tfrac{\partial(F,G)}{\partial(x,p)}=0-1\cdot(-2)=2$, $\tfrac{\partial(F,G)}{\partial(y,q)}=(-1)(1)-0=-1$, the $z$-terms vanish: $[F,G]=1\ne0$. Directly: $z_{xy}=1$ but $z_{yx}=2$.
Note In (a) the bracket is not identically zero — only zero on the set $F=G=0$. That is all Theorem 3.11 requires.

Now fix $F$ and regard $[F,G]=0$ as an equation for the unknown $G$. Expanding the Jacobians and collecting terms, $$-[F,G]=F_p\,G_x+F_q\,G_y+(pF_p+qF_q)\,G_z$$ $$-(F_x+pF_z)\,G_p-(F_y+qF_z)\,G_q,$$ a linear homogeneous first-order PDE for $G$ in the five variables $x,y,z,p,q$. By the theorem on homogeneous linear equations of §3.1 (which holds verbatim in five variables), its solutions are exactly the first integrals of the associated characteristic system.

📐 Theorem 3.12 — Charpit's method
Let $F\in C^2$. Charpit's auxiliary equations for $F(x,y,z,p,q)=0$ are $$\frac{dx}{F_p}=\frac{dy}{F_q}=\frac{dz}{pF_p+qF_q}$$ $$=\frac{dp}{-(F_x+pF_z)}=\frac{dq}{-(F_y+qF_z)} .$$ $F$ itself is a first integral of this system. If $G(x,y,z,p,q)$ is another first integral with $\partial(F,G)/\partial(p,q)\ne0$, then for each constant $a$ the equations $F=0$, $G=a$ are compatible; solving them for $p=\varphi(x,y,z,a)$, $q=\psi(x,y,z,a)$ and integrating $dz=\varphi\,dx+\psi\,dy$ gives a family of solutions $z=\phi(x,y,a,b)$ of $F=0$ — a complete integral when the rank condition of Definition 3.4 holds.

Proof. Let $X$ be the vector field $\big(F_p,F_q,pF_p+qF_q,-(F_x+pF_z),-(F_y+qF_z)\big)$. The displayed formula says $[F,G]=-X\cdot\nabla G$. Taking $G=F$: $$X\cdot\nabla F=F_pF_x+F_qF_y+(pF_p+qF_q)F_z$$ $$-(F_x+pF_z)F_p-(F_y+qF_z)F_q=0,$$ so $F$ is a first integral. If $G$ is a first integral, $X\cdot\nabla G=0$, so $[F,G-a]=[F,G]=0$ everywhere; by Theorem 3.11, $F=0$ and $G=a$ are compatible, so $dz=\varphi\,dx+\psi\,dy$ is integrable (Theorem 3.10) and its integral surfaces, which contain one more constant $b$, satisfy $F=0$ because $(z_x,z_y)=(\varphi,\psi)$ solves $F=0$. ∎

🧭 Problem-solving strategy — Charpit's method
  1. Write $F=0$ and compute $F_x,F_y,F_z,F_p,F_q$.
  2. Write Charpit's equations and look for the simplest integrable combination that contains $p$ or $q$ — often one of $\frac{dp}{\cdots}$, $\frac{dq}{\cdots}$ alone (if $F_x+pF_z=0$, then $p=a$).
  3. Check $\partial(F,G)/\partial(p,q)\ne0$ (an integral free of $p$ and $q$, such as one from $\frac{dx}{F_p}=\frac{dy}{F_q}$ alone, is useless).
  4. Solve $F=0$, $G=a$ for $p,q$; substitute into $dz=p\,dx+q\,dy$ and integrate (look for exact combinations such as $d(z-\cdots)$).
  5. Check the complete integral in the PDE. Then find the singular integral (Theorem 3.9, then verify) and, if asked, a general integral.
✏️ Example 3 — Charpit's method for $x\,p^2+q=z$
Find a complete integral of $x\,p^2+q=z$ in the region $x\gt0$.
Derivatives $F=xp^2+q-z$: $F_x=p^2$, $F_y=0$, $F_z=-1$, $F_p=2xp$, $F_q=1$.
Charpit $$\frac{dx}{2xp}=\frac{dy}{1}=\frac{dz}{2xp^2+q}$$ $$=\frac{dp}{-(p^2-p)}=\frac{dq}{-(0-q)} .$$ The second and last fractions give $dy=\dfrac{dq}{q}$, so $q=a\,e^{y}$. Here $G=qe^{-y}$ and $\partial(F,G)/\partial(p,q)=2xp\,e^{-y}\ne0$ when $p\ne0$.
Solve for $p$ $xp^2=z-ae^y$, so $p=\sqrt{\dfrac{z-ae^y}{x}}$ (where $z\gt ae^y$).
Integrate $dz=p\,dx+q\,dy$ becomes $dz-ae^y\,dy=\sqrt{\tfrac{z-ae^y}{x}}\,dx$. With $w=z-ae^y$: $\dfrac{dw}{\sqrt w}=\dfrac{dx}{\sqrt x}$, so $\sqrt w=\sqrt x+b$: $$z=a\,e^{y}+\big(\sqrt x+b\big)^2 .$$
Check $p=\dfrac{\sqrt x+b}{\sqrt x}$, so $xp^2=(\sqrt x+b)^2$, and $q=ae^y$: $xp^2+q=z$ ✓. The rank condition holds: $\phi_a=e^y$, $\phi_b=2(\sqrt x+b)$, $\phi_{ya}=e^y$, $\phi_{yb}=0$, $\phi_{xb}=1/\sqrt x$, and the minor $\begin{vmatrix}\phi_a&\phi_{xa}\\ \phi_b&\phi_{xb}\end{vmatrix}=e^y/\sqrt x\ne0$.

Jacobi's method (brief)

Jacobi's method is Charpit's idea for an equation in which the unknown does not appear: $F(x_1,x_2,x_3,p_1,p_2,p_3)=0$ with $p_i=\partial u/\partial x_i$. (An equation $F(x,y,z,p,q)=0$ can be brought to this form by looking for the solution implicitly as $u(x,y,z)=0$, with $p=-u_x/u_z$, $q=-u_y/u_z$.) With the bracket $(F,G)=\sum_i\big(F_{x_i}G_{p_i}-F_{p_i}G_{x_i}\big)$, one looks for two further equations $F_1=a_1$, $F_2=a_2$ with $(F,F_1)=(F,F_2)=(F_1,F_2)=0$, found from Jacobi's auxiliary equations $$\frac{dx_1}{F_{p_1}}=\frac{dx_2}{F_{p_2}}=\frac{dx_3}{F_{p_3}}$$ $$=\frac{dp_1}{-F_{x_1}}=\frac{dp_2}{-F_{x_2}}=\frac{dp_3}{-F_{x_3}} .$$ Then $du=p_1dx_1+p_2dx_2+p_3dx_3$ is integrable and gives a complete integral $u=\Phi(x_1,x_2,x_3,a_1,a_2)+a_3$.

