Aspects of Automorphic Induction
DISSERTATION
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By Edward Michael Belfanti, Jr., B.A. Graduate Program in Mathematics
The Ohio State University 2018
Dissertation Committee: James W. Cogdell, Advisor
Sachin Gautam
Roman Holowinsky c Copyright by
Edward Michael Belfanti, Jr.
2018 Abstract
Langlands’ functoriality conjectures predict how automorphic representations of different groups are related to one another. Automorphic induction is a basic case of functoriality motivated by Galois theory. Let F be a local field of characteristic 0. The local Langlands correspondence for GL(N) states that there is a bijection between N-dimensional, complex
0 representations of the Weil-Deligne group WF and irreducible, admissible representations of GL(N,F ). Given an operation on Weil group representations, one can ask what the corresponding operation is for representations of GL(N,F ). Automorphic induction is the operation on representations of GL(N,F ) corresponding to induction of representations of the
Weil group. Given a cyclic extension E ⊃ F of degree D, automorphic induction is a mapping of representations of GL(M,E) to representations of GL(MD,F ). Once automorphic induction has been established for local fields, one can ask if the operation applied at each local component of an automorphic representation produces another automorphic representation.
The automorphic induction problem was first considered in detail in [Kaz84]. Kazhdan proved the local automorphic induction map exists in the of M = 1. The next major result was [AC89], where it was shown that a global automorphic induction operation exists for prime degree extensions. The local theory was completed in [HH95]; Henniart and Herb showed that the local automorphic induction map exists for arbitrary M and cyclic extensions of arbitrary degree. Moreover, Henniart showed in [Hen01] that the local automorphic induction map constructed in [HH95] is consistent with induction of Weil group representations and the local Langlands correspondence. Finally, Henniart extended the results of [AC89] in [Hen12] to cyclic extensions of any degree and verified that the resulting
ii mapping of automorphic representations is consistent with the local lifting of [HH95]. All of these results rely on some version of the trace formula and powerful theorems about
L-functions for GL(N).
The goal of this thesis is to give a different proof of local and global automorphic induction when M = 1 and to emphasize some different aspects of the theory. As with the results previously mentioned, the main technical tool is a trace formula, specifically the full global trace formula of Arthur. The proof relies more on the trace formula and less on
L-functions than previous proofs.
Chapter 1 consists of background and preliminaries needed to state the main theorems on local and global automorphic induction. We also discuss various results that are needed to state Arthur’s trace formula.
In Chapter 2, we give a statement of the relevant trace formulas. We also discuss some computations in general rank that could be used in a generalization of our techniques to the case of M > 1.
The main local and global theorems are established in Chapter 3. As with most trace formula arguments, the proof boils down to an application of linear independence of characters.
iii For my family, friends, and all those who helped
along the way
iv Acknowledgments
A dissertation is the work of many people, some of whom deserve special thanks.
Melissa Grasso has been with me through not one, but two final semesters during which
I was convinced I would never graduate. Her love, support and companionship over the past seven years, and through some particularly trying months, has helped me more than I can ever repay. Her own successes and diligent hard work provided the motivation I needed to see this project through to the end; I only hope that the completion of this degree justifies in some small way the innumerable sacrifices, big and small, she has made for me. With love, thank you.
My parents, Ed and Teresa, have provided boundless love, generosity and advice, not just these past seven years but the past twenty-nine. I feel fortunate every day for their support, and it is an honor to be their son and share any success I may find with them.
Thank you always.
My brothers, Nick and Dom, have provided the support and encouragement that their older brother is supposed to provide. Thank you both for never doubting me.
Thanks to Nana, Papa, and Grandma, for the love and support unique to grandparents.
Thanks to my Grandpa, who I know would be proud of me. Thanks to my aunts and uncles
Amy, Paul and Mary, and my cousins William, Charles and Kate, for a fun Christmas before getting back to the grind. I’ve received many kinds words from my aunts Nancy,
Mary, Donna and Julie, which were especially welcome over the last weekend of work on this document.
I would like to thank Linda James, John Grasso, Jennifer Grasso, Erika James and Kevin
Riopelle, for always making me feel welcome in Sudbury.
v Stephen Eigenmann deserves thanks for always helping me keep my priorities straight.
