AMS FALL SECTIONAL SAMPLER

From left to right: Christine Heitsch, Jonathan R. Kujawa, Govind Menon, Kevin Pilgrim, and Bruce Sagan.

Make the time to visit any of the AMS Fall Sectional Meetings.
In this sampler, the speakers above have kindly provided introductions to their Invited
Addresses for the upcoming AMS Fall Sectional Meetings.

Semigroups of Branched Mapping Classes:

September 9–10, 2017 Dynamics and Geometry
Central Meeting; (Saturday–Sunday) by Kevin M. Pilgrim (Indiana University,
University of North Texas, Denton, TX Bloomington)
This meeting's location was erroneously listed as University of Texas in
page 824
the print edition. Notices apologizes for any confusion.

Building Polyhedra by Self-Assembly

September 16–17, 2017
by Govind Menon (Brown University)
Eastern Meeting; (Saturday–Sunday) page 822
State University of New York–Buffalo,
Buffalo, NY The Protean Chromatic Polynomial
by Bruce Sagan (Michigan State University)
page 828

Strings, Trees, and RNA Folding

September 23–24, 2017
by Christine Heitsch (Georgia Institute
Southeastern; (Saturday–Sunday)
of Technology)
University of Central Florida–Orlando,
page 817
Orlando, FL
Realizing the Spectrum of Tensor Categories
*A sampler from the Fall Western Sectional Meeting will
appear in the November issue of Notices.
by Jonathan R. Kujawa (University of
Oklahoma)
page 820

For permission to reprint this article, please contact:

reprint-permission@[Link].
DOI: [Link]
 AMS FALL SECTION SAMPLER
Christine Heitsch
Strings, Trees, and RNA Folding
We highlight some challenges and opportunities at the
interface of discrete mathematics and molecular biology,
illustrating that this interaction motivates new com-
binatorial theorems as well as advancing biomedical
applications.
High school teaches us that RNA’s role is to mediate
the production of proteins from DNA. Closer inspection,
however, reveals a vast complexity of structure and
diversity of function. As illustrated in Figures 1 and 2,
RNA molecules are essential to cellular processes ranging
from bacterial communication to viral capsid assembly.
It goes almost without saying that advancing knowledge
of how RNA functions in these diverse roles has the
potential for tremendous scientific impact.

Figure 2. How do viral capsids assemble? The RNA

genome (gold) is partially visible inside the quasi-
icosahedral protein capsid (purple). The resolution
of this detailed crystal structure [2] is such that the
number and approximate length of many runs of
stacked base pairs are known, but the composition
and connectivity of these helices are not. New knowl-
edge of RNA branching configurations is needed to
understand how this viral sequence folds into its
dodecahedral cage.

must first know how it is structured. Generically, an RNA

molecule is a single nucleotide sequence which folds into a
3D conformation via a set of noncrossing, intrasequence
Figure 1. How do bacteria communicate? Four RNA base pairings known as a secondary structure. Given
molecules (diagrammed here through marine bio- that the goal of 3D experimental determination remains
luminescence) are essential to the quorum sensing inaccessible for most RNA structures, computational
process by which bacteria regulate collective behav- predictions of possible base pairing configurations, such
ior, ranging from this benign light display to cholera as the four outlined in Figure 1, are essential for generating
toxicity. Despite high sequence similarity and known functional hypotheses.
functional redundancy [1], the branching of the four Yet accurate prediction of the branching of these
possible RNA structures can vary significantly. New structures remains a fundamental open question. Given
results in combinatorics and its applications provide the combinatorial nature of the problem, this is an
insight into this important, yet difficult-to-determine, opportunity for discrete models, methods, and analyses
molecular characteristic. to provide insights into a complex biomolecular process.
To begin to appreciate the challenge, first consider the
A basic biological principle is that “function follows model sequence g6 a4 c6 = gggggg aaaa cccccc. Under
form”; that is, to understand what a molecule does, one the Watson-Crick association of g and c, it folds into a
structure with one helix (run of 6 stacked base pairs) and
Christine Heitsch is professor of mathematics at the Georgia one loop (single-stranded region) known as a “hairpin.”
Institute of Technology. Her e-mail address is heitsch@math Thus, the lower left structure in Figure 1 has 3 hairpins,
.[Link]. 1 internal loop, and the central loop from which branch
For permission to reprint this article, please contact: 3 helical arms.
reprint-permission@[Link]. To analyze this coding of 2D branching information in
DOI: [Link] a linear biochemical chain, we model an RNA structure

September 2017 Notices of the AMS 817

 AMS FALL SECTION SAMPLER

Figure 3. How do branching configurations relate? Five arrangements of loops and helices for the RNA se-
quence A4 (G6 A4 C6 A4 )3 , determined by the pairing of G (labeled 1, 3, 5) and C (labeled 2, 4, 6) segments.
Arrows indicate movement between structures under allowable pairing exchanges. Two configurations with
a graph geodesic of length 2 comprise a meander; the corresponding noncrossing perfect matchings form a
single closed loop when drawn on the same endpoints, one above and the other below, as illustrated in the in-
set figure. Although these closed meanders arise in various mathematical settings, their exact enumeration
problem remains open.