📐 Theorem 3.13 — Jacobi's method (statement)
Let $F,F_1,F_2$ be $C^1$ functions of $(x_1,x_2,x_3,p_1,p_2,p_3)$ with pairwise vanishing brackets $(F,F_1)=(F,F_2)=(F_1,F_2)=0$ and with $\dfrac{\partial(F,F_1,F_2)}{\partial(p_1,p_2,p_3)}\ne0$. Solving $F=0$, $F_1=a_1$, $F_2=a_2$ for $p_i=\varphi_i(x,a_1,a_2)$, the form $\sum\varphi_i\,dx_i$ is exact, and $u=\int\sum\varphi_i\,dx_i+a_3$ is a complete integral of $F=0$.

Proof (sketch). Exactness means $\partial\varphi_i/\partial x_j=\partial\varphi_j/\partial x_i$. Differentiating the three identities $F(x,\varphi)=0$, $F_1(x,\varphi)=a_1$, $F_2(x,\varphi)=a_2$ and eliminating the derivatives of $\varphi$ by Cramer's rule (as in Theorem 3.11) shows that the symmetry of the matrix $(\partial\varphi_i/\partial x_j)$ is equivalent to the vanishing of the three brackets. ∎

✏️ Example 4 — Jacobi's method for $p_1x_1+p_2x_2=p_3^2$
Find a complete integral of $x_1\,\dfrac{\partial u}{\partial x_1}+x_2\,\dfrac{\partial u}{\partial x_2}=\left(\dfrac{\partial u}{\partial x_3}\right)^2$ in the region $x_1,x_2\gt0$.
Auxiliary equations $F=p_1x_1+p_2x_2-p_3^2$: $F_{x_1}=p_1$, $F_{x_2}=p_2$, $F_{x_3}=0$, $F_{p_1}=x_1$, $F_{p_2}=x_2$, $F_{p_3}=-2p_3$. $$\frac{dx_1}{x_1}=\frac{dx_2}{x_2}=\frac{dx_3}{-2p_3}$$ $$=\frac{dp_1}{-p_1}=\frac{dp_2}{-p_2}=\frac{dp_3}{0}.$$
Two integrals $dp_3=0$: $F_1=p_3=a$. $\frac{dx_1}{x_1}=-\frac{dp_1}{p_1}$: $F_2=p_1x_1=b$. Brackets: $(F,F_1)=F_{x_3}\cdot1=0$; $(F,F_2)=F_{x_1}x_1-F_{p_1}p_1=p_1x_1-x_1p_1=0$; $(F_1,F_2)=0$ (no common variables).
Solve $p_3=a$, $p_1=b/x_1$, and from $F=0$: $p_2x_2=a^2-b$, $p_2=(a^2-b)/x_2$.
Integrate $du=\frac{b}{x_1}dx_1+\frac{a^2-b}{x_2}dx_2+a\,dx_3$: $$u=b\log x_1+(a^2-b)\log x_2+a\,x_3+c .$$ Check: $p_1x_1+p_2x_2=b+a^2-b=a^2=p_3^2$ ✓.

Standard Forms

Four types of equation occur again and again; for each, Charpit's equations integrate at once and the answer can be written down.

📐 Theorem 3.14 — Complete integrals of the standard forms
  1. $f(p,q)=0$ (only $p,q$ occur): $z=ax+h(a)\,y+c$, where $b=h(a)$ solves $f(a,b)=0$.
  2. Clairaut form $z=px+qy+f(p,q)$: $z=ax+by+f(a,b)$; the singular integral, if any, is its envelope (eliminate $a,b$ from $x+f_a=0$, $y+f_b=0$).
  3. $f(z,p,q)=0$ ($x,y$ absent): put $q=ap$; then $p=\Pi(z,a)$ from $f(z,p,ap)=0$, and $\displaystyle\int\frac{dz}{\Pi(z,a)}=x+ay+b$. Equivalently, look for $z=Z(X)$ with $X=x+ay$: $f(Z,Z',aZ')=0$ is an ODE.
  4. Separable $f_1(x,p)=f_2(y,q)$: set both sides equal to $a$, solve $p=g_1(x,a)$, $q=g_2(y,a)$; then $z=\int g_1(x,a)\,dx+\int g_2(y,a)\,dy+b$.

Proof. In each case we exhibit a first integral of Charpit's equations (Theorem 3.12) and integrate. (1) $F_x=F_y=F_z=0$, so $dp=dq=0$: $p=a$, and $q=h(a)$ by $F=0$; then $dz=a\,dx+h(a)\,dy$. (2) $F=px+qy+f(p,q)-z$ has $F_x+pF_z=p-p=0$ and $F_y+qF_z=0$, so $p=a$, $q=b$ are both first integrals, and $F=0$ gives $z=ax+by+f(a,b)$ directly. (3) $F_x=F_y=0$, so $\frac{dp}{-pF_z}=\frac{dq}{-qF_z}$, i.e. $\frac{dp}{p}=\frac{dq}{q}$, $q=ap$; then $dz=p\,(dx+a\,dy)=\Pi(z,a)\,d(x+ay)$. (4) $F=f_1(x,p)-f_2(y,q)$: $\frac{dx}{F_p}=\frac{dp}{-F_x}$ reads $\partial_xf_1\,dx+\partial_pf_1\,dp=0$, so $f_1(x,p)=a$, and then $f_2(y,q)=a$. ∎

