First order equations

[Math 619 home]

Contents

The constant velocity transport equation — characteristics

We consider the equation

ut+cux=0.(CVT)\hspace{100pt} \frac{\partial u}{\partial t} +c \frac{\partial u}{\partial x} =0. \hspace{100pt}\textrm{(CVT)}

If u(x,t)u(x, t) is a continuously differentiable solution, then by the several variable chain rule

du(x0+ct,t)dt=uxd(x0+ct)dt+ut=cux+ut=0\frac{du(x_0+ct, t)}{dt} = \frac{\partial u}{\partial x}\frac{d(x_0+ct)}{dt} + \frac{\partial u}{\partial t}=cu_x+u_t=0

for any xo,tRx_o, t\in \mathbb{R}. This means that any solution is constant along the lines x(t)=x0+ct  (tR)x(t)=x_0+ct \;(t\in\mathbb{R}). These lines are called the characteristics of the equation.

The initial value problem

Suppose we are given the values u(x,0)=F(x)u(x, 0) = F(x) of the solution at time t=0t=0 for all xRx\in\mathbb{R}. Then for any (x,t)R2(x,t)\in\mathbb{R}^2 one has

u(x,t)=u(xct,0)=F(xct).u(x, t) = u(x-ct, 0) = F(x-ct).

In other words, if there is a solution with the prescribed initial values then it must be u(x,t)=F(xct)u(x, t) = F(x-ct).

On the other hand, if F:RRF:\mathbb{R}\to\mathbb{R} is C1C^1 then you can verify by substituting that u(x,t)=F(xct)u(x, t)=F(x-ct) satisfies the transport equation (CVT), i.e. ut+cux=0u_t+cu_x=0.

Solutions without derivatives

Much of the modern theory of partial differential equations deals with “generalized solution” of one kind or another. Here is an example that begins to show why one would want to do that.

Consider the solutions with initial values un(x,0)=arctan(nx)u_n(x, 0) = \arctan(nx). The solutions are

un(x,t)=arctan(n(xct)).u_n(x, t) = \arctan(n(x-ct)).

What happens if we let nn\to\infty? We get

u(x,0)=limnun(x,0)=limnarctannx={+π/2(x>0),π/2(x<0),u_\infty(x, 0) = \lim_{n\to\infty} u_n(x, 0) = \lim_{n\to\infty} \arctan{nx} =\begin{cases} +\pi/2 &(x\gt0), \\ -\pi/2 & (x\lt0), \end{cases}

and

u(x,t)=limnun(x,t)=limnarctann(xct)={+π/2(x>ct),π/2(x<ct),u_\infty(x, t) = \lim_{n\to\infty} u_n(x, t) = \lim_{n\to\infty} \arctan{n(x-ct)} =\begin{cases} +\pi/2 &(x\gt ct), \\ -\pi/2 & (x\lt ct), \end{cases}

The function u(x,t)u_\infty(x, t) is the limit of solutions, but it is not continuous and therefore not differentiable because it has a jump discontinuity at x=ctx=ct. It can't be a solution of the PDE ut+cux=0u_t+cu_x=0 because the derivatives in the equation aren't defined at all (x,t)(x, t).

Nevertheless, uu_\infty is a function of the form F(xct)F(x-ct) and it arises as the limit of actual solutions, so we would like to call it a solution. The discontinuous solution uu_\infty is itself not a solution of the PDE, but it does give a good description of all smooth solutions that are close to uu_\infty.

This leads to the question: how do we change the definition of solution to allow for discontinuous solutions?

Method of characteristics

There is a general method, called the method of characteristics for finding the solutions to a first order equation of the form

ut+a(x,t)ux=f(x,t,u)(1)\frac{\partial u}{\partial t} + a(x, t) \frac{\partial u}{\partial x} = f(x, t, u) \tag{1}

in which a:R2Ra:\mathbb{R}^2\to\mathbb{R} and f:R3Rf:\mathbb{R}^3\to\mathbb{R} are continuously differentiable functions.

Definition. A characteristic for the equation (1) is a function xc:[t0,t1]Rx_c:[t_0,t_1]\to\R that is defined on some interval [t0,t1]R[t_0,t_1]\subset\R and that satisfies the characteristic equation

dxc(t)dt=a(xc(t),t)for all t[t0,t1](2)\frac{dx_c(t)}{dt} = a\bigl(x_c(t), t\bigr) \qquad \text{for all }t\in [t_0, t_1]\tag{2}

Often the graph of xcx_c, i.e. the curve {(t,xc(t))R2t0tt1}\{(t, x_c(t))\in\R^2 \mid t_0\leqslant t\leqslant t_1\} is also called a characterstic of the equation (1).

