Graham Higman’s lectures on januarials

Marston Conder∗
Department of Mathematics, University of Auckland,
Private Bag 92019, Auckland 1142, New Zealand

Timothy Riley†

Department of Mathematics, 310 Malott Hall,

Cornell University, Ithaca, NY 14853, USA

Abstract

This is an account of a series of lectures of Graham Higman on januarials, namely

coset graphs for actions of triangle groups which become 2-face maps when em-
bedded in orientable surfaces.

Spilt Milk The Nineteenth Century and After

We that have done and thought, Though the great song return no more
That have thought and done, There’s keen delight in what we have:
Must ramble, and thin out The rattle of pebbles on the shore
Like milk spilt on a stone. Under the receding wave.

from The Winding Stair and Other Poems, W.B.Yeats, 1933

1 Preamble

Graham Higman gave the lectures on which this article is based, in Oxford in 2001.
They are likely to have been the final lectures he gave; he died in April 2008, at the
age of 91. He introduced them with the quotes from W.B. Yeats reproduced above, and
∗ [email protected], http://www.math.auckland.ac.nz/∼conder/
† [email protected], http://www.math.cornell.edu/∼riley/

1
described the work in preparing them as “justifying my old age” and “keeping me rel-
atively sane.” The second author attended the lectures, and the first author remembers
Higman’s work on related topics some years earlier; this account is developed from
recollections and from notes taken at the time. As such, any errors are ours, and the
presentation and the proofs offered may not be as Higman had in mind. At various
points, and as indicated, we have extended Higman’s treatment; we also include some
of our own observations in an afterword.
Januarials, which we will define in Section 2, are 2-complexes with two distinguished
faces, that result from embedding coset graphs for the actions of triangle groups into
compact orientable surfaces. They can be viewed as being assembled from two sub-
surfaces (essentially those two distinguished faces); we give appropriate definitions
and tools to explore the complexity of this assembly in Section 3. In Section 4 we give
sufficient conditions for actions of PSL(2, p) on projective lines to give rise to januari-
als. This leads to a number of examples presented in Section 5. Finally Section 6, our
afterword, contains some remarks on the coset graph appearing in Norman Blamey’s
1984 portrait of Higman, and some further examples of januarials.
It appears that Higman’s study of januarials was sparked by his work on Hurwitz
groups, which are non-trivial finite quotients of the (2, 3, 7)-triangle group. Higman
used coset diagrams to show that for all sufficiently large n, the alternating group Alt(n)
is a Hurwitz group, and his work was taken further by the first author to determine ex-
actly which Alt(n) are Hurwitz, in [1].

Funding. This work was partially supported by the New Zealand Marsden Fund [grant
UOA1015] to MDEC; and the United States National Sciences Foundation [grant DMS–
1101651] to TRR; and the Simons Foundation [collaboration grant 208567] to TRR.

Acknowledgements. We are grateful to Martin Bridson, Harald Helfgott and Peter Neu-
mann for encouragement and guidance in the writing of this account, and to Stephen
Blamey and Sam Howison (Chairman of the Oxford Mathematical Institute) for per-
mission to reproduce Norman Blamey’s portrait of Graham Higman in Figure 11. We
also thank the referee for carefully reading this article and making some very helpful
suggestions.

2 Coset graphs, face maps, januarials, and surfaces

Suppose S is a set endowed with an action s 7→ sg by a group G, and A is a generating

set for G. Define Γ to be the graph with vertex set S , and with an oriented edge labelled
by a ∈ A (called an a-edge) from vertex u to vertex v whenever ua = v. We will
be concerned with situations where G acts transitively on S , so that Γ is connected.
In that event we can identify S with the right cosets { Hg : g ∈ G } of the stabiliser
H = G s = StabG (s) of any particular s ∈ S , and for this reason, Γ is known as a coset
graph or Schreier graph, or sometimes coset diagram, for the action of G on S with

2
respect to A. When the action of G on S is also regular, we can identify S with the
underlying set of G, in which case Γ is the Cayley graph of G with respect to A.
Paths in the coset graph may be labelled with words on the generating set A (which can
be thought of as an alphabet). Suppose that a word w on A±1 represents g ∈ G, and
that s ∈ S . Let γ be the path in Γ obtained by concatenating the unique edge-paths in Γ
i i+1
from sg to sg , for each i ∈ Z, along which one reads w. This tours an orbit of hgi and
is a (closed) circuit precisely when that orbit is finite. There is one such path for each
orbit.

∞ 0

12

2 7

11 6
9 4 10 3

8 5

Figure 1: A coset graph arising from an action of the group PSL(2, 13) on F13 ∪ {∞},
with x : z 7→ −z and y : z 7→ (z − 1)/z, and its companion graph. This results in a
3-januarial of genus 0 and simple type (1, 0, 0).

∞ 0

16

2 9

15 8
6 3

10 12

7 5
14 11

13 4

Figure 2: A coset graph arising from an action of the group PSL(2, 17) on F17 ∪ {∞},
with x : z 7→ −z and y : z 7→ (z − 1)/z, and its companion graph. This results in a
3-januarial of genus 1 and simple type (1, 1, 0).

A map is a 2-cell embedding of a connected (multi)graph in some closed surface, with

its faces (the components of the complement of the graph in the surface) being home-
omorphic to open disks in R2 . Examples include triangulations and quadrangulations
of the torus, and the Platonic solids (which may be viewed as highly symmetric maps

3
on the sphere), with all vertices having the same valence and all faces having the same
size.
A januarial is a special instance of a map constructed from embedding a coset graph
for an action of the the (2, k, l) triangle group

∆(2, k, l) = h x, y | x2 = yk = (xy)l = 1 i,

with A = {x, y}. Because x2 = 1, the x-edges in such a coset graph Γ coming from
non-trivial cycles of x occur in pairs: whenever there is an x-edge from u to v, there is
an x-edge from v to u. We may identify each such pair, so as to leave an unoriented
x-edge between u and v. Then for each fixed point s of x, we attach a 2-cell (which we
will call an x-monogon) along the x-edge which forms a loop at s. Similarly, for each
orbit of hyi, we attach a polygon (which we call a y-face) along the path γ given by hyi
as described above. This gives a 2-complex, many examples of which appear in this
article; see Figures 1, 2, 6, 7, 8, 9, and 10. These and similar figures can be displayed
without labels on the edges, because we may shade the y-faces so that y-edges are
identifiable as those in the boundaries of y-faces, while all the remaining edges are x-
edges. We need not indicate orientations on the edges: the x-edges for the reason given
above, and the y-edges because we may adopt a convention that all y-edges are oriented
anti-clockwise around the corresponding y-faces. Note that the length (the number of
sides) of each y-face divides k.
Next, attach a polygon (which we call an xy-face) around each orbit of hxyi. As shown
by the following lemma, the resulting 2-complex J is homeomorphic to a closed ori-
entable surface. We may call the corresponding embedding of Γ an m-face map, where
m is the number of orbits of hxyi.
A januarial (and more precisely, a k-januarial) is the instance when m = 2 and the
orbits of hxyi have the same size | S | /2. Two examples of 3-januarials are given in
Figure 3.
Lemma 2.1. The 2-complex J defined above is homeomorphic to a compact orientable
surface without boundary.

