The Pennsylvania State University The Graduate School Eberly College of Science
PRIME GEODESIC THEOREMS FOR COMPLEXES OF PGL3(F ) AND
PGSp4(F )
A Dissertation in Mathematics by João Correia Matias
© 2018 João Correia Matias
Submitted in Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy
December 2018 The dissertation of João Correia Matias was reviewed and approved∗ by the following:
Wen-Ching Winnie Li Distinguished Professor of Mathematics Dissertation Advisor, Chair of Committee
A. Kirsten Eisenträger Professor of Mathematics
Yuri Zarhin Professor of Mathematics
Martin Fürer Professor of Computer Science and Engineering
Carina Curto Associate Professor of Mathematics Co-Director of Graduates Studies
∗Signatures are on file in the Graduate School.
ii Abstract
In all history of Number Theory, prime numbers have been one of the greatest sources of questions, displaying many interesting properties. The distributions of prime numbers, and prime ideals of number fields are given by two famous results in Number Theory: the Prime Number Theorem and the Chebotarev Density Theorem. Moreover, prime numbers exhibit deep connections with the Riemann zeta function and the Riemann Hypothesis. Starting in the twentieth century, mathematicians have studied many objects analogous to the ones mentioned above in other mathematical contexts, including: algebraic varieties, Riemann surfaces, graphs and simplicial complexes. In this dissertation, we focus on studying closed geodesic paths in combinatorial complexes. Now, let F be a nonarchimedean local field. The equivalence classes of primitive closed geodesic paths, under cyclic rotation, in surfaces or multicomplexes are called primes. We describe the distribution of the primes in finite multicomplexes covered by the buildings associated to PGL3(F ) or PGSp4(F ). In the case of Ramanujan complexes covered by the building associated to PGL3(F ), we give more precise estimates, including bounds for the error term. Additionally, we study finite coverings of these multicomplexes, in which case the groups of automorphisms are analogous to the Galois groups of extensions of number fields. As a consequence, we derive results analogous to the Chebotarev Density Theorem in various contexts. Lastly, and throughout the dissertation, we define Zeta functions and L-functions attached to such multicomplexes. We show that they are rational functions and relate them to certain adjacency operators.
iii Table of Contents
List of Figures viii
Acknowledgments ix
Chapter 1 Introduction 1 1.1 Zeta functions and the distribution of primes – a historical review . . . . 1 1.1.1 Prime numbers and the Riemann zeta function ...... 2 1.1.2 Rational points and Zeta functions of algebraic varieties ...... 2 1.1.3 Primitive closed geodesics and the Selberg zeta function ...... 3 1.1.4 Primitive geodesic cycles and the Ihara zeta function ...... 4 1.2 Ramanujan graphs ...... 5 1.3 Primes and Zeta functions for finite quotients of buildings ...... 6 1.4 Chebotarev Density Theorem ...... 8 1.4.1 In Number Theory ...... 8 1.4.2 For Riemann surfaces ...... 8 1.4.3 For graphs ...... 9 1.5 Main results ...... 10
Chapter 2 Preliminaries 12 2.1 Graphs ...... 12 2.2 Simplicial complexes and multicomplexes ...... 13 2.3 Primes in finite quotients of X ...... 14 2.3.1 Zeta functions and L-functions ...... 16 2.3.2 Distribution of primes of given dimension and type ...... 19 2.3.3 Finite coverings of multicomplexes ...... 21 2.3.4 Cumulative distribution of primes ...... 24
Chapter 3 Analysis of PGL3(F ) 26 3.1 The PGL3(F ) building ...... 26
iv 3.1.1 Vertices ...... 26 3.1.2 Edges ...... 27 3.1.3 Pointed chambers ...... 27 3.1.4 Apartments ...... 28 3.2 Finite quotients of X ...... 29 3.3 Type 1 edge zeta function ...... 29 3.3.1 Adjacency operator for type 1 edges ...... 29 3.3.2 Edge zeta function ...... 30 3.3.3 Edge L-functions ...... 31 3.3.4 Main results for 1-dimensional primes of type 1 ...... 31 3.4 Chamber zeta function ...... 32 3.4.1 Adjacency of chambers ...... 32 3.4.2 Adjacency operator for directed chambers ...... 33 3.4.3 Chamber zeta function ...... 34 3.4.4 Chamber L-functions ...... 35 3.4.5 Main results for 2-dimensional primes of type 1 ...... 35 3.5 Proofs of Theorems 11 and 12 ...... 36 3.5.1 Edge representatives ...... 36 3.5.2 Successor map ...... 38 3.5.3 Properties of AE,Γ ...... 39 3.5.4 Proof of Theorem 11 ...... 41 3.5.5 Proof of Theorem 12 ...... 41 3.6 Proofs of Theorems 14 and 15 ...... 42 3.6.1 Pointed chamber representatives ...... 42 3.6.2 Successor map ...... 43 3.6.3 Properties of AB,Γ ...... 44 3.6.4 Proof of Theorem 14 ...... 46 3.6.5 Proof of Theorem 15 ...... 46 3.7 Error terms of Ramanujan complexes ...... 47
Chapter 4 Analysis of PGSp4(F ) 49 4.1 The symplectic group ...... 49 4.1.1 Definitions ...... 49 4.1.2 Lattices ...... 50 4.1.3 Arithmetic of GSp4(F ) ...... 52 4.2 The Bruhat-Tits building associated to Sp4(F ) ...... 53 4.2.1 Vertices, edges and chambers ...... 53 4.2.2 Types of vertices and edges ...... 53 4.2.3 Types of chambers ...... 55 4.2.4 Apartments ...... 57 4.3 Finite quotients of X ...... 58
v 4.4 Type 1 edge zeta functions ...... 59 4.4.1 Adjacency operator for type 1 edges ...... 59 4.4.2 Type 1 edge zeta function ...... 60 4.4.3 Type 1 edge L-functions ...... 60 4.4.4 Main results for 1-dimensional primes of type 1 ...... 60 4.5 Type 2 edge zeta function ...... 61 4.5.1 Adjacency of type 2 edges ...... 61 4.5.2 Type 2 edge zeta functions ...... 63 4.5.3 Long type 2 edges ...... 63 4.5.4 Type 2 edge L-functions ...... 65 4.5.5 Main results for 1-dimensional primes of type 2 ...... 65 4.6 Spin type chamber zeta function ...... 66 4.6.1 Adjacency operator ...... 66 4.6.2 Spin type chamber zeta function ...... 68 4.6.3 Spin type chamber L-functions ...... 69 4.6.4 Main results for 2-dimensional primes of spin type ...... 69 4.7 Standard type chamber zeta function ...... 70 4.7.1 Translation of chambers ...... 70 4.7.2 Standard type chamber zeta function ...... 72 4.7.3 Standard type chamber L-functions ...... 73 4.7.4 Main results for 2-dimensional primes of standard type ...... 73 4.8 Proofs of Theorems 19 and 20 ...... 74 4.8.1 Representatives of type 1 edges ...... 75 4.8.2 Successor map ...... 76