as a plane tree—that is, a rooted tree whose subtrees among different types of loop structures? Is there a typical
are linearly ordered, by mapping loops to vertices and degree of loop branching? What is the dependence on the
helices to edges. In Figures 1 or 3 this would correspond thermodynamic optimization parameters? Since the accu-
to collapsing each circular region to a point and merging racy of computational base pairing predictions decays
two parallel lines into one. This abstraction preserves rapidly with sequence length, these theoretical results
the basic structural arrangement, including information help separate structural signals from thermodynamic
about the sequence ordering and the energetic types of noise, thereby supporting alternative hypotheses in viral
different loop structures. capsid assembly.1
We then generalize Conversely, under a suitable abstraction, the space
our toy example from
Mathematics is a above to consider sat-
of branching configurations reveals significant combi-
natorial structure. The challenge of understanding the
vital source of new urated (fully paired,
hence the lowest free
different possible low-energy secondary structures for
an RNA sequence motivates a new local move on plane
structural insights. energy) structures for trees/noncrossing perfect matchings. This yields graphs,
𝑅 = a4 (g6 a4 c6 a4 )𝑛 as in Figure 3, isomorphic to the Hasse diagram for the
and their correspond- lattice of noncrossing partitions. By recapitulating this
ing plane trees. This highlights one of the critical issues: well-known combinatorial structure, we gain insights that
there can be exponentially many different possible low-
now allow us to count and characterize the orbits under
energy branching configurations for an arbitrary RNA
the Kreweras complementation operator. This result then
sequence.
naturally leads to considering some new approaches to
Hence, even in a situation, as in Figure 2, where
the challenging open problem of meander enumeration.
some experimental information about the 3D structure is
In this way, we illustrate that the interaction of the two
known, mathematics is a vital source of new structural
disciplines is fruitful for mathematics as well as beneficial
insights. As we will discuss, by using strings and trees as
for biology.
a combinatorial model of RNA folding, we can analyze dif-
ferent possible branching configurations at viral genome
length scales. We prove theorems, using methods from 1
For more on the mathematics of molecular machines, see “Build-
enumerative, probabilistic, and geometric combinatorics, ing Polyhedra by Self-Assembly” by Govind Menon in this issue
which address questions such as: What are the trade-offs (page 822).

818 Notices of the AMS Volume 64, Number 8

 AMS FALL SECTION SAMPLER
References
[1] D. H. Lenz, K. C. Mok, B. N. Lilley, R. V. Kulkarni, N. S.
Wingreen, and B. L. Bassler, The small RNA chaperone Hfq
and multiple small RNAs control quorum sensing in Vibrio
harveyi and Vibrio cholerae, Cell 117 (2004), no. 1, 69–82.
[2] L. Tang, K. N. Johnson, L. A. Ball, T. Lin, M. Yeager, and
J. E. Johnson, The structure of Pariacoto virus reveals a do-
decahedral cage of duplex RNA, Nat. Struct. Biol. 8 (2001),
77–83.

Image Credits
Figure 1 courtesy of Henke and Bassler, Princeton Univer-
sity.
Figure 2 courtesy of Nature Structural and Molecular Biol-
ogy, Nature Publishing Group, reprinted with permis-
sion [2].
Figure 3 courtesy of Christine Heitsch, Georgia Institute
of Technology. Secondary structures predicted and
drawn by mfold software, available via
[Link].
Photo of Christine Heitsch courtesy of James Heitsch,
University of Illinois at Chicago.

