Alfio Quarteroni

Modeling
Reality with
Mathematics
Modeling Reality with Mathematics
 Alfio Quarteroni

Modeling Reality
with
Mathematics
Alfio Quarteroni
École Polytechnique Fédérale de Lausanne (EPFL)
Lausanne, Switzerland
Politecnico di Milano
Milan, Italy

Translated by
Simon G. Chiossi
Departamento de Matemática Aplicada
Universidade Federal Fluminense
Niterói, Rio de Janeiro, Brazil

Translation from the Italian language edition: “Le equazioni del cuore, della
pioggia e delle vele. Modelli matematici per simulare la realtà” by Alfio
Quarteroni, © Zanichelli 2020. Published by Zanichelli. All Rights Reserved.

ISBN 978-3-030-96161-9    ISBN 978-3-030-96162-6 (eBook)

https://doi.org/10.1007/978-3-030-96162-6

© The Editor(s) (if applicable) and The Author(s), under exclusive licence to Springer
Nature Switzerland AG 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the
Publisher, whether the whole or part of the material is concerned, specifically the
rights of reprinting, reuse of illustrations, recitation, broadcasting, reproduction on
microfilms or in any other physical way, and transmission or information storage and
retrieval, electronic adaptation, computer software, or by similar or dissimilar meth-
odology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc.
in this publication does not imply, even in the absence of a specific statement, that
such names are exempt from the relevant protective laws and regulations and therefore
free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and
information in this book are believed to be true and accurate at the date of publica-
tion. Neither the publisher nor the authors or the editors give a warranty, expressed or
implied, with respect to the material contained herein or for any errors or omissions
that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Cover illustration: Topping Heart ©Adobe Stock | #30115349

This Springer imprint is published by the registered company Springer Nature

Switzerland AG.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
to Lara, Luca, and Bianca Sofia,
my little models
 Preface

Following the dramatic spread of the COVID-19 pandemic

at the start of 2020, mathematics has never been so much at
the centre of everybody’s attention.
Expressions like exponential, logistic function, extrema and
inflection, which until that moment we thought were con-
fined to classrooms and university halls, all of a sudden have
entered the political debate. This was all the more true in
the initial stages of the epidemic, which were characterised
by a major uncertainty regarding the contagion’s evolution.
Yet even a casual observer must have noticed that math-
ematics is present in an increasingly pervasive way in our
daily lives. There is chatter in the media about intangible
algorithms for finding a soulmate, concocting the perfect
diet or taking any decision whatsoever. The news report of
billions of euros going up in flames on stock exchanges
around the world due to an algorithm that went amok.
They boast how big data (which everybody talks about, al-
though very few truly know about) is essential for economic
and technological advancement.

vii
viii Preface

In this book, as you may have gathered from the title, I

intend to present a version of mathematics that is less hos-
tile and obscure. Or rather, I will present an area, called
mathematical modelling, that over the last few decades has
taken the front row. Whether we are aware of it or not (as is
more often the case), we all regularly benefit from mathe-
matical models and algorithms, and without them our lives
would be very different. Here are some examples. Without
mathematical models we would not know what the weather
will look like tomorrow. We would not be able to share
photos and videos on our phones, nor browse the web as
fast as we do. We could not blindly rely on sat-navs to find
our way through cities we have never been to before. Our
cars would not be so silent, comfortable and efficient. We
could not use CAT scans to take a look inside our bodies,
nor would our favourite football team have a legion of
match analysts studying strategies to boost competitiveness
using the omnipresent big data. The list of similar stunning
achievements could go on and on.
Mathematics is—even if you do not know, it is easy to
guess so—an abstract science. And that is precisely one of
its secret weapons. Abstraction allows us to study problems
in their full generality, and helps us understand their key
and innermost features. Abstraction tickles our imagination
and imagination nurtures creativity, which in turn allows us
to discover the best path towards solving our problems. We
rely on abstraction to formulate far-fetching conjectures,
which sometimes are so intricate that they withstand our
attempts to solve (or disprove) them for centuries. Fermat’s
last theorem, for instance, was stated in 1637 and only
proved by the British mathematician Andrew Wiles in
1994. The Riemann hypothesis was first published in 1859
and is still unsolved.
 Preface ix

Mathematical models are characterised by being con-

crete, because they need to be both useful and of broad in-
terest. As strange as it may sound, they are born and they
flourish precisely because of abstraction. To better under-
stand this crucial fact and reveal it in all its might, in the
next pages we shall attempt to clarify what a mathematical
model truly is, and present some examples.
Before that, though, we should recall the pivotal role of
two core players in mathematics, which all of us have met
in school, namely numbers and equations. Numbers allow
us to quantify distances, weights, time intervals and so on.
Equations describe in a general way the relationships that
govern natural processes (the movement of a glacier, the
flooding caused by a river’s surge, the propagation of seis-
mic waves but also solar combustion and even how a forest
grows). Equations allow us to compute the trajectories of
satellites and the courses of Formula 1 racing cars; they
manage industrial processes and regulate the negotiation of
complex financial instruments. Because of equations we are
able to produce wonderful animated films (whose charac-
ters and events, albeit amazingly realistic, are solutions to
mathematical equations), we can study the best tactic for a
volleyball team and we can even simulate how vital organs
like the heart or the brain work. All of this is possible be-
cause equations translate into symbols and mathematical
relations physical laws, biological processes and social be-
haviours. In this way, they allow us to construct a virtual
world (the model) starting from the real world. By solving
equations we can make predictions (think of weather fore-
casts), simulate the progression of an illness or calculate
how a volcanic eruption will disperse its ashes in the atmo-
sphere. Summing up, a mathematical model’s equations are
telescopes pointed into the future.
x Preface

Mathematics is the oldest among the sciences (the earli-

est written documents were found in ancient Egypt and
Mesopotamia roughly 2000 BCE), and we can bet on it
being the one that will survive them all. Mathematics is the
only field capable of developing autonomously, since all
other sciences require the language and the tools of mathe-
matics to express themselves. We had better acknowledge
its role then. If you are willing to accompany me on this
journey, I intend to help you look at mathematics through
a different lens. Hopefully, I will awaken (once more?) the
interest of those of you who “In school I did understand
maths, but there was a teacher who made me hate it.” (In
everyone’s life there is always a teacher who made us love or
hate a subject.)
So let the adventure begin: enjoy!

Milan, Italy, 2022 Alfio Quarteroni

 Acknowledgements

I wish to thank my colleagues:

• Luca Bonaventura for the chapter WEATHER FORE-

CAST MODELS
• Luca Dede’ and Christian Vergara for the chapter A
MATHEMATICAL HEART
• Nicola Parolini for the chapter MATHEMATICS
IN THE WIND
• Gilles Fourestey, Nicola Parolini, Christophe Prud-
homme and Gianluigi Rozza for the chapter FLYING
ON SUN POWER
• Luca Paglieri for the chapter THE TASTE FOR
MATHEMATICS.

A special thank to Francesca Bonadei from Springer

Milan: her help and support are greatly appreciated.

xi
 Contents

1 
The Model, aka the Magic Box ������������������������������������  1
1.1 Initial Data��������������������������������������������������������������  5
1.2 Approximate in Order to Solve������������������������������  5
1.3 How Many Models for One Problem? How
Many Problems for a Single Model?����������������������  8
1.4 The Phases of a Model��������������������������������������������  9
2 Weather Forecast Models���������������������������������������������� 11
2.1 A Model Based on … Thin Air������������������������������ 12
2.2 The Physical Quantities Relevant to
Meteorology������������������������������������������������������������ 14
2.3 Physics Comes to the Rescue���������������������������������� 15
2.4 The Initial Data and the Boundary Data ���������������� 18
2.5 Numerical Models, from D-Day to
von Neumann ������������������������������������������������������ 19
2.6 Increasingly Sophisticated Models: Lorenz’s
Butterflies���������������������������������������������������������������� 22
2.7 Weather Forecasting Today������������������������������������ 25
References������������������������������������������������������������������������ 26
3 
Epidemics: The Mathematics of Contagion���������������� 29
3.1 Preys and Predators������������������������������������������������ 30
3.2 The Epidemiological Models���������������������������������� 32

xiii
xiv Contents

3.3 The Population’s Critical Size: The Case of

Measles ������������������������������������������������������������������ 33
3.4 A World of Susceptible Individuals������������������������ 35
3.5 The Equations of the Contagion ���������������������������� 38
3.6 The Peak, the Plateau, the Breakneck Slopes
(and the Climb-Ups) ���������������������������������������������� 43
References������������������������������������������������������������������������ 45
4 A Mathematical Heart�������������������������������������������������� 47
4.1 How Does the Cardiovascular System Work?
An Eternal Challenge for Philosophers,
Doctors, and Mathematicians���������������������������������� 49
4.2 The Models Today�������������������������������������������������� 51
4.3 Mathematics in the Operating Room���������������������� 55
4.4 The Blood’s Equations�������������������������������������������� 57
4.5 At the Heart of the Problem������������������������������������ 60
References������������������������������������������������������������������������ 65
5 Mathematics in the Wind���������������������������������������������� 67
5.1 A Sports Trophy with a Glorious History �������������� 68
5.2 The Swiss Outsider and the Mathematics
of Sails�������������������������������������������������������������������� 70
5.3 The Numerical Simulations������������������������������������ 79
5.4 How Did It End Up? ���������������������������������������������� 83
References������������������������������������������������������������������������ 84
6 Flying
 on Sun Power������������������������������������������������������ 85
6.1 The Piccards, a Family of Explorers���������������������� 87
6.2 Ending the Fossil-Fuel Era�������������������������������������� 89
6.3 The Solar Impulse Mission: The Challenges���������� 91
6.4 Mathematics Comes into Play�������������������������������� 93
6.5 Multidisciplinary Optimisation������������������������������ 97
6.6 An Example of Multi-objective Optimisation�������� 99
References������������������������������������������������������������������������102
7 The Taste for Mathematics��������������������������������������������103
7.1 Food Preparation����������������������������������������������������104
7.2 Mathematics and the Brain ������������������������������������104
7.3 The “Formula of Flavour”��������������������������������������107
 Contents xv

7.4 Optimising the Industrial Production of Food��������109

7.5 Mathematical Packaging ����������������������������������������111
7.6 Mathematics and Health ����������������������������������������113
References������������������������������������������������������������������������114
8 The Future Awaiting Us������������������������������������������������117
Reference ������������������������������������������������������������������������123
 List of Figures

Fig. 1.1 The riverbed divided into a collection of tiny

cubes of the same size 7
Fig. 2.1 Lorenz’s butterfly-shaped diagram. Photo:
zentilia/Shutterstock24
Fig. 3.1 The behaviour of the populations of prey and
predators in time, according to the
Lotka-Volterra model 32
Fig. 3.2 The spread of the contagion for an epidemic
with R0 = 2 37
Fig. 3.3 A patient’s symptomatic and contagious trend
in time 40
Fig. 3.4 The SUIHTER epidemiological model:
S = Susceptibles, U = Undetected, I = Isolating
at home, R = Recovered, H = Hospitalised,
T = threat of life, E = expired 41
Fig. 3.5 The contagion curves with no intervention
(blue) and containment measures (red).
The integral between 0 and t, that is, the
shaded area below each curve, provides the
total number of people infected from day 0
to day t43

xvii
xviii List of Figures

Fig. 3.6 Epidemic trend in Italy for the compartments

hospitalised, daily positives, and hosted in ICUs
(SUIHTER model) 45
Fig. 4.1 Mathematical simulation of an artery’s
deformation52
Fig. 4.2 Blood pulse inside the carotid 53
Fig. 4.3 Mathematical simulation of the effects of a
stent [insertion] on the elasticity and
stiffness properties of the arterial wall 56
Fig. 4.4 Drug-eluting stent simulation 57
Fig. 4.5 A conceptual map that illustrates the different
physical processes characterising the
functioning of the left ventricle and their
interactions62
Fig. 4.6 A mathematical model of the heart’s activity 63
Fig. 5.1 Team Alinghi. 2003 America’s Cup, Auckland,
New Zealand. Picture: ChameleonsEye/
Shutterstock72
Fig. 5.2 Aerial and submerged components in a
sailboat of America’s Cup class 72
Fig. 5.3 Detail of the surface grid on the submerged
part of the vessel 75
Fig. 5.4 Forces and momenta acting on the boat 76
Fig. 5.5 Flow lines around and past the bulb 79
Fig. 5.6 Pressure distribution on the surface of the
appendices and flow lines around the winglets 80
Fig. 5.7 Pressure distribution on the two boats in
downwind leg 82
Fig. 5.8 Aerodynamical interaction of two boats sailing
downwind82
Fig. 6.1 Solar Impulse, a project for the first
solar-powered circumnavigation of the globe.
Photo: Frederic Legrand—COMEO/
Shutterstock86
Fig. 6.2 The Breitling Orbiter 3. Foto: Cedric
Favero / VWPics / Alamy Foto Stock 89
Fig. 6.3 Example of computational mesh 95
 List of Figures xix

Fig. 6.4 Two examples of 3D airflow simulations

around the aircraft 96
Fig. 6.5 Two Pareto optimal solutions A and B 99
Fig. 6.6 The problem’s Pareto front 100
Fig. 6.7 Flight profile corresponding to the first
endpoint of the Pareto front 101
Fig. 6.8 Flight profile corresponding to the second
endpoint of the Pareto front 102
Fig. 7.1 When a food product encounters mathematics 105
 1
The Model, aka the Magic Box

Abstract A mathematical model works like a box. The re-

searcher puts in the data, the observations and the measure-
ments, and the box returns the solutions to the model, that
is, the expected values of the physical quantities describing
the phenomenon.

In the past few days it has rained a lot and the river is full of
water. How do we predict whether it will burst and flood
the town centre?
There is no time to waste and we must act swiftly. We
cannot study the river’s behaviour for hours and try and
guess what will happen. Neither can we base ourselves on
similar past events, because floods are not a phenomenon
that repeats itself cyclically and regularly. There are many
random circumstances making every occurrence different
from the previous ones.
We must know in advance the water’s level at various
points along the river to understand if and when it will

© The Author(s), under exclusive license to Springer Nature 1

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_1
2 A. Quarteroni

overflow. We need, in other terms, a sensible and fast pre-

diction. Can we rely on mathematics for this? And how,
exactly?
Let us think of the case at hand, the river’s tipping point.
We will need to know the following set of data:

• the shape of the river bed;

• the flow rate outside the city;
• how rough the surface of bed and banks is.

Now, before we start formulating the model’s equations, we

should determine which variables (the problem’s solutions,
or unknowns) describe in the most comprehensive way the
process we want to examine. In our case we would like the
model to provide the following (unknown) quantities:

• the speed of the flood wave at every point along the river
course and at every instant in the successive hours;
• the water pressure at each point and at each instant of
time (in particular, the pressure on the river banks);
• the shape of the wave’s surface, which in turn will give us
the water level reached at each point of the basin, at each
time. (From that we will deduce if and when the river
will burst.)

These solutions will be expressible in terms of mathematical

functions, which are laws associating to each point in space
and to each instant the numerical value of the following
quantities:

• the speed (measured in metres per second);

• the pressure (in newtons);
• the elevation of the water crest (in metres).
 1 The Model, aka the Magic Box 3

(Note that if we had considered a water pipe instead of a

river, like those downstream mountain dams, the unknowns
would be the speed and the pressure. It would not make
sense to consider the elevation, since one could assume that
in a full pipe the water adheres to the inner walls.)
Now that we have established the data and the unknown
variables, the model describing the physical phenomenon
(the water flow, in our case) should allow us to pass from
the data to the solutions. It does so thanks to equations,
which are equalities between algebraic expressions that
mathematically encode the relevant physical laws (fluid dy-
namics, in particular of water) and link the unknown quan-
tities (the solutions) to the known ones (the data).
Often, physical laws are based on universal principles,
such as conservation laws. Let us consider for instance the
conservation of mass. In our case it translates into the con-
servation of the total quantity of water contained in a cer-
tain portion of the river basin. Consider the water mass
contained in a 100 m stretch at 1 pm. It will equal the mass
present at 12 pm, plus the water that flowed in upstream,
minus the water that flowed out either downstream or from
the banks in case of an outburst, during this hour. Or let us
consider the conservation of force: in order to respect it, we
will have to impose that the product of the water mass times
its acceleration equals the sum of all forces acting on
the water.
And that is not the end: other laws will impose further
relationships/equations.
The real challenge is to express these laws on an infini-
tesimal volume of fluid, which in a limiting process be-
comes a point. Mathematically speaking, in fact, these laws
describe the rate of change of the quantities involved in
space and time. More precisely, they describe infinitesimal
variations, meaning variations that are as small as we want.
4 A. Quarteroni

A variation is just a difference: when we fall ill and our body

temperature rises from 37 to 38 °C we say there has been a
variation of one degree. If our firm’s revenue last year
jumped from 90 k to 100 k Euros, we say there was a varia-
tion of 10,000 Euros, and so forth. If we introduce a dy-
namical factor, and pass from an absolute variation to a
variation measured against time, we obtain a rate of change.
When the reference time intervals become smaller and
smaller, we obtain the infinitesimal rate of change.
The most familiar example is probably measuring speed
on a motorway. Suppose it takes us 5 minutes to drive be-
tween two speeding cameras placed 10 km apart. The sys-
tem will record an average speed of 120 km/h (5 minutes
correspond to 1/12 of an hour). To measure the instanta-
neous speed at a given point, instead, which is what we read
on the dashboard, ideally we should bring the cameras
closer, until their positions coincide. This is the practical
rendition of a mathematical procedure called limit: the in-
stantaneous speed is the limit value of the speeds measured
over increasingly smaller travel times. Speed is an example
of what in mathematics we call a derivative, a function that
expresses the rate of change of the initial quantity. (If these
notions are still a bit obscure, be patient; in a short while we
will clarify them with concrete examples.)
Since change rates are expressed by derivatives, the mod-
el’s equation will contain relationships between the solu-
tions and their derivatives. These types of equations are
called differential equations.
 1 The Model, aka the Magic Box 5

1.1 Initial Data

Suppose we are addressing, as in this case, a dynamical
model, or evolution model, which describes a changing situ-
ation. In order to make predictions on future behaviours, it
is necessary to include in the data the system’s initial state.
In our example this entails knowing the water’s speed and
depth at any point of the river bed at time zero. Time zero
refers to the instant at which the simulation of the process
is supposed to begin. For example, if we observe the water
flow from today at 12 pm (time zero) until tomorrow at
12 pm (final time), in principle we would also need the full
knowledge of the water’s speed and depth at today’s 12 pm.
This would be if we lived in an idealised world. In the real
world, any measurement and observation will only provide
us with some of these values (for instance, only in the prox-
imity of detection instruments or speeding cameras).
Typically, the missing data are provided (in an approximate
way, inevitably) by surrogate numerical models, called ini-
tialisation models. This is a rather complicated story that we
will not investigate here; we shall, nevertheless, describe a
few examples when later dealing with more specific
situations.

