Partial Differential Equations,

An Introduction to Theory and Applications

by
Michael Shearer and Rachel Levy

Solutions Manual
by
Jeffrey Wong and Michael Shearer
Department of Mathematics, University of California, Los Angeles, CA 90095, USA
Department of Mathematics, North Carolina State University, Raleigh, NC 27695, USA
Chapter 1 Solutions
1. For u(x, t) = f (x − 3t), ut = −3f 0 (x − 3t), ux = f 0 (x − 3t). Thus, ut + 3ux = 0.

2. For u(x, t) = emt sin(nx), ut − kuxx = mu + kn2 u = 0 when m = −kn2 .

3. For u(x, t) = sin(mt) sin(nx), utt = −m2 u; uxx = −n2 u. Thus, utt = c2 uxx implies m2 = c2 n2 .

4. For u(x, t) = a(t)e2x +b(t)ex +c(t), we find ut = uxx when a(t) = e4t , b(t) = et , c(t) = constant.

5. k(u) = mum−1 .

6. u(x, t) = (1 + (x − 4t)2 )−1 .

7. 
 0 if x ≥ 4t,
u(x, t) =
 (t − x/4)e−(t−x/4) if x ≤ 4t

There is no solution if the PDE is changed to ut − 4ux = 0 because then the general solution
would have the form u(x, t) = f (x + 4t). The initial condition u(x, 0) = 0, x > 0 gives f (x) =
0, x > 0, which is incompatible with the boundary condition (along the t axis, x = 0), since
u(0, t) = f (4t) = 0, t > 0.

8. (a) Verification follows from the hints.

(b) (i) φ(0) = ψ(0); (ii) ψ 0 (0) = −cφ0 (0).

9. By the chain rule, if u = u(x, t), u = f (x − ut), then ux = f 0 (x − ut)(1 − ux t), so that
ux = f 0 /(1 + tf 0 ). Similarly, since ut = f 0 (x − ut)(−ut t − u), we have ut = −f 0 f /(1 + tf 0 ).
Thus, ut + uux = −f 0 f /(1 + tf 0 ) + f f 0 /(1 + tf 0 ) = 0.

10. Let u0 (y) = 1 − y 2 if −1 ≤ y ≤ 1, and u0 (y) = 0 otherwise.

(i) Use the implicit solution (4.12) to find an explicit formula for u = u(x, t),
with −1 < x < 1, 0 < t < 21 .
(ii) Verify that u(1, t) = 0, 0 < t < 21 .
−
(iii) Differentiate your formula to find ux (1− , t), and deduce that ux (1− , t) → −∞ as t → 12 .
−
Note: ux (x, t) is discontinuous at x = ±1; the notation ux (1 , t) means the one-sided limit:
−
ux (1− , t) = lim ux (x, t). Similarly, t → 12 means t → 21 , with t < 12 .
x<1,x→1

1
Chapter 2 Solutions
1. (a) Hyperbolic.
(b) Parabolic if α = 1/4; Hyperbolic if α < 1/4; Elliptic if α > 1/4.
(c) Parabolic.
(d) Parabolic on the curve y = x3 /4 − 1 and on the y axis x = 0. These curves divide the x, y
plane into 4 regions. If x > 0 and y < x3 /4 − 1, or x < 0 and y > x3 /4 − 1, then the equation
is hyperbolic. In the other two regions, the equation is elliptic.

2. Hint: write everything out carefully using subscripts for components of vectors and matrices.
Since you are given the result, you can work both forwards from what is given and backwards
from the end result to verify that they give the same expressions.

3. This exercise is a long-winded calculation using Taylor series, combining (2.6)-(2.9) and the
definition of G in terms of a, b, c, f. Best to use Maple or Mathematica, but the calculation of
u2 , and then u3 , is achievable on paper.

4. Try differentiating a few times to see a pattern in the derivatives. Use the fact that the
exponential decays to zero faster than any power of x.

5. (a) σ(ξ) = ±iξ 2 . The beam equation is dispersive and not dissipative because λ is imaginary,
and is nonlinear in ξ. For the wave equation, λ is also imaginary, but it is linear in ξ, so that
waves with different spatial frequencies ξ all travel with the same speed. The beam equation
traveling waves travel withR speed s(ξ) = ξ that depends on ξ. The beam equation does not
dissipate energy E(t) = 12 {(ut )2 + (uxx )2 }dx : E 0 (t) = 0 (integrating by parts twice, with
boundary conditions that render boundary terms equal to zero).
(b) σ(ξ) = −icξ/(1 − βξ 2 ). The speed of traveling waves is s(ξ) = c/(1 − βξ 2 ). This dependence
on ξ differs from the quadratic dependence of the KdV √ equation traveling waves in that s(ξ)
is bounded if β < 0, and has a singularity at ξ = 1/ β if β > 0.