Proof. In the construction of J we identified oppositely oriented x-edges in pairs. For

this proof, however, it is convenient to revert to the pairs of oriented x-edges, and insert
a digon (which we call an x-digon) between each pair. We will show that the resulting
complex gives an orientable surface without boundary. It will follow that the same is
true of a januarial, because we have an embedding in the same surface when the digons
(any two of which have no x-edge in common) are successively collapsed to single
edges.
Now in this complex, each vertex has valence four: it has both an incoming and an
outgoing x-edge (coming from an edge-loop in the event that x fixes the vertex), and
both an incoming and an outgoing y-edge (which, similarly, may come from a loop).
Each x-edge is incident with exactly one x-face (that is, an x-monogon or an x-digon),
and one xy-face. Each y-edge is incident with exactly one y-face and one xy-face. It
follows that the complex gives a surface without boundary. Moreover, the surface is

4
 0
6
∞

1 12 7
3 14 6
5
10
2 10 2
15
13 7
∞
1 9
4 16

11 9 4 0 8
5 12

8 3
11

Figure 3: The 3-januarials arising from the coset graphs in Figures 1 and 2. The unori-
ented edges are x-edges and oriented edges are y-edges.

orientable because the directions of the edges give consistent orientations around the
perimeters of all the faces. Finally, since S is finite, the surface is compact. 

3 The topological complexity of januarials

Higman gave a notion of topological complexity which we call simple type below. It
concerns how a januarial J is assembled from the subspaces S 1 and S 2 that are the
closures of its two xy-faces. He recognised that some januarials are beyond the scope
of this notion; indeed, he made some ad hoc calculations for the examples in Figures 8
and 10 which show as much. Accordingly, below, we define a more general notion
which we call type, which applies to all januarials, and we explain how to calculate it
in general.
Topological features of J, S 1 and S 2 come into clearer focus when we collapse each
x-monogon and each y-face in J to a point. Any two y-faces in a januarial are disjoint.
The same is true of any two x-monogons. And an x-monogon can only meet a y-face
at a single vertex. So these collapses do not change the homeomorphism types of J, S 1
or S 2 .
Let Γ be the 1-skeleton of J — that is, the coset graph. Let J, S 1 , S 2 and Γ be the
images of J, S 1 , S 2 and Γ under these collapses. We call Γ a companion graph. Then
J = S 1 ∪ S 2 is a closed surface obtained by some identification of S 1 and S 2 along
their boundaries. Taking another perspective, J is the result of adding two faces to Γ,
via attaching maps ρ1 and ρ2 induced by the maps that attach the xy-faces to Γ.
Examples of such Γ and Γ appear in Figures 1, 2, 7, 8, 10, 12, and 13. Each one is
drawn in such a way that the cyclic order in which edges emanate from vertices agrees
with that in which x-edges meet y-faces in Γ. So, as the y-edges in Γ are oriented

5
anti-clockwise around the y-faces, one can read off ρi by following successive edges in
Γ in such a way that on arriving at a vertex, one exits along the right-most of all the
remaining edges (except if the vertex has valence one, in which case one exits along
the edge by which one arrived).
As hxyi yields exactly two orbits when acting on S , together ρ1 and ρ2 traverse each
edge in Γ twice, once in each direction. The edges comprising the subgraph G :=
S 1 ∩ S 2 , shown in blue in the figures, are traversed by ρ1 in one direction and ρ2 in the
other. Those traversed by ρ1 (resp. ρ2 ) in both directions are shown in red (resp. green).
The collapses carrying J to J leave only the two xy-faces, those x-edges which are
not loops, and one vertex for each y-face in J. These collapses do not alter the Euler
characteristic. Since J is a closed orientable surface, we find that the genus of J is
readily calculated as follows.
Lemma 3.1. Twice the genus of a januarial equals the number of x-edges which are
not loops minus the number of y-faces.

For example, this is 6 − 6 = 0 in the left-hand example of Figure 3 and is 8 − 6 = 2 in

the right-hand example, giving genera 0 and 1, respectively.
Now we turn to genera associated to S 1 and S 2 , or their images S 1 and S 2 . Defining
these requires care, since S 1 and S 2 may fail to be sub-surfaces of J (and likewise S 1
and S 2 fail to be sub-surfaces of J): they are closed surfaces from which the interiors of
some collection of discs have been removed, but the boundaries of those discs need not
be disjoint. (Figures 8 and 10 provide such examples.) But if we take a small closed
neighbourhood Ri of S i in J, we get a genuine sub-surface which serves as a suitable
proxy:
Lemma 3.2. R1 and R2 are orientable surfaces, and they retract to S 1 and S 2 , respec-
tively.

Proof. A small closed neighbourhood of G (or indeed of any subgraph of the 1-skeleton
of a finite cellulation of a closed surface) is a sub-surface with boundary and retracts
to G. Similarly Ri , which is the union of S i with a small closed neighbourhood of G, is
orientable and retracts to S i . It is orientable because J is orientable. 

