Lecture 21 Power Series Method at Singular Points – Frobenius
Theory
10/28/2011
Review.
The usual power series method, that is setting y = ∞ n
P
• n=0 an (x − x0) , breaks down if x0 is a singular
point. Here “breaks down” means “cannot find all solutions”.
• It’s possible to completely solve one class of DE
a x2 y ′′ + b x y ′ + c y = 0 (1)
– Euler equations – at singular points. The solutions involve xr (r not 0, 1, 2, ) and ln x, which are
P∞
not analytic at x0 = 0 and therefore cannot be represented as n=0 an (x − x0)n.
Remark 1. Note that xα+iβ = xα (cos β ln x + sinβ ln x).
• It turns out, once we include these two new ingredients into our ansatz, we can solve equations at
singular points as long as those singular points are “regular singular”:
x0 is “regular singular” point if
◦ x0 is singular;
◦ (x − x0) p(x) and (x − x0)2 q(x) are analytic at x0.
The Method of Frobenius.
The method of Frobenius is a modification to the power series method guided by the above observation.
This method is effective at regular singular points. The basic idea is to look for solutions of the form
∞
X
(x − x0)r an (x − x0)n. (2)
n=0
Consider the equation
y ′′ + p(x) y ′ + q(x) y = 0. (3)
Let x0 be a regular singular point. That is
∞
X ∞
X
p(x) (x − x0) = pn (x − x0)n , q(x) (x − x0)2 = qn (x − x0)n (4)
n=0 n=0
with certain radii of convergence.
To make the following discussion easier to read, we assume x0 = 0.
Substitute the expansion
∞
X
y = xr an x n (5)
n=0
into the equation we get
∞
!′′ ∞
!′ ∞
X X X
r n r n
x an x + p(x) x an x + q(x) xr an xn = 0. (6)
n=0 n=0 n=0
1
2 Lecture 21 Power Series Method at Singular Points – Frobenius Theory
Now compute
∞
!′′ ∞
!′′
X X
xr an x n = an xn+r
n=0 n=0
∞
X
= (n + r) (n + r − 1) an xn+r −2. (7)
n=0
∞
!′ ∞
!′
X X
p(x) xr an xn = p(x) an xn+r
n=0 n=0 !
X∞
= p(x) (n + r) an xn+r −1
n=0 !
∞
X
= (p(x) x) (n + r) an xn+r −2
n=0
! !
∞
X ∞
X
= pn x n (n + r) an xn+r −2
n=0
( n=0 )
∞
X n
X
= pn−m (m + r) am xn+r −2. (8)
n=0 m=0
∞ ∞
! ∞
!
X X X
r n r −2 n n
q(x) x an x = x qn x an x
n=0 " n=0 #n=0
∞
X n
X
= qn−m am xn+r −2. (9)
n=0 m=0
Now the equation becomes
∞
( n
)
X X
(n + r) (n + r − 1) an + [(m + r) pn−m + qn−m] am xn+r −2 = 0. (10)
n=0 m=0
Or equivalently
∞
( n−1
)
X X
[(n + r) (n + r − 1) + (n + r) p0 + q0] an + [(m + r) pn−m + qn−m] am xn+r −2 = 0. (11)
n=0 m=0
This leads to the following equations:
(n = 0): [r (r − 1) + p0 r + q0] a0 = 0, (12)
n−1
X
(n > 1): [(n + r) (n + r − 1) + (n + r) p0 + q0] an + [(m + r) pn−m + qn−m] am = 0. (13)
m=0
0 equation is singled out because if we require a0 0 (which is natural as when a0 = 0, we have
The n = P
y = xr+1 ∞ m
m=0 bm x where bm = am+1.), then it becomes a condition on r:
r (r − 1) + p0 r + q0 = 0. (14)
This is called the indicial equation and will provide us with two roots r1, r2 (Some complicated situation
may arise, we will discuss them later). These two roots are called exponents of the regular singular point
x = 0. After deciding r, the n > 1 relations provide us with a way to determine an one by one.
It turns out that there are three cases: r1 r2 with r1 − r2 not an integer; r1 = r2; r1 − r2 is an integer.
Before we discuss these cases in a bit more detail, let’s state the following theorem which summarizes the
method of Frobenius in its full glory.
10/28/2011 3
Theorem 2. Consider the equation
y ′′ + p(x) y ′ + q(x) y = 0 (15)
at an regular singular point x0. Let ρ be no bigger than the radius of convergence of either (x − x0) p or
(x − x0)2 q. Let r1, r2 solve the indicial equation
r (r − 1) + p0 r + q0 = 0. (16)
Then
1. If r1 r2 and r1 − r2 is not an integer, then the two linearly independent solutions are given by
∞
X ∞
X
y1(x) = |x − x0|r1 an (x − x0)n , y2(x) = |x − x0|r2 ān (x − x0)n. (17)
n=0 n=0
The coefficients an and ān should be determined through the recursive relation
n−1
X
[(n + r) (n + r − 1) + (n + r) p0 + q0] an + [(m + r) pn−m + qn−m] am = 0. (18)
m=0
2. If r1 = r2 , then y1 is given by the same formula as above, and y2 is of the form
∞
X
y2(x) = y1(x) ln |x − x0| + |x − x0|r1 dn (x − x0)n. (19)
n=1
3. If r1 − r2 is an integer, then take r1 to be the larger root (More precisely, when r1, r2 are both complex,
take r1 to be the one with larger real part, that is Re(r1) > Re(r2)). Then y1 is still the same, while
∞
X
y2(x) = c y1(x) ln |x − x0| + |x − x0|r2 en (x − x0)n. (20)
n=0
Note that c may be 0.
