The Spectrum of the Laplacian in Riemannian

Geometry
Martin Vito Cruz
25 November 2003

1 Introduction
To any compact Riemannian manifold (M, g) (with or without boundary), we
can associate a second-order partial differential operator, the Laplace operator
∆, defined by ∆(f ) = −div(grad(f )) for f ∈ L2 (M, g). We will also sometimes
write ∆g for ∆ if we want to emphasize which metric the Laplace operator is
associated with. The set of eigenvalues of ∆ (the spectrum of ∆, or of M ),
which we will write as spec(∆) or spec(M, g), then forms a discrete sequence
0 = λ0 ≤ λ1 ≤ λ2 ≤ . . . → ∞. For simplicity, we will assume that M is
connected; this will for example imply that the smallest eigenvalue, λ0 , occurs
with multiplicity 1.
Note that the Laplacian also acts on p-forms in addition to functions via
the definition ∆ = −(dδ + δd), where δ is the adjoint of d with respect to the
Riemannian structure on the manifold. This aspect of the Laplacian will not be
treated in this paper, the focus being the ordinary Laplacian acting on functions.
With that in mind, there are two broad questions that are at the heart of
spectral geometry:
(i) What can we say about the spectrum of M given the geometry?
(ii) What can we say about the geometry of M given the spectrum?
What we will attempt to cover in this survey are some selected aspects of those
two questions. In section 2, we will discuss two aspects of the first question:
what types of sequences can occur as spectra and what restrictions does geome-
try impose on the first nonzero eigenvalue of the Laplacian. In section 3, we will
discuss some positive results relevant to the second question: what geometric
information can we extract from the spectrum? Finally, in section 4, we will
talk about some negative results relevant to the second question, focusing on ex-
amples of manifolds where the spectrum fails to allow us to distinguish between
them at all. Good background references on the topics covered in this paper
(and on other topics not covered) are the books [2] and [7], and the survey [9].

1
2 Estimates on the eigenvalues
Obviously, the geometry of a Riemannian manifold completely determines the
spectrum: the metric determines the Laplace operator and hence the spectrum.
On the other hand, there are only few examples of manifolds where the spec-
trum is known explicitly. In this section, we will examine the restrictions that
geometry can impose on the spectrum.

2.1 What sequences can be spectra?

An interesting first question to ask is what types of sequences can occur as
the spectra of manifolds. A version of this question has been answered: what
finite sequences can occur as the initial part of the spectra of manifolds? Colin
de Verdière showed that there are essentially no restrictions at all. If M is a
closed connected manifold of dimension greater than or equal to 3, then any
preassigned finite sequence 0 = λ0 < λ1 ≤ λ2 ≤ . . . ≤ λk is the sequence of
first k + 1 eigenvalues of ∆g for some choice of the metric g on M ([5], [6]).
In particular, this means that for closed connected manifolds of dimension 3 or
greater, there are no restrictions on the multiplicities of the eigenvalues λi for
i > 0.
In dimension 2, there are some restrictions on the multiplicities of the eigen-
values. Let M be a closed connected 2-manifold with Euler characteristic χ(M ),
and let mj be the multiplicity of the j-th eigenvalue, j > 0, of the Laplace op-
erator associated to a metric on M . Then
(i) if M is the unit sphere, then mj ≤ 2j + 1;
(ii) if M is the real projective plane, then mj ≤ 2j + 3;
(iii) if M is the torus, then mj ≤ 2j + 4;
(iv) if M is the Klein bottle, then mj ≤ 2j + 3;
(v) if χ(M ) < 0, then mj ≤ 2j − 2χ(M ) + 3.
See [18] for details. For finite sequences 0 = λ0 < λ1 < λ2 < . . . < λk however,
the result by Colin de Verdière holds even in dimension 2.

2.2 Estimates on the first eigenvalue

The geometry of a manifold affects more than just the multiplicities of the
eigenvalues. Here we will focus on bounds on the first non-zero eigenvalue λ1
imposed by the geometry.
The first lower bound is due to Lichnerowicz [16]:
Theorem 1 Let (M, g) be a closed Riemannian manifold of dimension n ≥ 2
and let Ric be its Ricci tensor field. If
Ric(X, X) ≥ (n − 1)k > 0
for some constant k > 0 and for all X ∈ T (M ), then λ1 ≥ nk.