4.8.3 Properties of AP1,Γ ...... 78 4.8.4 Proof of Theorem 19 ...... 80 4.8.5 Proof of Theorem 20 ...... 80 4.9 Proofs of Theorems 22 and 23 ...... 81 4.9.1 Representatives of long type 2 edges ...... 81 s 0 4.9.2 Successor map P2 ...... 82 A 0 4.9.3 Properties of P2,Γ ...... 85 4.9.4 Proof of Theorem 22 ...... 88 4.9.5 Proof of Theorem 23 ...... 88 4.10 Proofs of Theorems 26 and 27 ...... 89 4.10.1 Representatives of spin type chambers ...... 89
4.10.2 Successor map sI1 ...... 90
4.10.3 Properties of AI1,Γ ...... 91 4.10.4 Proof of Theorem 26 ...... 93 4.10.5 Proof of Theorem 27 ...... 94 4.11 Proofs of Theorems 30 and 31 ...... 94 4.11.1 Standard type chamber representatives ...... 94
4.11.2 Successor map sI2 ...... 95
vi 4.11.3 Properties of AI2,Γ ...... 97 4.11.4 Proof of Theorem 30 ...... 99 4.11.5 Proof of Theorem 31 ...... 100
Appendix Perron-Frobenius Theorem for nonnegative matrices 101
Bibliography 103
vii List of Figures
3.1 Part of the standard apartment, with labelled vertices...... 28
3.2 Some edges in the standard apartment...... 30
3.3 A gallery in the standard apartment, containing F0...... 34
4.1 Some of the chambers in the standard apartment...... 56
4.2 Part of the standard apartment, with labelled vertices...... 58
4.3 Some edges in the standard apartment...... 62
4.4 Part of a spin type geodesic gallery in the standard apartment...... 68
4.5 Part of a geodesic gallery in the standard apartment...... 72
viii Acknowledgments
First and foremost, I would like to thank my advisor, Professor Winnie Li, for suggesting me this project and for her guidance throughout graduate school. Her support was crucial to complete this dissertation and I am most grateful for her endless patience during our regular discussions. She also questioned each of my results and listened to my explanations with interest, pushing me constantly to do better. Finally, I am also thankful to her for always giving me a great example both as a person and as a mathematician. I would also like to thank Professors Kirsten Eisenträger, Martin Fürer, Mihran Papikian and Yuri Zahrin, for serving in my Ph.D. committee, for the comments they provided to this dissertation, and for all that they taught me both in their classes and by their example. I was very fortunate for coming upon amazing people at Penn State that I will cherish for the rest of my life. These include, in particular, everyone in the Mathematics department that contributed to one of the best communities in Penn State. I am also very thankful to all my friends and peers for their camaraderie, and for all the good moments that we shared. Last, but definitely not least, I am forever indebted to my parents Ilda and Jorge for their unending support and love, for understanding me and for their genuine care when I was feeling low.
ix Dedication
Para o meu pai e a minha mãe.
x Chapter 1 | Introduction
In this dissertation we study the distribution of primitive closed geodesic paths in finite 2-dimensional simplicial complexes or, in general, multicomplexes (as in [LLR18]) covered by the buildings associated to PGL3(F ) or PGSp4(F ), where F is a local nonarchimedean field. We analyze paths in the 1-skeleton or 2-skeleton of multicomplexes, respectively. We shall refer to equivalence classes of primitive closed geodesic paths, under cyclic rotations, as primes. From this set of primes we define Zeta functions and L-functions whose analytic properties can be used to describe the distribution of primes, just like what happens to prime numbers and the Riemann zeta function. These Zeta functions are, in fact, rational functions and have a simple formula in terms of adjacency operators acting on the finite dimensional vector spaces of functions on the 1-skeleton or 2-skeleton of multicommplexes, respectively. This connection is useful, as it allows us to write the poles of the associated Zeta functions in terms of the eigenvalues of a linear transformation. The main contribution to the number of primes of given length comes from the largest eigenvalues, in absolute value, of the adjacency operators, and determining them is a central problem in the dissertation. This gives rise to the first main result of this dissertation: versions of the Prime Number Theorem for primes on finite complexes. The second main result is a collection of different versions of the Chebotarev Density Theorem, and gives a finer distribution of the primes, when they are partitioned by the associated Frobenius conjugacy class in the Galois group of a finite unramified Galois cover.
1.1 Zeta functions and the distribution of primes – a historical review
Before we present our main results, in this and the next sections, we give a brief historical review of the zeta functions, Prime Number Theorem and the Chebotarev Density
1 Theorem concerning the distribution of primes in different settings.
1.1.1 Prime numbers and the Riemann zeta function
Prime numbers have fascinated mathematicians for centuries and they have inspired many questions in all fields of mathematics. Moreover, there are still many open problems in Number Theory related to prime numbers. One of the most notorious problems, that is now a theorem, is the Prime Number Theorem, which roughly states that the number of prime numbers in the interval [1, x], denoted by π(x), is approximately x/ log x. We draw attention to the fact that the Prime Number Theorem may be derived from the analytic properties of the Riemann zeta function defined as
∞ X 1 Y ζ(s) = = (1 − p−s)−1 , 1 . ns n=1 p prime number
Here, the second equality is a consequence of the unique prime decomposition of natural numbers. A more precise version of the Prime Number Theorem states (cf. [MV07]) that the prime counting function, π(x), satisfies
π(x) = Li(x) + O(xβ log x) , where Z x 1 Li(x) = dt , and β = sup{
1.1.2 Rational points and Zeta functions of algebraic varieties
In his 1923 dissertation, Emil Artin defined and studied Zeta functions associated to algebraic curves over finite fields. These functions count the number of rational points of a given algebraic curve over different finite extensions of a given finite field. His definition is easily extended to Zeta functions of algebraic varieties, which we explain succinctly.