1.2 Approximate in Order to Solve

Let us go back to the model. If we have done the job prop-
erly, we should have managed to find the relationships be-
tween the unknowns (speed of the current, height of the
water) and the data (shape of the river, roughness of the bed
surface, upstream flow rate, status at time zero), thanks to
mathematical laws (the differential equations) that express
6 A. Quarteroni

the fundamental and universal physical principles that rep-

resent the process under exam (the water flow in a basin).
Often (actually, almost always) this lands us on an ex-
tremely complex mathematical problem, much different
from those we have seen in school. This problem will be so
complicated that it will not be possible to solve it by hand.
We will therefore need to use a computer. Before that,
though, we shall need to approximate our model, that is,
turn it into a similar model that can be translated into an
algorithm, hence solvable by computer. For this reason we
call the latter a numerical model. Clearly we should make
sure that the solution to the numerical model is very close
to the one we might have found if we had been able to solve
the original mathematical model.
Speaking in general, the overall number of unknown
variables is called the dimension of the system. Numerical
models, which a computer must be able to solve, have one
feature that cannot be waived: they must have finite dimen-
sion. This means they must have a finite number of un-
knowns, and that the algorithm computing them must
consist of finitely many steps. (This fact is not obvious. For
instance, calculating the sum of the inverses of all integers
squared from 1 onwards requires, in theory, infinitely
many steps.)
The passage from the infinite dimension of the mathe-
matical model to the finite dimension of the numerical
model is typical of the entire process. How do we go about
it? It is not simple, and there is no standard way to do it. In
fact there is an entire branch of mathematics (called nu-
merical mathematics, numerical analysis, or approximation
theory) that develops and studies constructive procedures
for the various types of physical problems. Let us discuss an
example within the river picture.
 1 The Model, aka the Magic Box 7

Fig. 1.1 The riverbed divided into a collection of tiny cubes of the
same size

The mathematically exact solution should give the re-

quired values (for the speed etc.) at each one of the infinitely
many points in space and time. We shall however content
ourselves with finding the values at a number of selected
points: the vertices (and perhaps also the centres) of a grid
obtained dividing 3D space in cubes of very small but finite
size (Fig. 1.1). The same should happen for time: we will
find the values corresponding to finitely many instants sep-
arated by regular intervals.
Back to our example, let us take cubes with edges of 1 m
and 10 minute intervals. Suppose the river is 1 km long,
the river bed and banks are completely smooth and the
basin’s cross section is constant, say 10 m wide and 5 m
deep on average (ok, no river is that regular, but we are
simplifying, right?). If we intend to predict the surge over
a 24 h period (from 12 pm today to 12 pm tomorrow) we
will need to find the unknown values of 24 × 6 = 144 time
instants (one hour is made of six 10-minute intervals), cor-
responding to 1000 × 10 × 5 = 50,000 water cubes. Then
on each cube we should solve the numerical model for the
three components of the velocity (because at any point the
velocity of a water particle is determined by its compo-
nents along the three spatial directions), for the pressure
8 A. Quarteroni

and for the elevation (at least for the cubes on the bottom
surface). I will spare you the calculations: just be aware that
the final problem has 210,000 unknowns and must be
solved 144 times. The numerical model has therefore di-
mension 210,000 × 144 = 30,240,000. This goes to show
that a relatively simple situation may lead to models of
whopping dimension.

1.3 How Many Models for One

Problem? How Many Problems
for a Single Model?
The chosen model has 1 m wide cubes, which might not be
enough for an accurate forecast. We might need to take
edges of 50 cm, and in case of a fast moving surge we might
want to monitor the situation every minute rather than ev-
ery ten. The kind of approximation will not be made em-
pirically, but on the basis of rigorous mathematical argu-
ments. Put differently, we have to ensure that our numerical
model is able to provide accurate solutions that allow for a
faithful representation of what will happen in reality. No
surprise then that there may exist several models to describe
the same physical process. The construction of a model de-
pends on the quantity of knowledge over the phenomenon
we wish to incorporate. In other words it depends on the
degree of simplification we deem acceptable for our ends.
Modelling is actually synonymous to approximating: often,
reality is far more complex than our capacity to represent it.
On the other hand it may happen that a single model is
apt to represent distinct physical processes. As surprising as it
may seem, the equations describing the water surge in a
river can also describe the evolution of the weather in a
certain region, and with a little imagination (this is where
 1 The Model, aka the Magic Box 9

abstraction comes into play!) they help to quantify a hedge

fund’s financial risk. Needless to say, the data, and even
more the variables representing the model’s solutions, will
have a different meaning from those of the flood model.
The common thread is the model’s mathematical structure.
This unifying capacity makes the approach via models ex-
tremely versatile, general and powerful.

1.4 The Phases of a Model

To build a model therefore means to complete a process
consisting of many phases, which we may summarise as
follows:

• understand the most relevant aspects of the concrete

physical problem (in the example, the speed of water, its
pressure, the height of the surge at each point of the ba-
sin, the expected time of a prediction);
• determine the essential data (the river’s shape, the basin’s
smoothness, the initial flow rate, the unknown variables’
values at time zero);
• translate this collection of information into a system of
equations in the unknown quantities—the solutions,
which are interrelated and depend on the data: this is
precisely the mathematical model;
• transform the mathematical model into an approximate
numerical model;
• build an algorithm to solve the numerical model, that is
a finite number of steps, each of which consists of unam-
biguous mathematical procedures implementable on
a computer;
• let the computer execute the algorithm, using a program-
ming language (Matlab, Fortran, C++, Python, Java etc.);
10 A. Quarteroni

• verify that the solutions provided by the computer agree

with the observations of the initial problem: this phase is
called validation and is essential to guarantee we have
operated properly.

In the ensuing chapters we shall examine several models in

action, on problems that display a rather different nature.
 2
Weather Forecast Models

Abstract Predicting tomorrow’s weather is a mathematical

problem. The main quantities at stake are the air density, its
temperature, pressure and speed. The current meteorologi-
cal models are increasingly reliable, because of the advance-
ments in atmospheric physics and the growing computa-
tional power of supercomputers.

If we did not have a model to predict the weather maybe

the course of history would have taken a different turn. The
decision to land in Normandy on that particular day (D-­
Day, 6th of June 1944) was taken based on the first math-
ematical model capable of making forecasts with a certain
accuracy.
The weather has always played a key role in all human
activities. Not only regarding our leisure (shall we organise
a picnic in the countryside next Sunday?) or tourism (is it
likely to rain at a given seaside resort at the end of July?),
but also for organising the countless jobs that take place

© The Author(s), under exclusive license to Springer Nature 11

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_2
12 A. Quarteroni

outdoors. Just think about agriculture, for which it is fun-

damental to decide when to do certain jobs like sowing and
harvesting, farming, planning when herds go out grazing,
or to plan transportation, be it of people or goods, by land,
sea or air. Weather forecasts are helpful in a number of other
potentially dangerous situations like predicting the chances
of snow avalanches, floods or tropical cyclones.
Indeed, atmospheric phenomena have always been con-
sidered the paradigm of unpredictability. Nowadays, even
technically advanced societies can suffer severe damages
caused by extreme meteorological events. The impact is
even more critical on lesser developed countries, where the
quantity and nature of rainfall is crucial for the actual sur-
vival of the population. Over the last decades, moreover,
forecasting the weather has become increasingly tied to the
problems of anticipating climate change (that is, the trends
in air and sea average temperatures and rainfall over decades
if not centuries) and predicting air pollution levels.
In this chapter we will present a brief history of the
mathematical models for weather forecasting, and the nu-
merical models developed in the twentieth century with the
aim of producing effective and accurate predictions, based
on a coherent mathematical description of the atmosphere.

2.1 A Model Based on … Thin Air

Addressing the weather forecast problem rigorously is not
easy, as attested by the fact that the first comprehensive
mathematical description only goes back to the start of the
twentieth century. In 1904 the Norwegian mathematical
physicist Vilhelm Bjerknes (1862–1951) suggested that we
describe the motion of the atmosphere by modifying the
equations governing a special class of gases, called ideal
 2 Weather Forecast Models 13

gases for their behaviour and considered ideal from many

points of view [1]. We could say that modern meteorology
was born at that moment. Bjerknes is the founder of the
Bergen School of Meteorology, named after the town where
he worked (is it not interesting that modern weather fore-
casting was born in a very rainy town?)
How can atmospheric air circulation be possibly mod-
elled? How can we capture winds and clouds in mathemat-
ical equations? To begin with, we must understand what we
should focus on. We are certainly interested in air and water
in all their various manifestations: clouds, oceans, rivers,
snow, ice, rain and wind. Notice that all these are fluid ob-
jects: they occupy the available space in a homogeneous and
continuous way, without interruptions or empty parts. We
must in fact begin from the space these elements occupy.
Let us consider the atmosphere, the casing of gases en-
closing our planet that is held in place by the earth’s gravity.
We can divide it in varying layers distinguished by their
composition: the troposphere (the closest layer to us, in di-
rect contact with the earth’s crust), the stratosphere, the me-
sosphere, the thermosphere, the ionosphere and the exo-
sphere. The troposphere reaches the altitude of 12 km above
sea level (to have an idea, airliners typically cruise between
10,000 and 12,000 m). The atmosphere’s stratification is
determined by the phenomenon known as vertical thermal
inversion: in the troposphere the temperature decreases
with the altitude, by about 6.5 °C every 1000 m, until we
reach the layer’s outer boundary where it begins to increase.
In the stratosphere, thermosphere and exosphere the tem-
perature increases with the altitude because of the direct
absorption of the solar radiation. The transition between
neighbouring spherical strata occurs across a discontinuity
surface called pause.
14 A. Quarteroni

To create forecasting models one concentrates on the tro-

posphere and the lower stratosphere (13–15 km), for it is at
these altitudes that high-level clouds (cirrus and cumulo-
nimbus clouds) form.

2.2 The Physical Quantities Relevant

to Meteorology
Weather forecasting involves several physical quantities.
Just to mention a few, air density and pressure, wind speed,
gravitational pull, heat exchanges between the air and the
oceans, orography (the features of the earth surface: plains,
hills, mountains, deserts, oceans, glaciers, forests, tundras
etc.). Some of these will feed the model in the form of data,
that is, known quantities (for instance the topographic re-
lief, gravity, or the planet’s angular acceleration). Others
will be unknowns representing the model’s solutions (most
notably, air pressure, temperature and wind speed).
As we have seen in the previous chapter (recall the river
current), we shall try to construct a model starting from the
basic principles of physics: the conservation laws of mass
and force. We shall apply them to infinitesimal fluid ele-
ments, which we may imagine as small cubes piled on one
another in long columns, from the base (the earth’s crust) to
the highest altitude at which we wish to study the atmo-
spheric motion (the stratosphere’s lower boundary, say). All
in all we will study hypothetical cubes made of, erm,
thin air.
Let us begin from the balance equations: the mass of the
fluid must be preserved, and the forces acting on the fluid
should be in equilibrium. These laws must hold in each
cube. After, we apply the mathematical procedure called
limit, so we make the cubes smaller and smaller and find
 2 Weather Forecast Models 15

equations that we can assume hold at all points of the phys-

ical space and at all times: from the initial instant (when the
model starts) to the final instant, the maximum time we
wish to cover with the forecast.
This process is far from elementary, and we shall defi-
nitely not give the details here. Instead, we shall briefly de-
scribe the balance equations that need to be imposed, and
discuss the type of equations that arise and their variables,
in other words the unknown quantities.

2.3 Physics Comes to the Rescue

First of all, let us summon the principle of classical mechan-
ics (perhaps better known in the formulation for chemical
reactions due to Antoine-Laurent de Lavoisier) whereby
“Nothing is lost, nothing is created, everything is transformed”.
We apply it in our situation to every infinitesimal volume.
At each instant the mass inside a cube equals the initial mass
present, plus the ingoing mass flux, minus the outgoing flux
(these fluxes are computed across the walls, the lateral sur-
faces of each cube). For example, if at the initial time the
cube contains a million air particles and during one second
50,000 have entered and 100,000 have exited, then after
one second there will be 950,000 particles. But how can we
know how many go in and out? These are unknown quanti-
ties. It is reasonable to assume that the flux of particles
through the cube walls depends not only on the speed of the
air but also on its density. If the speed is constant, in fact,
then the thinner (less dense) the air is, the greater the flux
will be. Speed and density will be our first unknowns: the
balance of the mass thus translates into an equation depend-
ing on the speed and the density. Starting with the infini-
tesimal volumes, the usual limiting process mentioned above
produces an equation containing not just our unknown
16 A. Quarteroni

quantities but also an indication of how they change, instant

by instant: these are the quantities’ derivatives. (To be abso-
lutely precise, the unknowns are the density and the flow rate
of the mass, namely the mass at each instant, which is ob-
tained by multiplying the speed and the density).
So now we have an equation in two unknowns. This is
not enough to solve the problem, and we must have an-
other equation. To find it we impose another physical law,
Newton’s second law of motion, according to which at any
point in space and time the mass of fluid times its accelera-
tion equals the net force acting on the fluid.
The net force is the sum of two kinds of forces: volumetric
forces and surface forces. The former concern the entire mass
contained in our cube: this includes gravity, the Coriolis force
(due to the earth’s rotation) and also the cohesive attraction
between molecules. The nature of molecular cohesion de-
pends on the specific material under exam (a gaseous fluid in
our case) and is determined through suitable experiments.
Mathematically speaking, they are expressed by constitutive
equations, relationships among quantities such as the stress
forces applied to the material and the deformations they gen-
erate. Surface forces, instead, are exerted on the walls of the air
cube only. Putting everything together we end up with an
equation depending on several quantities: acceleration, the
rate of change of pressure between arbitrarily close points, the
earth’s gravitational potential, plus other more technical terms
(more precisely: one term which represents the dissipation
due to atmospheric turbulence and one accounting for the
various heat sources that feed energy into the cube of fluid).
In order to write down this equation, though, a new vari-
able was introduced: the pressure. So here we go again: to
solve the problem we need yet another equation. Normally
this is a suitable equation of state, an algebraic relation be-
tween density and pressure. But there is a but. That rela-
tionship is expressed by a term proportional to the fluid’s
temperature, so we are back to the start. Another variable
 2 Weather Forecast Models 17

appears, the temperature, which requires one equation more.

So we impose the condition that the fluid is in thermal
equilibrium (thermal balance equation) as prescribed by the
first law of thermodynamics. And that is it, at long last!
Gathering all the equations found so far we have a system
with several unknowns.
Conservation of mass:
density spatial divergence
 
  
       0
t

Newton’s second law of motion:

angular velocity of the oth

her forces acting
non-inertial reference frame pressure on the fluid
   1   
v     
 v  v  2   v   p   
 F
t 
(Earth) gravitational potential energy

Equation of state:
universal gas constant

pRT
temperature

Thermal balance equation:

specific heat capacity at constant volume

  T     
cv   v  T   p  v  Q
 t 
18 A. Quarteroni

2.4 The Initial Data

and the Boundary Data
This is a monster system, one that no mathematician would
ever be able to solve explicitly and provide a clear and defi-
nite expression for the unknown variables. There is a glim-
mer of hope though: under certain conditions on the prob-
lem’s data, it can be proved that, theoretically, there is a
solution (and that it is unique: there cannot be others—at
least within a short time frame). These conditions are about
the initial data, meaning the values of the dynamical vari-
ables, namely the speed, temperature and density of the air
at the initial time of the forecast (this is the initialisation
mentioned in Chap. 1), and also the boundary data, those at
the frontier of the region we are examining. The equations
subsuming the latter are called boundary conditions.
Right, but where is this boundary? Or rather, how big is
the region we are working in? Before we attempt to answer
that let us note there are two different types of weather fore-
casting models. There are global models (continental, or
world-wide) and local models (with regional reach, aka lim-
ited area models). The latter usually cover a country, like
Italy, or specific regions, such as Sicily. For LAMs the
boundary is the entire perimeter of Italy or Sicily. The
boundary values are provided directly by the global models
(like the continental one for Europe), that should in prin-
ciple compute the unknown variables on the boundary of
any sub-region of the planet. The global planetary model,
on the contrary, has no boundaries: after all, the earth is
(roughly) a sphere. In this case the boundary conditions
subsume the spherical shape. If we started from any point
on the surface of the earth and travelled along a meridian
line or the parallel through that point, after a full circum-
navigation we would return to the initial point, and clearly
 2 Weather Forecast Models 19

we would find the same values for the pressure, the tem-
perature etc. Furthermore, if we wanted to limit the simula-
tion to an altitude of, say, 13 km at most, hence to a spher-
ical shell contained in the stratosphere, under the simplest
possible assumptions we could prescribe a uniform value
for temperature, density and wind speed in the upper layer
of the shell.
The above system is, from a strictly mathematical view-
point, solvable once the initial data and boundary condi-
tions are given. That said, in practice finding this solution is
a massive endeavour. First of all, the initial dataset relative
to the atmosphere status is available only at a relatively
small number of points (those where the detectors are posi-
tioned), and they refer to physical variables that differ from
one another and vary from point to point. Furthermore,
they are affected by measuring errors that cannot be ne-
glected. Secondly, a realistic description of any meteoro-
logical phenomenon cannot refrain from the atmospheric
distribution of water vapour, its phase changes (from gas to
liquid and vice versa) and the ensuing rainfall. All these as-
pects complicate a lot the task of making accurate, quanti-
tative forecasts.

2.5 Numerical Models, from D-Day

to von Neumann
Summarising, there is no hope to solve our system by hand.
It therefore becomes inevitable to resort to a suitable nu-
merical approximation, so to render the system translatable
into an algorithm, and eventually solve it with a (powerful)
computer.
The first attempt to tackle the numerical resolution of
the equations of motion is due to the British scientist Lewis
Fry Richardson (1881–1953). During the first half of the
20 A. Quarteroni

twentieth century Richardson managed to describe the

main phases of weather forecasting based on a mathemati-
cal model and on the numerical approximation of the equa-
tions of motion. He also introduced new concepts and tools
in many mathematical subjects like numerical analysis, lin-
ear algebra and fluid mechanics. For his contributions,
nowadays considered classical results, the name Richardson
is familiar to many mathematicians, although few are aware
of the peculiar context (meteorology) in which his ideas
came to fruition.
Richardson’s pioneering effort, told in a 1922 book [2],
culminated in the first concrete numerical solution of a sys-
tem of equations governing the atmospheric motion over a
region as wide as the entire Western Europe. The author
himself did all computations by hand, over a number of
years and often in near-fictional circumstances. During
World War I Richardson, a conscientious objector on reli-
gious grounds, was a nurse on the French front. He used
every free moment to wrap up a part of the massive amount
of arithmetical computations required.
Unfortunately his work did not have any immediate
practical fallout. Despite the impressive computing effort,
his results produced a completely wrong prediction. On the
other hand, his contributions laid the ground for modern
weather forecasting. For the first time in history someone
had analysed every component of a numerical model for
atmospheric circulation, following a conceptual scheme
that is not dissimilar to the one used at present. Furthermore,
it was the first time that a problem with no analytical solu-
tion (that is, exact, explicitly representable on paper) had
been tackled using the strategy of applying numerical tech-
niques to a simplified situation. In the most visionary and
renowned passage of his work, Richardson imagined a gi-
gantic amphitheatre filled with tens of thousands of
 2 Weather Forecast Models 21

(human!) calculators performing in parallel (independently

from one another) the arithmetical computations of his at-
mospheric model on different continents, with the aim of
generating in real time a mathematically sound forecast
based on the computing technology available at his time. A
precursor, some might say, of today’s parallel computing [3].
Decisive in making Richardson’s vision concrete was the
contribution of the heirs of the Bergen School, in particular
that of the Swedish meteorologist Carl-­ Gustaf Rossby
(1898–1957). After studying under Bjerknes, Rossby emi-
grated to the US in the 1920s. In America he held top posi-
tions at the US Whether Bureau, MIT and the University
of Chicago. He contributed to the founding of the weather
division working for civil aviation and the army during
WWII. In some sense, the decision to land in Normandy
exactly on D-Day (June 6, 1944) is also due to his work.
Rossby identified some of the characteristic features of
large-scale atmospheric circulation, such as jet streams and
what we now call Rossby waves [4]. He was also responsible
for significantly simplifying the general equations of atmo-
spheric motion [5]. The use of these simplified models was
also key to the first forecast made by a computer [6], a result
of the Princeton collaboration between John von Neumann
and Jules Charney, a former student of Rossby in Chicago,
in the late 1940s. With that system, the errors in the mea-
surements needed to recover the problem’s initial and
boundary data had a smaller impact on the quality of nu-
merical solutions.
In 1950 it became possible to predict the weather on a
region almost as large as North America, using a simplified
model describing the atmosphere as a single stratum of a
uniform fluid. In order to forecast over the next 24 hours an
equal amount of time was necessary on the only existing
computer of the time, the ENIAC. Nonetheless, the work
22 A. Quarteroni

of Charney and von Neumann proved, for the first time,

that a prediction based solely on a numerical model was
able to achieve results both qualitatively and quantitatively
not far from the forecast obtained by an experienced meteo-
rologist using the same dataset.

