J. Differential Geometry 95 (2013) 71-119: Received 5/25/2012

j.
differential geometry
95 (2013) 71-119
ANALYTIC TORSION AND L2 -TORSION OF

COMPACT LOCALLY SYMMETRIC MANIFOLDS
Werner Müller & Jonathan Pfaff
Abstract
In this paper we study the analytic torsion and the L2 -torsion
of compact locally symmetric manifolds. We consider the analytic
torsion with respect to representations of the fundamental group
which are obtained by restriction of irreducible representations of
the group of isometries of the underlying symmetric space. The
main purpose is to study the asymptotic behavior of the analytic
torsion with respect to sequences of representations associated to
rays of highest weights.
1. Introduction
Let G be a real, connected, linear semisimple Lie group with finite
center and of noncompact type. Let K ⊂ G be a maximal compact sub-
group. Then X e = G/K is a Riemannian symmetric space of the noncom-
pact type. Let Γ ⊂ G be a discrete, torsion-free, cocompact subgroup.
Then X = Γ\X e is a compact oriented locally symmetric manifold. Let
d = dim X. Let τ be a finite-dimensional irreducible representation of G
on a complex vector space Vτ . Denote by Eτ the flat vector bundle over
X associated to the representation τ |Γ of Γ. By [MtM, Lemma 3.1],
Eτ can be equipped with a distinguished Hermitian fiber metric, called
admissible. Let ∆p (τ ) be the Laplace operator acting on Eτ -valued p-
forms on X. Denote by ζp (s; τ ) the zeta function of ∆p (τ ) (see [Sh]).
Then the analytic torsion TX (τ ) ∈ R+ is defined by
d
1X d
(1.1) log TX (τ ) = (−1)p p ζp (s; τ )s=0
2 p=0 ds
(see [RS], [Mu2]). Since we have chosen distinguished metrics, we don’t

indicate the metric dependence of TX (τ ). We also consider the L2 -
(2)
torsion TX (τ ) which is defined as in [Lo], using the Γ-trace of the
heat operators on X.e
The main purpose of this paper is to study the asymptotic behavior
(2)
of TX (τ ) and TX (τ ) for certain sequences of representations τ of G.
Received 5/25/2012.
71
72 W. MÜLLER & J. PFAFF
This problem was first studied in [Mu3] in the context of hyperbolic

3-manifolds. The method used in this paper was based on the study of
the twisted Ruelle zeta function. In [MP] we have developed a different
and simpler method which we used to extend the results of [Mu3] to
compact hyperbolic manifolds of any dimension. In the present paper,
we generalize the results of the previous papers to arbitrary compact
locally symmetric spaces. Recently, Bismut, Ma, and Zhang [BMZ1],
[BMZ2] studied the asymptotic behavior of the analytic torsion by a dif-
ferent method and in the more general context of analytic torsion forms
on arbitrary compact manifolds. Furthermore, Bergeron and Venkatesh
[BV] studied the asymptotic behavior of the analytic torsion if the flat
bundle is kept fixed, but the discrete group varies in a tower {ΓN }N ∈N
of normal subgroups of finite index of Γ. They used this to study the
growth of the torsion subgroup in the cohomology of arithmetic groups.
In [MaM] the results of [Mu3] have been used to study the growth of
the torsion in the cohomology of arithmetic hyperbolic 3-manifolds, if
the lattice is kept fixed and the flat bundle varies. The results of the
present paper will be used to study the growth of the torsion in the
cohomology of arithmetic groups in higher rank cases.
Now we explain our results in more detail. Let δ(X) e = rankC (G) −
rankC (K). Occasionally we will denote this number by δ(G). Let g be
the Lie algebra of G. Let h ⊂ g be a fundamental Cartan subalgebra.
Let GC denote the connected complex linear Lie group corresponding
to the complexification gC of g and let U be a compact real form of GC
such that hC is the complexification of a Cartan-subalgebra of U . Then
the irreducible finite dimensional complex representations of G can be
identified with the irreducible finite dimensional complex representa-
tions of U . Fix positive roots ∆+ (gC , hC ). Let θ : g → g be the Cartan
involution. For a highest weight λ ∈ h∗C which we always assume to be
analytically integral with respect to U , we let τλ be the irreducible rep-
resentation of G corresponding to the representation of U with highest
weight λ. We will also say that τλ is the representation of G of highest
weight λ. Then we denote by λθ ∈ h∗C the highest weight of τλ ◦ θ, where
we regard θ as an involution on G. Our main result is the following
theorem.
Theorem 1.1. (i) Let X e be even dimensional or let δ(X)

e 6= 1. Then
TX (τ ) = 1 for all finite-dimensional representations τ of G.
e be odd-dimensional with δ(X)
(ii) Let X e = 1. Let λ ∈ h∗ be a highest
C
weight with λθ 6= λ. For m ∈ N, let τλ (m) be the irreducible represen-
tation of G with highest weight mλ. There exist constants c > 0 and
CXe 6= 0, which depends on X, e and a polynomial Pλ (m), which depends
on λ, such that

(1.2) log TX (τλ (m)) = CXe vol(X) · Pλ (m) + O e−cm
ANALYTIC TORSION AND L2 -TORSION OF MANIFOLDS 73
as m → ∞. Furthermore, there is a constant Cλ > 0 such that

(1.3) Pλ (m) = Cλ · m dim(τλ (m)) + Rλ (m),
where Rλ (m) is a polynomial whose degree equals the degree of the poly-
nomial dim(τλ (m)).
We note that (1.2) provides a complete asymptotic expansion for
log TX (τλ (m)). If one is only interested in the leading term, one can use
(1.3) which implies that there exists a constant C = C(X, e λ) 6= 0, which
depends on X e and λ, such that
(1.4) log TX (τλ (m)) = C vol(X) · m dim(τλ (m)) + O (dim(τλ (m)))

as m → ∞. Now the coefficient of the highest power can be determined
by Weyl’s dimension formula.
The condition λ 6= λθ is essential for our method to work. It implies
the existence of an increasing spectral gap for the corresponding Laplace
operators (see Corollary 7.6). It is a challenging and very interesting
problem to extend Theorem 1.1 to the case λ = λθ .
For hyperbolic manifolds, we proved the vanishing result (i) of The-
orem 1.1 in [MP, Proposition 1.7]. In general it was first proved by
Bismut, Ma, and Zhang [BMZ2]. It extends a result of Moscovici and
Stanton [MS1] who showed that TX (ρ) = 1, if δ(X) e ≥ 2 and ρ is a uni-
tary representation of Γ. Our proof is different from the previous proofs
and, as we believe, also simpler. It does not rely on the use of orbital
integrals or the Fourier inversion formula.
Part (ii) is a consequence of the following two propositions. The first
one shows that the asymptotic behavior of the analytic torsion with
respect to the representations τλ (m) is determined by the asymptotic
behavior of the L2 -torsion.
Proposition 1.2. Let X e be odd-dimensional with δ(X) e = 1. Let
∗
λ ∈ hC be a highest weight. Assume that λθ 6= λ. For m ∈ N let τλ (m)
be the irreducible representation of G with highest weight mλ. Then there
exists c > 0 such that
(2)
(1.5) log TX (τλ (m)) = log TX (τλ (m)) + O e−cm
for all m ∈ N.
This result was first proved in [MP] for hyperbolic manifolds. It was
also proved in [BMZ2] in the more general context of this paper (see
[BMZ2, Remark 7.8]). Our method of proof of (1.5) follows the method
developed in [MP].
The key result on which part (ii) of Theorem 1.1 relies is the com-
putation of the L2 -torsion. The computation is based on the Plancherel
formula. It gives
Proposition 1.3. Let the assumptions be as in Proposition 1.2. There

e and a polynomial Pλ (m),
exists a constant CXe , which depends on X,
which depends on λ, such that
(2)
(1.6) log TX (τλ (m)) = CXe vol(X) · Pλ (m), m ∈ N.
Moreover, there is a constant Cλ > 0 such that
(1.7) Pλ (m) = Cλ · m · dim(τλ (m)) + O (dim(τλ (m))
as m → ∞.
Finally, we note that if one specializes the main result of [BMZ2,
Theorem 1.1] to the case of analytic torsion of a locally symmetric space,
one can also determine the leading term of the asymptotic expansion of
(1.4). This has been carried out in [BMZ2] in the case of hyperbolic
3-manifolds.
If we consider one of the odd-dimensional irreducible symmetric spaces
Xe with δ(X)e = 1 and choose λ to be (an integral multiple of) a funda-
mental weight, the statements can be made more explicit.
Let G = SO0 (p, q), K = SO(p) × SO(q), p > 1, p, q odd, p ≥ q,
and let X e := G/K. Let n := (p + q − 2)/2. There are two fundamental
± ± ±
weights ω̃f,n which are not invariant under θ, and we let ωf,n := 2ω̃f,n
− +
(see (6.15)). One has ωf,n = (ωf,n )θ . By (6.21), it suffices to consider
+
the weight ωf,n . For m ∈ N let τ (m) be the representation with highest
+
weight mωf,n . By Weyl’s dimension formula, there exists a constant
C > 0 such that
n(n+1)
n(n+1)
(1.8) dim(τ (m)) = Cm 2 + O m 2 −1
as m → ∞. Let X ed be the compact dual of X.

e We let ǫ(q) := 0 for q = 1
and ǫ(q) := 1 for q > 1, and we let
pq−1
(−1) 2 2ǫ(q) π n
(1.9) Cp,q := p−1 .
ed )
vol(X 2
Corollary 1.4. Let p, q odd and let X e = SO0 (p, q)/ SO(p) × SO(q)
e With respect to the above notation we have
and X = Γ\X.
n(n+1)
log TX (τ (m)) = Cp,q vol(X) · m dim(τ (m)) + O m 2
as m → ∞.
The case q = 1 was treated in [MP] and the case p = 3, q = 1 in
[Mu3]. In the latter case we have Spin(3, 1) ∼
= SL(2, C). The irreducible
representation of Spin(3, 1) with highest weight 21 (m, m) corresponds to
the m-th symmetric power of the standard representation SL(2, C) on
C2 and we have
1
− log TX (τ (m)) = vol(X)m2 + O(m).
4π
The remaining case is X e = SL(3, R)/ SO(3). There are two funda-
mental weights ωi , i = 1, 2. Both are non-invariant under θ. Let τi (m),
i = 1, 2, be the irreducible representation with highest weight mωi . By
Weyl’s dimension formula, one has
1
dim τi (m) = m2 + O(m)
2
ed be the compact dual of X.
as m → ∞. Let X e
e = SL(3, R)/ SO(3) and X = Γ\X.
Corollary 1.5. Let X e We have
4π vol(X)
log TX (τi (m)) = m dim(τi (m)) + O(m2 )
ed )
9 vol(X
as m → ∞.
Using the equality of analytic and Reidemeister torsion [Mu2], we ob-
tain corresponding statements for the Reidemeister torsion τX (τλ (m)).
Especially we have
Corollary 1.6. Let X = Γ\X e be a compact odd-dimensional locally
symmetric manifold with δ(X) e = 1. Let λ ∈ h∗ be a highest weight which
C
satisfies λθ 6= λ. Let τX (τλ (m)) be the Reidemeister torsion of X with
respect to the representation τλ (m). Then vol(X) is determined by the
set {τX (τλ (m)) : m ∈ N}.
Finally we note that Bergeron and Venkatesh [BV] proved results
e = 1. Let Γ ⊃
of a similar nature, but in a different aspect. Let δ(X)
Γ1 ⊃ · · · ⊃ ΓN ⊃ · · · be a tower of subgroups of finite index with
∩N ΓN = {e}. A representation τ of G is called strongly acyclic if the
spectrum of the Laplacians ∆p (τ ) on ΓN \X e stays uniformly bounded
away from zero. Then for a strongly acyclic representation τ , they show
that there is a constant cG,τ > 0 such that
log TΓN \Xe (τ )
lim e
= cG,τ vol(Γ\X).
N →∞ [Γ : ΓN ]
Next we explain our methods to prove Theorem 1.1. The first step is
the proof of Proposition 1.2. We follow the proof used in [MP]. For an
irreducible representation τ of G and t > 0, put
d
X
K(t, τ ) := (−1)p p Tr e−t∆p (τ ) .
p=0
Assume that τ |Γ is acyclic, that is, H ∗ (X, Eτ ) = 0. Then the analytic

torsion is given by
Z ∞
1 d 1
(1.10) log TX (τ ) := t K(t, τ ) dt .
s−1
2 ds Γ(s) 0 s=0
Now the key ingredient of the proof of Proposition 1.2 is the following
lower bound for the spectrum of the Laplacians. For every highest weight
λ which satisfies λθ 6= λ, there exist C1 , C2 > 0 such that
(1.11) ∆p (τλ (m)) ≥ C1 m2 − C2 , m∈N
(see Corollary 7.6). Since τλ (m) is acyclic and dim X is odd, TX (τλ (m))
is metric independent [Mu2]. Especially, it is invariant under rescaling
1
of the metric. So we can replace ∆p (τλ (m)) by m ∆p (τλ (m)). Then
Z 1
1 d 1 t
log TX (τ (m)) = s−1
t K , τ (m) dt
2 ds Γ(s) 0 m s=0
(1.12) Z ∞
1 t
+ t−1 K , τ (m) dt.
2 1 m
It follows from (1.11) and standard estimations of the heat kernel that
m
the second term on the right is O(e− 8 ) as m → ∞. To deal with the
first term, we use a preliminary form of the Selberg trace formula. It
turns out that the contribution of the nontrivial conjugacy classes to the
trace formula is also exponentially decreasing in m. Finally, the identity
(2)
contribution equals log TX (τλ (m)) up to a term, which is exponentially
decreasing in m. This implies Proposition 1.2.
(2)
To deal with the L2 -torsion, we recall that for any τ , log TX (τ ) is
e
defined in terms of the Γ-trace of the heat operators e−t∆p (τ ) on the
e
universal covering [Lo]. In our case, e−t∆p (τ ) is a convolution operator
and its Γ-trace equals the contribution of the identity to the spectral
side of the Selberg trace formula applied to e−t∆p (τ ) . It follows that
(2) (2)
log TX (τ ) = vol(X) · t e (τ ),
X
(2) (2)
e and τ . To compute t (τ ), we factor-
where t e (τ ) depends only on X e
X X
e as X
ize X e = X e0 × X
e1 , where δ(Xe0 ) = 0 and X
e1 is irreducible with
δ(Xe1 ) = 1. Let τ = τ0 ⊗ τ1 be the corresponding decomposition of τ .
Let Xe0,d be the compact dual symmetric space of X e0 . Using a formula
similar to [Lo, Proposition 11], we get
(2) e χ(Xe0,d ) (2)

t e (τ ) = (−1)dim(X0 )/2 dim(τ0 ) · t e (τ1 ).
X vol(Xe0,d ) X1
(2)
This reduces the computation of t e (τ ) to the case of an irreducible
X
symmetric space X e with δ(X)
e = 1, which is odd-dimensional. From the
classification of simple Lie groups, it follows that the only possibilities
for Xe are X e = SL(3, R)/ SO(3) or X e = SO0 (k, l)/ SO(k) × SO(l), k, l
(2)
odd. Using the Plancherel formula, t e (τ ) can be computed explicitly
X
for these cases. Combined with Weyl’s dimension formula, it follows that
(2)
t e (τλ (m)) is a polynomial in m. In this way we obtain our main result.
X
The paper is organized as follows. In section 2 we collect some facts
about representations of reductive Lie groups. Section 3 is concerned
with Bochner-Laplace operators on locally symmetric spaces. The main
result is the estimation of the heat kernel of a Bochner-Laplace operator.
In section 4 we consider the analytic torsion in general. The main result
of this section is Proposition 4.2, which establishes part (i) of Theorem
1.1. Section 5 is devoted to the study of the L2 -torsion. We reduce the
study of the L2 -torsion to the case of an irreducible symmetric space
Xe with δ(X) e = 1. This case is then treated in section 6. Especially
we establish Proposition 1.3 in this case. In section 7 we prove a lower
bound for the spectrum of the twisted Laplace operators. This is the
key result for the proof of Proposition 1.2. In the final section 8, we
prove our main result, Theorem 1.1.
2. Preliminaries
In this section we summarize some facts about representations of
reductive Lie groups.
2.1. Let G be a real reductive Lie group in the sense of [Kn2, p.