Let V be an algebraic variety over a finite field k, and let Nν be the number of rational points of V over the field extension of k with degree ν, kν. Then, the Zeta function
2 attached to V is defined by X Nν Y Z(V, u) = exp uν = (1 − udeg v)−1 . ν ν≥1 closed point v
Here, a closed point is the orbit of a point x in V (k¯) under Gal(k/k¯ ) and deg v is the number of points in this orbit. The closed points play the role of primes for V . Part of the relevance of this function is due to four significant results known as the Weil Conjectures. These are four properties of the Zeta functions associated to algebraic varieties, conjectured by Weil [Wei49], and that had already been proved in the case of algebraic curves. The general case was proved by Deligne [Del74], Dwork [Dwo60] and Grothendieck [Gro65]. Two of those properties are the facts that these Zeta functions are rational and satisfy a version of the Riemann Hypothesis, which are in the same theme as this dissertation. 1 When V is the projective line P over k, the affine closed points are in bijective correspondence with the monic irreducible polynomials over k. Assume that k has size q.
One can verify that the number Mν of irreducible monic polynomials over k with degree 1 ν, in other words, the number of primes of P of degree ν, satisfies
qν qν/2 M = + O , for ν large . ν ν ν
This is the Prime Number Theorem for the projective line over a finite field with q elements.
1.1.3 Primitive closed geodesics and the Selberg zeta function
Recall that PSL2(R) acts transitively on the hyperbolic upper half-plane H by fractional linear transformations. These are also all the orientation preserving isometries of H.
In [Sel56], Selberg considered the compact Riemann surfaces of the form XΓ = Γ\H, where Γ is a discrete, torsion-free, cocompact subgroup of PSL2(R). The equivalence classes of primitive closed geodesics in this compact surface play the role of primes of XΓ.
Using these primes, Selberg defined a Zeta function attached to XΓ as
∞ Y Y −(s+j)l(C) Z(XΓ, s) = 1 − e , [C] j=1 where [C] runs over all primes of XΓ and l(C) is the length of C under the hyperbolic metric. He proved that this function and the Riemann zeta function have similar analytic
3 behaviors, but, unlike ζ(s), all except finitely many real zeros of Z(XΓ, s), in the strip 0 <
Theorem (Prime Geodesic Theorem for compact Riemann surfaces). Let
π(XΓ, x) = |{[C]: C is a primitive geodesic with length ≤ x}|,
1 and let s1, . . . , sk be the real zeros of Z(XΓ, s) in the interval ( 2 , 1). Then
3 s1 sk 2 π(XΓ, x) = Li(x) + Li(x ) + ··· + Li(x ) + O(x 4 (log x) ) .
1.1.4 Primitive geodesic cycles and the Ihara zeta function
Let F be a nonarchimedean local field with uniformizer π, ring of integers OF and q elements in its residue field. Now, observe that H = SL2(R)/ SO2(R). In [Iha66], Ihara defined an analog of Selberg’s zeta function over p-adic fields. For that purpose, let Γ be a discrete, torsion-free, cocompact subgroup of PGL2(F ). We say that an element γ of Γ is primitive if it generates its own centralizer in Γ. Likewise, a conjugacy class of Γ is primitive if its elements are primitive. Furthermore, for a conjugacy class C of Γ, we define the number deg C as the minimum value of ordπ(det g), when g is a matrix with entries in OF that represents an element of C. Ihara defined the following Zeta function
−1 Y deg P Z(XΓ, u) = 1 − u , P where P runs over all primitive conjugacy classes of Γ. He also proved that the Ihara zeta function is a rational function.
Theorem (Ihara [Iha66]). Let Γ be a discrete, cocompact, torsion-free subgroup of
PGL2(F ). Then, the Ihara zeta function of XΓ = Γ\ PGL2(F )/ PGL2(OF ) satisfies
(1 − u2)χ(XΓ) Z(XΓ, u) = 2 , det(I − AΓu + qu ) where AΓ is a Hecke operator on the space of functions on Γ\ PGL2(F )/ PGL2(OF ), and
χ(XΓ) is the Euler characteristic of XΓ.
Observe that rationality is a feature that Ihara zeta functions share with the Zeta
4 functions of algebraic varieties over finite fields, as is known from the Weil conjectures, alluded to before.
Serre [Ser80] realized that the double-quotient Γ\ PGL2(F )/ PGL2(OF ) may be in- terpreted as a finite (q + 1)-regular graph, and generalized Ihara’s zeta function to any finite regular graph H, using the fact that primitive conjugacy classes of Γ correspond bijectively to equivalence classes of primitive geodesic cycles, called primes of H, of the same length. Bass [Bas92] extended Ihara’s Theorem to irregular graphs. In [Has89], Hashimoto also showed that the Ihara zeta function, as suggested by Serre, is a rational function by relating it with the directed edge adjacency matrix.
Theorem (Hashimoto [Has89]). Let H be a finite graph with directed edge adjacency T . Then, the Ihara zeta function of H satisfies
1 Z(H, u) = . det(I − T u)
In [Has92], Hashimoto showed a version of the Prime Geodesic Theorem counting the number of primes of a finite graph.
Theorem (Prime Geodesic Theorem for graphs, Hashimoto [Has92]). Let H be a con- nected graph with directed edge adjacency matrix T . Let λ be the largest eigenvalue, in absolute value, of T and let δH be the greatest common divisor of the lengths of the geodesic cycles of H. Then,
λδH l |{[C]: C is a primitive geodesic cycle of H with length δ l}| ∼ , (1.1) H l as l → ∞ .
1.2 Ramanujan graphs
It is known that the expansion properties of a regular graph are related to the size of the gap between the two largest eigenvalues of its vertex adjacency matrix. As a result, this gives a relation between the expansion properties of a graph and the poles of its Ihara zeta function. This fact has applications to the construction of expanders, that is, families of graphs where the distance between any two vertices is, on average, small. Examples of these are given by Ramanujan graphs, that is, (q + 1)-regular graphs for which all √ nontrivial eigenvalues satisfy |λ| ≤ 2 q. This class of graphs was first mentioned in papers of Lubotzky-Phillips-Sarnak [LPS88] and Margulis [Mar88] that, independently, constructed families of (q + 1)-regular Ramanujan graphs, with q prime, using Hecke
5 operators on quaternion algebras. In their paper, the authors verified that all nontrivial √ eigenvalues of their examples satisfy |λ| ≤ 2 q using the Ramanujan-Petersson conjecture, proved by Deligne [Del74]. This is equivalent to the Zeta function attached to the graph satisfying the Riemann Hypothesis. A theorem of Alon-Boppana [LPS88] shows that this bound cannot be smaller for (q + 1)-regular graphs with arbitrarily many vertices. Lastly, we observe that if H is a (q + 1)-regular Ramanujan graph, then λ = q, in the equation (1.1). Furthermore, we have a good control of the error term of the estimate in δH l λ 2 (1.1), which is given by O l .
