Вы находитесь на странице: 1из 4

92.

445/545 Partial Differential Equations


Classification of Second Order Linear PDEs and Reduction to Canonical Form

A second order pde in 2 independent variables is linear if it can be written in the form
a(x, y)uxx + 2b(x, y)uxy + c(x, y)uyy + d(x, y)ux + e(x, y)uy + f (x, y)u = g(x, y)

(1)

This pde is said to be hyperbolic at the point (x, y) if b2 ac > 0, parabolic at (x, y) if b2 ac = 0,
or elliptic at (x, y) if b2 ac < 0.
The pde is hyperbolic (or parabolic or elliptic) on a region D if the pde is hyperbolic (or parabolic
or elliptic) at each point of D.
A second order linear pde can be reduced to so-called canonical form by an appropriate change of
variables = (x, y), = (x, y).



The Jacobian of this transformation is defined to be J = x
x

y
y




= x y x y .

The Jacobian should be nonzero to ensure that the transformation is invertible. In that case, we can,
at least in principle, solve for x and y as functions of and . We let w(, ) = u(x(, ), y(, )).
Using the Chain Rule, one can show that equation (1) takes the following form when expressed in
terms of the variables and :
Aw + 2Bw + Cw + Dw + Ew + F w = G(, )

(2)

where
A = ax2 + 2bxy + cy2
B = axx + b (x y + y x ) + cy y
C

= ax2 + 2bxy + cy2

D = axx + 2bxy + cyy + dx + ey


E = axx + 2bxy + cyy + dx + ey
F

= f (x(, ), y(, ))

G = g(x(, ), y(, ))
As shown in Pinchover & Rubinsteins book, the type of the equation is not affected
by the change of variables. If equation (1) is hyperbolic (or parabolic, or elliptic) at
the point (x, y), then equation (2) is also hyperbolic (or parabolic, or elliptic) at the
point (, ).
Note that the expressions for A and C can be factored:
A =
C

ih
i
p
p
1h
ax + (b + b2 ac)y ax + (b b2 ac)y
a
ih
i
p
p
1h
ax + (b + b2 ac)y ax (b + b2 ac)y
a

(3)
(4)

1. Hyperbolic Equations
The canonical form of a hyperbolic equation is
+ Ew
+ F w = G(,
)
w + Dw

(5)

The canonical variables and for a hyperbolic pde satisfy the equations


ax + b +
ax + b

b2 ac y


(6)

(7)

and

b2 ac y

making coefficients A and C in (2) zero by virtue of (3) and (4).


The families of curves = constant and = constant are the characteristic curves. Hyperbolic equations have two families of characteristic curves.
Example. Consider the pde uxx + 4uxy + ux = 0. In this problem, a = 1, 2b = 4, and c = 0,
so b2 ac = 22 (1)(0) = 4 > 0, and the given pde is hyperbolic on the entire xy plane.
Equations (6) and (7) reduce to x + 4y = 0 and x = 0. Solving these equations by the
method of characteristics, we find that = f (4x y) and = g(y). For simplicity we take
= 4x y and = y . We therefore have
ux = w x + w x = 4w
uxx = 4 [w x + w x ] = 16w
uxy

= 4 [w y + w y ] = 4w + 4w

Therefore, the given pde uxx + 4uxy + ux = 0 becomes


1
[16w ] + 4 [4w + 4w ] + [4w ] = 0, or 16w + 4w = 0, or w + w = 0 .
4
2. Parabolic Equations
The canonical form of a parabolic equation is
+ Ew
+ F w = G(,
)
w + Dw

(8)

For a parabolic equation, b2 ac = 0 so equations (3) and (4) reduce to the same equation:
A =
C =

1
[ax + by ]2
a
1
[ax + by ]2
a

(9)
(10)

Instead of two equations like (6) and (7) for hyperbolic equations, we have just the single
equation ax + by = 0 (or ax + by = 0). Parabolic equations have only one family
of characteristic curves.
We choose the canonical variable to be a solution of the equation
ax + by = 0

(11)

and we choose to be any function which makes the Jacobian x y y x nonzero. The
choice of makes C = 0. Because B 2 AC = 0, that makes B = 0 and therefore the only
nonzero second derivative term in the pde (2) is Aw .