Thanks to Jack Marczak, with whom I started studying math in high school. Thanks also to Larry Marczak, without whose help I may never have mastered logarithms.
I would like to thank the group chat consisting of Matt Flagg, Eric Kim, Rob Monaco and Tomas Castella for providing needed humor over the past seven years and for keeping me apprised of the news and scores I missed while completing this work. Matt deserves special thanks for his commiseration and support throughout more than ten years of school.
John Lynch and I have been friends ever since we realized we were in so many classes together that it was getting weird for us not to talk. Carney forever.
I would like to thank my friends and former roommates Steph Wilker, Scott Jaffee and Matt Jaffee, who managed to tolerate me through my early years in grad school, and qualifying exams in particular.
Thanks to Catherine Glover and Lyman Gillispie for their friendship, and for getting me out of my office. A special thanks to Lyman, for always taking seriously my terrible opinions about books that I don’t understand.
I would like to thank my colleague Yilong Wang for interesting conversations, mathe- matical and otherwise, and for encouragement over the last few weeks of writing.
Dan Moore, PhD, has been a close friend ever since he got over my gum snapping. His technical support kept me sane at a very strenuous time during the completion of this document. Thanks for the many, let’s say bizarre and enlightening, conversations. We’ll always have that Culver’s on the outskirts of middle of nowhere Indiana. Dan could not have supported me without the support he received from Mariel Colman, who deserves special thanks for her patience with us.
Thanks to Solomon Friedberg, whose advice and encouragement when I was at Boston
College inspired me to go to graduate school, and study representation theory in particular.
Jim Cogdell, my advisor, deserves thanks for many hours of conversation about repre- sentation theory, as well as for frantically reading this thesis and providing feedback as I pushed every conceivable deadline. It is impossible to believe this document could exist without his wisdom and advice throughout the research and writing process. He deserves
vi special thanks for helping me through a difficult project and situation not of his design.
vii Vita
2007–2011 ...... Boston College B.A. in Mathematics 2011–Present ...... The Ohio State University Graduate Teaching Associate, Graduate Research Associate Fields of Study
Major Field: Mathematics
Studies in Automorphic Representation Theory: James W. Cogdell
viii Table of Contents
Page Abstract...... ii Dedication...... iv Acknowledgments...... v Vita...... viii
Chapters
1 Preliminaries ...... 1 1.1 L-groups...... 1 1.2 Twisted endoscopy for (GN , ω) ...... 11 1.3 Representations of GN ...... 22 1.4 A substitute for global parameters ...... 44 1.5 Langlands-Shelstad-Kottwitz transfer...... 64 1.6 Local intertwining operators...... 79 1.7 Statement of the main theorems ...... 99
2 Trace formulas ...... 108 2.1 The discrete part of the trace formula ...... 108 2.2 Stabilization...... 114 2.3 Contribution of a parameter ψ ...... 121 2.4 A preliminary comparison...... 128 2.5 The stable multiplicity formula...... 135 2.6 An elliptic computation ...... 146 2.7 Globalizing local representations ...... 153 2.8 Orthogonality relations...... 164
3 The case of M = 1...... 170 3.1 Stable multiplicity when M = 1...... 170 3.2 Orthogonality relations when M = 1...... 171 3.3 Spherical characters and weak lifting...... 173 3.4 Main theorems ...... 178
Bibliography ...... 191
ix Appendices
A Odds and ends ...... 201 A.1 Restriction of scalars and endoscopy ...... 201 A.2 The Langlands classification...... 205 A.3 Global induced representations ...... 208 A.4 More on the Local Langlands correspondence ...... 219
x Chapter 1 Preliminaries
We collect various results that are needed for the statement of the trace formula and the main theorems.
1.1 L-groups
In this section, we will introduce the L-groups used throughout. We will start with some basic definitions and then specialize to our case. We follow the discussion from [Art05, §26],
[Bor79, Ch. I] and [KS99, §§1,2].