2.6 Increasingly Sophisticated

Models: Lorenz’s Butterflies
With this background, and thanks to the advancements in
computer science technology and mathematical modelling,
the fifty years ensuing the first work of Charney and von
Neumann witnessed a continuous improvement in fore-
casting as regards both accuracy and reliability. This is not
the place for a detailed account, but if we wanted to iden-
tify a few key elements that prompted a major progress in
weather forecast, apart from the obvious spectacular in-
crease in computer performance, we should focus on these
fundamental aspects: the use of high-precision numerical
methods to approximate numerically the equations of mo-
tion; the advancement of data-assimilation techniques (to
which we shall return); the systematic and increasingly per-
vasive use of satellite detection [7, 8].
Another essential contribution to making finer forecasts
came about when we eventually understood better the pe-
culiarities of non-linear and chaotic dynamical systems, such
as those governing the atmosphere. The notion of a chaotic
system, that later became so relevant and almost hip, was
born out of the work of a meteorologist, Edward N. Lorenz
(1917–2008) [9, 10]. By simplifying the equations of at-
mospheric motion Lorenz wrote down a small system of
two equations (subsequently turned famous under its
 2 Weather Forecast Models 23

creator’s name) with the aim of verifying the possible insta-

bility of the atmosphere. Lorenz, more precisely, wanted to
study a precise phenomenon: that the atmosphere’s dynam-
ics may amplify an even very small uncertainty in the initial
data to the point of preventing any sensible prediction over
its future state. It is for this reason that a two-week weather
forecast, or longer, cannot be accurate.
As a matter of fact Lorenz was the first person, in 1962,
to analyse the so-called butterfly effect. An excerpt from a
paper published by the New York Academy of Sciences [11]
reads One meteorologist remarked that if the theory were cor-
rect, one flap of a seagull’s wings would be enough to alter the
course of the weather forever. Lorenz discovered the effect
while observing that the tiniest change in the initial data of
his non-linear system would have produced a completely
different outcome. In later speeches and papers Lorenz used
the more elegant butterfly as a symbol, perhaps inspired by
the plot generated by a special instance of his system, the
so-called Lorenz attractor, that does indeed look like a but-
terfly (Fig. 2.1). The official birth date of the butterfly met-
aphor might be 1972, when Lorenz gave a talk in Brazil
entitled Does the flap of a butterfly’s wings in Brazil set off a
tornado in Texas?
One merit of this kind of study, in the field of chaotic
dynamics, is to have highlighted the limitations of a purely
deterministic set-up for the weather forecasting problem,
thus paving the way for probabilistic techniques. Put differ-
ently, today we no longer say tomorrow it will rain. We
rather tell the odds of it raining. Predictions, whether global
or local [12], are now done in this way.
What is more, the detailed analysis of the phenomenon
whereby everything depends on the quality of the initial
dataset has stimulated the development of novel techniques
24 A. Quarteroni

Fig. 2.1 Lorenz’s butterfly-shaped diagram. Photo: zentilia/

Shutterstock

of data assimilation [13, 14]. These techniques allow to in-

corporate in the numerical model millions of single mea-
surements made on a daily basis throughout the world us-
ing the most disparate means. In fact, even leaving aside the
significant theoretical advances, there is no doubt that the
better quality of weather forecasts is strongly related to the
great improvement, both qualitative and quantitative, of
the data available. Starting from the 1960s, in particular,
besides the measuring stations on the surface (of which,
paradoxically, there are fewer due to the high running
costs), we also use satellite surveying systematically. These
provide the bulk of the data used to initialise numerical
models, nowadays.
 2 Weather Forecast Models 25

2.7 Weather Forecasting Today

The impact of the scientific and technological advances we
have discussed has been impressive. To get an idea of the
degree of resolution and efficiency of the current models we
might consider for instance some of the characteristics of
the IFS, a global model of the European Centre for
Medium-Range Weather Forecast (ECMWF) in Reading,
UK. This prediction model, considered one of the best in
the world, uses a grid with average horizontal spatial resolu-
tion of 22 km and 90 vertical levels. This means it employs
cubes of fluid with a square base of 22 km per edge, and
height of the order of 100 m, so it allows for the inclusion
in the model of part of the stratosphere. The IFS can make
10-day forecasts on a modern parallel supercomputer in less
than an hour, to which one must add a few hours for the
complicated data-­assimilation process. The prediction’s ac-
curacy is much better now compared to earlier models: just
think that the average number of days covered by a reliable
IFS forecast for Europe has passed from 5.5 in 1980 to 7.5
today. Even greater has been the impact of the new tech-
niques over tropical zones and the southern hemisphere,
where the scant network of measuring centres on the ground
did not permit, in the past, reliable forecasts beyond one
day. Today, instead, we have reached reliability levels com-
parable to the richest and most densely populated areas,
due to the improved integration and assimilation of satellite
observations.
Moreover, probabilistic predictions are nowadays a func-
tional reality, one that provides a conceptually more accu-
rate description of the evolution of a chaotic system such as
the planet’s atmosphere, and at the same time allows to
quantify its margins of error in a given meteorological
situation.
26 A. Quarteroni

Among the other global models in use today we count

the United States Global Forecast System (GFS), the United
Kingdom Met Office (UK), the GME (Germany), the GEM
(Canada), the JMA (Japan), the NAVGEM (US Navy), the
SEMBAC (US) and the WMC (Russia). These models dif-
fer in several parameters, like the horizontal resolution, the
orography involved, the mathematical model employed,
the spatial numerical resolution and many more. In some
cases we can only speculate, since often these models tech-
nical aspects have not been fully disclosed to users.
If one is interested in a smaller area (like a region), local-
scale LAMs can be used. They possess a higher horizontal
resolution than global models. Some LAMs have resolu-
tions of the order of 1 km, which allows them to distinguish
the forecasts of places 1 km apart. Moreover, they are suit-
able for short-term predictions: from a few hours (nowcast-
ing, as opposed to forecasting) to two/three days tops.
To sum up, it is because of these spectacular achieve-
ments of mathematical models that, today, if we want to go
on a Sunday outing we may look up the weather forecast
and decide whether to head east, south, west or north to be
sure we will find the sunniest meadow for a picnic or the
windiest beach for windsurfing.

References
1. Bjerknes V. (1904), Das Problem der Wettervorhersage, be-
trachtet vom Standpunkte der Mechanik und der Physik,
Meteorologische Zeitschrift, 21, pp. 1-7.
2. Richardson L.F. (1922), Weather Prediction by Numerical
Process, Cambridge University Press, Cambridge.
3. Lynch P. (1992), Richardson’s Barotropic Forecast: A
Reappraisal, Bulletin of the American Meteorological Society,
73, pp. 35-47.
 2 Weather Forecast Models 27

4. Rossby C.G. (1939), Relations Between Variations in the

Intensity of the Zonal Circulation of the Atmosphere and the
Displacements of the Semipermanent Centers of Action,
Journal of Marine Research, 2, pp. 38-55.
5. Haurwitz J. (1940), The Motion of Atmospheric Disturbances
on the Spherical Earth, Journal of Marine Research, 3,
pp. 254-267.
6. Charney J.G., Fjrtoft R., von Neumann J. (1950), Numerical
Integration of the Barotropic Vorticity Equation, Tellus, 2,
pp. 237-254.
7. Robert A.J. (1966), The Integration of a Low Order Spectral
Form of the Primitive Meteorological Equations, Journal of
the Meteorological Society of Japan, 44, pp. 237-245.
8. Bourke W. (1972), An Efficient, One-level, Primitive
Equation Spectral Model, Monthly Weather Review, 100,
pp. 683-689.
9. Lorenz E. (1963), Deterministic Nonperiodic Flow, Journal
of the Atmospheric Sciences, 20, pp. 130-141.
10. Lorenz E. (1969), The Predictability of a Flow Which
Possesses Many Scales of Motion, Tellus, 21, pp. 289-307.
11. Lorenz E. (1963), The Predictability of Hydrodynamic Flow,
Transactions of the New York Academy of Sciences, 25(4),
pp. 409-432.
12. Molteni F., Buizza R., Palmer T.N., Petroliagis T. (1996), The
ecmwf Ensemble Prediction System: Methodology and
Validation, Quarterly Journal of the Royal Meteorological
Society, 122, pp. 72-119.
13. Le Dimet, F.X., Talagrand O. (1986), Variational Algorithms
for Analysis and Assimilation of Meteorological Observations:
Theoretical Aspects, Tellus a, 38, pp. 97-110.
14. Rabier F., Thepaut J.F., Courtier P. (1998), Extended
Assimilation and Forecast Experiments with a Four-
Dimensional Variational Assimilation System, Quarterly
Journal of the Royal Meteorological Society, 124, pp. 1-39.
 3
Epidemics: The Mathematics
of Contagion

Abstract Studying how a pathogen spreads among the

population requires accurate mathematical models. The
classical SIR epidemiological model divides the population
in susceptibles, infected and recovered. Once the initial
conditions are given, the numbers in each group vary in
time under simple differential equations. A generalisation
of the basic model is very well suited to describe the evolu-
tion of the current Covid-19 pandemic.

As we saw in the previous chapter, the first mathematical

models (as we understand the notion today) were devel-
oped to study the weather, a phenomenon so ingrained in
the day-to-day that it deserves our undivided attention. A
few years after the birth of the first meteorological models
mathematicians thought they should address another type
of problems which, alas, has been with us since the dawn of
time: the spread of infectious diseases through epidemics.

© The Author(s), under exclusive license to Springer Nature 29

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_3
30 A. Quarteroni

Epidemiological models, too, are a hundred years old by

now, and may be thus considered classics in applied
mathematics.
Before we examine the most famous among these models,
called SIR, we must begin from a field that is only apparently
distant: the ecological interactions between preys and preda-
tors, which we will see display behaviours similar to contagions.

3.1 Preys and Predators

Prey-predator interactions are as old as life on earth. Since
the dawn of time there exist those who hunt and those who
are hunted, in forests, steppes, seas and skies. The equilib-
rium of the species in a system depends on the ratio be-
tween the available resources and the number of consumers
of those resources, hence it also depends on the ratio preys
to predators. The whole picture has a complex dynamics.
No surprise, then, that mathematicians have been keen on
modelling this phenomenon for more than a century.
Whenever a mathematician is involved, everything is ob-
viously always turned into equations. The most classical and
renowned are the Lotka-Volterra equations, named after the
mathematicians Alfred Lotka, who was born in the Ukraine
and later became a US citizen, and Vito Volterra, from Italy.
Volterra got involved in the problem around 1925 by
his son-in-law, the biologist and naturalist Umberto
D’Ancona. While studying the Adriatic Sea fish fauna
D’Ancona noticed that between 1915 and 1918 the sharp
drop in fishing caused by World War I had altered the bio-
logical equilibria. Sharks, stingrays and other predatory
species had increased their numbers at the expense of their
preys, the fish that feed on plankton or small invertebrates
(incidentally, these fish are what consumers most look for,
 3 Epidemics: The Mathematics of Contagion 31

so their diminished numbers had important economic

consequences). Volterra, after examining the data, came
up with a model based on two equations: one for the
number of exemplars of the predatory species evolving as
time goes by, the other providing the number of exem-
plars of preys. Clearly the two equations are tightly related
(or coupled, as mathematicians say), because the two num-
bers vary according to the interactions between the two
species.
In their simplest form the equations translate into math-
ematical terms several facts: predator y only eats prey x, the
number of predator-prey encounters is proportional to xy,
the variation of x is determined by the existing population
and by the action of y, and conversely. In symbols:

So what happened during the war? When fishing reduces to

almost nothing, predator fish have a larger number of prey
available, they feast happily and their number grows. This
obviously reduces the number of preys until the number of
predators reaches a peak (let us remember this peak, we
shall encounter it again in this chapter) and then starts to
fall quickly. At this point prey fish can increase in number,
32 A. Quarteroni

Fig. 3.1 The behaviour of the populations of prey and predators

in time, according to the Lotka-Volterra model

but they have enemies, and so forth. What we have in prac-

tice is a cyclic (periodic) pattern in which the two curves of
preys and predators intertwine and separate, in a life-and-
death orchestrated dance (Fig. 3.1).
Needless to say this is a primitive model, which we may
render more realistic by including terms that account for
the season and climate conditions (animal reproduction is
typically subject to fixed temporal cycles and is affected by
temperature), the reproduction rates of the species involved,
their longevity, the presence of external factors such as hu-
man intervention through fishing etc.

3.2 The Epidemiological Models

From a purely mathematical viewpoint, the study of the dy-
namics with which a pathogen (a virus, a bacterium or other)
spreads among the population is not that different from the
study of the interactions between preys and predators. We
could think of pathogens as predators on the hunt for preys,
meaning individuals to infect, and then proceed accordingly.
The earliest mathematical model in epidemiology was
probably formulated over 250 years ago by Daniel Bernoulli
(from a renowned family of mathematicians that vastly
 3 Epidemics: The Mathematics of Contagion 33

contributed to many areas), who wanted to endorse the

cause for smallpox vaccination with his work. Bernoulli
successfully demonstrated, mathematically, that if the en-
tire French population got vaccinated, the average life ex-
pectancy would increase by more than three years. At the
time life expectancy at birth was not high, and those extra
years were a big change.
After the work of William Hamer and Ronald Ross in
the early 1900s, the next cornerstone is 1927, when the
Scots William Kermack and Anderson McKendrick pro-
posed the famous model that bears their names. It was des-
tined to become a reference work in the field of epidemio-
logical modelling for years to come. It is a differential model
of SIR type (in a short while we will see what this means),
designed to explain the rapid growth and successive de-
crease in the number of infected people observed in certain
epidemics, especially the plague and cholera. Since then the
mathematical field of epidemiology has blossomed [1, 2].
One major step forward was passing from deterministic
models to models that subsume casual fluctuations.
But one step at a time.

3.3 The Population’s Critical Size:

The Case of Measles
Before the role of pathogens in spreading infectious diseases
was discovered, and way before the arrival of the first math-
ematical models, an apparently inexplicable phenomenon
became manifest. Certain diseases (like the flu, measles or
diphtheria) seemed to explode cyclically, alternating epi-
demic bursts and latent periods. Outbreaks appeared and
disappeared periodically, especially in large cities. Why
was that?
34 A. Quarteroni

The discovery of the mechanism whereby epidemics

seem to flare up and fizzle out is mostly a consequence of
studies conducted on measles. Measles is a disease with
which mankind has cohabited for millennia. Before the dis-
covery of its vaccine it was endemic throughout the world
and used to cause hundreds of thousands of deaths every
year. (It has not been eradicated to the day, but the number
of deaths has decreased a lot.)
In 1846 the Danish physician Peter Ludvig Panum ob-
served a measles epidemic on the Faroe Islands directly. He
was the first person to understand certain mechanisms gov-
erning the transmission, at a time when the role of patho-
genic micro-organisms was still not known. In 1906 the
English doctor William Hamer presented a striking result
of his study: the interval between measles epidemics ap-
peared to be constant. He collected the data in London,
which at the time had a population of 5 million and was
witnessing measles outbreaks every 18 months. Hamer also
noticed that in smaller towns this phenomenon did not oc-
cur, and the disease’s manifestations were more episodic.
The secret, as was discovered mid-century, lies in a num-
ber called critical community size (CCS). The critical com-
munity size is the smallest number of people in which the
virus can persist endemically. The CCS varies with the dis-
ease and reflects its characteristics (transmission efficacy,
virulence, permanent immunity of those who have recov-
ered etc.). Measles has a CCS roughly equal to 500,000. In
communities larger than half a million people (and in ab-
sence of a vaccine, naturally) the disease never disappears
completely and returns periodically. Smaller communities,
on the other hand, are reached sporadically, usually from
outside, and after a while the infection dies out. Today the
phenomenon is less accentuated in countries with strong
global connections, but it still remains important in Africa,
South America and parts of Asia [3].
 3 Epidemics: The Mathematics of Contagion 35

3.4 A World of Susceptible

Individuals
When dealing with an emerging virus (SARS, or the like)
everything is more complicated. On one hand the virus’s
specifics are unknown: virologists do not know its genetic
features, specialists in infectious diseases are unaware of its
virulence. On the other hand the virus, being a newcomer,
has the potential to infect the entire world population, since
no human being has ever been in contact with the pathogen
and did not have the chance to develop an immune
response. We then say that the entire population is suscep-
tible. This word refers to a person that has not contracted
the disease, but might. Once a pathogen has found its first
host (patient zero) and has infected them, this individual is
potentially able to spread the disease according to the spe-
cifics of the virus (SARS infects through the airways, for
HIV the transmission is of sexual nature). Susceptible indi-
viduals that come in contact with patient zero can get in-
fected, and this is how the epidemic spreads.
The basic Kermack-McKendrick model starts by cate-
gorising the population in two groups: infected people (I)
and susceptibles (S). Clearly, sooner or later some of the
infected get better, and then pass to form a third category,
the recovered (R). This group includes people who no longer
transmit the contagion (so either they have healed com-
pletely, or they have died of it). This sort of model is called
compartmental and is labelled by the acronym SIR.
Kermack and McKendrick introduced a fundamental
parameter, known as the basic reproduction number R0,
that expresses the average number of susceptibles one indi-
vidual might infect in the first phase of the contagion. The
larger R0 is, the more virulent the epidemic. The reproduc-
tion number of measles, for instance, is thought to be
36 A. Quarteroni

around 15. That is a very large number: an infected person

will pass the virus to 15 others, on average, if none are vac-
cinated. For comparison, smallpox has an R0 roughly equal
to 7, for mumps R0 is about 10, for chickenpox 8.5. But we
need to be careful: we are talking about average values,
computed over very large populations. In reality there are
many infected people who do not infect anybody else, and
also few so-called superspreaders. As regards SARS-COV-2,
the coronavirus that was detected in China in 2019 and
then infected the world over, the estimates vary a lot with
the place and the moment. In Italy, for example, it is be-
lieved that R0 is close to 2.5. This seems a low number com-
pared to other viruses, but it is still very bad news. Just
think that the R0 of the Spanish Flu, that between 1918 and
1920 infected 500 million people and caused tens of mil-
lions of deaths, has been estimated, retrospectively, to be
about 2.1.
As a matter of fact this is evident also mathematically. An
epidemic with R0 = 2 that is left to spread without counter-­
measures would lead to catastrophe. Starting from patient
zero, we would then have 2, 4, 8, 16 infected people (re-
member these are averages), so 2n infections after n itera-
tions. The growth curve of infections would be exponential
(Fig. 3.2). Although this model is simplistic, as we shall see
in Fig. 3.4 it is reflected in the real data.
A dynamics of this kind can, without any intervention,
lead to a massive percentage of infections (40–80%, de-
pending on the pathogen), with unsustainable social costs.
We may ask why we do not reach 100%, given that the
growth is exponential. Luckily enough, there are natural
obstacles to the spread of an epidemic, like the so-called
herd immunity. This term refers to the phenomenon
whereby the healed and those that are immune for whatever
reason, once they reach a certain critical number, start
 3 Epidemics: The Mathematics of Contagion 37

Fig. 3.2 The spread of the contagion for an epidemic with R0 = 2

protecting the rest of the susceptible population, because

they act as spacers. Moreover, there exist random variables
(in this case favourable ones) that fight the spread of
the virus.
Clearly, as the contagion progresses, the introduction of
containment measures (also called NPIs, for non-­
pharmaceutical interventions) cause the number R0 to de-
crease from its initial value, and we speak about the effective
reproduction number Rt.
The critical value of Rt is 1: below this threshold every
infected person infects on average less than one other indi-
vidual, and the epidemic will eventually die out. When
Rt = 0.5 for example, after n steps the average number of
infected individuals is 1/2n, a number that becomes increas-
ingly smaller as n grows. So if we want to contain an epi-
demic we should make sure Rt stays below that crucial value.
If there is a vaccine, to ensure the epidemic will extin-
guish itself we must vaccinate a fraction of the population
38 A. Quarteroni

larger than or equal to (1 − 1/R0). By putting in this formula

specific values of R0, we conclude that to eradicate measles
we need to vaccinate 95% of the world population (and we
are quite far from that), for poliomyelitis and smallpox
about 85% (we are almost there for the former, whereas
smallpox is considered to be completely eradicated), for
SARS-COV-2 around 60%. In the absence of a vaccine we
must rely on NPIs, which include the use of face masks,
social distancing, shielding and so on. A quick calculation
tells that if every individual came into contact with no more
than 0.6 strangers on average, the Covid-19 contagion
would stop. It is for this reason that one resorts to quaran-
tines, social distancing, contact-tracing etc.

