Second-order differential equations with constant coefficients

 Generally, a 2nd order differential equation can be written in the following form:
F(x; y, y’, y’’) = 0
where y = y(x) is the unknown function to solve for, and y’ = y’(x), y’’=y’’(x) are its 1 st
and 2nd derivatives.
 The function φ(x) will be the solution if it has derivatives φ’(x), φ’’(x) such that
for any x the following equality holds:
F(x; φ(x), φ’(x), φ’’(x)) = 0

 For any two initial conditions: y0 = y(x0), y’0 = y’(x0)

there exists only one solution of the homogeneous equation which satisfies the initial
conditions.
This constitutes the basic contents of Cauchy’s theorem*.

*The Cauchy–Kowalevski theorem is the main theorem on existence and uniqueness of solution for ordinary differential
equations (functions of a single variable, a subclass of partial differential equations). January 1, 2025 1
Second-Order Differential Equations
 Homogeneous linear 2nd order differential equations
Given: y’’ + by’ + cy = f(x), the homogenous equation is:
yt’’ + byt’ + cyt = 0 (homogeneous)
Searching for the solution using examples:
Example 1
yt’’ – yt = 0
It is easy to verify that the function: y = ex is a solution.
We can show that for any two constants c1, c2, the function:
yt = c1ex + c2e-x is also a solution to (1), since:
yt’ = c1ex - c2e-x
yt’’ = c1ex + c2e-x = yt
so, yt’’ - yt = 0
From Cauchy’s theorem there is no other solution to the homogeneous
equation.
Second-Order Differential Equations January 1, 2025 2
 Generalizing:
Example 2
y’’ – 4y = 0 (1)
We can find the solution in the form: y = eλx, where λ is an unknown number.
Substitute this and its second derivative into (1):
y’ = λeλx, and, y’’ = λ2eλx, then:
λ2eλx - 4eλx = 0 → (λ2 – 4)eλx = 0
therefore, a function of the form eλx satisfies equation (1) if and only if: (λ2 – 4) = 0.
The roots that satisfy the equation are λ1= 2 and λ2= -2.
Therefore, the solution is:
y = c1e2x + c2e-2x

Second-Order Differential Equations January 1, 2025 3

 Generalization: Characteristic equation
The above examples suggest an idea to use a solution of the form: eλx whenever we look
for a solution of a 2nd order differential equation.
 Given a homogeneous linear equation with constant coefficients:
y’’ + by’ + cy = 0 (1)

Substitute eλx as well as its first and second derivatives into (1):
y’’ + by’ + cy = 0
λ2eλx + bλeλx + ceλx = 0 → eλx(λ2 + bλ + c) = 0

The function eλx will be a solution of (1) for a λ which satisfies:

(λ2 + bλ + c) = 0 (2)
Equation (2) is called the characteristic equation of (1).
Proposition:
 If φ1 and φ2 are both solutions of (1), then so is the function φ = c 1φ1 + c2φ2,
which is the general solution of the homogeneous equation.
Second-Order Differential Equations January 1, 2025 4
 Then: yc = c1eλ1x + c2eλ2x

 Now, use time as the independent variable (t instead of x), so y’ is

dy/dt, y’’ = d2y/dt2, etc.:
Three types of roots for the characteristic equation:
(1) Case of two distinct real roots
Example 1 (no initial conditions):
y’’ + 4y’ + 3y = 0
The characteristic equation is:
λ2 + 4λ + 3 = 0, and λ1,2 = [-4 ± (√16 - 12)]/2, so:
λ1 = -1, λ2 = -3, then:
yt = c1e-t + c2e-3t

Second-Order Differential Equations January 1, 2025

5
 An example with initial conditions:
Example 2:
d2y/dt2 - 3(dy/dt) + 2y = 0, y’(0) = 6, y(0) = 4.
or: y’’ - 3y’ + 2y = 0,
The characteristic equation is: λ2 - 3λ + 2 = 0, and: λ1,2 = (3 ± 1)/2 = 2, 1.
The complimentary (and general solution) is: yt = c1e2t + c2et
Then, to satisfy the initial conditions:
y(0) = c1e0 + c2e0 = c1 + c2 = 4
y’(t) = 2c1e2t + c2et and y’(0) = 2c1e0 + c2e0 = 2c1 + c2 = 6

solving, c1 = 2, c2 = 2. The solution of the homogeneous equation is:

yt = 2e2t + 2et (definite solution)

Second-Order Differential Equations January 1, 2025 6

 (2) Case of multiple roots (repeated root)
y’’ + by’ + cy = 0 (1)
Characteristic equation:
λ2 + bλ + c
We have repeated roots if the discriminant is 0.
λ = [-b ± √0]/2 = -b/2
Note that, b = -2λ, c = λ2 [since b2 = 4c; c = b2/4 = λ2].

