VARIETIES OF MATHEMATICAL UNDERSTANDING

arXiv:2310.20100v1 [math.HO] 31 Oct 2023

JEREMY AVIGAD

Abstract. This essay considers ways that recent uses of computers in math-
ematics challenge contemporary views on the nature of mathematical under-
standing. It also puts these challenges in a historical perspective and offers
speculation as to a possible resolution.

1. Introduction
The goal of this essay is to explore some of the ways that recent uses of computers
in mathematics challenge conventional views as to the nature of mathematical un-
derstanding. Section 2 surveys four developments in discrete geometry, all of which
rely on the use of computation, although in markedly different ways. Section 3
aims to clarify the view that an important goal of mathematics is to obtain a deep
conceptual understanding. Section 4 highlights some of the differences between
such a conceptual understanding and the kinds of computational understanding we
get from the results in discrete geometry. Finally, Section 5 reflects on the history
of mathematics to support speculation as to whether some sort of reconciliation is
possible, and, if so, what form it might take.

2. Some results in discrete geometry

Discrete geometry studies combinatorial properties of discrete geometric objects.
A number of important problems in the field have to do with optimal or other-
wise interesting arrangements of such objects in space. Here, “space” might mean
the Euclidean plane or three-dimensional Euclidean space, but it can also refer to
higher-dimensional and non-Euclidean spaces. In this section, we will consider four
illustrative developments, each centered on a representative theorem.
2.1. The Kepler conjecture. The first result we will consider was known for a
long time as the Kepler conjecture, but now perhaps should go by the name Hales’
theorem. Imagine filling three-dimensional space with infinitely many nonoverlap-
ping spheres, all the same size. One can arrange them in hexagonal layers, with
each successive layer nestled into the one beneath it. At each layer, there are two
choices of how to fit the one above it into the crevices, but they are equally efficient;
the face-centered cubic packing is one such arrangement.
√ It is not hard to show that
the fraction of space filled in this way is π/3 2, about 0.7405.
In 1611, Johannes Kepler asserted that this density is optimal, but, for centuries,
the claim went unproved. In 1998, Thomas Hales announced a proof, obtained with

Date: February 21, 2021.

2020 Mathematics Subject Classification. Primary 00A35, 01A67; Secondary 52-02, 52-08.
I am grateful to Robert Y. Lewis, Patrick Massot, and the anonymous referee for comments,
corrections, and helpful suggestions. This work has been partially supported by AFOSR grant
FA9550-18-1-0120 and the Sloan Foundation.
1
2 JEREMY AVIGAD

a contribution from a student of his, Samuel Ferguson [19]. Hales’ strategy was to
reduce the problem to a complex set of sufficient conditions that could be verified
by an extensive body of computation. The proof shows that any counterexam-
ple to the Kepler conjecture would imply the existence of a finite arrangement of
spheres satisfying certain properties, giving rise, in turn, to a certain combinato-
rial structure. Computer code then produced an exhaustive enumeration of the
possible combinatorial structures; to be realized geometrically, any such structure
would have to satisfy certain inequalities. Using branch-and-bound methods, these
inequalities were relaxed to linear ones, at which point linear programming meth-
ods were used to demonstrate their infeasibility. In other words, the computations
showed that there is no finite arrangement of spheres of the kind guaranteed by
a putative counterexample. The proof thus consisted of a traditional mathemati-
cal argument (about 250 pages at the time) combined with a substantial body of
computer code used to carry out the calculations.
Theorem 2.1. The optimal density achieved by a packing of nonoverlapping equally
sized spheres in Euclidean space is the one attained by the face-centered cubic pack-
ing.
Hales’ Flyspeck project, completed in 2014, used computational proof assistants to
verify the correctness of the result in a formal axiomatic framework [18].
2.2. Packings of regular tetrahedra. The second theorem we will consider has
to do with the optimal packing density of regular tetrahedra instead of spheres. In
2006, John Conway and Salvatore Torquato obtained a packing with a density of
about 0.7175, and their paper sparked a flurry of interest in the problem [10]. The
best result to date is due to Elizabeth Chen, Michael Engel, and Sharon Glotzer
[7]:
Theorem 2.2. There is a packing of equally sized regular tetrahedra in Euclidean
space with density 4000/4671.
That fraction is equal to about 0.8563.
Theorem 2.2 emerged from an impressive constellation of approaches to the prob-
lem. Conway and Torquato came up with their packing by starting with a configu-
ration of irregular tetrahedra that fills space exactly, approximating it with regular
tetrahedra, and then tweaking the configuration to increase the density. Other re-
searchers adopted the strategy of parameterizing families of packings that seemed
promising. For example, one can focus on packings based on a two-tetrahedron
dimer the configuration formed by gluing two tetrahedra together so that their
bases match. With such restrictions, researchers could use various computer search
techniques to optimize the choice of parameters.
Yet a third approach used Monte Carlo simulations of roughly the following
physical act: put a lot of tetrahedra into a big box, jiggle the box, and then
compress it as the configuration settles. This often results in arrangements that
exhibit a surprisingly orderly quasi-crystalline structure. There is a thermodynamic
explanation: because pockets of order are locally more efficient, their components
have more freedom to move around, resulting in a system with higher entropy
overall. (Intuitively, there are lots of ways to perturb the components, so such
arrangements dominate the space of configurations, and a random process is more
likely to end up there.) In 2009, Glotzer and colleagues used such a simulation (and
 VARIETIES OF MATHEMATICAL UNDERSTANDING 3

subsequent tuning) to find a packing with a density of 0.8503, a record at the time
[17]. The current record of 0.8563 was found using a parameterized family instead,
but similar experimental techniques were used to increase faith in that the result
is optimal at least with respect to nearby alternatives.