6. (a) v(u) should be a decreasing function, so that more dense traffic moves more slowly.
(b) vmax = v(0), v(umax ) = 0.
(c) Q0 (u) = uv 0 + v = 0 has at least one solution by the Intermediate Value Theorem, since
Q0 (0) = vmax > 0 whereas Q0 (umax ) = umax v 0 (umax ) < 0. At least one of these solutions is a
maximum since Q(u) ≥ 0, and Q(0) = Q(umax ) = 0.
(d) Yes, you can choose a quartic function Q(u) = u(umax − u)q(u), in which q(u) > 0 is
quadratic, v(u) = (umax − u)q(u) is decreasing and Q0 (u) = 0 has three solutions between
u = 0 and u = 1.

2
Chapter 3 Solutions
1. u(x, y) = 52 (1 − e5(x−y) ).

2. u(x, t) = sin(x − arctan t).

3. u(x, t) = 61 e(2x+t) + (x − t)e−3t − 16 e(2x−5t) .

7. u(x, y, t) = x − 12 St2 − yt2 + y + St / 1 − St2 + 2St . This solution does not depend on

the domain 0 < x, 0 < y < 1 and does not satisfy the inlet boundary condition at x = 0.

−1
8. (a) u(x, t) = 1 − t + (x − 12 t2 )2 .
(b) As t → 1− , u(x, t) → (x − 12 )−2 , which is singular at x = 21 . The solution is finite for
0 ≤ t < 1.

9. Q(0) = 0. There is no flux at zero density because there are no cars.

Q(α) = 0. At maximum density, the traffic is stationary, so there is no flux.

10. The problem should refer to Example 5, Chapter 2. Let v(ρ) = aρ2 + bρ + vm , where vm is the
maximum speed (at zero density). Then 2aρ + b < 0, and there are two additional parameters,
the maximum density ρm and either a or b, related through v(ρm ) = 0 : b = −aρm − vm /ρm .
The flux is easily made non-concave, i.e., convex over an interval.
√ For example, choosing
b = 0, a = −vm /ρm gives Q00 (ρ) > 0 in the interval 0 ≤ ρ < ρm / 3.

11. Let F (u, x, t) = u − u0 (x − ut). Then F is smooth, F (u0 (x0 ), x0 , 0) = 0, and Fu (u0 (x0 ), x0 , 0) =
1. Thus, the implicit function theorem applies in a neighborhood of (u, x, t) = (u0 (x0 ), x0 , 0),
to give u = u(x, t) satisfying F (u(x, t), x, t) = 0.
1  √
12. u(x, t) = 2
2xt − 4xt + 1 .
2t
14. Consider the characteristic x = u0 (x0 )t + x0 , on which u(x, t) = u0 (x0 ) is constant. For fixed
x, t with t > 0, let g(y) = u0 (y)t + y − x. Then g 0 (y) = u00 (y) + 1 > 0. Moreover, g(y) → ±∞
as y → ±∞. Consequently, the equation g(y) = 0 has a unique smooth solution y = y(x, t).
Then u(x, t) = u0 (y(x, t)).

15. u = u0 (x − f 0 (u)t).

3
Chapter 4 Solutions
R3
1. 1 ψ(x) dx = 0.

2. (a) et = utt ut + uxt ux = (ut ux )x = px and similarly for ex .

(b) ett = pxt = exx and ptt = ext = pxx .

3. (a) ∂tt (u(x − y, t)) = utt (x − y, t) = uxx (x − y, t) = ∂xx (u(x − y, t)).

(b) (ux )tt = (utt )x = (uxx )x .
(c) ∂tt (u(ax, at)) = a2 utt (ax, at). and ∂xx (u(ax, at)) = a2 uxx (ax, at).

4. (a) Hint: Use the general solution (4.9).

(b) u(x + 2h, t + k) + u(x − 2h, t − k) = u(x − 2k, t − h) + u(x + 2k, t + h).

5. The wave is travels left for t < 1/2, reflects at t = 1 and travels right for t > 3/2.

6. (a) If x > t, use D’Alembert’s solution. If x < t,R take even extensions
R t−x of φ and ψ.
1 1 x+t
and obtain u(x, t) = 2 (φ(x + t) + φ(t − x)) + 2 t−x ψ(s) ds + 0 ψ(s) ds.
(b) u = 0 in the first quadrant if x + t < 1 or x > t + 2.
(c) This follows from (a).
(d) u is continuous if φ is continuous and u ∈ C 1 if ψ is continuous and φ0 (0) = 0.