✏️ Example 5 — Three standard forms
Find a complete integral of (a) $p+q=pq$; (b) $z=px+qy-pq$, together with its singular integral; (c) $x\,p^2=y\,q$ in the quadrant $x,y\gt0$.
(a) Form 1 $a+b=ab$ gives $b=\dfrac{a}{a-1}$ ($a\ne1$): $$z=ax+\frac{a}{a-1}\,y+c .$$ Check: $p+q=a+\tfrac{a}{a-1}=\tfrac{a^2}{a-1}=pq$ ✓. No singular integral: $F_p=1-q$, $F_q=1-p$ vanish only at $p=q=1$, where $F=1\ne0$.
(b) Clairaut Complete integral $z=ax+by-ab$ — planes. Envelope: $\partial_a$: $x-b=0$; $\partial_b$: $y-a=0$. So $a=y$, $b=x$ and $z=xy+xy-xy$: $$z=xy,$$ a hyperbolic paraboloid whose tangent planes are exactly the planes of the complete integral. Check: $p=y$, $q=x$, $px+qy-pq=xy$ ✓. Theorem 3.9 agrees: $F_p=x-q=0$, $F_q=y-p=0$.
(c) Separable $x\,p^2=a=y\,q$ with $a\gt0$: $p=\sqrt{a/x}$, $q=a/y$, so $$z=2\sqrt{a\,x}+a\log y+b .$$ Check: $xp^2=a=yq$ ✓.
✏️ Example 6 — Prove or disprove: "a complete integral is unique"
For $z=pq$: (a) find a complete integral by the method for $f(z,p,q)=0$; (b) find another by Charpit's equations; (c) find the singular integral; (d) decide whether the statement in the title is true, and whether $z=-\tfrac{(x-y)^2}{4}$ is a solution.
(a) $q=ap$: $z=ap^2$, $p=\sqrt{z/a}$, and $\dfrac{dz}{\sqrt{z/a}}=d(x+ay)$ gives $2\sqrt{az}=x+ay+b$: $$4az=(x+ay+b)^2\qquad(a\gt0).$$
(b) $F=pq-z$: $F_x=F_y=0$, $F_z=-1$, $F_p=q$, $F_q=p$. Charpit: $\dfrac{dx}{q}=\dfrac{dy}{p}=\dfrac{dz}{2pq}=\dfrac{dp}{p}=\dfrac{dq}{q}$. From $\frac{dx}{q}=\frac{dq}{q}$: $q=x+a$; from $\frac{dy}{p}=\frac{dp}{p}$: $p=y+b$. Then $dz=(y+b)\,dx+(x+a)\,dy=d\big((x+a)(y+b)\big)$: $$z=(x+a)(y+b).$$ Check: $pq=(y+b)(x+a)=z$ ✓.
(c) $F_p=q=0$, $F_q=p=0$, so $z=pq=0$; and $z\equiv0$ is a solution. It is also the envelope of either family (e.g. $\partial_a,\partial_b$ of $(x+a)(y+b)$ give $y+b=x+a=0$).
(d) False The two families are different: the zero set of $(x+a)(y+b)$ is a pair of crossing lines, while the zero set of $(x+ay+b)^2/(4a)$ is a single line, so no member of one family is a member of the other. Yet they are linked by envelopes: $b=a$ in (b) gives $\partial_a\big[(x+a)(y+a)\big]=x+y+2a=0$ and $z=\tfrac{x-y}{2}\cdot\tfrac{y-x}{2}=-\tfrac{(x-y)^2}{4}$, a solution ($p=-\tfrac{x-y}2$, $q=\tfrac{x-y}2$, $pq=z$ ✓) which is the member $a=-1$, $b=0$ of family (a) (allowing $a\lt0$, $z\le0$). A complete integral is a convenient two-parameter family, not the general solution.
Interactive 1 — Wavefronts of $p^2+q^2=1$: Rays, Envelopes and Caustics

Solve $p^2+q^2=1$ with $z=0$ on the red curve. The complete integral consists of the planes $z=x\cos\theta+y\sin\theta+c$; the solution is the envelope of those planes that vanish along the curve, and its level set $z=c$ is the envelope of circles of radius $c$ centred on the curve (dashed, Huygens' principle). The characteristics are straight rays leaving the curve along its normal (grey). Press Play to advance $c$. Where neighbouring rays cross, the smooth solution ends: the wavefront develops cusps and swallowtails.

level z = c—
first crossing of rays—
status—

The Cauchy Problem

We now make precise the question raised by Example 7 of §3.1: which curves lie on exactly one integral surface?

📘 Definition 3.6 — Cauchy problem; transversality

Let $\Gamma:\ s\mapsto\big(x_0(s),y_0(s),z_0(s)\big)$, $s\in I$ (an open interval), be a $C^1$ curve whose projection $\Gamma_0:\ s\mapsto(x_0(s),y_0(s))$ is regular ($(x_0',y_0')\ne0$). The Cauchy problem for $Pp+Qq=R$ asks for a solution $z(x,y)$, defined 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 projected curve $\Gamma_0$ is not tangent to the projected characteristic direction $(P,Q)$ there. $\Gamma$ is called non-characteristic if $J\ne0$ on $I$.

📐 Theorem 3.15 — Local existence and uniqueness for the Cauchy problem (quasi-linear case)
Let $P,Q,R\in C^1(\Omega)$, $\Omega\subseteq\mathbb{R}^3$ open, let $\Gamma\subseteq\Omega$ be as in Definition 3.6, 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 $s^*$. Any other $C^1$ solution with these data coincides with $z$ near $\big(x_0(s^*),y_0(s^*)\big)$.

Proof. Step 1 (characteristics from the data). For each $s$ let $t\mapsto\big(x(s,t),y(s,t),z(s,t)\big)$ solve $$x_t=P,\quad y_t=Q,\quad z_t=R,$$ $$\big(x,y,z\big)(s,0)=\big(x_0(s),y_0(s),z_0(s)\big).$$ By the existence theorem and the theorem on differentiable dependence on initial values (Chapter 1), these functions are defined and $C^1$ on a neighbourhood of $(s^*,0)$.

Step 2 (inverting the map). At $t=0$: $x_s=x_0'$, $y_s=y_0'$, $x_t=P$, $y_t=Q$, 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, $(s,t)\mapsto(x,y)$ maps a neighbourhood $U$ of $(s^*,0)$ $C^1$-diffeomorphically onto a neighbourhood $N$ of $(x_0(s^*),y_0(s^*))$. Let $(s,t)=(S(x,y),T(x,y))$ be the inverse, and define $Z(x,y)=z\big(S(x,y),T(x,y)\big)$, a $C^1$ function on $N$.

Step 3 (it is a solution). By construction $z(s,t)=Z\big(x(s,t),y(s,t)\big)$ on $U$. Differentiating in $t$: $z_t=Z_xx_t+Z_yy_t$, that is $R=P\,Z_x+Q\,Z_y$ at every point of the graph of $Z$. At $t=0$, $Z(x_0(s),y_0(s))=z_0(s)$: the data are satisfied.

Step 4 (uniqueness). If $\tilde Z$ is another solution with the same data near the point, then by the theorem of §3.1 that integral surfaces are woven from characteristics, the characteristic through $\Gamma(s)$ lies on the graph of $\tilde Z$ for small $|t|$ (uniformly for $s$ near $s^*$, by continuous dependence). Hence $\tilde Z(x(s,t),y(s,t))=z(s,t)=Z(x(s,t),y(s,t))$ for $(s,t)$ near $(s^*,0)$, and these points $(x,y)$ fill a neighbourhood of $(x_0(s^*),y_0(s^*))$. ∎

📐 Theorem 3.16 — Characteristic data: none or infinitely many
In the setting of Definition 3.6 suppose $J\equiv0$ on $I$ and $(P,Q)\ne(0,0)$ along $\Gamma$.
  1. If a $C^1$ solution of the Cauchy problem exists near $\Gamma_0$, then $\Gamma$ is a characteristic curve (its tangent is parallel to $(P,Q,R)$).
  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 $(x_0',y_0')=\lambda(s)\,(P,Q)$ for some scalar $\lambda(s)\ne0$. If $z$ solves the problem, differentiate $z(x_0(s),y_0(s))=z_0(s)$: $$z_0'=p\,x_0'+q\,y_0'=\lambda\,(pP+qQ)=\lambda R .$$ So $(x_0',y_0',z_0')=\lambda(P,Q,R)$: $\Gamma$ is tangent to the characteristic field, hence (by uniqueness for the characteristic ODE) a reparametrised characteristic. (2) Take a point $\Gamma(s^*)$ and a short curve $\tilde\Gamma$ through it whose projection is transversal to $(P,Q)$, carrying data $\tilde z_0$ with $\tilde z_0=z_0(s^*)$ at the common point but otherwise arbitrary. Theorem 3.15 gives a solution containing $\tilde\Gamma$; by the same theorem of §3.1 it contains the characteristic through $\Gamma(s^*)$, which is $\Gamma$ itself near that point. Different choices of $\tilde z_0$ give different solutions. ∎