446]. Let K ⊂ G be the associated maximal compact subgroup. Then
G has only finitely many connected components. Denote by G0 the
component of the identity. Let g and k denote the Lie algebras of G and
K, respectively. Let g = k ⊕ p be the Cartan decomposition.
We denote by Ĝ the unitary dual and by Ĝd the discrete series of
G. By Rep(G) we denote the equivalence classes of irreducible finite-
dimensional representations of G.
Let Q be a standard parabolic subgroup of G [Kn2, VII.7]. Then Q
has a Langlands decomposition Q = M AN , where M is reductive and
A is abelian. Q is called cuspidal if M̂d 6= ∅. Let KM = K ∩ M . Then
KM is a maximal compact subgroup of M .
Let Q = M AN be cuspidal. For (ξ, Wξ ) ∈ M̂d and ν ∈ a∗C , let
(2.1) πξ,ν = IndG ν
Q (ξ ⊗ e ⊗ Id)
be the induced representation acting by the left regular representation

on the Hilbert space

Hξ,ν = f : G → Wξ : f (gman) = e−(iν+ρQ )(log a) ξ(m)−1 f (g),
(2.2)
∀ m ∈ M, a ∈ A, n ∈ N, g ∈ G, f |K ∈ L2 (K, Wξ )
with norm given by
Z
kf k2 = |f (k)|2Wξ dk.
K
If ν ∈ a∗ , then πξ,ν is unitarily induced. Denote by Θξ,ν the global

character of πξ,ν .
2.2. Next we recall some facts concerning the discrete series. Let G be
a linear semisimple connected Lie group with finite center. Let K ⊂ G
be a maximal compact subgroup. Assume that δ(G) = 0. Then G/K is
even-dimensional. Let n = dim(G/K)/2. Let t ⊂ k be a compact Cartan
subalgebra of g. Let ∆(gC , tC ), ∆(kC , tC ) be the corresponding roots with
Weyl-groups WG , WK . Then one can regard WK as a subgroup of WG .
Let P be the weight lattice in it∗ . Let h·, ·i be the inner product on
it∗ induced by the Killing form. Recall that Λ ∈ P is called regular
if hΛ, αi = 6 0 for all α ∈ ∆(gC , tC ). Then Ĝd is parametrized by the
WK -orbits of the regular elements of P , where WK is the Weyl group of
∆(kC , tC ) [Kn1, Theorem 12.20, Theorem 9.20]. If Λ is a regular element
of P , the corresponding discrete series will be denoted by ωΛ . For π ∈ Ĝ
we denote by χπ the infinitesimal character of π. For a regular element
Λ ∈ h∗C let χΛ be the homomorphism of Z(gC ), defined by [Kn1, (8.32)].
By [Kn1, Theorem 9.20], the infinitesimal character of ωΛ is given by
χΛ . Fix positive roots ∆+ (gC , tC ) and let P + be the corresponding set of
dominant weights. Let ρG be the half-sum of the elements of ∆+ (gC , tC ).
Then we have the following proposition.
Proposition 2.1. Let τ ∈ Rep(G). Then for π ∈ Ĝd one has
(
p 1, χπ = χτ̌ , p = n;
dim (H (g, K; Hπ,K ⊗ Vτ )) =
0, else.
Moreover, there are exactly |WG |/|WK | distinct elements of Ĝd with
infinitesimal character χτ̌ , where τ̌ is the contragredient representation
of τ .
Proof. Let Λ(τ̌ ) ∈ P + be the highest weight of τ . Clearly Λ(τ̌ ) +
ρG is regular. Thus, since WG acts freely on the regular elements, the
proposition follows from [BW, Theorem I.5.3] and the above remarks
on infinitesimal characters. q.e.d.
2.3. Let Q = M AN be a standard parabolic subgroup. In general,
M is neither semisimple nor connected. But M is reductive in the
sense of [Kn2, p. 466]. Let KM = K ∩ M , let KM 0 be the compo-
nent of the identity, and let km := k ∩ m be its Lie algebra. Assume that
rank(M ) = rank(KM ). Then M has a nonempty discrete series, which
is defined as in [Kn1, XII,§8]. The explicit parametrization is given in
[Kn1, Proposition 12.32], [Wa2, section 8.7.1].
3. Bochner Laplace operators

Let G be a semisimple connected Lie group with finite center. Let
K ⊂ G be a maximal compact subgroup. Let X e = G/K. Let Γ be a
e
torsion-free, cocompact discrete subgroup of G and let X = Γ\X.
Let ν be a finite-dimensional unitary representation of K on the space

(Vν , h·, ·iν ). Let
eν := G ×ν Vν
E
be the associated homogeneous vector bundle over X. e Let Rg : E
eν → Eeν
be the action of g ∈ G. The inner product h·, ·iν induces a G-invariant
fiber metric ehν on Eeν . Let ∇
e ν be the connection on E
eν induced by the
canonical connection on the principal K-fiber bundle G → G/K. Then
∇e ν is G-invariant. Let
Eν := Γ\Eeν
be the associated locally homogeneous bundle over X. Since h̃ν and

e ν are G-invariant, they can be pushed down to a metric hν and a
∇
connection ∇ν on Eν . Let C ∞ (X, e Eeν ), resp. C ∞ (X, Eν ), denote the
e
space of smooth sections of Eν , resp. of Eν . Let
C ∞ (G, ν) := {f : G → Vν : f ∈ C ∞ ,
(3.1) f (gk) = ν(k−1 )f (g), ∀g ∈ G, ∀k ∈ K}.
Let L2 (G, ν) be the corresponding L2 -space. There is a canonical iso-
morphism
(3.2) e E
A : C ∞ (X, eν ) ∼
= C ∞ (G, ν)
which is defined by Af (g) = Rg−1 (f (gK)). It extends to an isometry
(3.3) e E
A : L2 (X, eν ) ∼
= L2 (G, ν).
Let
(3.4)
C ∞ (Γ\G, ν) := {f ∈ C ∞ (G, ν) : f (γg) = f (g), ∀g ∈ G, ∀γ ∈ Γ}
and let L2 (Γ\G, ν) be the corresponding L2 -space. The isomorphisms
(3.2) and (3.3) descend to isomorphisms
(3.5) A : C ∞ (X, Eν ) ∼
= C ∞ (Γ\G, ν), L2 (X, Eν ) ∼
= L2 (Γ\G, ν).
Let ∆ eν = ∇ fν ∗ ∇
e ν be the Bochner-Laplace operator of E eν . Since Xe
e
is complete, ∆ν with domain the space of smooth compactly supported
sections is essentially self-adjoint [LM, p. 155]. Its self-adjoint extension
will be denoted by ∆ e ν too. With respect to the isomorphism (3.2), one
has
(3.6) e ν = −R(Ω) + ν(ΩK ),
∆
where R denotes the right regular representation of G on C ∞ (G, ν) (see
[Mi1, Proposition 1.1]). The heat operator
˜
e−t∆ν : L2 (G, ν) → L2 (G, ν)
commutes with the action of G. Therefore, it is of the form

Z
−t∆˜ν
(3.7) (e φ)(g) = Htν (g−1 g′ )(φ(g′ )) dg ′
G
where
Htν : G → End(Vν )
is in C ∞ ∩ L2 and satisfies the covariance property
(3.8) Htν (k−1 gk′ ) = ν(k)−1 ◦ Htν (g) ◦ ν(k′ ), ∀k, k ′ ∈ K, ∀g ∈ G.
It follows as in [BM, Proposition 2.4] that Htν belongs to all Harish-
Chandra Schwartz spaces (C q (G) ⊗ End(Vν )), q > 0.
e 0 be the
Now let kHtν (g)k be the sup-norm of Htν (g) in End(Vν ). Let ∆
Laplacian on functions on X e and let Ht0 be the associated heat kernel
as above. We may use the principle of semigroup domination to bound
kHtν (g)k by the scalar heat kernel. Indeed we have
Proposition 3.1. Let ν ∈ K̂. Then we have
kHtν (g)k ≤ Ht0 (g)
for all t ∈ R+ and g ∈ G.
e e Eeν ).
Proof. Let Kν (t, x, y) be the kernel of e−t∆ν , acting in L2 (X,
Denote by |Kν (t, x, y)| the norm of the homomorphism

Kν (t, x, y) ∈ Hom (Eeν )y , (E
eν )x .
It was proved in [Mu1, p. 325] that in the sense of distributions, one

has
∂ e 0 |Kν (t, x, y)| ≤ 0,
+∆
∂t
where ∆e 0 acts in the x-variable. Using (3.15) in [Mu1], one can proceed
as in the proof of Theorem 4.3 of [DL] to show that
(3.9) |Kν (t, x, y)| ≤ K0 (t, x, y), e
t ∈ R+ , x, y ∈ X,
e
where K0 (t, x, y) is the kernel of e−t∆0 . See also [Gu, p. 7]. Now observe
that
Htν (g−1 g ′ ) = Rg−1 ◦Kν (t, gK, g′ K)◦Rg′ andHt0 (g −1 g′ ) = K0 (t, gK, g′ K).
e Rg : (E
Since for each x ∈ X, eν )x → (E
eν )g(x) is an isometry, the propo-
sition follows from (3.9). q.e.d.
Now we pass to the quotient X = Γ\X. e Let ∆ν = ∇ν ∗ ∇ν be the
Bochner-Laplace operator. It is essentially self-adjoint. Let RΓ be the
right regular representation of G on C ∞ (Γ\G, ν). By (3.6) it follows
that with respect to the isomorphism (3.5) we have
(3.10) ∆ν = −RΓ (Ω) + ν(ΩK ).
Let e−t∆ν be the heat semigroup of ∆ν , acting on L2 (Γ\G, ν). Then

e−t∆ν is represented by the smooth kernel
X
(3.11) Hν (t, g, g′ ) := Htν (g−1 γg ′ ).
γ∈Γ
The convergence of the series in (3.11) can be established, for example,

using Proposition 3.1 and the methods from the proof of Proposition
3.2 below. Put
(3.12) hνt (g) := tr Htν (g), g ∈ G,
where tr : End(Vν ) → C is the matrix trace. Then the trace of the heat
operator e−t∆ν is given by
Z Z X
(3.13) Tr(e−t∆ν ) = tr Hν (t, g, g) dġ = hνt (g−1 γg)dġ.
Γ\G Γ\G γ∈Γ
Using results of Donnelly, we now prove an estimate for the heat kernel
e 0 acting on C ∞ (X).
Ht0 of the Laplacian ∆ e
Proposition 3.2. There exist constants C0 and c0 such that for every
t ∈ (0, 1] and every g ∈ G, one has
X
Ht0 (g−1 γg) ≤ C0 e−c0 /t .
γ∈Γ
γ6=1
Proof. For x, y ∈ Xe let ρ(x, y) denote the geodesic distance of x, y.

e
Since K(t, gK, g′ K) = Ht0 (g −1 g′ ) is the kernel of e−t∆0 , it follows from
[Do1, Theorem 3.3] that there exists a constant C1 such that for every
g ∈ G and every t ∈ (0, 1], one has
2
0 − d2 ρ (gK, 1K)
(3.14) Ht (g) ≤ C1 t exp − .
4t
e and let BR (x) be the metric ball around x of radius R. Let
Let x ∈ X
h > 0 be the topological entropy of the geodesic flow of X (see [Ma]).
There exists C2 > 0 such that
(3.15) vol BR (x) ≤ C2 ehR , R>0
[Ma]. Since Γ is cocompact and torsion-free, there exists an ǫ > 0 such
that Bǫ (x) ∩ γBǫ (x) = ∅ for every γ ∈ Γ − {1} and every x ∈ X. e Thus
for every x ∈ X e the sets γBǫ (x), where γ run over all γ in Γ satisfying
ρ(x, γx) ≤ R, are mutually disjoint and contained in BR+ǫ (x). Using
(3.15), it follows that there exists a constant C3 such that for every
e one has
x ∈ X,
#{γ ∈ Γ : ρ(x, γx) ≤ R} ≤ C3 ehR .
e one
Hence there exists a constant C4 > 0 such that for every x ∈ X,
has
X ρ2 (γx,x)
(3.16) e− 8 ≤ C4 .
γ∈Γ
γ6=1
Now let
e
c1 := inf{ρ(x, γx) : γ ∈ Γ − {1}, x ∈ X}.
We have c1 > 0. Using (3.14) and (3.16), it follows that there are con-
stants c0 > 0 and C0 > 0 such that for every g ∈ G and 0 < t ≤ 1, we
have
X d 2
X 2
Ht0 (g−1 γg) ≤ C1 t− 2 e−c1 /(8t) e−ρ (γgK,gK)/8 ≤ C0 e−c0 /t .
γ∈Γ γ∈Γ
γ6=1 γ6=1
q.e.d.
4. The analytic torsion

Let τ be an irreducible finite-dimensional representation of G on Vτ .
Let Eτ be the flat vector bundle over X associated to the restriction of
e τ be the homogeneous vector bundle associated to τ |K and
τ to Γ. Let E
τ
let E := Γ\E eτ . There is a canonical isomorphism
(4.1) Eτ ∼
= Eτ
[MtM, Proposition 3.1]. By [MtM, Lemma 3.1], there exists an inner
product h·, ·i on Vτ such that
1) hτ (Y )u, vi = − hu, τ (Y )vi for all Y ∈ k, u, v ∈ Vτ
2) hτ (Y )u, vi = hu, τ (Y )vi for all Y ∈ p, u, v ∈ Vτ .
Such an inner product is called admissible. It is unique up to scaling.
Fix an admissible inner product. Since τ |K is unitary with respect to
this inner product, it induces a metric on E τ , and by (4.1) on Eτ , which
we also call admissible. Let Λp (Eτ ) = Λp T ∗ (X) ⊗ Eτ . Let
(4.2) νp (τ ) := Λp Ad∗ ⊗τ : K → GL(Λp p∗ ⊗ Vτ ).
Then there is a canonical isomorphism
(4.3) Λp (Eτ ) ∼
= Γ\(G ×ν (τ ) Λp p∗ ⊗ Vτ )
p
of locally homogeneous vector bundles. Let Λp (X, Eτ ) be the space of

the smooth Eτ -valued p-forms on X. The isomorphism (4.3) induces an
isomorphism
(4.4) Λp (X, Eτ ) ∼
= C ∞ (Γ\G, νp (τ )),
where the latter space is defined as in (3.4). A corresponding isomor-
phism also holds for the spaces of L2 -sections. Let ∆p (τ ) be the Hodge-
Laplacian on Λp (X, Eτ ) with respect to the admissible metric in Eτ . Let
RΓ denote the right regular representation of G in L2 (Γ\G). By [MtM,

(6.9)], it follows that with respect to the isomorphism (4.4), one has
∆p (τ )f = −RΓ (Ω)f + τ (Ω) Id f, f ∈ C ∞ (Γ\G, νp (τ )).
We remark that in [MtM] it is not assumed that G does not have
compact factors (see the remark on page 372 of [MtM]), and so we do
not make this assumption either. Let
d
X
(4.5) K(t, τ ) := (−1)p p Tr(e−t∆p (τ ) )
p=1
and
d
X
(4.6) h(τ ) := (−1)p p dim H p (X, Eτ ).
p=1
Then K(t, τ ) − h(τ ) decays exponentially as t → ∞, and it follows from

(1.1) that
Z ∞
1 d 1
(4.7) log TX (τ ) = t (K(t, τ ) − h(τ )) dt ,
s−1
2 ds Γ(s) 0 s=0
where the right hand side is defined near s = 0 by analytic continuation
of the Mellin transform. Let E eν (τ ) := G ×ν (τ ) Λp p∗ ⊗ Vτ and let ∆
e p (τ )
p p
e E
be the lift of ∆p (τ ) to C ∞ (X, eν (τ ) ). Then again it follows from [MtM,
p
∞
(6.9)] that on C (G, νp (τ )) one has
(4.8) ˜ p (τ ) = −RΓ (Ω) + τ (Ω) Id .
∆
˜
Let e−t∆p (τ ) be the corresponding heat semigroup on L2 (G, νp (τ )). It is
a smoothing operator which commutes with the action of G. Therefore,
it is of the form
Z
˜
e−t∆p (τ ) φ (g) = Htτ,p (g−1 g′ )φ(g′ ) dg ′ , φ ∈ (L2 (G, νp (τ )), g ∈ G,
G
where the kernel
(4.9) Htτ,p : G → End(Λp p∗ ⊗ Vτ )
belongs to C ∞ ∩ L2 and satisfies the covariance property
(4.10) Htτ,p (k−1 gk′ ) = νp (τ )(k)−1 Htτ,p (g)νp (τ )(k′ )
with respect to the representation (4.2). Moreover, for all q > 0 we have
(4.11) Htτ,p ∈ (C q (G) ⊗ End(Λp p∗ ⊗ Vτ ))K×K ,
where C q (G) denotes Harish-Chandra’s Lq -Schwartz space. The proof is
similar to the proof of Proposition 2.4 in [BM]. Now we come to the
heat kernel of ∆p (τ ). First the integral kernel of e−t∆p (τ ) , regarded as

an operator in L2 (Γ\G, νp (τ )), is given by
X τ,p
(4.12) H τ,p (t; g, g′ ) := Ht (g−1 γg ′ ).
γ∈Γ
As in section 3, this series converges absolutely and locally uniformly.

