The University of Sydney

MOOC Introduction to Calculus

Notes for ‘The logistic function’

Important Ideas and Useful Facts:

(i) The exponential growth model: Consider a population of size x = x(t) > 0, which is a
function of time t. The exponential growth model assumes that the growth rate of x with
respect to t is proportional to x, that is,
dx
= kx
dt
for some constant k > 0. This makes precise the idea that, the more there is of the
population or colony, the faster it should grow. One can show that the only solutions to
this differential equation (an equation involving a derivative) have the form

x = x(t) = Aekt

for some constant A > 0, which is the usual rule for exponential growth.

(ii) The logistic growth model: The exponential growth model is modified to take into ac-
count constraints on the growth of the population caused by limited resources. A more
sophisticated model presumes there is a largest sustainable population size, say M > 0,
and adjusts the differential
( ) equation above for x = x(t), where k > 0, by including an
inhibition factor 1 − M , which is close to one when x is small, but close to zero when
x

x is close to M : (
dx x)
= kx 1 − .
dt M
This (differential equation) is called the logistic equation. Its solutions are called logistic
functions, and they all have the form
M
x = x(t) =
1 + Ke−kt
for some constant K > 0. The graph of x (where x labels the vertical axis, whilst t
labels the horizontal axis) has a sigmoid shape, concave up followed by concave down.
The curve is increasing, with horizontal asymptotes x = 0 and x = M , and has 180◦
rotational symmetry about a point of inflection that occurs when x = M2 and t = lnkK .
x
M
x= 1+Ke−kt
M

M b
2 inflection

ln K t
k

1
Examples and derivations:
1 1 A B
1. Find integers A and B such that = = + .
119 7 × 17 7 17
Solution: We want 1 = 17A + 7B. We look for multiples of 17 and 7 that differ by 1, say
34 = 2 × 17 and 35 = 5 × 7, so take A = −2 and B = 5, and then we have
1 −2 5
= + .
119 7 17
2. Find constants A and B such that
1 1 A B
= = + .
2x − x2 x(2 − x) x 2−x

Solution: After rewriting the right-hand side with a common denominator, we find that
we need
1 = A(2 − x) + Bx = 2A + (B − A)x .
To get constants to match, we want 1 = 2A, so that A = 12 . It follows that B = 1
2
also,
so we have
1 1
1 1 2 2
= = + .
2x − x2 x(2 − x) x 2−x
∫
dx
3. Find the indefinite integral (assuming 0 < x < 2).
2−x
Solution: Put u = 2 − x, so that du
dx
= −1. Hence du = −dx and dx = −du, and we get
∫ ∫ ∫
dx −du du
= = − = − ln u + C = − ln(2 − x) + C ,
2−x u u
noting that u > 0, for the third step.
∫
dx
4. Find the indefinite integral (assuming 0 < x < 2).
2x − x2
Solution: Using the solutions to the two previous exercises, we get
∫ ∫ (1 1 ) (∫ ∫ )
dx 2 2 1 dx dx
= + dx = +
2x − x2 x 2−x 2 x 2−x
( ) ( )
1 1 x
= ln x − ln(2 − x) + C = ln +C .
2 2 2−x

5. Consider the logistic function

M
x = x(t) = ,
1 + Ke−kt
where M, K, k > 0. Use interval notation to describe the range and show that the
derivative x′ (t) is maximised when
ln K
t = ,
k
M
given that this occurs when x = 2
.

2
 Solution: The curve is sandwiched between horizontal asymptotes x = 0 and x = M , so
that the range is the interval (0, M ). The derivative is maximised when
M M
= ,
2 1 + Ke−kt
so that 2 = 1 + Ke−kt , giving Ke−kt = 1, so that K = ekt . Taking natural logarithms
and rearranging, we get
ln K
t = .
k
6. A rumour has been spread, amongst a population of 50,000 people, that calculus is fun.
It was seeded by ten people, and after 10 days, it is estimated that 1,000 people know
about the rumour. Let x(t) denote the number of people that know about the rumour
after t days, and assume a logistic model, so that
50, 000
x = x(t) =
1 + Ke−kt
for some positive constants K and k. Find
(a) exact expressions for K and k,
(b) the number of days after which the rumour is being spread most rapidly, and
(c) the number of days after which 40, 000 people know about the rumour.
Solution: (a) We know that
50, 000 50, 000
10 = x(0) = 0
= ,
1 + Ke 1+K
so that 1 + K = 5000, so that K = 4999.
We also know that
50, 000
1, 000 = x(10) = .
1 + Ke−10k
Hence 1 + Ke−10k = 50, so that Ke−10k = 49, which becomes e10k = K
49
. Taking natural
logarithms and rearranging, we get
( ) ( )
ln K/49 ln 4999/49
k = = .
10 10
(b) The rumour is being spread most rapidly when
ln K 10 ln(4999)
t = = ( ) ≈ 18.4 ,
k ln 4999/49
which is after about 18 days (to the nearest day).
(c) We want t such that
50, 000
40, 000 =
1 + Ke−kt
so that 1 + Ke−kt = 54 , that is, Ke−kt = 14 , so that ekt = 4K. This becomes
ln(4K) 10 ln(4 × 4999)
t = = ( ) ≈ 21.4 ,
k ln 4999/49
which is after about 21 days (to the nearest day).

