Skip to content

Interpreting differential equation models

A differential equation describes a quantity through its rate of change. In a population model, for example, the equation does not state the population directly. It states how quickly the population changes. Solving the equation recovers possible population functions; an initial condition selects the one relevant to the situation.

This lesson focuses on interpreting models. For the integration technique itself, first study separable differential equations.

You should be able to:

  • interpret dydx\dfrac{dy}{dx} as an instantaneous rate of change
  • separate variables and integrate
  • use ln\ln and exponential functions
  • apply an initial condition to find a constant
  • sketch exponential graphs and solve exponential equations

Review integration basics, growth and decay and A-level calculus and differentiation if needed.

Let Q(t)Q(t) be a quantity depending on time. Translate the wording one piece at a time.

WordsMathematical statement
the rate of change of QQdQdt\dfrac{dQ}{dt}
proportional to QQdQdt=kQ\dfrac{dQ}{dt}=kQ
proportional to the square of QQdQdt=kQ2\dfrac{dQ}{dt}=kQ^2
proportional to AQA-QdQdt=k(AQ)\dfrac{dQ}{dt}=k(A-Q)
decreases at a constant rate rrdQdt=r\dfrac{dQ}{dt}=-r

The constant of proportionality carries the sign and the units required by the model. If Q>0Q>0 and dQdt=kQ\dfrac{dQ}{dt}=kQ, then k>0k>0 gives growth and k<0k<0 gives decay.

Worked example 1: form the equation before solving

Section titled “Worked example 1: form the equation before solving”

A culture contains P(t)P(t) bacteria, where tt is measured in hours. Its rate of growth is proportional to the number present.

“Rate of growth” means dPdt\dfrac{dP}{dt} and “proportional to the number present” means kPkP. Therefore

dPdt=kP,k>0.\boxed{\frac{dP}{dt}=kP},\qquad k>0.

If the culture were dying at a rate proportional to its population, one could write either

dPdt=kP,k>0,\frac{dP}{dt}=-kP,\qquad k>0,

or dPdt=kP\dfrac{dP}{dt}=kP with k<0k<0. Both conventions work, but the meaning of kk must be stated consistently.

Misconception: confusing an amount with its rate

Section titled “Misconception: confusing an amount with its rate”

“The population is proportional to time” would mean P=ktP=kt. It does not mean dPdt=kP\dfrac{dP}{dt}=kP. Always identify the quantity on the left before translating the rest of the sentence.

A solution family and a particular solution

Section titled “A solution family and a particular solution”

Solving

dPdt=kP\frac{dP}{dt}=kP

gives a family of solutions. Separating variables,

1PdPdt=k,\frac1P\frac{dP}{dt}=k,

so

1PdP=kdt.\int\frac1P\,dP=\int k\,dt.

Hence

lnP=kt+C.\ln|P|=kt+C.

Exponentiating gives

P=eCekt.|P|=e^Ce^{kt}.

The arbitrary positive factor eCe^C can be absorbed into a new constant. Allowing that constant to have either sign gives

P=Aekt.P=Ae^{kt}.

For a population P>0P>0, so A>0A>0. At t=0t=0,

P(0)=Ae0=A.P(0)=Ae^0=A.

Thus AA is the initial population. The differential equation determines the shape of the family; the initial condition determines which member of the family models the culture.

Worked example 2: use data and interpret the constants

Section titled “Worked example 2: use data and interpret the constants”

A population satisfies

dPdt=kP.\frac{dP}{dt}=kP.

Initially there are 800800 organisms, and after 33 hours there are 12001200. Find the model.

The general positive solution is

P=Aekt.P=Ae^{kt}.

Using P(0)=800P(0)=800 gives A=800A=800. Now use P(3)=1200P(3)=1200:

1200=800e3k.1200=800e^{3k}.

Therefore

e3k=32,3k=ln32,k=13ln32.e^{3k}=\frac32, \qquad 3k=\ln\frac32, \qquad k=\frac13\ln\frac32.

The exact model is

P=800e(ln(3/2))t/3.\boxed{P=800e^{(\ln(3/2))t/3}}.

