Further Modelling

Jamie Frost
www.drfrost.org
@DrFrostMaths

Contact the resource team:

[email protected]
@DrFrostResource

Dr Frost Learning is a registered charity

Last modified: 5th April 2024 in England and Wales (no 1194954)
Teacher Notes
Prerequisite Knowledge
• Sine waves.
• Transformations of functions.
• Exponential functions and determining
and using two points on the curve.
• Log graphs.
• Piecewise functions with linear segments.

Key: Key Points Solution step – All slides include pedagogical

click to reveal detail in the ‘Notes’ section
! To be written for each slide.
in books Question/Discussion
Dr Frost Learning is a registered charity
Prompt in England and Wales (no 1194954)
How to use these slides
Though many slides in this resource will have titles specific to the topic, the slide titles in the table below
are used consistently within DFL resources for specific pedagogical purposes.
Any atypical use of a slide type, including any change of animation* or intended use, will be outlined in the
Teacher Notes for the slide.
Slide Title Explanation Default Animations*
To be used as a prior knowledge check or to review prerequisite
Recap Green click-to-reveal boxes.
knowledge. Can be used as a starter or as part of the main lesson.
To be used to highlight key concepts or theorems. This could include
the ‘why’ of the topic - including “real-life” contextual scenarios, or Usually in sequence with some
The Big Idea
putting into context of other mathematical concepts (past and green click-to-reveal boxes.
future).
Example To be modelled by the teacher. Solution animates in sequence.
Green click-to-reveal boxes.
Test Your To be completed by students and used for Assessment for Learning, For multi-step answers, reveal in
Understanding primarily using mini-whiteboards. parts or click final answer to
reveal full solution.
To be used as ‘Example’ &‘Test Your Understanding’ above, within the Example animates in sequence.
Example same slide to provide scaffold via visible modelled solution. Click the header to reveal TYU
Problem Pair TYU column is blank initially, to focus attention on example. question, then green click-to-
Reveal question by clicking ‘Test Your Understanding’ banner. reveal boxes.
Green click-to-reveal boxes. For
To be used as fluency practice. Multiple questions in rapid succession,
Quickfire multi-step answers, reveal in
for calculations that can be completed mentally. Often used for
Questions parts or click final line to reveal
shorter questions/ formulae or to isolate a small part of the method.
full solution.
To be used as a diagnostic question. Multiple choice questions, with
Multi-choice
plausible distractors, to allow teachers to diagnose misconceptions Arrow points to answer, on click.
Question
and errors in student thinking, then adapt their lesson accordingly.
Exam Question To be completed by teacher or student. Green click-to-reveal boxes.
Dr Frost Learning is a registered charity in England and Wales (no 1194954)
Recap of exponential modelling Show all
solutions

1 The two key properties of

exponential growth are:
𝑃 (rabbits) ( 5 , 1600 )
a) The value is multiplied by the
? after each
same scale factor
unit of time.
b) The rate of growth is
proportional to?the current
value, i.e.

( 2 , 200 )
Give some example scenarios where
2
we might use an exponential model:
(years)
• Population growth/decline
• Size of investment over time,
• ?
Radioactive decay. 3 A suitable model for the above
scenario would be:

𝒕
𝑷 =𝒂 𝒃 ?
Recap of exponential modelling Show all
solutions

4 Use the data to determine the best

values of the parameters and in
𝑃 (rabbits) ( 5 , 1600 )
Substitute the
data points into
the model.

2 ①
200=𝑎 𝑏 ②
3 ②÷①
8=𝑏 ? ( 2 , 200 )

200=𝑎 (2 ) 2 Substitute all

known values
back into one of
the equation.
(years)

5 If we had more than 2 data points with

Use your calculator’s statistics
the points only approximately following
a curve, how do we obtain the best
? exponential
mode to find the
regression line y=abx
possible model this time?
Alternative form of exponential models

𝑘𝑥
𝑦 =𝑎 𝑏
𝑥
𝑦=𝐴𝑒
It’s custom to use a capital
when as the coefficient of an
Why might we want to exponential term with base .
write in the second form?

is cleaner when differentiating, particularly

when finding the second derivative:

Why are two forms

In addition, when solving: equivalent?

the value emerges rather than a more arbitrary

base: As can be any arbitrary positive
? value, we could replace it with .
Note this assumes?that is positive
(which is a valid assumption of the
base of a power).
Application to half life
Radioactive substances decay over time, losing mass.