Therefore the trace of the heat operator e−t∆p (τ ) is given by
Z
−t∆p (τ )
Tr e = tr H τ,p (t; g, g) dġ,
Γ\G
where tr denotes the trace tr : End(Vν ) → C. Let

(4.13) hτ,p τ,p
t (g) := tr Ht (g).
Using (4.12) we obtain

Z X
(4.14) Tr e−t∆p (τ ) = hτ,p −1
t (g γg) dġ.
Γ\G γ∈Γ
Put
d
X
(4.15) ktτ = (−1)p p hτ,p
t .
p=1
Then it follows that

Z X
(4.16) K(t, τ ) = ktτ (g−1 γg) dġ.
Γ\G γ∈Γ
Let RΓ be the right regular representation of G on L2 (Γ\G). Then (4.16)

can be written as
(4.17) K(t, τ ) = Tr RΓ (ktτ ).
We shall now compute the Fourier transform of ktτ . To begin with, let
π be an admissible unitary representation of G on a Hilbert space Hπ .
Set Z
τ,p
π̃(Ht ) = π(g) ⊗ Htτ,p (g) dg.
G
This defines a bounded operator on Hπ ⊗ Λp p∗ ⊗ Vτ . As in [BM, pp.
160–161], it follows from (4.10) that relative to the splitting
h i⊥
Hπ ⊗ Λp p∗ ⊗ Vτ = (Hπ ⊗ Λp p∗ ⊗ Vτ )K ⊕ (Hπ ⊗ Λp p∗ ⊗ Vτ )K ,
π̃(Htτ,p ) has the form

π(Htτ,p ) 0
π̃(Htτ,p ) =
0 0
with π(Htτ,p ) acting on (Hπ ⊗ Λp p∗ ⊗ Vτ )K . Using (4.8), it follows as in

[BM, Corollary 2.2] that
(4.18) π(Htτ,p ) = et(π(Ω)−τ (Ω)) Id
on (Hπ ⊗ Λp p∗ ⊗ Vτ )K . Let {ξn }n∈N and {ej }m

j=1 be orthonormal bases
of Hπ and Λp p∗ ⊗ Vτ , respectively. Then we have
∞ X
X m
Tr π(Htτ,p ) = hπ(Htτ,p )(ξn ⊗ ej ), (ξn ⊗ ej )i
n=1 j=1
X∞ X m Z
= hπ(g)ξn , ξn ihHtτ,p (g)ej , ej i dg
(4.19) n=1 j=1 G
X∞ Z
= hτ,p
t (g)hπ(g)ξn , ξn i dg
n=1 G
= Tr π(hτ,p
t ).
Let π ∈ Ĝ and let Θπ denote its character. Then it follows from (4.15),
(4.18), and (4.19) that
d
X
(4.20) Θπ (ktτ ) =et(π(Ω)−τ (Ω))
(−1)p p · dim(Hπ ⊗ Λp p∗ ⊗ Vτ )K .
p=1
Now we consider the case of a principle series representation. Let Q be a

standard cuspidal parabolic subgroup. Let Q = M AN be the Langlands
decomposition of Q. Denote by a the Lie algebra of A. Let KM = K ∩M .
Let (ξ, Wξ ) be a discrete series representation of M and let ν ∈ a∗C . Let
πξ,ν be the induced representation and let Θξ,ν be the character of πξ,ν
(see section 2).
Proposition 4.1. Let Y ∈ a be a unit vector and let pY be the

orthogonal complement of Y in p. Then
K
(i) Θξ,ν (ktτ ) = et(πξ,ν (Ω)−τ (Ω)) dim Wξ ⊗ (Λodd p∗Y − Λev p∗Y ) ⊗ Vτ M ,
(ii) Θξ,ν (ktτ ) = 0 if dim aq ≥ 2.
Proof. By Frobenius reciprocity [Kn1, p. 208] and (4.20), we get

d
X
Θξ,ν (ktτ ) = et(πξ,ν (Ω)−τ (Ω)) (−1)p p dim (Wξ ⊗ Λp p∗ ⊗ Vτ )KM .
p=1
Now
p∗ = RY ∗ ⊕ p∗Y
as KM -modules. Therefore, in the Grothendieck ring of KM we have

d
X d
X
(−1)p pΛp p∗ = (−1)p pΛp p∗Y ⊕ Λp−1 p∗Y
p=1 p=1
d
X d−1
X
(4.21) = (−1)p pΛp p∗Y + (−1)p+1 (p + 1)Λp p∗Y
p=1 p=0
d
X
= (−1)p+1 Λp p∗Y .
p=0
Tensoring with Wξ and Vτ and taking KM -invariants, we obtain (i).

To prove (ii), suppose that there is a nonzero H ∈ a ∩ pY . Since M
centralizes H, ε(H)+i(H) is a KM intertwining operator between Λev p∗Y
and Λodd p∗Y , and non-trivial since H 6= 0. Hence Λev p∗Y and Λodd p∗Y are
equivalent as KM -modules, and (ii) follows. q.e.d.
Proposition 4.2. Assume that δ(X̃) ≥ 2 or assume that X̃ is even-

dimensional. Then TX (τ ) = 1 for all finite-dimensional irreducible rep-
resentations τ of G.
Proof. Let
M
RΓ = mΓ (π)π
π∈Ĝ
be the decomposition of the right regular representation RΓ of G on

L2 (Γ\G), see [Wa1, section 1]. Now insert ktπ on both sides and apply
the trace. If we recall that Θπ (ktτ ) = tr π(ktτ ) and use (4.17), we get
X
(4.22) K(t, τ ) = mΓ (π)Θπ (ktτ ).
π∈Ĝ
The series on the right hand side is absolutely convergent. First assume
that δ(X) ≥ 2. By [De, section 2.2] the Grothendieck group of all ad-
missible representations of G is generated by the representations πξ,λ ,
where πξ,λ is associated to some standard cuspidal parabolic subgroup
Q of G, as in (2.1). Since δ(X) ≥ 2, one has Θξ,λ (ktτ ) = 0 for every
such representation by Proposition 4.1. Thus one has Θπ (ktτ ) = 0 for
every irreducible unitary representation of G. By (4.22) it follows that
K(t, τ ) = 0. Let h(τ ) be as in (4.6). Since K(t, τ ) − h(τ ) decays expo-
nentially as t → ∞, it follows that K(t, τ ) − h(τ ) = 0, and using (4.7),
the first statement follows.
Now assume that d = dim X e is even. Note that as K-modules we
have
Λp p∗ ∼
= Λd−p p∗ , p = 0, . . . , d.
Since d is even, it follows that in the representation ring R(K), we have

the following equality:
d
X d
p p ∗ dX
(−1) pΛ p = (−1)p Λp p∗ .
2
p=0 p=0
Let (π, Hπ ) ∈ Ĝ. Then it follows from (4.20) that

d
d t(π(Ω)−τ (Ω) X
Θπ (ktτ ) = e (−1)p dim(Hπ ⊗ Λp p∗ ⊗ Vτ )K .
2 p=0
Let Hπ,K be the subspace of Hπ consisting of all smooth K-finite vectors.

Then
(Hπ,K ⊗ Λp p∗ ⊗ Vτ )K = (Hπ ⊗ Λp p∗ ⊗ Vτ )K .
Thus the (g, K)-cohomology H ∗ (g, K; Hπ,K ⊗ Vτ ) is computed from the
Lie algebra cohomology complex ([Hπ ⊗ Λp p∗ ⊗ Vτ ]K , d) (see [BW]).
Using the Poincaré principle, we get
d
d t(π(Ω)−τ (Ω) X
(4.23) Θπ (ktτ ) = e (−1)p dim H p (g, K; Hπ,K ⊗ Vτ ).
2
p=0
Now by [BW, II.3.1, I.5.3] we have

(
[Hπ ⊗ Λp p∗ ⊗ Vτ ]K , π(Ω) = τ (Ω);
(4.24) H p (g, K; Hπ,K ⊗ Vτ ) =
0, 6 τ (Ω).
π(Ω) =
Hence for every π ∈ Ĝ, one has Θπ (ktτ ) ∈ Z, and Θπ (ktτ ) is independent
of t > 0. Since the series on the right hand side of (4.22) converges
absolutely, there exist only finitely many π ∈ Ĝ with mΓ (π) 6= 0 and
Θπ (ktτ ) 6= 0. Thus K(t, τ ) is independent of t > 0. Let h(τ ) be defined by
(4.6). Then K(t, τ ) − h(τ ) = O(e−ct ) as t → ∞. Hence K(t, τ ) = h(τ ).
By (4.7) it follows that TX (τ ) = 1. q.e.d.
5. L2 -torsion
(2)
In this section we study the L2 -torsion TX (τ ). For its definition we
refer to [Lo]. Actually, in [Lo] only the case of the trivial representa-
tion τ0 has been discussed. However, the extension to a nontrivial τ is
straightforward. The definition is based on the Γ-trace of the heat oper-
e e (see [Lo]). For our purposes,
ator e−t∆p (τ ) on the universal covering X
2
it suffices to introduce the L -torsion for representations τ on Xe which
satisfy τθ =∼
6 τ.
Let hτ,p
t be the function defined by (4.13). By homogeneity it follows
that in our case the Γ-trace is given by

e
(5.1) TrΓ e−t∆p (τ ) = vol(X)hτ,p
t (1).
In order to define the L2 -torsion, we need to know the asymptotic be-

havior of hτ,p
t (1) as t → 0 and t → ∞. First we consider the behavior as
t → 0. Using (4.14) we have
Z X
τ,p
(5.2) vol(X)ht (1) = Tr e −t∆p (τ )
− hτ,p −1
t (g γg) dġ.
Γ\G γ∈Γ−{1}
To deal with the second term on the right, we consider the representation
νp (τ ) of K which is defined by (4.2), and for p = 0, . . . , n we put
(5.3) Ep (τ ) := τ (Ω) Id −νp (τ )(ΩK ),
which we regard as an endomorphism of Λp p∗ ⊗ Vτ . It defines an endo-
morphism of Λp T ∗ (X) ⊗ Eτ . By (3.6) and (4.8) we have
(5.4) ∆p (τ ) = ∆νp (τ ) + Ep (τ ).
Let νp (τ ) = ⊕σ∈K̂ m(σ)σ be the decomposition of νp (τ ) into irreducible
representations. This induces a corresponding decomposition of the ho-
mogeneous vector bundle
M
Eeν (τ ) = m(σ)Eeσ .
p
σ∈K̂
With respect to this decomposition, we have

M
(5.5) Ep (τ ) = m(σ) (τ (Ω) − σ(ΩK )) IdVσ ,
σ∈K̂
where σ(ΩK ) is the Casimir eigenvalue of σ and Vσ is the representation

space of σ, and
M
(5.6) ∆νp (τ ) = m(σ)∆σ .
σ∈K̂
ν (τ )
This shows that ∆νp (τ ) (τ ) commutes with Ep (τ ). Let Ht p be the
e ν (τ ) e
−t∆
kernel of e p and let Htτ,p be the kernel of e−t∆p (τ ) . Using (5.4) we
get
ν (τ )
(5.7) Htτ,p (g) = e−tEp (τ ) ◦ Ht p (g), g ∈ G.
Let c ∈ R be such that Ep (τ ) ≥ c. By Proposition 3.1 it follows that
(5.8) kHtτ,p (g)k ≤ e−ct Ht0 (g), t ∈ R+ , g ∈ G.
Taking the trace in End(Λp p∗ ⊗ Vτ ), we get
X
|hτ,p −1
t (g γg)|
γ∈Γ−{1}
(5.9) X
d
≤ dim(τ )e−ct Ht0 (g−1 γg), t ∈ R+ , g ∈ G.
p
γ∈Γ−{1}
By Proposition 3.2, there exist C1 , c1 > 0 such that

Z X
(5.10) |hτ,p −1
t (g γg)| dġ ≤ C1 e
−c1 /t
Γ\G γ∈Γ−{1}
for 0 < t ≤ 1. Thus by (5.2),

1
hτ,p
t (1) = Tr e−t∆p (τ ) + O(e−c1 /t )
vol(X)

for 0 < t ≤ 1. Using the asymptotic expansion of Tr e−t∆p (τ ) (see
[Gi]), it follows that there is an asymptotic expansion
∞
X
(5.11) hτ,p
t (1) ∼ aj t−d/2+j
j=0
as t → 0. To study the behavior of hτ,p t (1) as t → ∞, we use the

Plancherel theorem, which can be applied since hτ,p
t is a K-finite Schwarz
function. Let π be an admissible unitary representation of G on a Hilbert
space Hπ . It follows from (4.18) and (4.19) that
Tr π(hτ,p
t )=e
t(π(Ω)−τ (Ω))
dim (Hπ ⊗ Λp p∗ ⊗ Vτ )K .
Let Q = M AN be a standard parabolic subgroup of G. Let (ξ, Wξ ) be

a discrete series representation of M . Let h·, i denote the inner product
on the real vector space a∗ induced by the Killing form. Fix positive
restricted roots of a and let ρa denote the corresponding half-sum of
these roots. Define a constant c(ξ) by
(5.12) c(ξ) := −hρa , ρa i + ξ(ΩM ).
Recall that for ν ∈ a∗ one has
(5.13) πξ,ν (Ω) = −hν, νi + c(ξ).
Then by the Plancherel theorem, [HC, Theorem 3], and (5.13) we have
hτ,p
t (1)
X X Z
2
= e−t(τ (Ω)−c(ξ))
e−tkνk dim (Hξ,iν ⊗ Λp p∗ ⊗ Vτ )K pξ (iν) dν.
Q ξ∈M̂d a∗
Here the outer sum is over all association classes of standard cuspidal
parabolic subgroups of G, and pξ (iν), the Plancherel density, is of poly-
nomial growth in ν. Let KM = K ∩ M . By Frobenius reciprocity we
have
(5.14) dim (Hξ,ν ⊗ Λp p∗ ⊗ Vτ )K = dim (Wξ ⊗ Λp p∗ ⊗ Vτ )KM .
Thus we get
(5.15)
hτ,p
t (1)
X X Z
p ∗ KM −t(τ (Ω)−c(ξ)) 2
= dim (Wξ ⊗ Λ p ⊗ Vτ ) e e−tkνk pξ (iν) dν.
Q ξ∈M̂d a∗
The exponents of the exponential factors in front of the integrals are

controlled by the following lemma.
Lemma 5.1. Let (τ, Vτ ) ∈ Rep(G). Assume that τ ∼ 6 τθ . Let P =
=
M AN be a cuspidal parabolic subgroup of G. Let ξ ∈ M̂d and assume
that dim (Wξ ⊗ Λp p∗ ⊗ Vτ )KM 6= 0. Then one has
τ (Ω) − c(ξ) > 0.
Proof. Assume that τ (Ω) − c(ξ) ≤ 0. Then by (5.13) there exists a
ν0 ∈ a∗ such that
πξ,ν0 (Ω) = τ (Ω).
Together with (5.14), our assumption, and [BW, Proposition II.3.1], it
follows that
dim (H p (g, K; Hξ,ν0 ,K ⊗ Vτ )) 6= 0,
where Hξ,ν0 ,K are the K-finite vectors in Hξ,ν0 . Since τ ∼
6 τθ , this is a
=
contradiction to the first statement of [BW, Proposition II.6.12]. q.e.d.
Let τ be an irreducible representation of G which satisfies τ ∼

6 τθ . It
=
follows from (5.15) and Lemma 5.1 that there exists c > 0 such that

(5.16) hτ,p
t (1) = O e
−ct
as t → ∞. Using (5.11) and (5.16), it follows from standard methods

(see for example [Gi]), that the Mellin transform
Z ∞
hτ,p
t (1)t
s−1
dt
0
converges absolutely and uniformly on compact subsets of the half-plane
Re(s) > d/2 and admits a meromorphic extension to C which is holo-
morphic at s = 0 if d = dim(X̃) is odd and has at most a simple
pole at s = 0 for d = dim(X̃) even. Thus we can define the L2 -torsion
(2)
TX (τ ) ∈ R+ by
(2)
log TX (τ )
d Z
(5.17) 1X d 1 ∞
e p (τ )
:= (−1)p p TrΓ e −t∆
t s−1
dt ,
2 p=1 ds Γ(s) 0 s=0
where the right hand side is defined near s = 0 by analytic continuation.