Here 800800 is the population at t=0t=0. The parameter

k=13ln320.135 h1k=\frac13\ln\frac32\approx0.135\text{ h}^{-1}

is the continuous proportional growth constant. Its unit is per hour because the exponent ktkt must have no units.

It is tempting to call kk a 13.5%13.5\% hourly increase. That is only an approximation. The exact multiplication factor over one hour is

ek=(32)1/31.1447,e^k=\left(\frac32\right)^{1/3}\approx1.1447,

so the population increases by about 14.5%14.5\% in one hour under this model.

Read information directly from the equation

Section titled “Read information directly from the equation”

Before solving, inspect the sign and zeros of the rate.

  • If dydt>0\dfrac{dy}{dt}>0, then yy is increasing.
  • If dydt<0\dfrac{dy}{dt}<0, then yy is decreasing.
  • If dydt=0\dfrac{dy}{dt}=0, the model has an equilibrium value, provided a constant solution at that value is valid.
  • A large dydt\left|\dfrac{dy}{dt}\right| means a steep solution curve, not necessarily a large value of yy.

Worked example 3: approach to an equilibrium

Section titled “Worked example 3: approach to an equilibrium”

The temperature TT in degrees Celsius of an object is modelled by

dTdt=0.2(T20),\frac{dT}{dt}=-0.2(T-20),

where tt is in minutes and T(0)=80T(0)=80.

The rate is zero when T=20T=20, so 20C20^\circ\text{C} is an equilibrium temperature. If T>20T>20, then T20>0T-20>0 and dTdt<0\dfrac{dT}{dt}<0, so the object cools. If T<20T<20, then dTdt>0\dfrac{dT}{dt}>0, so it warms. In both cases the rate points towards 2020.

Now solve. Write u=T20u=T-20. Since dudt=dTdt\dfrac{du}{dt}=\dfrac{dT}{dt},

dudt=0.2u.\frac{du}{dt}=-0.2u.

Therefore

u=Ae0.2t,u=Ae^{-0.2t},

and hence

T=20+Ae0.2t.T=20+Ae^{-0.2t}.

Using T(0)=80T(0)=80 gives 80=20+A80=20+A, so A=60A=60. Thus

T=20+60e0.2t.\boxed{T=20+60e^{-0.2t}}.

The parts of the answer have distinct meanings:

  • 2020 is the surrounding temperature and the long term equilibrium
  • 60=802060=80-20 is the initial temperature difference
  • 0.2 min10.2\text{ min}^{-1} controls how rapidly that difference decays

As tt\to\infty, e0.2t0e^{-0.2t}\to0, so T20T\to20. The model predicts that TT approaches 20C20^\circ\text{C} but does not equal it at any finite time, because 60e0.2t>060e^{-0.2t}>0.

For the model

dNdt=0.4N(1N500),\frac{dN}{dt}=0.4N\left(1-\frac{N}{500}\right),

find the equilibrium values and determine whether NN increases or decreases when (a) 0<N<5000<N<500 and (b) N>500N>500.

Answer

The rate is zero when

N=0or1N500=0,N=0\quad\text{or}\quad1-\frac{N}{500}=0,

so the equilibria are N=0N=0 and N=500N=500.

For 0<N<5000<N<500, both factors are positive, so dNdt>0\dfrac{dN}{dt}>0 and NN increases. For N>500N>500, the second factor is negative, so dNdt<0\dfrac{dN}{dt}<0 and NN decreases. The rate therefore points towards N=500N=500 from either side.

You can extract all of this without solving the differential equation.

Initial conditions are not optional decoration

Section titled “Initial conditions are not optional decoration”

An initial condition usually has the form

y=y0whenx=x0,y=y_0\quad\text{when}\quad x=x_0,

or equivalently y(x0)=y0y(x_0)=y_0. Substitute both coordinates into the general solution.

Worked example 4: a reciprocal model and its valid domain

Section titled “Worked example 4: a reciprocal model and its valid domain”

The mass MM of a crystal satisfies

dMdt=0.02M2,M(0)=10,\frac{dM}{dt}=0.02M^2, \qquad M(0)=10,

where MM is in grams and tt is in days.