Theorem. If xc:[t0,t1]Rx_c:[t_0, t_1]\to\R is a characteristic of equation (1) then the function uc(t)=u(x(t),t)u_c(t)=u(x(t), t) satisfies

uc(t)=f(xc(t),t,uc(t)).(3)u_c'(t) = f\bigl(x_c(t), t, u_c(t)\bigr).\tag{3}

In the special case where f(x,t,u)=0f(x, t, u)=0 for all (x,t,u)(x, t, u) it follows that uc(t)=uc(t0)u_c(t)=u_c(t_0) for all t[t0,t1]t\in[t_0,t_1], i.e. if the right hand side ff in the equation (1) is just 00, then any C1C^1 solution uu is constant along a characteristic.

Note that uc(t)u_c(t) is a function of one variable and (3) is an ordinary differential equation.

Proof of the Theorem. To derive (3) let xc:RRx_c:\mathbb{R}\to\mathbb{R} be a characteristic of the PDE (1), so that xc:RRx_c:\mathbb{R}\to\mathbb{R} satisfies (2). Then the several variable chain rule implies

du(xc(t),t)dt=ux(xc(t),t)xc(t)+ut(xc(t),t).\frac{du(x_c(t), t)}{dt} = u_x\bigl(x_c(t), t\bigr)\cdot x_c'(t) + u_t\bigl(x_c(t), t\bigr).

Since uu satisfies (1), we can write this as

du(xc(t),t)dt=uxxc(t)a(xc(t),t)ux+f(xc(t),t,u(xc(t),t)).\frac{du(x_c(t), t)}{dt} = u_x \cdot x_c'(t) - a(x_c(t), t)\cdot u_x + f(x_c(t), t, u(x_c(t), t)).

The characteristic equation (2) implies xc(t)=a(xc(t),t)x_c'(t)=a(x_c(t), t) so the first two terms on the left cancel. This proves (3).

We next show that if f(x,t,u)=0f(x, t, u)=0 for all x,t,ux, t, u, then solutions are constant along characteristics. If f=0f=0 and if xc:[t0,t1]Rx_c:[t_0, t_1]\to\R is a characteristic, then (3) implies that uc(t)=u(xc(t),t)u_c(t) = u(x_c(t), t) is a function of one variable that is differentiable for all t[t0,t1]t\in[t_0, t_1], and that satisfies uc(t)=0u_c'(t)=0 for all t[t0,t1]t\in [t_0,t_1]. This implies that uc(t)u_c(t) does not depend on tt, i.e. uc(t)=uc(t0)u_c(t)=u_c(t_0) for all t[t0,t1]t\in[t_0, t_1].                  Q.E.D.

The recipe for solving initial value problems for 1st order PDE.

We can use the above Theorem to solve an initial value problem for the partial differential equation (1).

Special case where the right hand side f(x,t,u)=0f(x, t, u)=0.
In this case (1) is the equation ut+a(x,t)ux=0u_t+a(x, t)u_x=0, and we can proceed as follows.
If we are given the initial values u0(x)=u(x,0)u_0(x) = u(x, 0) of a solution for all xRx\in\mathbb{R}, and we want to find value of the solution uu at some point (x1,t1)(x_1, t_1) in space and time, then

xc(t)=a(xc(t),t),xc(t1)=x1x_c'(t) = a(x_c(t), t), \qquad x_c(t_1) = x_1

u(x1,t1)=u(xc(t1),t1)=u(xc(0),0)=u0(xc(0)).u(x_1, t_1) = u\bigl(x_c(t_1), t_1\bigr)= u\bigl(x_c(0), 0\bigr) = u_0(x_c(0)).

General case with arbitrary right hand side f(x,t,u)f(x, t, u).

xc(t)=a(xc(t),t),xc(t1)=x1x_c'(t) = a(x_c(t), t), \qquad x_c(t_1) = x_1

uc(t)=f(xc(t),t,uc(t)),uc(0)=u0(xc(0)).u_c'(t) = f(x_c(t), t, u_c(t)), \qquad u_c(0) = u_0(x_c(0)).

u(x1,t1)=u(xc(t1),t1)=uc(t1).u(x_1, t_1) = u\bigl(x_c(t_1), t_1\bigr)= u_c(t_1).