September 2017 Notices of the AMS 819

 AMS FALL SECTION SAMPLER
Jonathan R. Kujawa object of 𝒞, then 𝑀 ⊗ 𝑁 is an object of 𝐼, and (2) 𝐴 ⊕ 𝐵
is an object of 𝐼 if and only if both 𝐴 and 𝐵 are objects
of 𝐼. The second condition is the requirement that 𝐼 be
a “thick” subcategory. Thinking of the kernels of ring
homomorphisms, the thick condition becomes plausible
once we notice that for any functor of tensor categories,
𝐹 ∶ 𝒞 → 𝒟, we have 𝐹(𝐴 ⊕ 𝐵) ≅ 0 if and only if 𝐹(𝐴) ≅ 0
and 𝐹(𝐵) ≅ 0. Finite-dimensional 𝑘-vector spaces is a
categorical version of a field in that it has no proper
ideals. Namely, if 𝑉 is a nonzero vector space in 𝐼, then by
Realizing the Spectrum of Tensor Categories writing it as a direct sum of one-dimensional subspaces
The usefulness of attaching geometry to algebraic objects and using the thick condition, it follows that 𝑘 lies in 𝐼.
goes back at least to Descartes. Using geometry we can Therefore every vector space 𝑊 ≅ 𝑊 ⊗ 𝑘 lies in the tensor
obtain qualitative information about our original algebraic ideal 𝐼.
object. Given a polynomial with real coefficients, 𝑝(𝑥), we Things become more interesting when the category is
teach schoolchildren to look at the graph of 𝑦 = 𝑝(𝑥) in not semisimple. In this case the structure of the category
ℝ2 . By examining the x-intercepts, y-intercepts, and end can be quite complicated. Just classifying the objects is
behavior they can say things about the degree, leading already a hopeless task in all but the easiest examples.
coefficient, the constant term, and so on. A more reasonable goal is to describe the tensor ideals.
A more modern example is the prime ideal spectrum, Doing so gives us an idea of the coarse structure of the
Spec(𝑅), of a commutative ring 𝑅. More generally, given a category and, in particular, gives information about when
finitely generated 𝑅-module 𝑀, we can define the support one object can be obtained from another by direct sums,
of 𝑀, supp(𝑀), to be the subset of Spec(𝑅) consisting direct summands, and tensor products.
of all prime ideals 𝑃 such that 𝑀 localized at 𝑃 does In easy cases, the tensor ideals can be described by
not vanish. The geometry of the spectrum and support hand. For example, let 𝑘 be a fixed ground field which
again captures algebraic information. For example, for is algebraically closed and of characteristic 𝑝 > 0. Let
two modules the support of a direct sum is the union of 𝐶𝑝 be the cyclic group of order 𝑝 and let 𝐶𝑝 -mod be
the supports, and the support of the tensor product is the category of finite-dimensional 𝐶𝑝 -modules, that is,
the intersection of the supports. finite-dimensional 𝑘-vector spaces with a linear action by
Now suppose 𝒞 is a category which admits both a direct the elements of 𝐶𝑝 . Then 𝐶𝑝 -mod again admits a tensor
sum and a tensor product. We also assume the tensor product. Namely, given 𝐶𝑝 -modules 𝑀 and 𝑁, define
product is “commutative” in that there are canonical 𝑀 ⊗ 𝑁 to be the tensor product as vector spaces with 𝐶𝑝
isomorphisms 𝑋 ⊗ 𝑌 ≅ 𝑌 ⊗ 𝑋 for all pairs of objects 𝑋 action given by the formula 𝑔.(𝑚 ⊗ 𝑛) = (𝑔.𝑚) ⊗ (𝑔.𝑛)
and 𝑌. For example, 𝒞 could be 𝑘-vec, the category of finite- for 𝑔 ∈ 𝐶𝑝 . There are 𝑝 nonisomorphic 𝐶𝑝 -modules,
dimensional 𝑘-vector spaces over a fixed ground field 𝑘 𝑄1 , … , 𝑄𝑝 , which cannot be written as a direct sum of
with the usual direct sum and tensor product operations. smaller modules. The dimension of 𝑄𝑑 as a 𝑘-vector space
In this case the canonical isomorphism 𝑉 ⊗ 𝑊 → 𝑊 ⊗ 𝑉 is is 𝑑. In particular, 𝑄1 is the unique simple 𝐶𝑝 -module,
given by the “flip” map 𝑣 ⊗ 𝑤 ↦ 𝑤 ⊗ 𝑣. The ground field and 𝑄𝑝 is the unique projective indecomposable module.
acts as the identity for the tensor product in that there In this case, direct calculations show that there are two
are canonical isomorphisms 𝑘 ⊗ 𝑉 ≅ 𝑉 ≅ 𝑉 ⊗ 𝑘. Such tensor ideals: the entire category and the full subcategory
tensor categories are common throughout mathematics. consisting of projective modules. A more general and
Another elementary example is the category of closed, much more difficult problem is to classify the tensor
orientable surfaces with direct sum given by disjoint ideals of 𝐺-mod when 𝐺 is any finite group in which 𝑝
union and tensor product given by connected sum. divides the order of 𝐺.
Such a category can be thought of as a categorical Now let 𝒦 be a tensor triangulated category consisting
analogue of a commutative ring, with the direct sum as the of compact objects. That is, 𝒦 is a triangulated category
“addition” and the tensor product as the “multiplication.” with a compatible tensor product and subject to suitable
With this in mind it is natural to ask for the notion of finiteness assumptions. Approximately ten years ago Paul
an ideal. A tensor ideal of 𝒞 is a full subcategory 𝐼 which Balmer introduced geometry by defining the spectrum of
has the property that (1) if 𝑀 is an object of 𝐼 and 𝑁 an 𝒦 in the spirit of commutative ring theory. In this setting
a tensor ideal is taken to be a full triangulated subcategory
Jonathan R. Kujawa is professor of mathematics at the University with properties (1) and (2) as above. A tensor ideal 𝐼 is
of Oklahoma. His e-mail address is kujawa@[Link]. called prime if it is a proper ideal and if whenever 𝐴 ⊗ 𝐵 is
The research of the author was partially supported by NSF grant an object of 𝐼, either 𝐴 or 𝐵 is an object of 𝐼. The spectrum
DMS-1160763 and NSA grant H98230-16-0055. of 𝒦, Spc(𝒦), is then the collection of all prime tensor
For permission to reprint this article, please contact: ideals with the Zariski topology. Balmer also defined the
reprint-permission@[Link]. support of any object 𝑀 in 𝒦 as the set of all prime
DOI: [Link] tensor ideals which do not contain 𝑀.

820 Notices of the AMS Volume 64, Number 8

 AMS FALL SECTION SAMPLER
Balmer proved that
the spectrum and
support for 𝒦 are
Tensor triangular
universal in a precise geometry is a
sense among support
theories for 𝒦. He also beautiful and
proved that it provides
a classification of the powerful theory.
thick tensor ideals of
𝒦 and, hence, that it provides a geometric answer to our
earlier question. Balmer, collaborators, and others have
gone on to show that tensor triangular geometry is a rich
theory which brings valuable new tools and insights to
a variety of settings. Those who are able to attend the
November Western Sectional Meeting at the University of
California, Riverside, will have the opportunity to hear
this story in person at Balmer’s Invited Address “An
invitation to tensor-triangular geometry.”
Tensor triangular geometry is a beautiful and powerful
theory. However, for tensor triangular categories of inter-
est it is desirable to have a concrete description of the
spectrum. Such a realization is both useful for applica-
tions and to connect it to existing theories. For example,
when he introduced tensor triangular geometry, Balmer
proved that if 𝒦 is the homotopy category of bounded
complexes of finitely generated projective 𝑅-modules,
then the spectrum and support recover Spec(𝑅) and its
support. He also showed that if 𝒦 is the stable module
category for 𝐺-mod when 𝐺 is a finite group, then the
spectrum and support match the long-studied spectrum
of the cohomology ring of 𝐺 and cohomological support
varieties for 𝐺-modules. In particular, this recovers the
classification of thick tensor ideals in this setting first
obtained by Benson-Carlson-Rickard twenty years ago.
This shows that tensor triangular geometry encompasses
known theories. Additional examples have since been
computed. Nevertheless, it remains a challenging prob-
lem to give an explicit realization of the spectrum and
support for tensor triangulated categories of interest.
In joint work with Brian Boe and Daniel Nakano, we
provide a description of the spectrum for several tensor
triangulated categories which appear in nature. We give
explicit, down-to-earth descriptions of the spectrum for
the stable category of finite-dimensional modules for
the complex Lie superalgebra 𝔤𝔩(𝑚|𝑛) and for the stable
category of finite-dimensional modules for quantized
enveloping algebras at a root of unity. The goal of the
talk will be to describe these results through a gentle
introduction involving plenty of examples.