Solution: eλt is a solution to equation (1).

We can show that in this case (of repeated root), teλt is also a solution.
We have:
From: y = teλt → y’ = eλt + λteλt, y’’ = λeλt + λeλt + λ2teλt

Second-Order Differential Equations January 1, 2025 7

 or:
y’ = eλt + λteλt and y’’ = 2λeλt + λ2teλt , so:
y’’ + by’ + cy = 2λeλt + λ2teλt + beλt + btλeλt + cteλt
= eλt (2λ + b) + teλt (λ2 + bλ + c)
= eλt (2λ - 2λ) + teλt (λ2 - 2λ2 + λ2) = 0
Therefore, the solution of the homogeneous equation is:
yt = c1eλt + c2 teλt

Example:
y’’ – 2y’ + y = 0
λ1,2 = 1 (repeated root)
solution of this homogeneous eq. is:
yt = c1et + c2tet
Second-Order Differential Equations January 1, 2025 8
 (3) Case of complex roots
y’’ + by’ + cy = 0 (1)
λ2 + bλ + c (2)
When: b2 - 4c < 0, the roots are complex.
Given the conjugate complex roots: λ1,2 = h ± iv (where h is the real part of the
root), then the general solution will be:

yt = c1e(h + iv)t + c2e(h - iv)t

which can be written as:
yt = eht(c1eivt + c2e-ivt) [we’ll come back to this later]

Second-Order Differential Equations January 1, 2025

9
 Particular solution (for any case of roots)
Suppose the right-hand side is a constant.
y’’ + by’ + cy = d (1)

Consider a particular solution of the form: yp = k, then:

y’ = 0, y’’ = 0, and k = d/c = yp

Second-Order Differential Equations January 1, 2025

10
 Example (real roots):
y’’ - 2y’ - 3y = 10, y’(0) = 6, y(0) = 4.
→ λ1,2 = 3, -1

yp = k → 0 – 0 – 3k = 10→0 → k = -10/3
then, general solution:
yt = c1e3t + c2e-t – 10/3

Using the additional informtion:

y(0) = c1e0 + c2e0 – 10/3 = c1 + c2 – 10/3 = 4
Since yt’ = 3c1e3t – c2e-t
y’(0) = 3c1 – c2 = 6, and
c1 = 10/3, c2 = 4
The definite solution is:
Second-Order DifferentialyEquations
t = (10/3)e + 4e-t – 10/3
3t
January 1, 2025
11
 Example (complex roots):
y’’ + 2y’ + 17y = 34, y(0) = 3, y’(0) = 11
 First, particular solution: yp = 34/17 = 2.
Characteristic equation: λ2 + 2λ + 17 = 0 gives complex roots (h + vi and h – vi): (-1 + 4i) and (-1 – 4i)

Generally, the complimentary solution is:

yt = eht(A1eivt + A2e-ivt) (1)

To facilitate interpretation of imaginary exponetial functions: using their equivalint circular functions:
= eht [A1(cos vt + isin vt) + A2(cos vt - isin vt)
= eht [(A1 + A2)cos vt + (A1 – A2) isin vt]
= eht [(A5)cos vt + (A6) sin vt] (where A6 = (A1-A2)i

 In this example the time path (general solution) is given by:

yt = e-t(A1ei4t + A2e-i4t) + 2
= e-t(A5 cos4t + A6 sin4t) + 2 (were: A5=A1+A2, A6 = (A1-A2)i

Using the initial condition, we can find the definite solution.

However, you are not required to know how to find the definite solution under complex roots. It
requires
Second-Order remembering
Differential properties of sin/cos functions and their derivatives.
Equations January 1, 2025 12
 Stability of equilibrium (first and second order)
1. First-order differential equations
The solution is of the form: ceλt. Since e > 1, et will grow at an ever-increasing
rate if λ > 0. Thus, if λ > 0, the solution is non-covergent (explosive solution).

If λ < 0, the solution is convergent.

 If λ = 0, then eλt = 1 and the solution remains constant over time.

Since the constant c can be positive or negative, we identify 6 possibilities for

the time path of y (for complimentary solution):

Stability of Equilibrium January 1, 2025

13
 Yt

0 t

January 1, 2025
14
Stability of Equilibrium
 2. Second-order differential equations - Stability of time path
yt = c1eλ1t + c2eλ2t + yp
Two real roots:
If the largest root is λi, eventually the solution will approximate: yt = ci(eλit), since
the other term becomes relatively less important.
The time path is unstable when there is at least one positive root.

But “eventually” may be a long time and we may be more interested what happens
over a shorter period, we can answer this if the roots are of the same sort, i.e., both
positive or negative. When this is not the case, what will happen will depend on
the relative values of the constants, c1 and c2.