For t > 0 let
d
X
(5.18) K (2) (t, τ ) := (−1)p phτ,p
t (1).
p=1
Put
Z ∞
(2) 1 d 1
(5.19) t e (τ ) := K (2)
(t, τ )t s−1
dt .
X 2 ds Γ(s) 0 s=0
(2) e and τ , and we
Then t e (τ ) depends only on the symmetric space X
X
have
(2) (2)
(5.20) log TX (τ ) = vol(X) · t e (τ ).
X
Next we establish an auxiliary result concerning the twisted Eu-
ler characteristic. We let τ ∈ Rep(G) be arbitrary. Let Hp (X, Eτ ) :=
ker ∆p (τ ) be the space of Eτ -valued harmonic p-forms. Let
d
X
χ(X, Eτ ) := (−1)p dim Hp (X, Eτ )
p=0
be the twisted Euler characteristic. Furthermore, let X̃d denote the com-
pact dual of X̃. The following proposition is a familar consequence of
the index theorem. For the convenience of the reader, we include an
independent proof.
Proposition 5.2. If δ(X̃) 6= 0, we have χ(X, Eτ ) = 0. If δ(X̃) = 0,
one has
χ(X̃d )
(5.21) χ(X, Eτ ) = (−1)n vol(X) dim(τ ),
vol(X̃d )
e
where n = dim(X)/2.
Proof. Let π ∈ Ĝ. It follows from (4.18) and (4.20) that
d
X d
X
p
(−1) Θπ (hp,τ
t ) =e t(π(Ω)−τ (Ω))
(−1)p dim(Hπ ⊗ Λp p∗ ⊗ Vτ )K .
p=0 p=0
Using [BW, II.3.1] and the Poincaré principle as in the proof of Propo-
sition 4.2, we get
d
X d
X
(5.22) (−1)p Θπ (hp,τ
t )= (−1)p dim H p (g, K; Hπ,K ⊗ Vτ ).
p=0 p=0
Now by [BW, Theorem I.5.3], it follows that if H p (g, K; Hπ,K ⊗ Vτ ) 6=

0, then χπ = χτ̌ , where τ̌ is the contragredient representation of τ .
By [Kn1, Corollary 10.37, Corollary 9.2], there are only finitely many
representations π ∈ Ĝ with a given infinitesimal character. Thus if Q =
M AN is a fundamental parabolic subgroup with Q 6= G and if ξ ∈ M̂d ,

it follows that there are only finitely many λ ∈ a∗ such that
d
X
(5.23) (−1)p Θξ,λ (hp,τ
t ) 6= 0.
p=0
Hence by the Plancherel theorem, [HC, Theorem 3], and (5.22), we get
(5.24)
Xd d
X X
p p,τ
(−1) ht (1) = (−1)p d(π) dim H p (g, K; Hπ,K ⊗ Vτ ),
p=0 p=0 π∈Ĝd
where Ĝd denotes the discrete series of G, and d(π) denotes the formal
degree of π. The sum is finite. Let

˜ p (τ )
b(2)
p (X, Eτ ) := lim TrΓ e
−t∆
t→∞
be the L2 -Betti number. Using that (5.24) is independent of t and (5.1),

we get
d
X d
X
(5.25) vol(X) (−1)p hp,τ
t (1) = (−1)p b(2) (2)
p (X, Eτ ) = χ (X, Eτ ).
p=0 p=0
By the Γ-index theorem of Atiyah [At], we have χ(2) (X, Eτ ) = χ(X, Eτ ).

Hence by (5.24) and (5.25) we get
(5.26)
d
X X
χ(X, Eτ ) = vol(X) · (−1)p d(π) dim H p (g, K; Hπ,K ⊗ Vτ ).
p=0 π∈Ĝd
If δ(X̃) 6= 0, then Ĝd is empty, and hence this sum equals zero, which
proves the first statement. Now assume that δ(X̃) = 0. Then X̃ is even-
dimensional. Let dim(X̃) = 2n. We keep the notation from section 2.2.
By [Ol, Corollary 5.2] for Λ′ = w(Λ(τ̌ ) + ρG ), w ∈ WG /WK , one has
dim(τ )
d(ωΛ′ ) = ,
vol(X̃d )
and so together with Proposition 2.1 we get
d
X X
(−1)p d(π) dim H p (g, K; Hπ,K ⊗ Vτ )
(5.27) p=0 π∈Ĝd
1
=(−1)n #(WG /WK ) dim(τ ).
vol(X̃d )
Finally, by [Bo, page 175], one has
#(WG /WK ) = χ(X̃d ).
Applying (5.27), the proof of the proposition follows. q.e.d.
Remark 1. We remark that if X is Hermitian and τ is the triv-

ial representation, then (5.21) reduces to Hirzebruch’s proportionality
principle.
Now we assume that δ(X) e = 1 and that Xe is odd-dimensional. By

the classification of simple Lie groups, we have X e = X e0 × X
e1 , where
e e e 0
δ(X0 ) = 0 and X1 = SL(3, R)/ SO(3) or X1 = SO (p, q)/ SO(p)×SO(q),
p, q odd. Let X e0 = G0 /K0 and let G1 = SL(3, R), K1 = SO(3), or
0
G1 = SO (p, q), K1 = SO(p) × SO(q), p, q odd. Let G = G0 × G1 . Let τ
be a finite-dimensional irreducible representation of G, and assume that
τ 6∼
= τθ . Then τ = τ0 ⊗ τ1 , where τi is an irreducible representation of
Gi , i = 0, 1, and τ1 ∼
6 τ1,θ .
=
Proposition 5.3. Let δ(X) e = 1 and let X e be odd-dimensional. Let

e e e e
X = X0 × X1 , where X1 is an odd-dimensional irreducible symmetric
space with δ(Xe1 ) = 1. Let τ be a finite-dimensional irreducible repre-
sentation of G with τ ∼
6 τθ . Then
=
(2) e χ(Xe0,d ) (2)

t e (τ ) = (−1)dim X0 /2 dim τ0 · t e (τ1 ).
X e
vol(X0,d ) X1
Proof. Let E e→X e be the homogeneous vector bundle associated to

e
τ |K . Similarly, let Ei → X ei be the homogeneous vector bundle associ-
e∼
ated to τi |Ki , i = 0, 1. Then E =E e1 ⊠ E
e2 and
M
Λk (X, e ∼
e E) = e0 , E
Λp (X e0 ) ⊗ Λq (X
e1 , E
e1 ) .
p+q=k
With respect to this decomposition, we have

M
e k (τ ) =
∆ e p (τ0 ) ⊗ Id + Id ⊗∆
∆ e q (τ1 ) .
p+q=k
Let Htτ,k and Htτi ,p , i = 0, 1, be the corresponding heat kernels. Then it

follows that Htτ,k = ⊕p+q=k Htτ0 ,p ⊗ Htτ1 ,q . Hence for hτ,k
t = tr Htτ,k and
htτi ,p = tr Htτi ,p , i = 0, 1, we have
X
hτ,k
t = hτt 0 ,p · hτt 1 ,q .
p+q=k
Using this equality, we get

d
X d1 X
X d2
(−1)k k hτ,k
t (1) = (−1)p+q (p + q) hτt 1 ,p (1) · hτt 2 ,q (1)
k=0 p=0 q=0
d1
X d2
X
(5.28) = (−1)p hτt 1 ,p (1) · (−1)q q hτt 2 ,q (1)
p=0 q=0
d2
X d1
X
+ (−1)q
hτt 2 ,q (1) · (−1)p p hτt 1 ,p (1).
q=0 p=0
Let Γi ⊂ Gi , i = 0, 1, any cocompact, torsion-free discrete subgroup.

The existence of the Γi follows from our assumptions on the Gi stated
in the introduction and from results of Borel [Bor]. Put Xi = Γi \X ei
e
and Ei = Γ\Ei . By (5.25) and the remark following it, we have
d
X χ(Xi )
(5.29) (−1)p htτi ,p (1) = , i = 0, 1.
vol(Xi )
p=0
Taking the Mellin transform of (5.28) and using (5.29) and Proposition
5.2, the proposition follows. q.e.d.
e =1
6. The asymptotics of the L2 -torsion for δ(X)
In this section we study the asymptotic behavior of the L2 -torsion of
an odd-dimensional irreducible symmetric space X e with δ(X)
e = 1. Then
0
we can assume that G = SO (p, q), p, q odd, and K = SO(p) × SO(q),
or G = SL3 (R) and K = SO(3). To compute the L2 torsion in these
cases, we need some preparation. Let P = M AN be a fundamental
parabolic subgroup of G, i.e. we have dim(A) = 1. Let M 0 be the
identity component of M and let m be its Lie algebra. Then in our case
m is always semisimple. Let KM := K ∩ M , let KM 0 be the identity
component of KM , and let km := k ∩ m be its Lie algebra. Let t be a

Cartan subalgebra of km . Then t is also a Cartan subalgebra of m and
of k. Moreover, h := a ⊕ t is a Cartan subalgebra of g.
Let ∆(gC , hC ), ∆(mC , tC ), ∆((km )C , tC ) be the corresponding roots.
Then there is a canonical inclusion ∆(mC , tC ) ֒→ ∆(gC , hC ). Fix a pos-
itive restricted root e1 ∈ a∗ and fix positive roots ∆+ (mC , tC ). In this
way we obtain positive roots ∆+ (gC , hC ). Let ρG , resp. ρM , be the half-
sums of the elements of ∆+ (gC , hC ) and ∆+ (mC , tC ), respectively. By
our choices, we have ρG |m = ρM .
Let
T := {m ∈ KM : Ad(m)|t = Id}.
Then we have
T = {k ∈ K : Ad(k)|t = Id}.
Thus T is connected. Let NKM and NK 0 be the normalizers of t in KM
M
and KM0 , respectively. Let W
KM := NKM /T and let Wkm = NKM 0 /T be
the Weyl group of ∆((km )C , tC ). Moreover, we let Wm be the Weyl group

of ∆(mC , tC ). Finally we let
W (A) := {k ∈ K : Ad(k)a = a}/KM .
The following lemma is certainly well-known and has already been used
by Olbrich [Ol, page 15]. However, for the sake of completeness, we
include a proof here.
Lemma 6.1. One has
|WKM |
· |W (A)| = 2.
|Wkm |

Proof. By[Kn2, Proposition 7.19 (b)], the order # M/M 0 equals
# KM /KM 0 . Let k ∈ K . Then Ad(k)t is a maximal torus in k
M m
0 0
and thus there exists a k ∈ KM such that Ad(k)t = Ad(k )t. Hence0
every element of KM /KM 0 has a representative in N

KM , and thus there
are canonical isomorphisms KM /KM 0 ∼ N ∼
= KM /NKM 0 = WKM /Wkm . In
other words, |WKM |/|Wkm | equals the number of components of M . Let
ap be a maximal abelian subspace of p containing a, let ∆ap be the
corresponding restricted roots, and let W (∆ap ) be the corresponding
Weyl-group. One has W (∆ap ) = NK (ap )/ZK (ap ), where NK (ap ), resp.
ZK (ap ), are the normalizer resp. centralizer, of ap in K. Moreover, by
[Kn2, Proposition 8.85] each element of W (A) has a representative in
NK (ap ), i.e can be extended to an element of W (∆ap ) which fixes a.
Now a case-by-case study easily implies that W (∆ap ) contains such an
element which is non-trivial if and only if G = SO0 (p, 1). In this case,
M is connected. In all other cases, M has exactly two components. This
proves the lemma. q.e.d.
Let H1 ∈ a with e1 (H1 ) = 1. Then we normalize the Killing form
B by 1/B(H1 , H1 ). We let k·k be the corresponding norm on the real
vector-space it∗ ⊕ a∗ . Let Ω be the Casimir element with respect to the
normalized Killing form. Then for τ ∈ Rep(G) with highest weight Λ(τ ),
we have
(6.1) τ (Ω) = kΛ(τ ) + ρG k2 − kρG k2 .
The restriction of h·, ·i to m is non-degenerate and Ad-invariant. Let ΩM
be the corresponding Casimir element. For σ ∈ Rep(M 0 ) with highest
weight Λ(σ) ∈ it∗ , we define
(6.2) c(σ) := kΛ(σ) + ρM k2 − kρG k2 .
Then one has c(σ) = χσ (ΩM ) − kρG |a k2 , and thus one has
(6.3) c(σ) = c(σ̌)
for every σ ∈ Rep(M 0 ). Let Wg := W (gC , hC ) be the Weyl group of
∆(gC , hC ), and for w ∈ Wg , let ℓ(w) be its length with respect to the
simple roots defined by ∆+ (gC , hC ). Finally let
W 1 := {w ∈ Wg : w−1 α > 0 ∀α ∈ ∆+ (mC , tC )}.
The subspace n is even-dimensional and we write dim(n) = 2n. For
k = 0, . . . , 2n let H k (n; Vτ ) be the Lie-algebra cohomology of n with
coefficients in Vτ . Then the H k (n; Vτ ) are M 0 A-modules and their de-
composition into irreducible M 0 A-components can be described by the
following theorem of Kostant.
Proposition 6.2. In the sense of M 0 A-modules, one has
X
H k (n; Vτ ) ∼
= Vτ (w) ,
w∈W 1
ℓ(w)=k
where Vτ (w) is the M 0 A module with highest weight w(Λ(τ ) + ρG ) − ρG .

