You are on page 1of 22

Power Series Solution

And
Frobenius Method
(Mathematical Physics Assignment)
(Submitted to Dr. Abhishek Kumar Singh)
Name: Rajneil Baruah
Class:B.Sc(H) Physics
Roll: 5015
Year: 2ndYear

Series Solution:
Consider a Linear, Second order, Homogeneous, Ordinary Differential
Equation(ODE), in the form of:
d
d2
y + P (x) y + Q(x)y = 0
2
dx
dx

(1)

The above eq n (1) has two linearly independent solutions. Using

Wronskian, it can be shown that for a second order, Linear, Homogeneous
ODE, NO third linearly independent solution exists.
Thus, the most general soln of eq n (1) is the linear combination of the two
independent soln of the eq n .
If y1 (x) and y2 (x) are the independent soln , then the general soln to
eq n (1) is given by:
y = c1 y1 (x) + c2 y2 (x)
(2)
Our aim in this section is to obtain the solutions to the ODEs as a power
series.

Power Series of a Function:

Any function f (x) can be expressed as the sum of powers of 0 x0 about any
point x = x0 Thus, the series expansion of f (x) about x = x0 is given by:-

f (x) =

fm xm = f0 (x x0 )0 + f1 (x x0 )2 + f2 (x x0 )3 .......... (3)

m=0

For x = x0 = 0

f (x) =

fm xm = f0 x0 + f1 x2 + f2 x3 ..........

m=0

The general expansion is given by:

f (x) =

fm (x x0 )m

m=0

However, eq n (3) holds good only if the the functionf (x) is analytic at
0
x = x0 . Thus, if f (x0 ) = undef ined(
, 0 , , etc.), then,
f (x) 6=

fm (x x0 )m

m=0

f (x) =

fm (x x1 )m

m=0

Thus, for the function f (x),

1. x0 is a singular point.
2. x1 is an ordinary point.
Thus from eq n (1), we haveP (x) and Q(x). If P (x)|x=x0 and Q(x)|x=x0 are
NOT exploding, then P (x) and Q(x) are called analytic functions.
If P (x) and Q(x) are analytic functions in eq n (1), then the soln for the
ODE in the eq n can be given by:
y(x) =

Cm x m

(4)

m=0

[considering for x0 = 0]
Differentiating both sides twice w.r.t. x,

y 0 (x) =

mCm xm1

(5)

(6)

m=1

y 00 (x) =

X
m=2

m(m 1)Cm x

m2

+ P (x)

mCm x

m1

+ Q(x)

m=1

m=2

C m xm = 0

m=0

Shifting the indexes of the first and second term, we have,

(m+1)(m+2)Cm+2 x +P (x)

m=0

(m+1)Cm+1 x +Q(x)

m=0

X
m=0

C m xm = 0

Since, P (x) and Q(x) are analytic about x = x0 = 0, thus, the above eq n
changes to:

(7)

m=0

Since xm 6= 0

(8)

Rearranging eq n (viii), we have-

Cm+2 =

q0 Cm p0 (m + 1)Cm+1
(m + 1)(m + 2)

(9)

For m = 0, we getC2 in terms of C0 and C1 . Again, for m = 1, we can obtain the value of C3 in terms of C1 and C2 . Since,C2 be expressed in terms
of C1 and C0 , therefore, C3 can also be expressed in terms of C1 and C0 .
Similarly, all other constants Cm can be expressed in therms of C0 and
C1 .
Thus,

y=

C m xm

m=0
2

y = C0 (1 + a1 x + a2 x + ........) + C1 (1 + b1 x + b2 x2 + ......)
NOTE:
a1 , a2 ...an and b1 , b2 ...bn are calculated by putting the values of m in
eq n (ix). But C0 and C1 are constants whose values depend upon the initial
condition(or boundary condition).
However, the solution given is based on the assumption that P (x) and
Q(x) are analytic at x = x0 = 0. When atx = 0, P (x) or Q(x) blow up,
they become singular.
Hence, a simple power series solution cannot be found.