We define the type of J to be the pair ((h1 , g1 ), (h2 , g2 )), where gi and hi are the genus
of Ri and the number of connected components of the boundary of Ri respectively, for
i = 1, 2. We will not distinguish between types ((h1 , g1 ), (h2 , g2 )) and ((h2 , g2 ), (h1 , g1 )).
The most straightforward way in which R1 and R2 can be assembled to make J occurs
when R1 ∩ R2 is a disjoint union of h annuli, where h = h1 = h2 , or in other words,
when J is homeomorphic to a join of R1 and R2 in which the boundary components
are paired off and identified. In this case, we say that the januarial J is of simple type
(h, g1, g2 ). We do not distinguish between the simple types (h, g1, g2 ) and (h, g2 , g1 ).
Maps in which the graph is embedded in a suitably non-pathological manner (for in-
stance as a subgraph of the 1-skeleton of a finite cellulation of the surface) have the

6
property that a small neighbourhood is a disjoint union of annuli if and only if the
graph is a collection of disjoint simple circuits. So, as R1 ∩ R2 is a small neighbour-
hood of G, one can recognise simple type from the graph Γ :
Lemma 3.3. J is of simple type if and only if G is a collection of disjoint simple circuits.
In that case, if J has simple type (h, g1 , g2 ) then h is the number of circuits.

The genus of a januarial J (equivalently, of J) of simple type is present in the data

(h, g1, g2 ). When the handles (that is, the h annuli from R1 ∩ R2 ) that connect R1 and R2
are severed one-by-one, the genus falls by 1 each time, until we only have one handle
connecting R1 and R2 , and hence a surface of genus g1 + g2 . Since J has h handles to
begin with, this gives the following:
Lemma 3.4. The genus of a januarial J of simple type is g1 + g2 + h − 1.

Figures 1, 2, 7 and 12 show examples of coset graphs which give januarials of simple
type, and Figures 8, 10 and 13 show examples which give januarials that are not of
simple type. In each case, the caption indicates the genus of the januarial and the
details of the type. The genus of the januarial can be established in each case via an
Euler characteristic calculation (as per Lemma 3.4 for those of simple type).
For the examples of simple type, h is immediately evident from the companion graph Γ
on account of Lemma 3.3. For those not of simple type, our next lemma gives a means
of identifying h1 and h2 from Γ. Examples of partitions of G into circuits in the sense
of this lemma can be seen in Figures 8, 10 and 13.
Lemma 3.5. Let P be the set of all paths that traverse successive edges in G in the
directions they are traversed by ρ1 (resp. ρ2 ), in such a way that whenever such a path
reaches a vertex, it continues along the right-most of the other edges in G incident
with that vertex. (The next edge is necessarily traversed by ρ1 (resp. ρ2 ) in that direc-
tion.) All such paths close up into circuits, and P partitions G, in the sense that the
union of the circuits is G and no two share an edge. The cardinality of P is h2 (resp. h1 ).

Proof. We will prove the result for ρ1 . The same argument holds for ρ2 with the sub-
scripts 1 and 2 interchanged.
By construction, the portion of the circuit ρ1 that falls in G runs close alongside the
boundaries of the h2 holes in R2 . Consider the situation where ρ1 is traversing an edge
e in G, and let B denote the boundary of the hole that runs alongside — see Figure 4.
At the terminal vertex v of e, because of our convention for drawing companion graphs,
ρ1 will continue along the right-most of the other incident edges in Γ. If that edge e′
is in G, it also runs alongside B. (This happens at u in the figure.) Suppose, on the
other hand, that e′ is not in G. Then ρ1 does not run alongside B, but rather heads into
the interior of R2 . (This happens at v in the figure.) At some later time, ρ1 must return
along e′ in the opposite direction (perhaps visiting another portion of B in the interim)
since the edges ρ1 traverses exactly once are precisely those in G. Hence ρ1 arrives
back at v and then continues along the (new) right-most edge — which will either be
alongside B, or take it back into the interior of R2 , again to return eventually along that

7
same edge. Repeating this, we eventually find the next edge in G incident with v that
continues alongside B. It follows that however many detours into the interior of R2
are required, it is the right-most of the edges in G incident with v (aside from e) that
continues alongside B. So the circuits traversed as explained in the statement of the
lemma are those that run alongside the boundaries of the holes in R2 . The remaining
claims easily follow from this. 

B
v S1
u

Figure 4: Tracking the boundary of one of the holes in R2 .

Given hi (for i = 1 or 2), one can determine gi from Γ via the following observation:
Lemma 3.6. The genus gi of Ri satisfies 2 − 2gi = Vi − Ei + hi + 1 where Vi and Ei
denote the number of vertices and edges, respectively, in the subgraph of Γ visited by
the attaching map of the face of S i .

Proof. By Lemma 3.2, filling the hi holes in Ri with discs gives a closed orientable
surface of genus gi which is homotopic to S i with hi discs attached along circuits in its
1-skeleton. Hence the Euler characteristic 2 − 2gi of Ri is the same as that of S i with
the hi discs attached, namely Vi − Ei + hi + 1. 
Questions 3.7. Higman asked the following questions concerning k-januarials of sim-
ple type. For a given k, what are the possible values for and interrelationships between
g1 , g2 and h? Are there arbitrarily large values of k for which there exist examples with
h = 1? How large can h be, for given k? Similar questions can be asked also about
januarials that are not of simple type.

4 Januarials from PSL(2, q)

4.1 PGL(2, q), PSL(2, q) and the classical modular group

The projective linear group PGL(n, F) over a field F is the quotient GL(n, F)/Z of the
group of invertible n × n matrices by its centre Z = { aIn | a ∈ F r {0}}. Its subgroup,
the projective special linear group PSL(n, F), is the quotient of SL(n, F), the group of

8
all n × n matrices over F of determinant one, by its subgroup of all scalar matrices of
determinant one.
There is a natural isomorphism between
! PGL(2, F) and a group of Möbius transforma-
a b az + c
tions, under which the matrix corresponds to the transformation z 7→ ,
c d bz + d
when multiplication of transformations is read from left to right. This gives actions of
PGL(2, F) and PSL(2, F) on the projective line PL(F) = F ∪ {∞}. Also if F is finite, of
order q, then PGL(2, F) and PSL(2, F) are denoted by PGL(2, q) and PSL(2, q).
A search for 3-januarials may begin with the classical modular group

PSL(2, Z) = h x, y | x2 = y3 = 1i

which acts on Q ∪ {∞} by Möbius transformations with x : z 7→ −1/z and y : z 7→

(z − 1)/z. Notice that xy : z 7→ z + 1. A portion of the resulting coset diagram is shown
in Figure 5.