Rξ
7. F (−ξ) = G(ξ) − 0 h(s) ds for ξ > 0 with G from D’Alembert’s solution, so for x < t we have
R t−x
u = G(t − x) + G(x + t) − 0 h(s) ds. Easy to check ux (0, t) = h(t).
The solution is continuous if ψ, φ, h are. If h(0) = φ0 (0), then ux , ut are continuous.
R∞
8. E 0 (t) = − −∞ µu2t dx.
R∞
9. (a) E(t) = 0 2c u2t + 12 u2x dx is conserved if ψ(0) = 0 or ux (0, t) = 0.
(b) E 0 (t) = ux ut |x=0 = ψ(0)h(0).
1
Rt R x+c(t−s)
10. Write u(x, t) = 2c 0 ũ(x, t, s) ds where ũ(x, t, s) = x−c(t−s) f (y, s) dy. Compute
Rt Rt
utt = ∂t ũ(x, t, t) + 0 ũtt (x, t, s) ds and ∂xx u = 0 ∂xx g(x, t, s) ds and use that ũtt + c2 ũxx = 0.

11. (a) Note that the integrals in (4.23) depend only on |x|.
(b) If ct −  ≤ |x| ≤ ct +  then S(x, ct) ∩ {|x| < } = ∅ so u(x, t) = 0.

4
Chapter 5 Solutions
1. (a) Follow the hint.
(b) Bound the integrand by something easy to integrate.

2. Note that for x ∈ Rn , ∇φ = xr φr and ∇ · (x/r) = (n − 1)/r.

3. Observe that Φ(x, t) is even in x.

   
1 1+x 1 1−x
4. u(x, t) = Erf √ − Erf √ .
4 4kt 4 4kt
2 kt+px)
5. u(x, t) = e(p .
  
1 kt−x 2kt − x
6. u(x, t) = e 1 − Erf √ .
2 4kt
∂
( 21 u2 dx) = u∂x (kux ) dx = − ku2x dx if u ∈ L2 .
R R R
7. ∂t

8. a) Commute x and t derivatives on u.

b) By uniqueness, v(x, t) = 0 for all t > 0, so u(x, t) = x2 f (t) + xg(t) + h(t). Substituting into
the initial condition and the heat equation leads to u(x, t) = x2 + 2kt.

9. (a) We find that vt = kvxx with v(x, 0) = g(x).

(b) du acts as a source
Rt term.
(c) Set v = u exp( 0 d(s) ds).

10. Set v(x, t) = u(x + ct, t).

1 2 c2 2 dx.
R R
11. Define the energy E(t) = 2 Ω ut dx + 2 Ω |∇u| We get
Z
E 0 (t) = c2 ut (∇u · n) dS
∂Ω

where n is the outward unit normal; this is zero if u = const. or ∇u · n = 0 on ∂Ω.

5
Chapter 6 Solutions
1. u(x, t) = e−4t sin x − 3e−100t sin 5x.
2
2. u(x, t) = (2 cos 3πt + 3π sin 3πt) sin πx + 7 cos 9πt sin 3πx.

3. λ = 0 is an eigenvalue if and only if a0 + La0 aL + aL = 0 with eigenfunction u(x) = a0 x + 1.

4. a) Set λ = −k 2 . Then k satisfies tanh kL = k|a0 + aL |/(k 2 + a0 aL ) := f (k). This is maximized

√
at k = a0 aL with f (k ∗ ) ≥ 1, so it must intersect tanh kL at some value k > 0.
b) If f 0 (0) < L then f crosses tanh kL twice to produce two eigenvalues (which is also sufficient).
This is equivalent to a0 + aL + La0 aL > 0.

5. a) Formalize the graphical arguments in the section.

b) βn = tan−1 (f (βn )) + (n − 1)π/L but f (βn ) → 0 as n → ∞ so θn := βn − (n − 1)π/L → 0.
c) Note that f (Cn + θn ) = (a0 + aL )/(Cn) + O(1/n2 ) and tan(1/n) = 1/n + O(1/n2 ), which
gives that θn = (a0 + aL )/πn + O(1/n2 ).

6. There are two eigenvalues if a0 + aL + La0 aL > 0, one if it equals zero and none otherwise.
Neumann boundary conditions correspond to a0 = aL = 0 and Dirichlet to a0 = aL = ±∞.

7. Set αi = ri /pi and λ = k 2 . Continuity of u and u0 at x = m gives two equations, along

p

with two more for the boundary conditions. Solving separately for u = u1 (x), x < m, u =
u2 (x), x > m we find u1 (x) = A1 sin(kα1 x), u2 (x) = A2 sin(kα2 (L − x)). Let A = A2 /A1 . Then
u1 (m) = u2 (m) gives A = sin(kα1 m)/ sin(kα2 (L − m)). Similarly, continuity of u0 at x = m
gives kα1 cos(kα1 m) = −Akα2 cos(kα2 (L − m)). Thus, the equation that determines values of
k, and hence λ = k 2 , is
tan(kα1 m) α1
=− .
tan(kα2 (L − m)) α2

8. a) Look for values of µ > 0 such that cos µL = −1/ cosh µL, by graphing the functions of µL
on each side of this equation. Note that cos µL has zeros at µL = nπ + π/2 and has minima
at µL = π + 2nπ; solutions exist between minima and adjacent zeros.
b) Pretty clearly from the graph, Lµn ∼ (n − 1/2)π as n → ∞. More precisely, define θn by
Lµn = (n − 1/2)π + θn . Then, expanding cos(Lµn ) and using −1/ cosh(µn L) ∼ −2e−µn L we
are led to θn ∼ 2(−1)n+1 eπ/2 e−πn .