Frobenius Method:
The Solution that we obtained in the previous section was strictly based
on the assumption that P (x) and Q(x) are analytic atx = x0 = 0.
Now, let us consider a differential equation of the form-

d2
c(x)
b(x) d
y+ 2 y=0
y+
2
dx
x dx
x
b(x)

Here, P (x) = x and Q(x) =

points are defined as-

c(x)
.
x2

(10)

lim xP (x) = f inite

and
lim x2 Q(x) = f inite

Since,P (x) and Q(x) are non-analytical aboutx = 0, thus, checking the
c(x)
above condition shows that b(x)
x and x2 are functions which make x = 0 a
regular singular point.
Thus, the expression of y that we obtained as a solution:
y=

Cm x m

m=0

is NOT a solution to the given form of equation (i).

Frobenius Method allows us to calculate a series solution around a regular
singular point. Eq n (10) [whereb(x) and c(x) are analytic about x = 0], has
at least one solution1 given by:

y(x) = xr

C m xm

(11)

m=0

y(x) = xr (C0 + C1 x + C2 x2 + C3 x3 ........)

(C0 6= 0)
Fuchs Theorem1
The theorem asserts that, we can always obtain at least one power series
solution, provided we are expanding about a point which an ordinary point
or at worst, a regular singular point.

Indicial Equation
For the Differential Equationx2 y + xb(x)y 0 + c(x)y = 0
expanding b(x) and c(x) as a power series:
b(x) = b0 + b1 x + b2 x2 + b3 x3 .........
c(x) = c0 + c1 x + c2 x2 + c3 x3 .........

(12)

From Fuchs Theorem, we have one power series solution at least. The
power series solution proposed by Frobenius is:

y=

Cm xm+r

m=0

Differentiating w.r.t. x:
0

y =

(m + r)Cm xm+r1

m=0

y 00 =

m=0

[(m + r 1)(m + r)Cm xm+r2 ]+

m=0

x(b0 + b1 x + b2 x ....)

[(m + r)Cm xm+r1 ]+

m=0

(c0 + c1 x + c2 x2 ....)

X
m=0

Cm xm+r = 0

(13)

xr [r(r 1)C0 + ....]+

(b0 + b1 x + b2 x2 ....)xr (rC0 + ....)+
(c0 + c1 x + c2 x2 ....)xr (C0 + C1 x....) = 0
Equation the coefficients of xr :
[r(r 1) + b0 r + c0 ]C0 = 0]
Since we assumed that C0 6= 0,
r(r 1) + b0 + c0 = 0

(14)

The above eq n (14) is called the indicial equation of the differential

equation.
Since eq n (v) is a quadratic equation, it has two solutions for the value of r.
Only one solution will be of the form as given in eq n (ii). Other solutions
will depend on the roots of the indicial equation.
There can be a total of three cases:

1. The roots are real and distinct roots, NOT differing by an integer.
2. The roots are equal (double root).
3. The roots are real and distinct, differing by an integer.
CASE: I
(When the roots are real and distinct and DO NOT differ by an integer.)
In such a case, both the solutions can be directly obtained on the basis of
the roots of the indicial equation. If r1 and r2 are the two roots of the
indicial equation (r1 6= r2 ), then the solutions are given by:

y1 (x) =

Cm xm+r1

m=0

y2 (x) =

X
m=0

Dm xm+r2

CASE: II
(When both the roots are real and equal)
The indicial equation is given by:
r(r 1) + b0 r + c0 = 0
there can be a double root if:
(b0 1)2 4c0 = 0
If r1 and r2 are the two roots such that r1 = r2 = r0 , then from Fuchs
Theorem, one solution is given by:
y1 (x) = xr0 [C0 + C1 x + C2 x2 + C3 X 3 ........]
Let the second solution y2 (x) be given by,
y2 (x) = u(x)y1 (x),
such that y2 (x) is a solution to the differential equation.
Differentiating y2 (x) w.r.t. x, we have:
y20 (x) = uy10 + uy10
And
y200 (x) = u00 y1 + 2uy10 + uy100
Since y2 (x) is a solution :
x2 y200 (x) + xb(x)y20 + c(x)y = 0
x2 (u00 y1 + 2uy10 + uy100 ) + xb(x)(uy10 + uy10 ) + c(x)(uy1 ) = 0
x2 y1 u00 + 2x2 y1 u0 + b(x)y1 u0 = 0
(Since, x2 y100 + xb(x)uy10 + c(x)uy1 = 0)
Dividing both sides by x2 y1 y0
u00 + [2 y1 + bx0 + ......]u0 = 0

