An ODE
  | 
(1) | 
 
has singularities for finite 
 under the following conditions: (a) If either 
 or 
 diverges as 
, but 
 and 
 remain finite as 
, then 
 is called a regular or nonessential
singular point.  (b) If 
 diverges faster than 
 so that 
 as 
, or
 diverges faster than 
 so that 
 as 
, then 
 is called an
irregular or essential singularity.
Singularities of equation (1) at infinity are investigated by making the substitution 
, so 
, giving
  | 
(2) | 
 
Then (1) becomes
![\begin{displaymath}
z^4 {d^2y\over dz^2} + [2z^3-z^2P(z)]{dy\over dz} + Q(z)y = 0.
\end{displaymath}](o_728.gif)  | 
(4) | 
 
Case (a): If 
remain finite at 
 (
), then the point is ordinary.  Case (b): If either 
 diverges no more
rapidly than 
 or 
 diverges no more rapidly than 
, then the point is a regular singular point.  
Case (c): Otherwise, the point is an irregular singular point.
Morse and Feshbach (1953, pp. 667-674) give the canonical forms and solutions for second-order ODEs classified by
types of singular points.
For special classes of second-order linear ordinary differential equations, variable Coefficients can
be transformed into constant Coefficients. Given a second-order linear ODE with variable
Coefficients
  | 
(7) | 
 
Define a function 
,
  | 
(8) | 
 
  | 
(9) | 
 
![\begin{displaymath}
\left({dz\over dx}\right)^2 {d^2y\over dz^2} + \left[{{d^2z\over dx^2} + p(x){dz\over dx}}\right]{dy\over dz} + q(x)y = 0
\end{displaymath}](o_743.gif)  | 
(10) | 
 
![\begin{displaymath}
{d^2y\over dz^2} + \left[{{d^2z\over dx^2} + p(x){dz\over dx...
...^2}\right]y \equiv {d^2y\over dz^2} + A {dy\over dz} + By = 0.
\end{displaymath}](o_744.gif)  | 
(11) | 
 
This will have constant Coefficients if 
 and 
 are not functions of 
.  But we are free to set
 to an arbitrary Positive constant for 
 by defining 
 as
![\begin{displaymath}
z \equiv B^{-1/2}\int [q(x)]^{1/2}\,dx.
\end{displaymath}](o_746.gif)  | 
(12) | 
 
Then
![\begin{displaymath}
{dz\over dx} = B^{-1/2}[q(x)]^{1/2}
\end{displaymath}](o_747.gif)  | 
(13) | 
 
![\begin{displaymath}
{d^2z\over dx^2} = {\textstyle{1\over 2}}B^{-1/2}[q(x)]^{-1/2}q'(x),
\end{displaymath}](o_748.gif)  | 
(14) | 
 
and
Equation (11) therefore becomes
![\begin{displaymath}
{d^2y\over dz^2} + {q'(x)+2p(x)q(x)\over 2[q(x)]^{3/2}} B^{1/2} {dy\over dz} + By = 0,
\end{displaymath}](o_751.gif)  | 
(16) | 
 
which has constant Coefficients provided that
![\begin{displaymath}
A \equiv {q'(x)+2p(x)q(x)\over 2[q(x)]^{3/2}} B^{1/2} = {\rm [constant]}.
\end{displaymath}](o_752.gif)  | 
(17) | 
 
Eliminating constants, this gives
![\begin{displaymath}
A'\equiv {q'(x)+2p(x)q(x)\over [q(x)]^{3/2}} = {\rm [constant]}.
\end{displaymath}](o_753.gif)  | 
(18) | 
 
So for an ordinary differential equation in which 
 is a constant, the solution is given by solving the second-order
linear ODE with constant Coefficients
  | 
(19) | 
 
for 
, where 
 is defined as above.
A linear second-order homogeneous differential equation of the general form
  | 
(20) | 
 
can be transformed into standard form 
  | 
(21) | 
 
with the first-order term eliminated using the substitution
  | 
(22) | 
 
Then
  | 
(23) | 
 
  | 
(24) | 
 
  | 
(25) | 
 
so
Therefore,
![\begin{displaymath}
z''+[Q(x)-{\textstyle{1\over 2}}P'(x)-{\textstyle{1\over 4}}P^2(x)]z \equiv z''(x)+q(x)z = 0,
\end{displaymath}](o_765.gif)  | 
(28) | 
 
