2 Frobenius Series Solution of Ordinary Differential Equations
2 Frobenius Series Solution of Ordinary Differential Equations
2 Frobenius Series Solution of Ordinary Differential Equations
At the start of the differential equation section of the 1B21 course last year, you met the linear first-order
separable equation
dy
= y , (2.1)
dx
where is a constant. You were also shown how to integrate the equation to get the solution
y = A e x , (2.2)
where A is an arbitrary integration constant. The solution can be expanded in a power series in x and I want
to show explicitly that this power series does indeed satisfy Eq. (2.1):
1 2 2 1 3 3 1 n1 n1 1 n n
y = A 1 + x + x + x + + x + x +
2 6 (n 1)! n!
dy 2 1 3 2 1 n n1
= A 0 + + x + x + + x + .
dx 2 (n 1)!
Notice that, although similar terms are pushed one to the right, y 0 is clearly equal to y.
Let us now see how this is handled using the summation notation. Here
X 1 n n
y = A x , (2.3)
n=0
n!
dy X 1 X 1
= A n xn1 = A n1 xn1 . (2.4)
dx n=0
(n 1)! n=0
(n 1)!
Eq. (2.4) does not yet look quite like times Eq. (2.3) because one sees xn1 rather than xn . This is a reflection
of all the terms being pushed one over, as noted in the long-hand representation above. To show the equivalence,
first note that the n = 0 term is actually absent from the series for y 0 and so the series effectively starts at
n = 1. Secondly, the n in the series is a dummy variable, like an integration variable, which does not occur in
the final answer. We can therefore make the substitution m = n 1 in Eq. (2.4) to find
dy X 1 m m
= A x , (2.5)
dx m=0
m!
which is exactly y.
Generally the boot is on the other foot. We often end up with a differential equation (e.g. the Legendre
equation) which we cannot solve by inspection as we have done here. We want then to develop techniques for
finding directly series solutions for differential equations.
Let us try for a solution of Eq. (2.1) in the form
X
X
y = xk an xn = an xn+k , (2.6)
n=0 n=0
where a0 6= 0. Thus the value of the index k is defined by the condition that the first term a0 doesnt vanish.
Differentiating term-by-term leads to
dy X
= (n + k) an xn+k1 . (2.7)
dx n=0
Inserting Eqs. (2.6) and (2.7) into the differential equation, we find
X
X
(m + k) am xm+k1 = an xn+k (2.8)
m=0 n=0
29
where, to avoid confusion, I have used a different dummy variable m on the left from the n on the right. Since
this equation has to be true for a range of values of x around the origin, it must be true separately for each
power of x. We must therefore compare the coefficients of different powers of x on the two sides of Eq. (2.8).
To facilitate the comparison, let m = n + 1 on the left hand side.
X
X
n+k
(n + k + 1) an+1 x = an xn+k . (2.9)
n=1 n=0
Since we now have powers xn+k explicitly on both sides of the equation, it is simple to deduce that (n + k +
1) an+1 = an , i.e.
an+1 = an . (2.10)
n+k+1
This is a very simple example of a recurrence relation, which allows us to evaluate all the higher coefficients
from the first one. It does however contain the unknown index k. How is its value fixed?
The lowest power of x on the left hand side of Eq. (2.9) is a0 k xk1 (corresponding to n = 1), and there is
nothing like this on the right hand side because the sum there starts with n = 0. The term must therefore be
made to cancel
a0 k = 0 . (2.11)
Since a0 6= 0, this can only happen if k = 0. Eq. (2.11) is a very simple example of what is called an indicial
equation; it fixes the index k.
With k = 0, the recurrence relation becomes
an+1 = an . (2.12)
n+1
To use the same notation as in the previous solution, let a0 = A. Then
a1 = A ,
2
a2 = a1 = A,
2 2
3
a3 = A etc.
6
In general
n+1
an+1 = A = A. (2.13)
n+1 n n1 1 (n + 1)!
30
Looking first at the lowest term, corresponding to n = 0, we see that
a0 ( k) xk1 = 0 .
an ( k n) = n an = 0 for n > 0 .
y = A e/x ,
which has an essential singularity at x = 0 and for which no power series in x is possible in this region. How
does this manifest itself in the Frobenius method?
X
y = an xn+k , hence
n=0
dy X
= (n + k) an xn+k1 ,
dx n=0
y X
= an xn+k2 .
x2 n=0
The lowest power of x is xk2 and this is multiplied by a0 . Provided that 6= 0, the only solution would
require a0 = 0. But the value of k was determined by requiring that a0 6= 0. These two conditions are in mutual
contradiction and so there is no power series solution in x.
Example 3
Consider the differential equation
dy y
= ,
dx 1x
which has solution y = A/(1 x). By the series method we would obtain
X
X
(1 x) an (n + k) xn+k1 = an xn+k .
n=0 n=0
Multiplying out by the x factor and taking it over to the other side,
X
X
an (n + k) xn+k1 = an (n + k + 1) xn+k .
n=0 n=0
The indicial equation is obtained by demanding that the lowest power, corresponding to n = 1 on the left
hand side, should vanish. This requires k = 0. The recurrence relation is then
an+1 = an ,
31
which means that all the coefficients are equal to a0 . The full solution is therefore
X
y = a0 xn .
n=0
This solution is fine provided that the series converges and you saw in the 1B21 course that this geometric
series diverges if | x |> 1. This is not entirely unexpected because the equation has a regular singular point at
x = 1. In summary:
a: You cannot make any power series expansion about an irregular point, i.e. one where y 0 diverges faster than
1/x.
b: The power series may not converge if x is too large.
