Solving Homogeneous Second Order Linear ODEs with Constant

Coefficients
R. C. Daileda

A homogeneous second order linear differential equation with constant coefficients has the form

ay 00 + by 0 + cy = 0 (1)

where a, b, c are real constants and a 6= 0. To solve this differential equation, we begin by introducing the
characteristic equation
ar2 + br + c = 0 (2)
and find that we can use the solutions of this simple algebraic equation to produce a pair of independent
solutions y1 and y2 to (1). We are then able to use these two so-called fundamental solutions to produce
every solution to (1) by appealing to the following result.

Theorem 1. If y1 and y2 are linearly independent solutions to (1) then the general solution to (1) is given
by
y = c1 y1 + c2 y2 ,
where c1 and c2 are arbitrary real numbers.
While this provides us with a very convenient and efficient means of finding the general solution to
a differential equation of the form (1), it is also somewhat unsatisfactory because no attempt to justify
Theorem 1 is made. There’s a very good reason for this. While it isn’t hard to check that the functions
y described in Theorem 1 are indeed solutions to (1), checking that every solution actually has this form
is much more difficult. This should be compared with the procedures that we have for solving first order
separable or linear equations. In these cases the completeness of our solutions is immediate, since they are
obtained from a step by step algebraic manipulation of the original differential equation.
The goal of this note is to bridge this gap in our knowledge by proving Theorem 1. This we will do by
establishing the somewhat weaker result stated below.

Theorem 2. If the characteristic equation (2) has two distinct real roots r1 and r2 then the general solution
to (1) is given by
y = c1 er1 x + c2 er2 x (3)
where c1 and c2 are arbitrary constants. If the characteristic equation (2) has only a single (repeated) real
root r then the general solution to (1) is given by

y = (c1 x + c2 )erx (4)

where c1 and c2 are arbitrary constants. And if the characteristic equation (2) has nonreal (complex) roots
α ± βi then the general solution to (1) is given by

y = eαx (c1 cos βx + c2 sin βx) (5)

where c1 and c2 are arbitrary constants.

Note that for our purposes Theorem 2 suffices to provide us with the general solution to (1). As such,
anyone who wishes may skip the derivation of Theorem 1 from Theorem 2.
Proof of Theorem 2. It is easy (but perhaps a bit tedious) to check that the functions given in (3) to (5) do,
indeed, yield solutions to (1) in each case. The question is: are these the only solutions? We show in every
case that they are.
We can deal with the first two cases simultaneously. Suppose that y is a solution to (1) and that r1 is a
real root of (2). Define a new function u by

u = ay 0 + (ar1 + b)y. (6)

Since y and y 0 are differentiable so is u, and in fact

u0 = ay 00 + (ar1 + b)y 0 .

Therefore

u0 − r1 u = ay 00 + (ar1 + b)y 0 − r1 ay 0 − r1 (ar1 + b)y

= ay 00 + by 0 − (ar12 + br1 )y.

However, we know that ar12 + br1 + c = 0 so that ar12 + br1 = −c. That is

u0 − r1 u = ay 00 + by 0 − (ar12 + br1 )y = ay 00 + by 0 + cy = 0.

Therefore, u is a solution to the first order linear differential equation u0 − r1 u = 0! Solving this using our
earlier techniques we find that we must have u = C1 er1 x for some constant C1 . If we put this back into (6)
we find that
ay 0 + (ar1 + b)y = C1 er1 x
which is another first order linear equation, this time in y! Dividing by a (which we know to be nonzero) it
becomes  
0 b
y + r1 + y = C 2 e r1 x (7)
a
where C2 = C1 /a.
If r2 is the other root of ar2 + br + c = 0 (which may, in fact, equal r1 ), then the polynomial ax2 + bx + c
must factor as a(x − r1 )(x − r2 ). If we multiply out the latter polynomial and compare its x coefficient with
the first, we find that b = −a(r1 + r2 ), or −r2 = r1 + b/a.1 So we can rewrite (7) as

y 0 − r2 y = C2 er1 x .

Using the integrating factor e−r2 x and solving we find that y must be given by
Z
y = C2 er2 x e(r1 −r2 )x dx.

There are now two cases. If r1 6= r2 (i.e. the characteristic equation has two distinct real solutions) then
r1 − r2 6= 0 and so we have
e(r1 −r2 )x
Z
e(r1 −r2 )x dx = + C3 (8)
r1 − r2
which means that
e(r1 −r2 )x
 
y = C 2 e r2 x + C3 = C4 er1 x + C3 er2 x ,
r1 − r2
where we have set C4 = C2 /(r1 − r2 ). Up to the names of the constants (which are arbitrary anyway), this
is exactly what we needed! But what if r1 = r2 so that (8) isn’t valid? In this case we have
Z Z
y = C2 er2 x e(r1 −r2 )x dx = C2 er1 x dx = C2 er1 x (x + C3 ) = (C2 x + C4 )er1 x ,