Equations for the derivatives of a solution

If uu is a solution of

ut+sin(x)ux=u2u_t+\sin(x)u_x=u^2

then the derivative v=uxv=u_x also satisfies a partial differential equation. We get this equation by differentiating the equation for uu with respect to xx on both sides:

(ut+sin(x)ux)x=u2x    utx+cos(x)ux+sin(x)uxx=2uux\frac{\partial \bigl(u_t+\sin(x)u_x\bigr)}{\partial x} = \frac{\partial u^2}{\partial x} \implies\frac{\partial u_t}{\partial x}+\cos(x)\frac{\partial u}{\partial x}+ \sin(x)\frac{\partial u_x}{\partial x} =2u\frac{\partial u}{\partial x}

Rewrite the second order derivatives that appeared as derivatives of vv using

uxx=vxand utx=2uxt=2utx=uxt=vt.\frac{\partial u_x}{\partial x} = \frac{\partial v}{\partial x}\qquad \text {and }\qquad \frac{\partial u_t}{\partial x}= \frac{\partial^2 u}{\partial x\partial t}= \frac{\partial^2 u}{\partial t\partial x} =\frac{\partial u_x}{\partial t} = v_t.

We get

vt+cos(x)ux+sin(x)vx=2uuxv_t + \cos(x)\frac{\partial u}{\partial x}+\sin(x)\frac{\partial v}{\partial x} = 2u \frac{\partial u}{\partial x}

Finally, replace every ux\frac{\partial u}{\partial x} by vv:

vt+sin(x)vx=2uvcos(x)vv_t + \sin(x) v_x = 2uv - \cos(x) v

A nonlinear equation

Consider the so-called inviscid Burger's equation

ut+uux=0.\frac{\partial u}{\partial t} + u\frac{\partial u}{\partial x} = 0.

If u(x,t)u(x, t) is a C1C^1 solution then any level set of uu on which ux0u_x\neq 0 is a straight line.

Suppose a level set is a graph x=x(t)x=x(t), i.e. suppose that for some function x=x(t)x=x(t) one has u(x(t),t)=cu(x(t), t) = c for all tt. Then

0=dcdt=du(x(t),t)dt=ux(x(t),t)x(t)+ut(x(t),t)=ux(x(t),t)x(t)u(x(t),t)ux(x(t),t)=ux(x(t),t)x(t)cux(x(t),t)=(x(t)c)ux(x(t),t)}    x(t)=c.\left. \begin{aligned} 0& = \frac{dc}{dt}=\frac{du(x(t), t)}{dt} \\ &= u_x(x(t), t)x'(t) + u_t(x(t), t) \\ &= u_x(x(t), t)x'(t) - u(x(t), t) u_x(x(t), t)\\ &= u_x(x(t), t)x'(t) - c u_x(x(t), t)\\ &= \bigl(x'(t) - c \bigr) u_x(x(t), t) \end{aligned} \right\} \quad \implies x'(t) = c.

Problems

Suppose u:[0,)×[0,)Ru:[0,\infty)\times[0, \infty) \to \mathbb{R} is a C1C^1 solution of

ut+xux=0.u_t +x u_x = 0.

  1. Find the characteristics of the equation.
  2. If you are given the initial condition u(x,0)=u0(x)u(x, 0) = u_0(x) for all x[0,)x\in [0, \infty), can you compute u(x,t)u(x, t) for all x,t0x, t\geq 0?
  3. Show that the function v(x,t)=ux(x,t)v(x, t) = u_x(x, t) satisfies the partial differential equation vt+xvx=vv_t+xv_x=-v.
  4. Show that along a characteristic x(t)x(t) of the equation for vv one has ux(x(t),t)=AeBtu_x(x(t), t) = Ae^{Bt} and find A,BA,B.

Solutions:

1: The characteristic equation is dxdt=x\frac{dx}{dt}=x, so the characteristics are xc(t)=xc(0)etx_c(t) = x_c(0)e^t.