1.1.1 Let F be either a local or global field of characteristic 0. Let G be a connected, reductive, linear algebraic group defined over F . Assume in addition that G is quasi-split and that the derived group of G is simply connected. These assumptions will simplify the discussion and include many classical situations. Throughout this work, a “general” group will refer to such a G. A Borel subgroup B ⊂ G and maximal torus T ⊂ B will often be referred to as a pair (B,T ). To each pair (B,T ), the theory of algebraic groups (over an algebraically closed field of characteristic 0) assigns a based root datum
∨ ∨ Ψ(B,T ) = (XT , ∆B,T ,XT , ∆B,T ).
See [Spr79, §2] for a more detailed summary. Here XT is the lattice of characters of T ,
∨ ∆B,T is the set of simple roots of T relative to B, X is the lattice of cocharacters of T and
∨ ∆B,T is the set of simple coroots of T relative to B. Any two pairs (B,T ) and (B1,T1) are
1 conjugate to one another. Moreover, any inner automorphism of G which takes (B,T ) to
(B1,T1) induces an isomorphism between Ψ(B,T ) and Ψ(B1,T1). Thus we obtain
Ψ(G) = (X, ∆,X∨, ∆∨), the canonical based root datum of G.
Unless stated otherwise, for now all homomorphisms of algebraic groups are defined over
F . Let Aut(G) denote the set of automorphisms of G. For g ∈ G, let Int(g) denote the automorphism of G given by
Int(g)(x) = gxg−1, x ∈ G.
Then Int(G) ⊂ Aut(G) will denote the image of the map
g 7→ Int(g): G → Aut(G).
The group Out(G) of outer automorphisms is defined to be the quotient
Out(G) = Aut(G)/ Int(G).
The datum Ψ(G) satisfies
Aut(Ψ(G)) ' Out(G).
Recall that a splitting of G is a pair (B,T ) together with a set {Xα : α ∈ ∆} of nonzero vectors in the associated root spaces gα ⊂ g, the Lie algebra of G. Suppose a group Γ acts on G by automorphisms. If the Γ-action preserves the splitting (B,T, {Xα}), we call it an L-action relative to (B,T, {Xα}). A Γ-action is called an L-action if it is an L-action relative to some splitting. For any splitting (B,T, {Xα}), there are isomorphisms
Aut(Ψ(G)) ' Out(G)
'{Automorphisms of G which preserve (B,T, {Xα})}.
See [Spr79, Proposition 2.13] for more details.
2 An action of ΓF = Gal(F /F ) on Ψ(G) can be obtained in the following way. Let
ψ∗ : G → G∗ be an isomorphism of G with a split group G∗ defined over F . Such an isomorphism is of course not necessarily defined over F . Then the map
∗−1 ∗ σ 7→ ψ ◦ σ(ψ ):ΓF → Out(G)
defines an action of ΓF on Ψ(G) denoted by ρG; it does not depend on ψ∗. By fixing a splitting of G, we obtain an L-action of ΓF on G also denoted by ρG. The datum
Ψ(G)∨ = (X∨, ∆∨,X, ∆)
∨ is the dual root datum of G. The group ΓF acts on Ψ(G) via the dual action of ρG which we denote by ρ∨ . An L-group datum for G is a triple (G, ρ , η ) where G b Gb G
1. Gb is a connected reductive group over C,
2. ρ :Γ → Out(G) is an L-action of Γ on G and Gb F b F b
3.
∨ ηG : Ψ(G) → Ψ(Gb)
is a ΓF -equivariant bijection between canonical based root data. In other words, ηG intertwines the actions ρ∨ and ρ . G Gb
For each G, we fix a ΓF -splitting
spl(Gb) = (B,b T,b {Xbα∨ }) of Gb.AΓF -action on Ψ(Gb) then gives an L-action fixing spl(Gb). We define the semidirect product
L G = Gb o ΓF via the L-action ρ discussed above. Let W be the Weil group of F . See [Tat79] for the Gb F definition of WF and its basic properties. Via the natural map WF → ΓF , we obtain the 3 Weil form of the L-group
Gb o WF .
Let E be a finite extension of F over which G is split. Then the action ρ | is trivial. Gb WE Thus we may also consider the finite form of the L-group,
Gb o ΓE/F , for a finite extension E of F over which G is split.