3.5 The Equations of the Contagion

Let us now see how one gets to the equations of an epide-
miological model. For the easiest model we need a few sim-
plifying assumptions. The first is called homogeneous mix-
ing: every individual has the same odds of infecting any
other person, irrespective of their previous contacts. It is
obviously a rather unlikely hypothesis, since we all have
more frequent contacts with family members, colleagues
and friends, and we meet less frequently whoever does not
live nearby or associates with other groups. Secondly, in the
simple model we assume that the incubation period (the
time between the infection and the appearance of the symp-
toms) and the course of the disease (the period between the
first symptoms and recovery, or death) are constant. To
make the model more realistic we should consider on one
hand heterogeneous mixing, and on the other also contem-
plate the possibility that infected people are prevented from
 3 Epidemics: The Mathematics of Contagion 39

mingling at different rates, as a consequence of full recov-

ery, immunity acquisition, isolation or death.
Let us stick with the simple situation. Susceptible indi-
viduals S that move out of the infected compartment I will
widen the compartment of the recovered R, according to
the pattern S → I → R. The probability of transition S → I
is related to the number of contacts between one suscepti-
ble and the whole infected community. The same reasoning
applies to the transition from I to R. The model should
then contemplate the average times spent in phases S and I.
Let us indicate by s, i and r the relative quantities of peo-
ple in each compartment, and by N the population’s size.
S I R
Then= s = ,i and r = . Under these assumptions
N N N
we deduce that the three quantities are related by the fol-
lowing equations:

ds
   si
dt
di
  si   i
dt
dr
 i
dt

where ds/dt, di/dt, dr/dt indicate the rates of change, that is,
the derivatives, of the quantities s, i, r over time. (Shortly we
will see what β and γ mean.)
To the above equations we must add the initial condi-
tions. We should, in other words, establish the starting val-
ues s0, i0 and r0 of the three quantities, those defining the
situation in the population at the beginning of the period
considered.
40 A. Quarteroni

The values s0, i0, r0 are positive or zero. Moreover, because

of how s, i, r have been defined, we have s0 + i0 + r0 = 1, and
a similar identity s + i + r = 1 holds at any time.
By convention, 1/β is the average time between succes-
sive contagions. In this way the coefficient β in the first two
equations denotes the average number of contagions per
unit of time. The term −βsi expresses the speed of transmis-
sion from compartment S to compartment I. We indicate
by 1 the average time spent in compartment I. Recalling
that R0 was the average number of susceptibles an infected
person can pass the virus to, we will have R0 = (average con-
tagions per unit of time) × (average infectious period) = β/γ.
How do we assign a value to β and γ? One possibility is
to start from the experimental data coming from animal
infections. This procedure though is rather difficult and in-
efficient in the case of new epidemics. What happens in
practice is that one estimates with statistical methods two
quantities related to the infectious, or communicability, pe-
riod. One is the incubation period mentioned earlier, that
is, the time between the contagion and the manifestation of
the symptoms. The other is the serial interval, which refers
to the time between the appearance of symptoms in succes-
sive cases (Fig. 3.3). In the early days of SARS-­COV-­2, for

Fig. 3.3 A patient’s symptomatic and contagious trend in time

 3 Epidemics: The Mathematics of Contagion 41

example, Chinese scientists managed to estimate these two

parameters quickly but only approximatively: an average
incubation time of 5.2 days and an average serial interval of
7.5 days. A second possibility for gauging the parameters β,
γ during an epidemic consists in comparing the model’s re-
sults over a given time interval with the real data observed,
and then choose the values and that minimise the differ-
ence. This procedure is called model calibration.
The presence of a time delay between the emergence of
the infection in a new patient and the start of the commu-
nicability period prompts us to make a first generalisation
of the SIR model, where we add a fourth compartment.
This is the so-called SEIR model, with E indicating the ex-
posed: infected individuals that are still not contagious.
Several other generalisations are available. Figure 3.4, for
instance, shows the SUIHTER model (named after the ini-
tials of its compartments) that has been successfully used
for the simulation and forecast of the spread of COVID-­19 in
Italy starting from February 2020 [5,6].

Fig. 3.4 The SUIHTER epidemiological model: S = Susceptibles, U

= Undetected, I = Isolating at home, R = Recovered, H = Hospital-
ised, T = threat of life, E = expired
42 A. Quarteroni

The main issue with these epidemiological models is

their deterministic nature. The model only considers average
values, and does not account for the fact that the contagion
happens in a nonhomogeneous way, with random compo-
nents. In practice a deterministic model such as SIR is only
valid asymptotically, if we suppose there are infinitely many
individuals (and it is for this reason that it works less badly
with large populations).
A solution is to introduce stochastic equations, in which
the unknowns s, i, r depend on random variables as well, to
reflect the intrinsic uncertainty of the phenomenon mod-
elled. But this complicates the calculations.
Other more sophisticated deterministic models sub-
sume the population inhomogeneity in age and census,
or in spatial distribution. If vaccines are already available,
it becomes necessary to introduce an additional variable
measuring the percentage of vaccinated population (or
additional variables if vaccines are administered in
several doses).
Also the computation of the CCS, which we discussed
for measles, is based on compartmental epidemiological
models derived from SIR. The more advanced models
include seasonality, so for example there are less chances
of contagion when schools are closed. A very much stud-
ied case is that of the African nation Niger, where the
rainy season hampers movements (and hence the occa-
sions of contagion) to the point that, during those
months, the CCS of measles passes from 500,000 to 1.5
million [4].
 3 Epidemics: The Mathematics of Contagion 43

3.6 The Peak, the Plateau,

the Breakneck Slopes (and
the Climb-Ups)
As we have observed, in the very first phase of the epidemic
the models describe a contagion curve of exponential type.
Even in absence of containment measures, the curve does
not climb indefinitely (if this happened the entire popula-
tion would go extinct). At a certain point the curve reaches
a maximum, or peak, and then either goes back down or it
stabilises on an endemic value. Containment measures are
essentially in place to lower the peak value. The latter is
sometimes less pronounced, looking more like a plateau
than a proper peak. Once this value has been reached there
is a slow descent in the number of new infections (Fig. 3.5).
The curve smoothes out according to the harshness and
success of the containment measures, which can slow down
and flatten the growth in contagions, thus attaining a peak

Fig. 3.5 The contagion curves with no intervention (blue) and

containment measures (red). The integral between 0 and t, that is,
the shaded area below each curve, provides the total number of
people infected from day 0 to day t
44 A. Quarteroni

that is lower and happens later. This slowdown is essential

to allow the health system to prepare and be ready to treat a
number of infected people not exceeding what the actual
resources can take on. From an exquisitely mathematical
point of view, pursuing that goal means introducing in the
model further variables that represent the mitigation ef-
fects. The extra variables also act as control parameters, that
is, parameters that can be tweaked in order to direct the
system towards more acceptable results. (In mathematical
jargon, one must ensure the solution to the SIR equations
minimises the function counting to overall infection num-
ber.) See Fig. 3.6 for an example: the extent of the NPIs is
described at the top using the colours adopted by Italian
authorities (white-yellow-orange-red-black, from mildest
to harshest). In this way, the scenarios analysed by mathe-
matical models provide a vital support to the health au-
thorities and civil defence. Such contributions are invalu-
able especially during the contagion’s acute phase (where
the contagion curve is exponential). But they can be very
beneficial in later stages too. Often in fact there are back-
lashes in an epidemic, with ensuing new increases of the
curve (albeit less violent) and successive adjustments. Let us
not fool ourselves, though: very virulent epidemics are not
easily tamed by mathematical models. Unpredictability
lurks everywhere, and often there is no precise data avail-
able. Models allow to highlight trends as they happen,
without pretending to give accurate numbers. Yet qualita-
tive considerations can help politicians devise more efficient
NPIs, raise awareness among the population about what is
happening and prompt people to act responsibly.
 3 Epidemics: The Mathematics of Contagion 45

Fig. 3.6 Epidemic trend in Italy for the compartments hospital-

ised, daily positives, and hosted in ICUs (SUIHTER model)

References
1. Chen D., Moulin B., Wu J. (a cura di) (2015), Analyzing and
Modeling Spatial and Temporal Dynamics of Infectious
Diseases, Wiley, Hoboken, nj.
2. Pastore y Piontti A., Perra N., Rossi L., Samay N., Vespignani
A. (2019), Charting the Next Pandemic: Modeling Infectious
Disease Spreading in the Data Science Age, Springer, Cham.
46 A. Quarteroni

3. Bartlett M.S. (1960), The Critical Community Size for Measles

in the United States, Journal of the Royal Statistical Society,
Series A (General), 123 (1), pp. 37-44.
4. Ferrari M.J. et al. (2010), Rural-Urban Gradient in Seasonal
Forcing of Measles Transmission in Niger, Proceedings of the
Royal Society B: Biological Sciences, 277 (1695),
pp. 2775-2782.
5. Parolini N., Dede’ L., Antonietti P.F., Ardenghi G., Miglio E.,
Manzoni A., Pugliese A., Verani M., Quarteroni A. (2021),
SUIHTER: A new mathematical model for COVID-19.
Application to the analysis of the second epidemic outbreak in
Italy. https://www.arxiv.org/abs/2101.03369. Proceedings of
the Royal Society A, 477 (2253), https://doi.org/10.1098/
rspa.2021.0027
6. Parolini N., Dede’ L., Ardenghi G., Quarteroni A. (2022),
Modelling the COVID-19 and the vaccination campaign in
Italy by the SUIHTER model, Infectious Disease Modelling,
7 (2), pp. 45-63.
 4
A Mathematical Heart

Abstract Modelling the circulation of blood is a fluid dy-

namics problem, made complicated by the elasticity of the
arterial walls and the formation of turbulence at the arter-
ies’ bifurcations. Yet the most stimulating challenge is trans-
lating into equations the heart’s behaviour. A complete
mathematical model has been recently developed at
Politecnico di Milano and is being used for diagnosing and
treatment of cardiac dysfunctions.

Twenty-five-or-so years ago I met with a young doctor spe-

cialising in vascular diseases who wanted my advice on a
physically-related matter regarding blood circulation. I was
rather taken aback, because the subject was very far from
my usual field of work, but I immediately realised the prob-
lem was fascinating to a mathematician as well.
How do you go about in such cases, when you have to
face areas of knowledge that are far from your own exper-
tise? The first question one should ask, in total honesty, is

© The Author(s), under exclusive license to Springer Nature 47

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_4
48 A. Quarteroni

whether one can say something relevant on the matter. The

person who asked for advice wants concrete answers, and
ignorance is no excuse. One should therefore understand
the physical and biological phenomena at the heart of the
problem, and then turn them into a model that phrases
them mathematically. Is the model capable of answering the
precise question we were asked? And if so, is it possible, af-
terwards, to return from the model to the concrete situation?
In the case of blood circulation the answer to all above
questions is a sound yes. If we think about it, there is a tight
relationship between medicine and numerical sciences. If
we look at physiological phenomena such as muscle con-
traction, respiration, or blood circulation, we soon realise
how any action, voluntary or involuntary, is the result of
opposing forces produced by an engine and regulated at
various levels by sophisticated mechanisms. Medical sci-
ence, after long observing the natural processes governing
the human body, became concerned with discovering the
causes of ailments and studying their prevention. It does so
through models of healthy behaviours and the early identi-
fication of risk factors. For its part, today mathematics is
able to produce models that translate physiological mecha-
nisms into equations, in some cases simulate the evolution
of diseases, analyse the processes the body uses to absorb
drugs, and so on. As we shall see further on, mathematics
has recently started to enter operating rooms, by helping
surgeons prepare delicate interventions.
In this chapter we shall focus on the circulatory system
which, besides creating exceptional endeavours for mathe-
matics, has an enormous social impact. Vascular diseases, in
fact, are the main cause of death in the western world, being
responsible for roughly 31% of all deaths world-wide (the
percentage is above 40% in Europe). In particular, isch-
aemic heart disease is the primary cause of death in Italy,
representing 29% of the total. More than cancer. Those
 4 A Mathematical Heart 49

who survive a heart attack often become chronic patients.

The disease changes the patient’s way of life and is a big
economic burden on society. In Italy the latest National
Institute for Social Security estimates (INPS) quantify the
direct costs in 16 billion euros per year, plus another 5 due
to indirect costs related to productivity loss. Cardiovascular
diseases represent 19% of the pension scheme’s total budget.
We shall first briefly examine, also from a historical per-
spective, what we know about the vascular system. Then we
will model it in physical-mathematical terms. We will begin
with the arteries, then pass to the engine, the heart, of which
my research team at Politecnico di Milano is trying to create
a digital model, in a project called iHEART financed by the
prestigious European Research Council.

4.1 How Does the Cardiovascular

System Work? An Eternal
Challenge for Philosophers,
Doctors, and Mathematicians
Understanding what the heart’s purpose is and how it works
is a challenge that has fascinated humanity since the dawn
of time. In the third century BCE Aristoteles believed blood
vessels transported vital heat from the heart to the periph-
ery. A little later Praxagoras of Cos naively understood the
different role of arteries and veins, and conjectured that ar-
teries transported air whereas veins carried blood. We have
to wait until Galen (second century CE), the forefather of
western medicine, before it was recognised that arteries
carry blood, too.
Jumping forward 11 centuries, between 1487 and 1513
the great Leonardo da Vinci made his studies of human
anatomy. He was fascinated by the machine of perfections, as
50 A. Quarteroni

he described the human body. Leonardo’s anatomical draw-

ings are the most extraordinary piece of work his times al-
lowed for. He was the first to define, with great precision,
the four cardiac chambers, thus distinguishing the ventri-
cles from the atria (formerly, auricles) and accurately de-
scribing the heart’s cycle. According to Leonardo, this cycle
was caused by the atria’s contraction upon the diastole of
the ventricles, whereas the ventricles’ systole coincided with
the atria’s refilling. He called them motions of the heart
(moti del core).
Another big leap forward bring us to the seventeenth
century, when William Harvey inaugurated modern vascu-
lar research in his treatise Exercitatio anatomica de motu
cardis et sanguinis animalibus (Anatomical exercise on the
motion of the heart and blood in living creatures), which
contains the first-ever correct description of circulation.
Thanks to the beating-heart vivisection of various animals
Harvey saw that the cardiac valves in the veins were de-
signed to open only when blood flowed towards the heart,
and concluded that blood is in motion, and it moves in
a cycle.
In the subsequent century the great Swiss mathemati-
cians Euler and Daniel Bernoulli made decisive advance-
ments in the fluid-dynamical study of blood. In 1730
Bernoulli, then professor of Mathematics and Anatomy in
Basel, formulated, whilst studying blood pressure, the fa-
mous equation of living forces (vis viva in Latin). This equa-
tion establishes the relationship obeyed by the blood’s pres-
sure, density and velocity (more precisely: half the density
times the velocity squared, plus the pressure, is constant).
In 1775 Euler, in the paper Principia pro motu sanguinis per
arteria determinando (Defining the principles for the mo-
tion of blood through arteries), wrote a system of differen-
tial equations (still fundamental and employed in various
 4 A Mathematical Heart 51

contexts, like the study of gases inside a pipe or even air-

plane design) describing the evolution of the mass flow rate
and pressure in a cylindrical, straight ideal blood vessel. A
generalisation of his equations represent today the basic
tool for simulating the blood flow inside the vascular sys-
tem’s intricate network of arteries and veins.
In the nineteenth century, whilst examining the arterial
flow, the French surgeon and physicist Jean Léonard Marie
Poiseuille derived the first simplified mathematical model
of a stationary fluid (not varying with time) inside a cylin-
drical pipe, a model that still bears his name. More or less
around the same time the British polymath Thomas Young
made a cardinal contribution to the study of the properties
of arterial tissues and the nature of the elastic waves propa-
gating through them.
Among the twentieth century contributors we mention
the American physiologist Otto Frank, who put forward a
circulatory model based on the similarities with electrical cir-
cuits, and the British scientist John Womersley, who in study-
ing the blood flow found the non-stationary (varying in
time) equivalent of the Poiseuille flux, and precisely described
the pressure variations throughout the entire cardiac cycle.
But perhaps the greatest turning point of the twentieth cen-
tury occurred in the applications of all the accumulated
knowledge: cardiac surgery, thanks to open-­heart operations,
transplants and bypasses, has saved the lives of countless peo-
ple, who had no hope of surviving before its introduction.

4.2 The Models Today

The past three decades have witnessed a vigorous growth of
mathematical models for the circulatory system. Today we
can build a virtual (mathematical) world in which we can
solve equations that reproduce with great authenticity and
52 A. Quarteroni

precision the complex mechanical and biochemical pro-

cesses occurring in the arteries, veins, and even inside the
heart, as we shall see shortly [1]. An important physical
characteristic is that the arteries’ walls are elastic, and de-
form to facilitate the passage of blood.
A numerical simulation is illustrated in Fig. 4.1. Arter-
ies are made of three layers, called (from inside outwards)
tunica intima, tunica media and tunica adventitia. Across
the arterial walls and the blood flow there is a constant
energy exchange that mathematical models nowadays can
simulate. This allows, in particular, to see how blood in-
teracts with the endothelial cells that coat the vessels and
the heart’s innermost tissue, and how this affects their

Fig. 4.1 Mathematical simulation of an artery’s deformation

 4 A Mathematical Heart 53

direction, deformation and possible damages. This is im-

portant also in view of understanding how degenerative
processes arise, like those creating obstructions or the for-
mation of plaques.
Furthermore, the blood flow equations consent to de-
scribe the transport, diffusion and absorption of several
chemical components (like oxygen, lipids and drugs) in the
various layers of the arterial wall, and ultimately allow to
understand whether our tissues are properly nourished.
More generally, today’s models and simulations can re-
construct a number of anomalous behaviours displayed by
the blood flow, which are possible contributing factors to
the start of a pathology. An important case, just as an ex-
ample, are the vortices occurring past the carotid bifurca-
tion (Fig. 4.2). The two carotids (left and right) are arteries
that originate from the aorta immediately after the latter
exits the heart, at the top of the aortic arch. Inside the neck
each carotid splits into two branches, one of which brings

Fig. 4.2 Blood pulse inside the carotid

54 A. Quarteroni

blood directly to the brain. It is here that the vortices occur.

The problem is very general: vortices happen in vessels with
large bends (the aortic arch for instance, or the coronaries)
or whenever the pressure the blood exerts on the wall jumps
in the fraction of a second passing between the systole (the
heart’s contraction) and the diastole (the dilation).
Nowadays, also thanks to mathematical models, we begin
to understand in detail phenomena like haemodynamics at
a local level, the effects a modification in the walls has on
blood flow, or the medium/long-term gradual adaptation of
the whole system after a surgery (for instance after the re-
moval of plaques).
Another case where mathematical models have made
massive advancements, with useful applications, is coronar-
ies. Coronaries draw blood from the first aortic tract, just
outside the heart, and they convey it back to the heart or
better, to the cardiac muscle, thus guaranteeing myocardial
cells are nourished (the myocardium is the heart’s muscular
tissue).
Unfortunately, the coronaries have the tendency to oc-
clude and hence deprive the myocardium of the necessary
irrigation (doctors call it perfusion), which in the worst
cases leads to a heart attack. In the majority of cases surgery
is imperative to restore the flow, and this can happen in
many ways: with a bypass, an angioplasty or by placing
a stent.
In order to create a bypass one uses a portion of one of
the patient’s healthy arteries (or veins) to create an alterna-
tive route for blood. The deviation is sown on the vessel
before and after the obstruction. There are many ways for
doing this, and we may ask what will be the optimal way,
the one benefitting the patient most. Obviously the con-
cept of optimality is very subjective. Can mathematics help
in this respect?
 4 A Mathematical Heart 55

Despite this kind of surgery has become a routine, each

year 8% of patients with a bypass implant risk a new occlu-
sion. Moreover, after 10 years from the first operation a sig-
nificant percentage must undergo surgery again (10–15%
of arterial bypasses and 50% of venous ones; the latter
grows to 85% after 15 years). Performing surgery a second
time involves a high risk of complications, so one should be
very careful to avoid problems after the first operation. In
this respect, the simulation of the circulation in a coronary
bypass, especially downstream from the reattachment point,
can make us understand how much the arterial morphology
affects the flow and hence the post-op evolution.

