Introduction The “naı̈ve” Frobenius method The general Frobenius method

The Method of Frobenius

R. C. Daileda

Trinity University

Partial Differential Equations

April 7, 2015

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Motivating example
Failure of the power series method

Consider the ODE 2xy ′′ + y ′ + y = 0. In standard form this is

1 ′ 1 1
y ′′ + y + y = 0 ⇒ p(x) = q(x) = , g (x) = 0.
2x 2x 2x
In exercise A.4.25 you showed that 1/x is analytic at any a > 0,
with radius R = a. Hence:
Every solution of 2xy ′′ + y ′ + y = 0 is analytic at a > 0
with radius R ≥ a (i.e. given by a PS for 0 < x < 2a).
However, since p, q, g are continuous for x > 0, general theory
guarantees that:

Every solution of 2xy ′′ + y ′ + y = 0 is defined for all x > 0.

Question: Can we find series solutions defined for all x > 0?

Daileda Frobenius’ Method
Introduction The “naı̈ve” Frobenius method The general Frobenius method

Even though p(x) = q(x) = 1/2x is not analytic at a = 0, we

nonetheless assume
∞
X
y= an x n (with positive radius)
n=0

and see what happens. Plugging into the ODE and collecting
common powers of x leads to
−an
an+1 = for n ≥ 1,
(n + 1)(2n + 1)

and then choosing a0 = 1 yields the first solution

∞ √
(−1)n 2n X (−1)n 2n 
an = ⇒ y1 = x n = cos 2x .
(2n)! (2n)!
n=0

But choosing a0 = 0 gives an = 0 for all n ≥ 0, so that y2 ≡ 0.

Daileda Frobenius’ Method
Introduction The “naı̈ve” Frobenius method The general Frobenius method

What now?
To find a second independent solution, we instead assume
∞
X ∞
X
y = xr an x n = an x n+r (a0 6= 0)
|n=0{z } n=0
PS with R>0

for some r ∈ R to be determined. Since

∞
X ∞
X
y′ = (n + r )an x n+r −1 , y ′′ = (n + r )(n + r − 1)an x n+r −2 ,
n=0 n=0

plugging into the ODE gives

∞
X ∞
X ∞
X
n+r −2 n+r −1
2x (n+r )(n+r −1)an x + (n+r )an x + an x n+r = 0.
n=0 n=0 n=0

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Distributing the 2x and setting m = n − 1 in the first two series

yields
∞
X ∞
X
m+r
2(m + r + 1)(m + r )am+1 x + (m + 1 + r )am+1 x m+r
m=−1 m=−1
X∞
+ an x n+r = 0
n=0

or, replacing m with n

(2r (r − 1) + r )a0 x r −1
| {z }
n=−1
∞
X
+ ((n + r + 1) (2(n + r ) + 1) an+1 + an ) x n+r = 0.
n=0

This requires the coefficients on each power of x to equal zero.

Daileda Frobenius’ Method
Introduction The “naı̈ve” Frobenius method The general Frobenius method

That is
1
r (2r − 1)a0 = 0 ⇒ r (2r − 1) = 0 ⇒ r = 0, ,
a0 6=0 2

and (n + r + 1)(2n + 2r + 1)an+1 + an = 0, or

−an
an+1 = for n ≥ 0.
(n + r + 1)(2n + 2r + 1)

Each value of r gives a different recurrence:

−an
r = 0 ⇒ an+1 = ,
(n + 1)(2n + 1)
1 −an −an
r= ⇒ an+1 = = .
2 (n + 3/2)(2n + 2) (2n + 3)(n + 1)

Notice that the first is the original recurrence!

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Taking a0 = 1 in the second we eventually find that

∞ √ 
(−1)n 2n X (−1)n 2n 1
an = ⇒ y2 = x 1/2 x n = √ sin 2x .
(2n + 1)! (2n + 1)! 2
n=0

This gives the second (linearly independent) solution to the ODE,

and we have the general solution
√  √ 
y = c1 y1 + c2 y2 = c1 cos 2x + c2′ sin 2x (x > 0).

Remarks:
The fact that both series yielded familiar functions is simply a
coincidence, and should not be expected in general.
One could also have obtained y2 from y1 (or vice-verse) using
a technique called reduction of order.

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Method of Frobenius - First Solution

When will the preceding technique work at an “extraordinary”

point? Here’s a partial answer:
Theorem
Suppose that at least one of p(x) or q(x) is not analytic at x = 0,
but that both of xp(x) and x 2 q(x) are. If

lim xp(x) = p0 and lim x 2 q(x) = q0 ,

x→0 x→0

then there is a solution to y ′′ + p(x)y ′ + q(x)y = 0 (x > 0) of the

form
X ∞
y = xr an x n (a0 6= 0),
n=0

where r is a root of the indicial equation r 2 + (p0 − 1)r + q0 = 0.

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Remarks

Under the hypotheses of the theorem, we say that a = 0 is a

regular singular point of the ODE.
Suppose the roots of the indicial equation are r1 and r2 .
If r1 − r2 6∈ Z, then both r = r1 and r = r2 yield (linearly
independent) solutions.
If r1 − r2 ∈ Z, then only r = max{r1 , r2 } is guaranteed to work.
The other may or may not.

If the PS for xp(x) and x 2 q(x) both converge for |x| < R, so
does the PS factor of y .
We can talk about regular singularities at any x = a by
instead considering (x − a)p(x), (x − a)2 q(x), lim , and
x→a
writing the solution in powers of (x − a) .

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Example
Find the general solution to x 2 y ′′ + xy ′ + (x − 2)y = 0.

