Chapter 5: Dierential equations

September 11, 2019

Motivation
Solving dierential equations, i.e. equations involving functions of
independent variable(s) and their derivatives, is required for
studying the changes over time in economic variables, notably in
macroeconomics.

We are interested in functions of time, and the models involve their

derivatives. Dierential equations link the values of the function
x (t ) to time t and its derivatives x 0 (t ) = dx (t )/dt ,
x (2) (t ) = d 2 x (t )/dt 2 , etc. Those take the form:
f (x (t ), x 0 (t ), . . . , t ) = 0
The solution we seek are some functions of time x ∗ (t ).
Terminology
An ordinary dierential equation has a single independent
variable (time here). The opposite case is called partial dierential
equation.

An ordinary dierential equation of order m involves the derivatives

of the unknown function x (t ) up to the mth order (denoted x (k ) ):
f (t , x , x 0 , x (2) , ..., x (m) ) = 0
A dierential equation of order m which may be written
x (m) = g (t , x , x 0 , x (2) , ..., x (m−1) ) is said normal.

Equations of the form f (x , x 0 , x (2) , ..., x (m) ) = 0 , in which t does

not appear as a distinct argument, are called autonomous.
First-order ordinary dierential equations
First-order linear equations
Separable equations
Changes of variables
Qualitative theory

Second-order ordinary dierential equations

Second-order linear equations
Constant coecients
Stability for linear equations
First-order ordinary dierential equations
A rst-order equation writes:
x 0 = F (t , x ) (1)

where F is a given function of 2 variables and x = x (t ) is the

unknown function.

Example 1: Natural growth: x 0 = ax ; growth towards a limit b :

x0 + ax = b; logistic growth: x 0 + ax = bx 2 .
A solution in an interval I ⊂ R is a dierentiable function ϕ
dened on I such that x = ϕ(t ) satises (1).

An initial value problem is given by (1) and an initial condition

x (t = 0) = x0 . Such problems are commonly encountered in
economics, e.g. capital accumulation in growth theory.
First-order linear equations
This corresponds to equations of the form

x 0 + a (t ) x = b (t ) , (2)

where a (t ) and b (t ) are continuous functions of t on a given

interval I .

x 0 + x = t ; x 0 + 3tx = 5t ; t 2 + 2 x 0 + e t x = t ln t .


Example 2:
First-order linear di. equations with constant coecients
Consider rst the simplest case of constant coecients where
a (t ) = a 6= 0 and b (t ) = b:
x 0 + ax = b
Multiply each member by the integrating factor e at to get:

e at x 0 + ax = be at

This can be rewritten as:

d (xe at )
= be at
dt
Integrate:

d (xe at ) b
Z Z
= be at ⇔ xe at = e at + C
dt a
Finally multiply by e −at to get:

b
x= + Ce −at
a
Example 3: Consider the equation

x 0 + 4x = 8
Then

x = Ce −4t + = Ce −4t + 2
8

When t → ∞,
x (t ) → 2
As a = 4 > 0, the equilibrium state, x = 2, is stable.
First-order linear di. eq. with variable right-hand term
Now consider dierential equations with a variable right-hand
term:
x 0 + ax = b (t )
Multiply by the integrating factor e at to get

d (xe at )
e at x 0 + ax = b (t ) e at ⇔ = b (t ) e at


dt
Integrate Z
xe at = b(τ )e aτ d τ + C

Multiplying by e −at yields:

Z 
x= b (τ ) e aτ d τ e −at + Ce −at (3)
First-order linear di. equations: general case
We now turn to the general case in which both a (t ) and b (t )
vary with t:
x 0 + a (t ) x = b (t )
The integrating factor is now e A(t ) where

Z t
A (t ) = a (τ ) d τ, with A0 (t ) = a (t )
t0
Multiply by the integrating factor to obtain

x 0 + a (t ) x e A(t ) = b (t ) e A(t )

The left-hand side is

d xe A(t )


= x 0 + a (t ) x e A(t )


dt
Integrate: Z
xe A(t ) = b (τ ) e A(τ ) d τ + C

Multiply by e −A(t ) to obtain the following result.