3
7. Suppose a yeast colony starts off with a population of 100 cells and is growing most rapidly
10 hours later, when it reaches 500 cells. Assuming logistic growth, find an upper bound
for the colony size, and how long it takes to reach 75% and then 90% of this maximum.
Solution: Let t be the number of hours and x = x(t) be the size of the colony, so that

M
x = ,
1 + Ke−kt
for some constants M, K, k > 0.
The population is growing most rapidly after t = 10 hours, at which time it reaches
500. This occurs at the point of inflection of the logistic curve, when the population has
reached half of its potential maximum, so that M2 = 500. Hence, an upper bound for the
colony size is M = 1, 000.
We have
M 1000
100 = x(0) = = ,
1 + Ke0 1+K
which quickly rearranges to give K = 9.
We also have
M 1000
500 = x(10) = −10k
= ,
1 + Ke 1 + 9e−10k
which quickly rearranges to give
1
e−10k = .
9
Hence e10k = 9, so that, taking natural logarithms and rearranging, we get
ln 9
k = .
10
1000
To reach 75% of the maximum, we need x = 750, so that 750 = , which
1 + 9e−kt
rearranges to give
1000 4
1 + 9e−kt = = .
750 3
Thus e−kt 1
= 27 , so that ekt = 27. Taking natural logarithms and rearranging, we get

ln(27) 10 ln(27) 10 ln(33 ) 10 × 3 ln 3

t = = = 2
= = 15 .
k ln 9 ln(3 ) 2 ln 3

1000
To reach 90% of the maximum, we need x = 900, so that 900 = , which
1 + 9e−kt
rearranges to give
1000 10
1 + 9e−kt = = .
900 9
Thus e−kt 1
= 81 , so that ekt = 81. Taking natural logarithms and rearranging, we get

ln(81) 10 ln(81) 10 ln(92 ) 10 × 2 ln 9

t = = = = = 20 .
k ln 9 ln 9 ln 9
Thus we expect to reach 75% of the maximum after 15 hours, and 90% after 20 hours.

4
8. We explain why the exponential growth model
dx
= kx ,
dt
for some constant k > 0, only has solutions of the form y = Aekt for some constant A > 0.
We assume throughout that x = x(t) is positive.
Proof: We multiply through by dt to get an equation involving differentials:

dx = kxdt

and then divide both sides by x to get

dx
= kdt .
x
We then integrate both sides, to get an equation linking antiderivatives:
∫ ∫
dx
= k dt = kt + C ,
x
for some constant of integration C. On the left-hand side, we integrate to get the natural
logarithm function, and can suppose that the constant of integration has been moved
across and absorbed in the constant on the right, and we get

ln x = kt + C .

Raising Euler’s number e to the power of each side of this equation, we get

x = eln x = ekt+C = eC ekt = Aekt ,

for some constant A, taking A = eC , completing the proof.

9. We explain why the logistic growth model

( )
dx x
= kx 1 − ,
dt M

for some constants k, M > 0, only has solutions of the form

M
x=
1 + Ke−kt
for some constant K > 0. We assume throughout that 0 < x < M .
Proof: We multiply through by dt to get an equation involving differentials:
( )
x
dx = kx 1 − dt
M
( )
and then divide both sides by x 1 − Mx
, putting everything involving x to the left-hand
side, to get
dx
( = kdt .
x 1− M x
)

5
 (This is an example of the technique of separating variables, which is explained in greater
generality in more advanced courses in calculus.) We then integrate both sides, to get an
equation linking antiderivatives:
∫ ∫
dx
( ) = k dt = kt + C , (1)
x 1− M x

for some constant of integration C. To unravel the left-hand side, we first observe that
1
1 1 1 1
( ) = + Mx = + .
x 1− x
M
x 1− M x M −x
It easy to check this equation holds. It is not so clear where it came from. It arises by
postulating a decomposition of the left-hand side as a sum of individual fractions with
denominators made up from each of the factors of the denominator of the left-hand side,
and then using a “guess and check” method. (A more formal, systematic approach uses
the method of partial fractions, explained in more advanced courses in calculus.)
We then get
∫ ∫ ∫ ( )
dx dx dx x
( ) = + = ln x − ln(M − x) + C = ln +C ,
x 1− M x
x M −x M −x
where the second antiderivative can be deduced quickly by a “guess and check” method
or using integration by substitution. Putting this together with (1), by combining the
constants of integration on the right, we get
( ) ( )
1 x
ln M = ln = kt + C .
x
−1 M −x
Raising Euler’s number e to the power of each side of this equation, we get
1
= ekt+C = eC ekt ,
M
x
− 1
so that, after reciprocating,
M
− 1 = e−C e−kt = Ke−kt ,
x
for some constant K, taking K = e−C , which, after rearranging, produces
M
x = ,
1 + Ke−kt
completing the proof.
10. It is worth noting, in the rule above for the logistic function, that when Ke−kt is large,
then it dominates the fraction, so that the addition of 1 becomes negligible, so that
( )
M M M kt
x = −kt
≈ −kt
= e .
1 + Ke Ke K
Thus, towards the left of the inflection point of the curve, the logistic function behaves
approximately as an exponential function.
This makes sense in terms of the model: for small x, the inhibition factor is close to zero,
and the logistic model is approximated well by the exponential growth model. As the
population increases, its growth rate is maximised when it reaches M2 , after which the
population continues to increase, but the growth rate decreases, and dwindles to a trickle
the closer the population gets to the maximum carrying capacity M .