Proof. See for example [Wr, Theorem 2.5.1.3]. q.e.d.
Corollary 6.3. As M 0 A-modules, we have
2n
M M
(−1)k Λk n∗ ⊗ Vτ = (−1)l(w) Vτ (w) .
k=0 w∈W 1
Proof. This follows from Proposition 6.2 and the Poincaré principle
[Ko, (7.2.3)]. q.e.d.
For w ∈ W 1 , let στ,w ∈ Rep(M 0 ) be the finite-dimensional irreducible
representation of M 0 with highest weight
(6.4) Λ(στ,w ) := w(Λ(τ ) + ρG )|t − ρM ,
and let λτ,w ∈ R be such that
(6.5) w(Λ(τ ) + ρG )|a = λτ,w e1 .
Then we have the following corollary about the Casimir eigenvalue.
Proposition 6.4. For every w ∈ W 1 , one has
τ (Ω) = λ2τ,w + c(στ,w ).
Proof. By (6.1) we have
τ (Ω) = kΛ(τ ) + ρG k2 − kρG k2 = kw(Λ(τ ) + ρG )k2 − kρG k2
= kλτ,w e1 k2 + kΛ(στ,w ) + ρM k2 − kρG k2 = λ2τ,w + c(στ,w ).
q.e.d.
Let ktτ be defined by (4.15). Our next goal is to compute the Fourier
transform of ktτ . Note that, since T is connected, it follows from [Wa2,
section 6.9, section 8.7.1] that for every discrete series representation ξ
of M over Wξ , there exists a discrete series representation ξ 0 of M 0 over
Wξ 0 such that ξ is induced from ξ 0 . Moreover, since M 0 is semisimple,
the discrete series of M 0 is parametrized as in section 2.2. By [Wa2,
section 8.7.1], two discrete series representations ξ10 and ξ20 of M 0 with
corresponding parameters Λξ10 , Λξ20 as in section 2.2 induce the same
discrete series representation of M if and only if Λξ10 and Λξ20 are WKM -
conjugate. For ξ ∈ M̂d and λ ∈ C, we let πξ,λ := πξ,λe1 , Θξ,λ := Θξ,λe1 .
Proposition 6.5. Let ξ ∈ M̂d with infinitesimal character χ(ξ). Let

pm := p ∩ m and let v := 21 dim pm . Then for λ ∈ C one has
X 2 2
Θξ,λ (ktτ ) = (−1)v (−1)ℓ(w)+1 e−t(λ +λτ,w ) .
w∈W 1
χ(ξ)=χ(σ̌τ,w )
Proof. Let ξ 0 , Λξ 0 be as above. Then one has
πξ,λ (Ω) = −λ2 + kΛξ k2 − kρG k2 .
Thus if σ ∈ Rep(M 0 ) is such that χσ = χξ , one has
(6.6) πξ,λ (Ω) = −λ2 + c(σ).
Moreover, ξ|KM is induced from ξ 0 |K 0 and so by Frobenius reciprocity,

M
one has
0
[Λp p∗ ⊗ Hξ ⊗ Vτ ]K = [Λp p∗ ⊗ Wξ ⊗ Vτ ]KM = [Λp p∗ ⊗ Wξ0 ⊗ Vτ ]KM .
Thus by (4.18) one has

d
X K 0
Θξ,λ (ktτ ) = et(πξ,λ (Ω)−τ (Ω)) (−1)p p Λp p∗ ⊗ Wξ 0 ⊗ Vτ M .
p=0
0
Let pY be as in Proposition 4.1. Since dim a = 1, it follows that as KM
modules, pY ∼ 0
= pm ⊕ n. Using (4.21), it follows that as KM modules, we
have
d
X d
X
(−1)p pΛp p∗ = (−1)p+1 Λp (p∗m ⊕ n∗ )
p=0 p=0
2n
X
= (−1)k+1 Λev p∗m − Λodd p∗m ⊗ Λk n∗ .
k=0
Thus, together with Corollary 6.3 and the Poincaré principle one gets
d
X K 0
(−1)p p Λp p∗ ⊗ Wξ 0 ⊗ Vτ M
p=0
X h iK 0
M
= (−1)ℓ(w)+1 Λev p∗m − Λodd p∗m ⊗ Wξ 0 ⊗ Vτ (w)
w∈W 1
X
= (−1)ℓ(w)+1 χ(m, KM 0 ; Wξ 0 ⊗ Vτ (w) ),
w∈W 1
where χ(m, KM0 ;W 0 ⊗ V

ξ τ (w) ) denotes the Euler-characteristic of the
(m, KM )-cohomology with coefficients in the M 0 -module Vτ (w) ⊗ Wξ 0 .
0
Thus the proposition follows from Proposition 2.1, Proposition 6.4, (6.6)
and (6.3). q.e.d.
Next we come to the Plancherel measures. For ξ ∈ M̂d , we let ξ 0 ∈

M̂d0 be as above. Fix a regular Λξ 0 ∈ it∗ corresponding to ξ 0 as in
section 2.2 and let Λξ := Λξ 0 . Choose positive roots ∆+ (mC , tC ; Λξ ) such
that Λξ is dominant with respect to these roots. Let ∆+ (gC , hC ; Λξ ) be
positive roots defined via ∆+ (mC , tC ; Λξ ) and e1 , and let ρG,Λξ be the
half-sum of the elements of ∆+ (mC , tC ; Λξ ). For λ ∈ R, we let µξ (λ) be
the Plancherel measure of πξ,λ . Then there exists a polynomial Pξ (z)
such that one has
(6.7) µξ (λ) = Pξ (iλ).
The polynomial Pξ (z) is given as follows. There exists a constant cX̃
which depends only on X̃ such that one has
Y hα, Λξ + ze1 i
(6.8) Pξ (z) = (−1)n cX̃

+
α, ρG,Λξ
α∈∆ (gC ,hC ;Λξ )
[Kn1, Theorem 13.11], [Wa3, Theorem 13.5.1]. By [Ol, Lemma 5.1]

and our normalizations, one has
1
(6.9) cX̃ = .
|W (A)| vol(X̃d )
Note that Pξ (z) is an even polynomial in z. Now let w ∈ Wm . We
regard Wm as a subgroup of Wg . Then if we replace Λξ by wΛξ , we have
to replace ∆+ (gC , tC ; Λξ ) by w∆+ (gC , tC ; Λξ ). This implies that Pξ (z)
depends only on the Wm -orbit of Λξ , or equivalently on the infinitesimal
character χ(ξ) of ξ. Thus if for σ ∈ Rep(M 0 ) with highest weight Λ(σ)
we let
Y hα, Λ(σ) + ρM + ze1 i
(6.10) Pσ (z) := (−1)n cX̃ ,
+
hα, ρG i
α∈∆ (gC ,hC )
where cX̃ is as in (6.8), it follows that Pξ (λ) = Pσ (λ) if χ(σ) = χ(ξ).

Putting everything together, we obtain the following corollary.
Proposition 6.6. Let τ ∈ Rep(G) and assume that τ ∼ 6 τθ . Then
=
one has
Z |λτ,w |
(2) v |Wm | X ℓ(w)
log TX (τ ) = (−1) π vol(X) (−1) Pσ̌τ,w (t)dt.
|WKM | 1 0
w∈W
Proof. For a given regular and integral Λ ∈ it∗ , there are exactly
|Wm |/|WKM | distinct elements of M̂d with infinitesimal character χΛ .
Thus if one combines the Plancherel theorem with Proposition 4.1,
Proposition 6.5, (6.7), and the previous remarks, one obtains
X Z
τ v |Wm | ℓ(w)+1 −tλ2τ,w 2
kt (1) = (−1) (−1) e e−tλ Pσ̌τ,w (iλ)dλ.
|WKM | 1 R
w∈W
We let
I(t, τ ) := vol(X)ktτ (1).
By the computations below, one has |λτ,w | > 0 for every w ∈ W 1 . Thus,
since is Pσ (λ) is an even polynomial of degree 2n for each σ ∈ M̂ 0 , for
s ∈ C with Re(s) > 2n + 1 the integral
Z ∞
MI(s, τ ) := ts−1 I(t, τ )dt
0
exists. Moreover, by [Fr, Lemma 2 and Lemma 3], MI(s, τ ) has a mero-
morphic continuation to C which is regular at 0, and if MI(τ ) denotes
its value at 0, one has
Z |λτ,w |
|Wm | X
MI(τ ) = 2π vol(X)(−1)v (−1)ℓ(w) Pσ̌τ,w (λ) dλ.
|WKM | 1 0
w∈W
By definition one has
(2) 1
log TX (τ ) = MI(τ )
2
and the proposition follows. q.e.d.
Now let G = SO0 (p, q), p > 1, p, q odd, p ≥ q, p = 2p1 + 1, q =
e = G/K. Then
2q1 + 1. Let n := p1 + q1 . Let K = SO(p) × SO(q) and X
e
dim(X) = 2n + 1. The normalized Killing form is given by
1
hX, Y i := B(X, Y ).
2n − 2
We equip X̃ with the Riemannian metric defined by the restriction of
h·, ·i to p. We have m ∼
= so(p − 1, q − 1). We realize the fundamental
Cartan subalgebra as follows. Let
(6.11) H1 := Ep,p+1 + Ep+1,p .
Then we put
a = RH1 .
Moreover, we let
(√
−1(E2i−3,2i−2 − E2i−2,2i−3 ), 2 ≤ i ≤ p1 + 1
(6.12) Hi := √
−1(E2i−1,2i − E2i,2i−1 ), p1 + 1 < i ≤ n + 1.
Then
n+1
M √
t := −1Hi
i=2
is a Cartan subalgebra of m and
h := a ⊕ t
is a Cartan subalgebra of g. Define ei ∈ h∗C , i = 1, . . . , n + 1, by
ei (Hj ) = δi,j , 1 ≤ i, j ≤ n + 1.
Then the sets of roots of (gC , hC ) and (mC , tC ) are given by
∆(gC , hC ) = {±ei ± ej , 1 ≤ i < j ≤ n + 1}
∆(mC , tC ) = {±ei ± ej , 2 ≤ i < j ≤ n + 1}.
We fix positive systems of roots by
∆+ (gC , hC ) := {ei + ej , i 6= j} ⊔ {ei − ej , i < j}
∆+ (mC , tC ) := {ei + ej , i 6= j, i, j ≥ 2} ⊔ {ei − ej , 2 ≤ i < j}.
We parametrize the finite-dimensional irreducible representations τ of
G by their highest weights
(6.13)
Λ(τ ) = k1 (τ )e1 + · · · + kn+1 (τ )en+1 , (k1 (τ ), . . . , kn+1 (τ )) ∈ Zn+1 ,
k1 (τ ) ≥ k2 (τ ) ≥ · · · ≥ kn (τ ) ≥ |kn+1 (τ )| .
If Λ is a weight as in (6.13), then
(6.14) Λθ = k1 (τ )e1 + · · · + kn (τ )en − kn+1 (τ )en+1 .
Now we let
n+1
X n
X
+ − +
(6.15) ωf,n := ej ; ωf,n := (ωf,n )θ = ej − en+1 .
j=1 j=1
±
Then 12 ωf,n are the fundamental weights of ∆+ (gC , hC ) which are not
invariant under θ. We parametrize the finite-dimensional irreducible rep-
resentations σ of M 0 by their highest weights
Λ(σ) =k2 (σ)e2 + · · · + kn+1 (σ)en+1 , (k2 (σ), . . . , kn+1 (σ)) ∈ Zn
(6.16)
k2 (σ) ≥ k3 (σ) ≥ · · · ≥ kn (σ) ≥ |kn+1 (σ)| ∈ Zn .
For σ ∈ Rep(M 0 ) with highest weight Λ(σ) as in (6.16), we let w0 σ ∈

Rep(M 0 ) be the representation with highest weight
Λ(w0 σ) := k2 (σ)e2 + · · · + kn (σ)en − kn+1 (σ)en+1 .
Then for every σ ∈ Rep(M 0 ), one has σ̌ = σ if n is even and σ̌ = w0 σ
if n is odd. Applying (6.10), this implies that
(6.17) Pσ (λ) = Pw0 σ (λ) = Pσ̌ (λ)
for every σ ∈ Rep(M 0 ).
Let τ ∈ Rep(G) with highest weight τ1 e1 + · · · + τn+1 en+1 . For k =
0, . . . , n, let
(6.18) λτ,k = τk+1 + n − k
and let στ,k be the representation of G with highest weight
(6.19)
Λστ,k := (τ1 + 1)e2 + · · · + (τk + 1)ek+1 + τk+2 ek+2 + · · · + τn+1 en+1 .
Then as in [MP, section 2.7], one has
(6.20)
{(λτ,w , στ,w , l(w)) : w ∈ W 1 } = {(λτ,k , στ,k , k) : k = 0, . . . , n}
⊔ {(−λτ,k , w0 στ,k , 2n − k) : k = 0, . . . , n}.
Combining (6.14), (6.17), (6.20), and Proposition 6.6, it follows that
(2) (2)
(6.21) TX (τ ) = TX (τθ )
for each τ ∈ Rep(G). Now for p, q ∈ N we let ǫ(q) := 0 for q = 1 and
ǫ(q) := 1 for q > 1, and we let
pq−1
(−1) 2 2ǫ(q) π p+q−2
2
(6.22) Cp,q := p−1 .
vol(X̃d ) 2
Then we have
Proposition 6.7. For p, q, odd, p ≥ q, let Xe = SO0 (p, q)/ SO(p) ×
e ∗
SO(q) and let X = Γ\X. Let Λ ∈ hC be a highest weight as in (6.13) and
assume that Λθ 6= Λ. For m ∈ N let τΛ (m) be the irreducible represen-
tation of SO0 (p, q) with highest weight mΛ. There exists a polynomial
PΛ (m) whose coefficients depend only on Λ, such that for all m ∈ N we
have
(2)
log TX (τΛ (m)) = Cp.q vol(X)PΛ (m).
Moreover, there is a constant CΛ > 0, which depends on Λ, such that
(6.23) PΛ (m) = CΛ · m dim(τΛ (m)) + O (dim(τΛ (m)))
± ±
as m → ∞. If Λ = ωf,n , where ωf,n are as in (6.15), then CΛ = 1.
Proof. Let Λ = τ1 e1 + · · · + τn+1 en+1 . By (6.14) and (6.21) we may

assume that τn+1 > 0. Put τ (m) := τΛ (m). Then
(6.24) λτ (m),k = mτk+1 + n − k, k = 0, . . . , n,
and by Proposition 6.6, (6.20), and (6.17), we have
(2)
log TX (τ (m))
n Z
|Wm | X λτ (m),k
=2π vol(X)(−1)v (−1)k Pστ (m),k (t) dt.
|WKM | 0
k=0
In the hyperbolic case, the term (−1)v |Wm |/|WKM | equals 1. Therefore
this equation agrees with [MP, (5.16), (5.17)]. Note that n = dim n. Let
cXe be defined by (6.9) and put
n Z λτ (m),k
(−1)n X k
(6.25) PΛ (m) := (−1) Pστ (m),k (t) dt.
cX̃ 0
k=0
Then it follows from (6.10) and (6.20) that PΛ is a polynomial in m
whose coefficients depend only on Λ. By definition, one has
(2) |Wm |
log TX (τ (m)) = 2π vol(X)(−1)v+n c PΛ (m).
|WKM | X̃
So it remains to compute the constant. By (6.9) and Lemma 6.1, one
has
|Wm | |Wm | 1
c = .
|WKM | X̃ |Wkm | 2 vol(X̃d )
Recall that mC ∼= so(2n, C), (km )C ∼
= so(2p1 , C) ⊕ so(2q1 , C) and so by
[Kn2, page 685], one has |Wm | = n!2n−1 , |Wkm | = p1 !q1 !2n−1−ǫ(q) , where
ǫ(q) is as above. Thus as in [Ol, Proposition 1.3], one has
p+q−2
|Wm | ǫ(q) 2
=2 p−1 .
|Wkm | 2
Furthermore, one has v = dim2 pm = (p−1)(q−1)

2 and thus we get v + n =
pq−1
2 . This proves the first part of the proposition.
To determine the highest order term of the polynomial PΛ (m), we
proceed as in [MP, Lemma 5.4] to show that
n
Y t2 − λ2τ (m),j
Pστ (m),k (t) = (−1)n+k cXe dim(τ (m)) .
λ2
j=0 τ (m),k
− λ2τ (m),j
j6=k
Denote the product on the right by Πk (t; m). Then it follows from (6.25)
that
Xn Z λτ (m),k
(6.26) PΛ (m) = dim(τ (m)) · Πk (t; m) dt.
k=0 0
To deal with the sum, we follow [BV, 5.9.1]. Put λτ (m),n+1 = 0. Then
λτ (m),k , k = 0, . . . , n + 1 is a strictly decreasing sequence. For k =
0, . . . , n set
Xk
Qk (t; m) := Πj (t; m).
j=0
Then Qk (t; m) is the unique even polynomial of degree ≤ 2n which

satisfies
(
1, if j ≤ k,
(6.27) Qk (±λτ (m),j ) =
0, if n ≥ j > k.
Moreover, we have
Xn Z λτ (m),k n Z
X λτ (m),k
(6.28) Πk (t; m) dt = Qk (t; m) dt.
k=0 0 k=0 λτ (m),k+1
As proved in [BV, section 5.9.1], each integral on the right is positive.