4/5
1/5 5/4
5
∞ 0
−1/4 −4
1/4
4 1
7/4 3/4
−1 −3/4 4/7
−3 4/3
−4/3 3 2
−1/3 −2 1/3
3/7 7/3
1/2
−1/2
2/3
3/2 −2/3
8/3 3/8
−3/2 −5/3
5/3

−3/5 3/5
5/8 5/2
2/5 8/5
−5/2
−2/5
7/2

7/5 2/7
5/7

t : z 7→ 1/z

Figure 5: The coset graph arising from the action of the modular group PSL(2, Z) =
h x, y | x2 = y3 = 1 i on Q ∪ {∞} via x : z 7→ −1/z and y : z 7→ (z − 1)/z.

Suppose p is a prime. Then the group PSL(2, p) is a homomorphic image of

∆(2, 3, p) = h x, y | x2 = y3 = (xy) p = 1 i,

whereby ∆(2, 3, p) acts on F p ∪{∞} via x : z 7→ −1/z and y : z 7→ (z−1)/z, with product
xy : z 7→ z + 1. The resulting coset diagrams Γ are quotients of the diagram in Figure 5;
for example, the diagrams for PSL(2, 13) and PSL(2, 17) are shown in Figure 6. But,
these coset diagrams do not immediately yield januarials, since the orbits of hxyi have
lengths 1 and p rather than both (p + 1)/2.

9
 ∞ 0
∞ 0

1
1
16
12
2 9
2 7

8 15
6 11 12 10
5 8

3 6
3 9

11 14
4 10 5 7

4 13
t : z 7→ 1/z
t : z 7→ 1/z

Figure 6: Coset graphs arising from actions of ∆(2, 3, 13) on F13 ∪ {∞} and ∆(2, 3, 17)
on F17 ∪ {∞}, both via x : z 7→ −1/z and y : z 7→ (z − 1)/z.

4.2 Associates

Here is a potential remedy for the failure of the coset diagrams constructed above from
PSL(2, p) to produce januarials. It applies in the general setting of a finite group G
acting on a set S and containing elements x and y satisfying x2 = yk = (xy)l = 1 for
some l ∈ Z. Let Γ be the resulting coset graph for the action of ∆(2, k, l) on S , via G,
with respect to the generating set {x, y}.
Suppose G has an element t of order 2 with the property that t−1 xt = x−1 (= x) and
t−1 yt = y−1 . Conjugation by such an element t reverses every cycle of y and preserves
every cycle of the involution x, and hence t induces a reflection of the coset graph Γ.
Note that (xt)2 = x(txt) = xx = 1, which allows us to consider the pair (xt, y) in place of
(x, y). If l′ is the order of xt y, then we have an action of ∆(2, 3, l′) on S via hxt, yi. The
resulting coset graph Γ′ is called an associate of Γ. This graph also admits a reflection
via the same involution t, since t−1 (xt)t = x−1 t = xt = (xt)−1 (and t−1 yt = y−1 ). For
more details about the correspondence (x, y, t) 7→ (xt, y, t), see [5]. The associate graph
Γ′ gives a new candidate for a januarial.
Examples 4.1. When G is PSL(2, p) for some prime p ≡ 1 mod 4, and x : z 7→ −1/z
and y : z 7→ (z − 1)/z, we can take t to be the transformation z 7→ 1/z, which has order
2 and satisfies t−1 xt = x−1 and t−1 yt = y−1 , as required. [The condition p ≡ 1 mod 4
ensures that −1 is a square mod p, so that the transformation t (and hence also xt) lies
in PSL(2, p).] In this case, xt is the transformation z 7→ −z. Hence, in particular, the
associates of the coset graphs in the cases p = 13 and 17 from Figure 6 are precisely
those in Figures 1 and 2. The transformation xty : z 7→ (z + 1)/z has two cycles of equal
length, and so in both cases the associates are januarials — specifically those depicted
in Figure 3.

10
Like Γ, the associate can fail to yield a januarial if the sizes of the hxyi-orbits are not
the requisite |S | /2, but it succeeds in many cases. In Section 4.5 we will explore when
it can be successfully applied to the examples from PSL(2, p). But first we need the
following study of conjugacy classes in PGL(2, q).

4.3 Classifying conjugacy classes in PGL(2, q)

Some of the details of the analysis in this section are similar to that carried out by
Macbeath in [6].
!
a b
Let q be any odd prime-power greater than 3, say q = p s . For M = ∈ GL(2, q),
c d
we may define
(tr M)2 (a + d)2
θ(M) = = .
det M ad − bc

The characteristic polynomial of M is det(xI2 − M) = x2 − tr(M)x + det(M), and tr(M)

and det(M), and therefore also θ(M), are invariant under conjugacy within GL(2, q).
Since θ(M) is invariant under scalar multiplication of M, it follows that this gives us
a well-defined function θ : PGL(2, q) → Fq that is constant on conjugacy classes of
PGL(2, q).
In fact, the function θ parametrises the conjugacy classes of PGL(2, q), as follows.
Proposition 4.2. Let g and h be elements of PGL(2, q). If θ(g) = θ(h) < {0, 4}, then
g and h are conjugate in PGL(2, q). In the exceptional cases, there are precisely two
conjugacy classes on which θ = 0, namely the class of involutions in PSL(2, q) and the
class of involutions in PGL(2, q) r PSL(2, q), and two classes on which θ = 4, namely
the class containing the identity element and the class of the transformation z 7→ z + 1.

This proposition can be proved using rational canonical forms, but also we can give a
direct proof for the generic case.

Proof for the case where θ(g) < {0, 4}. Suppose the transformation g ∈ PGL(2, q) is in-
duced by the matrix M ∈ GL(2, q), with tr(M) , 0. Then we can choose a vector u ∈ Fq2
such that u and uM are linearly independent
! over Fq . The matrix for g with respect to
0 1
the basis {u, uM} is the of the form , but since the trace is non-zero and a conju-
∗ ∗
gacy invariant, we can change the basis if necessary, so that the matrix for g has entry
1 in the lower-right corner. The matrix for g then becomes
!
0 1
M′ = ,
−∆ 1

where ∆ is the determinant. But then θ(g) = θ(M ′ ) = 1/∆, so ∆ = 1/θ(g), and it
follows that θ(g) determines the matrix. Since matrices representing the same linear

11
transformation with respect to different bases are conjugate within GL(2, q), we find
that θ(g) determines the conjugacy class of g. 