[Inserting the power series of b(x)]

y10
xr1 [rC0 + (r + 1)C1 x + ........]
=
y1
xr [C0 + C1 x + ......]
=

1 rC0 + (r + 1)C1 x......

[
]
x
C0 + C1 x + ......
=

Thus,
2y 0
u00 + [ y11 +

b0
x

r
+ ..........
x

+ ......]u0 = 0

0
u00 + [ 2r+b
+ ......]u0 = 0
x

u00
u0

1
x

+ ......

By integration, we get

lnu0 = lnx + ......

u = lnx + k1 x + k2 x2 + .....
Finally, we know that:
y2 (x) = u(x)y1 (x)
y2 (x) = y1 (x)lnx + y1 (x)

K m xm

m=0

And since y1 (x) is analytical:

y2 (x) = y1 (x)ln|x| + x

X
m=0

Am xm

(15)

CASE: III
(When both roots are real and distinct and they differ by an integer)
The initial steps will be same as in CASE: II. Thus,
u00 + [

2r + b0
+ ......]u0 = 0
x

From the indicial equation:

r(r 1) + b0 r + c0 = 0
if r1 and r2 are the roots of this indicial equation, such that:
r1 = r0

r2 = r0 N

[where N is a positive integer]

b0 1 = (r1 r2 )
2r + b0 = N + 1
Thus,

N +1
u00
= [
+ ........]
0
u
x
Integrating both sides, we get,
lnu0 = (N + 1)lnx + C0 + C1 x + .....

u0 = x(N +1) e(C0 +C1 x+......)

u0 = xN1+1 +
Finally,

K1
xN

+ .... +

KN
x

+ KN +1 + KP +2 x + ......

u = KN ln|x| +

Ki xi

i=N

Again from y2 (x) = u(x)y1 (x),

y2 (x) = KN y1 (x)ln|x| + xr2

Bk xk

(16)

Formula for Finding the Second Solution:

Consider a Euler-Cauchy Differential Equation:
y 00 + P (x)y 0 + Q(x)y = 0
Let us assume that:
y=

(1)

Cm xm+r

(2)

m=0

C0 6= 0; |x| > 0
be a soluion to (1)
Assuming that x = x0 = 0 is a regular singular point; this means that
xP (x) and x2 Q(x) are analytic functions;
i.e.
xP (x) =

Pm xm

m=0

x2 Q(x) =

Qm xm

m=0

(m + r)Cm xm+r1

m=0

y 00 =

(m + r)(m + r 1)Cm xm+r2

m=0

Now,

X
1 X
m
P (x)y = [
Pm x
(m + r)Cm xm+r1 ]
x
0

m=0

P (x)y 0 =

xr1
x

m=0
m
X

(n + r)Cn Pmn )xm ]

m=0 n=0

Similarly,
Q(x)y = xr2 [

m1
X
X
(
Cn Qmn + Q0 Cm )xm ]
m=0 n=0

10

(3)

Substituting the values of y 00 , P (x)y 0 and Q(x)y in (1), we get the equation
given below (after cancelling the common factor xr2 ):

[((m + r)(m + r 1) + (m + r)P0 + Q0 )Cm

m=0

m1
X

(4)
Cn ((n + r)Pmn + Qmn )]xm = 0

n=0

Equating the coefficient of xm :

[(m + r)(m + r 1) + (m + r)P0 + Q0 ]Cm

X
+
Cn [(n + r)Pmn + Qm n] = 0

(5)

n=0

Thus,
C0 [r(r 1) + rP0 + Q0 ] = 0
Let,
F (r) = r(r 1) + rP0 + Q0
Thus,
C0 F (r) = 0

