First Order Partial Differential Equations

1. The Method of Characteristics

A partial differential equation of order one in its most general form is an equation of the form 3 Fx, u, 4u = 0, 1.1

3 where the unknown is the function u = ux = ux 1 , ..., x n of n real variables. Here, we will not consider problems of such generality but will focus instead on a smaller class of problems. For example, the equation (1.1) is said to be a quasilinear equation in two variables if it is of the form at, x, u / t ux, t + bt, x, u / x ut, x = ft, x, u, 1.2

i.e., the equation is linear in the derivatives / t u and / x u but is nonlinear in u. If ft, x, u = 0, the equation is said to be homogeneous. In order to make the notation more convenient later, we are choosing to call the independent variables, t and x. Suppose u = ux, t is a smooth solution of (1.2) and let S = t, x, u 5 R 3 : u = ux, t. Then S is said to be a solution surface for (1.2). The smoothness of the solution u means that S has a tangent plane at each point t, x, u 5 S. The normal vector 3 to the tangent n plane has the direction numbers / t u, / x u, ?1; i.e., ux, t ? u = 0 is the equation of S and / t u dt + / x u dx ? du = 0 is the equation of the tangent plane. Now consider a curve C = t = ts, x = xs, u = us, s 5 I solution curve for the system dt = at, x, u, ds dx = bt, x, u, ds du = ft, x, u, ds in 3-space defined as a

1.3

If 3 denotes a vector tangent to C at t, x, u then the direction numbers of 3 must be T T T n T a, b, f. But then (1.2) implies that 3 e 3, which is to say, 3 lies in the tangent plane to the surface S. But if 3 lies in the tangent plane, then C must lie in S. Evidently, solution curves T of (1.2) lie in the solution surface S associated with (1.2). Such curves are called characteristic curves for (1.2). Note that if C is a solution curve for (1.3) then at, x, u / t ux, t + bt, x, u / x ux, t, u = / t ux, t dt + / x ux, t dx = du = ft, x, u, ds ds ds so the pde reduces to an ode along C. In general, the equations for C must be solved as a

system. We will recall now some notions from differential geometry that will clarify the procedure for solving the system (1.3).

3 A vector valued function, V = Pt, x, u, Qt, x, u, Rt, x, u is called a vector field if P,Q,R are all smooth functions and if P 2 + Q 2 + R 2 is never zero. A space curve, 3 C = t = ts, x = xs, u = us, s 5 I is said to be an integral curve or trajectory for V 3 if V is tangent to C at every point; i.e., if dt = Pt, x, u, ds or, equivalently, dx = Qt, x, u, ds dt = dx = du R P Q du = Rt, x, u, ds 1.4a 1.4b

Integral Curves for Vector Fields

3 A function d = dt, x, u is said to be a first integral for the vector field V = P, Q, R if Pt, x, u / t d + Qt, x, u / x d + Rt, x, u / u d = 0. 1.5

3 The trajectories C for V will be found by representing C as the intersection of level surfaces of first integrals. The level surfaces S j = t, x, u : d j t, x, u = C j j = 1, 2

intersect transversally at each point if their normals, 3 1 and 3 2 are never parallel. This n n situation occurs if d 1 and d 2 are such that the expression 4d 1 4d 2 is different from zero at each point. In this case the functions d 1 and d 2 are said to be functionally independent and their level surfaces S 1 and S 2 intersect in a curve C. Since C then lies in both of the surfaces, S 1 and S 2 , the tangent to C is normal to both 3 1 and 3 2 , that is to both n n 4d 1 and 4d 2 . This is the same thing as saying both d 1 and d 2 satisfy (1.5). We will illustrate with examples. Example 1.1 3 1. Consider the radial vector field V = t, x, u. A first integral must satisfy t / t dt, x, u + x / x dt, x, u + u / u dt, x, u = 0. To obtain a solution, we consider the following system of odes dt = dx = du x u t Then dt = dx x t or dt = dx x t and dx = du x u

leads to x = C 1 , t

and

dx = du implies x u

x u = C2.

