MATH 219

Spring 2021-22
Lecture 15
Lecture notes by Özgür Kişisel

Content: The method of undetermined coefficients.

Suggested Problems: (Boyce, Di Prima, 9th edition)
§4.3: 2, 7, 8, 11, 12, 14, 17, 18, 21

1 Method of undetermined coefficients

Let us now consider a non-homogenous nth order linear equation with constant
coefficients, namely an equation of the form

y (n) + a1 y (n−1) + . . . + an y = b(t).

We can solve any such equation in principle, by using the techniques developed
so far: First find all solutions of the corresponding homogenous equation (with
b(t) = 0), then convert the ODE into a first order n × n system and apply variation
of parameters. This approach will be followed in the next lecture. However, there is
a much simpler and direct method if b(t) has a special form. More specifically, let
us assume for this lecture that b(t) is a linear combination of functions of type

tk eλt , tk eat cos bt, tk eat sin bt (1)

for various values of the nonnegative integer k and real numbers a, b, λ. The strategy
can be outlined as follows:

1. Find all solutions yh (t) of the homogenous equation y (n) +a1 y (n−1) +. . .+an y =
0.

2. Guess the form of a particular solution yp (t) of the full equation y (n) +a1 y (n−1) +
. . . + an y = b(t) as a linear combination of functions in (1). Then find the
constants appearing in yp (t) by placing this function into the ODE.

1
The description of this method is admittedly vague at this point. Let us work on a
few examples first. Afterwards, we will clarify how a correct guess can be made and
we will outline a proof of why this method is guaranteed to work.

Example 1.1 Find all solutions of the ODE

y 00 − y = 6e3t .

Solution: The corresponding homogenous equation is y 00 − y = 0. Its characteristic

equation is λ2 − 1 = 0, so λ = ±1. Since the roots are distinct, we see that yh =
c1 et + c2 e−t . For a particular solution, it is natural to make a guess of the form
yp = Ae3t , since the derivatives of e3t will again give us multiples of itself and we
will have a chance of balancing the left and right hand sides. Plugging yp into the
ODE gives

(Ae3t )00 − Ae3t = 6e3t

8Ae3t = 6e3t
3
and it is clear that A = 4
works. Therefore

3
y = yh + yp = c1 et + c2 e−t + e3t .
4

Example 1.2 Find all solutions of the ODE

y 00 + 3y 0 + 2y = cos t.

Solution: Again, we first solve the homogenous equation y 00 + 3y 0 + 2y = 0. The

characteristic equation is λ2 + 3λ + 2 = 0 and its roots are λ1 = −2, λ2 = −1. Hence
yh = c1 e−2t + c2 e−t . Let us now think about yp . A guess of the form yp = A cos t
will not be a good idea this time, since the derivatives of this function will also give
us sin t terms and it will be impossible to balance them with the term on the right.
However, we can set yp = A cos t + B sin t and use the extra degree of freedom to
balance things out:

(A cos t + B sin t)00 + 3(A cos t + B sin t)0 + 2(A cos t + B sin t) = cos t
−A cos t − B sin t − 3A sin t + 3B cos t + 2A cos t + 2B sin t = cos t
(A + 3B) cos t + (B − 3A) sin t = cos t.

2
Since the functions cos t, sin t are linearly independent, such an equation holds for
all t if and only if the equalities A + 3B = 1 and B − 3A = 0 simultaneously hold.
This gives us A = 1/10 and B = 3/10. So we get
1 3
y = yh + yp = c1 e−2t + c2 e−t + cos t + sin t.
10 10

Example 1.3 Find all solutions of the ODE

y 00 − y = et .

Solution: Just as in the first example of this lecture, we get yh = c1 et + c2 e−t . For
the particular solution, a guess of the form yp = Aet is not a good idea: Such a
term is already present in yh . This means that if we plug it into the left hand side,
it will produce 0 and can never produce a nonzero term like et . Therefore we need
something else. Inspired by the discussion about repeated roots in previous lectures,
we can try yp = Atet . Then,
(Atet )00 − Atet = et
Atet + 2Aet − Atet = et
2Aet = et
and we see that A = 21 works. The important thing here is that the tet terms on the
left cancel; if they didn’t cancel, it would be impossible to balance the two sides. We
obtain
1
y = yh + yp = c1 et + c2 e−t + tet .
2

2 Differential operators and annihilators

The examples above give us a glimpse of the method of undetermined coefficients.
But at this point it is not absolutely clear how we can make the correct guess and
we have no proof that this method will work. In order to close this gap, we will talk
about differential operators and annihilators, which are important concepts on their
own as well.
Let us develop a convenient notation which is especially useful for constant coefficient
linear ODE’s. Set
d
D= .
dt
3
D is called the differentiation operator. We can compose D with itself any number
of times and take linear combinations of operators of this type. Such operators are
called differential operators.