Photo Credit
Photo of Jonathan Kujawa by Anne Dunn, courtesy
of Jonathan Kujawa.

September 2017 Notices of the AMS 821

 AMS FALL SECTION SAMPLER
Govind Menon
Building Polyhedra by Self-Assembly
Gromov begins an interesting—and speculative—recent
article [2] with the question, “Is there mathematics in
biology?” The answer, I think, is yes, but this is not
immediately apparent, since the real underlying question
is whether modern biology can inspire new forms of
mathematics in a way that compares to the deep ties
that bind mathematics and physics. If we believe that
an essential aspect of mathematics lies in the discovery
of abstract principles from empirical knowledge, there
is little doubt that biology today presents us with an
abundance of the “raw stuff.” What seems much harder
is to process this raw stuff into beautiful mathematics,
especially if one begins with the genetic code and the
theory of evolution.
The topic of my talk is not true biology, but an
instance of “synthetic biology.” All biological organisms
build themselves or “self-assemble.” This is, of course,
familiar to us from our everyday experience, but my
talk will be about much smaller organisms. For the
past twenty years, nanotechnologists have been trying Figure 1. A ribbon-diagram showing the structure
to manufacture devices by mimicking biological self- of the bacteriophage MS2. The coat protein exists
assembly and the exquisite design of molecular machines. in three distinct conformations (A, B, and C), which
The goal of my talk is to advertise one aspect of this rapidly merge in pairs into A/B dimers (blue/green) and C/C
growing field and to explain how an important biological dimers (maroon). A/B dimers cluster into pentamers
example—the self-assembly of viruses with icosahedral around the 5-fold axes of an icosahedron, three alter-
symmetry—can inspire and guide the development of nating A/B and C/C clusters form at the 3-fold axes,
self-assembly in technology. and the C/C dimers sit as axes of 2-fold symmetry.
Viruses are biological organisms that lack the cellu-
lar machinery necessary for independent existence. The
simplest viruses consist of genomes contained within a known since the mid-1970s, it is only recently that the
protein shield (the capsid). The capsid disassembles when intricate combinatorial structure of the co-assembly of
the virus attacks a host cell; the virus genome then hijacks the capsid with RNA folding was deciphered by Reidun
the host cell and uses it to make many more copies of Twarock and her colleagues [1].1
virus genome and proteins, which then rapidly reassemble The self-assembly of viruses has inspired many exam-
into new copies of the virus. The natural design of viruses ples of synthetic self-assembly. My work has mainly been
has two elegant features that should appeal to all math- in collaboration with David Gracias, an experimentalist
ematicians: genetic economy and structural symmetry. at Johns Hopkins University. Over the past fifteen years,
The genetic sequences of primitive viruses are very short. David has used photolithography to design many devices
For example, the genome of MS2, a well-studied virus, has and containers that fold themselves into a final shape
only 3,569 nucleotides that code for four proteins (lysis, once they are released from a substrate. The devices built
replicase, maturation, and coat protein), each of which in his lab are small (a hair’s width and smaller), but much
has a very specific function. The lysis enzyme degrades larger than viruses such as MS2. This allows us to observe
the cell wall of the host, and the replicase catalyzes the the pathways of self-folding, unlike the process of self-
reproduction of the virus. The other two proteins are used assembly of viruses, which must be inferred indirectly
to build the MS2 capsid: it consists of 180 copies of the (Figure 2).
coat protein, pinned at one end by the maturation protein, The unfolding of a polyhedron into a planar net is a
in a beautiful arrangement of dimers with icosahedral classical problem in discrete geometry, and our collabora-
symmetry (Figure 1). While the genome of MS2 has been tion began when David asked me what the best net should
be for a self-folding dodecahedron. The issue here is a
Govind Menon is professor of applied mathematics at Brown Uni- combinatorial explosion. The cube has only 11 nets, each
versity. His e-mail address is govind_menon@[Link].
For permission to reprint this article, please contact: 1
For more on connections between combinatorics and molecu-
reprint-permission@[Link]. lar biology, see “Strings, Trees, and RNA Folding” by Christine
DOI: [Link] Heitsch in this issue (page 817).

822 Notices of the AMS Volume 64, Number 8

 AMS FALL SECTION SAMPLER
dra by self-assembly: Theory and experiment, Artificial Life
(2014). MR 2774091
[4] S. Pandey, M. Ewing, A. Kunas, N. Nguyen, D. H. Gracias,
and G. Menon, Algorithmic design of self-folding polyhedra,
Proceedings of the National Academy of Sciences 108 (2011),
19885–19890.

Image Credits
Figure 1 ©Dr Neil Ranson, University of Leeds, UK. Used
under the Creative Commons Attribution-Share Alike 3.0
Unported License.
Figure 2 courtesy of Shivendra Pandey.
Photo of Govind Menon courtesy of Govind Menon.

Figure 2. Optical microscope images of surface-

tension-driven self-assembly of a dodecahedron from
a net. The sides of each face of the dodecahedron are
300 μm.

of which may be tested in the lab. However, the dodeca-

hedron has 43,380 nets, and, to my surprise and delight,
simple heuristics along with our computations revealed
the best nets in the lab [4]. Since then our work has evolved
into a study of the pathways of self-assembly [3]. This
has required some surprisingly sophisticated mathemat-
ics. My current goal is to understand the conformational
diffusion of polyhedral linkages. More formally, this in-
volves a rigorous formulation for Brownian motion on
algebraic varieties defined by polyhedral linkages, along
with effective algorithms for simulation.