4.3 Mathematics
in the Operating Room
Let us return to the model’s application. A procedure less
invasive than the coronary bypass seen above consists in the
intravenous introduction of a stent. A stent is a microstruc-
ture made of metallic threads interwoven into a suitable
shape. It is inserted closed, and then it is made to expand
around the obstruction until the artery returns to its origi-
nal diameter. This kind of device typically stays in place
forever.
The implant of a stent is a much less invasive and oner-
ous operation than a bypass, apart from being cheaper eco-
nomically. Inserting the device in a vessel (Fig. 4.3), whether
in a coronary, a carotid or another artery, changes the wall’s
stiffness and elastic properties. Therefore the treated part
will, after the operation, react in a completely different way
to the passage of the blood pumped by the heart. The stiff-
ness in particular will increase. This causes one component
of the propagating wave to be reflected and slowed down in
56 A. Quarteroni

Fig. 4.3 Mathematical simulation of the effects of a stent [inser-

tion] on the elasticity and stiffness properties of the arterial wall

the proximal area (near the heart), whereas it gets acceler-

ated in the distal area (towards the periphery, far from the
heart). Sometimes this generates a significant variation of
the force the blood pressure exerts on the arterial wall.
Mathematical models take this phenomenon in account
and are useful in the process of designing more effi-
cient stents.
Another source of problems is the fact that devices inter-
act with the cells of the vessels’ walls they are in direct con-
tact with. Metals such as iron, nickel etc., constituting cer-
tain stent types, can interact with the cells of the tunica
media and intima, thus causing an inflammatory response.
In turn, that may trigger a rampant proliferation of smooth-­
muscle cells inside the tunica media and a reduction of the
lumen (the hollow part of the vessel in which the blood
flows). To counter this phenomenon, research has devised
drug-eluting stents, which are micro-coated by a special
material capable of storing and slowly releasing an anti-­
inflammatory active principle. Tiny holes can be made in a
stent’s fibres and then filled with layers of different sub-
stances. To that end, mathematical modelling and com-
puter simulations allow to assess the behaviour of several
configurations (Fig. 4.4), both quickly and in a much
cheaper way compared with traditional testing [2, 3].
 4 A Mathematical Heart 57

Fig. 4.4 Drug-eluting stent simulation

4.4 The Blood’s Equations

Blood, in the eyes of physics, is not even a fluid, but a sus-
pension of solid particles (white cells, red cells and platelets)
in a fluid substance called plasma. In larger arteries (of di-
ameter exceeding 2 mm-like epicardial coronaries and ca-
rotids) though, the behaviour of blood is completely similar
to that of a homogeneous fluid, and therefore we shall treat
is as such.
The model employed in this situation is based on the
celebrated Navier-Stokes equations, which owe their name
to the French physicist Claude-Louis Navier (1785–1836)
and the Irish physicist George Gabriel Stokes (1819–1903).
One of the Navier-Stokes equations rephrases Newton’s sec-
ond law: the mass of a fluid times the acceleration, at each
point in the vessel and at each instant, equals the sum of all
forces acting at that point. But while in Chap. 2 we applied
the equations to the air (which is a fluid, let us not forget it)
58 A. Quarteroni

for the purpose of weather forecasting, here we need not

include the force due to the earth rotation, which at present
is negligible. We will, moreover, suppose the temperature is
constant (in reality it is not, but it makes for an acceptable
simplification), and so will the density. Consequently we
shall not need the temperature equation, and the mass con-
servation law will not contain the density’s variation in time.
The Navier-Stokes equations produce as solution the
blood pressure and its velocity along the three spatial direc-
tions. The difficulty arises from the fact they should be
solved in a small region (the arterial lumen, the cavity oc-
cupied by blood) that varies with time, because the arterial
walls are distensible (luckily for us). Their contraction and
relaxation, by the way, depend on the flow, so they repre-
sent further unknowns. To describe the walls’ motion we
will then need another equation, which will be coupled to
the Navier-Stokes equations governing the motion of blood.
In formulating the new relations we will take inspiration
from Newton’s second law once again, but we will have to
include the special structure of the arterial wall. As was
mentioned, the wall is made of three layers separated by
thin membranes, different in structure and mechanical be-
haviour. We therefore need another relation, called constitu-
tive equation, for translating that particular structure in a
system of relations between the forces acting on the walls
and the deformations of the walls themselves.
There remains to describe the interaction of the blood
flow with the arterial wall. We will do that using two cou-
pling equations, expressing a kinematic condition (relative
to the motion of blood) and a dynamical condition (relative
to the force blood applies to the walls). The former trans-
lates the property that every blood particle that hits the wall
will remain stuck to it, so from that moment it will move in
sync with the wall (because blood is a viscous fluid). The
dynamical equation, instead, states that the force exerted by
 4 A Mathematical Heart 59

blood on the wall is equal, but with opposite sign, to the

force that the wall exerts on blood when deforming.
Summarising, the mathematical model will consist of:
the Navier-Stokes equations for the fluid, the equations for
wall motion, the constitutive equation and the coupling
equations (kinematic and dynamical).
Navier-Stokes equations for the fluid (blood):


fluid (blood) velocity

viscosity density pressure

 v        
  v  
 v   v
    p 0
 
t
acceleration diffusionn inertial pressure
term term term gradient
 
 v 0

Elasto-dynamics equations for the arterial wall:

wall displacement
 
2 s     
   s    0

 
2
t
acceleration spatial Cauchy stress
term divergence tensor

Constitutive law:
elastic coefficients of the arterial wall

    s 
 
 s   f  s , , , 
 t 
60 A. Quarteroni

Kinematic condition:

 s
v
t

Dynamical condition:

unit vector orthogonal to the

epithelium (the lumen-artery interface)
    
v
  p n   s   n
n

4.5 At the Heart of the Problem

Until this point we have talked only about circulation. But
what about the system’s engine itself, the heart? Surely its
modelling must be an even more complex endeavour.
Well, on the 1st of December 2017 we inaugurated an
ambitious project iHEART, for Integrated Heart, at
Politecnico di Milano. The goal is to realise a virtual mathe-
matical model, made of a system of equations, capable of
integrating every one of the processes of the cardiac function,
and therefore able to reproduce the human heart’s behaviour.
If one could build a virtual model capable of simulating
all the functions and dysfunctions of the heart, there would
be enormous rewards in terms of prevention and assistance
to doctors. Even more if the study could be personalised, so
to create a virtual heart for every single individual.
Our hearts beat because every second a spontaneous elec-
trical impulse is generated in a specific region called the
sino-atrial node, which is made of cells that excite involun-
tarily and thus make for a natural pacemaker of sorts. From
there the sparkle produces an electric field propagating to
 4 A Mathematical Heart 61

the atria, then to the entire myocardium, thanks to a net-

work of special fibres called Purkinje fibres; they work as a
large network of fast lanes. This phenomenon is described
by a series of equations, called the electrophysiology model.
Because of the electric impulse all cardiomyocites (the
elementary cells of our heart) excite, depolarise and repola-
rise to give rise to a sequence of contractions and expan-
sions. These induce, at the macroscopic level, the move-
ment of the entire cardiac muscle, thus facilitating the
ventricles’ contraction and relaxation. The venous blood,
full of carbon dioxide and other toxic waste, is sucked in the
right atrium by the vena cava, and then passes in the right
ventricle to be pumped through the pulmonary arteries to
the lungs. Inside the lungs the venous blood releases its
toxic components (eliminated through the respiration) and
loads itself with oxygen. Through the pulmonary veins this
enriched blood reaches the right atrium, then the right ven-
tricle upon the atrium’s relaxation. The successive contrac-
tion of the ventricle makes the blood overcome the aortic
valve’s resistance and get pumped in the ascending aorta,
from where it reaches every corner of our body.
How can we represent all these processes mathemati-
cally? Divide and conquer: we begin by breaking down the
problem in its main components. In the case at hand we
will have: a) equations describing the electric potential cre-
ated across cell membranes (the so-called transmembrane
potential) and its propagation; b) equations describing the
intensity of current produced in each cell by depolarisation
and polarisation; c) equations describing the rate of con-
traction and relaxation of every single myofibril; d) me-
chanical equations allowing to simulate the motion of the
whole myocardium, similar to those we met for the interac-
tion between blood flow and arterial wall movement.
What is missing are the equations controlling how the
fluid (blood) behaves inside atria and ventricles, and the
62 A. Quarteroni

Fig. 4.5 A conceptual map that illustrates the different physical

processes characterising the functioning of the left ventricle and
their interactions

way our coronaries perfuse the myocardium. And that is

not all. We must describe the dynamics of the four valves
present in our hearts as well: the two that connect the upper
and lower chambers (mitral and tricuspid), the pulmonary
valve directing the venous blood to the lungs, and the aortic
valve through which the arterial blood is pumped into the
ascending aorta every second. There is no need to remark
that all these models are interdependent, meaning that
there are further equations that link the various unknowns
together (Fig. 4.5).
The resulting system is the virtual model of the heart. It
is made of 30 or so nonlinear and time-varying differential
equations, whose number after the numerical approxima-
tions balloons to tens, or hundreds, of millions [4, 5]. To
 4 A Mathematical Heart 63

Fig. 4.6 A mathematical model of the heart’s activity

solve them one needs a supercomputer capable of perform-

ing thousands of billions of algebraic operations per second
(Fig. 4.6).
For it to function, this model requires data, and of vary-
ing nature. To begin with, we should start from clinical im-
ages (such as magnetic resonance, CT-scans) to reconstruct
by geometrical techniques the shape of a specific patient’s
heart. But we will also need information on how the myo-
cardial fibres are arranged, plus other parameters describing
its electric conductivity. After a stroke, for example, there
are scarred areas of the myocardium where the conductivity
is, if not zero, much reduced.
As is easy to guess, these data are often hard to get hold
of. It is for this reason that the creation of an integrated
model of the human heart goes through the crucial collabo-
ration with doctors.
64 A. Quarteroni

From the mathematical point of view the heart poses the

typical problems of living structures, which are further
magnified by its intrinsic complexity. First and foremost is
the lack of boundary conditions and, in general, the fact
that not all relevant data are known or attainable. One must
also represent non-standard objects (each one of us has their
own heart and arteries), with nonrigid and geometrically
complicated shapes. It comes down to the mathematician’s
creativity to complete the model in a manner that is not too
invasive, that is, without assuming too many of the missing
data nor oversimplifying. It is at this juncture that the ex-
pertise of medics, radiologists, cardiologists and other spe-
cialists becomes fundamental.
The heart is such a complex organ that we will never
reach a completely exhaustive and precise description of it.
Building a virtual, mathematical, heart, only made of equa-
tions, is a tremendous challenge; one that is worth accept-
ing though, for our health would profit enormously from
every small victory.
What should we expect in 10 or 20 years’ time when do-
ing a cardiac checkup? Will we have a personalised model of
our own heart, that will allow for precise diagnoses and
most of all enable the specialist to assess future risks?
We are on the way towards that scenario. At Johns
Hopkins University in Baltimore, for instance, Professor
Natalia Trayanova goes in the operating room to assist car-
diologists in the difficult task of curing arrhythmias and
ventricular tachycardias. She uses the simulations she made
the day before with her fellow mathematicians and engi-
neers, starting from the clinical images (MRIs and CT-scans)
of the hearts of the patients that will undergo surgery.
The cases of collaboration between mathematicians and
doctors, like this one, are more frequent than what we may
think. Today mathematical simulations are employed to
 4 A Mathematical Heart 65

improve the treatment of patients in several areas, including

oncology, pathologies of the cardiovascular system, the fit-
ting of prostheses to replace damaged or compromised
joints, and so forth.
The dream is to pass from a heart that pounds because of
a maths test (who, as a student, did not experience this feel-
ing?) to making the heart beat in a healthier way thanks to
mathematics.

References
1. Formaggia L., Quarteroni A., Veneziani A., Eds. (2009),
Cardiovascular Mathematics. Modeling and Simulation of the
Circulatory System, Springer, Milano-New York.
2. Quarteroni A., Veneziani A., Zunino P. (2001), Mathematical
and Numerical Modelling of Solute Dynamics in Blood Flow
and Arterial Walls, siam Journal of Numerical Analysis, 39 (5),
pp. 1488-1511.
3. Prosi M., Zunino P., Perktold K., Quarteroni A. (2005),
Mathematical and Numerical Models for Transfer of Low-­
Density Lipoproteins Through the Arterial Walls: A New
Methodology for the Model Set Up with Applications to the
Study of Disturbed Lumenal Flow, Journal of Biomechanics,
38, pp. 903-917.
4. Quarteroni A., Dede’ L., Manzoni A., Vergara C. (2019),
Mathematical Modelling of the Human Cardiovascular System.
Data, Numerical Approximation, Clinical Applications,
Cambridge University Press, Cambridge.
5. Quarteroni A., Dede’ L., Regazzoni F., (2022), Modeling the
Cardiac Electromechanical Function: A Mathematical Journey,
Bullettin of the American Mathematical Society, 59 (3),
pp. 371-403.
 5
Mathematics in the Wind

Abstract Optimising the shape of sails, minimising the

drag of water waves and perfecting the aerodynamics of a
hull can guarantee a decisive advantage at the finish line:
Alinghi’s triumphs in the America’s Cup are also a victory
for mathematics.

In the autumn of 2001 the President of École Polytechnique

Fédérale de Lausanne (EPFL), where I taught, asked me to
take part in a meeting. One of the attendees was rather un-
expected: Russell Coutts, the legendary skipper who had
just led New Zealand’s Black Magic team to victory in the
29th and 30th editions of the America’s Cup, the world’s
most prestigious sailing race. During that meeting we were
told that Ernesto Bertarelli, a young successful entrepreneur
and sailing enthusiast, had decided to create a Swiss team,
captained by Coutts, to compete in the 31st edition of the
Cup. The team would be called Alinghi (after the name of

© The Author(s), under exclusive license to Springer Nature 67

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_5
68 A. Quarteroni

the boat Bertarelli had received as a present from his father

as a boy). The EPFL was asked to act as scientific consultant
on the project. I remember the President bantering on how
the media would cover the story (a team from a notoriously
mountainous, and landlocked, country competing in the
America’s Cup …), and then telling me the EPFL would
take on two tasks: the building materials (assigned to a col-
league of mine) and the fluid dynamics design. The latter
would be for me and my CMCS team (Mathematical
Modelling and Scientific Computing) to solve.
No need to say I knew nothing about sailing. To prepare
for the second meeting with Russell Coutts and Grant
Simmer (the famous skipper of Australia II, the first non-
American yacht to win the America’s Cup after 135 years of
uncontested US reign), I had to download from the web
the 50 keywords of the sailing lexicon.
I could not possibly know that from that moment, and
for the next 9 years, Alinghi would become a trusted travel-
ling companion. It was an adrenaline-­fuelled adventure,
that led us to win the Vuitton Cup and then the 31st
America’s Cup in March 2003 in Auckland (beating Black
Magic), the 32nd America’s Cup in July 2007 in Valencia
(defeating team Oracle), and to the February 2010 final,
again in Valencia, that we lost to team Oracle.
I am sure you will be asking yourselves: what does math-
ematics have to do with the America’s Cup? We will get
there soon, but before let us look back at some history.

5.1 A Sports Trophy

with a Glorious History
Apart from the Olympic Games, the America’s Cup is the
world’s oldest sports competition. The first edition took
place on 22 August 1851 at the Isle of Wight in the UK. The
 5 Mathematics in the Wind 69

British Royal Yacht Squadron, with 14 yachts, challenged

the New York Yacht Club (NYYC), which was participating
with one yacht only, a schooner called America. The
Americans won, finishing 8 minutes ahead of the next
yacht, the British Aurora, and were awarded the trophy that
had been prepared to celebrate the first universal exposition
in London.
The loss was a hard blow for the UK’s invincible naval
powerhouse, and the Brits called for a second match to get
even. By regulation this had to be played in waters chosen
by the NYYC. The Americans prevailed again, in one of an
incredible string of triumphs: they were unbeaten in 25
events over 132 years, the longest winning streak in the his-
tory of any sport.
The turning point was 1983, when six challengers (called
syndicates) signed up. To decide who would battle the
Americans over the trophy, it was decided to hold a series of
knock-out regattas, which became the Louis Vuitton Cup.
Also participating in the 1983 edition, for the first time,
was an Italian yacht, Azzurra. It ended up third and, most
relevantly, had the merit of publicising the existence of the
race among the Italian audience. If you happen to know an
Italian named Azzurra born in the mid-80s, now you know
why. The trophy was easily won by the Australian team
Royal Perth Yacht Club, with Australia II (this yacht was
equipped with technical innovations that were kept under
wraps until the end). What is more, Australia II won the
America’s Cup, thus ending the US team’s 132-year-
old reign.
Research and technology stated to play a decisive role in
that edition. The design of Australia II’s winged bulb, for
instance, was inspired by advanced aerodynamical notions,
already exploited in aviation. The New Zealand yacht that
came in second was the first 12-m class yacht with a fibre-
glass hull, rather than the usual aluminum.
70 A. Quarteroni

In 1988 a New Zealand syndicate lodged a surprise chal-

lenge with a whopping 36-m yacht (a return to the yacht
size of the early races). The challenger, the San Diego Yacht
Club, exploited a hidden weak spot in the rules and built
Start and Stripes, a small, 18-m catamaran, both agile and
very fast, that crushed the reigning champions.
From 1988 to 2007 the rules changed to allow only
12-m yachts to race. In 1992, America3 (read: America
cubed) beat the Italian challenger Il Moro di Venezia. In
1995 the Team New Zealand syndicate won the race, with
Russell Coutts as skipper of Black Magic. In 1999–2000, in
the waters of Auckland, Team New Zealand prevailed again
by defeating the Italian challenger Prada Challenge with
Luna Rossa. The Italian yacht, helmed by the Neapolitan
skipper Francesco de Angelis, had won the Louis Vuitton
Cup by beating in the final race the US yacht America One,
captained by Paul Cayard, formerly at the helm of Moro di
Venezia.