Example 2.1 By definition, it is clear that Dy = y 0 , D2 y = y 00 etc. Applications of

polynomial expressions in D to y also make sense. For instance:

(D3 + 4D − 7)y = y 000 + 4y 0 − 7y

Based on this new notation, our equation y (n) + a1 y (n−1) + . . . + an y = b(t) can be
rewritten as
(Dn + a1 Dn−1 + . . . + an )y = b(t).
If we set p(D) = Dn +a1 Dn−1 +. . .+an then notice that p(λ) = 0 is the characteristic
equation for the corresponding homogenous ODE. Say λ1 , . . . , λn are the roots of
the characteristic equation. Then the differential operator can be factorized and we
can write the ODE in the form

(D − λ1 )(D − λ2 ) . . . (D − λn )y = b(t)

2.1 Annihilators

Definition 2.1 Say L(D) = Dk + c1 Dk−1 + . . . + ck is a differential operator. Then

L(D) is called an annihilator of b(t) if

1. L(D)b(t) = 0,

2. L(D) has the lowest degree among all nonzero polynomial operators such that
the first condition is satisfied.

One can prove that the annihilator of a function is unique if it exists. On the other
hand, not all functions have polynomial annihilators. For instance, take f (t) = 1/t.
Then Df = −1/t2 , D2 f = 2/t3 , . . . , Dm f = (−1)m m!/tm+1 . . . All these functions
are linearly independent, so no nontrivial combination of them is zero. Therefore
f (t) = 1/t does not have a polynomial annihilator. Next, we will compute the
annihilators of some familiar functions.

4
Example 2.2 If f (t) = eλt then

(D − λ)eλt = λeλt − λeλt = 0.

Since D − λ is of first degree, it is necessarily the lowest order operator annihilating

eλt . Therefore the annihilator of eλt is D − λ. As a special case, note that the
annihilator of f (t) = 1 is D.

Example 2.3 Let us take f (t) = teλt . Then

(D − λ)teλt = eλt + λteλt − λteλt = eλt ,

therefore by the previous example

(D − λ)2 teλt = (D − λ)eλt = 0.

The operator (D − λ)2 is of order 2. It can be shown that no first order operator
annihilates teλt hence the annihilator of teλt is (D − λ)2 .
In a similar manner,

(D − λ)tk eλt = ktk−1 eλt + λtk eλt − λtk eλt = ktk−1 eλt .

Using this equality and by using induction one can show that the annihilator of tk eλt
is (D − λ)k+1 for any nonnegative integer k.

Example 2.4 Let f (t) = cos at. Then D(cos at) = −a sin at and D2 (cos at) =
−a2 cos at. It follows that
(D2 + a2 ) cos at = 0.
It can be shown that no first order differential operator annihilates cos at, therefore
the annihilator of cos at is D2 +a2 . A similar computation shows that the annihilator
of sin at is also D2 + a2 .

Notice that if L(D)f (t) = 0 and L(D)g(t) = 0 then L(D)(c1 f (t) + c2 g(t)) = 0
for any constants c1 , c2 . Furthermore, if L(D)f (t) = 0 and M (D)g(t) = 0 then
L(D)M (D)(c1 f (t) + c2 g(t)) = 0. Actually, one can be a little bit more economical:
If N (D) is the “least common multiple” of L(D) and M (D), then N (D)(c1 f (t) +
c2 g(t)) = 0.

5
Example 2.5 Find the annihilator of
√
f (t) = e−2t + t2 e−2t + cos( 3t) + t.
√
Solution: The annihilators of e−2t , t2 e−2t , cos( 3t) and t are D +2, (D +2)3 , D2 +3
and D2 respectively. The annihilator of f (t) is the least common multiple of all these
terms, which is:
(D + 2)3 (D2 + 3)D2 .

Example 2.6 The annihilator of e(a+ib)t is D−(a+ib) and the annihilator of e(a−ib)t
is D − (a − ib). Since
e(a+ib)t + e(a−ib)t
eat cos bt =
2
we see that (D − (a + ib))(D − (a − ib)) = D2 − 2aD + (a2 + b2 ) annihilates eat cos bt.
It can be shown that no first order operator annihilates this function, therefore we
found the actual annihilator. Likewise, eat sin bt has the same annihilator.
Multiplication by t has the same effect on such functions as for the case of real roots.
Therefore a similar analysis shows that the annihilators of tk eat cos bt and tk eat sin bt
are equal and given by
(D2 − 2a + (a2 + b2 ))k+1 .