1 This can also be seen directly by appealing to the quadratic formula, which expresses r1 and r2 in terms of a, b and c.
with C4 = C2 C3 , which gives us what we expected in this case, too.
Now let’s move on to the third case, in which the characteristic equation (2) has nonreal roots α±βi. Since
we’re after real-valued solutions in a situation which more naturally calls for complex numbers, things are a bit
more complicated. The first thing we need is a relationship between α, β and the coefficients of (2). As above,
knowing the roots of a polynomial allows us to factor it, so that ax2 + bx + c = a(x − (α + βi))(x − (α − βi)) =
a(x2 − 2αx + (α2 + β 2 )). Comparing coefficients in these expressions tells us that we must have
b = −2aα (9)
c = a(α2 + β 2 ). (10)
We’ll need these relationships shortly.
Now assume that y is a solution to (1). We’re going to perform a series of changes of variables to get (1)
into a friendlier form. We first set
u = e−αx y. (11)
Differentiating twice yields
u00 = e−αx y 00 − 2αy 0 + α2 y

so that
au00 + β 2 au = e−αx ay 00 − 2aαy 0 + a(α2 + β 2 )y

= e−αx (ay 00 + by 0 + cy)

= 0
where we have used (9), (10) and the fact that y solves (1). If we divide both sides of this equation by a,
which we know to be nonzero, we obtain the simple linear equation
u00 + β 2 u = 0. (12)
At this point it’s worth remembering that our independent variable has been assumed to be x. We now
make the substitution t = βx. According to the chain rule
du du dt du
u0 = = =β
dx dt dx dt 
d2 u d2 u dt d2 u

d du d du
u00 = = = β =β 2 = β2 2 .
dx dx dx dx dt dt dx dt
Therefore (12) becomes
d2 u
β2 + β 2 u = 0.
dt2
Since we know α + βi is definitely not real it must be that β 6= 0. We can thus divide both sides of the
equation above by β 2 which tells us that
d2 u
+ u = 0. (13)
dt2
To solve (13) for u we perform one final substitution, letting
du
w= + (tan t)u. (14)
dt
Because tan t is only defined on intervals of the form In = (nπ − π/2, nπ + π/2), where n is an integer, w is
only defined on these intervals. So, from this point on let’s assume that our t domain is a single In . We find
that
dw d2 u du du
− (tan t)w = + (sec2 t)u + tan t − tan t − (tan2 t)u
dt dt2 dt dt
d2 u
= + (sec2 t − tan2 t)u
dt2
d2 u
= +u
dt2
= 0.
That is, w satisfies the linear equation dw/dt − (tan t)w = 0, which is solved easily using the integrating
factor cos t. In fact
w = C1 sec t (15)
for some constant C1 . Referring back to (14), this means that
du
+ (tan t)u = C1 sec t,
dt
which is once again linear. The integrating factor sec t allows us to finally obtain

u = C1 sin t + C2 cos t

for some constant C2 .

At this point the back substitutions t = βx and y = eαx u tell us immediately that

y = eαx (C2 cos βx + C1 sin βx) , (16)

and we’re finished. Well, not quite. There’s still one technical point we need to address. When we defined
w in the previous paragraph we had to assume that its independent variable t was restricted to the interval
In . This means that the expression for y in (16) is only valid on these intervals, and that in principle the
constants C1 and C2 might vary as we vary n. However, u and du/dt are continuous (since they are both
differentiable), and we can take limits at the endpoints of each interval to find that the constants match
everywhere. For example, if we have

u = A1 cos t + B1 sin t for t ∈ I0

u = A2 cos t + B2 sin t for t ∈ I1

then π
B1 = lim − (A1 cos t + B1 sin t) = u = lim (A2 cos t + B2 sin t) = B2
t→π/2 2 t→π/2+

and a similar computation with du/dt gives A1 = A2 . The general case is left to the reader. This completes
the proof of the third case.
The method of proof we’ve just employed actually still works (in principle) if we modify (1) so that it is
no longer homogeneous. That is, if we replace (1) with

ay 00 + by 0 + cy = G(x)

where G(x) is nonzero, the substitution u = ay 0 + (ar1 + b)y of the proof (in the case of real roots) yields
the first order differential equation
u0 − r1 u = G(x)
for u, which is simply the inhomogeneous version of u0 − r1 u = 0. This equation is still linear, and if we
solve it we can back substitute into u = ay 0 + (ar1 + b)y and solve for y. The reader is encouraged to give
this a try in the cases where G(x) is a constant or an exponential function.
Proof of Theorem 1. According to Theorem 2, the solutions to (1) are given by y = c1 u1 + c2 u2 for some
pair of independent functions u1 , u2 . For example, in the case that the characteristic equation has two real
roots r1 and r2 we have u1 = er1 x and u2 = er2 x . Let y1 and y2 be any other pair of independent solutions.
Then there are constants a, b, c and d so that y1 = au1 + bu2 and y2 = cu1 + du2 . Because y1 and y2 are
linearly independent, we can invert this relationship to obtain u1 = Ay1 + By2 and u2 = Cy1 + Dy2 for some
A, B, C and D. Thus, if y = c1 u1 + c2 u2 is any solution to (1), then

y = c1 (Ay1 + By2 ) + c2 (Cy1 + Dy2 ) = (c1 A + c2 C)y1 + (c1 B + c2 D)y2 = d1 y1 + d2 y2

for some constants d1 and d2 . Since y was an arbitrary solution to (1), this shows that every solution has
the form stated in Theorem 1.