1.3 Primes and Zeta functions for finite quotients of buildings
Graphs are 1-dimensional simplicial complexes. So, it is natural to ask the same questions for higher dimensional complexes. In simplicial complexes with dimension greater than 1 we may construct paths over adjacent cells with dimension 2. We call such paths galleries. Recall that in Serre’s interpretation of Ihara’s zeta function, he uses graphs given by
finite quotients of the building associated to PGL2(F ). Here we study 2-dimensional complexes, that we will denote by XΓ, covered by the buildings associated to PGL3(F ) or PGSp4(F ). In Deitmar-Hoffman [DH06] and Kang-Li-Wang [KLW10], the authors study thor- oughly finite quotients, XΓ, of the 2-dimensional building associated to PGL3(F ), by discrete, torsion-free, cocompact subgroups Γ of PGL3(F ). They also study the Zeta functions counting the number of closed geodesic cycles and galleries in XΓ. There are two types for geodesic cycles and galleries, respectively. Then, in [KL14], Kang-Li prove formulas for the Zeta functions associated to the closed geodesic cycles of types 1 and 2 of XΓ, and denoted by Z1,1(XΓ, u) and Z1,2(XΓ, u), respectively. They also give formulas for these Zeta functions in terms of linear operators. If we let AE,Γ be the adjacency operator on type 1 edges, then
1 1 Z1,1(XΓ, u) = , and Z1,2(XΓ, u) = T 2 . det(I − AE,Γu) det(I − (AE,Γ) u )
T Here, (AE,Γ) is the transpose of AE,Γ. A similar formula is given for the Zeta function associated to the closed geodesic galleries of type 1. If we let AB,Γ be the adjacency operator on type 1 chambers, then
1 Z2,1(XΓ, u) = . det(I − AB,Γu)
6 Furthermore, they show an identity relating the various adjacency operators acting on the different i-skeletons of XΓ, analogous to Ihara’s Theorem for graphs.
Theorem (Kang-Li [KL14]). The adjacency operators of XΓ satisfy the identity
3 χ(X ) (1 − u ) Γ det(I + AB,Γu) 2 3 3 = T 2 , det(I − A1u + A2u − q u I) det(I − AE,Γu) det(I − (AE,Γ) u ) where χ(XΓ) is the Euler characteristic of XΓ, and A1 and A2 are Hecke operators 2 associated to the vertices of XΓ, acting on L (Γ\ PGL3(F )/ PGL3(OF )).
In Fang-Li-Wang [FLW13], the authors study another kind of 2-dimensional complexes, this time given by finite quotients, XΓ, of the building associated to PGSp4(F ), by discrete, torsion-free, cocompact subgroups Γ of PGSp4(F ). They prove formulas for the Zeta functions attached to XΓ in terms of adjacency operators. Like the case of PGL3(F ), there are two types for paths of directed edges and two types for paths of directed chambers. The edge zeta functions of types 1 and 2 are reciprocal of the characteristic polynomials of adjacency matrices AP1,Γ (resp. AP2,Γ) of type 1 (resp. type 2) edges:
1 1 Z1,1(XΓ, u) = , and Z1,2(XΓ, u) = , det(I − AP1,Γu) det(I − AP2,Γu) and the Zeta function attached to the closed geodesic galleries of chambers of spin type satisfies 1 Z2,spin(XΓ, u) = , det(I − AI1,Γu) where AI1,Γ is the adjacency matrix of chambers of spin type. Lastly, they show an identity relating the adjacency operators on different i-skeletons
2 χ(XΓ) 2 2 2Np−Nns (1 − u ) (1 − q u ) det(I − AI1,Γu) 2 3 2 6 4 = 2 , det(I − A1u + qA2u − q A1u + q Iu ) det(I − AP1,Γu) det(I − AP2,Γu ) where χ(XΓ) is the Euler characteristic of XΓ, 2Np is the number of special vertices in
XΓ, Nns is the number of nonspecial vertices in XΓ, and A1 and A2 are Hecke operators 2 on L (Γ\ PGSp4(F )/ PGSp4(OF )). More recently, Kang-Li-Wang [KLW] defined Zeta functions attached to the closed geodesic galleries of chambers of standard type and obtained identities relating the aforementioned with other Zeta functions of the building associated to PGSp4(F ).
7 1.4 Chebotarev Density Theorem
1.4.1 In Number Theory
Whenever L/E is a finite Galois extension of number fields, the group of automorphisms of L that fix every element of E is called the Galois group of the extension, and is denoted by Gal(L/E). As the rings of integers, OL in L and OE in E, are Dedekind domains, we know that each ideal I of OL or OE admits a decomposition into prime ideals of OL or
OE, respectively. In particular, if p is a prime ideal of OE, unramified in L, then pOL admits a decomposition
pOL = P1 ··· Pr into prime ideals of OL. In this case, we say that the prime ideals Pj, j = 1, . . . , r lie over p.
For each prime p in OE, we associate a specific conjugacy class Frobp of Gal(L/E), called the Frobenius conjugacy class at p. It consists of the elements of Gal(L/E) that induce the map x 7→ xq on OL/P, for some prime ideal P of OL that lies over p and with q = |OE/p|. Denote pr by SC (E) the set of unramified prime ideals p of OE such that Frobp = C. We can now state the famous Chebotarev Density Theorem.
Theorem (Chebotarev Density Theorem). Let L/E be a finite Galois extension of pr number fields with Galois group G. Then, for any conjugacy class C of G, the set SC (E) has natural density |C|/|G|.
1.4.2 For Riemann surfaces
In his thesis (cf. [Sar80]), Sarnak also studied a version of the Chebotarev Density Theorem for compact Riemann surfaces covered by the hyperbolic upper half-plane. Let 0 0 Γ and Γ be two discrete, torsion-free, cocompact subgroups of PSL2(R) such that Γ is a normal subgroup of Γ of finite index. This inclusion induces a finite covering map from 0 XΓ0 = Γ \H to XΓ = Γ\H, denoted by ϕ : XΓ0 → XΓ. If p is a primitive closed geodesic cycle in XΓ, then there are finitely many primitive closed geodesics P1,..., Pr, such that
ϕ(Pj) = p, j = 1, . . . , r.