(6)

Since we have already assumed that C0 6= 0, we have:

F (r) = 0
Thus,
r(r 1) + rP0 + Q0 = 0
which is the indicial equation.
Similarly we can show that the coefficient of x1 is zero.
(m = 1 and n = 0)
C1 [r(r + 1) + (r + 1)P0 + Q0 ] + C0 [rP1 + Q1 ] = 0
C1 F (r + 1) + C0 (rP1 + Q1 ) = 0
Similarly, the equating the coefficient of x2 to zero, we get:
C2 [(r + 1)(r + 2) + (r + 2)P0 + Q0 ]
+ C0 [rP2 + Q2 ] + C1 [(r + 1)P1 + Q1 ] = 0
C2 F (r + 2) + C0 (rP2 + Q2 ) + C1 ((r + 1)P1 + Q1 ) = 0
11

(7)

Finally when we will equate the coefficient of xm , we will get:

Cm F (m + r) + C0 [rPm + Qm ] + ......
...... + Cm1 [(m + r 1)P1 + Q1 ] = 0

(8)

From (6),(7) and (8), we can see that, the expressions for C1 ,C2 ......Cm can
be calculated in terms of C0
CASE I
(When the roots are real and distinct and DO NOT differ by an integer)
If the roots of the indicial equations are r1 and r2 such thatr1 r2 6= N
(where N is any integer), then the solutions of the differential equation (1)
are given by:
y1 (x) =
y2 (x) =

X
m=0

Cm (1)xm+r1
Cm (2)xm+r2

(9)

m=0

CASE II
(When the roots are real and equal)
If r1 and r2 are two solutions of the indicial equation such that
r1 = r2 = r0 , then according to F uch0 s T heorem, we can obtain one
solution as the following:
y1 (x) = x

r0

Cm x m

(10)

m=0

Since both roots are equal here(r0 ), we cannot obtain the second solution
directly as the first one.
From (4), we have,

m=0
m1
X

n=0

12

[r(r 1) + rP0 + Q0 ]C0 xr

X
[((m + r)(m + r 1) + (m + r)P0 + Q0 )Cm
m=1
m1
X

n=0

C0 F (r)x +

m=1
m1
X

Cn ((n + r)Pmn + Qmn )]xm+r = L(y(x, r))

n=0

In all the above equations L(y(x, r)) represents the differential equation
(1).
Since, the coefficients of xm+r be zero for m 1:
L(y(x, r)) = C0 F (r)xr

(11)

Since the roots of the equation are known(r0 ),

F (r) = (r r0 )2
Thus, equation (11) becomes L(y(x, r)) = C0 (r r0 )2 xr

(12)

We know,
L(y(x, r)) = y 00 + P (x)y 0 + Q(x)
= D2 y + P (x)Dy + Q(x)y

[D

L(y(x, r)) = [D2 + P (x) + Q(x)]y

From equations (12) and (13), we have:
[D2 + P (x) + Q(x)]y = C0 (r r0 )2 xr

13

d
]
dx
(13)

[D2 + P (x) + Q(x)]y =

C0 (r r0 )2 xr
r
r
Since, differential operators are commutative, the above equation can also
be written as:
[D2 + P (x) + Q(x)]

r

(14)

Thus,

[D2 + P (x) + Q(x)] y|r=r0 = 0

r

From the above equation, we can see that r y|r=r0 is also a solution to
differential equation (1).
We have,
y1 =

Cm xm+r

m=0

y1
|r=r0
r

X
y2 =
[
Cm xm+r ]
r
y2 =

(15)

m=0

Cm =

[C0 (rpm + Qm ) + ......Cm1 ((m + r 1)P1 + Q1 )]

F (m + r)

(16)

We know,
F (r) = (r r0 )2
F (m + r) = (m + r r0 )2
Putting the values of F (r) and F (m + r) in equations (16) and further in
(15):