5.2 The Swiss Outsider

and the Mathematics of Sails
This is the moment when the history of the America’s Cup
crosses paths with my scientific account. In December 2001
I set up an EPFL team of three, who until that fateful 2nd
of March, 2003 would stay in touch with Grant Simmer’s
design team and Russell Coutts’s sailors. Videoconferences
with the Alinghi team in Auckland (12 hours ahead) were
held weekly at the crack of dawn for us, the end of the eve-
ning for them.
How does one, as a scientist, tackle a subject so distant
from one’s own personal history? The first thing is, clearly,
understanding the problem, to decide whether one is able
 5 Mathematics in the Wind 71

to take it on. “Understanding” here means examining in

detail the physics and engineering aspects that are useful to
the construction of the model. Initially the work is mainly
theoretical, but one must then translate the knowledge ac-
quired into the client’s language, which also involves eco-
nomic aspects and deadlines: compared to a curiosity-
driven research, here there is the constraint of finite
resources and, what is even more compelling, finite time.
The Alinghi job seemed really challenging: a gigantic ob-
ject (24 tons in weight, almost 800 sq. m. of sails when fully
open, a 25-m mast), with a complex geometry, subject to
incredible forces and stresses. Two turbulent fluids are in-
volved: air and water. The conditions of motion vary a lot:
depending on the strength and direction of the wind, there
could be aerodynamic lift effects similar to those on aero-
plane wings, or drag as in parachutes, and so on. The extra-
scientific constraints are – hard to conceive as it may
be – even more severe. Large bags of money are at stake,
and the line between victory and defeat is very, very thin.
America’s Cup races are ferociously competitive: even after
a race that lasts hours, the two boats may cross the finish
line a few centimetres apart. In contrast to what happens in
industrial applications like aircraft design, where large secu-
rity margins are required, here it is essential to reach ex-
treme, razor-edge performances. Designers must push their
creation’s agility features to the limit, without exceeding the
limitations of natural resilience.
An America’s Cup monohull-type vessel (like those com-
peting until the 32nd edition) is made of various compo-
nents: in the water we have the hull, the keel, the bulb, plus
other appendices such as winglets and the rudder; outside
there are the mast and the sails—mainsail, genoa, spinnaker
and gennaker (see Figs. 5.1 and 5.2). In the past there were
very different shapes of hulls, keels and bulbs, but in the
72 A. Quarteroni

Fig. 5.1 Team Alinghi. 2003 America’s Cup, Auckland, New Zea-
land. Picture: ChameleonsEye/Shutterstock

Fig. 5.2 Aerial and submerged components in a sailboat of Amer-

ica’s Cup class
 5 Mathematics in the Wind 73

2000s (till 2007, the last year of monohull racing) a uni-

form standard was agreed upon. Hence it is the tiniest de-
tails that make a difference in terms of results. In 2003
Jerome Milgram, an MIT professor and a veteran consul-
tant to America’s Cup teams, said “America’s Cup yachts
require extreme precision in the design of the hull, the un-
derwater parts and the sails. An innovative hull offering a
1% drag reduction can warrant a gain of 30 seconds at the
finish line.”
Let us come to the mathematics, and divide the prob-
lems in sub-problems. We begin with the hull. To optimise
its performance we must solve the fluid dynamics equations
around the whole boat keeping into account the regatta’s
countless conditions; the variations, for instance, of wind
and sea waves, of the different sailing regimes (upwind,
downwind), of the position and motion of the opponent
vessel. We must also consider the dynamics of the interac-
tion between fluid and structural components, meaning
water and the submerged parts, as well as between air, sails
and mast. Finally, we must model with great precision the
shape and motion of the interface between water and air,
the so-called free surface.
Let us introduce the computational domain Ω, the 3D
region in which the equations must be solved. This domain
consists of a parallelepiped of 300 × 200 × 180 m. The
lower part is occupied by water, the top part by air, and the
boat sits in the middle. On the parallelepiped’s (ideal, not
physical) frontier we must impose boundary conditions re-
flecting the true, experimental boundary conditions needed
to solve the differential problem.
The equations governing the flow around the boat are
the dear old Navier-­Stokes equations, here in the version
for fluids with non-constant density (inhomogeneous flu-
ids). In fact there are two fluids in the game: water (with a
74 A. Quarteroni

density of 1 g per cubic centimetre, that is 1000 kg per cu-

bic metre), and the much less dense air (around 1.3 kg per
cubic metre at a pressure of 1 atmosphere, at sea level and
temperature of 4 degrees Celsius). The viscosity of the two
fluids is very different too.
The Navier-Stokes equations are over 150 years old, but
even today a legion of physicists, engineers and mathemati-
cians keep investigating their qualitative, analytical and
quantitative aspects. To the day, the solutions’ stability,
uniqueness, and direct numerical approximation are diffi-
cult and fascinating fields of investigation. In particular, the
larger the Reynolds number, a parameter equal to the prod-
uct of the boat’s speed times its length, times the inverse of
the fluid viscosity, the more complex the numerical approx-
imation will be. It is easy to imagine, then, that modelling
the motion of an America’s Cup sailing craft, whose
Reynolds number may equal a few million, is a very chal-
lenging goal.
As we have said already, we do not have an explicit for-
mula for the exact solution of the equations, so we have to
resort to their numerical approximation. To that end we
draw over the boat’s surface a grid (or mesh) made of hun-
dreds of thousands of small triangles (see Fig. 5.3, showing
a portion of the submerged part). The process generates a
lattice of millions of tetrahedra (called elements) in the com-
putational domain Ω.
This discretisation process—skipping all intermediate
passages—leads to a system of nonlinear algebraic equa-
tions, whose dimension is usually proportional to the num-
ber of elements. By the way, in a turbulent regime the ele-
ments should, in theory, be small enough to capture the
energy exchange mechanisms. These happen inside increas-
ingly smaller vortices, and when the latter become infini-
tesimal the energy within is dissipated as heat. This energy
 5 Mathematics in the Wind 75

Fig. 5.3 Detail of the surface grid on the submerged part of

the vessel

cascade derives its name from the famous Russian scientist

Andrej Kolmogorov.
The flow around a sailboat of America’s Cup class dis-
plays during a race a turbulent behaviour in proximity of
most of the wet surface. Turbulent flows are characterised
by a non-stationary behaviour, and exhibit coherent (recog-
nisable) vortex structures that interact with one another,
exchange energy and fluctuate in time and space. Under
these conditions it is unthinkable to simulate the Navier-
Stokes equations directly, that is, by an approximation in-
volving elements so small that allow to capture all the sig-
nificant coherent structures. We would need far too many
triangles and tetrahedra, so many that no existing computer
would be able to manage the problem. Hence one relies on
so-called turbulence models, obtained by taking suitable av-
erages in the Navier-Stokes equations.
76 A. Quarteroni

This is most certainly not the end of the story. We now

must account for the deformation of the sails, which are
elastic. The corresponding equation translates the equilib-
rium condition for the forces that act on the sails, which are
three- or two-dimensional structures (depending on
whether or not we take sail thickness into account).
At last, we must write the equations of rigid body dy-
namics applied to the vessel. For this the diagrams of forces
and momenta in Fig. 5.4 will be useful.
So here is our problem, whose solution will faithfully re-
construct the conditions of the regatta: fluid velocity and
pressure in each tiny element (of water and air), displace-
ment of every micro-portion of sail, drag forces (including

Fig. 5.4 Forces and momenta acting on the boat

 5 Mathematics in the Wind 77

the important forces caused by turbulence) acting on every

microportion of hull, keel, bulb and winglets [1, 2].
Navier-Stokes equations:

Cauchy stress gravitational

fluid density tensor acceleration
 
            
  v
      v  v       v, 
p g
t    
fluid vellocity fluid pressure


spatial divergence
  
    v   0
t
Sails equation:

density of fabric
sail displacement weight (force)
 
  d     
 
2
V     V d  fV

t2

acceleration term mechanical streess

Boat’s dynamics equation:

linear acceleration of
the yacht’s center of mass

 

 XG = F
m
mass of yacht
78 A. Quarteroni

transformation matrix
from fixed to inertial frame angularr velocity

   1        
T IT      T I T 1   M
G

force due to the

interaction with flow external force
    

F  Fflux  mg  Fext

momentum due to the

interaction with flow
 
   

MG  M flux  X ext  XG  Fext 
In truth, the Navier-Stokes equations should be rewritten
to account for turbulent effects, for instance by splitting the
variables into averaged terms plus a correction provided by
extra terms that represent the fluctuations. This involves
adapting the equation of conservation of momentum, where
a viscosity appears that depends non-linearly on the velocity,
and also new terms representing the turbulent stresses. To
complete the new system it is necessary to introduce two fur-
ther transport-diffusion equations, one for the turbulent ki-
netic energy and one for its dissipation rate [1, 2]. At last, let
us remark that the Navier-Stokes equations should be written
for both air and water, keeping in account the corresponding
densities and viscosities. This requires yet another two equa-
tions of dynamical and kinematic type at the boundary sepa-
rating air and water (the free surface), plus an extra equation
prescribing the motion of the boundary itself. As one can see,
the full mathematical model for the study of an America’s
Cup sailboat gives the chills to even the best sea-trained
mathematicians [3, 4].
 5 Mathematics in the Wind 79

5.3 The Numerical Simulations

Once we got a model we were able to address numerical
simulations, which mainly regarded three aspects: the
analysis of the various configurations of the keel appendi-
ces; the behaviour of the free surface around the hull (the
wave shape) and its interaction with the appendices; the
aerodynamic flow around the sails [5]. The computations
provided the design team with an ample assortment of
information, which were implemented at different stages
of the design cycle. They were even used once the race
had started, to suitably adapt the shape of several compo-
nents (Fig. 5.5).
The keel appendices are a key factor in the success of
an America’s Cup sailboat. They must guarantee high

Fig. 5.5 Flow lines around and past the bulb

80 A. Quarteroni

performances in different regatta conditions (upwind

and downwind) and in a wide range of wind conditions,
hence of boat velocity.
We considered several design parameters. Regarding the
shape of the bulb we analysed various lateral and vertical
profiles and many transversal sections, each one having its
own plusses and minuses. Elongated bulbs, for instance, al-
low to reduce the pressure drag, but due to the wider wet
surface they undergo a bigger friction.
Winglets were adopted for the first time by Australia II,
the 1983 champion. They have been extensively employed
by almost all participants to the following editions. Their
presence at the bottom of the keel (in analogy to the wing-
lets on aircraft wings) allows to reduce the vortices at those
points and decrease a component of the drag known as
(lift-)induced drag (Fig. 5.6). Despite winglets have been

Fig. 5.6 Pressure distribution on the surface of the appendices

and flow lines around the winglets
 5 Mathematics in the Wind 81

used by every team since, their optimisation with regards to

various parameters (longitudinal position, angle of attack,
twist and sweep angles and wingplan form) is still an open
problem.
In America’s Cup monohull boats, wave drag can repre-
sent a very significant part (up to 60%) of the total drag. An
accurate prediction of this component is therefore decisive
during the design phase. To do that numerical methods are
required that simulate the dynamics of the free surface sep-
arating the air from the water (the wave shape). Using our
numerical simulations we examined several winglet shapes,
focussing on the bow area.
The aerodynamics of sails is another determining factor.
We considered three specific aspects: the estimated forces
acting on the sails in downwind leg, the interaction of two
competing yachts’ sails, the sheltered region created by a
windward opponent. Thanks to the analysis of the propul-
sion force, the lateral force and the pressure distribution,
one deduces that the sails work as a combination of a para-
chute (when the lift is aligned with the direction of thrust)
and a vertical wing (with thrust-aligned drag), as was ob-
served by Richards [6].
On downwind leg, the boat sailing to windward has a
tactical advantage, since it can control the rival and keep
them in the sheltered region, thus significantly reducing
wind. We made numerical simulations of the flow around
the two boats in downwind leg, and we saw that the flow
witnessed by the leeward boat was vastly perturbed by the
other vessel. The pressure distributions on the two boats’
sails are substantially different (as shown in Figs. 5.7 and
5.8). Information of this kind, extracted from the numeri-
cal simulations, is useful during the race as well: it supports
the tactical decisions and helps maximise the overbearing
on the leeward yacht or curb the negative effects [7, 8].
82 A. Quarteroni

Fig. 5.7 Pressure distribution on the two boats in downwind leg

Fig. 5.8 Aerodynamical interaction of two boats sailing downwind

 5 Mathematics in the Wind 83

5.4 How Did It End Up?

On 19 January 2003, Alinghi, helmed by Russell Coutts,
defeats 4–1 Larry Ellison’s Oracle BMW Racing US yacht,
steered by another New Zealander, Chris Dickinson, and
conquers the Louis Vuitton Cup. At EPFL we are flabber-
gasted: it seems like we are living a dream.
The final race starts on 15 February and the challengers
are up against the Black Magic Kiwis. On the 2nd of March,
after five successful regattas, Alinghi triumphs: it is the first
European team to win the America’s Cup. And it is a team
from a mountainous, landlocked country!
The 32nd edition was held in Valencia, Spain, by deci-
sion of the winning ship owner (called the defender). The
challenger is Emirates Team New Zealand (after prevailing
5–0 over the Italian Luna Rossa in the Louis Vuitton Cup
final). The decisive race happens between 23 June–9 July
2007, and once again Alinghi wins.
In the 33rd edition, again held in the Valencia waters
from the 1st to the 25th of February 2010, Alinghi is de-
feated by team Oracle’s trimaran USA17 steered by James
Spithill. This happened after a complex legal battle that
gave the Americans a massive advantage from the outset.
The racing yachts were significantly different: USA17 had
three hulls whereas Alinghi 5 had two; a 68 m mast for the
Americans against a “mere” 62 m for the Swiss, a rigid wing
as opposed to a mobile sail, the US boat’s 17 tons against
Alinghi’s 11. In a nutshell, it was a technologically uneven
race, with an almost foregone outcome.
Alinghi’s adventure ends with that third final in seven years,
and so does my personal venture in the America’s Cup. In the
ensuing years I was contacted by other teams, but I preferred
to wrap it up with the Swiss experience. My mathematical
adventures in high-level sports did not end with sailing,
though. Alinghi was without a doubt an important stepping
stone, that later on allowed me to engage—mathematically
84 A. Quarteroni

speaking!—in various sports such as rowing, swimming, vol-

leyball and (of course) football. The general goal is to transfer
to many sports contexts the knowledge of mathematical mod-
elling, big data science, artificial intelligence and machine
learning. The aim is to improve athletes’ individual perfor-
mances and the strategies and tactics of team sports: an excit-
ing way to use mathematics at the highest levels.

References
1. Parolini N., Quarteroni A. (2005), Mathematical Models and
Simulation for the America’s Cup, Computer Methods in
Applied Mechanics and Engineering, 194, pp. 1001-1026.
2. Parolini N., Quarteroni A. (2004), Simulazione numerica per
la Coppa America di vela, Bollettino U.M.I., Serie VIII,
7A, pp. 1-15.
3. Mohammadi P., Pironneau O. (1994), Analysis of the
K-Epsilon Turbulence Model, Masson, Paris.
4. Cowles G., Parolini N., Sawley M. (2003), Numerical
Simulation using RANS-­ based Tools for America’s Cup
Design, Proceedings of the 16th Chesapeake Sailing Yacht
Symposium, Annapolis, MD.
5. Quarteroni A. (2009), Mathematical Models in Science and
Engineering, Notices of the AMS, 56 (1), pp. 10-19.
6. Richards P.J., Jonhson A., Stanton S. (2001), America’s Cup
Downwind Sails–Vertical Wings or Horizontal Parachutes?,
Journal of Wind Engineering and Industrial Aerodynamics,
89, pp. 1565-1577.
7. Quarteroni A., Sala M., Sawley M.L., Parolini N., Cowles
G. W. (2003), Mathematical Modelling and Visualisation of
Complex Three-dimensional Flows, in Hege H.C., Polthier
K. (Eds.), Visualisation and Mathematics III, Springer,
Heidelberg-Berlin.
8. Parolini N., Quarteroni A. (2004), Numerical Simulation for
Yacht Design, Proceedings of the 6th Conference on
Informatics and Mathematics, Athens, hercma 2003
(E.A. Lipitakis Ed.), 1, pp. 38-44.
 6
Flying on Sun Power

Abstract Technology and sustainability: this is the chal-

lenge of Solar Impulse, an experimental solar-powered air-
craft designed to circumnavigate the globe. Mathematical
techniques in multi-objective optimisation have allowed to
accommodate the multiple design needs arising from differ-
ent areas, like materials science, propulsion and aero-
dynamics.

The dream of flying is as old as the world: just think of

Leonardo da Vinci’s magnificent yet unachievable ma-
chines. More than a century after the Wright brothers’ first
flight on the dunes of Kitty Hawk, North Carolina, on 17
December 1903, and following the enormous expansion of
trade routes across the world, a new era in the history of
aviation may be beginning: the challenge of lighter and
cleaner aircrafts. On 4 March 2005 Steve Fossett, a man
used to great records, landed the Virgin Atlantic GlobalFlyer

© The Author(s), under exclusive license to Springer Nature 85

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_6
86 A. Quarteroni

in Salina, Kansas, after the first non-stop solo flight around

the globe. He flew 67 hours, 2 minutes and 38 seconds on
a single turbofan engine [1]. In parallel NASA inaugurated
in 1999 the Helios Prototype programme, aimed at building
a solar-powered, ultra-light experimental aircraft. In 2001
the prototype unofficially broke the world record by reach-
ing the altitude of 29,524 m. The enterprise did not have a
happy ending though: on 26 June 2003 the aircraft crashed
into the Pacific ocean due to severe instability issues.
The baton was picked up by Bertrand Piccard, who in
2003 launched the Solar Impulse project, a solar-powered
aircraft for circumnavigating the earth without stopover.
The challenge is blending mankind’s dream of adventure
and discovery with the respect for the environment, and
primarily promote the idea that technology can evolve with
clear objectives for sustainable development (Fig. 6.1).
The École Polytechnique Fédérale de Lausanne (EPFL)
acted as scientific consultant during the design phase of the

Fig. 6.1 Solar Impulse, a project for the first solar-powered cir-
cumnavigation of the globe. Photo: Frederic Legrand—COMEO/
Shutterstock
 6 Flying on Sun Power 87

Solar Impulse prototype, by providing human and scien-

tific resources from over ten fields of knowledge, including
ultralight materials, energy storage and transformation and
new man-machine interface paradigms [2].
My CMCS team came up with the mathematical model
in the earliest phase of conceptual design, using techniques
called multi-objective optimisation. These techniques are es-
sential to find solutions that harmonise design requirements
stemming from a number of different contexts (here, mate-
rials science, propulsion and aerodynamics), which pre-
cisely for their diversity might be conflicting. But before we
talk about models let us review some history.

6.1 The Piccards, a Family

of Explorers
The Piccards, from Auguste to his grandson Bertrand, pass-
ing through Jacques (son of Auguste), are intrepid by family
tradition. Auguste Piccard, born in 1884 in Basel, was pro-
fessor of physics at ETH, the Federal Institute of Technology
in Zurich (where Einstein studied and taught) and then at
Brussels University. A friend of Albert Einstein and Marie
Curie, he contributed to modern aviation and space explo-
ration, inventing the pressurised gondola and the strato-
spheric balloon. He explored the stratosphere in 1931 and
1932, reaching an altitude of 15,781 and 16,201 m respec-
tively, in order to study cosmic rays. On those occasions he
was the first man to observe the curvature of the Earth with
his own eyes. He applied the same principle of stratospheric
ballooning to the exploration of the oceans’ depths, and
built a revolutionary submarine called Bathyscaphe. Along
with his son Jacques, he dove in the bathyscaphe to the
depth of 3150 m, thus becoming the man of the two
88 A. Quarteroni

extremes: no one before him had flown to such an altitude

and explored the ocean abyss.
Jacques Piccard, born in Brussels in 1922, started out
reading economics. Because of his contacts in the world of
finance he was able to gather the money to fund his father’s
second bathyscaphe (called Trieste). He then abandoned his
career and started working with Auguste on the submarine’s
construction. Together with his father he broke several re-
cords, among which the deepest location ever reached: the
bottom of the Mariana Trench, at a depth of 10,196 m.
After his father’s death he carried out the family tradition
by constructing four mesoscaphes (submarines for medium
depths) including the famous Auguste Piccard, the world’s
first passenger submarine. (During Lausanne’s 1964 na-
tional exhibition 33,000 tourists explored the depths of
Lake Geneva in the Auguste.) Among other vehicles we
mention the Ben Franklin, with which Jacques studied the
Gulf current covering 3000 km in one month in 1969, and
the F.-A. Forel, a easily transportable submersible in which
he went on more than 2000 scientific and didactical mis-
sions in European lakes and in the Mediterranean.
Bertrand Piccard was born in Lausanne in 1958. Since
an early age he was interested in studying the human behav-
iour in extreme conditions. He was one of the pioneers of
hang gliding and micro-light flying in the 1970s, and be-
came European hang-glider aerobatics champion in 1985.
After switching to ballooning, in 1992 he won with Wim
Verstraeten the first transatlantic balloon race (the Chrysler
Challenge). He then launched the Breitling Orbiter project
(Fig. 6.2), a vehicle designed for flying non-stop around the
globe, and piloted all three successive attempts (alas, all
failed). In 1999, along with the British aviator Brian Jones,
Piccard completed the first balloon flight around the globe,
which was also the longest flight ever made for distance and
 6 Flying on Sun Power 89

Fig. 6.2 The Breitling Orbiter 3. Foto: Cedric Favero / VWPics /

Alamy Foto Stock

duration. They took off in Switzerland and landed in Egypt

after 19 days and 21 hours, from 1 to 20 March 1999 [3].
In 2003, as mentioned, he embarked on the Solar Impulse
enterprise, which besides an adventure is also a project bear-
ing important technological and social commitments.