We denote by G the group of automorphisms g of XΓ0 such that ϕ ◦ g = ϕ. One can ∼ 0 easily see that G = Γ/Γ , so that the cover ϕ is Galois. Each closed geodesic p in XΓ is associated to a conjugacy class in G that we denote by Frobp, which consists of the
8 elements in G that map the initial point of a lift of p in some Pj to the terminal point of pr that lift. Similarly to before, for a conjugacy class C of G, we denote by SC (Γ) the set of primitive closed geodesics p in XΓ such that Frobp = C. We now state the following theorem due to Sarnak.
Theorem (Sarnak [Sar80]). Let Γ0 and Γ be discrete, torsion-free, cocompact subgroups 0 of PSL2(R) such that Γ is a normal subgroup of Γ. Then, this inclusion induces a Galois ∼ 0 pr covering ϕ : XΓ0 → XΓ with group of automorphisms G = Γ/Γ , and the set SC (Γ) has natural density |C|/|G| for any conjugacy class C of G.
1.4.3 For graphs
We now present the case for coverings of graphs. Such coverings were studied thoroughly in Stark-Terras [ST00]. Let ϕ : X → Y be an unramified covering of finite graphs with d sheets. We further assume that the covering is normal, meaning that the group of automorphisms of ϕ acts transitively on each fiber, hence it has cardinality d. Call this group the Galois group G. We remark that Hashimoto [Has90] gives some insight to the case of ramified graph coverings. For each primitive geodesic cycle p in Y there are a number of primitive geodesic cycles P1,..., Pr in X, such that ϕ(Pj) = p, j = 1, . . . , r. We say that each prime
Pj lies over p. This was shown in the work of Terras [Ter11], Somodi [Som15], and Huang-Li [HL]. Furthermore, to each prime p in Y we associate a conjugacy class of G, that we denote by Frobp, in the same way as in section 1.4.2. Given a conjugacy class C of G, denote by pr SC (Y ) the set of primes p such that Frobp = C.
Theorem (Terras [Ter11]). Let ϕ : X → Y be a d-sheeted normal unramified covering of pr finite graphs with Galois group G. Then, for each conjugacy class C of G, the set SC (Y ) has Dirichlet density |C|/|G|.
Hashimoto [Has90] considered the above result in natural density. However, his argument was incomplete. Later, Huang-Li [HL] gave a necessary and sufficient condition for which the Chebotarev Density Theorem holds for natural density. To state their result, denote by δX and δY the greatest common divisors of all lengths of primes in X and Y , respectively.
pr Theorem (Huang-Li [HL]). In the same context as in the previous theorem, SC (Y ) has natural density if and only if δX = δY , in which case its value agrees with the Dirichlet density.
9 1.5 Main results
As before, let F be a nonarchimedean local field with uniformizer π, ring of integers
OF and q elements in its residue field. Our objective is to study versions of the Prime Geodesic Theorem and Chebotarev Density Theorem for equivalence classes of primitive closed geodesic cycles and galleries of given type for finite quotients of PGL3(F ) and
PGSp4(F ). Note that the asymptotic distribution of the 1 and 2-dimensional primes of given type is determined by the largest eigenvalues, in absolute value, of the corresponding adjacency operator. Moreover, as each cell in these buildings has the same number of adjacent cells, it is straightforward to determine the largest eigenvalue in absolute value. However, the asymptotic distribution depends on the multiplicity of such eigenvalues, which is not trivial to determine. The first main result of this dissertation delves with this question and gives a version of the Prime Geodesic Theorem for finite quotients of the buildings associated to PGL3(F ) and PGSp4(F ).
Theorem 1 (Prime Geodesic Theorem for 2-dimensional complexes). Let G be either the group PGL3(F ) or PGSp4(F ) and X the associated building. Let Γ be a discrete, cocompact, torsion-free subgroup of G, such that ordπ(det Γ) is divisible by 3 or 4, respec- tively. Given the dimension i and the type θ, denote by δ the greatest common divisor of the lengths of the i-dimensional primes with type θ. Then,
ηqκδl |{p | p is a type θ prime of dimension i in X with length δl}| ∼ , (1.2) Γ l as l → ∞, and
η qκδl |{p | p is a type θ prime of dimension i in XΓ with length < δl}| ∼ , qκδ − 1 l as l → ∞, where the value of each variable is given in the table below.
Group G PGL3(F ) PGSp4(F ) Dimension i 1 2 1 2 Type θ 1 1 2 spin standard η 1 1 2 1 2 δ 3 2 1 2 1 κ 2 1 3 4 2 3
For G = PGL3(F ), we can make Theorem 1 more precise when XΓ is Ramanujan. Ramanujan complexes were introduced in [Li04] and [LSV05b], and are the simplicial
10 complexes given by finite quotients of the building associated to PGLn(F ) whose spectra are contained in that of their universal cover. In particular, when n = 2, these are the (q + 1)-regular Ramanujan graphs. Furthermore, Kang-Li-Wang [KLW10] (Theorem 2) give equivalent conditions for Ramanujan complexes, in terms of the poles of the associated Zeta functions. Using that result, we can verify that the error term of (1.2) is κδl q 2 O l . 0 Now, let Γ and Γ be discrete, cocompact, torsion-free subgroups of PGL3(F ) or 0 PGSp4(F ) such that Γ is a normal subgroup of Γ of finite index. We further assume 0 that ordπ(det Γ ) and ordπ(det Γ) are divisible by 3 or 4, depending on whether they are subgroups of PGL3(F ) or PGSp4(F ), respectively. We denote by XΓ0 and XΓ the 0 quotients of the building associated to either PGL3(F ) or PGSp4(F ) by the groups Γ and Γ, respectively. The inclusion of subgroups induces a covering map ϕ : XΓ0 → XΓ with automorphism group Γ/Γ0. Similarly to what we did in the previous sections, for each i-dimensional prime p with type θ in XΓ we associate a conjugacy class from the 0 pr,θ group Γ/Γ , denoted by Frobp. We denote by Si,C (Γ) the set of i-dimensional primes p with type θ such that Frobp = C. We then have the following result.
pr,θ 0 Theorem 2. With notation as above, the set Si,C (Γ) has natural density |C|/|Γ/Γ |.