2.2 Method of undetermined coefficients revisited

Let us return to our discussion of a non-homogenous ODE

L(D)y = b(t) (2)

where L(D) is a linear differential operator with constant coefficients. If b(t) has a
polynomial annihilator, then we can solve this equation as follows:

1. Find the annihilator M (D) of b(t).

2. Apply M (D) to both sides of (2) in order to get

M (D)L(D)y = 0 (3)

6
 3. Find all solutions of (3) (this equation is homogenous, so in principle we know
how to do this).
4. Not all solutions of (3) are solutions of (2). To find out which ones are, plug
the solution obtained in the previous step back in (2).

The final step will reveal the values of some of the coefficients of the linear combi-
nation obtained in step 3. For this reason, the method is named “the method of
undetermined coefficients”.

Example 2.7 Solve the ODE

y 00 − y = e2t

Solution: The equation can be written as (D2 − 1)y = e2t or

(D − 1)(D + 1)y = e2t .

The annihilator of the right hand side is D − 2. Apply this operator to both sides in
order to get
(D − 2)(D − 1)(D + 1)y = 0.
The three roots of the characteristic equation are 2, 1 and −1. They are distinct.
So, the solutions of this equation are

y = c1 et + c2 e−t + c3 e2t .

Now let us put this function back in the original equation:

(D2 − 1)(c1 et + c2 e−t + c3 e2t ) = e2t

(D2 − 1)(c3 e2t ) = e2t
4c3 e2t − c3 e2t = e2t
3c3 = 1
c3 = 1/3.

The fact that D2 − 1 annihilates et and e−t was used in order to write the second
equality. Therefore the solutions of the equation are
1
y = c1 et + c2 e−t + e2t
3
where c1 , c2 ∈ R.

7
Example 2.8 Solve the initial value problem

y 000 + 4y 0 = t, y(0) = y 0 (0) = 0, y 0 (0) = 1.

Solution: Rewrite the ODE as

(D3 + 4D)y = t.

The annihilator of the right hand side is D2 . Apply it to both sides. We get

D3 (D2 + 4)y = 0.

The roots of the characteristic equation are 0, 0, 0, 2i, −2i. Therefore all solutions of
this new ODE are

y = c1 cos 2t + c2 sin 2t + c3 + c4 t + c5 t2 .

Plug this function back in the original ODE:

(D3 + 4D)(c1 cos 2t + c2 sin 2t + c3 + c4 t + c5 t2 ) = t

(D3 + 4D)(c4 t + c5 t2 ) = t
4c4 + 8c5 t = t.

From the last equality we deduce that c4 = 0, c5 = 1/8. Therefore all solutions of the
ODE are
1
y = c1 cos 2t + c2 sin 2t + c3 + t2 .
8
We still need to use the initial values in order to find the remaining constants. Com-
1 1
pute y 0 = −2c1 sin 2t + 2c2 cos 2t + t and y 00 = −4c1 cos 2t − 4c2 sin 2t + . The ini-
4 4
tial conditions imply

c1 + c3 = 0
2c2 = 0
1
−4c1 + = 1.
4
We get c1 = −3/16, c2 = 0 and c3 = 3/16 as the unique solution of this linear
system. Therefore
3 3 1
y = − cos 2t + + t2 .
16 16 8

8
Example 2.9 Find the form of the solutions of

y 00 + 2y 0 + 2y = 3e−t + 2e−t cos t + 4e−t t2 sin t.

In other words, write the solution as a linear combination of several terms and point
out which coefficients will be determined (do not compute them).
Solution: Rewrite the equation as

(D2 + 2D + 2)y = 3e−t + 2e−t cos t + 4e−t t2 sin t.

The annihilators of e−t , e−t cos t and e−t t2 sin t are D + 1, D2 + 2D + 2 and (D2 +
2D + 2)3 respectively. Therefore the annihilator of the right hand side is their least
common multiple, which is

(D + 1)(D2 + 2D + 2)3 .

Apply this operator to both sides in order to get

(D + 1)(D2 + 2D + 2)4 y = 0.

The solutions of this 9th order equation are

y = c1 e−t cos t + c2 e−t sin t + c3 te−t cos t + c4 te−t sin t + c5 t2 e−t cos t
+ c6 t2 e−t sin t + c7 t3 e−t cos t + c8 t3 e−t sin t + c9 e−t .

When we plug this back in the original equation, the first two summands will be
annihilated by D2 + 2D + 2, hence c1 and c2 will be free. The other 7 constants
c3 , c4 , . . . , c9 are to be determined.