The utility of the parameter θ is enhanced by the following lemma, which gives a
number of cases in which the order of an element y ∈ PGL(2, q) determines θ(y).
Elements with trace 0 give involutions in PGL(2, q), and conversely, while elements
with trace −1 and determinant 1 give elements of order! 3 in PGL(2, q), and parabolic
1 0
elements (which are the conjugates of the matrix , or equivalently, the elements
1 1
with trace 2 and determinant 1), give elements of order p in PGL(2, q). We also note
that every element of order (q + 1)/2 in PGL(2, p) is the square of an element of order
q+1 in PGL(2, q), and hence lies in PSL(2, q) and is the product of two cycles of length
(q + 1)/2 in the natural action of PGL(2, q) on Fq ∪ {∞}.
Lemma 4.3. If y is an element of order 1, 2, 3, 4 or 6 in PGL(2, q), then θ(y) = 4, 0, 1,
2 or 3, respectively. Also if y has order p (the prime divisor of q) then θ(y) = 4.
!
a b
Proof. Suppose y is induced by the element M = ∈ GL(2, q). Then the first
c d
three cases are easy consequences of the respective observations that in those cases, M
is scalar, or M has trace 0, or M has minimum polynomial x2 + x + 1.
!
2 a2 + bc b(a + d)
For the next two cases, we note that M = and therefore
c(a + d) d2 + bc

tr(M 2 ) = a2 + d2 + 2bc = (a + d)2 − 2(ad − bc) = (tr M)2 − 2 det M.

If g has order 4, then M 2 has order 2, and so 0 = tr(M 2 ) = (tr M)2 − 2 det M, which
gives θ(y) = θ(M) = (tr M)2 /det M = 2. Similarly, if y has order 6, then since M 2
induces an element of order 3 in PGL(2, q) we know that

((tr M)2 − 2 det M)2 = (tr(M 2 ))2 = det(M 2 ) = (det M)2 ,

and therefore (θ(M) − 2)2 = 1. But since y does not have order 3, we know that
θ(M) , 1, and so θ(M) − 2 = 1, which gives θ(y) = θ(M) = 3.
Finally, if y has order
! p, then y is parabolic and therefore induced by some conjugate
1 0
of the matrix , which implies that θ(y) = (1 + 1)2 /1 = 4. 
1 1

4.4 How many 3-januarials arise from PSL(2, p)?

We can now proceed further, to consider 3-januarials. Let φ be Euler’s totient function
— that is, let φ(n) be the number of integers in {1, . . . , n} that are coprime to n.

12
Lemma 4.4. The number of conjugacy classes of elements in PGL(2, q) of order
(q + 1)/2 is 21 φ((q + 1)/2). Moreover, if z is any element of order (q + 1)/2 in PGL(2, q),
then
n every
o one of these conjugacy classes intersects the subgroup generated by z in
zi , z−i for exactly one i coprime to (q + 1)/2.

Proof. This follows easily from the observation that every element ! of order (q + 1)/2 in
i
λ 0
SL(2, q) is conjugate in GL(2, q2) to one of the form , where λ is an element
0 λ−i
of order (q + 1)/2 in the field Fq2 , and the traces λi + λ−i are distinct in Fq . 
Lemma 4.5. For any given conjugacy class C of elements of PSL(2, q) of order l <
{1, 2, p}, there exists a triple (x, y, xy) of elements of PSL(2, q) such that x has order 2,
and y has order 3, and xy is in C. Moreover, this triple is unique up to conjugacy in
PGL(2, q) whenever l , 6.

Proof. Every element of order 3 in PSL(2, q) is conjugate

! in PGL(2, q) to the element
−1 1
y : z 7→ (z − 1)/z, induced by the matrix Y = , while any element of order
−1 0
!
a b
2 in PSL(2, q) is induced by a matrix X of the form X = , with trace 0 and
c −a
determinant −a2 − bc = !1. Now observe that for any such choice of X and Y, we
−a − b a
have XY = , which has trace tr(XY) = −a − b + c and determinant
a−c c
det(XY) = det(X)det(Y) = 1.
This can be turned around: we can show that for any given non-zero trace r, there
exist a, b and c in Fq such that −1 = a2 + bc and r = −a − b + c, and hence there
exists a matrix X of trace zero such that XY has trace r, giving a triple (x, y, xy) of the
required type. Note that we need c − b = r + a and bc = −(1 + a2 ), and therefore
(c + b)2 = (c − b)2 + 4bc = r2 + 2ar − 3a2 − 4. Now if p , 3, then we can multiply
this by 3 and it becomes 3(c + b)2 = 4r2 − 12 − (3a − r)2 ; and then since in Fq there are
(q + 1)/2 elements of the form 3u2 and (q + 1)/2 elements of the form s − v2 for any
given s ∈ Fq , and any two subsets of size (q + 1)/2 in Fq have non-empty intersection,
the latter equation can be solved in Fq for c + b and 3a − r, and hence for c + b and a
(and c − b = r − a), and hence for a, b and c (since q is odd). On the other hand, if
p = 3, then the equation becomes (c + b)2 = r2 + 2ar − 3a2 − 4 = r2 + 2ar − 1, which
is even easier to solve for a, b, c, provided that r , 0.
The main assertion now follows easily. Uniqueness up to conjugacy in PGL(2, q) when
l , 6 is left as an exercise. 

Since the automorphism group of PSL(2, p) is PGL(2, p) for every odd prime p, we
obtain the following:
Corollary 4.6. For any prime p > 3, the number of distinct 3-januarials constructible
from PSL(2, p) in the way described in sub-sections 4.1 and 4.2 is 12 φ((p + 1)/2).

13
For example, when p = 37 the number of 3-januarials is φ(19)/2 = 9, and when p = 53
the number is φ(27)/2 = 9 as well. A further attractive property is as follows:
Lemma 4.7. For every triple (x, y, xy) as in Lemma 4.5 with l , 6, there exists an
involution t in PGL(2, q) such that t−1 xt = x−1 and t−1 yt = y−1 .

Proof. We may suppose that x and y are induced by the matrices

! X and Y as in the
−(b + c) 2a + b
proof of Lemma 4.5. In that case, let T = , which has the property
2a − c b+c
that XT = T X −1 while YT = T Y −1 . The determinant of T is

−(b + c)2 − (2a + b)(2a − c) = −4a2 − b2 − c2 − 2ab + 2ac − bc,

which equals 3det(XY) − (tr(XY))2 , since det(XY) = −a2 − bc while tr(XY) = −a −

b + c, and therefore T is invertible if and only if θ(XY) = (tr(XY))2 /det(XY) , 3, or
equivalently, XY does not have order 6. Finally, note that if T is invertible, then since
its trace is zero, we have t2 = 1. 