2: To compute u(x,t)u(x, t) consider the characteristic for which xc(t)=xx_c(t) = x. For this characteristic we have x=xc(t)=xc(0)etx=x_c(t)=x_c(0)e^t and therefore xc(0)=xetx_c(0) = xe^{-t}.
Along the characteristic we have ddtu(xc(t),t)=0\frac{d}{dt}u(x_c(t), t)=0, so u(xc(t),t)=u(xc(0),0)u(x_c(t), t)=u(x_c(0), 0). Substitute xc(t)=xx_c(t)=x and xc(0)=xetx_c(0)=xe^{-t} and we get u(x,t)=u(xet,0)=u0(xet)u(x, t) = u(xe^{-t}, 0)=u_0(xe^{-t}).

3: Differentiate the equation ut+xux=0u_t+xu_x=0 with respect to xx:

(ut+xux)x=0    utx+xxux+xuxx=0(use utx=uxt)    (ux)t+v+xvx=0(use ux=v,uxx=vx)    vt+xvx=v.\begin{aligned} \frac{\partial (u_t+xu_x)}{\partial x}=0&\implies u_{tx} + \frac{\partial x}{\partial x} u_x + x u_{xx} =0 & (\text{use } u_{tx}=u_{xt})\\ \implies&(u_x)_t + v + x v_x=0 & (\text{use } u_x=v, u_{xx}=v_x)\\ \implies& v_t + xv_x=-v. \end{aligned}

4: The characteristics for the vv equation are determined by dxdt=x\frac{dx}{dt}=x, so they are the same as the characteristics for the equation for uu. They are given by xc(t)=xc(0)etx_c(t) = x_c(0)e^t. Along a characteristic we have dv(xc(t),t)dt=v(xc(t),t)\frac{d v(x_c(t), t)}{dt} = -v(x_c(t), t). Solving this ordinary differential equation we get v(xc(t),t)=v(xc(0),0)etv(x_c(t), t) = v(x_c(0), 0)e^{-t}, i.e.

ux(xc(t),t)=ux(xc(0),0)et.u_x(x_c(t), t) = u_x(x_c(0),0)e^{-t}.

Suppose u:[0,)×[0,)Ru:[0,\infty)\times[0, \infty) \to \mathbb{R} is a C1C^1 solution of

ut=x2ux.u_t = x^2 u_x.

  1. Find the characteristics of the equation for uu.
  2. If you are given the initial condition u(x,0)=u0(x)u(x, 0) = u_0(x) for all x[0,)x\in [0, \infty), can you compute u(x,t)u(x, t) for all x,t0x, t\geq 0?

Solution

The Partial Differential Equation for uu is utx2ux=0u_t-x^2u_x=0, so the differential equation for the characteristics is dxdt=x2\frac{dx}{dt}=-x^2… note the minus sign!
Solving this equation we get:

1x2dxdt=1    d  1xdt=t    1x=t+C    xc(t)=1t+C.-\frac{1}{x^2}\frac{dx}{dt}=1\implies \frac{d\; \dfrac{1}{x}}{dt}=t \implies \frac{1}{x}=t+C\implies x_c(t) = \frac{1}{t+C}.

To find the value of u(x,t)u(x, t) for given x,tx, t we need to know the characteristic for which xc(t)=xx_c(t)=x. The constant CC and the initial value xc(0)x_c(0) for this characteristic follow from

x=1t+C    C=1xt    xc(0)=10+C=11/xt=x1xt.x=\frac{1}{t+C} \implies C=\frac{1}{x}-t \implies x_c(0) = \frac{1}{0+C}= \frac{1}{1/x - t} = \frac{x}{1-xt}.

Along the characteristic we have

du(xc(t),t)dt=0(†)\frac{d u(x_c(t'), t')}{dt'} = 0\tag{\dag}

and we would like to conclude that u(xc(t),t)=u(xc(0),0)u(x_c(t), t) = u(x_c(0), 0). But we can only conclude this if we know that ()(\dag) holds for all t[0,t]t'\in[0, t]. The equation ()(\dag) holds at all tt' at which the characteristic xc(t)x_c(t') is defined. The characteristic xcx_c is given by

xc(t)=1t+Cx_c(t') = \frac{1}{t'+C}

so it is defined for all t[0,t]t'\in[0, t] if t+C0t'+C\neq 0 for all t[0,t]t'\in[0, t]. This holds if either C>0C>0 or C<tC<-t. Using C=1xtC=\dfrac{1}{x}-t we now check for which x,tx, t this is true.

It is given that x>0x>0 so C=1xt>tC=\dfrac{1}{x}-t>-t is true. We still have to see when C>0C>0. This holds if 1x>t\dfrac{1}{x}>t, i.e. if xt<1xt<1.