2.3. Sphere packing in higher dimensions. Packing problems are not restricted
to three dimensions, and the next results have to do with the densest packing of
spheres in n-dimensional Euclidean space. Surprisingly little is known about di-
mensions greater than three, but not for lack of effort. There are remarkably effi-
cient packings in eight and twenty-four dimensions. The eight dimensional packing
centers spheres on the points of a lattice known as E8 , while the twenty-four dimen-
sional packing centers spheres on the points of a lattice known as the Leech lattice,
discovered by Conway in 1967. In 2016, Maryna Viazovska announced a proof
that the density achieved by the E8 packing is optimal in eight dimensions [29],
and within a week of the announcement, Henry Cohn, Abhinav Kumar, Stephen
D. Miller, and Danylo Radchenko had joined her to extend the method to show the
optimality of the packing based on the Leech lattice in twenty-four dimensions [9].
Theorem 2.3. The optimal density of an 8-dimensional sphere packing is attained
by E8 , and the optimal density of a 24-dimensional sphere packing is attained by
the Leech lattice.
There is a nice exposition of these results and the methods behind them in an
article by Cohn in the Notices of the American Mathematical Society [8]. In a
paper in the Annals of Mathematics in 2003, Cohn and Noam Elkies provided a
simple Fourier-analytic method of providing upper bounds on the density of sphere
packings [8]. In short, they showed that the existence of a function with certain
properties gives rise to such a bound, thereby reducing the task to that of finding
functions with the right properties. Using numerical methods, they were able to
produce functions yielding the best known bounds for dimensions 4 through 36, and
noted that the upper bounds were especially tight in dimensions 8 and 24. This
gave rise to the hope that one could construct suitable “magic functions” in those
dimensions to show that the lower bounds given by E8 and the Leech lattice are
sharp.
As the Notices article explains, numeric calculations by Cohn and Miller pro-
vided hints as to what such functions should look like. Viazovska’s construction
for E8 combined these insights with mathematical intuition and expertise: draw-
ing on modular forms, properties of the Laplace transform, experimentation, and
guesswork, she was able to produce a function with the requisite properties. The
extension to 24 dimensions was based on the same techniques.

2.4. The Keller conjecture. The last result that we will consider starts with two
dimensions and moves to higher dimensions from there. It is clearly not possible to
tile the plane with unit squares in such a way that no two squares share a common
edge. Start with a square and then stack one on top of it, shifted to the right, say,
so that they will not have an edge in common. Put another square to the right
of the first one, shifted down to avoid sharing an edge. There is then a gap left
over that clearly cannot be filled. It is not hard to convince oneself that, similarly,
there is no way to fill space with unit cubes without two cubes sharing a face.
In 1930, Ott-Heinrich Keller conjectured that the corresponding claim holds in all
4 JEREMY AVIGAD

dimensions, that is, for every n, there is no tiling of n-dimensional space with unit
cubes such that no two squares share a common (n − 1)-dimensional face [21].
In 1940, Oskar Perron showed that the conjecture is true for n ≤ 6 [25, 26]. In
1992, Jeffrey Lagarias and Peter Shor showed, surprisingly, that it is false for n ≥ 10
[23]. (The failure of the conjecture in any dimension straightforwardly implies the
failure in higher dimensions.) In 2002, John Mackey showed the conjecture is false
for n ≥ 8 [24], leaving only the case n = 7 open. This was recently settled by
Joshua Brakensiek, Marijn Heule, John Mackey, and David Narváez [6].
Theorem 2.4. Keller’s conjecture is true up to and including dimension 7, but
fails for dimension 8 and above.
Once again, the final result builds on a heterogeneous mix of methods. An
important milestone in the attack on the conjecture was to reduce the original
geometric question to a discrete, graph-theoretic one. In 1986 Sándor Szabó reduced
the conjecture to periodic tilings [27], and in 1990 Keresztély Corrádi and Szabó
showed that the conjecture is equivalent to the statement that for every n, there is
no clique of size 2n in a certain finite graph, Γn . In fact, that was how the initial
counterexamples were obtained: Lagarias and Shor exhibited a clique of size 210
in Γ10 , and Mackey exhibited a clique of size 28 in Γ8 , each of which gave rise to
a counterexample of the corresponding dimension. In 2017, Andrzej Kisielewicz
generalized the construction to a family of graphs Gn,s to make the correspondence
with the dimension sharp: the Keller conjecture is true at dimension n if and only
if there is no clique in Gn,s for any s. (In this notation, the Corrádi–Szabó graph
Γn is Gn,2 .) He then showed that, to settle the only remaining case, it is enough
to rule out the existence of a clique in G7,6 , which has 127 nodes [22].
That is exactly what Brakensiek, Heule, Mackey, and Narváez did. To obtain the
result, they used propositional logic to encode the existence of a clique in G7,6 and
then used software known as a SAT solver to show that the resulting formula has
no solutions. Propositional satisfiability solvers have, in recent years, opened up
new possibilities for finding complex combinatorial configurations meeting various
sets of constraints, or showing that a given set of constraints cannot be met [3].
They are used to solve a wide range of problems in hardware verification, software
verification, and AI. A naive search would have overwhelmed the solver; to obtain
the solution, Brakensiek, Heule, Mackey, and Narváez had to develop new heuristics
to break symmetry and reduce the search space to a manageable size.

3. Conceptual understanding and depth

What do results like these contribute to mathematical understanding? Answer-
ing that question requires a conception of the kind of understanding that mathe-
matics is after. The subject comprises a range of pursuits, some more pure, and
some more applied; it reaches into all the physical and social sciences, including
engineering, statistics, and computer science. What I am trying to get at here is
something like the essence of pure mathematics, the things we value about math-
ematics itself. What makes something worthy of publication in a premier journal
like the Annals of Mathematics? What are the criteria at play in selecting a Fields
medalist? What are the topics most worthy of representation at the International
Congress of Mathematicians? In short, what is quintessentially important to the
subject?
 VARIETIES OF MATHEMATICAL UNDERSTANDING 5

It is common to say that the goal of mathematics is to obtain a conceptual

understanding of mathematical phenomena, and a deep understanding at that.
The Langlands program, which seeks to develop far-reaching connections between
number theory and geometry, is often held as a paradigm of conceptual depth.
What makes it so? To develop some intuitions, I will draw on informal writings
by Kevin Buzzard, a number theorist at Imperial College in London. In 2017,
Buzzard launched his Xena blog,1 in part to document his newfound interest in
the use of computational proof assistants to develop libraries of formally checked
mathematics. Over the years, he has used it as a venue to share what is on his
mind, sometimes explaining to computer scientists and the general public what is
important to him as a mathematician, and sometimes explaining to mathematicians
what he has learned about proof assistants. The first of these two activities is the
one that is relevant here. One does not have to look far to find other mathematicians
expressing similar values, but Buzzard has a gift for exposition, and it is especially
convenient to find so many thoughtful characterizations of mathematical activity
collected in one place.2
One observation is that deep mathematics is usually pretty complicated. Gener-
ally speaking, the deeper the result, the harder it is for the general public or even
mathematicians not directly involved with the research to appreciate it.
So what are the mathematicians I know interested in? Well, let’s
take the research staff in my department at Imperial College. They
are working on results about objects which in some cases take hun-
dreds of axioms to define, or are even more complicated: sometimes
even the definitions of the objects we study can only be formalised
once one has proved hard theorems. For example the definition of
the canonical model of a Shimura variety over a number field can
only be made once one has proved most of the theorems in Deligne’s
paper on canonical models, which in turn rely on the theory of CM
abelian varieties, which in turn rely on the theorems of global class
field theory. That’s the kind of definitions which mathematicians
in my department get excited about [. . . ]. I once went to an en-
tire 24 lecture course by John Coates which assumed local class
field theory and deduced the theorems of global class field theory.
I have read enough of the book by Shimura and Taniyama on CM
abelian varieties to know what’s going on there. I have been to a
study group on Deligne’s paper on canonical models. So after per-
haps 100 hours of study absorbing the prerequisites, I was ready for
the definition of a Shimura variety over a number field. And then
there is still the small matter of the definition of étale cohomology.
(Xena, July 6, 2018)