2
 More recently, in 1979, Li and Yau [15] proved that if M is a closed Rieman-
2
nian manifold whose Ricci curvature is nonnegative, then λ1 ≥ 2Dπ2 (M ) , where
D(M ) is the diameter of M . This was achieved via a gradient estimate on the
first eigenfunction. Zhong and Yang [24], using a similar method, improved on
2
this by showing that under the same conditions, λ1 ≥ D2π(M ) .
Even more recently, Yang [23] generalized the previous result:
Theorem 2 Let (M, g) be a closed Riemannian manifold. If

Ric(X, X) ≥ (n − 1)k ≥ 0

for some nonnegative constant k and for all X ∈ T (M ), then

(n − 1)k π2
λ1 ≥ + 2 .
4 D (M )

It is in general much easier to give upper bounds on λ1 than it is to give lower

bounds. The basic result in this area is a comparison theorem due to Cheng
[3]: in a complete Riemannian n-manifold whose Ricci curvature is ≥ (n − 1)k
(where k is some constant), λ1 for a geodesic ball of radius r in M is less than
or equal to λ1 for a geodesic ball of the same radius in a space of constant
curvature k. From this, we can obtain the bound:
Theorem 3 If M is a compact n-manifold with Ricci curvature ≥ (n − 1)(−k),
k > 0, then
(n − 1)2 cn
λ1 ≤ k+ 2 ,
4 D (M )
where cn is a positive constant depending only on n.

3 Geometric implications of the spectrum

The spectrum does not in general determine the geometry of a manifold. Nev-
ertheless, some geometric information can be extracted from the spectrum. In
what follows, we define a spectral invariant to be anything that is completely
determined by the spectrum.

3.1 Invariants from the heat equation

Let M be a Riemannian manifold. A heat kernel, or alternatively, a fundamental
solution to the heat equation, is a function K : (0, ∞) × M × M → M that
satisfies
(i) K(t, x, y) is C 1 in t and C 2 in x and y;
∂K
(ii) ∂t + ∆2 (K) = 0, where ∆2 is the Laplacian with respect to the second
variable (i.e., the first space variable);

3
 R
(iii) limt→0+ M K(t, x, y)f (y) dy = f (x) for any compactly supported function
f on M .
The heat kernel exists and is unique for compact Riemannian manifolds. Its
importance stems from the fact that the solution to the heat equation
∂u
+ ∆(u) = 0, u : [0, ∞) × M → R,
∂t
(where ∆ is the Laplacian with respect to the second variable) with initial
condition u(0, x) = f (x) is given by
Z
u(t, x) = K(t, x, y)f (y) dy.
M

If {λi } is the spectrum of M and {ξi } are the associated eigenfunctions

(normalized so that they form an orthonormal basis of L2 (M )), then we can
write X
K(t, x, y) = e−λi t ξi (x)ξi (y).
i

From this, it is clear that the heat trace, Z(t) = M K(t, x, x) dx = i e−λi t , is
R P
a spectral invariant. The heat trace has an asymptotic expansion as t → 0+ :
∞
X
dim(M )/2
Z(t) = (4πt) aj tj ,
j=1

where the aj are integrals over M of universal homogeneous polynomials in the

curvature and its covariant derivatives ([17], see [8] or [7] for details). The first
few of these are
Z Z
1 1
a0 = vol(M), a1 = S, a2 = (5S 2 − 2|Ric|2 − 10|Rm|2 ),
6 M 360 M
where S is the scalar curvature, Ric is the Ricci tensor, and Rm is the curvature
tensor. The dimension, the volume, and the total scalar curvature are thus
completely determined by the spectrum. If M is a surface, then the Gauss-
Bonnet theorem implies that the Euler characteristic of M is also a spectral
invariant.
A more in depth study of the heat trace can yield more information. It is
known for example that if M is a closed, connected Riemannian manifold of
dimension n ≤ 6, and if M has the same spectrum as the n-sphere S n with the
standard metric (resp. RP n ), then M is in fact isometric to S n (resp. RP n ).
More on this can be found in [7].

3.2 Other invariants

There are other invariants besides those mentioned above. For generic closed
Riemannian manifolds for example, the geodesic length spectrum — the set of
lengths of closed geodesics — is a spectral invariant [4].

4
 As another way to get spectral invariants, we can try to study the fundamen-
2
tal solution
P iλk t to the wave equation ∂∂t2u + ∆(u) = 0 and the associated wave trace
ke . The asymptotics of the wave trace near t = 0 turns out to give the
same information as the asymptotics of the heat trace. On the other hand, the
singularities of the wave trace can in some cases yield new invariants such as the
geodesic length spectrum mentioned above and the quantum Birkhoff normal
form for the Poincaré map about certain geodesics (in fact, the singularities of
the wave trace occur at the lengths of closed geodesics; for definitions and a
better overview, see [1] and [9]).

