Industrial Mathematics

Week 1 – Modelling
Dr Ben Goddard
b.goddard@ed.ac.uk
24th September 2021

This worksheet is designed to introduce you to the process of modelling. We are going to
work through the two questions in groups during the Workshop. The aim is not necessarily to
come up with a solution to the model (in fact, this is likely not remotely realistic in the time
available), rather it is to think about how to formulate a model that includes the important
information for a system, and which is likely to be simple enough to solve either analytically
or, more reasonably, numerically.
We will discuss approaches to the numerical solution of advection-diffusion equations next
week (although you may have already seen some of this, or related, material in other courses).
We will also cover the numerical solution of ‘agent-based models’, which consist of a number
of ‘agents’ (people, animals, etc) that interact in a certain way. The former are ‘continuous’
models, meaning that the solution lies on the whole of a given time-space domain; given
any x in space and t in time, the model predicts the true values. In contrast, agent-based
models are ‘discrete’, often both in space and time. Such a model could give, for example,
the positions of a group of people in a room at a particular time.
There are connections between these types of models. For example, if there were many
people in the room, then it may be reasonable to construct a ‘density’ of people (e.g., number
of people per square meter), and write down a partial differential equation for this. There
are many ways to go from a discrete model to a continuous one, and this is a frequently-used
tactic to reduce the computational requirements.
One of the things you’re going to be asked to think about in the two questions below
is what sort of model is most appropriate, both from a ‘physical’ perspective, and from
obtaining something that is computationally tractable.
You may want to remind yourself of the modelling process in Video 1.2 and the Week 1
synchronous activity. In this Workshop we’re going to be focussing on the first two steps,
namely how to formulate a ‘real’ model for a problem (or at least identify what is important),
and how to write down a mathematical version of the model. We’re going to postpone solving
the model and comparing to real data until later, if at all, for these examples.
We will be looking at two modelling problems:

1. The rise and fall of the Roman Empire;

2. Determining the optimal depth of a wine cellar.

1
 The former may not seem like an industrial problem, and it’s likely not. (There may
be industries (e.g. government agencies) who would like to be able to model this sort of
thing, but that’s not really the point here.) The reason it is included is to get you to think
about a problem which is outside your comfort area and perhaps not treated very widely in
undergraduate mathematics degrees. You’re probably used to seeing problems from biology,
engineering, or physics, but there’s absolutely no reason why such things can’t be modelled
mathematically.
The latter is a key example for the course. As mentioned in the Lecture, we are going
to work through the various stages of the modelling process, from an industrial perspective,
including how to formulate a model, solve a simplified version analytically to get some insight
into the solution, solve more complex versions numerically, include real-world data, interpret
the results, and write a report on the work.
1: The Rise and Fall of the Roman Empire
We’re going to go through a few steps of the modelling process. The idea is to discuss the
steps in groups, with the aid of your tutors, and try to produce at least an outline of the
sort of model you think may be useful (as well as identifying models which are less useful!).
Don’t be too concerned about getting the ‘right’ answer – there likely isn’t one! There are
also very few truly ‘wrong’ suggestions; it’s often hard to tell what will be important in a
model before trying it out. So, please discuss it in a friendly manner, take each others’ ideas
on board, and try to piece together a model.

Overall aim: Develop a mathematical model for the extent of the Roman Empire over time.

Suggested steps:

1. Watch the video at https://www.youtube.com/watch?v=WaxeyHAOTWA, which shows

the extent of the Roman Empire over time, as well as highlighting what are thought
to be some (but likely not all!) key points in its rise and fall. Feel free to also do some
reading online if you think this will help. The steps marked with a ∗ are things that
we should do when modelling, but you likely won’t get to them today.

2. Discuss what factors affect the development of an empire. At this stage you should
write down anything that may be important; we’ll narrow things down in the next
step.

3. Try to identify which factors are most important, which are likely unimportant, and
any that you’re not sure about.

4. Choose one or two of these factors and try to explain in words how you think they
should affect the extent of an empire over time.

5. Try to write down at least the outline of a mathematical model for what you described
in the previous step.

6. Discuss how computationally tractable you think this model will be. For example, how
many dimensions does it have; which parameters would you need to identify values for?

2
 * Obtain some real-world data for the parameters in your model.
* Solve the model (likely numerically?).
* Determine if you can say anything general about empires using your model. Is there
anything missing that you think should be included at the next iteration of the mod-
elling process?

2: Optimal Placement of a Wine Cellar

This exercise, again in groups, is going to get you to think about how to formulate a mathe-
matical model given some insight to the problem from an industrial client. The key things to
remember are (a) what the client wants to know/be able to do with the model; (b) that you
need to be able to solve your model either analytically or numerically; (c) you should include
the things that the client thinks are important, even if you have to approximate them, or to
show that they aren’t important after all.

In this case, we’re going to take a slightly more directed approach in what sort of model
we’re going to choose, and focus more on the nitty-gritty of writing down the equations and
trying to identify some of the parameters.

Industry Problem Description: Optimal Wine Cellars Ltd install and maintain wine
cellars for clients who wish to age their wine over the course of years or decades. A wine cellar
is a storage room, built underground, used to control important factors such as humidity and
temperature. There are two main types of wine cellar: active and passive. In an active wine
cellar, the desired conditions are maintained by a climate control system. In contrast, passive
cellars are built underground with the aim or reducing temperature variations over time;
Optimal Wine Cellars Ltd specialise in these passive cellars. They are principally interested in
how the depth of a wine cellar affects the temperature over the course of a year. It is believed
that the main cause of these variations is the daily and seasonal temperature changes on the
surface, which propagate down to the wine cellar. The extent of the temperature variations
is known to depend on the depth of the cellar as well as the type of soil or other material it
is built into.
There are two main criteria for an ‘optimal’ wine cellar:
• When considering daily variations of ±5◦ from the mean, and/or seasonal variations of
±15◦ , the temperature in the wine cellar should remain within ±2◦ of the mean;
• The temperature variation should not only be small, but also perfectly out of phase
with the seasonal variations (i.e., the cellar is colder in summer and warmer in winter).
[Note that this is actually the problem that inspired Joseph Fourier to invent Fourier
series.]

Suggested steps:
1. If you haven’t already watched the video introduction to advection-diffusion equations,
and/or read some of the associated literature (see Reading List) then now would be a
good time to at least watch the video.

3
2. Write down an advection-diffusion equation which may be a useful model. The key
things to think about are the domain, boundary conditions, parameters in the model,
and likely the initial condition, although the latter may be a bit harder.
In particular:

• What sort of domain should you consider? Can you reduce it to a 1D problem?
One boundary is likely to be the surface of the earth, but what should we do about
the other boundary?
• What should the boundary conditions be? How can we encode the daily and
season changes in temperature?
• What are sensible choices for the initial conditions? How much do you expect
them to matter?
• What parameters are important? For example, is there anything driving advection
in the system?
• What do the parameters in your model mean physically? Can you find suitable
values for them (e.g., on the internet)?
• Is it possible to solve the model analytically? If not, can you make some reasonable
approximations so that it is? What would you hope to be able to see in these
analytical solutions?
• What could you do to include real-world data, such as temperatures, in the model?
• How would you go about determining how ‘sensitive’ the model is to the parameter
choices? For example, if I change a parameter by a few percent, does this have a
small or large effect on the output of the model?