4.5 Necessary conditions for associates to yield januarials

For any positive integer n, define θn to be the set of all values of θ(g) for elements g of
order n in the group PGL(2, q). Consider the effect of the mapping r 7→ (r − 1)2 on the
elements of this set θn , for some n, as follows.
Suppose that r ∈ θn , where n is coprime to q, and let g be an element of order n in
PGL(2, q) with θ(g) = r. By taking a conjugate of g if necessary in PGL(q2 ), we may
assume that g is the transformation z 7→ ρz induced by the matrix
!
ρ 0
M= ,
0 1

where ρ is a primitive nth root of 1 in Fq or Fq2 . In this case, tr(M) = ρ + 1 while

det(M) = ρ, and so

(ρ + 1)2
r = θ(g) = θ(M) = = ρ + ρ−1 + 2.
ρ

It follows that (r − 2)2 = (ρ + ρ−1 )2 = ρ2 + ρ−2 + 2. But r is of order n and so ρ will

be a primitive nth root of unity. In particular, if n is odd then also ρ2 is a primitive nth
root of unity, in which case (r − 2)2 = ρ2 + ρ−2 + 2 = θ(M 2 ), which also belongs to θn .
Iterating the procedure then yields further elements of θn , until we reach a stage where
i i i
ρ2 = ρ±1 , and then ρ2 + ρ−2 + 2 = ρ + ρ−1 + 2 = r.

We now derive two necessary conditions on the values of θ(g) in θn in the special case
where n = (q + 1)/2.
Lemma 4.8. If g is an element of order (q + 1)/2 in PSL(2, q), then θ(g) is a square in
Fq while θ(g) − 4 is not a square in Fq .

14
 µ2
!
0
Proof. First, g is conjugate in PGL(2, q2 ) to the projective image of M =
0 1
where µ is a primitive (q + 1)th root of unity in Fq2 , and this gives

(tr M)2 (µ2 + 1)2

θ(g) = = = µ2 + 2 + µ−2 .
det M µ2

In particular, θ(g) = µ2 +2+µ−2 = (µ+µ−1 )2 , which is a square in Fq (since µ+µ−1 ∈ Fq ).

On the other hand, θ(g) − 4 = µ2 − 2 + µ−2 = (µ − µ−1 )2 , which is not a square in Fq ,
since µ − µ−1 = 2µ − (µ + µ−1 ) < Fq . 
Corollary 4.9. Suppose g ∈ PSL(2, p) has order (p + 1)/2.

(a) If θ(g) = −1, then p ≡ 13 or 17 mod 20.

(b) If θ(g) = −2, then p ≡ 17 or 19 mod 24.
(c) If θ(g) = −3, then p ≡ 13, 19 or 31 mod 42.

Proof. In case (a), by Lemma 4.8 we require that −1 = θ(g) is a square mod p while
−5 = θ(g) − 4 is not, and hence also 5 is not. Thus p ≡ 1 mod 4, and by quadratic
reciprocity, also p ≡ 2 or 3 mod 5, giving p ≡ 13 or 17 mod 20. Similarly, in case (b)
we require that −2 is a square mod p while 3 is not. It follows that p ≡ 1 or 3 mod 8,
while also p . ±1 mod 12, and therefore p ≡ 17 or 19 mod 24 (since we are assuming
p > 3). Finally, in case (c) we require that −3 is a square mod p while −7 is not, and
hence that p is a square mod 3 and a non-square modulo 7, giving p ≡ 10, 13 or 19
mod 21, and therefore p ≡ 13, 19 or 31 mod 42. 

Next, we give what Higman described as the ‘Pythagorean Lemma’. One motivation
for this choice of name is that for an element X of SO(3), we could define θ(X) to
be 4 cos2 (φ/2), where φ is the angle of rotation of X. Now let a, b and c be half-
turns about the three co-ordinate axes, and let d be a half-turn about any unit vector
(cos α, cos β, cos γ). Then the angle of rotation of ad is twice the angle between the
axes of a and d, namely 2α, and similarly the angles of rotation of bd and cd are 2β
and 2γ. Thus θ(ad) + θ(bd) + θ(cd) = 4 cos2 α + 4 cos2 β + 4 cos2 γ = 4.
Lemma 4.10 (Pythagorean Lemma). Suppose a, b and c are the non-identity elements
of a subgroup of PSL(2, q) isomorphic to the Klein 4-group. If d is any element of order
2 in PSL(2, q), then
θ(ad) + θ(bd) + θ(cd) = 4.

Proof. One can take a quadratic extension Fq2 of the ground field Fq , and then in
PSL(2, Fq2 ), all copies of the Klein 4-group are conjugate, since we are assuming that
q is odd. (See the classification of subgroups of PSL(2, p f ) in [4] for example.) Hence
we may assume that our Klein 4-group in PGL(2, q) is generated by
1 1
a : z 7→ −z, b : z 7→ − , and c : z 7→ .
z z

15
 αz + β
Now any involution d has trace zero and hence is of the form d : z 7→ .
γz − α
From this we find that
−αz + β −α + βz α + βz
ad : z 7→ , bd : z 7→ , and cd : z 7→ ,
−γz − α −γ − αz γ − αz
and therefore
4α2 (β − γ)2 (β + γ)2
θ(ad) + θ(bd) + θ(cd) = + + = 4,
α2 + βγ −βγ − α 2 βγ + α2
as required. 

Lemma 4.3, the Pythagorean Lemma and Corollary 4.9 combine to give us necessary
conditions for associates formed as in Section 4.2 to yield januarials. The scope of
this result is limited to k equal to 3, 4 or 6, since these are the only orders for which
Lemma 4.3 applies.
Corollary 4.11. Consider ∆(2, k, p) = h x, y | x2 = yk = (xy) p = 1 i acting on
F p ∪ {∞} via PSL(2, p) in such a way that xy is the transformation z 7→ z + 1. Let t be
an involution in PGL(2, p) such that t−1 xt = x−1 and t−1 yt = y−1 , and suppose that the
resulting associate coset diagram found by replacing x by xt yields a k-januarial for
PSL(2, p) or PGL(2, p), depending on whether or not t lies in PSL(2, p). Then
(a) if k = 3, then p ≡ 13 or 17 mod 20 ;
(b) if k = 4, then p ≡ 17 or 19 mod 24 ;
(c) if k = 6, then p ≡ 13, 19 or 31 mod 42.