4 Isospectral manifolds
As was alluded to earlier, geometry is not in general a spectral invariant. Two
manifolds are said to be isospectral if they have the same spectrum. The first
example of nonisometric isospectral manifolds was found in 1964 by John Milnor,
who exhibited two distinct but isospectral 16-dimensional manifolds. This was
followed by the construction in the 1980s and 1990s of many different examples
of nonisometric but isospectral manifolds. Among these are discrete families of
isospectral manifolds, continuous families of isospectral manifolds, isospectral
plane domains, and even isospectral conformally equivalent manifolds.
In general, there are three known methods to construct or discover these
examples of nonisometric isospectral manifolds. For a more complete overview
of these methods see [9].

4.1 Direct computation of the spectrum

The first of these is straightforward: direct computation. It is only rarely possi-
ble to explicitly compute the spectrum of a manifold, but the first examples of
isospectral manifolds were actually discovered via this method. Milnor’s exam-
ple mentioned above consists of two isospectral flat tori, flat tori — quotients
of Euclidean space by lattices of full rank — being one of the few examples
of Riemannian manifolds whose spectra can be computed explicitly. Spherical
space forms — quotients of spheres by finite groups of orthogonal transforma-
tions acting without fixed points — form another class of examples of manifolds
whose spectra is known. Some of the more remarkable examples of pairs of
isospectral manifolds are the pairs constructed by Ikeda of lens spaces which
are isospectral for the Laplacian acting on p-forms for p ≤ k but not for the
Laplacian acting on p-forms for p = k + 1 (recall that a lens space is a spherical
space form where the group is cyclic) [14].

4.2 Representation theoretic methods

The second method involves representation theoretic techniques. The two known
variants of this method actually produce examples of manifolds that are strongly
isospectral: any natural strongly elliptic operator on the manifolds (such as the

5
Laplace operator acting on p-forms) has the same spectrum. Let G be a Lie
group. We call subgroup Γ cocompact if Γ/G is compact. Define RΓ,a to be the
right translation operator on Γ/G, i.e., RΓ,a (Γx) = (Γxa). If we let L2 (Γ/G)
be the space of measurable functions that are square integrable with respect
to the Haar measure on Γ/G induced by the bi-invariant Haar measure on G,
and if we define ρΓ (a)f = f ◦ RΓ,a for f ∈ L2 (Γ/G) and a ∈ G, then ρΓ is a
unitary representation of G. Two cocompact discrete subgroups Γ1 and Γ2 of
G are said to be representation equivalent if there exists a unitary isomorphism
T : L2 (Γ1 /G) → L2 (Γ2 /G) such that T (ρΓ1 (x))T −1 = ρΓ2 (x) for all x ∈ G.
Isospectral manifolds can then be constructed via

Theorem 4 Let Γ1 and Γ2 be cocompact discrete subgroups of a Lie group G,

and let g be a left-invariant metric on G. If Γ1 and Γ2 are representation
equivalent, then
spec(Γ1 /G, g) = spec(Γ2 /G, g).
The first known examples of isospectral manifolds with different fundamental
groups were constructed via this method, as were the first examples of contin-
uous families of isospectral manifolds (see [9] for details).
The other general method of constructing isospectral manifolds involving
representation theoretic ideas was discovered by Sunada in 1985:
Theorem 5 Let Γ1 and Γ2 be representation equivalent subgroups of a finite
group G, which acts on a compact Riemannian manifold (M, g) on the left by
isometries. If the nonidentity elements of Γ1 and Γ2 act as fixed point free
isometries on M , then

spec(Γ1 /M, g) = spec(Γ2 /M, g).

(See [7] or [9] for a proof.) The construction of isospectral plane domains by
Gordon, Webb, and Wolpert [12] was achieved by a variant of this method;
this answered the question asked by Marc Kac: ”Can you hear the shape of a
drum?” in the negative.