Theorem (1)
The rst-order linear dierential equation:
x 0 + a (t ) x = b (t ) (4)

has for solution:

 Z 
x =e a(τ )d τ C + b (τ ) e a (τ )d τ
dτ
R R
−
(5)
Example 4: Consider the following dierential equation, with t>0,
α ϕ (t )
ϕ0 (t ) + =0
1−α t
Using the result above:

R dτ  Z 
ϕ (t ) = e
α
− 1−α τ C+ 0×e
A(τ ) d τ

[ln τ ]tt
= Ce − 1−α (ln t −ln t0 )
α α
= Ce
− 1−α
0
α
= ct − 1−α

If a given initial condition is ϕ (1) = 1, one obtains c = 1 and thus

α
ϕ (t ) = t − 1−α
Separable equations
Nonlinear rst-order dierential equations are dicult to
handle, except for special cases, one of which is of separable
equations.

Separable equations take the form:

x 0 = F (t , x ) = f (t ) g (x ) (6)

Example 5: x 0 = x 2 − 1 is separable; x 0 = xt + t = t (x + 1) is
separable; x = xt + t is not separable.
0 2

Note: if g (a) = 0, then x (t ) = a is a solution of (6).

A general method to solve separable equations proceeds as follows:

1. Write (6) as:

dx
= f (t ) g (x )
dt
1.1 Then separate the variables:
dx
= f (t ) dt
g (x )
1.2 Integrate to get:
dx
Z Z
= f (t ) dt
g (x )
1.3 Evaluate both integrals (if possible) to obtain the solution.
Example 6: Consider the equation

dx t3
= 6
dt x + 1
Separate the variables:

x 6 + 1 dx = t 3 dt

Integrate: Z Z
6
x + 1 dx = t 3 dt

Evaluate:
1 7 1
x + x = t4 + C
7 4

Without any initial (or nal, etc) condition, there is an innity of

solutions. The equation is solved once we obtain an equation of the
unknown function in which the derivatives do not appear.
Changes of variables
It is often very convenient to transform the variables appearing
in the considered nonlinear dierential equation.
Example 7: A Bernouilli equation has the form, with n real other
than 0 or 1,
x 0 + u (t )x = w (t )x n
Dividing by x n , we get
x −n x 0 + u (t )x 1−n = w (t )
Dening z = x 1−n , so that z 0 = (1 − n)x −n x 0 ,
z0
+ u (t )z = w (t )
(1 − n)

This is a linear rst-order dierential equation in z.

Qualitative theory
In most economic models, it is not possible to solve the obtained
dierential equations explicitly.

In many cases, it remains possible to get information about the

existence and behavior of solutions. In particular, we are interested
in their stability when t becomes arbitrarily large.

Consider the case of rst-order autonomous equations , i.e.

time-independent equations:

x 0 = F (x )
Phase plane analysis of autonomous equations
Phase plane analysis is useful to gain insights into the form of
the solutions of such equations.

A phase diagram plots the curve (x (t ), x 0 (t )):

Stability of equilibria
A point a represents an equilibrium (or stationary) state if and
only if F (a) = 0. In this case x (t ) = a is a solution of the equation.

An equilibrium state a is said locally asymptotically stable if x (t )

converges to a in a neighborhood of a. This occurs if F 0 (a) < 0. If
F 0 (a) > 0, the equilibrium is unstable. (See a1 and a2 in Figure 2.)

An equilibrium state a is said globally asymptotically stable if

x (t ) converges to a everywhere (case of a in Figure 1).
Example 8: An economy produces an output q from its capital
stock k according to the production function q = f (k ) where f is
an increasing and strictly concave function, with f (0) = 0 and
limk →∞ f 0 (k ) = 0 and limk →0 f 0 (k ) = b > 0.

A constant fraction of output is invested, so that gross investment

is sf (k ), and the stock of capital depreciates at the rate n > 0.
The net rate of capital accumulation is:

k 0 = sf (k ) − nk = g (k )
g 0 (k ) = sf 0 (k ) − n tends to sb − n (assumed positive) as k → 0
and to −n as k → +∞, so there exists k1 > 0 such that g (k1 ) = 0.

g 00 (k ) = sf 00 (k ) < 0, so that g is concave.