In standard form this ODE has

1 x −2
p(x) = and q(x) = ,
x x2
neither of which is analytic at x = 0. However, both

xp(x) = 1 and x 2 q(x) = x − 2

are analytic at x = 0, so we have a regular singularity with

p0 = lim xp(x) = 1 and q0 = lim x 2 q(x) = − 2.

x→0 x→0

The indicial equation is

√
r 2 + (1 − 1)r − 2 = 0 ⇒ r = ± 2.

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Applying the method of Frobenius, we set

∞
X ∞
X
r n
y =x an x = an x n+r (a0 6= 0)
n=0 n=0

and substitute into the ODE, obtaining

∞
X
2



r
(r − 2)a0 x + (n + r )2 − 2 an + an−1 x n+r = 0.
n=1

Hence we must have r 2 − 2 = 0 (which we already knew) and

−an−1 −an−1
an = 2
= for n ≥ 1.
(n + r ) − 2 n(n + 2r )
Taking a0 = 1 one readily sees that
(−1)n
an = .
n!(1 + 2r )(2 + 2r )(3 + 2r ) · · · (n + 2r )

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

√ √ √
Since the difference of the roots is 2 − (− 2) = 2 2 6∈ Z, the
two r -values give independent solutions:
∞
√
2
X (−1)n x n
y1 = x √ √ √ √ ,
n=0
n!(1 + 2 2)(2 + 2 2)(3 + 2 2) · · · (n + 2 2)
∞
√
− 2
X (−1)n x n
y2 = x √ √ √ √ ,
n=0
n!(1 − 2 2)(2 − 2 2)(3 − 2 2) · · · (n − 2 2)

and the general solution (for x > 0) is

y = c1 y1 + c2 y2 .

Remark: Because xp(x) = 1 and x 2 q(x) = x − 2 both have

infinite radius of convergence, so do both series above.

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Method of Frobenius - Second Solution

What do we do if the indicial roots differ by an integer?
Theorem
Suppose that x = 0 is a regular singular point of
y ′′ + p(x)y ′ + q(x)y = 0, and that the roots of the indicial
equation are r1 and r2 , with r1 − r2 ∈ N0 .
If r1 = r2 = r , the second solution has the form
∞
X
r
y2 = y1 ln x + x bn x n .
n=1

If r1 > r2 (so that y1 uses r1 ), the second solution has the form
∞
X
r2
y2 = ky1 ln x + x bn x n (b0 6= 0).
n=0
Daileda Frobenius’ Method
Introduction The “naı̈ve” Frobenius method The general Frobenius method

Example
Find the general solution to xy ′′ + (1 − x)y ′ + 2y = 0, x > 0.

In standard form we have

1−x 2
p(x) = and q(x) = ,
x x
which are non-analytic at x = 0, and

xp(x) = 1 − x and x 2 q(x) = 2x,

which are. This makes x = 0 a regular singularity with

p0 = lim 1 − x = 1 and lim 2x = 0,

x→0 x→0

and indicial equation

r 2 + (1 − 1)r + 0 = 0 ⇒ r = 0.

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Since r = 0 is a double root, we are guaranteed only one solution

of the form
∞
X ∞
X
y = xr an x n = an x n .
n=0 n=0

Plugging this into the ODE and simplifying leads to the recursion
(n − 2)an
an+1 = for n ≥ 0.
(n + 1)2
Taking a0 = 1 we find that
−2a0 −a1 1 0 · a2
a1 = 2
= − 2, a2 = 2 = , a3 = 2 = 0,
1 2 2 3
and hence a4 = a5 = a6 = · · · = 0 as well. So our first solution is

x2
y1 = 1 − 2x + .
2

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

According to the theorem, a second independent solution has the

form
∞
X
y2 = y1 ln x + x 0 bn x n ,
n=1
| {z }
w
and we need to solve for the bn . The product rule gives us
y1
y2′ = y1′ ln x + + w ′,
x
2y ′ y1
y2′′ = y1′′ ln x + 1 − 2 + w ′′ ,
x x
and plugging these into xy2′′ + (1 − x)y2′ + 2y2 = 0 we obtain


xy1′′ + (1 − x)y1′ + 2y1 ln x − y1 + 2y1′ + xw ′′ + (1 − x)w ′ + 2w = 0,
| {z }
=0
xw ′′ + (1 − x)w ′ + 2w = −2y1′ + y1 .

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

P∞
We now plug y1 = 1 − 2x + x 2 /2 and w = n=1 bn x
n into this
equation to obtain a recurrence for the bn :
∞
X
x2
b1 + (n + 1)2 bn+1 − (n − 2)bn x n = 5 − 4x + .
2
n=1

Hence
1
b1 = 5, 4b2 + b1 = −4, 9b3 = ,
2
and
(n − 2)bn 36b3
bn+1 = ⇒ bn = for n ≥ 3.
(n + 1)2 n(n − 1)(n − 2)n!
Thus, since b3 = 1/18,
  ∞
x2 9 2 X xn
y2 = 1 − 2x + ln x + 5x − x + 2 .
2 4 n(n − 1)(n − 2)n!
| {z } n=3
| {z }
y1 w

Daileda Frobenius’ Method

Introduction The “naı̈ve” Frobenius method The general Frobenius method

Finally, we have that the general solution is given by

y = c1 y1 + c2 y2 .

Remarks. Regarding the case r1 − r2 ∈ N0 :

When y1 has infinitely many nonzero coefficients, the general
recursion for bn will be more complicated.
If a closed form expression for the coefficients of y1 isn’t
available, the recursion relations for the an and bn still allow
us to compute as many terms as we need.
Similar computations and comments hold when r1 − r2 ∈ N,
except that one must also solve for k.
Because of the ln x factor, one can frequently conclude that
|y2 | → ∞ as x → 0+ , without explicitly computing the bn .
This will suffice for our applications.

Daileda Frobenius’ Method