If multiple (real) roots:

yt = c1eλt + c2teλt
If λ < 0, stable/convergent time path. Note that c2teλt is a decreasing function in t
(i.e., derivative with respect to t is < 0), hence it will tend to 0 over time if λ < 0.
Stability of Equilibrium January 1, 2025 15
 Complex roots:
When we have complex roots there will be a term in the solution of the form:
yt = eht(c1eivt + c2e-ivt)
The expression in parenthesis will repeat itself (trigonometric fluctuations).

Convergence or divergence of fluctuations will depend on what is outside the

parentheses, i.e., eht, where h is the real part of the complex root. Stability
requires h < 0.

The distinction between +ve and –ve real part of the root (h) is of particular
importance because it determines whether, the fluctuations that occur will become
damped (convergent) or explosive.
Stability of Equilibrium January 1, 2025 16
  There is an easy way to determine whether, if the roots are complex,
the real part is + ve or – ve, by inspecting the equation itself. From:
λ2 + bλ + c = 0,
The roots are: (-b ± √b2 – 4c )/2 = -b/2 ± (i√4c - b2 )/2
Hence, the real part of the complex root, h = –b/2 is -ve if b is + ve and
vice versa.

Stability of Equilibrium January 1, 2025

17
 Economic applications
(1) Dynamic models of market price
 Consider a market model where the market behaviour of individuals is
affected by expectations about the future price. The derivative dP/dt refers to
whether price is rising or falling over time. On the other hand, the second
derivative d2P/dt2 indicates whether price increases at an increasing rate (or
decreasing at a decreasing rate).
If Qd and Qs depend on d2P/dt2 in addition to dP/dt then:
Qd = D[P(t), dP/dt, d2P/dt2]
Qs = S[P(t), dP/dt, d2P/dt2]
Using linear D and S equations and the alternative notation:
Qd = α – βP + mP’ + nP’’ (α, β > 0)
Qs = -γ + δP + uP’ + wP’’ (γ, δ > 0)

Second-Order: Applications January 1, 2025 18

 Assume market always clears.
Equating D to S (i.e., the assumption that the market always clears) and
simplifying:
P’’ + bP’ + cP = d
where b, c, d are related to the original coefficients.

The particular solution is found by setting yp = k, then:

k = d/c

The general solution in the case of 2 roots is:

Pt = c1eλ t +c2eλ t + d/c
1 2

Second-Order: Applications January 1, 2025 19

 Example (from C & W)
Qd = 3P’’ + P’ – P + 9
Qs = 5P’’ – P’ + 4P – 1
P(0) = 4, P’(0) = 4

Equating Qd and Qs and simplifying:

P’’ – P’ +2.5P = 5
Particular solution: Pp = 2 ( -2.5P = -5 → P =2)
The characteristic equation is: λ2 – λ – 5/2 = 0 and has complex
roots:
½ ± (3/2)i, with h = ½ and v = 3/2.
(b = -1, so –b/2 is positive)
The general solution will be:
Pt = e1/2t( )

The path
Second-Order: Applications is non-convergent and price fluctuations are explosive.
January 1, 2025 20
 Another application: The dynamic Philips curve

Background:
Philips (1958) published: The relation between unemployment and the rate of
change of money wages in the UK, 1861-1957.

Found a statistically negative relationship between (wage) inflation and

unemployment (see graph).

Second-Order: Applications January 1, 2025 21

Graph from Wikipedia
January 1, 2025
22
  Similar patterns found in other countries, and in 1960 Samuelson and Solow
formalized this relationship between unemployment and inflation.
 During those years it was believed that there is a permanent/stable
relationship which could be exploited for policy purposes.
 However, since 1974, seven Nobel prizes have been given for work critical
of the original Philips Curve hypothesis. Most of this criticism was initiated
during the 1970s when the world observed high rates of both unemployment
and inflation.

Second-Order: Applications January 1, 2025

23
1961 - 2005
January 1, 2025
24
  Friedman argued the negative inflation-unemployment relationship was only
a short run phenomenon (in the long run workers will learn to anticipate
inflation, resulting in employment contracts with higher wage rates according
to anticipated inflation).
 In the long run unemployment will converge to the “natural rate” and
will be no trade off.
 In its modern version, policies can result in short run declines in
unemployment by increasing permanent inflation.
Inflation can lower unemployment temporarily, but not permanently.

Second-Order: Applications January 1, 2025 25

 In the short-run policymakers will face an inflation-unemployment tradeoff (Initial Short-Run
Phillips Curve). Policymakers can reduce the unemployment rate temporarily, moving from
point A to point B through expansionary policies.

However, exploiting this short-run

tradeoff will raise inflation
expectations, shifting the short-
run curve rightward to a "New
Short-Run Phillips Curve" and
moving the point of equilibrium
from B to C.