The graph of g (k ) cuts the horizontal axis at the two equilibrium
points k = 0 and k = k1 > 0. While the equilibrium at k = 0 is
unstable, the equilibrium at k1 is asymptotically stable.
Summary of main concepts
1. Linear rst-order equations are solved using integrating factors:

1.1 e at in the constant coecients or variable RHS terms cases

1.2 e A(t ) , with A(t ) = R a(τ )d τ , in the general case
2. Non-linear rst-order di. eq. can be dicult to solve, but:

2.1 Separable equations, i.e. of the form x 0 = f (t )g (x ), are solved

by separating the variables and integrating
2.2 Changes in variables can simplify some equations
2.3 Otherwise use phase-plane analysis to investigate the form of
solutions
3. An equilibrium state a is locally (globally) asymptotically stable
if the solution converges to a in its neighborhood (everywhere)
Second-order ordinary dierential equations
Second-order ordinary dierential equations are of the form

x 00 = F t , x , x 0 .


(7)

where the solutions are twice dierentiable functions dened on

an interval I.

Example 9: x 00 = k . Then x 0 = kx + C1 and x = k2 x 2 + C1 x + C2 .

For k 6= 0, the solution corresponds to a family of parabolas in the
(t , x ) plane.
Linear second-order dierential equations
We concentrate rst on linear dierential equations, i.e. of the form

x 00 + a (t ) x 0 + b (t ) x = f (t )
where a (t ) , b (t ) and f (t ) are continuous functions of t on an
interval I .

There is no explicit solution in the general case. However, it is

possible to characterize the structure of the general solution.
Homogeneous second-order linear dierential equations
Theorem (2)
The homogeneous linear dierential equation:
x 00 + a(t )x 0 + b(t )x = 0 (8)

has the general solution:

x = Au1 (t ) + Bu2 (t ) (9)

where u1 (t ) and u2 (t ) are two solutions which are not

proportional and A and B are any arbitrary constants.
A proof relying on linear algebra is given in Sydsaeter et al.'s text.
General second-order linear dierential equations
Theorem (3)
The non-homogeneous dierential equation:
x 00 + a (t ) x 0 + b (t ) x = f (t ) (10)

has the general solution

x = Au1 (t ) + Bu2 (t ) + u ∗ (t ) (11)

where Au1 (t ) + Bu2 (t ) is the general solution of the associated

homogeneous equation and u ∗ (t ) is any particular solution of the
non-homogeneous equation.
This result derives straightforwardly from the previous one.
Second-order linear di. eqs with constant coecients
We now turn to the important case where a (t ) = a and b (t ) = b.

We rst consider homogeneous equations of the form

x 00 + ax 0 + bx = 0. (12)

We can try solutions of the form x = e rt , with derivatives

x0 = re rt = rx and x
00 = r 2 e rt = r 2 x . In this case, the dierential
equation can be rewritten as

r 2 + ar + b = 0, (13)

which is simply a second-order polynomial. This is called the

characteristic equation of the dierential equation.
Three cases have to be distinguished depending on the sign of the
discriminant:
∆ = a2 − 4b
Case 1. If ∆ > 0, (13) has two dierent real solutions:

r
1 1
r1,2 = − a ± a2 − b
2 4

Hence e r1 t and e r2 t both satisfy (12) and:

x = Ae r1 t + Be r2 t
Case 2. If ∆ = 0, (13) has one real solution:

1
r =− a
2

Hence, e rt satises (12), but one can check that te rt (not

proportional to e rt ) also does. Therefore:

x = Ae rt + Bte rt = (A + Bt ) e rt
Case 3. If ∆ < 0, (13) has two complex solutions:

r
1 1 1
r1,2 = − a ± i |∆| = − a ± i b − a2
p
2 2 4

α = −a/2 and β = b − a2 /4, u1 (t ) = e αt cos β t

p
Putting and
u2 (t ) = e αt sin β t satisfy (12). Therefore:
x = (A cos β t + B sin β t ) e αt
Example 10: Solve x 00 + 2x 0 + x = 0 with x (0) = 1 and x (1) = 2.