References
[1] E. C. Dykeman, P. G. Stockley, and R. Twarock, Solving a
Levinthal’s paradox for virus assembly identifies a unique
antiviral strategy, Proceedings of the National Academy of
Sciences (2014).
[2] M. Gromov, Crystals, proteins, stability and isoperimetry,
Bull. Amer. Math. Soc. (N.S.) 48 (2011), 229–257.
[3] Ryan Kaplan, Joseph Klobušický, Shivendra Pandey,
David H. Gracias, and Govind Menon, Building polyhe-

September 2017 Notices of the AMS 823

 AMS FALL SECTION SAMPLER
Kevin M. Pilgrim

Semigroups of Branched Mapping Classes:

Dynamics and Geometry
A rational function of a single complex variable defines
τ −1 + τ τ
a continuous map of the Riemann sphere to itself. In the
early 1980s, W. Thurston gave a topological characteriza- 1 !
0 −1
"
tion of certain rational functions among the much larger 1 1
set of self-branched coverings of the sphere. Implicit in
his development are generalizations of mapping class Figure 1. Shown are two fundamental domains for
groups. These generalizations will be the focus of my the torus ℂ/⟨1, 𝜏⟩ where 𝜏 = 𝑒2𝜋𝑖/6 . The ℝ-linear map
talk. induced by 1 ↦ 𝜏 and 𝜏 ↦ −1 + 𝜏 is a rotation of
order 6 and descends to an isometry on the torus.
Mapping Class Groups
The simplest mapping class group is that of the torus.
The mapping class group of the torus 𝑇2 is the group Conjugacy
Mod(𝑇2 ) of orientation-preserving self-homeomorphisms Two elements 𝑓, 𝑔 ∈ Mod(𝑇2 ) are conjugate if 𝑔 = ℎ−1 𝑓ℎ
of the torus, where two such maps are identified if they for some ℎ ∈ Mod(𝑇2 ). When does this happen? Thinking
are isotopic, i.e. connected by a continuous path of dynamically, two maps 𝑓, 𝑔 are conjugate if they coincide
homeomorphisms. after a change of coordinates via an element ℎ of Mod(𝑇2 ).
Matrices provide lots of examples. Let 𝑇2 = ℝ2 /ℤ2 be Properties like being periodic, reducible, or irreducible are
the usual presentation of the torus as a quotient of the thus invariant under conjugacy. The conjugacy problem
plane. Suppose 𝐴 = [ 𝑎𝑐 𝑑𝑏 ]. The linear map ℝ2 → ℝ2 given asks, given 𝑓, 𝑔, can you tell if 𝑓 and 𝑔 are conjugate? And
𝑥 if the answer is “yes,” can you produce such an element
by [ 𝑦𝑥 ] ↦ 𝐴[ 𝑦 ] descends to a homeomorphism 𝑓 ∶ 𝑇2 → 𝑇2
that preserves orientation if and only if it sends ℤ2 onto ℎ? For Mod(𝑇2 ) the answer is “yes”; the classification of
itself and det(𝐴) > 0. Equivalently, 𝑎, 𝑏, 𝑐, 𝑑 ∈ ℤ and conjugacy classes goes back to Gauss.
det(𝐴) = 1, i.e. 𝐴 ∈ SL2 (ℤ). For example: Geometrization
• 𝐴 = [ 01 −11 ]. A calculation shows 𝐴 has order 6. There is another way to think about the classification
The map 𝑓 is periodic. of conjugacy classes, via geometrization. Roughly, ge-
• 𝐴 = [ 10 11 ]. Since 𝐴𝑛 = [ 10 𝑛1 ], 𝑓 has infinite order. ometrization is the problem of finding a “nice” or “optimal”
The 𝑥-axis is an eigenspace of 𝐴. Its image under geometric structure for some topological object. Think
projection to 𝑇2 gives a simple closed curve which about the (surface of the) bagel you ate this morning.
is preserved by 𝑓. Infinite order maps for which It inherits a metric from usual Euclidean 3-space. That
some iterate preserves a curve are called aperiodic metric has variable curvature. In contrast, the Euclidean
reducible. metric on the plane ℝ2 descends to a metric on 𝑇2 which is
• 𝐴 = [ 21 11 ]. Again 𝐴 has infinite order. The flat: the curvature is zero everywhere. The map 𝑓 induced
eigenspaces have irrational slopes and project by 𝐴 = [ 01 −1 1 ] seems simple enough, but it distorts this
to dense subsets of 𝑇2 . If we use suitable local Euclidean metric! There are, however, lots of flat metrics
coordinates (𝑢, 𝑣) given by the eigenspaces, 𝑓 on 𝑇2 . They can be fruitfully organized by points in the
looks like (𝑢, 𝑣) ↦ (𝜆𝑢, 𝜆−1 𝑣) where 𝜆 = 3+√5 2 is upper-half-plane ℍ ⊂ ℂ. Given 𝜏 ∈ ℍ, we can identify ℝ2
the larger eigenvalue. No iterate of 𝑓 fixes a curve. as a real vector space with ℂ via (1, 0) ↦ 1 and (0, 1) ↦ 𝜏,
Such 𝑓 are called irreducible. and transport the Euclidean metric on ℂ to one on ℝ2
It turns out that every element of Mod(𝑇2 ) arises in via this identification. If we take the Euclidean metric
this way: the mapping class group Mod(𝑇2 ) is naturally on ℂ and take 𝜏 = exp(2𝜋𝑖/6), 𝑓 becomes an isometric
isomorphic to SL2 (ℤ). Put another way, mapping classes rotation of order 6; see Figure 1.
of the torus are determined by their induced action on In general, if 𝑓 ∈ Mod(𝑇2 ) and if we equip 𝑇2 with
the fundamental group ℤ2 of 𝑇2 . a Euclidean metric corresponding to a point 𝜏 ∈ ℍ in
this way, then we can transport this metric via 𝑓 to
Kevin Pilgrim is professor of mathematics at Indiana University. get another metric on 𝑇2 . This leads to an action of
His research is partially supported by the Simons Foundation. His Mod(𝑇2 ) on the space of metrics ℍ. The action is given
e-mail address is pilgrim@[Link]. by [ 𝑎𝑐 𝑑𝑏 ].𝜏 =↦ 𝑎𝜏+𝑏
𝑐𝜏+𝑑
.
For permission to reprint this article, please contact: Given 𝜏1 , 𝜏2 ∈ ℍ, there is a natural notion of distance
reprint-permission@[Link]. 𝑑(𝜏1 , 𝜏2 ) between the corresponding flat metrics on 𝑇2
DOI: [Link] that turns out to be the hyperbolic metric on ℍ; its