1https://xenaproject.wordpress.com/
2
Philosophers of mathematics have weighed in as well. See, for example, the special issue of
Philosophia Mathematica devoted to the notion of mathematical depth (volume 23, issue 2, June
2015), or a recent survey by Hamami and Morris [20], which aims to provide mathematics educa-
tors with an overview of recent philosophical work on mathematical representation, explanation,
understanding, and design.
6 JEREMY AVIGAD

Perhaps more important than the complexity of statements and definitions is the
complexity of the proofs. A really deep proof often requires more background
knowledge than any one person can master.

To completely understand a proof of FLT (let’s say, for now, the

proof explained in the 1995 Darmon-Diamond-Taylor paper) you
will need to be a master of the techniques used by Langlands in
his proof of cyclic base change (and I know people who are), and a
master of Mazur’s work on the Eisenstein ideal (and I know people
who are). But what about the far less sexy technical stuff? To move
from the complex analytic theory of modular forms to the algebraic
theory you will need to know the delicate interplay between the
analytic and algebraic theory of moduli spaces of elliptic curves
(and I know people who know this—but I went through some of this
stuff once and it’s far more delicate than I had imagined, and there
are parts of it where the only useful reference seems to be Brian
Conrad’s brain). This last example is perhaps a good example of a
tedious technical issue which it’s very easy to forget about, because
the results are intuitive and the proofs can be technical. There are
many other subtleties which one would have to fully understand
because they’re on the syllabus. Is there really one human being
who would feel confident answering questions on all of this material?
I am really not sure at all. (Xena, September 27, 2019)

It would be a mistake, however, to equate depth with complexity, and other

postings on Xena make it clear that the complexity is only a means to an end.
Complex definitions and proofs are worth the effort when they provide answers to
questions that are judged by the community to be interesting and important. This
last standard is one that is hard to pin down, but, as Buzzard emphasizes, that
does not mean that it is arbitrary or subjective.

What mathematics is fashionable? Just take a look at the work of

the recent Fields Medalists. That’s a pretty good way of telling.
But fortunately, unlike many other fashions, “fashionable math-
ematics” is not controlled by the whim of big companies or some
cabal. Fashionable mathematics is mathematics which justifies it-
self by its ability to answer questions which have been previously
deemed interesting or important. Peter Scholze’s definition of a
perfectoid space opened a door. In the last ten years, perfectoid
spaces have been used to prove new cases of the monodromy–weight
conjecture, to prove the direct summand conjecture, to give a new
proof of purity for flat cohomology, a strengthened version of the
almost purity theorem and so on (and I didn’t even mention ap-
plications to the Langlands philosophy). These are results whose
statements do not mention perfectoid spaces, and some have been
open for decades. This is what makes this kind of mathematics
fashionable—it is giving my community new insights. (Xena, Feb-
ruary 9, 2020)
 VARIETIES OF MATHEMATICAL UNDERSTANDING 7

Buzzard’s posts highlight the importance of having the right definitions, as does
Scholze himself in a lovely quotation featured on Michael Harris’ Mathematics with-
out Apologies blog:3
What I care most about are definitions. For one thing, humans
describe mathematics through language, and, as always, we need
sharp words in order to articulate our ideas clearly. (For example,
for a long time, I had some idea of the concept of diamonds. But
only when I came up with a good name could I really start to think
about it, let alone communicate it to others. Finding the name
took several months (or even a year?). Then it took another two
or three years to finally write down the correct definition (among
many close variants). The essential difficulty in writing “Étale co-
homology of diamonds” was (by far) not giving the proofs, but
finding the definitions.) But even beyond mere language, we per-
ceive mathematical nature through the lenses given by definitions,
and it is critical that the definitions put the essential points into
focus. (Mathematicians without Apologies, June 2, 2018)
The last sentence shows that it isn’t even the definitions per se, but the intuitions
they convey. Sometimes this is cast in terms of providing new representations:
Mathematicians are so good at instantly switching between the var-
ious “obviously equivalent” ways that a mathematician looks at a
complicated algebraic object. (“It’s an equivalence relation! Now
it’s a partition! Now it’s an equivalence relation again! Let your
mental model jump freely to the point of view which makes what
I’m saying in this particular paragraph obvious!”, or “Matrices are
obviously associative under multiplication because functions are as-
sociative under composition.”) (Xena, July 23, 2020)
More generally, the value of a good definition is that it serves as the carrier for
mathematical insight and ideas.
Ideas are the purely artistic part of mathematics. That “lightbulb”
moment, the insight which enables you to solve a problem—this is
the elusive mathematical idea.
Ideas are the part of mathematics that I understand the least,
in a formal sense. Here are two questions:
• What is a group?
• How do you think about groups?
The first one is a precise “scientific” question. A group is a set
equipped with some extra structure, and which satisfies some ax-
ioms. The formal answer is on Wikipedia’s page on groups. A
group is a definition. But the second question is a different kind
of question. Different people think about groups in different ways.
Say G is a group generated by an element x satisfying x5 = x8 = 1.
What can you say about G? If you are a mathematics undergrad-
uate who has just seen the formal definition of a group, you can
probably say nothing. If you have a more mature understanding