If the exponential model is , show that the half-life

𝑦 (kg) When
In general, for
𝑦0 𝑘( 0 )
Initial mass exponential
models, is the 𝑦 0= 𝐴 𝑒
initial value of .
The time it takes for the
substance to decay to When
1 half of its mass is known
𝑦 as its half-life, written as 1 𝑘𝑡
2 0 𝑦 = 𝑦0 𝑒 0.5

2 0
1 𝑘𝑡
=𝑒 0.5

half-life
𝑥
(years)
ln
1
2 ( )
=𝑘 𝑡 0.5

− ln ( 2 )=𝑘 𝑡 0.5
In general,
by laws of logs. ln ( 2 )
𝑡 0.5=−
𝑘
More general exponential models
The background radiation in Chernobyl is mSv per year.
A radioactive substance is introduced to a lab, raising the radiation level to mSv per year.
The half-life of the substance is minutes.
Estimate the radiation level in the room after minutes.

We may additionally allow horizontal translation Note that the level of radiation will
of the exponential curve, to allow for a non-zero decay in the same way as mass.
‘long-term’ value (when exponential decay).
𝑘𝑑 𝐴=150 − 100
𝑦 (kg) 𝑦 = 𝑑+ 𝐴 𝑒 represents the ln 2
radioactivity of the 𝑡 0.5=−
𝑘
𝐴 substance, with either
ln 2
its initial mass or initial 30 =−
level of radiation. 𝑘
ln 2
𝑘=−
ln 2 30
𝑡 0.5=−
𝑘 ¿ − 0.0231
− 0.0231𝑡
𝑅=100+50 𝑒
𝑦 =𝑑 When minutes:

𝑥
(years)
Test Your Understanding
6 A radioactive burger weighing g is placed in a non-radioactive
burger bun weighing g.
If the half-life of the burger is days, what is the total mass of the
‘food’ after days?

? Model
When days:

? Mass
Logarithmic models
Logarithmic models are suitable when
a scale factor increase in the input
(e.g. ) only leads to a constant
increase in the output (e.g. ).

𝑦 =𝑎 +𝑏 ln 𝑥 Amplitude

For example, if the wave amplitude of an

earthquake is times as great, its value on
E.g. the Richter scale only increases by .

Richter scale

𝑓 ( 10𝑥 ) =𝑎+𝑏ln ( 10𝑥)

What if the
input was
+𝟏
times
greater? This is a
constant (amplitude)
increase. 𝑎 10 𝑎
×𝟏𝟎
This is the opposite of exponential models.
Example
Determine a function to convert the
concentration of hydrogen ions to
the pH level .
Water, pH 7, has a H+ concentration
0 7 14 of M and an acid of pH 1 has a
concentration of M.

pH is a measure of the acidity of liquids.

The scale is inverse logarithmic: if the

concentration of hydrogen ions (H +) is
times smaller, the pH increases by .

In , will be negative if it is inverse

logarithmic.

Substitute given Solving:

information into Model is therefore:
our function.

Note that if we had used to reflect the

scaling behaviour of the pH scale, the function
would be a much simpler
Test Your Understanding

7 A ‘Pressupum’ is a logarithmic scale to

represent someone’s ability to do
press-ups in one go.

A student who can do press-ups has a

Pressupum Index of . A student who
can do pressups in one go has a ?
Solving simultaneously:
Pressupum Index of .

Determine a function to express the

Pressupum Index in terms of number
of press-ups .

To do directly on an fx-
Pedantry corner: the examples presented for
CG50 calculator, press logarithmic models (e.g. pH/Richter) are not
Statistics, enter your really ‘models’, more just ‘conversion functions’
for a single underlying measure.
values (pressing EXE after Formulas for models typically try to represent a
each one), click CALC, relationship between different physical
then REG, then LOG. quantities, and converting between different
scales is not doing that.
Sinusoidal models
We can use sinusoidal models for quantities which oscillate in waves.