Proof. When k = 3, 4 or 6, Lemma 4.3 gives us θ(y) = 1, 2, 3, respectively, and in all

three cases θ(xy) = 4, since xy is the transformation z 7→ z + 1. Apply the Pythagorean
Lemma, by taking a, b, c and d as the four involutions t, x, xt and ty respectively, to give
θ(y) + θ(xty) + θ(xy) = 4. Note that θ(xty) = θ(xyt) because t−1 (xty)t = (t−1 xt)yt = xyt.
Thus we find θ(xyt) = θ(xty) = 4 − θ(y) − θ(xy) = −θ(y), which equals −1, −2, or −3,
respectively, when k is 3, 4, or 6. Finally, for the associate to be a januarial we need
xyt to have order (p + 1)/2, and so the constraints on p follow from Corollary 4.9. 

5 Examples

In this section we explore a number of examples of k-januarials guided by Corol-

lary 4.11.
We begin with k = 3. The eight smallest p satisfying the conditions of Corollary 4.11
are 13, 17, 37, 53, 73, 97, 113 and 137. We have already seen how the cases p = 13
and p = 17 yield the 3-januarials depicted in Figure 3. The cases where p is 37, 53,
73, 97 or 137 all yield januarials. On the other hand, the standard construction (with

16
 ∞ 0

72

38 2 36

71 37 25

49 3
35
48
58 32 16 34 39

15
26 59
47 40 42
14
27

46
60
28
13 33 31
57 41 45
61

6
29 43 17 68
44
5
30

56
69 67 52 22 66 12 18
10
4 21 7
55

62 8 64 53
24 70
19 65 9 50
11 20

54 63

51 23

Figure 7: The associate of the standard action of PSL(2, 73) on F73 ∪ {∞}, giving an
action of ∆(2, 3, 37) via z 7→ −z and z 7→ (z − 1)/z. This yields a 3-januarial of genus 5
and simple type (3, 2, 1).

x : z 7→ −1/z and y : z 7→ (z − 1)/z ) does not yield a januarial in the case p = 113,
since in that case xty : z 7→ (z + 1)/z, which has order 19, rather than (113 + 1)/2 = 57.
The action of PSL(2, 73) on F73 ∪ {∞} via x : z 7→ −1/z and y : z 7→ (z − 1)/z gives
a coset graph, the associate of which is depicted together with its companion graph in
Figure 7.
The number of x-edges which are not loops and y-faces in the associate are 36 and 26,
respectively, so the genus is (36 − 26)/2 = 5 by Lemma 3.1. The type of the januarial
is apparent from Γ. The blue subgraph (which is the common boundary of S 1 and S 2 )
consists of three disjoint simple closed curves, and so h = 3 by Lemma 3.3. There are
22 vertices on S 1 and 21 on S 2 ; there are 26 edges on S 1 (coloured green and blue),
and 27 edges on S 2 (coloured red and blue); and both S 1 and S 2 have 1 face and 3
holes. Hence by Lemma 3.6, the genera of S 1 and S 2 are (2 − 22 + 26 − 3 − 1)/2 = 1
and (2 − 21 + 27 − 3 − 1)/2 = 2, respectively. Accordingly, the 3-januarial is of type
(3, 2, 1), and this gives an alternative means of identifying the genus as 2 +1 +3 −1 = 5,
by Lemma 3.4.

17
 We now turn to 4-januarials.
For the standard construction (with x : z 7→ −1/z and y : z 7→ (z − 1)/z ) to give a
4-januarial for PSL(2, p), Corollary 4.11 requires p ≡ 17 or 19 mod 24. The eight
smallest values for the prime p satisfying this condition are 17, 19, 41, 43, 67, 89,
113 and 117. But also if we want xt to lie in PSL(2, p), we need t : z 7→ 1/z to lie
in PSL(2, p), and then we need −1 to be a square mod p, and so p ≡ 1 mod 4. The
cases p = 17, 89, 113 and 117 all give simple 4-januarials for PSL(2, p) in this way,
while the case p = 41 fails, since in that case xy has order 7 rather than the required
(41 + 1)/2 = 21.
On the other hand, if we are happy to construct 4-januarials for PGL(2, p) instead
of PSL(2, p), we can relax the requirements and allow x, y or t to lie in PGL(2, p) \
PSL(2, p). When we do that, we get simple 4-januarials for PSL(2, p) in the cases
p = 19 and 43 (but not for p = 67). We consider the case p = 43 in more detail below.

∞ 0

23 42 1 22 21 29

15 11 2 33
26 24
20 14

30 38
6 18
25 31
13 5

39 16
17 10
7 36 35 8 3 40
4 27

17 25 37 19
41 12 9 32

Figure 8: A coset diagram for the action of ∆(2, 4, 22) on F43 ∪{∞} given by z 7→ 21z/22
and y : z 7→ (2z − 1)/2z, its companion graph, and two partitions of the subgraph G into
circuits. This yields a 4-januarial of genus 5 and general type ((4, 3), (2, 2)).

Letting y be the transformation z 7→ (2z − 1)/2z, we can taken x as z 7→ 21/z and once
more get the product xy as the (parabolic) transformation z 7→ z + 1, which fixes ∞
and induces a 43-cycle on the remaining points. The generator y induces a permutation
with eleven 4-cycles and no fixed points, while the generator x fixes two points (namely
8 and 35) and induces 21 transpositions on the remaining points.
We have not drawn the resulting coset graph, but note that it is reflexible, via the trans-
formation t : z 7→ 22/z. Its associate, given by the triple (xt, y, xty), shown alongside
its companion graph in Figure 8, produces a 4-januarial, since xty is the transformation
z 7→ (z + 22)/z, which has two cycles of length 22.
The genus is (21 − 11)/2 = 5 by Lemma 3.1. We can use Lemma 3.5 to find h1 and h2 .