🧭 Problem-solving strategy — solving a Cauchy problem by characteristics
  1. Parametrise the data: $x=x_0(s)$, $y=y_0(s)$, $z=z_0(s)$. Compute $J(s)=x_0'Q-y_0'P$ and note where it vanishes.
  2. Solve the characteristic ODEs $x_t=P$, $y_t=Q$, $z_t=R$ with these initial values: $x(s,t)$, $y(s,t)$, $z(s,t)$.
  3. Invert $(s,t)\mapsto(x,y)$ (often $t$ is one of the variables) and substitute into $z(s,t)$.
  4. Check the PDE and the data. Determine where the inversion is valid: the solution lives where $\partial(x,y)/\partial(s,t)\ne0$; where it vanishes, characteristics cross or touch, and the solution may break down.
✏️ Example 7 — Burgers' equation: a solution that breaks
Solve $z_y+z\,z_x=0$ (i.e. $z\,p+q=0$; think of $y$ as time) with $z(x,0)=-x$. Where does the solution exist? For general data $z(x,0)=g(x)$, find the first time $y^*\gt0$ at which characteristics cross.
Transversality $P=z$, $Q=1$, $R=0$; $\Gamma:\ (s,0,-s)$. $J=x_0'Q-y_0'P=1\cdot1-0=1\ne0$: unique local solution near the whole $x$-axis.
Characteristics $x_t=z$, $y_t=1$, $z_t=0$ with $(x,y,z)(s,0)=(s,0,-s)$: $z=-s$, $y=t$, $x=s-st=s(1-y)$.
Invert For $y\ne1$: $s=\dfrac{x}{1-y}$, so $$z=-\frac{x}{1-y}\qquad(y\lt1).$$ Check: $z_y=-\tfrac{x}{(1-y)^2}$, $zz_x=\tfrac{x}{(1-y)^2}$, sum $0$ ✓; $z(x,0)=-x$ ✓.
Breaking All characteristic lines $x=s(1-y)$ pass through $(0,1)$: at $y=1$ the profile has become vertical ($z_x=-\tfrac1{1-y}\to-\infty$). No $C^1$ solution exists on any strip $\mathbb{R}\times[0,Y)$ with $Y\gt1$.
General data Characteristics: $x=s+g(s)\,y$, $z=g(s)$, so $z=g(x-zy)$ implicitly. $\dfrac{\partial x}{\partial s}=1+g'(s)\,y$ first vanishes at $$y^*=\frac{-1}{\min g'}\quad\text{if }\min g'\lt0,$$ and never (for $y\ge0$) if $g'\ge0$. Here $g'\equiv-1$ gives $y^*=1$. Waves whose speed $z$ decreases in $x$ (fast behind slow) must break — the mechanism behind shock waves and traffic jams.
Interactive 2 — Transversality: When Do Cauchy Data Determine a Solution?

Cauchy data are given on the red segment through $A=(1,1)$ at angle $\alpha$. Grey: projected characteristics; small blue arrows: the direction $(P,Q)$ along the segment; blue curves: the characteristics leaving the segment, which sweep out the region where the data determine the solution. The right panel plots $J(\sigma)=x_0'Q-y_0'P$ along the segment. Where $J=0$ (amber) the segment touches a projected characteristic and uniqueness or existence may fail there; if $J\equiv0$ the segment is a projected characteristic. Press Play to rotate the segment.

100°
J at A—
zeros of J on the segment—
verdict—

Nonlinear equations: characteristic strips. For $F(x,y,z,p,q)=0$ the characteristics are the solutions of Charpit's equations written with a parameter $t$: $$x_t=F_p,\quad y_t=F_q,\quad z_t=pF_p+qF_q,$$ $$p_t=-F_x-pF_z,\quad q_t=-F_y-qF_z .$$ Each solution is a curve $(x,y,z)$ together with a tangent plane (slopes $p,q$) at each of its points — a characteristic strip. The same steps as in Theorem 3.15 give the following result.

📐 Theorem 3.17 — Cauchy problem for nonlinear equations (statement)
Let $F\in C^2$ and let $\Gamma=(x_0,y_0,z_0)(s)$ be $C^1$ data. Suppose $C^1$ functions $p_0(s),q_0(s)$ satisfy, at $s=s^*$ and hence nearby, $$F(x_0,y_0,z_0,p_0,q_0)=0,$$ $$z_0'=p_0\,x_0'+q_0\,y_0'\ \ (\text{strip condition}),$$ $$x_0'F_q-y_0'F_p\ne0 .$$ Then there is a unique solution near $(x_0(s^*),y_0(s^*))$ that contains $\Gamma$ and has slopes $(p_0,q_0)$ along it; it is the union of the characteristic strips issuing from the initial strip $(x_0,y_0,z_0,p_0,q_0)$.

The first two conditions determine $(p_0,q_0)$ only up to a finite choice, and each choice can give a different solution. For $p^2+q^2=1$ with $z=0$ on the $x$-axis, the strip condition gives $p_0=0$ and then $q_0=\pm1$: the two solutions $z=y$ and $z=-y$. The characteristics of $f(p,q)=0$ are straight lines ($p,q$ are constant along them, since $F_x=F_y=F_z=0$), which is why the rays in Interactive 1 are straight.

⚠️ Counterexamples and pitfalls
  1. A complete integral is not unique. $z=(x+a)(y+b)$ and $4az=(x+ay+b)^2$ are both complete integrals of $z=pq$ (Example 6).
  2. The discriminant surface need not be a solution. For $p^2+q^2+z-x=0$, $F_p=F_q=0$ gives $p=q=0$ and then $z=x$; but $z=x$ has $p=1$, and $1+0+x-x=1\ne0$. Always check.
  3. Not every equation has a singular integral. If $F$ involves only $p,q$ and $(F_p,F_q)\ne0$ whenever $F=0$ (e.g. $p+q=pq$, $p^2+q^2=1$), there is none.
  4. A singular integral is not a particular integral. $z=xy$ is not of the form $ax+by-ab$ for any constants $a,b$, though it solves $z=px+qy-pq$.
  5. A general integral needs the envelope. Putting $b=h(a)$ into a complete integral gives only a one-parameter family of particular integrals; the general integral is the envelope of that family.
  6. Charpit's integral must contain $p$ or $q$. An integral of $\frac{dx}{F_p}=\frac{dy}{F_q}$ that involves only $x,y,z$ gives $\partial(F,G)/\partial(p,q)=0$ and cannot be solved for $p,q$ together with $F=0$.
  7. Compatibility is tested on $F=G=0$, not identically. For $p=z$, $q=z$ the bracket is $p-q$, which vanishes only on the common zero set; the system is compatible ($z=ce^{x+y}$).
  8. Transversality is a local condition. Burgers' equation $z_y+zz_x=0$ with smooth data $z(x,0)=-x$ satisfies $J\equiv1$, yet the solution $-x/(1-y)$ breaks at $y=1$. Theorem 3.15 is about a neighbourhood of the data curve only.
  9. $J=0$ at a single point is not harmless. For $y\,p-x\,q=0$ with data on the line $x=1$, $J=-s$ vanishes at $(1,0)$; a solution near that point exists only if the data are even in $y$, and inside the unit circle nothing is determined.
  10. Nonlinear Cauchy problems may have several solutions even when transversal: $p^2+q^2=1$, $z=0$ on $y=0$ has $z=y$ and $z=-y$.