14

[
y2 =
r

= [x ln|x|

n=0

Cn [(n + r)Pmn + Qmn ]xm xr

]r=r0
(m + r r0 )2
Cm xm ]r=r0

m=0
m1
X Cn ((n + r)Pmn + Qmn )
[x [ [
]]]r=r0 xm
x n=0
(m + r r0 )2
r

= y1 ln|x| + [xr
[Since, [xr [

Bk xk ]r = r0

k=0
m1
X

Cn ((n + r)Pmn + Qmn )

[
]]]r=r0 xm is analytical]
2
x n=0
(m + r r0 )

Finally we get,
y1 (x) =

Cm xm+r0

m=0

y2 (x) =

y
|r=r0
r

y2 (x) = y1 (x)ln|x| + x

r0

Ak xk

k=0

CASE III
(When the roots are real and distinct and they differ by an integer)
Let r1 = r0 and r2 = r0 N be the roots of the indicial equation
where (r1 > r2 ) and N > 0(an integer).
From F uch0 s T heorem, we know that we can obtain at least one of
the two solutions in terms of power series directly. But, to know
which root will give us the first solution, we must first analyze the
indicial equation and the recurrence relation.
From equation (6), we haveF (r) = 0

15

Since r1 and r2 are the solutions of the above indicial equation, we

have:
(r r1 )(r r2 ) = F (r)
F (r) = (r r0 )(r r0 + N )
Thus,
F (m + r) = (m + r r0 )(m + r r0 + N )

(17)

Cm =

F (m + r)

equationCm =

[C0 (rpm + Qm ) + ......Cm1 ((m + r 1)P1 + Q1 )]

(m + r r0 )(m + r r0 + N )

Looking at the denominator:

(m + r r0 )(m + r r0 + N ) = F (m + r)
Case A:
For r = r0 ,
F (m + r) = m(m + N ) which is always greater than 1
or
F (m + r)|r=r0 6= 0
Case B:
For r = r0 N
F (m + r) = m(m N )
The above term is non zero only for m 6= N .
For m = N ,
F(m+r)=0
Thus,
Cm |r=r0 N = undef ined
16

(18)

1. Since m and N are integers, (m N ) will always give an integer(including zero for m = N ). Therefore, although we obtain two distinct
roots, the integer difference between them makes a difference in the solution
and we CANNOT directly obtain the two linearly independent solutions.
2.For r = r0 [the larger root],F (m + r) always had a non zero value,
irrespective of the value of m. Thus, we obtain a finite value of Cm . However,
for r = r0 N [smaller root], F (m + r) = 0 for m = N . Thus, for m = N ,
the value of Cm blows up. Hence, we always obtain the first solution from
the larger root and NOT the smaller one.
Thus, the first solution is given by:

y1 (x) =

Cm xm+r0

(19)

m=0

y(x) =

X
m=0

[C0 (rpm + Qm ) + ......Cm1 ((m + r 1)P1 + Q1 )]

(m + r r0 )(m + r r0 + N )

Similar to equation (11) of the previous case, we have:

L(y(x, r)) = C0 F (r)xr
L(y(x, r)) = C0 (r r0 )(r r0 + N )
We know that,
L(y(x, r)) = [D2 + P (x)D + Q(x)]y
[D2 + P (x)D + Q(x)]y = C0 (r r0 )(r r0 + N )2
[D2 + P (x)D + Q(x)](r r0 + N )y = C0 (r r0 )(r r0 + N )2

(20)

[D2 + P (x)D + Q(x)](r r0 + N )y

r
= C0 [xr (r r0 )(r r0 + N 2 )ln|x|
+ xr [2(r r0 )(r r0 + N ) + (r r0 + N 2 ]]
(21)

17

(r r0 + N )y
r
= C0 [xr (r r0 )(r r0 + N 2 )ln|x|

[D2 + P (x)D + Q(x)]

+ xr [2(r r0 )(r r0 + N ) + (r r0 + N 2 ]]
The RHS of the previous equatin becomes zero for r = r0 N , Thus,
[D2 + P (x)D + Q(x)]
Hence it shows that
equation L(y(x,r)).

r (r

(r r0 + N )y|r=r0 N = 0
r

r0 + N )y|r=r0 N is a solution of the differential

Now we have,
y(x) = xr

Cm x m

m=0

(r r2 )y(x) = (r r2 )xr

Cm x m

m=0

We know that, Cm = Cm (r)(Cm is a function of r).