Daily hours
of sunlight

Jun 21st Jun 21st

16
𝑦 =𝑎 sin ( 𝑏 ( 𝑥 − 𝑐 ) ) + 𝑑
Dec 21st

(Days after Dec 21st )

• Objects which travel in circles (e.g. height of a person on a Ferris

Where might wheel) and indirectly (e.g. tides, hours of sunlight throughout the year)
• Objects moving with Simple Harmonic motion, e.g. springs, where
we see these in acceleration, acting towards the centre, is proportional to displacement,
real life? i.e.
• Moving waves in physical phenomena, e.g. displacement of
electromagnetic waves, sound waves and ripples/waves in water.
Vertical transformations
Knowledge of graph transformations can help us to find the values of and in
the sinusoidal model by comparing to the graph of

Finding the value of

𝑦
Jun 21st Jun 21st
16
4
1 Mar 20th Dec 21st
𝜋 2𝜋 8
𝑦 =sin 𝑥 𝑦 =𝑎 sin ( 𝑏 ( 𝑥 − 𝑐 ) ) + 𝑑
(Days after
Dec 21st)

The is the scale factor of the stretch

parallel to the vertical axis.
Since the amplitude of is , will be the
amplitude of the transformed wave.

𝒂= 𝟒?
Vertical transformations
Knowledge of graph transformations can help us to find the values of and in
the sinusoidal model by comparing to the graph of

Finding the value of 𝑦

Jun 21st Jun 21st
16
Mar 20th Dec 21st
𝜋 2𝜋 8
𝑦 =sin 𝑥 𝑦 =𝑎 sin ( 𝑏 ( 𝑥 − 𝑐 ) ) + 𝑑
(Days after
Dec 21st)

The midlines make clear

the vertical translation of The is translation in
the graph. the vertical direction.

𝒅=𝟏𝟐
?
Horizontal transformations
Knowledge of graph transformations can help us to find the values of and in
the sinusoidal model by comparing to the graph of

Finding the value of

𝑦
Jun 21st Jun 21st

Mar 20th Dec 21st

𝜋 2𝜋
𝑦 =sin 𝑥 𝑦 =𝑎 sin ( 𝑏 ( 𝑥 − 𝑐 ) ) + 𝑑
2 ?𝜋 91 (Days after
Dec 21st)

365
?
Recall that
The period of a sine wave is the results in a
width of one full oscillation. stretch parallel to
the -axis by scale
factor

𝟐𝝅
𝒃= ?
𝟑𝟔𝟓
Horizontal transformations
Knowledge of graph transformations can help us to find the values of and in
the sinusoidal model by comparing to the graph of

Finding the value of 𝑦

Jun 21st Jun 21st

Mar 20th Dec 21st

𝜋 2𝜋
𝑦 =sin 𝑥 𝑦 =𝑎 sin ( 𝑏 ( 𝑥 − 𝑐 ) ) + 𝑑
91 (Days after
Dec 21st)

The translates the graph

Uses the start of the wave, on horizontally by .
the midline, to determine . This is known as the phase shift.

𝒄 =𝟗𝟏
?

Final equation:
?
Summary of sinusoidal models

! For a sinusoidal model

• is the amplitude
• is the period
• is the phase shift
• is the equation of the midline

Why do we use rather than ?

The easy way to get the transformations involved

and their order is to consider how to get from
𝑏 ( 𝑥 −𝑐 ) 𝑏𝑥 −𝑐 the expression inside the function back to .

÷𝑏 +𝑐
e.g. to the values, i.e.
𝑥 −𝑐 𝑏𝑥 translation by
+𝑐 ÷𝑏
𝑥 𝑥 As this does the translation first, and the
values are subsequently divided by , this
translation effectively becomes , meaning
This does the scaling first, which means the phase shift will be not .
the subsequent horizontal translation (and
thus phase shift) actually will be by .
Example Test Your Understanding
A Ferris wheel has a period of minutes. 8 As a diver jumps into the water, the
A seat starts m of the ground and the top of depth of the water can initially be
the wheel is m of the ground. modelled as a sinusoidal function.
The Ferris wheel is rotating and a seat is The depth varies from m to m and takes
level with the centre of the wheel after seconds to go from minimum to
minutes. maximum depth.
Assuming the height of the seat can be The water starts from rest but displaces
modelled with sinusoidal motion, determine upwards second after the diver hits the
a model for the height (in metres) in terms water.
of the time a Determine a model for the depth of the
(in minutes) after moving. water in terms of the time after the
diver hits the water.
b Identify a weakness in this model.

3 secs is just half

the full wave.
The amplitude will be
the radius of the wheel.

?
( )
0 amplitude is the
𝝅
centre of the wheel. 𝒚 =𝟎 .𝟐𝟓 𝐬𝐢𝐧 ( 𝒙 − 𝟏 ) +𝟓 .𝟐𝟓
𝟑
For the first second the water will be at rest and so
will not follow the model.
A ‘piecewise’ model could handle this.
Logistic models
Consider bacteria growing in a petri dish.
What will happen:
(a) initially?
(b) after a longer period of time?