Separate and integrate:

M2dM=0.02dt,M^{-2}\,dM=0.02\,dt, M1=0.02t+C.-M^{-1}=0.02t+C.

It is convenient to rearrange before using the condition:

1M=C10.02t.\frac1M=C_1-0.02t.

Since M(0)=10M(0)=10,

110=C1.\frac1{10}=C_1.

Therefore

M=10.10.02t=505t.\boxed{M=\frac1{0.1-0.02t}=\frac{50}{5-t}}.

The algebraic expression is defined for t5t\ne5, but the physical solution beginning at t=0t=0 is meaningful only for

0t<5.0\leq t<5.

The model predicts MM\to\infty as t5t\to5^-, which is evidence that the model must fail before day 55. It does not predict a physically meaningful negative mass after day 55.

This is an important distinction: the domain of a formula and the domain in which a model is credible need not be the same.

To verify a model solution, check both requirements:

  1. Differentiate it and substitute into the differential equation.
  2. Substitute the initial condition.

Worked example 5: verification catches a sign error

Section titled “Worked example 5: verification catches a sign error”

Claim: y=3+2e4ty=3+2e^{-4t} solves

dydt=4(3y),y(0)=5.\frac{dy}{dt}=4(3-y), \qquad y(0)=5.

Differentiate:

dydt=8e4t.\frac{dy}{dt}=-8e^{-4t}.

The right hand side is

4(3y)=4(332e4t)=8e4t.4(3-y)=4(3-3-2e^{-4t})=-8e^{-4t}.

The two sides agree. Also,

y(0)=3+2=5.y(0)=3+2=5.

Therefore the claim is correct. Notice that checking only y(0)=5y(0)=5 would not establish that the function satisfies the differential equation.

Does Q=12e0.3tQ=12e^{-0.3t} satisfy

dQdt=0.3Q,Q(0)=10?\frac{dQ}{dt}=-0.3Q, \qquad Q(0)=10?
Answer

Differentiation gives

dQdt=3.6e0.3t=0.3Q,\frac{dQ}{dt}=-3.6e^{-0.3t}=-0.3Q,

so the differential equation is satisfied. However,

Q(0)=1210.Q(0)=12\ne10.

It is a member of the correct solution family, but it is not the particular solution satisfying the stated initial condition. That solution is Q=10e0.3tQ=10e^{-0.3t}.

The units on both sides of a differential equation must agree. If VV is measured in litres and tt in seconds, then

dVdt\frac{dV}{dt}

has units L s1\text{L s}^{-1}.

In

dVdt=kV,\frac{dV}{dt}=kV,

kk must have units s1\text{s}^{-1}. In

dVdt=kV2,\frac{dV}{dt}=kV^2,

kk must have units

L s1L2=L1s1.\frac{\text{L s}^{-1}}{\text{L}^2}=\text{L}^{-1}\text{s}^{-1}.

The exponent in an exponential must be dimensionless. Thus e0.4te^{-0.4t} is meaningful only if 0.4t0.4t has no units. When tt is in minutes, 0.40.4 has units min1\text{min}^{-1}.

The radius rr of a drop is measured in millimetres and time tt in seconds. In the model

drdt=kr2,\frac{dr}{dt}=-kr^2,

what are the units of kk?

Answer

The left side has units mm s1\text{mm s}^{-1}, so

[k]=mm s1mm2=mm1s1.[k]=\frac{\text{mm s}^{-1}}{\text{mm}^2} =\boxed{\text{mm}^{-1}\text{s}^{-1}}.

A model answer should name the quantity, include units, and respect the model’s domain.

For the cooling model

T=20+60e0.2t,T=20+60e^{-0.2t},

the time at which the object reaches 30C30^\circ\text{C} satisfies

30=20+60e0.2t.30=20+60e^{-0.2t}.

Therefore

e0.2t=16,0.2t=ln16=ln6,e^{-0.2t}=\frac16, \qquad -0.2t=\ln\frac16=-\ln6,

so

t=5ln68.96.t=5\ln6\approx8.96.