We recall [Tat79, Proposition (1.6.1)].
Proposition. Let iv : F → F v be an F -homomorphism. For each finite extension E of F in F , let Ev = i(E)Fv be the induced completion of E. There exists a continuous homomorphism θv : WFv → WF such that the following diagrams are commutative.
WFv / ΓFv
θv iv WF / ΓF
E× ∼ W ab v / Ev
nv θv × × ∼ ab E \AE / WE
× The map iv is induced by the embedding iv : F → F v. The map nv sends a ∈ Ev to the class of the idele whose v-component is a and whose other components are 1. The map θv is
0 unique up to conjugation by the kernel WF of the map WF → ΓF .
L L Using any form of the L-group, we obtain an embedding Gv ,→ G which is trivial on
Gb. Indeed, the action of ΓFv on Gbv = Gb is the restriction of the action of ΓF , and likewise for WFv and WF . As in the proposition, the embedding is to be regarded as a map up to 0 WF -conjugacy.
1.1.2 Let P(GF ) be the parabolic subgroups of G defined over F and let P(G) be the parabolic subgroups of G defined over F . Let p(GF ) be the conjugacy classes of parabolic
4 subgroups in P(GF ). (Conjugacy relative to G(F ) and G(F ) is the same.) Let p(G) be the conjugacy classes of parabolic subgroups in P(G). Since G is quasi-split, p(GF ) is the same as the set of conjugacy classes of parabolic subgroups (defined over F ) that are stable under
ΓF . There is a canonical bijection between p(G) and subsets of ∆. Since G is quasi-split, p(GF ) corresponds to ΓF -stable subsets of ∆.
Let Pb ⊂ Gb be a parabolic subgroup and consider the normalizer NLG(Pb). A parabolic subgroup of LG is such a normalizer that meets every class in LG modulo Gb. In other words, for NLG(Pb) to be a parabolic subgroup, the map NLG(Pb) → WF must be surjective. Let p(LG) be the set of conjugacy classes of parabolic subgroups of LG. Then the correspondence
∨ ∨ ∆ ↔ ∆ induces a bijection between ΓF -stable subsets of ∆ and ∆ and thus a bijection
L p(GF ) ↔ p( G).
Let P ∈ P(GF ) and let M ⊂ P be a Levi subgroup defined over F . Let Bb be the Borel subgroup determined by the dual root datum Ψ(G)∨ = Ψ(Gb). Let LP ⊃ LB be the standard parabolic subgroup in the class associated to P . Then LM can be identified with a Levi subgroup of LP , namely the normalizer in LP of a Levi subgroup Mc ⊂ Pb.
1.1.3 We will let GN denote the general linear group GL(N) regarded as a matrix group defined over F . We fix the standard pair (BN ,TN ) consisting of the group BN of upper triangular matrices and the group TN of diagonal matrices in GN . Let
χi : TN → Gm, 1 ≤ i ≤ N, and
λi : Gm → TN , 1 ≤ i ≤ N, denote the standard characters and cocharacters of TN . More precisely,
χi(diag(t1, . . . , tN )) = ti and
λi(t) = diag(1, . . . , t, . . . , 1)
5 with t in the ith place. Then
N N ! M N−1 M N−1 Ψ(BN ,TN ) = Zχi, {χi − χi+1}i=1 , Zλi, {λi − λi+1}i=1 . i=1 i=1
Let GbN = GL(N, C). The pair (BbN , TbN ), the datum Ψ(BbN , TbN ) and the characters χbi and cocharacters λbi are all defined similarly. ∨ Since GN is split, the action of ΓF on Ψ(GN ) is trivial. Thus the dual action on Ψ(GN ) is also trivial. By taking the trivial action on Ψ(GbN ), we see that
η = λ 7→ χ : X∨ → X GN i bi b
is a ΓF -equivariant bijection. The triple (GbN , triv, ηGN ) is an L-group datum for GN .
1.1.4 We recall some notions about induced groups and restriction of scalars. All of this material (and additional details) is from [Bor79, §§I.4,I.5]. Let W be a group and let W 0 ⊂ W be a subgroup of finite index. Suppose W 0 acts on a group M by automorphisms. Then we let
W W 0 0 0 0 IndW 0 (M) = IW 0 (M) = {f : W → M : f(w w) = w · f(w), w ∈ W , w ∈ W }.