Thus,

X
y(x) = xr
Cm x m
m=0

(r r0 + N )y(x) = xr

(r r0 + N )Cm (r)xm

m=0

Let,
(r r0 + N )Cm (r) = Km (r)
(r r0 + N )y(x) = xr

Km (r)xm

m=0

X
X

0
(r r0 + N )y(x) = xr ln|x|
Km (r)xm + xr
Km
(r)xm
r
m=0

18

m=0

Now, for r = r0 N ,

X
X

r0 N
m
r0 N
0

(rr0 +N )y(x) = x
ln|x|
Km (r0 N )x +x
Km
(r0 N )xm
r
m=0

m=0

From equation (20), it can be seen that (r r0 + N )y|r=r0 +N is also a

solution to the differential equation L(y(x, r)).
It can be found that:
xr0 N

Km (r0 N )xm = K

m=0

Cm xm

m=0

The above equation shows that both the solutions are linearly dependent.
If y1 (x) = xr0

m=0 Cm x

m,

then,
r0 N

X
k=0

the

19

Ak xk

**Proof of Linear Dependency of two solutions:

While solving a differential equations for power series solution, in the case
of integer difference of roots, we actually obtain three solutions. But
among the three, two are linearly dependent.
The three solutions are:

y1 (x) = x

Cm x m

m=0

y2 (x) = (r r0 + N )x

Cm x m

m=0

r
y3 (x) =
(r r0 + N )x
Cm x m
r
m=0

Among these three, y1 (x) and y2 (x) are linearly dependent on each other.

PROOF:
Let 1 , 2 .....N ..... be the constants of the summations when shown in
terms of C0 . Thus,
y(x) = C0 [xr +

N xr0 +N
1 xr+1
......
+......]
(r + 1 r)(r + 1 r + N )
(r + N r0 )(r + N r0 + N )

Put r = r0
y1 (x) = C0 [xr0 +

(2 xr0 +2 )
N xr0 +N
1 xr0 +1
+
+ ...... +
+ ......]
1(1 + N )
2(2 + N )
N (2N )

Similarly,
y2 (x) = C0 [xr +

1 (r r0 + N )xr+1
N (r r0 + N )xr
+......
+......]
(r + 1 r0 )(r + 1 r0 + N )
(r r0 + N )(r r0 + 2N )

Putting r = r0 + N :
y2 (x) = C0

20

N xr0
N

Thus, finally we have:

y1 (x) = C0 [xr0 +

1 xr0 +1
(2 xr0 +2 )
N xr0 +N
+
+ ...... +
+ ......]
1(1 + N )
2(2 + N )
N (2N )

N xr0
N
To show their Linear dependency we must find their Wronskian.
Wronskian of y1 (x) and y2 (x) is defined as:
y2 (x) = C0

(r0 + 1)1 xr0

(r0 + N )N xr0 +N 1
+ ....... +
+ ......]
1(1 + N )
N (2N )
y20 =

C0 r0 N xr0 +1
N

Thus,

W (y1 , y1 ) = y1 y20 y10 y2

W =

22 x2r0
C0 2 N 1 x2r0
[
+
+ .......]
N
1(N + 1) 1(1 + N )

Thus we can see that, W (y1 , y2 ) is zero for x = 0.

From the Theorem of Wronskian we know that:
If the Wronskian of the solution of a second order differential equation is
zero for any point in the domain, then, it is zero for the whole domain of
the solutions
Since the Wronskian of y1 and y2 ,W (y1 , y2 ) is zero, thus, it is proved that,
y1 and y2 are linearly dependent.
Hence,
x

r0 N

Km (r0 N )x

m=0

=K

X
m=0

holds good.

21

Cm x m