824 Notices of the AMS Volume 64, Number 8

 AMS FALL SECTION SAMPLER
2|𝑑𝜏|
infinitesimal form is given by 𝑑𝑠 = ℑ(𝜏)
. This yields that if |𝑧| > max{2, |𝑐|}, then this sequence converges
2
a natural invariant of 𝑓 ∈ Mod(𝑇 ) via its minimum to infinity. On the other hand, the quadratic formula
displacement inf{𝑑(𝜏, 𝐴.𝜏) ∶ 𝜏 ∈ ℍ}. This can be (1) zero shows that there are fixed points. So there is a nonempty
and realized by some fixed-point 𝜏, (2) zero and not compact subset 𝐾𝑐 consisting of points 𝑧 for which the
realized by any point, or (3) positive and realized by orbit of 𝑧 is bounded. The boundary 𝐽𝑐 of this locus is
points 𝜏 lying along some geodesic that is preserved the typically fractal Julia set and is the locus of chaotic
by the action of 𝑓. These correspond to the three cases behavior. We have 𝑧 ∈ 𝐽𝑐 if and only if the orbit of 𝑧 is
above. The first and third cases are called geometrizable: bounded, but there are arbitrarily small perturbations of
an optimal metric exists. For reducible cases, we can get 𝑧 for which the orbit becomes unbounded.
close to optimal, but we cannot achieve it. Looking at the polynomials 𝑝𝑐∘𝑛 (0) − 𝑝𝑐∘𝑚 (0), 𝑛, 𝑚 =
0, 1, 2, …, we find there is a countably infinite set of
General Mapping Class Groups
parameter values 𝑐 for which the forward orbit of the
Now suppose 𝑆 is an arbitrary closed orientable surface branch point at the origin is finite. These critically finite
and 𝑃 ⊂ 𝑆 is finite. The mapping class group Mod(𝑆, 𝑃) is parameters play a crucial role in complex dynamics
the group of orientation-preserving homeomorphisms 𝑓 ∶ and provide a rich source of examples of semigroups
𝑆 → 𝑆 for which 𝑓(𝑃) = 𝑃, where again two are identified BrMod(𝑆2 , 𝑃).
if they are isotopic. Roughly summarizing thousands of
An Example
pages of hard work by many mathematicians: the flavor
of the results for the torus outlined above extend to Let 𝑔(𝑧) = 𝑧2 +𝑖. There are branch points at the origin and
arbitrary mapping class groups. Elements of Mod(𝑆, 𝑃) are the point at infinity where 𝑔 is locally 2-to-1. The point at
determined by their action on the fundamental group. A infinity ∞ is fixed by 𝑔, and 0 ↦ 𝑖 ↦ 𝑖 − 1 ↦ −𝑖 ↦ 𝑖 − 1.
geometrizable element is either periodic and represented Let’s take 𝑃 ∶= {𝑖, 𝑖 − 1, −𝑖, ∞}; note that 𝑔(𝑃) ⊂ 𝑃. We
by an isometry or irreducible and represented by a map have just constructed an element of BrMod(𝑆2 , 𝑃).
which in (mildly singular local) Euclidean coordinates How might we get other elements? We can do so by
looks again like (𝑥, 𝑦) ↦ (𝜆𝑥, 𝜆−1 𝑦). A class that is not twisting via pre- and post-composition with elements of
geometrizable reduces canonically into geometrizable Mod(𝑆2 , 𝑃). Let’s look at all such elements:
pieces by cutting along some finite invariant collection ℋ𝑔 ∶= {[ℎ0 ∘𝑔∘ℎ1 ] ∶ ℎ0 , ℎ1 ∈ Mod(𝑆2 , 𝑃)} ⊂ BrMod(𝑆2 , 𝑃).
of pairwise disjoint curves. The conjugacy problem is
significantly harder but solvable. This set decomposes into conjugacy classes. How many
are there? As shown in a groundbreaking paper by L.
Branched Mappings Bartholdi and V. Nekrashevych, the set of conjugacy
We now drop the assumption that 𝑓 is a homeomorphism classes in ℋ𝑔 consists of two geometrizable elements
and require only that it is a finite branched covering classified by polynomials 𝑧2 ± 𝑖 and a ℤ’s worth of classes
unramified away from 𝑃. Equivalently, 𝑓 ∶ 𝑆 − 𝑓−1 (𝑃) → represented by reducible elements that fix a curve in a
𝑆 − 𝑃 is a covering map of some degree 𝑑 ≥ 1; we require way that provides an obstruction to geometrization.
still that 𝑓(𝑃) ⊂ 𝑃. The Riemann-Hurwitz formula implies What if we use a different polynomial? Now let 𝑓(𝑧) =
that if 𝑑 ≥ 2, then the surface 𝑆 is either the torus 𝑇2 𝑧2 + 𝑐 where 𝑐 is the unique parameter for which ℑ(𝑐) > 0
or the sphere 𝑆2 and that if 𝑆 is the torus, the map 𝑓 is and the origin is periodic of least period 3; now put
unramified. Composition descends to a well-defined map 𝑃 = {0, 𝑐, 𝑐2 + 𝑐, ∞}. The polynomial 𝑓 is called the
on isotopy classes, and we obtain a countable semigroup Douady Rabbit polynomial, since its Julia set (shown in
BrMod(𝑆, 𝑃). the left-hand side of Figure 4) looks a bit like a rabbit.
Again the case of the torus is instructive. The conjugacy Now the corresponding set of conjugacy classes consists
problem in this semigroup boils down to the question: of just three geometrizable elements, classified by the
Given a pair of 2 by 2 integral matrices 𝐴, 𝐵 of common three values of 𝑐 for which the origin has period 3 under
determinant larger than one, when is 𝐵 = 𝑃−1 𝐴𝑃 for iteration of 𝑧2 + 𝑐.
some 𝑃 ∈ SL2 (ℤ)? This question is equivalent to the G. Kelsey and R. Lodge have just announced an
classification of real quadratic fields and is still unsolved. extension of these results to all quadratics with #𝑃 = 4.
So I’ll focus on the case 𝑆 = 𝑆2 . It turns out the case
#𝑃 = 4 is already interesting. Highlights
There are several results about the semigroup
Complex Dynamics
BrMod(𝑆2 , 𝑃) that mirror results for the mapping
Complex dynamics provides a natural source of geometriz- class group Mod(𝑆, 𝑃).
able elements of BrMod(𝑆2 , 𝑃). Let’s identify the Riemann A first point—one exploited by L. Bartholdi and V.
sphere ℂ ̂ with 𝑆2 in the usual way via stereographic Nekrashevych—is that branched mapping classes are
projection. For a parameter 𝑐 ∈ ℂ let 𝑝𝑐 ∶ ℂ̂ →ℂ ̂ be given faithfully encoded by algebraic data known as wreath
by 𝑝𝑐 (𝑧) = 𝑧2 + 𝑐. Dynamics is concerned with iteration. recursions; like induced maps on fundamental groups,
We study sequences like {𝑧, 𝑝𝑐 (𝑧), 𝑝𝑐 (𝑝𝑐 (𝑧)), …}, called they are well defined up to conjugacy. As an illustration,
the orbit of 𝑧. On the one hand, an easy exercise shows equation (1) gives wreath recursion data for the Rabbit