3
https://mathematicswithoutapologies.wordpress.com
8 JEREMY AVIGAD

of group theory, you instantly know that this group is trivial, be-
cause you have a far more sophisticated model of what is going on.
Ideas are complicated, and human-dependent. A computer’s idea
of what a group is, is literally a copy of the definition in Wikipedia,
and this is one of the reasons that computers are currently bad at
proving new theorems by themselves. You can develop a com-
puter’s intuition by teaching it theorems about groups, or teach-
ing it examples of groups, or trying to write AI’s which figure out
group theory theorems or examples of groups automatically. But
intuition is a very subtle thing, and I do not understand it at all
well. . . . (Xena, June 20, 2020)
If ideas and representations are the message, the language of perfectoid spaces
and Shimura varieties is the medium. Conceptual understanding is couched in a
structural world-view that is a hallmark of modern mathematics. The approach can
be characterized in terms of all of the following: a focus on axiomatically described
structures; the use of certain tools, such as limit and quotient constructions, for
building new structures from old ones; and an emphasis on characterizing structures
and their properties in relation to other structures, for example, by studying the
morphisms between them and by viewing the structures themselves as elements of
even more abstract structures.
One can discern the first glimmerings of the conceptual view of mathematics in
Gauss’ monumental Disquisitiones Arithmeticae of 1801. That work appeared well
before the advent of modern algebra—for example, Cayley’s abstract definition of
a group did not appear until 1854—but Gauss’ second proof of the law of quadratic
reciprocity and his methods of classifying binary quadratic forms hint at such ab-
stractions. Early in the Disquisitiones Gauss provided a fairly modern treatment
of Wilson’s theorem, which states that the product of the nonzero residues modulo
a prime number p is congruent to −1 modulo p. He attributed Wilson’s inability to
prove the theorem to an excessive reliance on notation, and wrote: “in our opinion
truths of this kind should be drawn from the notions involved rather than from
notations.”4 The slogans “notions over notation” and “concepts over calculation”
were rallying cries for the nineteenth-century transformation of mathematics. Much
as been written about the tensions between the computational rigor of Kronecker,
Weierstrass, and the Berlin school, on the one hand, and more algebraic, geometric,
set-theoretic, and conceptual approaches developed by Riemann, Dedekind, Can-
tor, and Hilbert on the other.5 The conceptual movement grew to maturity in
the nineteenth century, with the work of Emmy Noether, the publication of van
der Waerden’s Modern Algebra in 1930, the founding of Bourbaki, and the work
of Grothendieck. The transformation of mathematics has been so thorough that
few mathematicians will even notice the extent to which the language of structures
pervades Buzzard’s posts. It is simply the way we talk about mathematics today.

4Clark’s translation [11, Section 76] uses “ideas” instead of “notions,” sacrificing the play on
words that is apparent in the original Latin: “At nostro quidem iudicio huiusmodi veritates ex
notionibus potius quam ex notationibus hauriri debebant.”
5Dedekind quoted Gauss’ slogan approvingly in 1895, in one of his later works on the theory
of ideals. For an analysis of how “concepts over calculation” played out in Dedekind’s work in
algebraic number theory, see [1].
 VARIETIES OF MATHEMATICAL UNDERSTANDING 9

Let me summarize some of the morals we can extract from Xena. What are
judged to be the important questions facing mathematics change over time, but
the changes are not random or arbitrary. Mathematics may have been designed
to help us make sense of our physical experiences, but doing so gives rise to new
mathematical experiences, which in turn give rise to new questions and new research
programs. The fact that questions are situated in a tradition that extends through
the centuries is important to judgments of depth. So is complexity: questions of-
ten turn out to be surprisingly hard, and when complex definitions and proofs are
deemed to be necessary to address them, their complexity is a further sign of depth.
The central challenge is to find the right definitions, which provide new represen-
tations, new formulations, and new contextualizations of previous knowledge. The
ability of a research program to draw on disparate intuitions is significant, as is the
ability to unify previously existing fields of study and open up fruitful new avenues
of research. Finally, social aspects of mathematics are important: the fact that
some results are too complex for any one person to master is evidence that we are
pushing against the boundaries of what can be known, and part of what we value
about mathematics is that it supports a collaborative understanding.
It may be helpful to think of conceptual mathematics as a technology that is
designed to extend our cognitive reach. Nineteenth-century algebraic methods
emerged to manage complexity in situations where calculational approaches had
become unwieldy. Structural methods provide a language for defining and reason-
ing about abstract objects, which in turn, provide an effective means of control-
ling information by modularizing arguments, making information salient when it
is needed, and suppressing it when it is a distraction. Depth reflects the extent
to which our conceptual engineering succeeds in that regard, enabling us to rea-
son better and more efficiently. Depth can also be viewed as a counterweight to
cleverness: when it comes to solving really hard problems, cleverness can only get
us so far, and conceptual infrastructure and scaffolding are often needed to get
the job done. To be sure, progress often requires clever tactical skirmishes, but
far-reaching conceptual strategy provides the most substantial means of extending
our knowledge.
We can now assess the Langlands program in these terms. The program has
served to unify several previously existing fields of mathematical research, includ-
ing local and global class field theory, the classical theory of modular forms and
Eisenstein series, the theory of L-functions, harmonic analysis on reductive groups,
and Shimura varieties. It has contributed to major results in number theory, in-
cluding Weil’s conjecture on Tamagawa numbers, the modularity theorem, Fermat’s
last theorem, and the Sato-Tate conjecture. With the associated Langlands conjec-
tures, it provides an ambitious but concrete research program, and it has inspired
major imitations, such as the geometric Langlands program and “mod p” variants.6
The program exhibits all the characteristics that are underscored by Xena. It has
led to the resolution of existing problems, it supports the transfer of ideas between
different fields of inquiry, and it has opened up new questions and new lines of
research. It is historically linked to the geometric and number-theoretic traditions
that stretch back to Euclid. It requires building massively elaborate networks
of structures, with definitions that are often too complex to be written down in
complete detail, and proofs that are even more so. The program draws on more
6I am grateful to the anonymous referee for the contents of this paragraph.
10 JEREMY AVIGAD

mathematics than any one mathematician can master, but with modularization of
concerns and the coordinated efforts, it enables us to make progress as a community.
These are all indications of depth.

4. Assessing discrete geometry

How do the theorems of Section 2 fare with respect to the values described in
Section 3? Not well, I am afraid, but it is illuminating to see why not. It is clear
from the start that the results fail to exhibit many of the features that are praised
in Buzzard’s posts. The definitions involved are not very complex and the proofs
are fairly self-contained; the average mathematician can generally come up to speed
with a week or two’s worth of background reading and a handful of prior journal
articles.
The results are further handicapped by virtue of their subject matter. Discrete
geometry is closely allied with combinatorics, a subject which, in the eyes of many
mathematicians, is a repository of cleverness rather than depth. William T. Gowers
has observed that, in the traditional division of mathematics into the two cultures
of “theory builders” and “problem solvers,” the former are generally held in higher
esteem [12]. He has eloquently defended combinatorics against the criticism that
it is an ad hoc compilation of clever but disjointed results, the result of beavering
away at insignificant problems rather than contributing to the grand mathematical
edifice. The status of combinatorics was raised by a number of impressive results
after the turn of the new millennium, including the celebrated Green–Tao result on
primes in an arithmetic progression [14], as well as generalizations and strengthen-
ings of Szemerédi’s theorem. Such work drew on ideas and methods from ergodic
theory and Fourier analysis, bringing combinatorics closer to the status of structural
mathematics, and making it, in Buzzard’s terminology, fashionable.7 But while this
gave combinatorics a welcome boost, discrete geometry did not rise with it; the lat-
ter still looks like combinatorics used to look before its structural makeover.
But the results of Section 2 are even more tainted by their heavy use of computa-
tion, including optimization techniques, combinatorial enumeration, validated nu-
merical computation, linear programming methods, Monte Carlo simulation, search
techniques, propositional satisfiability algorithms, and computer algebra. These are
mainstays of computer science, engineering, and applied mathematics, but they are
not seen as the purview of pure mathematics. So it hard to view the results of
Section 2 as part of the pure tradition.
When it comes to assessing the theorems of Section 2 with respect to the core
values of understanding and depth, Theorem 2.2, the lower bound on the opti-
mal density of tetrahedral packings, is the easiest to dismiss. Although the results
required insight and hard work, there is not much theory there—no complex defini-
tions and proofs, and no overarching conceptual framework. The result isn’t known
to be optimal, and there seems to be nothing special about the fraction 4000/4671.
And some of the related results, for example, the fact that certain types of random
processes settle into certain types of ordered configurations, are only experimentally
confirmed, though the claims are also supported by general physical principles. I
once heard Sharon Glotzer present an engaging survey of the work at a confer-
ence to celebrate Hales’ sixtieth birthday. At one point in her talk, after stating
7A use of the Hahn-Banach theorem by Gowers in the finite setting is a nice example of a
reconciliation of infinitary and finitary methods [13].
 VARIETIES OF MATHEMATICAL UNDERSTANDING 11