This can be seen as follows. By (6.27), the polynomial Q′k has a root
in each interval [λστ (m),j+1 , λστ (m),j ], [−λστ (m),j , −λστ (m),j+1 ] for 1 ≤ j <
n, j 6= k and a root in [−λστ (m),n , λστ (m),n ]. Since Q′k is of degree ≤
2n − 1, it follows that Qk is either constant or strictly increasing on
[λστ (m),k+1 , λστ (m),k ]. Furthermore, Qn (t; m) is a polynomial of degree
2n, which is equal to 1 at 2n + 2 pairwise distinct points. Hence Qn ≡ 1.
Thus by (6.24) and (6.28) we get
n
X
(n + 1)(mτ1 + n) = (n + 1)λτ (m),0 ≥ (λτ (m),k − λτ (m),k+1 )
k=0
Xn Z λτ (m),k
(6.29) ≥ Πk (t; m) dt ≥ τn+1 m.
k=0 0
Since PΛ (m) is a polynomial in m, it follows that there exists CΛ ≥

τn+1 > 0 such that (6.23) holds. If Λ is one of the fundamental weights
±
ωf,n , defined by (6.15), then it follows as in [MP, section 5] that CΛ = 1.
This proves the second part of the proposition. q.e.d.
Finally we turn to the case G = SL3 (R), K = SO(3). We define our

fundamental Cartan subalgebra as follows. Let
H1 := diag(1, 1, −2); a := RH1 .
Then we have m = sl2 (R), if sl2 (R) is embedded into g as an upper left
block. Let

0 1
H2 := , t := RT1 ,
−1 0
embedded into g as an upper left block. Then t is a Cartan subalgebra

of m and
(6.30) h := a ⊕ t
is a θ-stable fundamental Cartan subalgebra of g. Note that h is different
from the usual Cartan subalgebra h̃ of g which consists of all diagonal
matrices of trace 0. Define f1 ∈ a∗ and f2 ∈ it∗ by
f1 (H1 ) = 3; f2 (H2 ) = i.
We fix f1 as a positive restricted root of a. Then we can define positive
roots by
∆+ (gC , hC ) := {f1 − f2 , f1 + f2 , 2f2 }; ∆+ (mC , tC ) = {2f2 }.
Under our normalization, one has
1
(6.31) hf1 , f1 i = 1; hf2 , f2 i = ; hf1 , f2 i = 0.
3
One easily sees that dim n = 2; hence n = 1. Moreover, by [Kn2, page
485], one has |W (A)| = 1. For k ∈ N let σk ∈ Rep(M 0 ) be of highest
weight kf2 . Then it follows from (6.10) and (6.9) that
!
9 2 k+1 2
(6.32) Pσk (z) = − (k + 1) z − .
8 vol(X̃d ) 3
P
Define ei ∈ h̃∗C by ei (diag(t1 , t2 , t3 )) = j δi,j tj . Then one can choose
positive roots
(6.33) ∆+ (gC , h̃C ) := {e1 − e2 , e1 − e3 , e2 − e3 }
and there is a standard inner-automorphism Φ of gC which sends hC to
h̃C and which satisfies
(6.34)
Φ∗ (e1 − e2 ) = 2f2 ; Φ∗ (e1 − e3 ) = f1 + f2 ; Φ∗ (e2 − e3 ) = f1 − f2 .
The fundamental weights ω̃1 , ω̃2 ∈ h̃∗C are given by
2 1
ω̃1 = (e1 − e2 ) + (e2 − e3 )
3 3
and
1 2
ω̃2 = (e1 − e2 ) + (e2 − e3 ).
3 3
Thus the fundamental weights ω1 , ω2 ∈ h∗C are given by
1 2
(6.35) ω1 := Φ∗ (ω̃1 ) = f1 + f2 ; ω2 := Φ∗ (ω̃2 ) = f1 .
3 3
Let N0 := N ⊔ {0}. If Λ is a weight, Λ = τ1 ω1 + τ2 ω2 , τ1 , τ2 ∈ N0 , then
a standard computation shows that
(6.36) Λθ = τ2 ω1 + τ1 ω2 .
Now we fix τ1 , τ2 ∈ N0 , τ1 + τ2 > 0, and for m ∈ N we let τ (m) be the

representation of G with highest weight
(6.37) Λ(τ (m)) := mτ1 ω1 + mτ2 ω2 .
We let W̃g be the Weyl-group of ∆(gC , h̃C ). Then W̃g consists of all
permutations of e1 , e2 , e3 . Let
W̃ 1 := (Φ∗ )−1 W 1 = {w ∈ W̃g : w−1 (e1 − e2 ) > 0}.

Then one has
{(w, ℓ(w)); w ∈ W̃ 1 }

e1 e2 e3 e1 e2 e3
= (Id, 0); ,1 ; ,2 .
e1 e3 e2 e3 e1 e2
By a direct computation we get
(6.38)
{w(Λ(τ (m)) + ρ̃G ), ℓ(w); w ∈ W̃ 1 }

2mτ1 + mτ2 + 3 mτ1 + 2mτ2 + 3
= (e1 − e2 ) + (e2 − e3 ); 0 ,
3 3

2mτ1 + mτ2 + 3 mτ1 − mτ2
(e1 − e2 ) + (e2 − e3 ); 1 ,
3 3

−mτ1 + mτ2 −2mτ1 − mτ2 − 3
(e1 − e2 ) + (e2 − e3 ); 2 .
3 3
As in [BV, 5.9.2], we introduce the following constants:
mτ1 + 1 mτ1 + mτ2 + 2
A1 (τ (m)) := ; A2 (τ (m)) := ;
(6.39) 2 2
mτ2 + 1
A3 (τ (m)) :=
2
and
mτ1 + 2mτ2 + 3 mτ1 − mτ2
C1 (τ (m)) := ; C2 (τ ) := ;
(6.40) 3 3
2mτ1 + mτ2 + 3
C3 (τ ) := .
3
Note that on h̃∗C , one has ω̃1 = e1 ; ω̃2 = e1 + e2 , since the matrices in h̃∗C
have trace 0. Then, combining (6.34) and (6.38), we get

{ Λ(στ (m),w ), λτ (m),w , ℓ(w) ; w ∈ W 1 } =
{((2A1 (τ (m)) − 1)f2 , C1 (τ (m)), 0) , ((2A2 (τ (m)) − 1)f2 , C2 (τ (m)), 1) ,
((2A3 (τ (m)) − 1)f2 , −C3 (τ (m)), 2)}.
Thus if we apply (6.32), we obtain

X Z |λτ (m),w |
ℓ(w)
(−1) Pστ (m),w (t)dt
0
w∈W 1
3 Z
CSL3 (R) X k+1
|Ck (τ (m))|
9 2 2
=− (−1) Ak (τ (m)) t − Ak (τ (m)) dt
vol(X̃d ) k=1 0 4
(6.41)
X3
Ak (τ (m))|Ck (τ (m))|
=− (−1)k+1 3Ck (τ (m))2 − 4Ak (τ (m))2 .
k=1
4 vol(X̃d )
We can now prove our main result about the L2 -torsion for the case
G = SL3 (R).
Proposition 6.8. Let X e = SL(3, R)/ SO(3) and X = Γ\X. e Let
∗
Λ ∈ hC be a highest weight with Λθ 6= Λ. For m ∈ N let τΛ (m) be the
irreducible representation of SL(3, R) with highest weight mΛ. There
exists a polynomial PΛ whose coefficients depend only on Λ such that
(2) π vol(X)
log TX (τΛ (m)) = P (m).
vol(Xed ) Λ
Moreover, there exists a constant C(Λ) > 0 depending only on Λ such
that
PΛ (m) = C(Λ)m dim(τΛ (m)) + O(dim(τΛ (m)))
as m → ∞. If Λ equals one of the fundamental weights ωf,i , then C(Λ) =
4/9.
Proof. There exist τ1 , τ2 ∈ N0 , τ1 6= τ2 , such that Λ = τ1 ω1 + τ2 ω2 .
Put τ (m) := τΛ (m). Then by Proposition 6.6, (6.39), (6.40), and (6.41),
the first statement is proved and it remains to consider the asymptotic
behavior of the polynomial PΛ . We differentiate two cases. First we
assume that τ1 τ2 6= 0. Then if we put
( τ 4 2τ 3 τ τ 2τ 2
− 182 + 19 2 + 13 2 ; τ1 ≥ τ2
α4 (τ ) := τ4 2τ 3 τ τ 2τ 2
− 181 + 29 1 + 13 2 ; τ2 ≥ τ1 ,
an explicit computation using (6.39), (6.40), and (6.41) shows that
X Z |λτ (m),w |
ℓ(w) α4 (τ ) 4
(−1) Pστ (m),w (t)dt = − m + O(m3 )
w∈W 1 0 vol(X̃ d )
as m → ∞. Note that α4 (τ ) > 0 by our assumption on τ1 and τ2 . Now
we assume that τ1 τ2 = 0. Then if we define
2(τ13 + τ23 )
α3 (τ ) := ,
9
an explicit computation using (6.39), (6.40), and (6.41) gives

X Z λτ (m),w
ℓ(w) α3 (τ ) 3
(−1) Pστ (m),w (t)dt = − m + O(m2 )
w∈W 1
0 vol(X̃ d )
as m → ∞. For SL3 (R), one has v = 1, and using Lemma 6.1, one
|Wm |
gets |W KM |
= 1. Moreover, every element of Rep(M 0 ) is self-dual. Thus,
using Proposition 6.6, we obtain
(2) πα4 (τ ) 4
log TX (τ (m)) = vol(X) m + O(m3 )
vol(X̃d )
as m → ∞, if τ1 τ2 6= 0, and
(2) πα3 (τ ) 3
log TX (τ (m) = vol(X) m + O(m2 )
vol(X̃d )
as m → ∞, if τ1 τ2 = 0. Now we define constants

τ12 τ2 + τ22 τ1 4τ1 τ2 + τ12 + τ22
d3 (τ ) := ; d2 (τ ) := .
2 2
Then by Weyl’s dimension formula, one has
dim τ (m) = d3 (τ )m3 + d2 (τ )m2 + O(m)
as m → ∞. Note that d3 (τ ) > 0 for τ1 τ2 = 6 0 and that d3 (τ ) = 0,
d2 (τ ) > 0 for τ1 τ2 = 0. This completes the proof of the proposition.
q.e.d.
7. Lower bounds of the spectrum

In this section we assume that X e is odd-dimensional and that
e = 1. Our goal is to establish the lower bound (1.11) for the spec-
δ(X)
trum of the Laplace operators ∆p (τλ (m)). To this end we use (5.4),
which reduces the problem to the estimation from below of the endo-
morphism Ep (τλ (m)).
First we introduce some notation. Let X e = G/K. There is a de-
composition X e = X e0 × X e1 with δ(X e0 ) = 0 and Xe1 is an irreducible
e e
symmetric space with δ(X1 ) = 1. Since X0 is even-dimensional, the
dimension of Xe1 is odd. Let G = G0 × G1 be the corresponding decom-
position of G. Then δ(G0 ) = 0 and by the classification of simple Lie
groups, G1 = SO0 (p, q), p, q odd, or G1 = SL(3, R). Let gi , i = 0, 1 be
the Lie algebra of Gi . Let t0 ⊂ g0 be a compact Cartan subalgebra and
let h1 ⊂ g1 be a fundamental Cartan subalgebra. Then h1 is of split
rank one. Put
h := t0 ⊕ h1 .
Then h is a Cartan subalgebra of split rank one. Let (τ, Vτ ) ∈ Rep(G)
with highest weight λ ∈ h∗C . Then λ = λ0 + λ1 , where λ0 ∈ t∗0,C and λ1 ∈
h∗1,C are highest weights. Let θ : g → g be the Cartan involution. Assume

that λθ 6= λ. Then λ1 satisfies (λ1 )θ 6= λ1 . Let (τi , Vτi ) ∈ Rep(Gi ),
i = 0, 1, be the representations with highest weight λi . Then τ ∼ = τ0 ⊗τ1 .
Let
gi = ki ⊕ pi
be the Cartan decomposition of gi , i = 0, 1. We may choose p such that
p = p0 ⊕ p1 . Then we have
M
Λp p∗ ⊗ Vτ ∼
= (Λr p∗0 ⊗ Vτ0 ) ⊗ (Λs p∗1 ⊗ Vτ1 ) .
r+s=p
Let Ωi ∈ Z(gi,C ), i = 1, 2, be the Casimir operator of gi . Then Ω =

Ω0 ⊗ Id + Id ⊗Ω1 . Similarly, we have ΩK = Ω0,K ⊗ Id + Id ⊗Ω1,K . Set
νi,p (τi ) := Λp Ad∗pi ⊗τi : Ki → GL(Λp p∗i ⊗ Vτi ), i = 0, 1.
Let
(7.1) Ei,p (τi ) := τi (Ωi ) Idi −νp (τi )(Ωi,K ), i = 0, 1,
be the corresponding endomorphisms acting in Λp p∗i
⊗ Vτi . Then it fol-
lows that
M
(7.2) Ep (τ ) = (E0,r (τ0 ) ⊗ Id + Id ⊗E1,s (τ1 )) .
r+s=p
Therefore it suffices to estimate Ei,p (τi ), i = 0, 1.

