B
Similarity solutions
Similarity solutions to PDEs are solutions which depend on certain groupings
of the independent variables, rather than on each variable separately. Ill show
the method by a couple of examples, one linear, the other nonlinear.
B.1
Linear example: the heat equation
The heat equation in one dimension is
ut = uxx .
This form of equation arises often within boundary layers in a PDE: the firstorder derivative may be in an unstretched direction and the higher-order derivative come from the component of 2 in a stretched direction, if the coefficient
of 2 in the original equation was small (i.e. an advectiondiffusion equation
with weak diffusion).
We introduce the dilation transformation
z = a x,
s = b t,
v = c u
under which t = b s and so on, and the PDE becomes
bc vs = 2ac vzz .
We look for values under which our PDE is unchanged: in this case we have
b c = 2a c and so b = 2a. That tells us that, provided b = 2a, if u(x, t) is a
solution of the original equation, then so is c u(a x, b t). But what use is this
observation?
The key thing is to note that the combinations
vsc/b = c u(b t)c/b = utc/b
and
zsa/b = a x(b t)a/b = xta/b
are both unchanged by the transformation , which suggests we look for a solution
which combines these two forms:
u = tc/b f (xta/b ).
Returning to our specific example, we neededb = 2a which means the combination for the argument of f is xt1/2 = x/ t. We introduce a new variable
for this combination
= x/ t
u = tc/b f ()
and substitute into the original equation:
1 3/2
c
1
c c/b1
c/b
=
f () + t f () xt
f () f () tc/b1
ut = t
b
2
b
2
0 = ut uxx
uxx = tc/b1 f ()
1
c
f () f () f () tc/b1
=
b
2
15
We have reduced a constant-coefficient PDE to a variable-coefficient ODE:
1
c
f () + f () f () = 0.
2
b
For a linear equation like this, the ratio c/b is not determined by the equation
and we have some flexibility to use in meeting boundary conditions.
B.1.1
Fixed boundary conditions
Suppose our original equation came with the boundary conditions
u(x, 0) = 0, x > 0
u(x, t) 0, x
u(0, t) = u0 , t > 0.
Transforming these into the new variables gives
tc/b f () 0, , even as t 0
tc/b f (0) = u0 , t > 0.
The first of these gives two conditions: f () 0 as and also c/b 0.
The second, on the other hand, can only be satisfied if c/b = 0 and then we
have the transformation
= xt1/2
u = f ()
1
f () 0, as ,
f () + f () = 0
2
We can integrate this once to obtain
2
f () = C1 exp
4
f (0) = u0 .
2
p
1/2
f () = C1
exp
dp + C2 = C1 () erf
+ C2
4
2
0
Rx
2
where erf (x) := (2/ ) 0 et dt. Then the boundary conditions lead to
= u0 erfc
.
f () = u0 1 erf
2
2
Z
The solution of the original equation is
u = u0 erfc
B.1.2
x
.
2 t
Flux boundary conditions
On the other hand, if we have a flux boundary condition on u:
u(x, 0) = 0, x > 0
u(x, t) 0, x
ux (0, t) = Q, t > 0.
then we still have the conditions c/b 0 and f () 0 as , but now
tc/b1/2 f (0) = Q, t > 0,
16
which can only be satsfied by taking c/b = 1/2. The final transformation is
u = t1/2 f ()
= xt1/2
giving the ODE and boundary conditions
2f () + f () f () = 0
f () 0, ,
f (0) = Q.
It is easy to spot one solution to this equation: f () = C1 . So we use the
reduction-of-order trick and set f () = g() to get:
2g () + (4 + 2 )g () = 0
Now we can integrate once:
2
C1
g () = 2 exp
4
f () = C1
g() = C1
Z
2
1
p
dp + C2
exp
p2
4
2
1
p
dp + C2 .
exp
p2
4
Integrating by parts gives
2
f () = C1 exp
4
erf
+ C2 ,
2
2
and after applying the boundary conditions the solution becomes
2
.
f () = Q erfc
exp
4
B.2
Nonlinear example: KdV equation
The Kortewegde Vries equation is
ut + 6uux + uxxx = 0.
Setting z = a x, s = b t and v = c u gives
bc vs + 6a2c vvz + 3ac vzzz = 0,
which gives us the conditions for invariance:
b c = a 2c = 3a c :
b = 3a, c = 2a.
The transformation u = t2/3 f (), = xt1/3 converts the KdV equation to
2
t5/3 f () + f () 6f ()
f () = 0,
3
3
2
f () = 0.
f () + f () 6f ()
3
3
This ordinary differential equation can be shown to have the so called Painlev
e
property, meaning that it does not have a movable singular point. A movable
17
singular point is a point where the solution becomes singular, whose location
depends on the arbitrary constants of integration. For instance, the equation
y = y 2 has the solution y = (C )1 , which has a singular point whose location
depends on the arbitrary constant of integration, C. Then this equation does
not have the Painleve property. There is a conjecture3 , that PDEs that reduce
to ODEs having the Painleve property are integrable: that is, they admit soliton
solutions and are solvable by the inverse scattering transform.
Thus although we cant solve the ODE above in general, the act of deriving it
can give us useful information about the original PDE.
3 Ablowitz
et al., J. Math. Phys. 21, 715 (1980)
18