9. (a) Use (6.5) to obtain bn = 4/πn for n odd and zero otherwise. √
(b) Evaluating the series at x = π/4 gives 1 + 1/3 − 1/5 − 1/7 + 1/9 + · · · = π4 2.

6
Chapter 7 Solutions
1. Integrate by parts to obtain L∗ v = (av)00 − (bv)0 + cv.

2. The proof is as in Theorem 7.1, but using the weighted norm ||u||2r = r(x)|u(x)|2 dx, ie.
R

λ||u||2r = (λru, u) = (Lu, u) = · · · = λ||u||2r with (u, v) the unweighted L2 inner product.

3. Hint: If f is continuous except for a jump at x0 from a to b then f − (b − a)H(x − x0 ) is

continuous.

4. Let {vn } be the orthonormal basis for L2 ((−π, π)). Approximate f in L2 by h ∈ C 1 ([−π, π])
and use Bessel’s inequality to bound ||Sn (f ) − Sn (h)||L2 ≤ ||f − h||L2 . We have Sn (h) → h
uniformly (hence in L2 ) by Theorem 7.5.
By the triangle inequality, ||f − Sn (f )||L2 ≤ 2||f − h||L2 + ||h − Sn (h)||L2 .

5. Consider the difference quotient Dh f = h−1 (f (x + h) − f (x)). We have the estimate

||Dh (f ∗ g) − f 0 ∗ g||L1 ≤ ||Dh f − f 0 ||L1 ||g||L1 .

Now Dh f → f 0 in L1 by dominated convergence since Dh f → f 0 pointwise and

Z Z h
−1
h
||D f ||L1 = h f 0 (x + s) ds dx ≤ ||f 0 ||L1
0

using Fubini’s theorem. Thus Dh (f ∗ g) → f 0 ∗ g in L1 so (f ∗ g)0 = f 0 ∗ g.

R π ixn
6. (a) By the inversion formula, f (n) = 2π 1
−π e fˆ(x) dx, which are just the Fourier series
coefficients for f.
(b) Take the inverse transform of fˆ from (a) and use that the inverse transform of the function
equal to e−inω for |x| < π and zero otherwise is 1
π sin(π(x − n))/(x − n).

7. (a) Compute a2n = − π(4n42 −1) for n ≥ 1 and a0 = 4/π with a2n+1 = 0. Evaluate at x = π and
x = π/2 to obtain ∞
P 2
P∞ n 2
n=1 1/(4n − 1) = 1/2 and n=1 (−1) /(4n − 1) = (2 − π)/4.
2 n
(b) Take f (x) = x2 , which has Fourier series x2 = π3 + 4 n≥1 (−1)
P
n2
cos nx.

7
Chapter 8 Solutions
1. Let M = max∂Ω u(x), let  > 0 and v = u(x) + |x|2 . If v has a maximum at x0 ∈ Ω then
0 ≥ ∆v = 2n, a contradiction. Hence |v | ≤ M + O(). This implies that |u(x)| ≤ M + O()
so |u(x)| ≤ M.

2. If u is constant in U then clearly the weak maximum principle holds. Otherwise, by the strong
maximum principle, u(x) < maxy∈∂U u(y), which is a stronger condition.

3. (a)
R If u2 and vR are solutions then w = uR − v solves ∆w = 0 with ∂w/∂ν + αw = 0 on ∂U. Then
|∇w| = − w∆w + w ∂w 2
R
∂ν dS = − αw dS which implies that w = 0.
(b) If α = 0 then (a) implies ∇w = 0, i.e. u − v = const.
(c) Consider u00 = 0 in (0, 1) with boundary conditions u0 − 2u = 0 at x = 1 and −u0 − 2u = 0
at x = 0. Then u = b(1 − 2x) is a solution for any b ∈ R.

4. Compute the geometric series to obtain (1 − ar z)−1 + (1 − ar z)−1 − 1 where z = ei(θ−φ) . This
1−(r/a)2 zz
is then equal to 1−2r/a(z+z)+(r/a)2 zz
which gives Poisson’s formula.
R
5. Evaluate the integral Vr = B(0,r) 1 dx in polar coordinates.

6. If ∆u 6= 0 with u ∈ C 2 (U ) then ∆u > 0 (or ∆u < 0) in some ball B(x, r) ⊂ U . This

contradictions the calculation in (8.2).