∆ = 22 − 4 = 0. Hence: r = − 21 × 2 = −1 and
x = (A + Bt ) e −t .
Using the conditions:
x (0) = A = 1
x (1) = (1 + B ) e −1 = 2 ⇔ B = 2e − 1
we get:
x (t ) = (1 + (2e − 1) t ) e −t .
Example 11: Solve x 00 − x = 0

∆>0 and r = ±1 ⇒ x = Ae t + Be −t

Solve 3 x 00 − 30x 0 + 75x = 0

∆=0 and r = 5 ⇒ x = (A + Bt ) e 5t
Method of undetermined coecients
By the theorem above, solutions to non-homogeneous equations:
x 00 + ax 0 + bx = f (t ) (14)

are obtained as the sum of a solution to the associated

homogeneous equation and a particular solution to the
non-homogeneous equation.

To nd the particular solution of the non-homogeneous equation,

the method of undetermined coecients is useful
I Iff (t ) is constant, equal to A, a particular solution of (14) is
x = A/ b .
I If f (t ) is a polynomial of degree n, a particular solution is
n
u∗ = Ai t i .
X

i =0
Compute the derivatives, substitute them in the equation and
identify the coecients.
I If f (t ) = pe qt , and q2 + aq + b 6= 0 (e qt is not a solution of
the homogeneous equation), a particular solution is
p
u∗ = e qt
q + aq + b
2

I If f (t ) = p sin rt + q cos rt , a particular solution is

u ∗ = A sin rt + B cos rt with appropriate constants A and B .
Stability of 2nd-order dierential equations
Stability examines whether small changes in initial conditions have
any eects in the long run on the behaviors of the solutions or die
out with time.

The dierential equation x 00 + a (t ) x 0 + b (t ) x = f (t ) is globally

asymptotically stable if and only if every solution
Au1 (t ) + Bu2 (t ) of the associated homogeneous equation tends to
0 as t becomes arbitrarily large, for all values of A and B .

The constants A and B that depend on initial conditions will not

matter in this case.
Example 12: The following solution to a dierential equation:

x = Ae t + Be −t −
1
sin t
2

is not globally asympotically stable as limt →+∞ e t = +∞.

At the opposite, this solution to another dierential equation:

√ √ !
− 2t 3 3
x (t ) = e A cos t + B sin t + t 2 − 2t + 1
2 2

t
is stable as limt →+∞ e − 2 = 0, |cos a| ≤ 1 and |sin a| ≤ 1 for
every a. The equation is globally asymptotically stable.
characterization of global asymptotic stability
Theorem (4)
The dierential equation x 00 + ax 0 + bx = f (t ) is globally
asymptotically stable if and only if a > 0 and b > 0.

Proof: The two roots (real or complex) r1 and r2 of the quadratic

characteristic equation r + ar + b = 0 have the property that
2

r + ar + b = (r − r1 )(r − r2 )
2 = r 2 − (r1 + r2 )r + (r1 r2 ), so that
a = −r1 − r2 and b = r1 r2 . Then use the previous results:

1. If∆ > 0 of ∆ = 0, global asymptotic stability holds i r1 < 0

and r2 < 0, which is true i a > 0 and b > 0

2. If ∆ < 0, it holds if α = −a/2 < 0, which is also true i a>0

and b>0
Summary of main concepts
1. Linear second-order equations do not have explicit solutions in
the general case, but the structure of the solution is

1.1 x = Au1 (t ) + Bu2 (t ) , with u1 and u2 two non proportional

solutions, for an homogenous equation
1.2 x = Au1 (t ) + Bu2 (t ) + u∗ (t ) , with u∗ a particular solution,
for a non-homogenous equation
2. Linear second-order equations with constant coecients have
solutions whose form depends on the discriminant of the
characteristic equation

2.1 and a particular solution can be found using the method of

undetermined coecients
3. The dierential equations are globally asymptotically stable i
every solution of the homogenous equation tends to zero, so
that the initial conditions have no eect as t increases