Initially, like normal populations, the bacteria

will grow exponentially.
Eventually, the population? will be limited by the
space available in the petri dish, so will converge
to a limit.
Population ()
𝑃= 𝐿

is known as the
carrying capacity
? Graph

Time ()
Deeper Understanding
We want the gradient to be small if either:
(a) the existing population is small or
(b) the remaining capacity is small.

Can you think of a suitable differential equation?

𝒅𝑷
∝ 𝑷 (? 𝑳 − 𝑷 )
𝒅𝒕
Notice that will
be small if either
Population () is small or is close
to .
𝑃= 𝐿

is small, so the Any function with this ‘S’

gradient will again be shape and a limiting value is
small. known as a sigmoid function.
The logistic function is a
specific type that satisfies
this differential equation.

Time ()
is small, so the gradient
will be small.
Solving the differential equation
𝑑𝑃
∝ 𝑃 ( 𝐿− 𝑃 )
𝑑𝑡
𝑑𝑃
= 𝑘𝑃 ( 𝐿 − 𝑃 )
𝑑𝑡
1 𝑑𝑃 Get everything related to left of
= 𝑘 equation, and anything related
𝑃 ( 𝐿 − 𝑃 ) 𝑑𝑡 (which there isn’t) to right.
1 𝑑𝑃
∫ 𝑃 ( 𝐿− 𝑃 ) 𝑑𝑡 𝑑𝑡 =∫ 𝑘 𝑑𝑡
Integrate both sides.
1
∫ 𝑃 ( 𝐿− 𝑃 ) 𝑑𝑃=∫ 𝑘 𝑑𝑡
(
∫ 1/𝑃𝐿 + 𝐿−1/ 𝐿
𝑃 )
𝑑𝑃=∫ 𝑘 𝑑𝑡 Split into partial fractions
to make integratable.
1
𝐿
1
( +
1
)
∫ 𝑃 𝐿− 𝑃 𝑑𝑃 =∫ 𝑘 𝑑𝑡 Tidy up.
1
¿ Do the integration.
𝐿
ln | 𝑃
𝐿− 𝑃 |
= 𝐿𝑘𝑡 +𝑐 ′ Remove the . Typically would be
written as , where is a positive
𝑃 𝐿𝑘𝑡 +𝑐
′
𝐿𝑘𝑡 𝑐
′
𝑘 𝑡
′
constant.
=𝑒 =𝑒 𝑒 = 𝐴𝑒
𝐿− 𝑃 is itself a generic constant so
replaced with .
Solving the differential equation
Continuing…
𝑃 𝑘 𝑡
′

=𝐴𝑒
𝐿− 𝑃 For convenience we’ve
𝑘𝑡 used to mean .
𝑃 = ( 𝐿 − 𝑃 ) 𝐴𝑒 Strategy now is to make
𝑘𝑡 𝑘𝑡 the subject.
𝑃= 𝐴𝐿𝑒 − 𝐴𝑃 𝑒
𝑃 + 𝐴𝑃 𝑒 𝑘𝑡 = 𝐴𝐿 𝑒 𝑘𝑡
isolated on one
side so it can be
factorised out.
𝑃 ( 1+ 𝐴 𝑒𝑘𝑡 )= 𝐴𝐿𝑒 𝑘𝑡
𝐴𝐿 𝑒 𝑘𝑡
𝑃= 𝑘𝑡
1+ 𝐴 𝑒
𝐿
𝑃= − 𝑘𝑡
1 + 𝐶𝑒 ÷ 𝐴𝑒 𝑘𝑡
Key Points
! The logistic model is
𝑦
𝑦 =𝐿

where is the carrying capacity (maximum population) and

controls how fast the population grows.
It is useful for initially exponential growing values which have a 𝑥
maximum capacity.

Test Your Understanding:

A population of foxes in an urban area is modelled
using where is in years.

1 What is the initial population of foxes?

foxes ?
2 What is the longer-term limit on the number of foxes? ?
3 How many foxes are there after years?

foxes ?
Example Test Your Understanding
There are initially foxes in an urban 9 The population of Galapagos tortoises
area. The area can’t support more than on the Galapagos islands is initially ,
foxes. After one year there are foxes. but is limited to . After years the
Use logistic regression to estimate the population of tortoises is .
number of foxes after years.
𝐿 a Use logistic regression to estimate the
𝑦= − 𝑘𝑥
number of tortoises after years.
1 +𝐶 𝑒 b After how many years will the
When , When population be ?