We can therefore only determine the value of u(x,t)u(x, t) if xt<1xt<1 (assuming x,t0x, t\geq 0).

About the inviscid Burger’ equation.

Let u:R×[0,T)Ru:\mathbb{R}\times[0, T)\to\mathbb{R} be a C2C^2 solution to the inviscid Burger's equation. Let aRa\in\mathbb{R} be given, and define c=u(a,0)c=u(a, 0).

  1. What can you say about the level set {(x,t)u(x,t)=c}\{(x, t) \mid u(x, t)=c\}?
  2. Show that the function v(x,t)=ux(x,t)v(x, t) = u_x(x, t) satisfies the partial differential equation

    vt+uvx+v2=0.v_t+uv_x+v^2=0.

  3. Show that S(t)=ux(a+ct,t)S(t) = u_x(a+ct,t) satisfies S(t)=S(t)2S'(t)=-S(t)^2.
  4. Assume that u(x,0)=21+exu(x, 0)=\dfrac{2}{1+e^x} and compute ux(t,t)u_x(t,t) for all t>0t>0; show that there is a T>0T>0 such that

limtTux(t,t)=.\lim_{t\nearrow T} u_x(t, t) = -\infty.

Solutions.

1: On the level set {(x,t)u(x,t)=c}\{(x, t) \mid u(x,t)=c\} we have ut=cuxu_t=-cu_x. If ux0u_x\neq 0 along the level set, then the Implicit Function Theorem says that near each point (x0,t0)(x_0, t_0) with $u(x_0, t$$_0)=c$, the level set is the graph of a function x=xc(t)x=x_c(t). The derivative of this function follows from

0=dcdt=du(xc(t),t)dt=uxxc(t)+ut    xc(t)=utux=cuxux=c.0=\frac{dc}{dt} = \frac{d u(x_c(t), t)}{dt} =\frac{\partial u}{\partial x}x_c'(t) + \frac{\partial u}{\partial t} \implies x_c'(t) = -\frac{u_t}{u_x} = -\frac{-cu_x}{u_x} = c.

Thus the level set is a straight line with slope xc(t)=cx_c'(t)=c. Since u(a,0)=cu(a, 0)=c we also know that xc(0)=ax_c(0)=a, so xc(t)=a+ctx_c(t)=a+ct.

2: Differentiate the equation ut+uux=0u_t+uu_x=0 with respect to xx to get

utx+uxux+uuxx=0    uxt+uxux+uuxx=0    vt+v2+uvx=0.u_{tx} +u_x u_x + u u_{xx}=0\implies u_{xt} +u_x u_x + u u_{xx}=0\implies v_t +v^2 + uv_x=0 .

3: We have

S(t)=dv(a+ct,t)dt=cvx(a+ct,t)+vt(a+ct,t)S'(t) = \frac{d v(a+ct, t)}{dt} = cv_x(a+ct, t) + v_t(a+ct, t)

Furthermore u(a+ct,t)=cu(a+ct, t) = c, so

S(t)=u(a+ct,t)vx(a+ct,t)+vt(a+ct,t)=v(a+ct,t)2=S(t)2.S'(t) = u(a+ct, t)v_x(a+ct, t)+v_t(a+ct, t) =-v(a+ct, t)^2 = -S(t)^2.

4: The initial function is u(x,0)=21+exu(x, 0)=\dfrac{2}{1+e^x}, which satisfies u(0,0)=1u(0,0)=1. The level set through (0,0)(0,0) is therefore a line with slope 11, and is thus given by x(t)=tx(t)=t: the level set through (0,0)(0,0) is {(t,t)t0}\{(t,t) \mid t\geq 0\}.

By the previous problem S(t)=ux(t,t)S(t)=u_x(t,t) satisfies S(t)=S(t)2S'(t)=-S(t)^2.

At t=0t=0 we have

S(0)=ux(0,0)=(x21+ex)x=0=12.S(0) = u_x(0,0) = \left(\frac{\partial}{\partial x}\dfrac{2}{1+e^x}\right)_{x=0} = -\frac 12.

Solving S(t)=S(t)2S'(t) = -S(t)^2, S(0)=12S(0)=-\frac 12 we find S(t)=1t2S(t) = \dfrac{1}{t-2}. Hence limt2S(t)=\lim_{t\nearrow 2} S(t) = -\infty.