where
  | 
(29) | 
 
If 
, then the differential equation becomes
  | 
(30) | 
 
which can be solved by multiplying by
![\begin{displaymath}
\mathop{\rm exp}\nolimits \left[{\int^x P(x')\,dx'}\right]
\end{displaymath}](o_768.gif)  | 
(31) | 
 
to obtain
![\begin{displaymath}
0 = {d\over dx}\left\{{\mathop{\rm exp}\nolimits \left[{\int^x P(x')\,dx'}\right]{dy\over dx}}\right\}
\end{displaymath}](o_769.gif)  | 
(32) | 
 
![\begin{displaymath}
c_1 = \mathop{\rm exp}\nolimits \left[{\int^x P(x')\,dx'}\right]{dy\over dx}
\end{displaymath}](o_770.gif)  | 
(33) | 
 
![\begin{displaymath}
y = c_1\int^x {dx\over \mathop{\rm exp}\nolimits \left[{\int^x P(x')\,dx'}\right]}+c_2.
\end{displaymath}](o_771.gif)  | 
(34) | 
 
If one solution (
) to a second-order ODE is known, the other (
) may be found using the Reduction of
Order method.  From the Abel's Identity
  | 
(35) | 
 
where
  | 
(36) | 
 
  | 
(37) | 
 
![\begin{displaymath}
\ln\left[{W(x)\over W(a)}\right]= \int^x_a P(x')\,dx'
\end{displaymath}](o_777.gif)  | 
(38) | 
 
![\begin{displaymath}
W(x) = W(a)\mathop{\rm exp}\nolimits \left[{- \int^x_a P(x')\,dx'}\right].
\end{displaymath}](o_778.gif)  | 
(39) | 
 
But
  | 
(40) | 
 
Combining (39) and (40) yields
![\begin{displaymath}
{d\over dx}\left({y_2\over y1}\right)= W(a) {\mathop{\rm exp}\nolimits [- \int^x_a P(x')\,dx']\over y_1^2}
\end{displaymath}](o_780.gif)  | 
(41) | 
 
![\begin{displaymath}
y_2(x) = y_1(x)W(a)\int_b^x {\mathop{\rm exp}\nolimits [- \int^{x'}_a P(x'')\,dx'']\over [y_1(x')]^2}\,dx'.
\end{displaymath}](o_781.gif)  | 
(42) | 
 
Disregarding 
, since it is simply a multiplicative constant, and the constants 
 and 
, which will contribute a
solution which is not linearly independent of 
,
![\begin{displaymath}
y_2(x) = y_1(x)\int^x {\mathop{\rm exp}\nolimits \left[{- \int^{x'} P(x'')\,dx''}\right]\over [y_1(x')]^2}\,dx'.
\end{displaymath}](o_783.gif)  | 
(43) | 
 
If 
, this simplifies to
![\begin{displaymath}
y_2(x) = y_1(x) \int^x{dx'\over [y_1(x')]^2}.
\end{displaymath}](o_785.gif)  | 
(44) | 
 
For a nonhomogeneous second-order ODE in which the 
 term does not appear in the function 
,
  | 
(45) | 
 
let 
, then
  | 
(46) | 
 
So the first-order ODE
  | 
(47) | 
 
if linear, can be solved for 
 as a linear first-order ODE.  Once the solution is known,
  | 
(48) | 
 
  | 
(49) | 
 
On the other hand, if 
 is missing from 
,
  | 
(50) | 
 
let 
, then 
, and the equation reduces to
  | 
(51) | 
 
which, if linear, can be solved for 
 as a linear first-order ODE.  Once the solution is known,
  | 
(52) | 
 
See also Abel's Identity, Adjoint Operator
References
Arfken, G.  ``A Second Solution.''  §8.6 in Mathematical Methods for Physicists, 3rd ed.
  Orlando, FL: Academic Press, pp. 467-480, 1985.
Boyce, W. E. and DiPrima, R. C.  Elementary Differential Equations and Boundary Value Problems, 4th ed.
  New York: Wiley, 1986.
Morse, P. M. and Feshbach, H.  Methods of Theoretical Physics, Part I.  New York: McGraw-Hill, pp. 667-674, 1953.
© 1996-9 Eric W. Weisstein 
1999-05-26