W The group W acts on IW 0 (M) by right translation:
(wf)(w1) = f(w1w).
For the moment we let F ⊂ E be a finite extension in F . Let GE be a connected reductive E-group and set H = RE/F GE. By definition H is the functor from F -algebras to sets defined by the property that if A is an F -algebra then
H(A) = GE(A ⊗F E).
See [PR94, §2.1.2] for more details, in particular for the fact that H is represented by a linear algebraic group defined over F , also denoted by H. We will only use the interpretation
σ of H as a linear algebraic group. Let σ be an automorphism of F . Let GE be the algebraic
6 group defined by applying σ to the polynomials that define GE. Then
Y H(F ) ' IΓF (G (F )) ' Gσ (F ). ΓE E E σ∈ΓE \ΓF
The group ΓF acts on the right hand side by permuting the factors.
∨ ∨ Fix a pair (BE,TE) in GE and let Ψ(BE,TE) = (XE, ∆E,XE, ∆E) be the associated root datum. Fix a pair (BH ,TH ) in H with associated based root datum Ψ(BH ,TH ) = (X , ∆ ,X∨ , ∆∨ ). Fix in addition an isomorphism ψ : H → Q Gσ . This iso- H H H H σ∈ΓE \ΓF E morphism is not generally defined over F . Then (Q Bσ , Q T σ) is a pair in σ∈ΓE \ΓF E σ∈ΓE \ΓF E Q Gσ . We may choose ψ to take (B ,T ) to this pair. We have σ∈ΓE \ΓF E H H
XH = {χE ◦ σ ◦ ψ : χ ∈ XE, σ ∈ ΓF },
∆H = {αE ◦ σ ◦ ψ : αE ∈ ∆E, σ ∈ ΓF },
∨ −1 ∨ XH = {ψ ◦ σ ◦ λE : λE ∈ XE, σ ∈ ΓF },
∨ −1 ∨ ∨ ∨ ∆H = {ψ ◦ σ ◦ αE : αE ∈ XE, σ ∈ ΓF }.
The group Γ acts on Q Gσ , and composition with σ in the sets above denotes F σ∈ΓE \ΓF E composition with this action. There are isomorphisms
X ' IΓF (X ) ΓE E and
X∨ ' IΓF X∨. ΓE E
Let ρE denote the action of ΓE on GbE, described in 1.1.1, used to form the L-group L GbE. Consider the group IΓF G ΓE bE obtained by inducing ρE. The action of ΓF will be denoted by ρF . It is isomorphic (as a group) to the product of [E : F ] copies of GbE. A choice of a pair (BbE, TbE) with associated
∨ ∨ ΓF root datum (Xb , ∆b , Xb , ∆b ) induces a pair and datum on IΓ GbE by taking TbE BbE ,TbE TbE BbE ,TbE E the same pair and the same datum on each factor. In fact, the character lattice can be
7 identified with IΓF X as above. Let η denote the Γ -equivariant bijection between X∨ ΓE bE E E E and X . Then η induces a Γ -equivariant bijection between IΓF (X∨) and IΓF (X ). Thus bE E F ΓE E ΓE bE (IΓF G , ρ , η ) is an L-group datum for H. ΓE bE F F
The group GbE embeds in Hb diagonally. This embedding fits into a commutative diagram
L {1} / GbE / GE / ΓE / {1}
L {1} / Hb / H / ΓF / {1}.
(The map ΓE → ΓF is the identity embedding.)
If JE ⊆ ∆E is ΓE-stable then
[ J = {αE ◦ σ : αE ∈ JE}
σ∈ΓE/F is ΓF -stable. The map JE 7→ J is a bijection from the ΓE-stable subsets of ∆E to the
ΓF -stable subsets of ∆. The map PE 7→ RE/F PE is a bijection from the set of parabolic subgroups of GE defined over E to the set of parabolic subgroups of H defined over F ; it induces a bijection between the conjugacy classes of parabolic subgroups. In addition there