Exercises

I
Complete, general and singular integrals; compatibility.
  • 1 Show that $z=ax+by+a^2+b^2$ is a complete integral of $z=px+qy+p^2+q^2$, and find the singular integral.
  • 2 For the same equation, find the general integral corresponding to $b=a$.
  • 3 Find the singular integral of $z=px+qy+2\sqrt{1+p^2+q^2}$ and describe it geometrically.
  • 4 For $p^2+q^2+z-x=0$ eliminate $p,q$ from $F=F_p=F_q=0$ and show that the resulting surface is not a solution.
  • 5 Decide whether the pairs are compatible, and if so find the common solutions: (a) $p=x+y$, $q=x-y$; (b) $p=y$, $q=-x$.
  • 6 Show that $p=z$ and $q=z$ are compatible although $[F,G]$ is not identically $0$, and find the common solutions.
  • 7 From the complete integral $z=(x+a)(y+b)$ of $z=pq$, find the general integral corresponding to $b=a$ and verify it.
1 $p=a$, $q=b$ and $z=ax+by+a^2+b^2=px+qy+p^2+q^2$ ✓; rank: $\phi_{xa}=1$, $\phi_{yb}=1$, $\phi_{xb}=\phi_{ya}=0$, so the $2\times2$ minor is $1$. Envelope: $x+2a=0$, $y+2b=0$, so $$z=-\tfrac{x^2+y^2}{4}.$$ Check: $p=-x/2$, $q=-y/2$: $px+qy+p^2+q^2=-\tfrac{x^2+y^2}{2}+\tfrac{x^2+y^2}4=z$ ✓.
2 $z=a(x+y)+2a^2$; $\partial_a$: $x+y+4a=0$; envelope $$z=-\tfrac{(x+y)^2}{8}.$$ Check: $p=q=-\tfrac{x+y}4$: $px+qy+p^2+q^2=-\tfrac{(x+y)^2}{4}+\tfrac{(x+y)^2}{8}=z$ ✓.
3 Complete integral $z=ax+by+2s$, $s=\sqrt{1+a^2+b^2}$. Envelope: $x+2a/s=0$, $y+2b/s=0$, so $a=-xs/2$, $b=-ys/2$, $s^2=1+s^2\tfrac{x^2+y^2}{4}$, $s=\tfrac{2}{\sqrt{4-x^2-y^2}}$, and $z=-s\tfrac{x^2+y^2}{2}+2s=\sqrt{4-x^2-y^2}$. Singular integral: the hemisphere $x^2+y^2+z^2=4$, $z\gt0$. The planes $ax+by-z+2s=0$ are at distance $2s/\sqrt{a^2+b^2+1}=2$ from the origin: they are the tangent planes of the sphere of radius $2$.
4 $F_p=2p=0$, $F_q=2q=0$, then $z=x$. For $z=x$: $p=1$, $q=0$, $F=1+0+x-x=1\ne0$. Not a solution (Theorem 3.9 gives only a necessary condition).
5 (a) $[F,G]=\tfrac{\partial(F,G)}{\partial(x,p)}+\tfrac{\partial(F,G)}{\partial(y,q)}=1+(-1)=0$: compatible; $z_{xy}=1=z_{yx}$; common solutions $z=\tfrac{x^2}2+xy-\tfrac{y^2}2+c$. (b) $z_{xy}=1$, $z_{yx}=-1$; $[F,G]=-2\ne0$: not compatible.
6 $F=p-z$, $G=q-z$: $\tfrac{\partial(F,G)}{\partial(z,p)}=F_zG_p-F_pG_z=1$, $\tfrac{\partial(F,G)}{\partial(z,q)}=F_zG_q-F_qG_z=-1$, the others $0$: $[F,G]=p-q$, which is $0$ when $p=q=z$. Common solutions $z=c\,e^{x+y}$.
7 $z=(x+a)(y+a)$; $\partial_a$: $x+y+2a=0$; $z=-\tfrac{(x-y)^2}4$. Check: $p=-\tfrac{x-y}2$, $q=\tfrac{x-y}2$, $pq=-\tfrac{(x-y)^2}4=z$ ✓.
II
Charpit's method and the standard forms. Find a complete integral (and a singular integral where asked).
  • 1 $pq=1$.
  • 2 $p^2-q+2=0$.
  • 3 $z=px+qy+p^2q^2$; also the singular integral.
  • 4 $z=px+qy+\log(pq)$; also the singular integral.
  • 5 $z=p^2+q^2$.
  • 6 $p^2=z\,q$.
  • 7 $p-3x^2=q^2-y$.
  • 8 $y\,p^2=x\,q$ in the quadrant $x,y\gt0$.
  • 9 $z=px+q^2$ (use Charpit's equations).
1 Form 1: $ab=1$: $z=ax+\tfrac{y}{a}+c$.
2 Form 1: $b=a^2+2$: $z=ax+(a^2+2)y+c$.
3 Clairaut: $z=ax+by+a^2b^2$. Envelope: $x+2ab^2=0$, $y+2a^2b=0$. Then $xy=4a^3b^3$, so $ab=(xy/4)^{1/3}$, and $z=ax+by+a^2b^2=-2a^2b^2-2a^2b^2+a^2b^2=-3a^2b^2$: $$z=-3\left(\tfrac{xy}{4}\right)^{2/3},\quad\text{i.e.}\quad 16z^3+27x^2y^2=0 .$$
4 Clairaut: $z=ax+by+\log(ab)$ ($ab\gt0$). Envelope: $x+\tfrac1a=0$, $y+\tfrac1b=0$: $a=-\tfrac1x$, $b=-\tfrac1y$, and $$z=-2-\log(xy)\qquad(xy\gt0).$$ Check: $p=-\tfrac1x$, $q=-\tfrac1y$, $px+qy+\log(pq)=-2-\log(xy)$ ✓.
5 Form 3: $z=Z(X)$, $X=x+ay$: $Z=(1+a^2)Z'^2$, $Z'=\sqrt{Z/(1+a^2)}$, $2\sqrt Z=\tfrac{X+b}{\sqrt{1+a^2}}$: $$4(1+a^2)\,z=(x+ay+b)^2 .$$ (Singular integral: $F_p=2p$, $F_q=2q$ vanish at $p=q=0$, giving $z=0$, which is a solution.)
6 Form 3: $Z'^2=Z\cdot aZ'$, so $Z'=aZ$ (discarding $Z'=0$): $z=b\,e^{a(x+ay)}=b\,e^{ax+a^2y}$. Check: $p=az$, $q=a^2z$, $p^2=a^2z^2=zq$ ✓.
7 Form 4: $p-3x^2=a=q^2-y$: $p=3x^2+a$, $q=\sqrt{y+a}$: $$z=x^3+ax+\tfrac23(y+a)^{3/2}+b .$$
8 Form 4: $\tfrac{p^2}{x}=a=\tfrac{q}{y}$: $p=\sqrt{ax}$, $q=ay$: $$z=\tfrac23\sqrt a\,x^{3/2}+\tfrac a2y^2+b .$$
9 $F=px+q^2-z$: $F_x=p$, $F_z=-1$, so $F_x+pF_z=0$ and $dp=0$: $p=a$. Then $q=\sqrt{z-ax}$, and $dz=a\,dx+\sqrt{z-ax}\,dy$ gives $\tfrac{d(z-ax)}{\sqrt{z-ax}}=dy$, $2\sqrt{z-ax}=y+b$: $$z=ax+\tfrac{(y+b)^2}{4}.$$ Check: $px+q^2=ax+\tfrac{(y+b)^2}4=z$ ✓.
III
The Cauchy problem.
  • 1 Solve $z_x+2z_y=z$ with $z(x,0)=\sin x$.
  • 2 Solve $x\,p+y\,q=2z$ with $z(x,1)=x$.
  • 3 Solve $z_y+z\,z_x=0$ with $z(x,0)=2x+1$, and state where the solution is valid.
  • 4 For $z_y+z\,z_x=0$ with $z(x,0)=e^{-x^2}$, find the first $y^*\gt0$ at which characteristics cross.
  • 5 For $y\,p-x\,q=0$ with data $z(1,y)=g(y)$ on the line $x=1$, compute $J$, and show that a solution near $(1,0)$ exists only if $g$ is even near $0$.
  • 6 Solve $z_x+z_y=z^2$ with $z(x,0)=x$. Where does the solution blow up?
