Jim Lambers
MAT 606
Spring Semester 2015-16
Lecture 9 Notes
These notes correspond to Sections 3.3-3.5 in the text.
Solving Canonical Hyperbolic Equations
Suppose a hyperbolic PDE is transformed into the simple canonical form
uξη = 0.
Then, this PDE is easily solved by integration, which yields the general solution
u(ξ, η) = F (ξ) + G(η),
where F and G are functions that can be determined from the initial conditions. This will be used
to derive a solution formula for the wave equation.
Canonical Form of Parabolic Equations
We now investigate the transformation of a parabolic PDE into the canonical form
uξξ + `1 [u] = G,
where `1 is a first-order differential operator. Using the notation from our general discussion of
coordinate change, this transformation is accomplished by ensuring that the coefficients of the
transformed PDE satisfy B = C = 0. However, because we already know that the original and
transformed PDE are both parabolic, it is sufficient to ensure that C = 0, because any nonsingular
2
coordinate transformation will ensure that δ(L) = B − AC = 0, from which it follows that if
C = 0, then B = 0 as well. We therefore need to solve
1
C = Aηx2 + 2Bηx ηy + Cηy2 = (Aηx + Bηy )2 = 0,
A
where we have used the fact that δ(L) = B 2 − AC = 0. Along coordinate curves η = constant, we
have
dη = ηx dx + ηy dy = 0,
which yields
dy ηx B
=− = .
dx ηy A
Solving this ODE provides a formula for η in terms of x and y. The only requirement for the other
characteristic variable, ξ, is that the Jacobian of the coordinate change
J = ξx ηy − ηx ξy
is nonzero. Note that whereas hyperbolic PDE have two families of characteristics, defined by
ξ = constant and η = constant, parabolic PDE have only one.
1
Example 1 The PDE
x2 uxx − 2xyuxy + y 2 uyy + xux + yuy = 0, x > 0,
is parabolic, as B 2 − AC = (−xy)2 − x2 y 2 = 0. The change of variables that makes C̄ = 0 satisfies
the ODE
dy −xy y
= 2 =− .
dx x x
Rearranging yields the separable equation
1 1
dy = − dx,
y x
which can be integrated to obtain η = xy. For simplicity, we choose ξ = x, which ensures that
J = ξx ηy − ηx ξy = (1)(x) − y(0) = x > 0. We then obtain the PDE
1
uξξ + uξ = 0.
ξ
To solve this, we let w = uξ to obtain a first-order ODE
1
wξ + w = 0,
ξ
which has the solution
R f (η)
w(ξ, η) = e− 1/ξ dξ
f (η) = ,
ξ
where f is arbitrary. We then have
Z
v= w dξ = f (η) ln |ξ| + g(η),
where g is also arbitrary. This yields the general solution
u(x, y) = f (xy) ln |x| + g(xy).
Canonical Form of Elliptic Equations
For elliptic PDE, we seek the canonical form
uξξ + uηη + `1 [u] = G,
where, as before, `1 is a first-order differential operator. That is, we require A = C and B = 0.
That is,
A(ξx2 − ηx2 ) + 2B(ξx ξy − ηx ηy ) + C(ξy2 − ηy2 ) = 0, (1)
Aξx ηx + B[ξx ηy + ξy ηx ] + Cξy ηy = 0. (2)
√
If we define φ = ξ + iη, where i = −1, then we have
φ2x = ξx2 − ηx2 + 2iξx ηx ,
φx φy = ξx ξy − ηx ηy + i(ξx ηy + ξy ηx ),
φ2y = ξy2 − ηy2 + 2iξy ηy .
2
Substitution into the equations for (1), (2) yields
Aφ2x + 2Bφx φy + Cφ2y = 0,
which is the same equation solved in the hyperbolic case. Proceeding as in that case, we obtain
ODEs √
dy B ± i AC − B 2
= ,
dx A
which yield formulas for two families of complex characteristics, φ1 and φ2 , that are complex
conjugates of one another. We then obtain
ξ = Re φ1 , η = Im φ1 .
Expressing the original PDE in the new variables ξ and η yields the desired canonical form.
Example 2 The Tricomi equation
uxx + xuyy = 0, x > 0,
is elliptic, as B 2 − AC = 02 − 1(x) = −x < 0. The complex characteristics φ1 and φ2 arise from
solutions of the ODEs
dy √
= ±i x,
dx
which are
3
φ1 , φ2 = y ± ix3/2 .
2
It follows that the real characteristic variables are
3
ξ = y, η = x3/2 .
2
From the formulas for the transformed coefficients, we obtain the PDE
9 9 3
uxx + xuyy = xuξξ + xuηη + x−1/2 uη = 0,
4 4 4
which, in the characteristic variables, becomes
9 1
uξξ + uηη + uη = 0,
4 3η
which is in canonical form. 2