18
The partition arising from the attaching map ρ2 (following the green and blue edges)
comprises h1 = 4 circuits, while the partition for ρ1 (following the red and blue edges)
comprises h2 = 2 circuits. Both are indicated in the figure. By Lemma 3.6, we have
2 − 2g1 = 8 − 17 + 4 + 1 and 2 − 2g2 = 11 − 16 + 2 + 1, and so g1 = 3 and g2 = 2.
Finally in this section, we consider 6-januarials.
In this case, Corollary 4.11 requires p ≡ 13, 19 or 31 mod 42. Figure 9 shows a coset
graph for PGL(2, 31). Its associate, which yields a januarial J, is shown in Figure 10
together with the companion graph. By Lemma 3.1, the genus of J is (15 − 7)/2 =
4. Lemma 3.5 gives h1 and h2 : as shown in the figure, we find that h1 = 1 circuits
comprise the partition of G arising from the attaching map ρ2 (following the green
and blue edges), and h2 = 4 comprise the partition arising from ρ1 (following the red
and blue edges). Hence by Lemma 3.6, we find that 2 − 2g1 = 5 − 13 + 1 + 1 and
2 − 2g2 = 7 − 12 + 4 + 1, and therefore g1 = 4 and g2 = 1.
∞ 0
2 10 1 21 30 26

6 9 22
11 16 19
8 23
28 24
13 4

18 27

5 29
17 14
3 7
20 15
25 12

t : z 7→ 1/(3z)

Figure 9: A coset graph for the action of ∆(2, 6, 31) on F31 ∪ {∞} given by x : z 7→ 10/z
and y : z 7→ (z + 10)/z.

∞ 0
1 21

26 11 16 2

30 10
19 6

22 8
4
13
24 28
9 12 25 23
7 3
15 20

29 5
14 17
27 18

Figure 10: The associate of Figure 9, giving an action of ∆(2, 6, 16) on F31 ∪ {∞} via
x : z 7→ z/30 and y : z 7→ (z + 10)/z, together with its companion graph, and two
partitions of the subgraph G into circuits. The resulting 6-januarial is of genus 4 and
general type ((1, 4), (4, 1)).

19
6 Afterword

6.1 Higman’s portrait

Higman delivered the lectures on which this account is based in the Higman Room of
the Mathematical Institute at Oxford University. As he spoke, he could see an image
of himself looking on, from his 1984 portrait by Norman Blamey, which is reproduced
below.

Figure 11: Norman Blamey’s 1984 portrait of Graham Higman

The portrait shows Higman beside a coset graph for the action of the group PSL(2, 11)
on the cosets of a dihedral subgroup of order 10 and index 66. Equivalently, it gives
the natural action of PSL(2, 11) on the 66 unordered pairs of points on the projective
line over a field of order 11. The two generators x and y satisfy the relations x2 = y5 =
(xy)5 = 1, but also the diagram is reflexible about a vertical axis of symmetry, and the
reflection is achievable by conjugation by an involution t in the same group.
In fact x and yt may be taken as involutory generators of the stabilizer of the pair {0, ∞},
such as z 7→ −1/z and z 7→ 2/z, and t as the transformation z 7→ (z + 1)/(z − 1). These
choices make y the transformation z 7→ (z + 2)/(2 − z). The three generators x, y and t
then satisfy the relations written on the blackboard in the portrait, namely

x2 = y5 = (xy)5 = t2 = (xt)2 = (yt)2 = (xyt)5 = 1,

which are the defining relations for the group G5,5,5 in the notation of Coxeter [2].
Hence in particular, G5,5,5 is isomorphic to PSL(2, 11).

20
The diagram does not give a januarial, but rather a 13-face map. The associated surface
has genus 2, since there are 13 pentagons corresponding to the 5-cycles of hyi, and 28
edges between distinct pairs of such pentagons (from transpositions of x), and 13 faces
coming from the 5-cycles of hxyi, giving Euler characteristic 13 − 28 + 13 = −2. The
isomorphism with G5,5,5 also makes PSL(2, 11) the automorphism group of a regular
map of type {5, 5}5 on a non-orientable surface of Euler characteristic −33 (see [3]),
and hence also the automorphism group of a regular 3-polytope of type [5, 5].

6.2 Other sources of januarials

Many januarials can also be constructed from groups other than PSL(2, q) and PGL(2, q).
For example, the alternating group Alt(16) is generated by elements x = (2, 4)(3, 7)(6,
10)(8, 16)(9, 13)(11, 14) and y = (1, 2, 3)(4, 5, 6)(7, 8, 9) (10, 11, 12)(13, 14, 15), with
product xy = (1, 2, 5, 6, 11, 15, 13, 7)(3, 8, 16, 9, 14, 12, 10, 4), which has two cycles of
length 8. The resulting coset diagram is shown in Figure 12.
7 9

3 13
1 8

16
15

2 14

12
4
11

5
6 10

Figure 12: A coset graph for an action of Alt(16) of degree 16, together with its com-
panion graph. This gives a 3-januarial of genus 0 and simple type (1, 0, 0).

Other examples are obtainable from the groups PSL(2, q) and PGL(2, q) without taking
the approach that we did in Section 4 which had xy as the transformation z 7→ z + 1.
An example is given in Figure 13.

0 9
3 1
2
5

4 8
6 10
7 ∞

Figure 13: A coset graph for an action of PSL(2, 11) on F11 ∪ {∞}, via x : z 7→ −1/z
and y : z 7→ (8z − 8)/(z + 1), together with its companion graph, and two partitions of
the subgraph G. This results in a 6-januarial of genus 1 and general type ((2, 1), (2, 1)).

21
References
[1] M.D.E. Conder, Generators for alternating and symmetric groups, J. London
Math. Soc (2) 22, 1980, 75–86.
[2] H.S.M. Coxeter, The abstract groups Gm,n,p , Trans. Amer. Math. Soc. 45 (1939),
73–150.
[3] H.S.M. Coxeter and W.O.J. Moser, Generators and Relations for Discrete
Groups, 4th ed., Springer Berlin (1980).
[4] B. Huppert, Endliche Gruppen. I. Die Grundlehren der Mathematischen Wis-
senschaften, Band 134, Springer-Verlag, Berlin-New York 1967 xii+793 pp.
[5] G.A. Jones and J.S. Thornton, Automorphisms and congruence subgroups of the
extended modular group, J. London Math. Soc. (2) 34 (1986), 26–40.
[6] A.M. Macbeath, Generators of the linear fractional groups, in Number Theory,
Proc. Symposia in Pure Mathematics 12 (American Mathematical Society, Prov-
idence, R.I., 1969), pp. 14-32.

22