a result, she said “we can show that rigorously,” and then paused. Looking out
at an audience of discrete geometers, number theorists, representation theorists,
computer scientists, and logicians, she added “well, rigorously for us, not for you.”
This tongue-in-cheek remark was a gracious way of acknowledging methodological
differences between those fields and her own, chemical engineering.
Results like Theorem 2.2 are sometimes classified as experimental mathematics,
a term which describes a grab-bag of ways of using computation to explore mathe-
matical phenomena and discover new results. Sometimes results that are discovered
using such techniques can be justified rigorously, after the fact; sometimes the com-
putations are meant only to indicate patterns or lend plausibility to claims. The
field, surveyed in a number of accessible textbooks [4, 5], is lively and popular,
and there is even a journal, Experimental Mathematics, devoted to it. But to the
mathematical elite, the phrase “experimental mathematics” is itself a contradiction
in terms. Many mathematicians look down on such work as being recreational at
best, and, at worst, detrimental to the subject.
Theorem 2.4, the resolution of the Keller conjecture, is the next easiest to dismiss.
It has a few more things going in its favor: the conjecture had been around since
1930, algebraic insights were used to reduce the problem to a combinatorial one, and
the nature of the results—the fact that something counterintuitive and surprising
happens in higher dimensions—has the flavor of real mathematics. But once again,
the methods used to settle the last case put the final result squarely in the realm
of computer science. Propositional satisfiability solvers are used to solve problems
in a wide range of domains, including hardware and software verification as well as
discrete optimization. Mathematicians do not conventionally view search problems
as being a part of mathematics, especially not when the state space is finite. One
mathematician I know, upon hearing of the resolution of Keller’s conjecture, said to
me: “don’t expect mathematicians to be interested—this is a ‘finite computation.’ ”
Theorem 2.1, the Kepler conjecture, rises higher in the ranking. After all, it isn’t
easy to ignore the proof a conjecture that remained open for nearly four centuries.
The proof did not involve heuristic search, but it did involve lengthy calculations,
including brute force enumeration and numerical bounds on large numbers of non-
linear expressions. That is enough to convince many mathematics that the problem
is not really a mathematical one, but rather, a matter of engineering. Appel and
Haken’s proof of the four color theorem was met with a similar response, highlight-
ing the concern that computational methods do not deliver a suitable mathematical
understanding.
In 2018, I wrote an article, The Mechanization of Mathematics [2], that surveyed
recent logic-based uses of computers in mathematics. A colleague commented on a
draft:

. . . the most striking point is how your emphasis is absolutely or-

thogonal to what interests me (and other mathematicians . . . ).
Let me explain this, exaggerating slightly for comparison. First,
for the purpose of this discussion, what I call a traditional math-
ematician is someone who has a permanent position that involves
proving theorems . . . . What they do is try to get a conceptual
and rigorous understanding of which mathematical statements are
true. Both adjectives are important. To them, Hales’ proof of the
Kepler conjecture is nothing like solving the conjecture. And the
12 JEREMY AVIGAD

Flyspeck project is purely computer science. That doesn’t mean

it is not interesting; it is something different, because at least part
of the proof lacks the “conceptual” adjective. You will often hear
mathematicians, talking about their own proofs, saying things like
“there is nothing to understand here, it’s only a computation.”
To most mathematicians, the interesting part of a proof is the
high level structure, and the technical details are only a chore that
is necessary to make sure we haven’t missed anything important.
Notice that the passage emphasizes that the attitude it describes is not a comment
on how interesting and important the proof of the Kepler conjecture is, but, rather,
the extent to which it belongs to traditional mathematics.
Viazovska’s result on the optimality of E8 and the follow-up result on the Leech
lattice fare the best under the standards of Section 3. There are a number of strikes
against them: the proofs are short, they involve a fair amount of calculation, they
do not introduce complicated new definitions, and the discoveries build on intuitions
gleaned from numerical computation and experimentation. So what makes them
real mathematics? The fact that E8 and the Leech lattice are richly complicated
and highly symmetric objects, and the fact that they many have interesting alge-
braic properties, suggest that they, themselves, are conceptually deep. It also helps
that the proof draws on modular forms, representation theory, and harmonic anal-
ysis, themselves well established as central components of traditional mathematics.
Cohn’s description of Viazovska’s work emphasizes these features:
“. . . it’s wonderful to see a relatively simple proof of a deep theorem
in sphere packing.”
“Her proof is thus a notable contribution to the story of E8 , and
more generally the story of exceptional structures in mathematics.”
“Viazovska . . . establishes a new connection between modular
forms and discrete geometry.”
“Instead of justifying sphere packing by aspects of the problem
or its applications, we’ll justify it by its solutions: a question is
good if it has good answers. Sphere packing turns out to be a far
richer and more beautiful topic than the bare problem statement
suggests. From this perspective, the point of the subject is the
remarkable structures that arise as dense sphere packings.”
“. . . modular forms are deep and mysterious functions connected
with lattices, as are the magic functions, so wouldn’t it make sense
for them to be related?”
“Despite our lack of understanding [of other higher dimensional
sphere packing problems], the special role of eight and twenty-four
dimensions aligns with our experience elsewhere in mathematics.
Mathematics is full of exceptional or sporadic phenomena that oc-
cur in only finitely many cases, and the E8 and Leech lattices are
prototypical examples. These objects do not occur in isolation, but
rather in constellations of remarkable structures.”
In sum, what makes the results attractive are the structures involved and the rich
connections to other parts of mathematics. Even the title of the article, “A con-
ceptual breakthrough in sphere packing,” emphasizes the conceptual aspect.
 VARIETIES OF MATHEMATICAL UNDERSTANDING 13