Both these problems are present in second order equations, to which we now turn.
d2 y
+ 2 y = 0 . (2.14)
dx2
The most general solution is
y = A cos x + B sin x , (2.15)
where A and B are arbitrary constants to be fixed by the boundary conditions. A second order linear equation
has two arbitrary constants.
Note that the two individual solutions of this equation, viz cos x and sin x, are respectively even and odd
functions of the independent variable x. Why is this? First write down Eq. (2.14) for a function y = f (x) and
then let x x. We then have the two equations
d2 f (x)
+ 2 f (x) = 0 ,
dx2
d2 f (x)
+ 2 f (x) = 0 . (2.16)
dx2
Thus f (x) satisfies the same equation as f (x) and this is because all the operators in Eq. (2.14) are even;
d2
doesnt change when x x. Any linear combinations of f (x) and f (x) also satisfy the equations. In
dx2
particular, the even and odd combinations
1
fe (x) = 2 [f (x) + f (x)] , (2.17)
1
fo (x) = 2 [f (x) f (x)] (2.18)
also satisfy the equation. This is the real reason why cos x and sin x are solutions to the oscillator equation.
Of course, this argument does not say that the basic solutions have to be either even or odd, but one can always
choose them so to be. We will use this argument when we study the Legendre equation in detail.
Try now for a series solution of the oscillator equation;
X
y = an xn+k ,
n=0
dy X
= (n + k) an xn+k1 ,
dx n=0
2
d y X
= (n + k)(n + k 1) an xn+k2 . (2.19)
dx2 n=0
32
Inserting these into Eq. (2.14), we find that
X
X
(n + k)(n + k 1) an xn+k2 + 2 an xn+k = 0 . (2.20)
n=0 n=0
The indicial equation is obtained by looking at the coefficient of the lowest power here, viz xk2 . Since this only
occurs in the first sum (for n = 0), it must vanish:
k(k 1) a0 xk2 = 0 .
There are therefore two possible values of the index, k = 0 or k = 1, and this is quite typical for a second order
equation.
Changing the dummy index in the first sum by n n + 2,
X
X
(n + k + 2)(n + k + 1) an+2 xn+k + 2 an xn+k = 0 , (2.21)
n=2 n=0
all the powers now look the same and so we can compare coefficients to obtain the recurrence relation
and X 1 a0
y = a0 x (1)n/2 (x)n = sin x . (2.29)
n even
(n + 1)!
This gives the expected solutions and so we can now start using the technique to study more complicated
second order equations.
33
2.3 Singularities of Second Order Equations
Consider the general linear homogeneous second order equation
d2 y dy
2
+ p(x) + q(x) y = 0 , (2.30)
dx dx
where p(x) and q(x) are functions just of x. If we want to do a power series expansion near the point x0 , the
behaviour of p(x) and q(x) in the vicinity of this point is of crucial importance.
If p(x) and q(x) are finite, single valued and differentiable at x0 , then x0 is called a regular point or an
ordinary point. The equation is said to be regular at x0 . In this case lim p(x) and lim q(x) both exist, i.e.
xx0 xx0
are finite. For the harmonic oscillator equation, p(x) = 0 and q(x) = 2 , so that all points of the equation are
regular. If, on the other hand, either of the two limits is infinite, we say that x0 is a singular point; the equation
is singular at x0 .
Suppose that x0 is a singular point, but that
both exist, the differential equation is said to have a regular singularity at x0 . If either limit is infinite, there
is an essential singularity at x = x0 .
This classification is very important because if x0 is a regular point then we can always find two independent
series solutions of the form
X
y(x) = an (x x0 )n+k , (2.32)
n=0
which are convergent for all values of x between x0 and the nearest singular point. Furthermore, it can be shown
(Fuchs theorem) that if x0 is regular singular point then there exists at least one series solution of the form of
Eq. (2.32).
If, on the other hand, x0 is an essential singularity of the equation then no power series solution in x x0
exists.
Roots differing by an integer
Consider the equation
d2 y dy
x(x 1)2
+ 3x +y =0.
dx dx
Comparing this with the standard form, we see that p(x) = 3/(x 1) and q(x) = 1/x(x 1). Thus x = 0
and x = 1 are regular points of the differential equation and so we can expect to get at least one power series
solution in x. Inserting
X
y = an xn+k ,
n=0
dy X
= (n + k) an xn+k1 ,
dx n=0
d2 y X
= (n + k)(n + k 1) an xn+k2
dx2 n=0
Hence
X
X
n+k
an x [(n + k)(n + k 1) + 3(n + k) + 1] = (n + k + 1)(n + k) an+1 xn+k .
n=0 n=1
34
The indicial equation comes from looking at the lowest power of x, which is given by n = 1 on the right
hand side. This gives k(k 1) = 0, i.e. k = 1 or k = 0. The recurrence relation is
Note that this series converges for | x |< 1; the divergence at x = 1 is due to the singular point there.
On the other hand, when the index k = 0, we are in trouble because the recurrence relation is
n+1
an+1 = an .
n
If you try to calculate a1 by putting n = 0 you see that the whole thing blows up. Hence there is not a second
series solution at x = 0. Fuchs theorem only guaranteed that there would be one solution of this kind; the
other solution is going to be nasty at x = 0.
One can find the second (irregular) solution by letting y(x) = y1 (x) v(x) and getting a simpler equation for
v(x). Normally, as in this case, v(x) has a nasty `n(x) term in it. This is always the case if the indicial equation
has equal roots. This happens for Bessels equation which one often comes across in problems with cylindrical
symmetry.
35