x 3 That is, d 1 t, x, u = x and d 2 t, x, u = u are a pair of first integrals for V = t, x, u. We t can show that for any smooth function F of two variables, d 3 t, x, u = Fd 1 t, x, u, d 2 t, x, u 3 is also a first integral for V and d 3 is then viewed as an implicit representation for the most general solution of the first integral pde. 3 Problem 1.1 Show that if d 1 t, x, u and d 2 t, x, u are a pair of first integrals for V then d 3 t, x, u = Fd 1 t, x, u, d 2 t, x, u, where F denotes an arbitrary smooth function of two 3 variables, is also a first integral for V. . x Problem 1.2 Show that d 1 t, x, u = x and d 2 t, x, u = u are functionally independent. t 3 2. Consider the vector field V = t, u, x. The trajectories are solution curves for the following system of odes dt = dx = du u x t From or dt = du x t and dx = du u x

dx = du it follows that x 2 = u 2 + C 1 . Then u x dt = du = du and t = C 2 u + u 2 + C 1 = C 2 u + x. x t 2 u + C1 t 2 2 Evidently, d 1 = x ? u and d 2 = u + x are a pair of first integrals for the vector field 3 V = t, u, x. Each of these functions satisfies t / t dt, x, u + u / x dt, x, u + x / u dt, x, u = 0, and the most general solution of this equation can be written implicitly as t F x 2 ? u 2 , u + x = 0, where F denotes an arbitrary smooth function of two variables.

Characteristics for Quasilinear PDEs of Order 1

We are aware now that C is a characteristic curve for the quasilinear pde (1.2) if C is a 3 trajectory for the vector field V = a, b, ?f. Then solutions for the pde can be obtained from first integrals for the vector field. However, we are not usually interested in finding the most general solution for the pde but are instead interested in finding certain particular solutions. For example, we shall be interested in finding a solution for (1.2) that satisfies the additional condition that ux, t = gx, t on C I : x = xb, t = tb.

where the curve C I and the function g are given. A condition of this form is called a Cauchy condition, and the problem of finding a solution for (1.2) that satisfies the Cauchy condition is called a Cauchy initial value problem.

Example 1.2 1. Consider the following Cauchy problem / t ux, t + 4 / x ux, t = 0, ux, 0 = 1 . 1 + x2

dx = 4. This pair of equations is Let C 0 denote a solution curve for, dt = 1, ds ds equivalent to the single equation, dx = 4, for which the solution is, xt = 4t + x 0 . The dt solution curves, C 0 , are a family of straight lines all having the same slope. Then along any solution curve, / t ux, t + 4 / x ux, t = / t ux, t + / x ux, t dx = du = 0, dt dt from which it follows that u is a constant on each such curve. A general solution for the partial differential equation is given by ux, t = fx ? 4t where f = fz denotes an arbitrary smooth function of one variable. Then ux, 0 = fx and this, combined with the Cauchy initial condition, leads to the solution 1 ux, t = 1 + x ? 4t 2 for the Cauchy problem. Note that the initial value u 0 = ux 0 , 0 of the solution at the point x 0 propagates along the line x ? 4t = x 0 ; i.e., ux, t = u 0 at all points x, t such that x ? 4t = x 0 . As a result, if the initial data is specified only on the interval, say 0 < x < 10, then the solution is determined only in the strip, F = x, t : 4t < x < 4t + 10, t > 0. The strip F is the domain of influence of the initial interval I = 0 < x < 10. The pde in this example is linear which leads to the result that the characteristic system of odes uncouples. That is, we can solve the equation x v t = 4 separately from the equation u v t = 0. 2. Now consider a Cauchy problem for the variable coefficient equation / t ux, t + x t / x ux, t = 0, ux, 0 = sinx. The coefficients in this equation are functions of the independent variables in the problem but do not depend on the unknown function u. Hence the equation is a linear partial differential equation as was the equation in the previous example. The solution curves for the characteristic ode, dx = xt are given by, dt 2 t 2 /2 ln x = t /2 + c 0 , or x = c 1 e . Evidently, the solution curves are the level curves of dx, t = xe ?t /2 and since the pde reduces to the ode u v s = 0 along level curves of d, the solution u of the partial differential equation is constant along these curves. The most general such solution has the form 2 ux, t = f xe ?t /2 for an arbitrary smooth function of one variable f. Using this in the initial condition leads to, ux, 0 = fx = sinx and ux, t = sin xe ?t
2 /2 2