In particular, we remark that, unlike the graph case, we do not distinguish between the cases of Dirichlet or natural densities, since we know that the greatest common divisor of the lengths of primes with given dimension i and type θ is the same for both quotients
XΓ and XΓ0 . Hence the Chebotarev Density Theorem holds for the stronger natural density. This dissertation is organized as follows. Our main results are Theorems 1 and 2, that we further divide into Theorems 11, 12, 14, 15, 19, 20, 22, 23, 26, 27, 30 and 31. In chapter 2, we introduce the necessary concepts to understand the structure of simplicial complexes and multicomplexes, and we prove some preliminary results that we use multiple times in chapters 3 and 4, in various situations. Chapters 3 and 4 have a similar structure, and both begin with an explanation of the structure of the buildings attached to PGL3(F ) and PGSp4(F ), respectively. Then, we state the main results of the chapter, and give proofs for them in the end.
11 Chapter 2 | Preliminaries
2.1 Graphs
A graph is the classic abstract representation of a network. We define a directed graph −→ X as a set of vertices, V , together with a set of directed edges given by a subset E of −→ V × V − {(v, v): v ∈ V }. For each e = (u, v) ∈ E , we write e−1 = (v, u) to denote the edge with opposite direction from e, and we call an unordered pair of vertices {u, v} undirected edge. The set of unordered edges will be denoted by E. We will often refer to these just by edges, whenever it is clear from the context whether they are directed or undirected edges.
A path is a sequence of adjacent edges C = (e1, e2, . . . , el) where the terminal vertex of ei coincides with the initial vertex of ei+1, i = 1, . . . , l − 1. The length of C is l(C) = l.
Moreover, if the initial vertex of e1 is the same as the terminal vertex of el, we say that C is a closed path or that it is a cycle, and we denote by [C] the equivalence class containing 0 0 0 all cyclic rotations of C. If C = (e1, . . . , el) and C = (e1, . . . , el0 ) are two paths in X, 0 such that the terminal vertex of el is the same as the initial vertex of e1, we define their 0 0 0 concatenation as the path CC = (e1, . . . , el, e1, . . . , el0 ). We say that a cycle C is primitive if there does not exist another cycle C0 so that k C0 = C, for some integer k ≥ 2, and in this case [C] is also called primitive. Whenever −1 every pair of consecutives edges in C = (e1, . . . , el) satisfies ei 6= ei+1 we say that C is −1 backtrackless. Additionally, if e1 =6 el, we say that C is tailless. If V is finite and has size n, we define the adjacency matrix of X as the n × n matrix
AX = A = (auv) with indices in V , whose entries are given by ( −→ 1 , if (u, v) ∈ E auv = −→ . 0 , if (u, v) 6∈ E
12 −→ Moreover, if E is finite and has size 2m, we define the edge adjacency matrix of X as 0 −1 the 2m × 2m matrix TX = T = (tee0 ), where tee0 = 1 if e =6 e and the terminal vertex 0 of e coincides with the initial vertex of e , and tee0 = 0 otherwise. We wish to introduce a useful generalization of graphs. Multigraphs are similar to graphs except that they can have parallel edges. As a consequence, in this case, edges are not uniquely described by their endpoints. We define a multigraph Xe over a graph X as a set of vertices, V , together with a pair (E, m), where m : E → N is the multiplicity function defined on the undirected edges of X. The pairs (e, i) with e ∈ E and 1 ≤ i ≤ m(e) are the multiedges of Xe, and they form the set Ee. We also denote by ι : Xe → X the natural projection map that satisfies
ι(v) = v , v ∈ V and ι(e, i) = e , (e, i) ∈ E.e
2.2 Simplicial complexes and multicomplexes
Simplicial complexes are a generalization of graphs for higher dimensions. Again, use V to denote a set of vertices. We define a simplicial complex X with vertices in V as a collection of subsets of V , called cells, with the extra condition that if a = {v0, v1, . . . , vn} is a cell of X and b ⊂ a, then b is also a cell of X . The dimension of a cell a = {v0, v1, . . . , vn} is dim a = |a| − 1 = n. We also define the boundary of a cell a = {v0, . . . , vn}, denoted by ∂a, as the set of cells that result from removing a single vertex from a. For each i ∈ N we denote by Xi the i-skeleton of X, that is, the set of cells with dimension i in X or i-cells, for short. Finally, we define directed cells as the cells of X equipped with an orientation, that is, a tuple where each of its vertices appears exactly once. We present a further generalization of the concept of simplicial complex. Multi- complexes are to simplicial complexes like multigraphs are to graphs. This means that each cell is not uniquely determined by the vertices that it contains. Here we show the definition from [LLR18].
Definition (Multicomplex). Let X be a simplicial complex with set of vertices V , and m : X → N be a function that we will call multiplicity function. We define the set of pairs Xm = {(a, r): a ∈ X, r ∈ {1,..., m(a)}}, whose elements are multicells, and the forgetful map ι : Xm → X, by ι(a) = a, whenever a = (a, r). We also assume that m(v) = 1 if v ∈ V , and we define the dimension of a multicell as dim a = |ι(a)| − 1.
A gluing map g : {(a, σ) ∈ Xm × X : σ ∈ ∂ι(a)} → Xm is a map that satisfies (ι ◦ g)(a, σ) = σ for (a, σ) in the domain of g. In particular, we will see that g tells us how the multicells of a given dimension are glued together. We define the multiboundary
13 of a cell as the following set ( {g(a, σ): σ ∈ ∂ι(a)} if dim a ≥ 1 ∂ma = . ∅ if dim a = 0
The multicells obtained by taking multiboundaries of a multicell, one or more times, are called submulticells. Finally, we say that the triple Xe = (X, m, g) is a multicomplex if it satisfies all the conditions above and the following consistency property: if a = (a, r) is a multicell with two submulticells b = (b, r) and b0 = (b0, r0) such that dim b = dim b0 and the intersection of their projections onto X is ρ = σ∩σ0 ∈ Xdim b−1, then g(b, ρ) = g(b0, ρ). For a multicomplex Xe, we denote by Xei the set of multicells with dimension i or i-multicells, for short. We will call directed multicell to any multicell equipped with an orientation, that is, a tuple where each of its vertices appears exactly once.