Thus, the reduction in

unemployment below the
"Natural Rate" will be
temporary and lead only to
higher inflation in the long run.

Second-Order: Applications January 1, 2025 26

  What has remained influential is a modified form of the (short-run) Phillips
Curve that takes inflationary expectations into account and distinguishes
between a short-run vs a long-run PC. This "short-run Phillips curve" is also
called the "expectations-augmented Phillips curve", since it shifts up when
inflationary expectations rise.
 This seems to imply that in the long run, monetary policy cannot affect
unemployment. Unemployment adjusts back to its "natural rate", also called the
"NAIRU" (long-run PC is vertical).
 This conclusion relies on an assumption. We are later going to examine under
what conditions this is the case (i.e., long-run PC is vertical).

Second-Order: Applications January 1, 2025 27

 The dynamic Philips curve – The model
(1) Original Phillips Relation
[Define lowercase w, p representing rates of change of wages and prices,
while (capital) U is the rate of unemployment]
w = f(U) (f’(U) < 0)
Using a linear function: w = α – βU.
Price inflation: p = w – T (p is rate of change in price level; T is labour
productivity, induced by technology).
So, the linearized original Phillips curve for price inflation is:
p = α – T – βU (α, β > 0)

(2) Introducing expectations

w = f(U) + gπ (0 < g < 1) (π is the expected inflation rate)

Second-Order: Applications January 1, 2025 28

Then: p = α – T – βU + gπ (1)

 How inflation expectations (π) are adjusted? Assume adaptive expectations:

dπ/dt = j(p – π) (0 < j < 1) (2)
i.e.: If individuals under-predict the rate of inflation, they revise their expectations up
(dπ/dt > 0), and vice versa.

 Now we add a 3rd equation for unemployment. Assume (for simplicity) that the only
government policy that matters is monetary policy. Unemployment responds to the
growth of real money supply (M/P). Denoting the rate of growth of nominal money
supply by m:
dU/dt= -k(m-p) (3)
(where k > 0 and m-p is the growth of real money-supply, i.e., of M/P).

[Note that, if m and p are the growth rates, i.e., and the growth rate of real money
supply, is m-p]
According to (3), the change in the unemployment rate over time is determined by the
rate of growth or real money supply. January 1, 2025
Second-Order: Applications 29
 The time path of π (similarly for p or U):
Equations (1) – (3):
p = α – T – βU + gπ (1)
dπ/dt = j(p – π) (2)
dU/dt= -k(m-p) (3)
constitute a model with 3 endogenous variables, π, p and U.
We can collapse the model into (any) one equation we choose, i.e., in π, p, or U;
for example, if we choose to derive a differential equation in π:
Putting (1) into (2):
dπ/dt = j(α – T – βU + gπ – π), or
dπ/dt = j(α – T – βU) – j(1-g)π (4)
 Differentiate (4) with respect to time:
d2π/dt2 = -jβ(dU/dt) – j(1-g) (dπ/dt) (5)
Second-Order: Applications January 1, 2025 30
 Put (3) into (5) to replace dU/dt with -k(m-p):
d2π/dt2 = jβkm – jβkp – j(1-g) (dπ/dt) (5)’
We still need to eliminate p. From (2):
p = (1/j)(dπ/dt) + π
Substituting into (5)’:
d2π/dt2 + [βk + j(1-g)]dπ/dt + (jβk)π = jβkm
[2nd order linear differential eqn with constant coefficients and constant term]

The particular solution is: πp = (jβkm)/ (jβk) = m

Which means that the inter-temporal equilibrium value (equilibrium over

time) of expected inflation depends only on (i.e., converges to) the rate of
change of money supply.

Second-Order: Applications January 1, 2025 31

  For the complimentary solution (which will inform on existence and type of
equilibrium over time), we have two roots and depending on the type of roots we’ll
know if the time path is monotonic or with fluctuations.
 But since both α1[= βk + j(1-g)] and α2 [= jβk] > 0, it can be shown that the time path is
convergent (see book for demonstration). However, we don’t know beforehand if we
have 2 distinct real roots, repeated roots of complex roots.
 Similarly, the time path of p can be also found to converge to m. So, in the long-run
(given it enough time), the rate of change of the price level (inflation) will equal to the
rate of growth of nominal money supply.
 On the other hand, if we derive the time path of the unemployment rate, as long as the
coefficient g = 1 (i.e, all expected inflation is transferred to wages and prices), the
particular solution (where the unemployment rate converges over time) will be a
number independent of the policy variable m (rate of growth of money supply) and
regardless of the equilibrium rate of inflation.
This highlights the result that in the long-run Philips curve is vertical [Note the
critical assumption, i.e., g = 1].
Q5 in Tut. 9 deals with what happens if g < 1:
January 1, 2025
Second-Order: Applications 32