5. Varieties of understanding
In Section 3, I characterized a view of mathematical understanding, and in Sec-
tion 4, I observed that the results of Section 2 do not rank particularly well by that
standard. What are we to make of this?
Let’s clarify what is at stake. At issue is not the quality or importance of the
results; the theorems are impressive, and mathematicians are generally happy to
grant them the respect they deserve. The questions are natural, the theorems are
pretty, the methods are clever, and the insights are substantial. The mathematical
community is open-minded and generous, and it is quick to appreciate the value
of mathematical results in physics, chemistry, computer science, and even discrete
geometry, whether or not they are seen as core mathematics.
Nor is it a question of popularity. The results of Section 2 have all made it into the
popular press and have gained a level of recognition that very few core mathematical
results ever receive. It is not a matter of institutional support; most of the authors
cited in Section 2 have good jobs and stable academic positions. The question is
not about academic recognition; the results of Section 2 have high citation counts,
generally much higher than papers associated with, say, the Langlands problem.
Nor is it a question of funding; grants are generally larger and easier to come by in
applied mathematics, computer science, and engineering, and availability of funding
in areas of pure mathematics tends to track their proximity to the more applied
branches. By that standard, research in discrete geometry does just fine.
Moreover, I don’t see the mathematical elite, or anyone making pronouncements
as to what traditional mathematicians do, complaining about any of that. Mathe-
maticians are often proud to distance themselves from the claim that anything they
do should be considered practically useful. People don’t go into pure mathematics
because they are looking for fame and fortune. They do it because they find it
enjoyable, meaningful, and satisfying; because they admire and respect the power
of mathematical thought; because they see themselves as part of a long and worthy
tradition; and because they find the subject elegant and beautiful.
The question I am asking here is therefore more of a philosophical one. Insofar
as there is an essence of mathematical understanding, is it adequately captured
by the conceptual view described in Section 3? Or would a better articulation of
this essence leave more room for the results of Section 2? Nobody denies that the
results are rigorous, and nobody denies that they introduce interesting new ideas.
They certainly meet some of the criteria for good mathematics: they open up new
capacities for rigorous thought; they provide new means of solving hard problems;
they offer new representations, conceptualizations, and methods; and they raise
new questions and avenues for exploration.
Let me replace the broad philosophical question with two that are more focused:
• Can the kinds of computational methods used to obtain the results of Sec-
tion 2 contribute to core areas of mathematics and help answer some of the
questions that are currently viewed as important?
• Can/should/will computational methods like the ones used toward obtain-
ing the results of Section 2 become, in and of themselves, part of a proper
mathematical understanding?
The answer to the first question is likely to be “yes.” Computers fundamentally
change the kinds of reasoning we can perform and dramatically increase the level of
14 JEREMY AVIGAD

complexity of the inferences that we can carry out with any measure of reliability.
The Langlands program, for example, seeks correspondences between certain types
of number-theoretic and geometric data. Perhaps there is an aesthetic judgment
as to whether a proposed correspondence is sufficiently natural, but insofar as the
problem can be made precise, I see no reason to think that search techniques can’t
be used to supplement conventional mathematical insight.
This last claim may seem ludicrous to practitioners. Computers are certainly
nowhere near being able to construct the kinds of conceptual definitions and proofs
that experts do. But let’s recognize that we have been looking for those particular
kinds of definitions and proofs because we can; they are the particular means we
have developed to meet mathematical challenges with the cognitive resources at
our disposal. The computer is a new type of resource, and it can produce different
kinds of solutions. Depending on how you count, we have been using the conceptual
approach for a century or more. In contrast, we still have a lot to learn about
how computers can be used to address core mathematical problems. Maybe it will
require a modern-day Gauss to figure out how to harness the power of computation,
or maybe we will get there in incremental steps. But it is impossible to deny that
raw computing power extends our mathematical capacities in a significant way, and
it is only a matter of time before we learn how to make proper use of it.
Whereas the first question is pragmatic, the second is a matter of mathemat-
ical values. If it turns out that some Langlands-like questions can be answered
with the use of computation, there is always the possibility that the mathematical
community will interpret this as a demonstration that, in hindsight, the Langlands
program is not as deep as we thought it was. There is always room to say, “Aha!
Now we see that it is just a matter of computation.” The question is whether
computational methods might one day become fashionable in Buzzard’s sense. If
computational methods prove useful in solving problems that were antecedently
deemed important, questions about the new methods themselves might be judged
to be important in their own right. Can we imagine that happening?
The history of mathematics offers some useful lessons in that regard. The first
lesson is that mathematical values change over time, not just views as to what
counts as rigorous, but also views as to what counts as a proper understanding.
Consider, for example, the fact that in the Western tradition, from ancient times
to the seventeenth century, geometry was held to be the proper foundation for
mathematics. Today, we think of an equation like y = x2 + 2x + 1 as expressing a
relationship between elements of the real number structure, a relationship that can
be depicted geometrically by graphing it. For the early algebraists, the perspective
was inverted: variables stood for geometric magnitudes like lengths and areas,
and algebraic manipulations were abstract proxies for geometric construction. In
1637, Descartes’ landmark Géométrie opened up new opportunities for the use of
algebraic methods in solving geometric problems, but even for Descartes, such a
problem was not solved until the algebraic solution was reinterpreted as a geometric
construction.
Such a detour through algebraic methods was unacceptable to Newton, who felt
that although such methods could be used as a heuristic for finding solutions, a
proposition was not suitably demonstrated, and not properly understood, until it
was given a purely geometric presentation. It is well known that Newton developed
an algebraic version of calculus around 1665, but suppressed publication in favor
 VARIETIES OF MATHEMATICAL UNDERSTANDING 15