A complete conclusion is:

The model predicts that the object reaches 30C after about 8.96 minutes.\boxed{\text{The model predicts that the object reaches }30^\circ\text{C after about }8.96\text{ minutes}.}

The phrase “the model predicts” matters. A mathematical model is a simplified description, not a guarantee.

The form of a differential equation encodes assumptions. For example,

dPdt=kP\frac{dP}{dt}=kP

with constant k>0k>0 assumes that:

  • every individual has the same average contribution to growth
  • the proportional growth rate stays constant
  • resources, space and other limiting effects do not change the rate
  • migration or harvesting is absent, or already included in the net rate
  • treating a population as continuous is adequate

Its solution P=P0ektP=P_0e^{kt} grows without bound, so it may be useful over a short interval but unrealistic over a long one.

Similarly, Newton’s cooling model

dTdt=k(TTs)\frac{dT}{dt}=-k(T-T_s)

usually assumes constant surrounding temperature TsT_s, constant kk, and conditions in which heat transfer is approximately proportional to the temperature difference. Changing airflow, evaporation or a varying room temperature can invalidate those assumptions.

When criticising a model, connect the limitation to the equation. “The model is unrealistic” is too vague. “The constant proportional growth rate ignores resource limitations, so long term exponential growth is unlikely” is precise.

An indefinite integration produces a family. Keep the arbitrary constant until an initial condition determines it.

A proportionality constant generally has units. Check that both sides of the differential equation match.

Assuming every algebraic branch is physical

Section titled “Assuming every algebraic branch is physical”

Reject values only for a stated reason, such as t<0t<0, negative population, a singularity, or a restriction in the question.

In T=20+60e0.2tT=20+60e^{-0.2t}, the temperature approaches 2020 from above but never crosses it. A horizontal asymptote can represent an equilibrium.

Giving a percentage interpretation for kk without checking

Section titled “Giving a percentage interpretation for kkk without checking”

In Q=Q0ektQ=Q_0e^{kt}, the exact factor over one time unit is eke^k, so the exact percentage change is 100(ek1)%100(e^k-1)\%. The approximation 100k%100k\% is reliable only when k|k| is small.

A chemical mass MM grams satisfies

dMdt=kM,\frac{dM}{dt}=-kM,

where tt is in hours and k>0k>0. Initially M=50M=50, and after 44 hours M=30M=30.

  1. Find MM in terms of tt, giving kk exactly.
  2. Find the half-life.
  3. State the units of kk.
  4. Give one modelling assumption.
Answer

The general solution is

M=Aekt.M=Ae^{-kt}.

Since M(0)=50M(0)=50, A=50A=50. Using M(4)=30M(4)=30,

30=50e4k,e4k=35.30=50e^{-4k}, \qquad e^{-4k}=\frac35.

Taking logarithms,

4k=ln35,-4k=\ln\frac35,

so

k=14ln53.k=\frac14\ln\frac53.

Therefore

M=50e(ln(5/3))t/4.\boxed{M=50e^{-(\ln(5/3))t/4}}.

For the half-life, set M=25M=25:

25=50ektekt=12t=ln2k.25=50e^{-kt} \quad\Rightarrow\quad e^{-kt}=\frac12 \quad\Rightarrow\quad t=\frac{\ln2}{k}.

Hence

t=4ln2ln(5/3)5.43 hours.\boxed{t=\frac{4\ln2}{\ln(5/3)}\approx5.43\text{ hours}}.

The units of kk are h1\text{h}^{-1}. A possible assumption is that the proportional decay constant remains unchanged throughout the modelled period. Equivalently, external conditions affecting the decay are assumed constant.

For a differential equation model, ask:

  1. What does each variable represent, and what are its units?
  2. Has the wording been translated into a rate equation correctly?
  3. What does the sign of the derivative say before any solving?
  4. Are there equilibrium values where the derivative is zero?
  5. Has the general solution kept its arbitrary constant?
  6. Has every initial condition been applied?
  7. Does the answer satisfy both the equation and the conditions?
  8. What domain is mathematically valid and physically credible?
  9. What assumptions are encoded in the equation?

Next, consolidate the solving technique in separable differential equations, revisit growth and decay for exponential models, and practise translating conclusions through interpreting answers in context.