All the solutions constructed above converge at least for 0 < |x − x0| < ρ (Remember that x0 is a singular
point, so we cannot expect convergence there).
Remark 3. Note that, although ρ is given by radii of convergence of (x − x0) p and (x − x0)2 q, in practice,
it is the same as the distance from x0 to the nearest singular point of p and q – no (x − x0) factor needed.
Remark 4. The proof of this theorem is through careful estimate of the size of an using the recurrence
relation. See R. P. Agarwal & D. O’Regan “Ordinary and Partial Differential Equations: With Special
Functions, Fourier Series, and Boundary Value Problems” Lecture 5.
Remark 5. In fact the converse of this theorem is also true. That is if all solutions of the equation satisfies
lim |x − x0|r y(x) = 0 (21)
x→x0
for some r, then (x − x0) p and (x − x0)2 q are analytic at x0. This is called Fuchs’ Theorem. Its proof is
a tour de force of complex analysis and can be found in K. Yosida “Lectures on Differential and Integral
Equations”, pp. 37 – 40.
Examples.
In this class we will only require solving equations with r1 − r2 not an integer.
Example 6. Solve
2 ′′ 1 1
x y +x x− y′ + y=0 (22)
2 2
at x0 = 0.
4 Lecture 21 Power Series Method at Singular Points – Frobenius Theory
Solution. We first write it into the standard form
(x − 1/2) ′ 1
y ′′ + y + y = 0. (23)
x 2 x2
x − 1/2 1
Thus p(x) = x and q(x) = 2 x2 . It is clear that x p(x) and x2 q(x) are analytic so 0 is a regular singular
point, and the method of Frobenius applies.
Now we write
∞
X
y= an xn+r. (24)
n=0
Substitute into the equation, we have
∞
!′′ ∞
!′ ∞
X
n+r x − 1/2 X
n+r 1 X
an x + an x + an xn+r = 0. (25)
x 2 x2
n=0 n=0 n=0
As p and q are particularly simple, we write the equation as
∞
!′′ ∞
!′ ∞
!′ ∞
X
n+r
X
n+r 1 X n+r 1 X
an x + an x − an x + an xn+r = 0. (26)
2x 2 x2
n=0 n=0 n=0 n=0
Carrying out the differentiation, we reach
∞ ∞ ∞ ∞
X X 1 X X an n+r −2
(n + r) (n + r − 1) an xn+r −2 + (n + r) an xn+r −1 − (n + r) an xn+r −2 + x =
2 2
n=0 n=0 n=0 n=0
0. (27)
Shifting index:
∞
X ∞
X
(n + r) an xn+r −1 = (n + r − 1) an−1 xn+r −2. (28)
n=0 n=1
Now the equation becomes
∞
r 1 X 1 1
r (r − 1) − + a0 xr −2 + (n + r) (n + r − 1) − (n + r) + an + (n + r − 1) an−1 xn+r −2 =
2 2 2 2
n=1
0. (29)
The indicial equation is
r 1 1
r (r − 1) − + =0 r1 = 1, r2 = . (30)
2 2 2
Their difference is not an integer.
To find y1 we set r = r1 = 1. The recurrence relation
1 1
(n + r) (n + r − 1) − (n + r) + an + (n + r − 1) an−1 = 0 (31)
2 2
becomes
1 1
n (n + 1) − (n + 1) + an + n an−1 = 0 (32)
2 2
which simplifies to
2
an = − an−1. (33)
2n+1
This gives
2n
an = (−1)n a0. (34)
(2 n + 1) (2 n − 1) 3
Setting a0 = 1 we obtain
∞
X 2n
y1(x) = |x| (−1)n xn. (35)
(2 n + 1) (2 n − 1) 3
n=0
10/28/2011 5
To find y2 we set r = r2 = 1/2. The recurrence relation becomes
1
an = − an−1
n
an = (−1)n
1
n!
a0 (36)
so
∞
X 1 n
y2(x) = |x|1/2 (−1)n x = |x|1/2 e−x. (37)
n!
n=0
Finall the general solution is
∞
X 2n
y(x) = C1 |x| (−1)n xn + C2 |x|1/2 e−x. (38)
(2 n + 1) (2 n − 1) 3
n=0
Remark 7. Of course, for anyone who can remember the formulas, there is no need to do all these differ-
entiation and index-shifting.
Remark 8. After getting r1,2, one can also write y1, y2 explicitly and solve an by substituting them into
the equation. See homework 8 solution.
Remark 9. For detailed discussion of the why and how of the other two cases, see my Math 334 2010 note at
[Link] They are not required for Math
334 2011.