September 2017 Notices of the AMS 825

 AMS FALL SECTION SAMPLER

Figure 2. A computer program reads the group- Figure 3. The limit set of the subsemigroup of
theoretic data faithfully encoding Douady’s Rabbit. BrMod(𝑆2 , 𝑃) generated by Mod(𝑆2 , 𝑃) and the
It produces a numerical approximation for the poly- Douady Rabbit polynomial 𝑓(𝑧) = 𝑧2 + 𝑐, drawn in
nomial map. It then draws an approximation to its the disk model.
fractal Julia set on the sphere.

BrMod(𝑆2 , 𝑃) generated by the Rabbit polynomial 𝑓 and

the group Mod(𝑆2 , 𝑃).
polynomial 𝑓(𝑧); here 𝑎, 𝑏, 𝑐 are free generators for the
Finally, the subject is immensely rich. In joint work with
fundamental group of 𝑆2 − 𝑃. The upshot is that we can
W. Floyd, G. Kelsey, S. Koch, R. Lodge, W. Parry, and E. Saenz,
represent and compute with elements of BrMod(𝑆2 , 𝑃). A
we systematically study elements of BrMod(𝑆2 , 𝑃), #𝑃 = 4,
program by L. Bartholdi takes this algebraic data as input for which the branch points are simple. Such maps turn
and returns a numerical approximation for the rational out to be closely related to affine torus maps, and
geometrization (if it exists) and an approximation of its this observation leads to them being computationally
associated Julia set; see Figure 2: tractable. A database of tens of thousands of examples
𝑎 = ⟨𝑎−1 𝑏−1 , 𝑐𝑏𝑎⟩(12), is now online at [Link]/netmaps/[Link].
(1) 𝑏 = ⟨𝑎, 1⟩, Figure 4 illustrates the circle of ideas. The first two figures
𝑐 = ⟨𝑏, 1⟩. (top left) show respectively the fractal Julia set for the

A second point is that our understanding of the 1.5

(2, 1)
corresponding geometrization questions is advancing 1.0
0.5
rapidly. In our setting, typical geometrizable elements of (0, 0)

degree larger than one are represented by maps which are -1.5 -1.0 -0.5 0.5 1.0 1.5
-0.5 (0, -1)
expanding in a suitable sense. Critically finite polynomials
-1.0
and rational maps in this sense are expanding. However, -2
-2 -1.5 -1 -.5 0 .5
-1.5
there do exist nonrational expanding maps. L. Bartholdi
and D. Dudko give a characterization of expanding classes. α2

Recently N. Selinger and M. Yampolsky and independently α1 α3 σf α3 α2

α1
L. Bartholdi and D. Dudko have shown that a general map
decomposes algorithmically into geometrizable pieces.
-2 -1 0 −1 −1– 0
A third point is that the semigroup BrMod(𝑆2 , 𝑃) acts 2

naturally on the space of conformal structures on (𝑆2 , 𝑃). Figure 4. Follow the Rabbit from its fractal Julia set
When #𝑃 = 4 this is again the hyperbolic plane ℍ. Figure 3 to its induced action on the upper-half-plane.
shows the limit set of the action of the subsemigroup of

826 Notices of the AMS Volume 64, Number 8

 AMS FALL SECTION SAMPLER
Rabbit polynomial 𝑓 and a diagram encoding the relation
between 𝑓 and an affine torus map. The (right top) figure
deals with the geometry of the induced map 𝜎𝑓 ∶ ℍ → ℍ:
there is a unique fixed point in the white region. The final
pair (bottom) illustrates that 𝜎𝑓 maps an ideal triangle to
a Schwarz triangle and is extended by reflection.
Image Credits
Figure 1 courtesy of Kevin M. Pilgrim.
Figure 2 courtesy of L. Bartholdi.
Figure 3 courtesy of S. Koch.
Figure 4 courtesy of W. Floyd et al.
Photo of Kevin M. Pilgrim courtesy of Indiana University
College of Arts and Sciences.