  • 7 Show that the Cauchy problem $x\,p+y\,q=z$, $z=1$ on $x^2+y^2=1$, is non-characteristic everywhere, and solve it.
  • 8 Find all solutions of $p^2+q^2=1$ with $z(x,0)=x/2$.
1 $x=s+t$, $y=2t$, $z=e^t\sin s$. So $t=y/2$, $s=x-y/2$: $$z=e^{y/2}\sin\!\left(x-\tfrac y2\right).$$ ($J=1\cdot2-0=2\ne0$.)
2 $u=y/x$, $v=z/x^2$; on $y=1$: $u=1/x$, $v=1/x=u$. So $z/x^2=y/x$: $z=xy$. Check: $xy+yx=2z$ ✓.
3 $x=s+(2s+1)y$, $z=2s+1$. Then $x+\tfrac12=(s+\tfrac12)(1+2y)$, so $z=2(s+\tfrac12)=\dfrac{2x+1}{1+2y}$, valid for $y\gt-\tfrac12$ (all lines meet at $(-\tfrac12,-\tfrac12)$). For $y\ge0$ it exists for all time: $g'=2\gt0$.
4 $g'(x)=-2xe^{-x^2}$ has minimum at $x=1/\sqrt2$: $\min g'=-\sqrt2\,e^{-1/2}$. Hence $$y^*=\frac{e^{1/2}}{\sqrt2}=\sqrt{e/2}\approx1.166 .$$
5 $\Gamma=(1,s,g(s))$: $x_0'=0$, $y_0'=1$, $P=y=s$, $Q=-x=-1$: $J=0\cdot(-1)-1\cdot s=-s$, zero only at $s=0$. Solutions are constant on circles; the circle $x^2+y^2=1+s^2$ meets $x=1$ at $y=\pm s$, so $g(s)=g(-s)$ is necessary for any solution on a neighbourhood of $(1,0)$ (for small $s$ both points are near $(1,0)$, and the arc joining them through $(1+\dots,0)$ stays near it). Moreover the values inside the unit circle are not determined at all.
6 $x=s+t$, $y=t$, $z_t=z^2$, $z(0)=s$: $z=\tfrac{s}{1-st}$. With $s=x-y$, $t=y$: $$z=\frac{x-y}{1-y(x-y)} .$$ It blows up on the curve $y(x-y)=1$; for example along $x-y=1$ at $y=1$.
7 $\Gamma=(\cos s,\sin s,1)$: $J=x_0'Q-y_0'P=(-\sin s)\sin s-\cos s\cos s=-1\ne0$. With $u=y/x$, $v=z/x$ (or by homogeneity): $z=\sqrt{x^2+y^2}$ for $(x,y)\ne0$.
8 Strip: $p_0=z_0'=\tfrac12$, and $p_0^2+q_0^2=1$ gives $q_0=\pm\tfrac{\sqrt3}2$. The characteristics are straight lines with constant $(p,q)$ (Form 1), so $$z=\tfrac x2\pm\tfrac{\sqrt3}2\,y .$$ Two solutions, both planes. ($x_0'F_q-y_0'F_p=2q_0\ne0$: transversal.)
IV
Proofs and challenges.
  • 1 Prove that an equation $f(p,q)=0$ with $\nabla f\ne0$ on the set $f=0$ has no singular integral.
  • 2 Write Charpit's equations for $f(p,q)=0$ and show that the projected characteristics are straight lines along which $p$ and $q$ are constant.
  • 3 Show that the singular integral of a Clairaut equation $z=px+qy+f(p,q)$, when it exists, is a surface whose tangent planes are exactly the planes of the complete integral.
  • 4 Let $g\in C^1(\mathbb{R})$ be bounded with $g'\ge0$. Prove that $z_y+zz_x=0$, $z(x,0)=g(x)$, has a $C^1$ solution on $\mathbb{R}\times[0,\infty)$.
  • 5 Let $g\in C^1(\mathbb{R})$ with $g'(s_0)\lt0$. Prove that $z_y+zz_x=0$, $z(x,0)=g(x)$, has no $C^1$ solution on $\mathbb{R}\times[0,Y)$ for $Y\gt-1/g'(s_0)$.
  • 6 Use Jacobi's method to find a complete integral of $x_1p_1^2+x_2p_2^2=p_3$ ($x_1,x_2\gt0$).
1 By Theorem 3.9 a singular integral satisfies $F=F_p=F_q=0$ at its points; here $F_p=f_p$, $F_q=f_q$, so it would need $f=0$ and $\nabla f=0$ simultaneously, which the hypothesis excludes.
2 $\dfrac{dx}{f_p}=\dfrac{dy}{f_q}=\dfrac{dz}{pf_p+qf_q}=\dfrac{dp}0=\dfrac{dq}0$. So $p,q$ are constant along a characteristic, hence so are $f_p,f_q$, and $(x,y)$ moves with constant velocity $(f_p,f_q)$: a straight line; $z$ grows linearly along it.
3 If $Z=\phi(x,y,\alpha,\beta)$ with $\phi=ax+by+f(a,b)$, then by the proof of Theorem 3.8 $Z_x=a$, $Z_y=b$ at the point where the parameters are $(a,b)=(\alpha,\beta)$; the tangent plane there is $z-Z=a(x'-x)+b(y'-y)$, which is the plane $z=ax'+by'+f(a,b)$ (it passes through the point and has the same slopes). Conversely every plane of the family touching the envelope does so at the point where $x=-f_a$, $y=-f_b$.
4 Characteristics $x=X(s,y)=s+g(s)\,y$, $z=g(s)$. For $y\ge0$, $X_s=1+g'(s)y\ge1$, so $s\mapsto X(s,y)$ is strictly increasing, and since $g$ is bounded, $X\to\pm\infty$ as $s\to\pm\infty$: it is a $C^1$ bijection of $\mathbb{R}$ with $C^1$ inverse $S(x,y)$ (inverse function theorem). $z(x,y)=g(S(x,y))$ is $C^1$ and solves the equation (Theorem 3.15, Step 3, applied globally).
5 Suppose $z$ is a $C^1$ solution on $\mathbb{R}\times[0,Y)$. By Theorem 3.15 and the characteristic theorem of §3.1, $z$ is constant on the characteristic lines $x=s+g(s)y$ while they stay in the strip. Along the line from $s_0$, differentiate $z(s+g(s)y,y)=g(s)$ in $s$: $z_x\cdot(1+g'(s)y)=g'(s)$, so $z_x=\dfrac{g'(s_0)}{1+g'(s_0)y}\to-\infty$ as $y\uparrow-1/g'(s_0)\lt Y$, contradicting continuity of $z_x$ on the strip.
6 $F=x_1p_1^2+x_2p_2^2-p_3$. Jacobi: $\frac{dp_3}{-F_{x_3}}=\frac{dp_3}{0}$, so $p_3=a$. From $\frac{dx_1}{2x_1p_1}=\frac{dp_1}{-p_1^2}$: $p_1\,dx_1+2x_1\,dp_1=0$, i.e. $d(x_1p_1^2)=0$: $x_1p_1^2=b$. Then $x_2p_2^2=a-b$, and $$u=2\sqrt{b\,x_1}+2\sqrt{(a-b)\,x_2}+a\,x_3+c .$$ Check: $x_1p_1^2+x_2p_2^2=b+a-b=a=p_3$ ✓.
🎯 Exam corner CSIR Part BCSIR Part CGATE MA