of the more geometric methods of his monumental Philosophiæ Naturalis Principia

Mathematica, published in 1687. Newton’s antipathy to algebraic methods has been
well documented by Guicciardini [15, 16]. For example, in Newton’s writings of the
1690s, we find:
Men of recent times, eager to add to the discoveries of the ancients,
have united specious arithmetic with geometry. Benefitting from
that, progress has been broad and far reaching if your eye is on
the profuseness of output but the advance is less of a blessing if
you look at the complexity of its conclusions. For these computa-
tions, progressing by means of arithmetical operations alone, very
often express in an intolerably roundabout way quantities which
are designated by the drawing of single line.8
He stated such a position even more strongly in the mid-1670s, when he judged the
ancient approach to solving a problem by Pappus to be “more elegant by far than
the Cartesian one,” since Descartes
achieved the results by an algebraic calculus which, when trans-
posed into words (following the practice of the Ancients in their
writings) would prove to be so tedious and entangled as to produce
nausea, nor might it be understood.9
In late 1683 or early 1684, Newton lectured on the “treatment of composition,”
which Guicciardini describes as “the synthetic, constructive phase of the problem-
solving process.” According to Newton:
Analysis guides us to the composition, but true composition is not
achieved before it is freed from analysis. Let even the slightest trace
of analysis be present in the composition and you will not yet have
attained true composition.10
As Guicciardini’s scholarship shows, Newton’s views were based in part on some-
what mystical attributions to the wisdom of the ancients, but they were also moti-
vated by deep-seated views about mathematical meaning. Today, we tend to view
the translation of a visual argument into set-theoretic terms as a step toward rigor.
From Newton’s standpoint, this is a step away from a proper mathematical under-
standing. It may be tempting to write Newton off as more of a physicist than a
mathematician, but that is not easy to do: throughout his career he held mathe-
matics to be prior to its physical applications, and his insistence on rigor and purity
of method, as well as his ability to solve fiendishly difficult mathematical problems,
mark him as one of our own.
The nineteenth-century transition to the modern conceptual view was beset by
similar misgivings. The conceptual approach was designed to address situations
where calculational methods were straining to reign in complexity in the theory of
equations, in analysis, and in number theory. In analogy to the seventeenth-century
view of the use of algebraic methods in geometry, the new abstract methods were

8This passage, from Newton’s Mathematical Papers [31, vol. 7, p. 251], appears in Guicciardini
[16, Section 4.6], along with others of a similar character. See also [16, Chapter 2].
9From [31, vol. 4, p. 277], translated by Westfall [30, p. 379], and quoted by Guicciardini [15,
p. 29]
10From [31, vol. 4, p. 477], quoted by Guicciardini [16, pp.74–75].
16 JEREMY AVIGAD

seen as heuristic shortcuts to obtaining results that were properly justified by ex-
plicit calculation. Once again, the community was initially reluctant to stray too
far. Today it is hard to appreciate the extent to which calculation was central
to the lives of nineteenth-century mathematicians, even the conceptual innovators.
Gauss famously recounted, in a letter to his student, Johann Franz Encke, the
great pleasure it gave him to spend an idle quarter hour tabulating frequencies of
prime numbers.11 Riemann’s short paper on the zeta function was a tour-de-force
of calculation, and Dedekind motivated his algebraic approaches to number theory
by working out concrete examples. Calculation has always been central to math-
ematics; we can perhaps forgive nineteenth-century mathematicians for thinking
that symbolic representations and algorithms are an essential part of mathematical
understanding.
A second lesson from history is that when mathematics does change, precon-
ceived notions as to what constitutes understanding are invariably pushed aside
by mathematical success. Mathematics isn’t beholden to dogma; we may think we
know what constitutes good mathematics, but mathematics may tell us otherwise.
In the eighteenth century, the algebraic approach to calculus gave us Euler, the
Bernoullis, Lagrange, and Laplace, and it gave us advances in mechanics, proba-
bility, and physics. It is no wonder we were willing to turn our backs on geometry.
The abstract, infinitary methods of the nineteenth century gave us algebraic num-
ber fields, complex analysis, manifolds, Lie groups, and the beginnings of algebraic
geometry. Concerns about computational meaning simply took a back seat.
It is perhaps the fate of mathematicians through the ages to think that their
views of mathematics are timeless. But today the notion that mathematical truth is
grounded in geometry seems quaint, and contemporary mathematicians scoff at the
suggestion that the subject is fundamentally about calculation. That should remind
us that mathematicians of the future may one day judge our present view that the
essence of mathematics lies in a certain type of conceptual understanding—built on
networks of complicated algebraic definitions, locally surveyable, with communities
of experts straining to hold the pieces together—as equally quaint and outdated.
To be sure, the conceptual view has won us great success in answering traditional
questions and generating new ones. But as new methods arise and fashions change,
our views of mathematics will change as well.
And so our current preoccupation with conceptual methods could well give way
to a more expansive view of mathematical understanding, one in which the com-
puter plays a more central role. This isn’t to say that we will then abandon the
conceptual viewpoint entirely. A third lesson from the history of mathematics is
that, as the subject evolves, it has a prodigious capacity for reinterpreting its past.
We still recognize the importance of Euclid’s Elements, Archimedes’ calculations of
volumes, and Appolonius’ work on conics; we acknowledge the brilliance of Newton
and Leibniz in the uniform methods they developed to calculate areas and tangents;
we appreciate the delicate arguments used by Abel and Galois to demonstrate the
unsolvability of the quintic; and we value Gauss’ detailed manipulations of qua-
dratic forms. We see all of these as part of the modern, conceptual understanding,
or, at least, as an essential part of its history. If eventually we end up with a view of
mathematical understanding that can better accommodate the use of computation,
it will likely be seen as an inevitable refinement of the conceptual point of view.
11See, for example, [28].
 VARIETIES OF MATHEMATICAL UNDERSTANDING 17

6. Conclusions
By highlighting the fact that views of mathematical understanding change over
time, I do not mean to deny that mathematics has any fixed characteristics. The
longevity and continuity of the subject, as well as our ability to understand, ap-
preciate, and build on the results of the past, show that there is something about
mathematics that manages to transcend the vicissitudes of time, place, and culture.
It has always been concerned with solving hard problems, communicating ideas in
precise ways, and establishing stable foundations for scientific reasoning. The sub-
ject aims to establish coordination between practitioners not only at a given point
in time, but across centuries. Its remarkable success at doing so is evidence that
mathematics draws on something fundamental about the way we cognize the world
around us.
What changes, rather, are our judgments as to the specific ways we can best
achieve our mathematical goals. These judgments depend on historical circum-
stances, including the body of mathematical knowledge accumulated to date, the
demands of science, social and institutional factors, and, yes, technology. The only
sure-fire way of finding out where mathematics is going is to wait and see, but we
ourselves are part of the mathematical tradition, and our philosophical views play a
role in guiding its development. Being mindful of these views and discussing them
is an important part of trying to understand what constitutes good mathematics.
Potential uses of computational methods, in particular, should challenge us to
think about our mathematical goals and how they are best served. It is notable
that the headline photograph to Cohn’s Notices article features Viazovska at her
computer, immersed in thought, with one hand on a mouse and with a coffee cup
nearby. The caption reads “Maryna Viazovska solved the sphere packing problem
in eight dimensions.” For all we know, in the picture, she may be catching up
on email or surfing the web for sports results. The caption, however, suggests
that she is using the computer to unlock deep mathematical secrets. The iconic
trope of a mathematician standing before a blackboard covered with symbols and
arrows may be giving way to the image of a mathematician at a keyboard, coaxing
mathematical understanding from calculation, simulation, and search. That should
give us something to think about.