7. See Theorem 8.4 where a short proof is given.

8
Chapter 9 Solutions
1. (a) Integrate the ODE.
(b) If u, v are solutions then w = u − v satisfies w00 = 0 and w0 (0) = w0 (1) = 0, so w = C.
Rz RxRz
(c) u0 (z) = 0 f (y) dy so (up to a constant) u(x) = 0 0 f (y) dy dz = N (x, y)f (y) dy where
R

N (x, y) = x − y if x ≥ y and N (x, y) = 0 otherwise.

2. From (9.6) compute φy (x) = G(x, y) + 21 |x − y| = 12 (x + y) − xy, which is linear in x.

3. If v = e−cx u then v 0 = ecx δ = δ so v = H(x) + C for some constant C. The solutions satisfying
|u| → 0 as |x| → ∞ are u = e−cx H(x) for c > 0 and u = −e−cx H(−x) for c < 0.
R x+1
4. (g ∗ f )(x) = x g(y) dy which is piecewise quadratic for −2 < x < 1 and zero otherwise.

5. Linearity is immediate. If φn → φ in D then (fi , φn ) → (fi , φ) in each case since φn → φ

uniformly and the functions φn have shared compact support.

6. If φ ∈ Cc∞ then η  (y)φ(x − y) dy = η(y)φ(x − y) dy → φ(x) as  → 0 since η dy = 1,

R R R

φ(x − y) → φ(x) uniformly and the integral for each  is over a fixed compact set.

7. (a) (∂u/∂x, φ) = −(u, ∂φ/∂x) for φ ∈ D.

R ∞R ∂φ
(b) We have (∂u/∂x, φ) = −(H(y), ∂φ/∂x) = − 0 ∂x dx dy = 0 for any φ ∈ D.
1 −k|x|
8. (a) Let u(x) = 2k e . Then −u00 + k 2 u = δ in D0 . For φ ∈ D and  > 0,
00 + k 2 uφ dx = (uφ0 − u0 φ)|− = 1 −k 0
− φ0 (−)) + e−k φ()
R
|x|≥ −uφ 2k e (φ ()

which tends to φ(0) as  → 0.

1 −k|y| −kx −k|x−y|
(b) Solve (9.14) to obtain φy (x) = 2k e e and G(x, y) = 1
2k (e − e−k|y|−kx ).

9. The fundamental solution is u = 1 −k|x| . Using the change of

4π|x| e variable u = v/r we get
−v 00 /r
+ k 2 v/r
= f which suggests v = Ck e−kr . The integral in the source condition can then
be evaluated explicitly to obtain Ck = 1/4π.

10. u ∈ L2 (B) if and only if α < 1/2, which is the condition that ensures |x2 |−α ∈ L2 (B).

11. (a) ((gf )0 , φ) = −(f, gφ0 ) = −(f, (gφ)0 − g 0 φ) = (gf 0 , φ) + (g 0 f, φ).

(b) Apply (a) with f = δ and note that gδ = g(0)δ.
R
12. Compute u (x) = B(x,) η (x − y)u(y) dy = 0 rn−1 η (r) ∂B(x,r) u(y) dS dr and use the mean
R R
R
value property and the fact that B(x,) η (y) dy = 1.

9
Chapter 10 Solutions
1. (a) Use the triangle inequality for v = (u − v) + v and u = (v − u) + u.
(b) Take e.g. u, v of opposite signs.

2. f (a, b) = ap /p + bq /q − ab is smooth and has ∇f = (ap−1 − a, bp−1 − b) which has critical points
when a and b are 0 or 1. The Hessian is negative definite if p, q > 1 so the global minimum is
at f (0, 0) = 0.
−1 to (uv)w dx and then Hölder
R
3. (a) Apply Hölder using exponents r and s = (1/p + 1/q)
using p/s and q/s to (uv)s dx.
R

(b) Write the integral as uqλ uq(1−λ) dx and use Hölder with exponents p/(qλ) and r/(q(1−λ)).
R

R1
4. Show that U |∂u/∂xj |p = C 0 r−p(β+1)+n−1 dr so u ∈ W 1,p if β < n/p − 1.
R

5. Observe that u(α) = |x|1+α (sin |x|/|x|)α ∼ |x|1+α and ∂xj u(α) ∼ (xj /|x|)|x|α near x = 0. It
follows that u(α) ∈ H 1 (B) if and only if α > −1.