E1. CSIR Part B GATE MA The singular integral of $z=px+qy+3pq$ is (A) $z=\tfrac{xy}{3}$; (B) $3z+xy=0$; (C) $z=3xy$; (D) there is no singular integral.

Idea Clairaut form: complete integral $z=ax+by+3ab$; take its envelope.
Envelope $x+3b=0$, $y+3a=0$: $a=-\tfrac y3$, $b=-\tfrac x3$, $z=-\tfrac{xy}3-\tfrac{xy}3+\tfrac{xy}3=-\tfrac{xy}{3}$. Check: $p=-\tfrac y3$, $q=-\tfrac x3$, $px+qy+3pq=-\tfrac{2xy}3+\tfrac{xy}3=-\tfrac{xy}3$ ✓.
Answer (B).

E2. CSIR Part B A complete integral of $p-q=x-y$ is (A) $z=\tfrac{x^2+y^2}{2}+a(x+y)+b$; (B) $z=\tfrac{x^2-y^2}{2}+a(x-y)+b$; (C) $z=xy+ax+b$; (D) $z=a(x+y)+b$.

Idea Separable: $p-x=q-y=a$.
Integrate $p=x+a$, $q=y+a$: $z=\tfrac{x^2}2+ax+\tfrac{y^2}2+ay+b$. Check (B): $p-q=x+a-(-y-a)=x+y+2a\ne x-y$; (C): $p-q=y+a-x$; (D): $p-q=0$.
Answer (A).

E3. CSIR Part B For $z_y+z\,z_x=0$, $z(x,0)=-\dfrac{2x}{1+x^2}$, the smallest $y\gt0$ at which characteristics intersect is (A) $\tfrac14$; (B) $\tfrac12$; (C) $1$; (D) $2$.

Idea $y^*=-1/\min g'$ (Example 7).
Minimum $g'(x)=-\dfrac{2(1-x^2)}{(1+x^2)^2}$. With $w=x^2$, $\dfrac{1-w}{(1+w)^2}$ is decreasing for $0\le w\le3$ and then negative, so its maximum is $1$ at $w=0$: $\min g'=g'(0)=-2$.
Answer $y^*=\tfrac12$: (B).

E4. CSIR Part C For the equation $z=pq$, which of the following are true? (One or more options may be correct.) (A) $z=(x+a)(y+b)$ is a complete integral. (B) $z=0$ is a singular integral. (C) Every solution is of the form $(x+a)(y+b)$ for some constants $a,b$. (D) $z=\tfrac{(x+y)^2}{4}$ is a solution.

(A) True $p=y+b$, $q=x+a$, $pq=z$; the minor $\begin{vmatrix}\phi_{xa}&\phi_{ya}\\ \phi_{xb}&\phi_{yb}\end{vmatrix}=\begin{vmatrix}0&1\\1&0\end{vmatrix}=-1\ne0$.
(B) True Envelope: $y+b=x+a=0$ gives $z=0$; $F_p=q=0$, $F_q=p=0$; and $z\equiv0$ solves $z=pq$.
(C) False $z=-\tfrac{(x-y)^2}4$ (Exercise I.7) is a solution not of this form.
(D) True $p=q=\tfrac{x+y}2$, $pq=\tfrac{(x+y)^2}4=z$ (it is $4az=(x+ay+b)^2$ with $a=1$, $b=0$).
Answer (A), (B), (D).

E5. CSIR Part C Consider $p^2+q^2=1$. Which of the following are true? (A) $z=x\cos\alpha+y\sin\alpha+c$ is a complete integral. (B) $z=\sqrt{x^2+y^2}$ is a solution on $\mathbb{R}^2\setminus\{(0,0)\}$. (C) The equation has a singular integral. (D) There is a $C^1$ solution on $\mathbb{R}^2$ that vanishes on the unit circle.

(A) True Form 1 with $a=\cos\alpha$, $b=\sin\alpha$; parameters $\alpha,c$: $\phi_c=1$, $\phi_{x\alpha}=-\sin\alpha$, $\phi_{y\alpha}=\cos\alpha$, so the rows $(\phi_\alpha,-\sin\alpha,\cos\alpha)$ and $(1,0,0)$ are independent.
(B) True $p=x/r$, $q=y/r$, $p^2+q^2=1$.
(C) False $F_p=2p$, $F_q=2q$ vanish only at $p=q=0$, where $F=-1\ne0$ (Exercise IV.1).
(D) False Such a $z$ is continuous on the closed unit disc, so it attains a maximum and a minimum there. An interior extremum would have $\nabla z=0$, impossible since $|\nabla z|=1$. So both are attained on the circle, where $z=0$: $z\equiv0$ on the disc, and then $\nabla z=0$ inside — a contradiction.
Answer (A), (B).