Points worth noting about these two examples: the solution curves C 0 constructed for these two examples are not characteristic curves for the pdes in the examples, since they are plane curves lying in the x-t plane. In fact, they are the projections into the x-t plane of the characteristic curves for the pdes. Since the coefficients in the pdes in these linear examples do not depend on the solution u, the characteristic system 1.3 or 1.4 can be uncoupled and we can solve for x = xs, t = ts without solving for u. The solution curves are curves in the x-t plane. In order to distinguish these plane curves from the previously discussed characteristic curves in 3-space, we will refer to the plane solution curves as base characteristics and use the term space characteristics when referring to characteristic curves in 3-space. the domain of influence for an interval I = a, b is the strip 2 2 F = x, t : t > 0, a e t /2 < x < b e t /2 bounded by the base characteristics that originate at the endpoints of the initial interval. the space characteristic curves in these two examples are curves of the form x = xt, u = const ; i.e., they are plane curves lying in planes parallel to the x-t plane. This is due to the fact that in both examples, the partial differential equation is homogeneous and in such cases, the pde reduces to du/dt = 0 along base characteristics with the obvious result that solutions are constant along base characteristics and along space characteristics. constant coefficient equations have straight line base characteristics while linear equations with variable coefficients have base characteristics which are curves. However even when the base characteristics are curved, the family of curves is coherent; i.e., base characteristics which originate at distinct points can never cross. To see why characteristics for linear equations are coherent, let C 1 and C 2 denote distinct base characteristics of the equation at, x / t ux, t + bt, x / x ut, x, u = ft, x, u, which cross at some point x 0 , t 0 . Intersection at a point requires that the curves have noncoincident tangents at the point of intersection. But this is inconsistent with (1.2) which implies that dt dx = ax 0 , t 0 = bx 0 , t 0 dt dx

C1

C2

i.e., the tangents to C 1 and C 2 have equal slopes at the point of intersection. In the case of quasilinear equations, where the coefficients can depend on u, we will see that this coherence can fail. 3. Consider a Cauchy problem for a linear but inhomogeneous equation x / t ux, t ? 2xt / x ux, t = 2t u, The characteristics are the solutions of 5 ux, 0 = x 3 .

dt = x, ds

dx = ?2xt, ds

du = 2t u, ds

or

dt = ? dx , x 2xt

and

du = ? dx . 2tu 2xt

Since the equation is linear, the base characteristic curves can be obtained independent of the solution u. The solutions for the characteristic equations are given by x + t2 = C0 and x u = C1.

The general solution for the partial differential equation can be expressed as Fx + t 2 u = C1 = x x Fx and then ux, 0 = x = x 3 . It follows that Fx = x 4 hence ux, t = x + t 2 . x
4

When the partial differential equation is inhomogeneous, as in this example, the solution u is no longer constant along the characteristics, but in fact varies in a manner prescribed by the differential equation, u v s = 2tu, and the initial condition. Then the space characteristics are no longer plane curves but are truly space curves. Notice that the partial differential equation reduces to an ordinary differential equation along the characteristics. 4. Now consider the following quasilinear problem, / t ux, t + 4 / x ux, t = ux, t 2 , The characteristic system can be written as dt = dx = du 1 4 u2 or dt = dx and dt = du . 1 4 1 u2 ux, 0 = 1 . 1 + x2

The pde is linear in the leading terms (in fact, this special subclass of quasilinear problems is referred to as the class of semilinear problems) so that the characteristic system uncouples and the base characteristics can be found without knowing the solution u. Clearly the base characteristics are the family of straight lines x ? 4t = x 0 . The second characteristic equation u v t = ut 2 has the solution ut = C 1 ? t ?1 and it follows that the most general solution for the pde can be written as, ux, t = 1 . fx ? 4t ? t 1 = 1 , or fx = 1 + x 2 . Then the solution of fx 1 + x2

The initial condition implies ux, 0 = the Cauchy initial value problem is

ux, t =

1 x ? 4t 2 + 1 ? t

Note that in spite of the smoothness of the initial data, the solution develops a singularity at x = 4, t = 1. Plotting solution profiles at times approaching t = 1 shows that the solution behaves like a wave that propagates from left to right and sharpens to a spike at x = 4, as t approaches t = 1.

u(x,t) vs x for t=0, .2, .6, .9 We compute / x ux, t =

?2x ? 4t 1 ? t + x ? 4t 2

from which it is evident that the gradient, / x ux, t has an even stronger singularity at x = 4, t = 1. The developing gradient singularity can be seen in the following plot of gradient profiles for increasing t.