September 2017 Notices of the AMS 827

 AMS FALL SECTION SAMPLER
Bruce Sagan it was finally proved by Wolfgang Haken and Kenneth
Appel in 1976. Their proof caused quite a stir in the
The Protean Chromatic Polynomial mathematical community, because it was the first to use
a substantial amount of computing time, and the large
I am very excited to have the opportunity to share some
number of cases could not all be checked by hand.
of the ideas surrounding one of my favorite objects
The chromatic polynomial was introduced in 1912 by
in combinatorics, the chromatic polynomial, during my
George Birkhoff as a possible tool for proving the then
Invited Address. Let me start by defining each of the
Four Color Conjecture. Although it did not turn out to be
terms in my title.
useful for the eventual proof, it has more than justified
The Merriam-Webster Dictionary defines “protean” as
its existence through its many other applications. Let 𝑡 be
“of or resembling Proteus in having a varied nature or
a nonnegative integer. The chromatic polynomial, 𝑃(𝐺; 𝑡),
ability to assume different forms.” In Greek mythology,
is the number of proper colorings 𝑐 ∶ 𝑉 → {1, 2 … , 𝑡}. It
Proteus was one of the gods of the sea and thus was
is not apparent at first blush why this cardinality should
associated with its constantly changing nature. In a similar
be called a polynomial. However, this will become clearer
manner, the chromatic polynomial gives one information
if we compute 𝑃(𝐺; 𝑡) for the graph in Figure 1. Suppose
about many things which, a priori, have nothing to do
we color the vertices in the order 𝑢, 𝑣, 𝑤, 𝑥. Then there
with its original purpose, as described below.
are 𝑡 choices for the color of 𝑢 since it is the first vertex
𝑢 𝑣 𝑢 𝑣 to be colored. After that, there will be 𝑡 − 1 choices for
the color of 𝑣, since it cannot be the same color as 𝑢.
Similar reasoning shows that there are 𝑡 − 1 choices for
𝐺= 𝐺=
𝑤. Finally, 𝑥 is adjacent to both 𝑢 and 𝑣, and these two
vertices have different colors, so we can color 𝑥 in 𝑡 − 2
𝑥 𝑤 𝑥 𝑤 ways. The net result is that
𝑢 𝑣 𝑃(𝐺; 𝑡) = 𝑡(𝑡 − 1)2 (𝑡 − 2) = 𝑡4 − 4𝑡3 + 5𝑡2 − 2𝑡,
which is a polynomial in 𝑡, the number of colors!
𝐺= One can show that 𝑃(𝐺; 𝑡) is always a polynomial in 𝑡
and give nice characterizations of its degree, coefficients,
and other properties. Furthermore, it has connections with
𝑥 𝑤
many other objects of study, including acyclic orientations
Figure 1. A graph and two colorings of graphs, hyperplane arrangements, and even Chern
classes in algebraic geometry. I will explain these during
“Chromatic” refers to color, and our general topic is my lecture, as well as present some recent work with
the coloring of the vertices of a graph. A (combinatorial) Joshua Hallam and Jeremy Martin relating 𝑃(𝐺; 𝑡) to yet
graph, 𝐺, consists of a set of vertices 𝑉 and a set of edges another graphical concept, increasing spanning forests. If
𝐸 which connect pairs of vertices. For example, the graph you are at the Buffalo AMS Sectional Meeting in September,
in the upper left in Figure 1 has vertex set 𝑉 = {𝑢, 𝑣, 𝑤, 𝑥} I hope to see you at my talk.
and edge set 𝐸 = {𝑢𝑣, 𝑢𝑥, 𝑣𝑥, 𝑣𝑤}. A coloring of 𝐺 is a
function 𝑐 ∶ 𝑉 → 𝑆 where 𝑆 is called the color set. The Image Credits
coloring is proper if the endpoints of every edge have
different colors. The coloring in the upper right of Figure 1 Figure 1 courtesy of Bruce Sagan.
is proper, while the one on the bottom is not because the Photo of Bruce Sagan by Robert Chandler, courtesy
edge 𝑒 = 𝑣𝑤 has the same color on both endpoints. The of Bruce Sagan.
chromatic number, 𝜒(𝐺), is the smallest number of colors
needed to properly color 𝐺. The graph in Figure 1 has
𝜒(𝐺) = 3 since the upper right image exhibits a proper
coloring with three colors, and the triangle 𝑢𝑣𝑥 cannot
be colored with fewer colors. Maybe the most famous
theorem in graph theory is the Four Color Theorem,
which states that if a graph is planar (can be drawn in
the plane without edge crossings), then 𝜒(𝐺) ≤ 4. This
statement was a conjecture for over a hundred years until

Bruce Sagan is professor of mathematics at Michigan State Uni-

versity. His e-mail address is sagan@[Link].
For permission to reprint this article, please contact:
reprint-permission@[Link].
DOI: [Link]

828 Notices of the AMS Volume 64, Number 8