When
tortoises

b
After 3 years: ?
Solving: years.

To do the regression directly on a calculator, use a large value for for the limiting value (here 10000)
Further applications of logistic functions

Probability of
passing
1 Where the limiting value is ,
logistic functions are useful for
modelling the probability of
anything binary (two outcomes),
e.g. ‘passing an exam’ or ‘female
vs not female’.

Hours revised (

We can then use logistic regression to get the

Each data point might represent parameters of a function (where the allows
a student, where is the hours horizontal translation) to model the probability
revised for an exam, and if they of a student passing an exam.
passed, and 0 otherwise.
Application to neural networks
A neural network is a way to ‘learn’ a complex function.

For example , taking an input of a photo, and outputting a desirable value such as:
• the probability the photo is of a giraffe (known as a ‘binary classifier’, i.e. giraffe or not)
• a numerical value corresponding to the identified animal (1 = giraffe, 2 = lion, …)
Input

Hidden
Output

p(giraffe)
Turn into a
= 0.87
numerical ‘feature
vector’
A (feed-forward) neutral
network consists of connected
‘neurons’, modelling how
We somehow turn the input into a list neurons work in the brain.
of numbers, representing features of
the input. This might be the intensity x1
of individual pixels in an image. f(a0 + a1x1 + a2x2 + a3x3)
x2
x3

We take some weighted linear combination of the inputs using some

parameters (). This is then fed through an activation function. We
would typically use some sigmoid function (i.e. ‘S’ shape) such as a
logistic function (range: 0 to 1) or (range: -1 to 1)
Application to neural networks
x1
The would allow us to represent a boundary (in 2D a
f(a0 + a1x1 + a2x2 + a3x3) line, in 3D a plane), so that in 2D for example a negative
x2 or positive value could act as a ‘binary classifier’, i.e.
distinguishing between two things or ‘yes’ and ‘no’.
x3

As we add extra ‘layers’ of

𝑎 neurons (we might have one
Height () 0 +𝑎 Height () hidden layer, or more, which is
1 𝑥 classed as ‘deep learning’),
1 +𝑎 C 𝑝 ( 𝑐𝑎𝑡 ) ≈ 1 we can represent more
C
D 2 𝑥 C
C C
C
C
complex boundaries, with the
=D
D C
2 activation functions allowing
D C 0 C
C D C
D C
C for nice interpolation to get a
D C
D D D C probabilistic output rather
D C C D C C
than a hard ‘yes’ or ‘no’.
C
𝑝 ( 𝑐𝑎𝑡 ) ≈ 1
D D C
D D
𝑝 ( 𝑐𝑎𝑡 ) ≈ 0
D
𝑝 ( 𝑐𝑎𝑡 )=0.5 D

Weight
𝑝 ( 𝑐𝑎𝑡 ) ≈ 0 Weight

Using training data (e.g. pictures we know are

Note that for the logistic function , , so a point on the boundary would end
giraffes and not giraffes), we use regression to
up with an output of 0.5.
learn the weights in each neuron. This is known
The point of the activation function is to exploit the ‘S’ shape to get a ‘fuzzy
as supervised learning, because the training
boundary’, where the boundary may be a probability of 0.5 (i.e. ambiguous
data includes the desired output.
what to output) but we want the probability to be close to 0 or 1 (i.e. clearly
decisive) if we are not close to the boundary. has a similar effect but 0 stays
as 0 and values that stray too far from 0 become -1 or 1.
Piecewise model

We have previously seen piecewise models as a means to have different

functions for different parts of the domain.

The function should be continuous, so

𝑦 (3 ,7) the output at the ends of different
parts of the function should match up.

( 1 , 3)

𝑥
10

{
2 If it’s a quadratic with turning
𝑥 2𝑥+4
?−Suggest 0 ≤ 𝑥< 3 point , the equation would be of
𝑓 (𝑥 )= the form
10 −𝑥
? Suggest 𝑥 ≥3
Using either or allows us to
determine that .
Example Test Your Understanding
Let 10 Let

Determine the value of required to make Determine the value of required to

the function continuous. make the function continuous.
a
Determine the range of the function.
b Solve
c
The graph segments must join at
a
?

b
Range is ?