6.2 Ending the Fossil-Fuel Era

To the day the primary energy source and the bedrock of
the global economy is oil. Its derivatives are everywhere:
fuel for cars, aeroplanes and heating systems, asphalt for
roads, plastic components and so on. But this resource is
finite. It is estimated that the planet’s reserves amount to
3000 billion barrels. At the start of this millennium 1000
have been already depleted, 1000 have been discovered
and another 1000 are still to be found (this estimate has
the advantage of being easy: one third, one third,
one third).
90 A. Quarteroni

According to some studies, in certain regions of the

world oil production has already peaked (in the US, for
example, it might have happened in November 2017), and
is now declining. Some predict that the reserves will be ex-
hausted within 50 years. These studies are based on the as-
sumption that oil consumption has a constant growth in
time, so that a small fluctuation in the global demand
would suffice to change these numbers.
Perhaps the most plausible scenario is different, though,
and we will not reach the point of exhausting resources
completely. According to this model, the reserves’ depletion
will make the price of oil increase until it will be no longer
profitable to extract it. Another factor that will curb the use
of oil derivatives is the increase of greenhouse gases in the
atmosphere. Greenhouse gases, in particular water vapour
(H2O), ozone (O3) and carbon dioxide (CO2), usually con-
sidered the main culprits of global warming, are mostly
produced when burning fossil fuels. According to NASA’s
Earth Observatory, if we take into account the plausible sce-
narios of fossil energy consumption the average tempera-
ture on Earth may increase by 2–6 °C by the end of the
twenty-first century. Part of this warming will happen any-
way, even if we reduce greenhouse gas emissions, because of
the inertial effect due to the adjustment of system Earth to
environmental changes that have already occurred [4]. The
global climate will change radically: the polar caps will
melt, sea levels will rise, tropical storms will become more
severe, ecosystems will be destroyed. Given the complexity
and interdependence of the phenomena we are considering,
the degree of reliability of these analyses is clearly not the
highest.
One other thing is certain: at the end of this century our
society will be completely different from the one we know
now. We are all aware we must decide whether to maintain
 6 Flying on Sun Power 91

the current dynamics and bear the brunt (which might be

dramatic), or try to steer course with a gradual transition to
energy forms that are alternative to fossil fuels. For this to
happen it is necessary to promote the idea of sustainable
growth, that is, meeting the needs of the present without
compromising the chances of future generations of meeting
theirs. There is no doubt science will have to play a crucial
role in this transition phase.

6.3 The Solar Impulse Mission:

The Challenges
The Solar Impulse project was born, among other things,
precisely to promote sustainable growth. It uses renewable
energy, stimulates the susceptibility to environmental issues
and strengthens the idea that technology can help create a
novel model of growth. The project’s ultimate objective was
to fly around the globe an aircraft powered by green energy
only (in particular, solar energy) and free of polluting emis-
sions. Mission accomplished: the aircraft left Abu Dhabi on
9 March 2015 headed eastbound, and returned there on 26
July 2016, more than 16 months later (it had to make a
long stop-over to repair the significant battery damage
caused by heat on the longest stretch, from Japan to Hawaii,
in July 2015).
The design of an aircraft of this kind requires that we ad-
dress several technological challenges. The main one is to
optimise energy harvesting and consumption. Considering
atmospheric absorption and cloud reflection, the sunlight
power reaching the Earth is about 1020 kW/m2 (at sea
level). That is quite a number: the solar radiation hitting the
entire surface of the Earth every minute exceeds the annual
energy consumption of the whole planet. The Sun could
92 A. Quarteroni

therefore meet our energy demand, if only we were able to

exploit it properly. On the other hand it is clear that the
Sun is not a regular source of energy: it is obviously not
available in areas under the cover of darkness. So we must
find a way to store energy for later use, and one of the key
aspects of the project was exactly to maximise solar energy
accumulation. Wide wings were chosen to accommodate
large photovoltaic cells and high-capacity power batteries.
During the cruise of Solar Impulse, the harvested energy can
be handled in two ways: it is stored in the batteries to in-
crease the energy reserve for the night, or it gets used im-
mediately to power the electric engines.
Choosing a large wingspan allows to maximise solar en-
ergy harvest in daylight, but it reflects negatively on the
structural aspects, since the aircraft becomes heavier and
hence it uses up more energy. To remedy the problem it
then becomes essential to involve last-generation, ultralight
composite materials.
The aircraft is also subjected to big temperature varia-
tions during the mission, at different times of day and dif-
ferent altitudes. The temperature difference between the
upper and lower wing surface can reach 60 °C at peak expo-
sition to sunlight. Moreover, the batteries become less effi-
cient when it is too cold (typically, below 0 °C). So it is
necessary to isolate the batteries inside the wings and guar-
antee the materials do not undergo thermal shocks that
might compromise functionality.
A further (and certainly fundamental) challenge regards
the pilot’s safety. In daylight, Solar Impulse reaches altitudes
beyond 10,000 m, where the temperature goes below
−50 °C and the air is extremely rarefied. So one needs an
efficient thermal insulation and a suitable pressurisation
system. Moreover, an automatic steering system must be
 6 Flying on Sun Power 93

designed to avoid an excessive burden on the pilot over

long hauls.
Many challenges and demands, at times contrasting: all
of which has made the design of the solar aircraft a true
technological bet. Therefore every design choice was care-
fully examined, because it has a potentially great impact on
the general behaviour of the aircraft.

6.4 Mathematics Comes into Play

To design Solar Impulse it became necessary to literally re-
think the concept of aircraft. Its configuration could not be
extrapolated from existent ones or from standard design
strategies. One had to develop aerodynamical models anew,
for the structural behaviour’s analysis, the handling of en-
ergy and propulsion, and for integrating these under a
global optimisation framework.
In a case like this, it is not possible to keep the involved
fields separated and try to optimise the single components
independently, because the specific choices in one aspect of
the project could strongly affect the system’s behaviour in
other contexts. The only possible approach is then an inte-
grated optimisation strategy, that accounts for the different
components and their interactions (we will see some details
in the following section).
The Solar Impulse project is based on mathematical mod-
els of high complexity, which translate into algorithms with
great computational costs. Hence it becomes necessary to
come up with simplified models. Luckily, aerodynamics is
one typical field where it is possible to define a hierarchy of
models, each with different levels of completeness, precision
and computational complexity, that can be integrated ef-
ficiently.
94 A. Quarteroni

Let us consider for instance how to address the problem

of designing the profiles of wings and stabilisers, whose aim
is to reduce air drag and optimise the aircraft’s efficiency
and stability. For this we shall use a so-called potential-flow
model [5], which simplifies reality in one precise respect: it
is founded on the assumption that the flow is irrotational
(also known as curl-free), meaning there are no vortices. In
this way the model is very efficient computationally (calcu-
lations take little time). A potential-flow model is used to
study the wing’s aerodynamical behaviour, that is, to pre-
dict air drag, cruising speed and lift (which, being perpen-
dicular to the aircraft’s direction of motion, is in practice
the component of the force supporting the aircraft). It is
also employed to estimate how the air behaves along the
wing profile (whether the flow is laminar or turbulent). In
order to simulate the 3D flow’s behaviour around the air-
craft, instead, a hierarchically higher model is necessary: the
latter is more complete, and is based on the solution to the
Navier-Stokes equations [6] we met in the previous chapters.
The passage from the physical model to the computa-
tional one is facilitated by programs such as CAD (Computer
Aided Design), that allow to represent the entire aircraft us-
ing a collection of geometrical shapes and define several
computational domains. The flow-governing equations, as
we have seen elsewhere, must be discretised. To do that we
subdivide the three-dimensional domain into a computa-
tional grid (or mesh) made of elementary cells, shaped as
small pyramids (tetrahedra) or cubes (hexahedra), and we
impose that the equations hold locally on each element of
the grid.
In order for the simulation to be realistic, the mesh must
not be uniform, but should be refined suitably in the areas
with steep gradients, that is, where the variables describing
the solutions (especially velocity, pressure and air
 6 Flying on Sun Power 95

Fig. 6.3 Example of computational mesh

temperature) change suddenly. See Fig. 6.3 for an example.

This may lead to a very large number of elements (in the
order of millions, or tens of millions for the finest grids)
and to algebraic problems of prohibitive dimension, which
can be solved only by resorting to algorithms implemented
on large parallel-computer architectures.
The results of these numerical simulations are essential to
understand in detail how the air flow behaves around the
aircraft, and how, in particular, to capture the effects on the
overall performance of phenomena such as vortex shedding
or the presence of areas of air recirculation.
Figure 6.4 shows two examples of 3D simulations: in the
top picture we have the typical vortex at the wing tip, high-
lighted by the streamlines. Below we see the distribution of
pressure values on the wing surface and the engine gondola.
96 A. Quarteroni

Fig. 6.4 Two examples of 3D airflow simulations around the

aircraft
 6 Flying on Sun Power 97

6.5 Multidisciplinary Optimisation

In a complex project like Solar Impulse suitable trade-offs
must be found to satisfy the needs, at times conflicting, of
each component. Consider for example the simple question
“how many batteries should be fitted onboard?”. As regards
the propulsion, the answer is “as many as possible”, since
the electric energy produced by the photovoltaic cells is the
only source available, so we would want to store as much of
it as possible. On the other hand batteries are heavy, and
too many might compromise the mission, for a number of
reasons. First, a bigger weight requires more energy to take
the aircraft higher, or just to support it at a constant alti-
tude. Second, batteries must be kept at a temperature as
uniform as possible to work properly; increasing their num-
ber would require more energy to heat them (it is cold up
there). Third, a bigger load would require a more robust
structure and hence a further increase of the overall weight
of the aircraft. So it is clear that calculating the ideal num-
ber of batteries is not easy: we must optimise it and keep
many constraints in account at the same time.
Problems of this sort can be tackled using a multidisci-
plinary optimisation approach. The latter, also known as
multi-objective optimization, allows to integrate the models
used in several fields (here, structural mechanics, aerody-
namics, propulsion, thermal analysis). This situation shows
up in countless other contexts, say automotive design,
where one would like to maximise the performance of a
new car and at the same time minimise the consumption
and maximise comfort (by reducing noise and vibrations).
The objective functions are the quantities to be optimised
(either maximised or minimised). The problems typically
involve several objective functions. It is easy to imagine that
there is no single solution optimising all objective
98 A. Quarteroni

functions, in general. There exist, though, many solutions

(theoretically, infinitely many) that represent acceptable
trade-­offs. Objective functions are called Pareto optimal, or
nondominated, provided they satisfy the following property:
none can be improved without degrading some of the others.
We can express the above property more precisely in
mathematical terms. Suppose our multi-objective optimisa-
tion problem consists in finding a vector x of dimension
n ≥ 2 that minimises the objective functions f1(x), f2(x), ...,
fn(x). (Note that there is no loss in generality by assuming
we must minimise everything: if one objective function had
to be maximised, it would be enough to flip its sign and
then minimise. This is a standard trick of mathematicians.)
The design parameters are numerical values describing an
acceptable configuration. They may be quantities that vary
continuously (the wing span, the inclination of the wing
profile to the airflow), or in a discrete fashion (number of
solar cells, of batteries etc.).
As we said, in general there are no vectors x minimising
all objective functions simultaneously. So one looks for
trade-off solutions, called Pareto optimal solutions. Let us
start with a definition. We say a solution x1(Pareto) domi-
nates another solution x2 if fi(x1) is less than or equal to fi(x2)
for all indices i = 1,...,n, and if for at least one index j, fj(x1)
is strictly smaller than fj(x2). A solution is then Pareto opti-
mal whenever there is no solution dominating it. The col-
lection of optimal solutions is the so-called Pareto front.
Figure 6.5 shows a simple example with n = 2. The points
A and B, on the Pareto front, represent optimal solutions.
The point C and all lighter-coloured squares are non-­
optimal solutions.
As we observed, multi-objective optimisation is a process
that allows to select the best configuration in high-­
complexity systems, by keeping into account the
 6 Flying on Sun Power 99

Fig. 6.5 Two Pareto optimal solutions A and B

interactions among the project’s many aspects. In other

words, it aims to find a mathematical answer to the follow-
ing key question: how does one decide what to change in a
configuration, and when to change it, if every design pa-
rameter affects everything else?

6.6 An Example of Multi-objective

Optimisation
Now let us take a look at a concrete example. The question
is: what is the maximum weight allowed for the structure of
Solar Impulse if we want to fly at night above a pre-­
established altitude? Answering the question helps design-
ers determine, say, the type of composite material for the
wing structure. It is a real and concrete problem, which we
100 A. Quarteroni

addressed in the EPFL labs by considering the aircraft’s

structural, aerodynamical and electric characteristics.
Objective functions, design variables and constraints are
defined in Table 6.1.
The realisable configurations obtained from the model
are shown in Fig. 6.6. We examined 20,000 configurations
and saw that four thousand satisfied the constraints. The

Table 6.1 Objective Objective functions

functions, design Maximisation of minimal nocturnal
parameters and altitude
constraints to find the Maximisation of structure’s weight
maximum Design parameters
weight allowed Wingspan = 80 m
Wing surface = 230 sq. m
Battery weight = 450 kg
Structure’s weight = 800–1000 kg
Constraints
Minimal nocturnal altitude >
3000 m

Fig. 6.6 The problem’s Pareto front

 6 Flying on Sun Power 101

rightmost boundary of the cluster of points defines the

problem’s Pareto front.
As we said, selecting configurations on the Pareto front
corresponds to fixing a weight (or a coefficient) for each one
of the objective functions considered. In mathematical
terms, it means finding the coefficients w1,...,wn that opti-
mise the function w1f1+...+wnfn. In this case the curve shows
the influence of the structure’s weight on the minimum al-
titude reached at night. We discover that, for example,
above 1000 kg the constraints are never satisfied, and that if
we decided, for safety reasons, to fly above 4000 m at night,
the maximum structural weight allowed would be 860 kg.
The flight profiles (in terms of altitude and speed) corre-
sponding to the two ends of the curve are shown in Figs. 6.7

Fig. 6.7 Flight profile corresponding to the first endpoint of the

Pareto front
102 A. Quarteroni

Fig. 6.8 Flight profile corresponding to the second endpoint of

the Pareto front

and 6.8. This computation is an example of how multi-­

objective optimisation can be employed to support design
choices.

References
1. www.stevefossett.com *
2. www.solarimpulse.com *
3. www.bertrandpiccard.com *
4. earthobservatory.nasa.gov/features/GlobalWarming *
5. Drela M. (1989), Xfoil: An Analysis and Design System for
Low Reynolds Number Airfoils, Lecture Notes in Engineering
54, Springer, New York.
6. Peyret R. (a cura di) (1996), Handbook of Computational
Fluid Mechanics, Academic Press, London.
 7
The Taste for Mathematics

Abstract Mathematics is useful to model the consistency

of chocolate as its temperature varies, to simulate new pro-
duction chains and to optimise the packaging, storage and
transport of food products. This also includes the optimisa-
tion of the nutritional content of food in view of maximis-
ing flavour perception.

The appreciation of a good meal, the art of cooking, food

science, the technology of food preparation: various facets
of one of the basic needs of the human being, namely, find-
ing a source of enough energy to make the body function
on a daily basis. Food processing has passed from being ar-
tisanal to becoming an important industry that sees the
growing involvement of multinationals operating on a
global scale. Although it may seem surprising, in recent
years the food industry has formed an increasingly tighter

© The Author(s), under exclusive license to Springer Nature 103

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_7
104 A. Quarteroni

bond with mathematics, which is often (wrongly) seen as

the science most detached from the primary needs of hu-
man beings.

7.1 Food Preparation

The food industry has been, for a while now, a testing
ground for the application of mathematical models. Before
we go to concrete examples, let us use a scheme to sum-
marise the typical path of a product in the food supply
chain. Even if simplistic, it will give us an idea of the com-
plexity of the process. A specific product may be thought of
as the creation of a team of specialised researchers, who
work to devise better food for nutritional value, flavour, as-
pect and attractiveness to consumers. Apart from its intrin-
sic characteristics, it is useful to study the way this product
will be prepared, its packaging, the conservation of its prop-
erties along the route that includes distribution, sale and
arrival to the end consumer’s home. Figure 7.1 shows three
steps, where we have highlighted the contributions mathe-
matics can offer.
Even after sale and consumption, the product can be
studied. For instance, we can examine how much the con-
sumer liked it and which effects it has on health (good or
bad). In these circumstances, too, mathematics can reveal
very useful. In the following sections we will see some
examples.

7.2 Mathematics and the Brain

What links the perception of taste through the taste buds to
the corresponding sensation elaborated by the brain? It is a
process we pay no attention to, that repeats daily. Yet it is
 7
The Taste for Mathematics

Fig. 7.1 When a food product encounters mathematics

105
106 A. Quarteroni

extraordinarily complex, for it involves a particularly diffi-

cult organ to model, the brain. The Brain and Mind Institute
at EPFL [1] aims to create models capable of emulating the
brain activity, using a supercomputer that performs more
than 300,000 billion operations per second. That is an in-
credible speed, necessary to reproduce the actions and in-
teractions of hundreds of thousands of nerve cells at a time.
One aspect of this very ambitious study is directed at simu-
lating the processes that govern the perception and evalua-
tion of taste. These models compute the brain’s energy re-
quirements and try to optimise the composition of nutrients
in food, with particular attention for two consumer groups
in critical phases of life: small children and elderly people.
It is known that the first assessment of a food happens
through the eyes, and that our choices concerning food are
often guided by the appearance of what stands in front of
us. The Blue Brain [2] project has the objective of analysing
the cognitive processes of sight, beside the more primary
processes relative to taste and smell.
Thanks to Blue Brain neuroscientists will be able to con-
duct computer experiments to test diagnostic and thera-
peutical tools for neurodegenerative diseases. Numerical
models allow to make simulations of the biophysical pro-
cesses happening at the various levels of the cerebral archi-
tecture, thus integrating the morphological and structural
data of a single brain with a universal mathematical model
based on a representation of the biochemical and electric
activity at the cellular level. Today, modelling a single nerve
cell requires we solve a system of 20,000 ordinary differen-
tial equations, roughly. This number mushrooms to more
than 100 billion if we want to model the entire brain activ-
ity. A goal that is still very far away.
 7 The Taste for Mathematics 107

7.3 The “Formula of Flavour”

Taste is the most immediate and important sense presiding
over our relationship with food. Therefore managing fla-
vour is the primary task, and the most complex one, in the
industrial design of food. A product’s flavour is created by
mixing several ingredients, among which there may be
preservants, colour additives or artificial aromas, which
clearly do not appear in the homemade version of the same
recipe. Understanding how a certain quantity of a certain
substance will affect the final taste of a product is almost
never an easy process. For that reason some companies rely
on mathematics to gain more insight on the matter, prefer-
ably in a deterministic, and hence reproducible, way.
Researchers at Glasgow University, for instance, have devel-
oped the formula of flavour for blackcurrant drinks, which
provides the expected result when using the basic ingredi-
ents [3]. The starting point is based on neural networks, to
which we shall return in the next chapter, and allows to
calculate the intensity of blackcurrant flavour as one tweaks
the relative ingredients’ quantities. It also takes into account
variables such as the berries’ geographical provenance and
the harvest period. Another mathematical model has been
developed, again at the University of Glasgow, to study the
relationship between the sweet flavour of lager and the pres-
ence of volatile substances other than sugar in the mix (in
this case, too, with the help of neural networks) [4].
The success of an industrially processed food is not only
a matter of flavour, though. Also texture plays an important
role in steering consumer choices. To this end researchers at
Birmingham University studied heat conduction in choco-
late to predict the product’s consistency in function of its
temperature [5]. The hardening process of chocolate has
several critical components, and depends not only on the
108 A. Quarteroni

initial and final temperatures, but also on the cooling time

and on the end product’s desired shape. Everything starts
with the heat equation [6], which predicts how temperature
will propagate in the chocolate. In particular, it is a good
idea to control the cooling strategy to guarantee that the
transition from the liquid phase to the solid phase is as ef-
ficient as possible.
The physics of heat transfer also governs a process par-
ticularly dear to Italians: the cooking of pasta, an (almost)
daily action for many of us.
The necessary chemical and physical transformations
(starch gelatinisation, denaturation, and coagulation of glu-
ten, essentially) happen only if the water reaches a certain
temperature. Contrary to the common belief, this is not
100 degrees Celsius, the boiling point; it suffices, in fact, to
maintain the liquid at 80–85 degrees. The ideal cooking
time depends on a number of factors, not least our personal
taste. But it is possible to obtain a simple formula [7], dif-
ferent for each type of pasta. For spaghetti we have

t  ar 2  b

where a is a coefficient that depends on the manufacturing,

r is the spaghetti radius and b is another coefficient reflect-
ing the desired cooking level (small for al dente pasta, large
if we want it well cooked). The formula for bucatini, in-
stead, accounts for the hollow shape. In this case

aD  d
2

t b
2

where D and d are the outer and inner diameters respectively.