gradient of u(x,t) vs x for t=0, .2, .6, .9 Since this pde is semilinear (i.e., linear in its leading terms) and has smooth initial data, the spontaneous singular behavior in the solution must be due to the nonlinear term on the right side of the equation. Note also that semilinear equations permit the base characteristics (i.e. the solution curves for t v s = a, x v s = b ) to be found independent of the solution u = ux, t and then the PDE reduces to a nonlinear ODE along the base characteristics. 5. Consider the quasilinear Cauchy problem,

/ t ux, t + u / x ux, t = 0,

ux, 0 =

1 = fx. 1 + x2

Here the coefficients of / t ux, t and / x ux, t depend on u so the characteristic equations become dt = 1, ds dx = u ds and du = 0, ds or dx = u dt and du = 0. dt

In this case these equations do not uncouple. In order to solve the equations as a system, we note first the equation u v t = 0. This implies u = const along solutions of the other equation, dx/dt = u. This is to say, u = u 0 along the straight lines, x ? u 0 t = x 0 . It follows from the initial condition that the initial value u 0 originating at x 0 , 0 is given by ux 0 , 0 = u 0 = fx 0 . The solution u can then be expressed implicitly by writing 1 ux, t = 1 + x ? ut 2 and this equation can be solved for u in terms of x and t. The result, obtainable using Maple or Mathematica, is a complicated function of x, t. As in the previous example, the quasilinearity of the pde produces some singular behavior in the solution. For instance, suppose u = u 0 = fx 0 along u = u 1 = fx 1 along x ? u0 t = x0 x ? u1 t = x1 x0.

and

If u 1 u 0 , then the two straight lines intersect at x D , t D where x tD = ? u1 ? x0 , 1 ? u0 x x D = u 1u 0 ? u 0 x 1 . 1 ? u0

Note that the time of intersection, t D , is positive if x 1 > x 0 and u 1 < u 0 . At such a point of intersection, ux D , t D has the impossible requirement of being simultaneously equal to the distinct values u 1 u 0 . We conclude that the solution breaks down in some way at this point. Note further that ux, t = fx ? ut leads to / x ux, t = f v x ? ut1 ? t / x ux, t; i.e. f v x ? ut . / x ux, t = 1 ? t f v x ? ut Evidently the gradient / x ux, t becomes undefined at any point where, 1 ? t f v x ? ut = 0, another indication that the solution breaks down at some finite time. By plotting the solution profiles versus x for a sequence of increasing times, it can be seen that the initial wave form moves to the right, deforming as it propagates. It is apparent that at some point, the tangent line to the profile becomes vertical and the solution is no longer single valued beyond this point.

Solution Profiles at t=0, 1/2, 1, 3/2 The spontaneously occuring singularity is also evident from plotting the gradient versus x at increasing times. We can see that as time increases, the gradient begins to develop a negatively infinite singularity as the tangent line tends toward the vertical. Here, as in the previous example, the singularity is due to the nonlinearity of the equation (which in this case leads to colliding characteristics).

Gradient Profiles at t=0, 1/2, 1, 3/2 In each of the last two examples we have seen equations with smooth coefficients and initial data develop spontaneous singularities due to the nonlinearity of the equations. The solutions in these two examples break down at some finite time and no classical solution for the initial value problems exists past this point of breakdown. It will be necessary to weaken the notion of solution in order for these nonlinear problems to be solvable globally.

Problem 1.3 Consider the Cauchy initial value problem 1 . 1 + x2 Solve the characteristic equations and find the equation of the base characteristic that passes through the point 1, 2. Find the most general solution of the pde. / t ux, t + x / x ux, t = xt ux, t, ux, 0 =

Find the particular solution that satisfies the initial condition and describe the set of points x, t where the solution is determined if the initial values are specified only for ?1 < x < 1.

Problem 1.4 Consider the Cauchy initial value problem ux, 0 = x 2 ?4x / t ux, t + 9t / x ux, t = xt , u Solve the characteristic equations to find two first integrals of the associated vector field. Find the most general solution of the pde Find the particular solution that satisfies the initial condition. Does the solution have any singular behavior?

Problem 1.5 Consider the Cauchy initial value problem u / t ux, t + x / x ux, t = t ,

u1, t = 2t

Solve the characteristic equations to find two first integrals of the associated vector field. Find the most general solution of the pde. Find the particular solution that satisfies the initial condition. Does the solution have any singular behavior?

10