Let us first recall the general formula for the Casimir eigenvalues.
We let g be a semisimple real Lie algebra with Cartan decomposition
g = k ⊕ p. Let t be a Cartan subalgebra of k and let h = t ⊕ b, b ⊂ p be a
θ-stable Cartan subalgebra of g containing t. Let the associated groups
G and K be as in the introduction. Let k·k denote the norm induced by
the (suitably normalized) Killing form on the real vector space b∗ ⊕ it∗ .
Fix positive roots ∆+ (gC , hC ), ∆+ (kC , tC ) and let ρG , resp. ρK , be the
half-sum of the positive roots. Let τ be an irreducible finite-dimensional
complex representation of G with highest weight Λ(τ ) ∈ b∗ ⊕ it∗ and
let ν be an irreducible unitary representation of K with highest weight
Λ(ν) ∈ it∗ . Then we have
(7.3)
τ (Ω) = kΛ(τ ) + ρG k2 − kρG k2 ; ν(ΩK ) = kΛ(ν) + ρK k2 − kρK k2 .
Now we prove the following general bound, which we use to deal with
E0,p (τ0 ).
Lemma 7.1. Let λ ∈ h∗C be a highest weight. Given m ∈ N, let
τλ (m) be the irreducible representation with highest weight mλ. There
exists C > 0 such that
Ep (τλ (m)) ≥ −Cm
for all p = 0, . . . , d and m ∈ N.
Proof. Let τ ∈ Rep(G) be of highest weight Λ(τ ). Let ν ′ ∈ K̂ with

highest weight Λ(ν ′ ) ∈ it∗ . Assume that [τ |K : ν ′ ] 6= 0. We claim that
there is a weight λ of τ such that Λ(ν ′ ) = λ|t . To see this, let Vτ be
the space of the representation τ and let Vτ (Λ(ν ′ )) be the eigenspace
of t with eigenvalue Λ(ν ′ ). Then Vτ (Λ(ν ′ )) is invariant under h. So it
decomposes into joint eigenspaces of h. Let λ be the weight of one of
these eigenspaces. Then λ|t = Λ(ν ′ ). Now we note that as a weight of τ ,
λ belongs to the convex hull of the Weyl group orbit of Λ(τ ) (see [Ha,
Theorem 7.41]). Thus we get
(7.4) kΛ(τ )k ≥ kλk ≥ kλ|t k = kΛ(ν ′ )k.
Now let ν ∈ K b with [νp (τ ) : ν] 6= 0. Then by [Kn2, Proposition 9.72]
′
there exists ν ∈ K b with [τ |K : ν ′ ] 6= 0 of highest weight Λ(ν ′ ) ∈ it∗ and
µ ∈ it∗ , which is a weight of νp such that the highest weight Λ(ν) of ν
is given by µ + Λ(ν ′ ). Since Λ(τ ) is dominant, we have
kΛ(τ ) + ρG k2 ≥ kΛ(τ )k2 .
Thus by (7.4) we get
kΛ(τ ) + ρG k2 − kΛ(ν) + ρK k2
≥ kΛ(τ )k2 − kΛ(ν ′ )k2 − 2kµ + ρK k · kΛ(ν ′ )k − kµ + ρK k2
≥ −2kµ + ρK k · kΛ(τ )k − kµ + ρK k2 .
There is C > 0 such that kµ + ρK k ≤ C for all weights µ of νp . Hence
there is C1 > 0 such that for all τ ∈ Rep(G), one has
(7.5) kΛ(τ ) + ρG k2 − kΛ(ν) + ρK k2 ≥ −C1 (kΛ(τ )k + 1)
for all ν ∈ K b with [νp (τ ) : ν] 6= 0. Now we apply this to τλ (m). By
definition of τλ (m), we have Λ(τλ (m)) = mλ. Using equations (7.5) and
(7.3), the lemma follows. q.e.d.
Now we turn to the estimation of E1,p (τ1 ). In this case we have either
G1 = SO0 (p, q), p, q odd, or G = SL(3, R). We deal with these cases
separately.
7.1. The case G = SO0 (p, q). Let p = 2p1 + 1, q = 2q1 + 1. Let
n := p1 + q1 . Let K = SO(p) × SO(q) and X e = G/K. We let t and h
be as in section 6. Also the Killing form will be normalized as in this
section. Then we have the following lemma.
Lemma 7.2. Let Λ ∈ h∗C be given as Λ = k1 e1 + · · · + kn+1 en+1 ,
k1 ≥ k2 ≥ · · · ≥ kn+1 ≥ 0. Let Λ′ ∈ h∗C belong to the convex hull of the
set {wΛ, w ∈ WG } and let λ ∈ it∗ be given by λ := Λ′ |t . Then one has
n
X
kλk2 ≤ ki2 .
i=1
Proof. Recall that the Weyl group WG consists of permutations and

even sign α1 , . . . , αm ∈
Pmchanges of the e1 , . . . , en+1 . Thus there exist n+1
(0, 1), j=1 αj = 1, and for each j = 1, . . . , m, a σj ∈ S , the sym-
metric group, and a sequence ǫj,1 , . . . , ǫj,n+1 ∈ {±1} such that
m n+1
!
X X
Λ′ = αj ǫj,i ki eσj (i) .
j=1 i=1
Thus one has

 
m
X n+1
X
 
λ= αj 
 ǫj,i ki eσj (i) 

j=1 i=1
σj (i)6=p1 +1
and so one gets

v
Xm n+1 X m u n+1
X u X
kλk ≤ αj
ǫ k e
j,i i σj (i) = α u
jt ki2
j=1
σj (i)6i=1
=p1 +1 j=1 i=1
σj (i)6=p1 +1
v v
m u n u n
X uX uX
≤ αj t ki = t
2 ki2 .
j=1 i=1 i=1
q.e.d.
Now we let Λ(τ ) ∈ h∗C be given by

Λ(τ ) := τ1 e1 + · · · + τn+1 en+1 , τ1 ≥ τ2 ≥ · · · ≥ τn+1 > 0.
For m ∈ N we let τ (m) be the representation of G with highest weight
Λ(τ (m)) := mΛ(τ ).
Then we have the following proposition.
Proposition 7.3. There exists a constant C such that
Ep (τ (m)) ≥ m2 τn+1 − Cm
for all m.
Proof. Recall that νp (τ (m)) = τ (m)|K ⊗ νp . Let ν ∈ K̂ be such that
[νp (τ (m)) : ν] 6= 0. By [Kn2, Proposition 9.72], there exists a ν ′ ∈ K̂
with [τ (m) : ν ′ ] 6= 0 of highest weight λ(ν ′ ) ∈ b∗C , and a µ ∈ b∗C which
is a weight of νp such that the highest weight λ(ν) of ν is given by
µ + λ(ν ′ ). Now there is a Λ e ∈ h∗ which is a weight of τ (m) such that
C
e t . By [Ha, Theorem 7.41], Λ
λ(ν ′ ) = Λ| e belongs to the convex hull of the
Weyl group orbit of Λ(τ (m)). Thus applying (7.3) and Lemma 7.2, we
obtain constants C1,2 which are independent of m such that
2
ν(ΩK ) = kλ(ν) + ρK k2 − kρK k2 ≤ λ(ν ′ ) + C1 (1 + λ(ν ′ ) )
 
Xn
≤ m2  τj2  + C2 m.
j=1
On the other hand, by (7.3) we have

τ (m)(Ω) = kΛ(τ (m)) + ρG k2 − kρG k2
n+1
X n+1
X n+1
X
2 2 2
= (mτj + n + 1 − j) − (n + 1 − j) ≥ m τj2 .
j=1 j=1 j=1
This implies the proposition. q.e.d.
7.2. The case G = SL(3, R). We use the notation of section 6. We
choose the Cartan subalgebra h ⊂ g, which is defined by (6.30). The
fundamental weights ωi ∈ h∗C , i = 1, 2, are given by (6.35). Let Λ ∈ h∗C be
a highest weight. For m ∈ N let τΛ (m) be the irreducible representation
with highest weight mΛ.
Proposition 7.4. Assume that Λ satisfies Λθ 6= Λ. Then there exists
CΛ > 0 such that
1
Ep (τΛ (m)) ≥ m2 − CΛ m
9
for all m ∈ N and p = 0, . . . , 5.
Proof. There exist τ1 , τ2 ∈ N0 such that Λ = τ1 ω1 + τ2 ω2 . Note that
by (6.33) and (6.34), one has ρG = f1 + f2 . Then by (6.35) and (6.31)
we get
τΛ (m)(Ω) = kmΛ + ρG k2 − kρG k2
4(τ12 + τ1 τ2 + τ22 ) 2 4(τ1 + τ2 )
= m + m.
9 3
Next recall that there is a natural isomorphism kC ∼ = su(2)C = sl(2, C)
(see [Ha, section 4.9]). Furthermore, if we embed sl(2, C) into gC as an
upper left block, then tC is isomorphic to a Cartan subalgebra of sl(2, C).
For j ∈ N we let νj denote the representation of kC with highest weight
jf2 . Then we deduce from the branching law from GL3 (C) to GL2 (C)
[GW, Theorem 8.1.1] that
M mτ2
mτ1 M
τΛ (m)|kC = νj+k .
j=0 k=0
If we use
j2 2
νj (ΩK ) = + j
3 3
and argue as in the proof of Proposition 7.3, we obtain a constant C

which is independent of τ1 , τ2 and m such that for every ν ∈ K̂ with
[νp (τ (m)) : ν] 6= 0 for some p, one has
(m(τ1 + τ2 ) + C)2 2 (m(τ1 + τ2 ) + C)

ν(ΩK ) ≤ + .
3 3
Thus we obtain a constant CΛ such that for every m and every p, one
has
(τ1 − τ2 )2 2
Ep (τΛ (m)) ≥ m − CΛ m.
9
By (6.36) the condition Λθ 6= Λ is equivalent to τ1 6= τ2 . This proves the
proposition. q.e.d.
Now we can summarize our results.

Proposition 7.5. Let δ(X) e = 1, X e odd-dimensional. Let λ ∈ h∗ be
C
a highest weight with λθ 6= λ. For m ∈ N let τλ (m) be the irreducible
representation of G with highest weight mλ. There exist C1 , C2 > 0 such
that
Ep (τλ (m)) ≥ C1 m2 − C2
for all p = 0, . . . , d and m ∈ N.
Proof. Let λ = λ0 + λ1 with λ0 ∈ t∗0,C and λ1 ∈ h∗1,C highest weights,
and (λ1 )θ 6= λ1 . Let τi (m), i = 0, 1, be the irreducible representa-
tions of Gi with highest weight mλi . Then τ (m) = τ0 (m) ⊗ τ1 (m).
Let E0,p (τ0 (m)) and E1,p (τ1 (m)) be defined by (7.1). By Lemma 7.1,
there exists C > 0 such that
E0,p (τ0 (m)) ≥ −Cm
for all p = 0, . . . , d and m ∈ N. Furthermore, by Proposition 7.3 and
Proposition 7.4, there exist C3 , C4 > 0 such that
E1,p (τ1 (m)) ≥ C3 m2 − C4
for all p = 0, . . . , d and m ∈ N. Combined with (7.2), the proof follows.
q.e.d.
Corollary 7.6. Let the assumptions be as in Proposition 7.5. There
exist constants C1 , C2 > 0 such that
∆p (τλ (m)) ≥ C1 m2 − C2
for all p = 0, . . . , d and m ∈ N.
Proof. Recall that the Bochner-Laplace operator satisfies ∆νp (τ (m)) ≥
0. Hence the corollary follows from (5.4) and Proposition 7.5. q.e.d.
8. Proof of the main results

First assume that δ(X) e 6= 0. Note that δ(X)
e = 0 implies that dim X e
is even. Hence, it follows from Proposition 4.2 that TX (τ ) = 1 for all
finite-dimensional irreducible representations of G, which proves part
(i) of Theorem 1.1.
Now assume that δ(X) e = 1. Let h ⊂ g be a fundamental Cartan
∗
subalgebra. Let λ ∈ hC be a highest weight with λθ 6= λ. For m ∈ N
let τ (m) be the irreducible representation of G with highest weight
mλ. Then τ (m) 6∼ = τ (m)θ for all m ∈ N. Hence by [BW, Chapter VII,
Theorem 6.7] we have H p (X, Eτ (m) ) = 0 for all p = 0, . . . , d. Then by
(4.7) we have
Z ∞
1 d 1
(8.1) log TX (τ (m)) = t K(t, τ (m)) dt .
s−1
2 ds Γ(s) 0 s=0
Since τ (m) is acyclic and dim X is odd, TX (τ (m)) is metric indepen-

√
dent [Mu2, Corollary 2.7]. Especially, we can rescale the metric by m
without changing TX (τ (m)). Equivalently, we can replace ∆p (τ (m)) by
1
m ∆p (τ (m)). Using (8.1) we get
Z ∞
1 d 1 t
t K , τ (m) dt .
2 ds Γ(s) 0 m s=0
To continue, we split the t-integral into the integral over [0, 1] and the
integral over [1, ∞). This leads to
Z 1
1 d 1 t
t K , τ (m) dt
2 ds Γ(s) 0 m s=0
(8.2) Z ∞
1 t
+ t−1 K , τ (m) dt.
2 1 m
We first consider the second term on the right hand side. To this end,
we need the following lemma.
τ (m), p
Lemma 8.1. Let ht be defined by (4.13) and let Ht0 be the heat
kernel of the Laplacian ∆ e 0 on C ∞ (X).
e There exist m0 ∈ N and C > 0
such that for all m ≥ m0 , g ∈ G, t ∈ (0, ∞), and p ∈ {0, . . . , d}, one
has
m2
τ (m), p
ht (g) ≤ C dim(τ (m))e−t 2 Ht0 (g).
ν (τ (m)) e
Proof. Let p ∈ {0, . . . , n}. Let Ht p be the kernel of e−t∆νp (τ (m))
τ (m), p e
and let Ht be the kernel of e−t∆p (τ (m)) . By (5.7) we have
τ (m), p ν (τ (m))
Ht (g) = e−tEp (τ (m)) ◦ Ht p (g).
Thus by Proposition 3.1 and Proposition 7.5 there exists an m0 such

that for m ≥ m0 , one has
m2
τ (m), p
(8.3) Ht (g) ≤ e−t 2 Ht0 (g).
Taking the trace in End(Λp p∗ ⊗ Vτ (m) ) for every p ∈ {0, . . . , d}, the
lemma follows. q.e.d.
Using (4.16), (4.15), and Lemma 8.1, we obtain
Z
t X
K , τ (m) ≤ Ce− m2 t dim(τ (m)) 0
Ht/m (g−1 γg) dġ
m
Γ\G γ∈Γ
−m t t
= Ce 2 dim(τ (m)) Tr(e− m ∆0 ).
Furthermore, by the heat asymptotic [Gi] we have
1

Tr(e− m ∆0 ) = Cd vol(X)md/2 + O m(d−1)/2
as m → ∞. Hence there exists C1 > 0 such that

K t , τ (m) ≤ C1 md/2 dim(τ (m))e− m2 t , t ≥ 1.
m
Thus we obtain
Z ∞ Z ∞
t m
t −1
K , τ (m) dt ≤ C1 md/2 dim(τ (m))e−m/4 t−1 e− 4 t dt.
m
1 1
Using Weyl’s dimension formula, it follows that
Z ∞
−1 t
(8.4) t K , τ (m) dt = O e−m/8 .
1 m
Now we turn to the first term on the right hand side of (8.2). We need
to estimate K(t, τ (m)) for 0 < t ≤ 1. To this end, we use (4.16) to
decompose K(t, τ (m)) into the sum of two terms: the contribution of
the identity
τ (m)
(8.5) I(t, τ (m)) := vol(X)kt (1),
τ (m)
where kt is defined by (4.15), and the remaining term
Z X τ (m)
H(t, τ (m)) := kt (g−1 γg) dġ.
Γ\G γ∈Γ
γ6=1
First we consider H(t, τ (m)). Using Proposition 8.1 and Proposition 3.2,
it follows that for every m ≥ m0 and every t ∈ (0, 1], we have
X τ (m)
m2 X
kt (g−1 γg) ≤ Ce−t 2 dim(τ (m)) Ht0 (g−1 γg)
γ∈Γ γ∈Γ
γ6=1 γ6=1
m2
≤ C1 dim(τ (m))e−t 2 e−c0 /t .
Hence, using Weyl’s dimension formula, we get

H t , τ (m) ≤ C2 e−c1 m e−c1 /t , 0 < t ≤ 1.
m
This implies that there is c2 > 0 such that