10
Chapter 11 Solutions
1. If u ∈ H01 (U ) then |u|2 ≤ C |∇u|2 = 0 by Poincare’s inequality, a contradiction.
R R

2. If Av = λv then λ|v|2 = v T Av ≥ θ|v|2 so λ ≥ θ.

R1 R1
3. We have 0 u02 dx ≥ C 0 u2 dx for u ∈ H01 ((0, 1)) where C = π 2 is the smallest eigenvalue of
−d2 /dx2 . Thus Theorem 11.8 applies with γ = −c − π 2 .
 
x −y − 1/2
4. is positive definite if and only if x > 12 (1 + y/2)2 , which defines the
−y − 1/2 2
region on which L is uniformly elliptic.
 
x 1/2
5. is positive definite when x > 1/4 with smallest eigenvalue x − 1/4 so c = 1/4.
1/2 1

6. From the triangle inequality, (u, u) + 2(u, v) + (v, v) = ||u + v||2 ≤ ||u||2 + ||v||2 + 2||u|| ||v||.

11
Chapter 12 Solutions
1. (a) u0 = 12 (u − u− )(u − u+ ).
(b) Solutions travel from the right to the left so we need u− > u+ .
(c) The solution satisfies (u − u+ )/(u − u− ) = Ceβξ where β = (u− − u+ )/2. After some
rearrangement, we get u = u− − β(1 + tanh( 21 βξ)).
(d) The wave speed is s = 1/2.

2. Equation (12.5): ut + uux + γuxxx = 0.

∂ ∂ 1 3
u2 dx = −2 + u2x ) = 0 and
R R
(a) Under the appropriate decay conditions, ∂t ∂x ( 3 u
R R R ∂ 2
ut = −uux − γuxxx = − ∂x (u ) = 0.
(b) The identity (u2 uxx )x = u2 uxxx + 2uux uxx gives
Z ∞ Z ∞ Z
∂ 1 2 3
( u − ηu dx) = −uux uxx − 3ηγu uxxx dx = (1/2 − 3η)u2 uxxx dx
2
∂t −∞ 2 x −∞

so η = 1/6γ produces the desired invariant.

√
3. Take scales t = αumax t̂, x = dT x̂ and u = umax û.
√
5. Solve v 0 = kv(v − 1) to obtain v(x − st) with s = (2a − 1)/ 2 as in Example 1. The traveling
wave is then u(x, t) = (1 − Cek(x−st) )−1 .

6. (a) Using (b), the ODE is βv 00 = −αv 0 + 21 (v − v+ )(v − v− ).

(b) The boundary conditions are consistent if and only if 12 v+2 − sv = 1 v 2 − sv = C which
+ 2 − −
1
gives s = 2 (v+ + v− ). The right equilibrium point is a saddle point, while the other is a stable
node or spiral so we need v− > v+ . For sufficiency, show that one branch of the unstable
manifold of (v− , 0) must cross the v-axis.
(c) The solution is monotonic if and only if (v + , 0) is a nodal sink (rather than a spiral), which
holds if and only if α2 > 2(v − − v + )β. Write the ODE as a system and show that eigenvalues
of the equilibria at v = v± , v 0 = 0 are real and distinct if and only if the condition on α, β
holds.

7. After integrating once, the ODE for u(ξ) is u0 /(b − su + f (u)) = 1. Integrating again gives the
desired equation.

12
Chapter 13 Solutions
1. The PDE is xux + tut = 0 with characteristics (x(s), t(s)) = (As2 , Bs2 ), i.e. x/t = const. u is
constant on characteristics, so u is a function of x/t.

2. Hint: Take locally linear approximations of u near (x0 , t0 ) ∈ C on either side of the curve
x = γ(t) and evaluate at a point on the curve. Show that they agree only when γ 0 (t) = f 0 (u)
by differentiating u(γ(t)+, t) = u(γ(t)−, t) and using the pde ut + f 0 (u)ux = 0.

3. We consider f (u) only for 0 ≤ u ≤ 1, since u is a volume fraction. Set A = k0 u2 + kw (1 − u)2 .

A direct computation (using MAPLE for example) shows that A2 f 0 (u) = 2k0 kw u(1 − u) and
f 00 (u)A3 = 2kw k0 ((kw + k0 )(2u3 − 3u2 ) + kw ). Thus, f (u) is positive for 0 < u ≤ 1, and
f ”(0) > 0, f ”(1) < 0. Thus, there is at least one zero in the interval 0 < u < 1. (You can also
use Rolle’s Theorem.) Since the cubic in f ”(u)A3 is readily seen to be a decreasing function
in this interval, the zero of f ” is unique.

4. For v fixed, f (u) = 21 (u−v)f 0 (u)+ 12 (u−v)f 0 (v)−f (v) which, as an ODE for f (u), has solution
f (u) = A(u − v)2 + 2f 0 (v)u − f (v) − vf 0 (v). In particular, f is quadratic. This characterizes
when a scalar conservation law has shock speed equal to the average of the characteristic
speeds.
√
5. u(x, t) = x/t for 0 < x/t < 2t−1/2 and = 0 otherwise. The shock height decays as t−1/2 so
√ u−1/2
u(x, t) → 0 uniformly with max |u| ≤ 2t .