Example. Consider the pde x2 uxx 2xyuxy + y 2 uyy + xux + yuy = 0 for x > 0. (Pinchover
& Rubinstein p. 70). In this problem, a = x2 , b = xy, and c = y 2 so
b2 ac = (xy)2 x2 y 2 = 0 and the given pde is parabolic on the half-plane x > 0. Equation
(11) becomes x2 x xyy = 0, or xx yy = 0. Using the method of characteristics, we
find that = f (xy). For simplicity we take = xy. If we just take = x, the Jacobian of
the transformation becomes xy y x = (1)(x) (0)(y) = x > 0. We can therefore take
= x and = xy . With this choice, we obtain
ux = w x + w x = w + yw
uy = w y + w y = 0 w + xw = xw

uxx = [w x + w x ] + y [w x + w x ] = w + 2yw + y 2 w
uxy = [w y + w y ] + w + y [w y + w y ] = w + xw + xyw
uyy

{z

Product Rule

= x [w y + w y ] = x2 w

Therefore,
the given pdeix2 uxx 2xyuxy + y 2 uyy + xux + hyuy = 0i becomes
h
x2 w + 2yw + y 2 w 2xy [w + xw + xyw ] + y 2 x2 w + x [w + yw ] + y [xw ] = 0,

1
or x2 w + xw = 0 or w + w = 0 . (Here we have used the fact that = x.)

3. Elliptic Equations
The canonical form of an elliptic equation is
+ Ew
+ F w = G(,
)
w + w + Dw

(12)

For an elliptic equation, b2 ac < 0 so equations (3) and (4) contain complex coefficients and
have no real solutions. Elliptic equations have no characteristic curves.
In order for (2) to reduce to (12), we must have A = C and B = 0, or A C = 0 and B = 0:


a x2 x2 + 2b (xy xy ) + c y2 y2

= 0 and

ax x + b (x y + y x ) + cy y = 0

Adding the first of these equations to i times the second, we obtain


a2x + 2bx y + c2y = 0

(13)

where = + i. Factoring equation (13), we obtain


 ih

 i

p
p
1h
ax + b + i ac b2 y ax + b i ac b2 y = 0
a

(14)

We will take to be the solution of




ax + b + i ac b2 y = 0
and then we will use the change of variables given by = <() and = =().

(15)

Example. Consider the pde uxx + xuyy = 0 for x > 0. (Pinchover & Rubinstein p. 72).
In this problem, a = 1, b = 0, and c = x so b2 ac = 02 (1)(x) = x < 0 and the

given pde is elliptic on the half-plane x > 0. Equation (15) becomes x + i xy = 0.


We take the initial data curve to be the x axis, so the intial curve can be parameterized

dx
dy
as x = s, y = 0. The characteristic curves satisfy the conditions
= 1 and
= i x.
dt
dt
dx
= 1 x = t + f (s). Because x = s on (where t = 0), f (s) must equal s. Therefore,
dt

dy
2
= i x = i t + s y = i (t + s)3/2 + g(s).
x = t + s and
dt
3
2
2
Because y = 0 on (where t = 0), g(s) must equal i 23 s3/2 . Therefore, y = i (t + s)3/2 i s3/2
3
3


2i
2i
3y
y = x3/2 s3/2 s3/2 = x3/2 + i
.
3
3
2

d
dx dy
=
+
= x 1 + y i x = x + i xy = 0 from the given
On characteristics,
dt
x dt
y dt
 
d
3y
3/2
3/2
3/2
pde.
= 0 = h(s). For simplicity we take h(s) = s . Therefore, = s = x + i
dt
2
3y
. With this choice, we obtain
so = <() = x3/2 and = =() =
2
3 1/2
3
x w + w 0 = x1/2 w
2
2
3
3
= w y + w y = 0 w + w = w
2
2


3 1/2
3 1/2
3
3
3
3
9x
=
x
w + x [w x + w x] = x1/2 w + x1/2 x1/2 w = x1/2 w +
w
4
2
4
2
2
4
4


3
3 3
9
=
[w y + w y ] =
w = w
2
2 2
4

ux = w x + w x =
uy
uxx
uyy

3 1/2
9x
9
x
w +
w + x w = 0, or
4
4
4
1
1
9x [w + w ] + 3x1/2 w = 0, or w + w + 3/2 w = 0, or w + w + w = 0.
3
3x
Therefore, the given pde uxx + xuyy = 0 becomes

Here we have used the fact that = x3/2 .

Вам также может понравиться