 7 The Taste for Mathematics 109

7.4 Optimising the Industrial

Production of Food
Industrial food production consists, concretely, in a series
of chemical and physical transformations of the basic ingre-
dients. Each one of its phases can therefore be examined
with the aid of mathematics. Let us see some examples.
The Mechanical Engineering Department at University
of Padua has analysed the molecular viscosity of dough in-
side an industrial mixer, and the study’s results have
prompted the modification of the machine to improve its
performance. The dough for fresh pasta, like many other
foods, behaves like a non-Newtonian fluid. This is a fluid
whose viscosity is not constant (as opposed to the classical
case of a Newtonian fluid, like air, or water) but depends on
its velocity. Hence the classical behavioural models of viscid
fluids inside machines such as mixers should be substituted
by new models, which account for this feature and treat
viscosity as a variable rather than as a constant. Non-­
Newtonian fluids, by the way, have become familiar to the
general public in the last few years because they lend them-
selves well to spectacular demonstrations. Most of us have
come across videos on the internet of researchers walking on
water, that is, walking without sinking over a pool filled
with a milky-looking liquid. Usually that liquid is a mixture
of water and corn or potato starch. The stunt is possible
because the force applied by the foot pressing onto the sur-
face makes the fluid behave like a solid rather than a liquid.
A weaker force, like that of a spoon in a liquid, does not
cause any apparent change in state [8].
The Department of Mathematics at University of
Florence, instead, concerned itself with coffee. A research
group studied the way in which pressurised water filters
through the coffee powder, to establish which conditions
110 A. Quarteroni

lead to the formation of lumps. This phenomenon alters the

transfer of the aroma from the coffee powder to the water,
and hence it changes the taste of the beverage [9]. While
Italian universities study pasta and coffee (no surprise
there), in California, as one might imagine, the focus is on
hamburgers and on finding out the best way to cook them
to optimise flavour, digestibility and food safety. Cooking a
hamburger is a less trivial task than we might expect. One
must determine precisely the temperature and cooking
time, so to eliminate the possible pathogens present in the
meat without altering too much its nutritional value and
(most of all) the mouth-watering taste. The Food and Drug
Administration (the American agency responsible for the
safety of food and drugs) has issued precise regulations re-
garding the minimum temperature and cooking time of
hamburgers prepared by the food industry or restaurants.
At the root of it all lies a physics problem: how can one
make the heat propagate inside the meat so that each part
reaches the minimum required temperature? It is easy to
imagine that the solution depends in an essential way on
the hamburger’s shape and thickness. The corresponding
mathematical model is provided by the following equation,
which does not look very complicated but in truth requires
non-trivial solution techniques:

H   T  1   T 
 k   r r  rk r .
t z  z   

Here H is the enthalpy (the sum of the internal energy and

the product of the pressure p by the volume V ), k is the
thermal conductivity inside the meat (in turn depending on
the enthalpy), t is time, T the temperature, r the radial co-
ordinate (going from the centre to the periphery) and z the
vertical coordinate (measuring thickness).
 7 The Taste for Mathematics 111

The model must be completed with several other equa-

tions, expressing the outward heat flow from the crust, and
suitable transfer conditions between the hamburger’s frozen
and defrosted parts (a complex free-boundary mathematical
problem!). According to this model, a standard double grill
heated at 200 degrees Celsius needs about 120 seconds to
make the centre of a 1.1 cm thick hamburger reach 75 de-
grees, exactly matching the experimental evidence! [10].
In the last few years other American teams have studied
the thermal behaviour of several liquids (milk, apple juice,
tomato sauce etc.) when they are hit by a microwave beam,
a very common procedure in the food industry [11]. To
tackle that problem it is necessary to solve Maxwell’s equa-
tions, which lie at the heart of every electromagnetic phe-
nomenon. Important domestic-appliance companies study
how to optimise irradiation so that a specific food (say,
chicken) is cooked in the most uniform possible way de-
pending on its shape and weight.

7.5 Mathematical Packaging

A freshly produced food still has a long way to go before it
reaches our dinner table: it must be packaged, distributed
to sellers and preserved appropriately. All jobs that mathe-
matics can make more efficient. Let us take packaging. It is
necessary to optimise the shape of the packaging and its
material to make it robust but light, and able to prevent
external contamination. The researchers at Technische
Universiteit Eindhoven in the Netherlands are specialised
in, thanks to modelling, designing glass bottles [12]. This is
another common object that, even if not manifestly, in-
volves advanced technological solutions. Mathematics con-
tributes to the optimisation of shape, weight and rigidity,
112 A. Quarteroni

and helps maximise storage and transport. Ideal bottles, in

fact, must be piled up and packaged so to minimise the
waste of space. Clearly, one must consider the usability con-
straints: producing cubical bottles, for instance, would solve
the stacking problem mathematically, but certainly not the
ease of use.
A research group in Amsterdam, once again in the
Netherlands, has applied mathematics to the study of pre-
serving vegetables in modified atmosphere, with the aim of
prolonging shelf life [13]. A significant improvement of
food preservation techniques can have important conse-
quences on distribution. Thanks to these studies it is now
possible to transport fresh produce by sea instead of air,
thus lowering the ensuing overall costs.
In general, the task of finding the best way for food to go
from the producer to the consumer may be seen as an ap-
plication of a classical problem in operations research: the
travelling salesman problem [14]. This problem is easy to
state but difficult to solve. Imagine a salesman has to visit a
given number of clients scattered in a region in a non-­
uniform way, and suppose he know the distances between
each client, the network of available routes and the travel-
ling times of each. The salesman will want to minimise the
effort, that is, travel the least possible. In the past the expe-
rienced salesman would have relied on experience, but to-
day we are able of providing precise mathematical solutions
to the problem.
We could give other examples of how mathematics ap-
plies to packaging and logistics, but now we are hungry: the
food has arrived on the table and we can enjoy it.
 7 The Taste for Mathematics 113

7.6 Mathematics and Health

A food continues to be the object of mathematical investi-
gation even after it has been eaten. A product’s success
should be assessed against consumer satisfaction, sure, but
also based on its effects, good or bad, on the health of
who eat it.
To make a simple example let us mention a study, done
with statistical methods, in which a team at University
College London has determined the features of the perfect
product [15]. If you think that in the US almost 90% of
newly launched foods are called back after a while because
of low sales, it becomes clear that studies of this kind are a
new and powerful tool for the people coming up with new
products in an ever more competitive market.
Consumers, on the other hand, might judge a product’s
nutritional value more important than its commercial
success.
Another aspect of food distribution that has worldwide
relevance is waste prevention. It is estimated that every year
1.3 billion tons of food go wasted: slightly more than half
during production and harvest, the rest during processing,
distribution and consumption. At the same time 870 mil-
lion people go hungry every day. By 2050, when the world
population will pass the 9 billion threshold, food produc-
tion will have to increase 70% to feed the entire planet [16].
For this reason the École Polytechnique Fédérale de
Lausanne has created the Integrative Food and Nutrition
Center: to solve the challenges posed by a sustainable future
of food and nutrition. The studies conducted at EPFL range
from precision agriculture to more efficient packaging sys-
tems and a better handling of waste [17].
Today the relationship between eating and mathematics
is very tight, as we have seen. In this chapter we have only
114 A. Quarteroni

scratched the surface with the help of examples showing

how modelling is omnipresent in every phase of a food
product’s life. It is interesting to observe that the mathemat-
ics necessary to address the problems described here is very
complex. These examples reflect a picture of mathematics
that is very different from the (wrong) idea of an abstract
science, crystallised in a list of theorems that are totally ir-
relevant to the lay person. It is a mathematics that can ac-
company us inside the kitchen, advise us and surprise us in
ways we could not have imagined [18].

References
1. www.epfl.ch/schools/sv/bmi/ *
2. bluebrain.epfl.ch *
3. Boccorh R.K., Paterson A. (2002), An Artificial Neural Net-
work Model for Predicting Flavour Intensity in Blackcurrant
Concentrates , Food Quality and Preference, 13, pp. 117-128.
4. Techakriengkrai I., Paterson A., Piggot J.R. (2004), Relation-
ships of Sweetness in Lager to Selected Volatile Congeners,
Journal Institute Brewing, 110 (4), pp. 360-366.
5. Tewkesbury H., Stapley A.G.F., Fryer P.J. (2000), Modelling
Temperature Distributions in Cooling Chocolate Moulds,
Chemical Engineering Science, 55, pp. 3123-3132.
6. Quarteroni A. Numerical Models of Differential Problems,
3rd edition, Springer Series MS&A, Vol 16, 2017 (xvii+681p.).
7. Rigamonti A., Varlamov A. (2007), Magico caleidoscopio
della fisica, La Goliardica Pavese, Pavia.
8. Owens R.G., Phillips T.N. (2002), Computational Rheology,
Imperial College Press, London.
9. Fasano A. (1996), Some Non-Standard One-­Dimensional
Filtration Problems, The Bulletin of Faculty of Education,
Chiba University (iii, Natural Sciences), 44, pp. 5-29
10. Singh R.P. (2000), Moving Boundaries in Food Engineering,
Food Technology, 54 (2), pp. 44-53.
 7 The Taste for Mathematics 115

11. Zhu J., Kuznetsov A.V., Sandeep K.P. (2007), Mathematical

Modelling of Continuous Flow Microwave Heating of Liq-
uids (Effects of Dielectric Properties and Design Parameters),
International Journal of Thermal Science, 46, pp. 328-341.
12. Laevsky K., Mattheij R.M.M. (2000), Mathematical Model-
ling of Some Glass Problems, in Fasano A., Complex Flows
in Industrial Processes, Birkhäuser, Basel.
13. Rijgersberg, H., Top J.L. (2003), An Engineering Model of
Modified Atmosphere Packaging for Vegetables, 2003 Inter-
national Conference on Bond Graph Modeling and Simula-
tion, Miami, FL.
14. Applegate D.L., Bixby R.E., Chvátal V., Cook W.J. (2006),
The Traveling Salesman Problem: A Computational Study,
Princeton University Press, Princeton, NJ.
15. Corney D. (2000), Designing food with Bayesian Belief Net-
works, Proceedings of ACDM2000 – Adaptive Computing
in Design and Manufacture, Plymouth, UK.
16. Report 2013, Food and Agriculture Organization of The
United Nations.
17. www.epfl.ch/research/domains/nutrition-­center1*
18. Giusti E. (2004), La matematica in cucina, Bollati Boringh-
ieri, Torino.
 8
The Future Awaiting Us

Abstract Not just models but also artificial intelligence,

machine learning and big data. Thanks to the advance-
ments in scientific knowledge and in the computational
power of supercomputers, the mathematical models of the
future will be increasingly accurate.

Hopefully I have managed to convince the reader that

mathematics, just like many other disciplines, is useful,
beautiful and creative, and at the same time contributed to
debunk a number of misconceptions.
Myth number one: mathematics is a self-referential disci-
pline, practised by individuals that are often incapable of
communicating with the world. In reality we have seen that
all cases examined required a strong interaction with profes-
sionals from other disciplines, whether doctors, meteorolo-
gists, athletes or other. For doing good applied mathematics
it is necessary to immerse oneself in the problems, capture

© The Author(s), under exclusive license to Springer Nature 117

Switzerland AG 2022
A. Quarteroni, Modeling Reality with Mathematics,
https://doi.org/10.1007/978-3-030-96162-6_8
118 A. Quarteroni

their essential features and hold regular conversations with

other people. My own personal motto in this respect is: tell
me what answer you expect and we mathematicians will try
to formulate the mathematically most correct question.
Myth number two: there is a clear-cut distinction be-
tween pure and applied mathematics. No. There is only one
sort of mathematics that makes sense doing—the good
sort, and the deeper we know it the better we will manage
to apply it to problem-solving. What is true is that there is
a difference between pure and applied mathematicians, due
to a different attitude when doing mathematics. Applied
mathematicians put the problem to be solved at the centre
of their attention (for they feel the need to give the real
world answers) and do not accept simplifying compromises
if these undermine the result’s significance. Pure mathema-
ticians, instead, are interested in the problem’s structure, in
unearthing the theoretical properties, thus possibly paving
the way to the creation of new theories that may lead far
away from the study’s original motivation. The latter is
curiosity-­driven research, as opposed to the former’s
problem-­driven research. But the line separating the two
scopes is not clear-cut, and the background knowledge is
the same. I myself have often moaned about not knowing
enough mathematics: sometimes, certain apparently very
abstract areas would have helped me build better models,
and therefore find more efficient solutions to real problems.
Myth number three: mathematics can be done only by
hand. Nothing could be more wrong. What if we did not
have computers? Actually: supercomputers, which perform
up to one billion billion operations per second (just like
Frontier at Oak Ridge National Laboratories in the US, the
fastest supercomputer in the world at the time of writing).
Almost every problem I have discussed in the book has been
 8 The Future Awaiting Us 119

solved with supercomputers. In order to translate these ma-

chines’ potential into a real advantage new algorithms had
to be developed, called parallel algorithms. It is because of
these novel instruments that we can get the most out of
machines made by clusters of several thousands of indepen-
dent logical units, also known as cores. This is but one of
many examples where technology and research go hand in
hand, one amplifying the other’s success.
One aspect I did not mention, that nonetheless deserves
a few words, is the difference between deterministic mod-
els and stochastic models. Here in fact I have concentrated
on the former kind exclusively. A process is called deter-
ministic if it produces the same solution each time we feed
it the same data. We call it stochastic, instead, if it keeps
into account that the input data may fluctuate in a more
or less casual way, and therefore it provides solutions only
within a certain probability margin. Examples of stochas-
tic processes are those describing the behaviour of bio-
logical systems of human actions (in social phenomena,
financial choices, decisions regarding games and competi-
tions). In certain applied contexts resorting to stochastic
models is inevitable. Think about climate: foreseeing the
average temperature increase in 30–50 years’ time neces-
sarily depends on the human footprint in the coming
years, for instance, a variable that can only be represented
in probabilistic terms. It goes without saying that solving
a non-deterministic model is way more complicated than
solving a deterministic model.
About that, nowadays there is a very popular technique
among mathematicians, called uncertainty quantification
(UQ). In practice one starts from a deterministic model
and assumes that the data oscillate within a confidence in-
terval (represented, say, by a Gaussian distribution). At this
120 A. Quarteroni

point one computes how the uncertainty of the data rever-

berates on the solution, and then one determines, retroac-
tively, the acceptable confidence interval. These techniques
are for instance used in models of the circulatory system
(see chapter four) where a lot of data have an intrinsic de-
gree of uncertainty (think of the lack of accuracy in recon-
structing the shape of organs from medical images like CT
scans or MRIs). They are also employed when the data can-
not be generated: this is the case of quantities such as the
myocardium’s electric conductivity, which is just impossible
to quantify on a living being.
The reader who made it to this point might wonder why
I did not speak about subjects like artificial intelligence
(AI), machine learning or big data, which dominate the me-
dia so much they have become topics of small talk. I have
no bias against them whatsoever. I wanted to talk about
models simply because it is an approach to problems some-
how alternative to that of artificial intelligence. AI is a fasci-
nating and complex subject that would deserve a whole
new discussion, but I will try to summarise what I mean.
Let us begin from the definitions: artificial intelligence is
a computer’s ability to emulate the cognitive functions of
the human mind, such as learning and problem solving.
Machine learning is a computer’s ability to learn and per-
form a specific action without prior programming via ex-
plicit instructions. Although not new (the first definition of
AI goes back at least 60 years), these concepts have had a
groundbreaking impact only recently. The crucial leap for-
ward is probably due to big data, a term referring to the
availability of immense quantities of data, both complex
and heterogeneous, and the corresponding evolution of sta-
tistical techniques for their analysis and classification. The
breakthrough we mentioned may also be due to
 8 The Future Awaiting Us 121

supercomputers, incredibly powerful machines able to

crunch big data in real time.
Among the countless areas where AI can successfully be
applied we can mention artificial vision, text and speech
recognition, expert systems, robotics, self-driving cars, au-
tomatic public-transport operation, even competitive sports.
Oh yes, football too. Nowadays the top clubs make use
of mathematics to optimise their performance. During any
match of the Italian Serie A, the Spanish Liga or the English
Premier League, special cameras record 20–30 times per
second the coordinates of the ball and of the 22 players on
the pitch. This totals to over 10 million positional data gen-
erated during a match. Big data algorithms manage to mon-
itor them all, and highlight an enormous number of indica-
tors and useful evaluations. We are not talking about the
statistics on ball possession that fill TV programmes at the
end of a match, but of instruments capable of reconstruct-
ing every single event, like a shot, a pass, a dribble or ball
control, all events whose effectiveness mathematics can
analyse. AI algorithms can provide in real time operational
indications regarding what the manager and their staff want
to monitor during the match. It is like having an additional
virtual assistant.
The new technologies are revolutionising the labour mar-
ket as well. According to some analysts, by 2030 AI will
create between 555 and 890 million new jobs across the
world, and for the same reason 400–800 million people
with have to change jobs.
The mathematical abstraction enabling machine learning
is the Artificial Neural Network, made by several decision
centres (artificial neurons) connected by logical/mathemat-
ical operations that allow to pass from an input to an out-
put without any knowledge of the underlying process.
Learning and decision making are developed from the
122 A. Quarteroni

accumulation and comparison of large quantities of data.

The network, in practice, is trained to perform certain tasks,
and executes them efficiently but without thinking.
Therefore no new knowledge is created; one only uses the
knowledge that can be extracted from the data.
In the light of this, then, AI is complementary to human
intelligence, and it is only the latter that oversees the cre-
ative act (at least until now: it is difficult to foresee what the
future has in store, given the prodigious advances machine
learning has made us familiar with in recent years). In de-
scribing the world around us, physical laws express exactly
the creation of knowledge. These laws are equations and
models that do not emerge ``automatically” from the col-
lection of lots of data, but are born out of human reasoning
and do not change with the context. Newton’s laws, the
Navier-Stokes equations, the second principle of thermody-
namics are but a few remarkable instances. Once discov-
ered, they become part of the universal heritage, which is
invariant in time and space. Put in another way, the model
for the atmospheric circulation is valid whether it is applied
to the Mediterranean sea, the Arizona desert or the Siberian
tundra. The data change, the equations do not. Similarly,
the heart model works, in principle, for any human being:
it is the data derived from clinical imaging, for example,
that will differentiate the heart of one patient from another.
The validity of AI is based on the analysed data, those
used for its training. Its results do not permit (yet) the ab-
straction of laws and models. With this perspective the two
approaches may be considered as being alternative. When
we examine them more critically (and with more
 8 The Future Awaiting Us 123

imagination) though, we catch a glimpse of the many op-

portunities to make them work in a complementary way
as well.
I believe, actually, that this path will lead to great prog-
ress in the years to come [1].

Reference
1. A. Quarteroni, Algorithms for a New World, Springer
Nature, 2022