6. The PDE ut + uux = 0 provides the equations c = 1 and ab = 2, using the assumption that
a, b, c, d are all positive. The Rankine-Hugoniot
√ condition holds (for the shock at x(t) = −dt2
with speed s(t) = −2dt) if a(1 + c − bd) = 4d. Substituting b = 2/a and isolating the square
root, we find an equation for d/a, leading to d = 3a/8. Thus, b = 2/a, c = 1, d = 3a/8, a > 0
is the solution. Moreover, the entropy condition follows from d > 0, on the t > 0 branch of
the shock.

7. (a) The Rankine-Hugoniot condition is 3s = u2+ + u+ u− + u2− .

(b) The entropy condition is 3 > 1 + u+ + u2+ > 3u2+ , which holds when −1/2 < u+ < 1.
√
(c) The rarefaction u(x, t) = û(x/t) satisfies û0 (ξ) = 1/f 00 (û(ξ)) which gives û(ξ) = ξ. The
solution is a rarefaction when u+ > 1 (and a shock-rarefaction pair if u+ < −1/2).

8. Along a characteristic, v = ux satisfies v 0 = −f 00 (u)v 2 which has solution v = 1/(C − f 00 (u)t).

Hence solutions remain smooth for all time if f 00 (u) < 0 and blow up if f 00 (u) > 0. The time
of blow up is t∗ = inf x∈R ux (x, 0)/f 00 (u(x, 0)) (it depends on the slope of the initial data).

9. (a) First note that x2 = −ρ̄t2 /(ρm vm ) and t2 = t1 (1 − ρ̄/ρm )−1 . For t > t2 the rarefaction
and the Rankine-Hugoniot condition provide an ODE for γ(t) which can be solved to obtain
1/2
γ(t) = 2vm (1 − ρ̄/ρm )t − vm t2 (2 − ρ̄/ρm )t1/2 .
t1
(b) Using (a), the time t3 = 4 (2 − ρ̄/ρm )2 (1 − ρ̄/ρm )−2 .

13
10. The density to the right of the slowed car will become zero, and cars will reach a density ρ1
behind it. Since the velocity-density relation is v = vm (1 − ρ/ρm ), we find ρ1 = ρm (1 − 2vv0m .)
There will then be a trailing shock connecting the original density ρ̄ = ρm (1 − vvm0 .) to ρ1 with
speed given by the Rankine-Hugoniot condition:

ρ̄v(ρ̄) − ρ1 v(ρ1 ) (ρ̄ + ρ1 )

s̄ = = vm (1 − ).
ρ̄ − ρ1 ρm

11. (a) Taking u(x, t) → u(−x, t) reverses (u2 )x and the dispersive term. To undo the (u2 )x sign
change we can take u → −u in the KdV-Burgers equation but not the modified equation.
(b) The system is u0 = w and w0 = γw − (u − u− )(u − u+ )(u − u0 ) with u0 = −(u− + u+ ).
If u+ < u0 < u− then u− and u+ are unstable spirals and u0 is a saddle point. Thus any
heteroclinic orbit must be from u± to u0 . The condition u+ < u0 < u− guarantees that the
entropy condition holds for such shocks.
(c) The corresponding shocks here are Lax shocks, but are undercompressive in the example.
Note that here the outer equilibria are unstable spirals rather than saddle nodes.
√
12. (a) If uR < −1+ 2γ/3 = uM then we have an undercompressive shock from uL to uM followed
by a rarefaction û from uM to uR given by û(ξ) = (u2M + 31 (ξ − s))1/2 for s < ξ < 3u2R .
√ √
(b) Assume uR > − 2/3γ and γ is such that 2 2/3γ < 1 = uL . Then uR > −1/2 so the
shock uL → uR satisfies
√ the entropy√condition. The traveling wave exists by Theorem 13.3.
When uM = −1 + 2γ/3 < uR < − 2γ/3, there is a shock from uL to uM (by Theorem 13.3)
and then a second from uM to uR which is a Lax shock since uM < uR < 0.

13. By the remarks after (13.39), Γ(v) for v0 < 0 is the least value of w0 such that q(t) has a
0
positive zero. Such a zero exists
√ √ only if q (0) < 0 and min q ≤ 0, which
when v0 < 0 if and
hold precisely when w0 < − 2v0 . Hence Γ(v) = − 2v.

14
Chapter 14 Solutions
1. The system is genuinely non-linear since ∇λ± · r± = ± 32 (gh)1/2 .

2. If T is convex and T 0 (1) > 0 then T 0 (ξ) ≥ ξ−1

1
T (ξ) > T (ξ)/ξ and the system is hyperbolic
by (14.18). If T is concave then T (ξ)/ξ > T (ξ) but there is some ξ ∗ such that T 0 (ξ) ≤ 0 for
0

ξ ≥ ξ ∗ and hyperbolicity is lost.