2.3 Primes in finite quotients of X
Let X be an infinite simply connected simplicial complex. The neighborhood of a cell a in X consists of the cells b, with the same dimension as a, such that a ∩ b is a subcell of a and b with dimension dim a − 1. In the sequel, the central object of our study will be backtrackless and tailless paths in Xi, i = 1, 2. However, we want to focus on paths that minimize the distance between cells, that is, geodesic paths. To help us in that pursuit, we will use directed cells and we introduce the concept of out-neighbors of a directed cell. −→ Furthermore, we shall denote by Xi the set of directed cells contained in Xi. The out-neighbors of a directed cell c are the directed cells with the same dimension, that share a given face with c. The orientation of c determines which of its faces it has in common with the out-neighbors and the orientation of the out-neighbors. For each directed cell c, we will also consider a subset R(c) of its out-neighbors, that we will call proper out-neighbors (R stands for “right multiplication”, which should become natural from the concrete examples considered later on). To give an idea, these adjacent cells give rise to straight lines in the apartments of buildings. We will describe these in more detail in each of the cases considered in the next chapters.
A geodesic path of i-cells is a sequence of directed i-cells C = (c1, c2, . . . , cl), such that cj+1 is a proper out-neighbor of cj for any 1 ≤ j ≤ l − 1, or equivalently, cj+1 ∈ R(cj), and its length is l(C) = l. We call a path of 1-cells path and a path of 2-cells gallery.
Whenever C is closed, that is c1 ∈ R(cl), we denote by [C] the equivalence class of all rotation shifts of C. If such C is composed of 1-cells then we say that C is a geodesic cycle, and if it is composed of 2-cells we say that C is a closed geodesic gallery. Furthermore,
14 we say that [C] is a 1 or 2-dimensional prime. We shall study different kinds of geodesic paths, that may be characterized by their shape or direction and the dimension of their directed cells. Likewise, we shall assign a type θ to each directed cell or variations of it, so that each geodesic contains cells of a given dimension and type. We denote by Xi,θ the set of directed cells with dimension i and type θ in X. We assume that a directed cell and its out-neighbors have the same type. We denote by Aut(X) the group of automorphisms of X and let Γ be a subgroup of Aut(X), such that Γ has no torsion, the quotient space Γ\X is finite, the stabilizer of any vertex v of X is trivial, and Γv does not contain two adjacent vertices for 0 any v ∈ X . In this situation, XΓ = Γ\X has a structure of simplicial complex or multicomplex induced by the orbits of cells in X under the action of Γ. In more detail, 0 XΓ is defined over the simplicial complex ι(XΓ), which has Γ\X as set of vertices and whose cells are given by the subsets {Γv0,..., Γvn} such that {v0, . . . , vn} is a cell of X.
Then, XΓ is a multicomplex over ι(XΓ) where the multiplicity function at a given cell
{Γv0,..., Γvn} is equal to the number of orbits of cells Γ{w0, . . . , wn} of X such that
{Γw0,..., Γwn} = {Γv0,..., Γvn}. The boundary of a multicell Γ{v0, . . . , vn} is given −→ i,θ by the orbits of the cells in ∂{v , . . . , v }. Finally, we define the sets Xi , Xi and X 0 n −→ Γ Γ Γ similarly to before, by taking the orbits of Xi, Xi and Xi,θ, respectively, under Γ. The proper out-neighbors and geodesic paths in XΓ are given by the projection of proper out-neighbors or geodesic paths in X under the projection map ϕ : X → XΓ.
Additionally, in order to guarantee that each path of i-multicells of XΓ has a unique lift, we assume that Γ satisfies the following condition.
(C) for any cells σ, a and b of X, with σ ∈ ∂a and σ ∈ ∂b, then Γ(a, σ) = Γ(b, σ) if and only if (a, σ) = (b, σ).
In this context, Γ is isomorphic to the fundamental group of XΓ. θ θ 1,θ 2,θ Let S1,Γ and S2,Γ be sets of representatives of the orbits in Γ\X and Γ\X , respectively. Consider C a closed geodesic path of i-multicells with i = 1 or 2, in XΓ. C θ has a unique lift to X starting from an i-multicell in Si,Γ, like the following
s0 → γ1s1 → γ2s2 → · · · → γl−1sl−1 → γlsl = γls0 ,
θ where sj ∈ Si,Γ, j = 0, . . . , l, and we take s0 = sl, because C is closed. Thus, the initial and terminal cells of the lift of C differ by the action of γl ∈ Γ. Note that if we replace
15 the initial cell by γs0 ∈ Γs0, the lift of C becomes
−1 −1 −1 −1 γs0 → (γ · γ1 · γ )γs1 → · · · → (γ · γl−1 · γ )γsl−1 → (γ · γl · γ )γsl = (γ · γl · γ )γs0 , so the initial and terminal cells differ by the action of a conjugate of γl. On the other hand, if we apply a rotation shift to C, the resulting path or gallery has the following lift
γ1s1 → γ2s2 → · · · → γl−1sl−1 → γlsl = γls0 → γl(γ1s1) .
Consequently, the conjugacy class [γl] is well-defined in terms of the equivalence class [C].
In this situation, we define the Frobenius conjugacy class of [C] as Frob[C] = [γl]. From here on, the equivalence classes, under cyclic rotation, of primitive closed geodesic cycles or galleries in XΓ of type θ will be called i-dimensional primes of type θ.
2.3.1 Zeta functions and L-functions
We will define counting functions of primes of given dimension and type, known as Zeta functions. These functions have many interesting features, and their analytic properties have consequences to the distribution of closed geodesic cycles and galleries in XΓ. We start by introducing adjacency operators in the i-skeleton of X, using the proper out-neighbors. We define the following operator on C(Xi,θ), the set of complex valued functions on Xi,θ, θ X 0 (Ai f)(c) = f(c ) . c→c0 proper
θ Then, we may define a similar operator, Ai,Γ, on the complex valued functions on directed θ cells of the finite quotient XΓ, by taking the restriction of Ai to the finite dimensional space θ i,θ i,θ Vi,Γ = {f : X → C | f(γc) = f(c), ∀c ∈ X , γ ∈ Γ}.
A Zeta function is given by a product over i-dimensional primes of type θ. Given the set i,θ of primes in XΓ , the corresponding Zeta function is given by
−1 Y l(p) Zi,θ(XΓ, u) = 1 − u . p type θ prime of dimension i
In the next section, we will prove that this Zeta function is a rational function in u.