E6. CSIR Part C GATE MA Consider the Cauchy problem $z_y+z\,z_x=0$, $z(x,0)=x$. Which of the following are true? (A) $z=\dfrac{x}{1+y}$ for $y\gt-1$. (B) A $C^1$ solution exists on $\mathbb{R}\times[0,\infty)$. (C) $z$ is constant along each line $x=s(1+y)$. (D) The projected characteristics through distinct points of the $x$-axis meet at some point with $y\gt0$.

(A) True Characteristics $x=s+sy$, $z=s$; so $s=\tfrac{x}{1+y}$. Check: $z_y+zz_x=-\tfrac{x}{(1+y)^2}+\tfrac{x}{(1+y)^2}=0$.
(B) True $g'=1\gt0$ (Exercise IV.4, or directly from (A)).
(C) True These are the characteristic lines, and $z=s$ on each.
(D) False The lines $x=s(1+y)$ all pass through $(0,-1)$ only; for $y\gt-1$ distinct lines are disjoint (the rarefaction fans out instead of breaking).
Answer (A), (B), (C).
🧑‍🏫 For the teacher

Suggested plan — 3 lectures × 60 min.

  1. Kinds of integrals. Recall the sphere family of §3.1 and turn it round: Definition 3.4, Theorem 3.8 (the "touching" argument is the whole proof), Theorem 3.9 and Example 1. Draw the rolling-ball cylinder on the board. Pitfall 2 as a warning. Then Theorem 3.14(1)–(2) and Example 5(a), (b).
  2. Charpit. Theorem 3.10 (necessity in two lines, the $w_y=\psi_zw$ trick for sufficiency), Theorem 3.11 and Example 2, then Charpit's equations as "the characteristic equations of $[F,G]=0$" and Example 3. Standard forms (3), (4) with Example 5(c) and Example 6 — let the class discover the second complete integral of $z=pq$. Jacobi in ten minutes with Example 4.
  3. The Cauchy problem. Definition 3.6, Theorem 3.15 with the picture of the $(s,t)$-grid being mapped to the plane; project Interactive 2 (circles case: rotate until $J$ has a zero on the segment; rays case: $J$ is constant along the segment!). Theorem 3.16, Example 7 and the teaser of the chapter introduction for breaking; finish with Interactive 1 and the remark on strips, then E4–E6 live.

Discussion questions. (1) Why is the envelope of solutions a solution, while the envelope of a family of non-solutions need not be anything? (2) In Example 6, two "complete" integrals look completely different. In what sense do they contain the same solutions? (3) Burgers' equation has a smooth solution near the initial line for any smooth data, yet it can break. Which hypothesis of Theorem 3.15 tells you "near"?

Misconceptions to watch for. Treating the discriminant eliminant as automatically a solution; putting $b=h(a)$ and forgetting the envelope; using an integral of Charpit's equations that contains neither $p$ nor $q$; checking compatibility identically instead of on $F=G=0$; believing that a transversal Cauchy problem has a global solution; assuming nonlinear Cauchy problems have unique solutions.

Suggested homework. I(1, 3, 4, 6), II(3, 5, 6, 8, 9), III(1, 3, 4, 6, 8), IV(2, 4, 5); Exam corner E1, E3, E5.

🔬 Research–Teaching NexusResearch-tutored · optional, about 45 minutes

Research question. A grass fire starts along a line of burning brush and spreads at a fixed speed in every direction. When does it reach a given farmhouse, and where on the fire front do sharp corners appear?

Why it matters. Fire-spread simulators used in Australia, California and the Mediterranean, seismic travel-time tomography and the "fast marching" methods of computer vision all solve the eikonal equation $|\nabla T|=1/v$ for an arrival time $T(x,y)$. Its characteristics are the rays of geometrical optics, and the corners of the front are where rays cross — the caustics of Interactive 1.

What to do.

  1. With constant speed $v$ and the fire starting at time $0$ on a curve $\Gamma$, show that $T=d/v$, where $d$ is the distance to $\Gamma$, solves $T_x^2+T_y^2=1/v^2$ wherever it is differentiable, and that the level curves of $T$ are the wavefronts of Interactive 1.
  2. Take $\Gamma$ to be the parabola $y=x^2/2$ (burning inside, i.e. upwards) and $v=0.5\ \text{m/s}$, lengths in units of $100\ \text{m}$. Using Charpit's equations for $p^2+q^2=1$, show that the ray from $(s,s^2/2)$ is $x=s-\tfrac{s\,c}{\sqrt{1+s^2}}$, $y=\tfrac{s^2}2+\tfrac{c}{\sqrt{1+s^2}}$ at distance $c$.
  3. Find where rays first cross (the cusp of the wavefront) and the time at which the fire first develops a sharp corner. Compute the arrival time at a farmhouse at $(0,3)$.
  4. Explore numerically: replace $v$ by a speed depending on direction (wind). Which part of the theory survives?

Tools: Interactive 1 of this section, pen and paper, and a few lines of Python to draw rays.

Hand in: the ray formula, the crossing distance and time, the arrival time at the farmhouse, and one figure of rays and fronts.

Going further. Read about viscosity solutions (Crandall and Lions), the notion of solution that selects the physically correct non-smooth front after rays cross, and about anisotropic fire-spread models.

Instructor note: the parabola $y=x^2/2$ has radius of curvature $1$ at its vertex, the minimum over the curve, so rays first cross at distance $c=1$ (at the centre of curvature $(0,1)$ of the vertex, the cusp of the evolute $27x^2=8(y-1)^3$): at time $100\ \text{m}/0.5\ \text{m/s}=200\ \text{s}$. The farmhouse at $(0,3)$ lies on the axis; the nearest point of the parabola is found from $s^2=2(3-1)=4$, $s=\pm2$, at distance $\sqrt{4+(2-3)^2}=\sqrt5\approx2.236$ units, i.e. $223.6\ \text{m}$, reached after about $447\ \text{s}$ (the vertical ray from the vertex has length $3\gt\sqrt5$ and is not the first arrival — a nice illustration of crossing rays).

🧪 Self-check quiz

0/5 correct

Q1. The singular integral of $z=px+qy-pq$ is

Q2. In Charpit's method for $F(x,y,z,p,q)=0$, the fraction paired with $dp$ is

Q3. A complete integral of $f(p,q)=0$ has the form

Q4. For the Cauchy problem $Pp+Qq=R$ with data on $(x_0(s),y_0(s),z_0(s))$, a unique local solution is guaranteed when

Q5. For $z_y+zz_x=0$ with $z(x,0)=-3x$, the solution exists for $0\le y\lt Y$ with the largest possible $Y$ equal to