3/2
3. The rarefaction curves are û(ξ) = (± √38 ξ + u0 )2/3 and v̂(ξ) = −ξ + v0 for ξ < 0.
√
4. (a) DF has eigenvalues λ = ± v 2 − αu2 so the system is strictly hyperbolic iff α ≥ 0.
(b) The conditions are s(u− −u+ ) = u2− −u2+ +α(v−
2 −v 2 ) and −2(u v −u v ) = s(v −v ).
+ − − + + − +

(c) From (b), we need v+ = 0 or 1 + 2u+ − 3u2+ + v+

2 = 0.

(d) The equations are u0 = (u − 1)(u + 2) − v 2 and v 0 = v(1 − 2u). An orbit with v = 0
connects (1, 0) to (−2, 0) for the wave speed s = −1. However, λ1 (U ± ) < s < λ2 (U ± ), which
violates the entropy condition.
√
5. Setting r = u2 + v 2 , the system becomes ut + (f (r)u)x = 0 and vt + (f (r)v)x = 0.
(a) λ1 = f (r) and λ2 = f (r) + rf 0 (r) = (rf (r))0 and eigenvectors r1 = (v, −u) and r2 = (u, v).
(b) Since ∇λ1 · r1 = r−1 f 0 (r)(u, v) · (v, −u) = 0 the first field is linearly degenerate. If
(rf (r))00 6= 0 then ∇λ2 · r2 = 1r (rf (r)00 (u, v) · (u, v) = r(rf (r))00 6= 0 for r > 0.
(c) Rotation invariance implies that the eigenvalues depend only on r and the eigenvectors
must be (u, v) and (v, −u), so one field is automatically linearly degenerate.
(d) The Rankine-Hugoniot conditions for the shock (1, 0) → (u, v) give u(u2 +v 2 )−1 = s(u−1)
and v(u2 +v 2 ) = sv which has a solution when u2 +v 2 = s = 1 (a contact discontinuity) or when
v = 0 and u2 + u + 1 = s (a shock). For shocks, the entropy condition requires −1 < u < 0.
Rarefactions exist from (1, 0) to (uR , 0) for any uR > 1.

6. (a) For λT± the eigenvectors (using the calculations in (14.17)) are r± T = (−q, p, ∓qλ, ±λp).

Since λ± is a function only of ξ it is parallel to (p, q, 0, 0) and thus ∇λT± · r±

T T = 0.

T 00 (ξ) 2
(b) For λT± we have r±
L = (p, q, ±λp, ±λq) and ∇λL · r L =
± ± ξ (p + q 2 ) = ξT 00 (ξ) so the field
is genuinely nonlinear provided T 00 6= 0.

7. (a) If ψ + (U ) is a Riemann invariant of a 2 × 2 system for λ+ then ∇ψ + is a left eigenvector

for λ− I − DF. It follows that ψt+ + λ− ψx+ = ∇ψ + · ut + λ− ∇ψ · ux = ∇ψ · (−DF + λ+ I)ux = 0.
Rup
(b) Riemann invariants for the p-system are ψ ± = 0 σ 0 (u) ds ± v and it easily checked that
they satisfy ψt± ∓ σ 0 (u)ψx± = 0.
p

8. The normalization for rarefaction waves is chosen so that ∇λk · rk = 1, which reduces (14.39)
to s̃0 (0) = 1/2. This implies that s̃(ξ) < s(0) = λk (U− ) for ξ < 0 small. Since s̃0 (0) < 1, (14.40)
implies that λk (U + ) < s(ξ) for ξ < 0 small.

15
Chapter 15 Solutions
1. For an incompressible flow, ∇(u ⊗ u) = (∇ · u)u + (u · ∇)u = (u · ∇)u. If we set E = 21 ρ|u|2
then ∂E/∂t + ∇ · ((E + p)u) = 0, which is the same as the energy in the compressible case (if
an external force G = ∇V arises from a potential V then E = 12 ρ|u|2 + V ).

2. The equations are, with G = (G1 , G2 , G3 ) and subscripts denoting derivatives,

1
ut + uux + vuy + wuz = −px + (uxx + uyy + uzz ) + G1
Re
1
vt + uvx + vvy + wvz = −py + (vxx + vyy + vzz ) + G2 ,
Re
1
wt + uwx + vwy + wwz = −pz + (wxx + wyy + wzz ) + G3 ,
Re
ux + vy + wz = 0.

3. For a steady state flow with no y-dependence we assume v = 0 and all derivatives with respect
to t or x are zero. Incompressibility, ux +wz = 0 then implies w = 0, so only u is non-zero. The
resulting equations are simply uzz = 0 with u(z = h) = U and u(z = 0) = 0 so the solution is
u = U z/h.

16