16 Theorem 3. The type θ Zeta function on the i-skeleton of XΓ satisfies
1 Zi,θ(XΓ, u) = θ . (2.1) det(I − Ai,Γu)
Let ρ be a d-dimensional unitary representation of Γ, with representation space Vρ. i,θ For dim X ≥ i ≥ 1, we define the following space of Vρ-valued functions in X
θ i,θ i,θ Vi,ρ = {f : X → Vρ : f(γc) = ρ(γ)f(c), ∀c ∈ X , γ ∈ Γ} ,
i,θ which is a subspace of C(X ) ⊗ Vρ equipped with the following Γ action
−1 i,θ i,θ (γf)(c) = ρ(γ)f(γ c) , ∀c ∈ X , f ∈ C(X ) ⊗ Vρ . (2.2)
θ i,θ Using the action of Γ we may also define Vi,ρ as the subspace of C(X ) ⊗ Vρ fixed by Γ, that is, θ i,θ Γ Vi,ρ = (C(X ) ⊗ Vρ) .
Denote by idρ : Vρ → Vρ the identity endomorphism on Vρ. Then, we introduce the θ θ θ adjacency operator Ai,ρ as the restriction of Ai ⊗ idρ to the subspace Vi,ρ. This definition θ θ is equivalent to saying that Ai,ρ is an operator that sends f ∈ Vi,ρ to
X 0 (Ai,ρf)(c) = f(c ) . c→c0 proper
We define the L-function associated to ρ as
−1 Y l(p) Li,θ(XΓ, ρ, u) = det I − ρ(Frobp)u . p type θ prime of dimension i
It turns out that this function is also a rational function, as shown below.
Theorem 4. The type θ L-function associated to XΓ and ρ satisfies
1 Li,θ(XΓ, ρ, u) = θ . (2.3) det(I − Ai,ρu)
Remark. When ρ = 1 is the trivial representation, this L-function becomes the type θ
Zeta function on the i-skeleton of XΓ.
Proof. The proof follows the same structure as the one for Theorem 1 in [Has89] and
17 Theorem 1.1 in [KL15]. We use the identity tr(log(I − A)) = log(det(I − A)). In order to prove the desired equality we apply this identity to both sides of (2.3). θ We first analyze the right hand side of (2.3). Consider Si,Γ = {s1, . . . , sn} a set of i,θ i,θ representatives of Γ\X and fix a basis v1, . . . , vd of Vρ. The functions fjk : X → Vρ, j = 1, . . . , d, k = 1, . . . , n, given by ( ρ(γ)vj if k = t fjk(γst) = , for γ ∈ Γ , 0 otherwise
θ θ l form a basis for Vi,ρ with size dn. Note that the operator (Ai,ρ) is a sum over geodesic paths of i-cells in X with length l + 1, like the following
st = st0 → γ1st1 → · · · → γlstl .
Denote by fk = a1f1k + ··· + adfdk an arbitrary linear combination of elements in the θ θ l basis of Vi,ρ that are zero outside Γsk. The components f•k of (Ai,ρ) fk are determined by
θ l X (Ai,ρ) fk (st) = fk(γst) st→···→γst proper, with length l X = ρ(γ)fk(st) , st→···→γst proper, with length l when t = k. As a result, we obtain the following lemma.
θ l Lemma 1. The trace of (Ai,ρ) may be given in terms of the i-dimensional primes of type θ in XΓ, as follows
θ l X X l/k tr (Ai,ρ) = k tr ρ(Frobp) . (2.4) k|l p type θ prime of dimension i l(p)=k
Using Lemma 1, we see that the logarithm of the right hand side of (2.3) can also be written as θ l l X (Ai,ρ) u − tr log(I − Aθ u) = tr i,ρ l l≥1
18 X 1 X X l/k l l = tr k tr ρ(Frobp) u , t = l k l≥1 k|l p type θ prime of dimension i l(p)=k t l(p)t X X tr ρ(Frobp) u = . t p type θ prime t≥1 of dimension i
On the other hand, taking the logarithm of the left hand side of (2.3) and expanding, we obtain
X l(p) X l(p) − log det(I − ρ(Frobp)u ) = − tr log(I − ρ(Frobp)u ) p type θ prime p type θ prime of dimension i of dimension i k l(p)k X X ρ(Frobp) u = tr k p type θ prime k≥1 of dimension i k l(p)k X X tr ρ(Frobp) u = . k p type θ prime k≥1 of dimension i
This concludes the proof.
2.3.2 Distribution of primes of given dimension and type
Recall that in the previous section we saw that the Zeta functions of XΓ may be given by rational functions of the form
1 Zi,θ(XΓ, u) = θ . det(I − Ai,Γu)
θ θ i,θ Let λi,Γ = max{|λ| : λ ∈ C is an eigenvalue of Ai,Γ} . Let n = ni,θ = |XΓ | and let θ α1, . . . , αt, β1, . . . , βn−t be all the eigenvalues of Ai,Γ counted with multiplicity, such that θ θ |αr| = λi,Γ, r = 1, . . . , t, and |βj| < λi,Γ, j = 1, . . . , n − t . We can write,
1 Z X , u . i,θ( Γ ) = Qt Qn−t (2.5) r=1(1 − αru) j=1(1 − βju)
19 Taking the logarithmic derivative of the left hand side of (2.5), we obtain
d −u log Z (X , u) du i,θ Γ d Y −1 X ul(p) = −u log 1 − ul(p) = − l(p) du − ul(p) p type θ prime p type θ prime 1 of dimension i of dimension i
X X l(p)k X X l = − l(p)u = − l(p) u p type θ prime k≥1 l≥1 p type θ prime of dimension i of dimension i l(p) | l
And calculating the Möbius inversion for the coefficients in the last expression, we have
X X − l(p)ul . (2.6) l≥1 p type θ prime of dimension i l(p)=l
On the other hand, we expect to get an equivalent expression if we apply the same operations to the right hand side of (2.5). Taking the logarithmic derivative, we obtain
! t n−t d 1 X αru X βju −u log = − − du Qt Qn−t 1 − α u 1 − β u r=1(1 − αru) j=1(1 − βju) r=1 r j=1 j
t n−t t n−t X X l X X l X X l X l l = − (αru) − (βju) = − αr + βj u . r=1 l≥1 j=1 l≥1 l≥1 r=1 j=1
And calculating the Möbius inversion for the coefficients in the last expression, we have
t n−t X X l X X − µ αk+ βk ul k r j l≥1 k|l r=1 j=1 t n−t t n−t X X l X l X l X k X k l = − α + β + µ α + β u r j k r j l≥1 r=1 j=1 k|l r=1 j=1 k