Z 1
d 1 t
s−1
t H , τ (m) dt
ds Γ(s) 0 m s=0
(8.6) Z 1
t
= t−1 H , τ (m) dt = O e−c2 m
0 m
as m → ∞.
It remains to consider the contribution of the identity I(t, τ (m)). By
Lemma 8.1 there exists C > 0 such that for all m ≥ m0 and p = 0, . . . , d
we have
τ (m),p m2
|ht (1)| ≤ C dim(τ (m))e−t 2 Ht0 (1).
Next we estimate Ht0 (1) using the Plancherel theorem. Since the func-
tion Ht0 (1) is K-biinvariant, the Plancherel theorem for Ht0 (1) reduces to
the spherical Plancherel theorem [He, Theorem 7.5]. Thus if Q = M AN
is a fixed minimal standard parabolic subgroup, it follows from (5.13)
that
Z
0 −tkρa k2 2
Ht (1) = e e−tkνk β(ν)dν,
a∗
where β(ν) is the spherical Plancherel density. Thus there exists C1 > 0
such that |Ht0 (1)| ≤ C1 for t ≥ 1. Hence by (4.15) we get
τ (m) m2
|kt (1)| ≤ C2 dim(τ (m))e−t 2
for t ≥ 1 and m ≥ m0 . By (8.5) and Weyl’s dimension formula, it follows

that there exist C, c > 0 such that

t

(8.7) I , τ (m) ≤ Ce−cm e−ct
m
for t ≥ 1 and m ≥ m0 . Hence we get

Z 1
d 1 t
ts−1 I , τ (m) dt
ds Γ(s) 0 m
s=0
(8.8) Z ∞
d 1 t
= ts−1 I , τ (m) dt + O e−cm
ds Γ(s) 0 m
s=0
for m ≥ m0 . Since we are assuming that δ(X) e = 1, dim(X) is odd. Then

τ (m) τ (m)
it follows from (5.11) and the definition of kt by (4.15) that kt (1)
has an asymptotic expansion of the form

∞
X
τ (m)
kt (1) ∼ cj t−d/2+j .
j=0
Since d is odd, the expansion has no constant term. This implies that
Z ∞
ts−1 I(t, τ (m)) dt
0
is holomorphic at s = 0. Therefore we get
Z ∞
d 1 t
ts−1 I , τ (m) dt
ds Γ(s) 0 m
s=0
Z ∞
d 1
= ts−1 I (t, τ (m)) dt .
ds Γ(s) 0
s=0
(2) (2)
By definition, the right hand side equals where TX (τ (m))
log TX (τ (m)),
is the L2 -torsion. Combined with (8.2), (8.4), and (8.6), we obtain
(2)
(8.9) log TX (τ (m)) = log TX (τ (m)) + O e−cm .
This proves Proposition 1.2. q.e.d.
Combining Proposition 5.3 with Proposition 6.7 and Proposition 6.8,
we obtain Proposition 1.3. Together with Proposition 1.2, we obtain
part (ii) of Theorem 1.1.
Corollary 1.4 follows from Proposition 6.7 and Corollary 1.5 follows
from Proposition 6.8.
References
[At] M.F. Atiyah, Elliptic operators, discrete groups and von Neumann alge-
bras, Colloque “Analyse et Topologie” en l’Honneur de Henri Cartan (Or-
say, 1974), pp. 43–72, Asterisque, No. 32-33, Soc. Math. France, Paris,
1976, MR 0420729, Zbl 0323.58015.
[BM] D. Barbasch & H. Moscovici, L2 -index and the trace formula, J. Funct.
An. 53 (1983), no. 2, 151–201, MR 0722507, Zbl 0537.58039.
[BMZ1] J.-M. Bismut, X. Ma & W. Zhang, Operateurs de Toeplitz et torsion ana-
lytique asymptotique, C. R. Math. Acad. Sci. Paris 349 (2011), no. 17–18,
977–981, MR 2838248, Zbl 1227.58010.
[BMZ2] J.-M. Bismut, X. Ma & W. Zhang, Asymptotic torsion and Toeplitz oper-
ators, Preprint, 2011.
[Bor] A. Borel, Compact Clifford-Klein forms of symmetric spaces, Topology 2
(1963), 111–122, MR 0447474, Zbl 0116.38603.
[Bo] R. Bott, The index theorem for homogeneous differential operators, 1965
Differential and Combinatorial Topology (A Symposium in Honor of
Marston Morse), pp. 167–186, Princeton Univiversity Press, Princeton,
N.J., MR 0182022, Zbl 0173.26001.
[BV] N. Bergeron & A. Venkatesh, The asymptotic growth of torsion homology

for arithmetic groups, http://arxiv.org/abs/1004.1083 (2010).
[BW] A. Borel & N. Wallach, Continuous cohomology, discrete subgroups, and
representations of reductive groups, Princeton University Press, Princeton,
1980, MR 0554917, Zbl 0443.22010.
[De] P. Delorme, Formules limites et formules asymptotiques pour les multi-
plicites dans L2 (G/Γ), Duke Math. J. 53 (1986), no. 3, 691–731, MR
0860667, Zbl 0623.22012.
[DL] H. Donnelly & P. Li, Lower bounds for the eigenvalues of Riemann-
ian manifolds, Michigan Math. J. 29 (1982), 149–161, MR 0654476, Zbl
0488.58022.
[Do1] H. Donnelly, Asymptotic expansions for the compact quotients of properly
discontinuous group actions, Illinois J. Math. 23 (1979), no. 3, 485–496,
MR 0537804, Zbl 0411.53033.
[Fr] D. Fried, Analytic torsion and closed geodesics on hyperbolic manifolds,
Invent. Math. 84 (1986), no. 3, 523–540, MR 0837526, Zbl 0411.53033.
[Gi] P.B. Gilkey, Invariance theory, the heat equation, and the Atiyah-Singer
index theorem, Second edition, Studies in Advanced Mathematics. CRC
Press, Boca Raton, FL, 1995, MR 1396308, Zbl 0856.58001.
[GW] R. Goodman & N. Wallach, Representations and invariants of the classical
groups, Encyclopedia of Mathematics and Its Applications, 68. Cambridge
University Press, Cambridge, 1998, MR 1606831, Zbl 0901.22001.
[Gu] B. Güneysu, Kato’s inequality and form boundedness of Kato potentials
on arbitrary Riemannian manifolds, 2011, arXiv:1105.0532v3.
[Ha] B.C. Hall, Lie groups, Lie algebras, and representations, Graduate Texts in
Mathematics, 222, Springer-Verlag, Berlin, New York, 2003, MR 1997306,
Zbl 1026.22001.
[He] S. Helgason, Groups and geometric analysis, Integral geometry, invari-
ant differential operators, and spherical functions, Corrected reprint of
the 1984 original Mathematical Surveys and Monographs, 83, American
Mathematical Society, Providence, RI, 2000, MR 1790156, Zbl 0965.43007.
[HC] Harish-Chandra, Harmonic analysis on real reductive groups. III, The
Maass-Selberg relations and the Plancherel formula. Ann. of Math. (2)
104 (1976), no. 1, 117–201, MR 0439994, Zbl 0331.22007.
[Kn1] A. Knapp, Representation theory of semisimple groups, Princeton Univer-
sity Press, Princeton, 2001, MR 1880691, Zbl 0993.22001.
[Kn2] A. Knapp, Lie groups beyond an introduction, Second Edition, Progress
in Mathematics, 140. Birkhäuser Boston, Inc., Boston, MA, 2002, MR
1920389, Zbl 1075.22501.
[Ko] B. Kostant, Lie algebra cohomology and the generalized Borel-Weil the-
orem, Annals of Math. (2) 74 (1961), 329–378, MR 0142696, Zbl
0134.03501.
[Lo] J. Lott, Heat kernels on covering spaces and topological invariants, J. Dif-
ferential Geom. 35 (1992), no. 2, 471–510, MR 1158345, Zbl 0770.58040.
[LM] H.B. Lawson & M.-L. Michelsohn, Spin geometry, Princeton Mathematical
Series, 38. Princeton University Press, Princeton, NJ, 1989, MR 1031992,
Zbl 0688.57001.
[Ma] A. Manning, Topological entropy for geodesic flows, Ann. of Math. (2) 110
(1979), no. 3, 567–573, MR 0554385, Zbl 0426.58016.
[MaM] S. Marshall & W. Müller, On the torsion in the cohomology of arithmetic
hyperbolic 3-manifolds, Preprint 2011, arXiv:1103.2262.
[MtM] Y. Matsushima & S. Murakami, On vector bundle valued harmonic forms
and automorphic forms on symmetric Riemannian manifolds, Ann. of
Math. (2) 78 (1963), 365–416, MR 0153028, Zbl 0125.10702.
[Mi1] R.J. Miatello, The Minakshisundaram-Pleijel coefficients for the vector-
valued heat kernel on compact locally symmetric spaces of negative curva-
ture, Trans. Amer. Math. Soc. 260 (1980), no. 1, 1–33, MR 0570777, Zbl
0444.58015.
[Mi] J. Milnor, Whitehead torsion, Bull. Amer. Math. Soc. 72 (1966), 358–426,
MR 0196736, Zbl 0147.23104.
[MS1] H. Moscovici & R. Stanton, Eta invariants of Dirac operators on locally
symmetric manifolds, Inv. Math. 95 (1989), 629–666, MR 0979370, Zbl
0672.58043.
[MS2] H. Moscovici & R. Stanton, R-torsion and zeta functions for locally sym-
metric manifolds, Inv. Math. 105 (1991), 185–216, MR 1109626, Zbl
0789.58073.
[Mu1] W. Müller, The trace class conjecture in the theory of automorphic forms,
II. Geom. Funct. Anal. 8 (1998), no. 2, 315–355, MR 1616155, Zbl
1073.1151.
[Mu2] W. Müller, Analytic torsion and R-torsion for unimodular representations,
J. Amer. Math. Soc. 6 (1993), 721–753, MR 1189689, Zbl 0789.58071.
[Mu3] W. Müller, The asymptotics of the Ray-Singer analytic torsion of hyper-
bolic 3 manifolds, Preprint 2010, arXiv:1003.5168, to appear in: “Met-
ric and Differential Geometry,” The Jeff Cheeger Anniversary Volume,
Progress in Math. Vol. 297, pp. 317–352, Birkhäuser, 2012.
[MP] W. Müller, J. Pfaff, The asymptotics of the Ray-Singer analytic tor-
sion for hyperbolic manifolds, Intern. Math. Research Notices, 2012, doi:
10.1093/imrn/rns130.
[Ol] M. Olbrich, L2 -invariants of locally symmetric spaces, Doc. Math. 7
(2002), 219–237, MR 1938121, Zbl 1029.58019.
[RS] D.B. Ray & I.M. Singer, R-torsion and the Laplacian on Riemannian man-
ifolds, Advances in Math. 7 (1971), 145–210, MR 0295381, Zbl 0239.58014.
[Sh] M.A. Shubin, Pseudodifferential operators and spectral theory, Second edi-
tion. Springer-Verlag, Berlin, 2001, MR 1852334, Zbl 0980.35180.
[Wa1] N. Wallach, On the Selberg trace formula in the case of compact quo-
tient, Bull. Amer. Math. Soc. 82 (1976), no. 2, 171–195, MR 0404533, Zbl
0351.22008.
[Wa2] N. Wallach, Real reductive groups I, Pure and Applied Mathematics, 132,
Academic Press, Inc., Boston, MA, 1988, MR 0929683, Zbl 0666.22002.
[Wa3] N. Wallach, Real reductive groups II, Pure and Applied Mathematics, 132-
II, Academic Press, Inc., Boston, MA, 1992, MR 1170566, Zbl 0785.22001.
[Wr] G. Warner, Harmonic analysis on semi-simple Lie groups, I, Die
Grundlehren der mathematischen Wissenschaften, Band 188, Springer-
Verlag, New York-Heidelberg, 1972, MR 0498999, Zbl 0265.22020.
Universität Bonn
Mathematisches Institut
Endenicher Allee 60
D–53115 Bonn, Germany
E-mail address: mueller@math.uni-bonn.de
Universität Bonn
Mathematisches Institut
Endenicher Alle 60
D–53115 Bonn, Germany
E-mail address: pfaff@math.uni-bonn.de

J. Differential Geometry 95 (2013) 71-119: Received 5/25/2012

Uploaded by

Copyright:

Available Formats

J. Differential Geometry 95 (2013) 71-119: Received 5/25/2012

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

J. Differential Geometry 95 (2013) 71-119: Received 5/25/2012

Uploaded by

Copyright:

Available Formats

j.

ANALYTIC TORSION AND L2 -TORSION OF

Werner Müller & Jonathan Pfaff

(see [RS], [Mu2]). Since we have chosen distinguished metrics, we don’t

This problem was first studied in [Mu3] in the context of hyperbolic

Theorem 1.1. (i) Let X e be even dimensional or let δ(X)

as m → ∞. Furthermore, there is a constant Cλ > 0 such that

(1.4) log TX (τλ (m)) = C vol(X) · m dim(τλ (m)) + O (dim(τλ (m)))

Proposition 1.3. Let the assumptions be as in Proposition 1.2. There

as m → ∞. Let X ed be the compact dual of X.

Assume that τ |Γ is acyclic, that is, H ∗ (X, Eτ ) = 0. Then the analytic

(2) e χ(Xe0,d ) (2)

2.1. Let G be a real reductive Lie group in the sense of [Kn2, p.

be the induced representation acting by the left regular representation

If ν ∈ a∗ , then πξ,ν is unitarily induced. Denote by Θξ,ν the global

3. Bochner Laplace operators

Let ν be a finite-dimensional unitary representation of K on the space

be the associated locally homogeneous bundle over X. Since h̃ν and

commutes with the action of G. Therefore, it is of the form

It was proved in [Mu1, p. 325] that in the sense of distributions, one

Let e−t∆ν be the heat semigroup of ∆ν , acting on L2 (Γ\G, ν). Then

The convergence of the series in (3.11) can be established, for example,

Proof. For x, y ∈ Xe let ρ(x, y) denote the geodesic distance of x, y.

4. The analytic torsion

of locally homogeneous vector bundles. Let Λp (X, Eτ ) be the space of

RΓ denote the right regular representation of G in L2 (Γ\G). By [MtM,

Then K(t, τ ) − h(τ ) decays exponentially as t → ∞, and it follows from

heat kernel of ∆p (τ ). First the integral kernel of e−t∆p (τ ) , regarded as

As in section 3, this series converges absolutely and locally uniformly.

where tr denotes the trace tr : End(Vν ) → C. Let

Using (4.12) we obtain

Then it follows that

Let RΓ be the right regular representation of G on L2 (Γ\G). Then (4.16)

π̃(Htτ,p ) has the form

with π(Htτ,p ) acting on (Hπ ⊗ Λp p∗ ⊗ Vτ )K . Using (4.8), it follows as in

(4.18) π(Htτ,p ) = et(π(Ω)−τ (Ω)) Id

on (Hπ ⊗ Λp p∗ ⊗ Vτ )K . Let {ξn }n∈N and {ej }m

Now we consider the case of a principle series representation. Let Q be a

Proposition 4.1. Let Y ∈ a be a unit vector and let pY be the

Proof. By Frobenius reciprocity [Kn1, p. 208] and (4.20), we get

as KM -modules. Therefore, in the Grothendieck ring of KM we have

Tensoring with Wξ and Vτ and taking KM -invariants, we obtain (i).

Proposition 4.2. Assume that δ(X̃) ≥ 2 or assume that X̃ is even-

be the decomposition of the right regular representation RΓ of G on

Since d is even, it follows that in the representation ring R(K), we have

Let (π, Hπ ) ∈ Ĝ. Then it follows from (4.20) that

Let Hπ,K be the subspace of Hπ consisting of all smooth K-finite vectors.

Now by [BW, II.3.1, I.5.3] we have

In order to define the L2 -torsion, we need to know the asymptotic be-

With respect to this decomposition, we have

where σ(ΩK ) is the Casimir eigenvalue of σ and Vσ is the representation

By Proposition 3.2, there exist C1 , c1 > 0 such that

for 0 < t ≤ 1. Thus by (5.2),

as t → 0. To study the behavior of hτ,p t (1) as t → ∞, we use the

Let Q = M AN be a standard parabolic subgroup of G. Let (ξ, Wξ ) be

(5.12) c(ξ) := −hρa , ρa i + ξ(ΩM ).

Recall that for ν ∈ a∗ one has

(5.13) πξ,ν (Ω) = −hν, νi + c(ξ).

The exponents of the exponential factors in front of the integrals are