References
1. Jeremy Avigad, Methodology and metaphysics in the development of Dedekind’s theory of
ideals, The Architecture of Modern Mathematics (José Ferreirós and Jeremy Gray, eds.),
Oxford Univ. Press, Oxford, 2006, pp. 159–186. MR 2258019
2. , The mechanization of mathematics, Notices Amer. Math. Soc. 65 (2018), no. 6,
681–690. MR 3792862
3. Armin Biere, Marijn Heule, Hans van Maaren, and Toby Walsh (eds.), Handbook of satisfia-
bility, IOS Press, 2009.
4. Jonathan Borwein and David Bailey, Mathematics by experiment, second ed., A K Peters,
Ltd., Wellesley, MA, 2008, Plausible reasoning in the 21st Century. MR 2473161
5. Jonathan Borwein and Keith Devlin, The computer as crucible: An introduction to experi-
mental mathematics, A K Peters, Ltd., Wellesley, MA, 2009. MR 2464847
6. Joshua Brakensiek, Marijn Heule, John Mackey, and David Narváez, The resolution of Keller’s
conjecture, International Joint Conference on Automated Reasoning (IJCAR) 2020 (Nicolas
Peltier and Viorica Sofronie-Stokkermans, eds.), Springer, 2020, pp. 48–65.
7. Elizabeth R. Chen, Michael Engel, and Sharon C. Glotzer, Dense crystalline dimer packings
of regular tetrahedra, Discrete Comput. Geom. 44 (2010), no. 2, 253–280. MR 2671012
18 JEREMY AVIGAD

8. Henry Cohn, A conceptual breakthrough in sphere packing, Notices Amer. Math. Soc. 64
(2017), no. 2, 102–115. MR 3587715
9. Henry Cohn, Abhinav Kumar, Stephen D. Miller, Danylo Radchenko, and Maryna Viazovska,
The sphere packing problem in dimension 24, Ann. of Math. (2) 185 (2017), no. 3, 1017–1033.
MR 3664817
10. J. H. Conway and S. Torquato, Packing, tiling, and covering with tetrahedra, Proc. Natl.
Acad. Sci. USA 103 (2006), no. 28, 10612–10617. MR 2242647
11. Carl Friedrich Gauss, Disquisitiones arithmeticae, Yale University Press, New Haven, Conn.-
London, 1966, Translated into English by Arthur A. Clarke. MR 0197380
12. W. T. Gowers, The two cultures of mathematics, Mathematics: frontiers and perspectives,
Amer. Math. Soc., Providence, RI, 2000, pp. 65–78. MR 1754768
13. , Decompositions, approximate structure, transference, and the Hahn-Banach theorem,
Bull. Lond. Math. Soc. 42 (2010), no. 4, 573–606. MR 2669681
14. Ben Green and Terence Tao, The primes contain arbitrarily long arithmetic progressions,
Ann. of Math. (2) 167 (2008), no. 2, 481–547. MR 2415379
15. Niccolò Guicciardini, Reading the Principia: The debate on Newton’s mathematical methods
for natural philosophy from 1687 to 1736, Cambridge University Press, Cambridge, 1999.
MR 1725356
16. , Isaac Newton on mathematical certainty and method, MIT Press, Cambridge, MA,
2009. MR 2562084
17. Amir Haji-Akbari, Michael Engel, Aaron S. Keys, Xiaoyu Zheng, Rolfe G. Petschek, Peter
Palffy-Muhoray, and Sharon C. Glotzer, Disordered, quasicrystalline and crystalline phases
of densely packed tetrahedra, Nature 462 (2009), 773–777.
18. Thomas Hales, Mark Adams, Gertrud Bauer, Tat Dat Dang, John Harrison, Le Truong
Hoang, Cezary Kaliszyk, Victor Magron, Sean McLaughlin, Tat Thang Nguyen, Quang Truong
Nguyen, Tobias Nipkow, Steven Obua, Joseph Pleso, Jason Rute, Alexey Solovyev, Thi
Hoai An Ta, Nam Trung Tran, Thi Diep Trieu, Josef Urban, Ky Vu, and Roland Zumkeller,
A formal proof of the Kepler conjecture, Forum Math. Pi 5 (2017), e2, 29. MR 3659768
19. Thomas C. Hales, A proof of the Kepler conjecture, Ann. of Math. (2) 162 (2005), no. 3,
1065–1185. MR 2179728
20. Yacin Hamami and Rebecca L. Morris, Philosophy of mathematical practice: a primer for
mathematics educators, ZDM Mathematics Education 52 (2020), 1113–1126.
21. Ott-Heinrich Keller, Über die lückenlose Erfüllung des Raumes mit Würfeln, J. Reine Angew.
Math. 163 (1930), 231–248. MR 1581241
22. Andrzej P. Kisielewicz, Rigid polyboxes and Keller’s conjecture, Adv. Geom. 17 (2017), no. 2,
203–230. MR 3652241
23. Jeffrey C. Lagarias and Peter W. Shor, Keller’s cube-tiling conjecture is false in high dimen-
sions, Bull. Amer. Math. Soc. (N.S.) 27 (1992), no. 2, 279–283. MR 1155280
24. John Mackey, A cube tiling of dimension eight with no facesharing, Discrete Comput. Geom.
28 (2002), no. 2, 275–279. MR 1920144
25. Oskar Perron, Über lückenlose Ausfüllung des n-dimensionalen Raumes durch kongruente
Würfel, Math. Z. 46 (1940), 1–26. MR 3041
26. , Über lückenlose Ausfüllung des n-dimensionalen Raumes durch kongruente Würfel.
II, Math. Z. 46 (1940), 161–180. MR 2185
27. S. Szabó, A reduction of Keller’s conjecture, Period. Math. Hungar. 17 (1986), no. 4, 265–277.
MR 866636
28. Yuri Tschinkel, About the cover: on the distribution of primes—Gauss’ tables, Bull. Amer.
Math. Soc. (N.S.) 43 (2006), no. 1, 89–91. MR 2201552
29. Maryna S. Viazovska, The sphere packing problem in dimension 8, Ann. of Math. (2) 185
(2017), no. 3, 991–1015. MR 3664816
30. Richard S. Westfall, Never at rest: A biography of Isaac Newton, Cambridge University Press,
Cambridge, 1980. MR 741027
31. D. T. Whiteside (ed.), The mathematical papers of Isaac Newton, Cambridge University Press,
Cambridge, 1967–1981, in seven volumes. MR 0505130

Department of Philosophy, Baker Hall 161, Carnegie Mellon University, Pitts-

burgh, PA 15213, USA
Email address: [email protected]