4.3 Riemannian submersions

The last known general technique for constructing isospectral manifolds involves
Riemannian submersions. A submersion π : M → N of a Riemannian manifold
M into another Riemannian manifold N is said to be a Riemannian submersion
if the differential π∗ maps the horizontal space at any p ∈ M isometrically to
Tπ(p) N . The fibers (of the submersion) are said to be totally geodesic if any
geodesic in M which starts tangent to a fiber stays in the fiber. The main result
is
Theorem 6 Let π : M → N be a Riemannian submersion with totally geodesic
fibers. Then the Laplacians on M and N , ∆M and ∆N , satisfy π ∗ ∆N = ∆M π ∗ ,
so that in particular, spec(∆N ) = spec(∆M |π∗ C ∞ (N ) ).

6
Note that this can be combined with the previous methods to construct new
pairs of isospectral manifolds by starting from known pairs of isospectral man-
ifolds in some circumstances.
The two previously discussed methods cannot produce examples of isospec-
tral manifolds with different local geometry because they depend on looking
at manifolds that have common Riemannian covers. The interest in this final
method lies in the fact that until very recently, all known examples of isospectral
manifolds with different local geometry arise from a theorem that is a conse-
quence of the previous theorem:
Theorem 7 Let T be an n-dimensional torus for n > 1. Assume M1 and M2
are principal T -bundles and assume that the fibers, with the induced Riemannian
metrics, are totally geodesic flat tori. Further assume that the quotient manifolds
M1 /S and M2 /S, with the induced metric, are isospectral for every subtorus S
of T of codimension ≤ 1. Then M1 and M2 are isospectral.
Among the examples of isospectral manifolds constructed using this theorem are
continuous isospectral deformations of metrics on S m−1 × T 2 ; in some of these,
the maximum scalar curvature changes during the deformation [11]. Other
examples are isospectral deformations of metrics on S 3 × S 3 × S 5 — these
provide the first constructions of isospectral simply connected manifolds [19].

4.4 Recent constructions

A wealth of examples of isospectral manifolds are covered in the survey [9].
What we will attempt here is to briefly mention some of the more recent con-
structions that postdate that survey.
One of the more striking achievements in recent years has been the construc-
tion of isospectral metrics on spheres. By relaxing the conditions of the previous
theorem, Gordon was able to construct isospectral metrics on spheres S n≥8 and
balls B n≥9 [10]. This was achieved by reformulating the theorem so that it no
longer requires the torus T to act freely on M1 and M2 . A slight variation of
this was used by Schüth to produce isospectral metrics on S 5 and B 6 [21].
A different approach by Szabó also produces isospectral metrics on spheres
and balls [22]. Let v and z be Euclidean vector spaces. Then every linear map
j : z → so(v) defines a Lie bracket [, ] : v × v → z by h[X, Y ], Zi = hj(Z)X, Y i
for X, Y ∈ v and Z ∈ z. This makes v ⊕ z a two-step nilpotent Lie algebra
(extend [, ] to v⊕z by setting it to 0 if one of its arguments is in z). Denote by N j
the corresponding simply connected Lie group, and let g j be the left-invariant
metric that arises from the given inner product on its Lie algebra v ⊕ z. Two
linear maps j1 , j2 : z → so(v) are said to be isospectral if for each Z ∈ z,
j1 (Z) is conjugate to j2 (Z). Gordon and Wilson used isospectral pairs of such
maps to construct isospectral pairs of balls and tori [13]. To produce isospectral
spheres and balls, Szabó showed that under certain conditions, one can obtain
isospectral spheres and balls in the Lie groups (N j1 , g j1 ) and (N j2 , g j2 ) and in
their 1-dimensional solvable extensions. What is required is that there exist an
orthogonal endomorphism σ of v commuting with each j(Z) such that j2 (Z) =

7
σj1 (Z) for all Z, and that there exist a nondegenerate anticommutator A ∈ j(z)
(i.e., Aj(Z) = −j(Z)A for all j(Z) orthogonal to A in so(v)). Among the
more interesting examples produced by this method are pairs of isospectral
metrics on spheres S 4k−1 , k ≥ 3, one of which is homogeneous, the other locally
nonhomogeneous.
Along different lines, a variation of the theorem in section 4.3 allowed Schüth
to construct
(i) non-locally isometric isospectral 4-manifolds,
(ii) isospectral left-invariant metrics on compact Lie groups, and isospectral
simply connected irreducible manifolds,
(iii) non-locally isometric isospectral conformally equivalent manifoolds.
See [20] for details.

5 Conclusion
The study of the spectrum of the Laplacian, spectral geometry, remains a very
active field of research. It is impossible to cover all of its aspects in such a
short survey and so this paper attempts only to discuss a small selection of
topics. The references mentioned in the introduction are good places to look for
information on topics not covered here. While some fundamental questions have
been answered (e.g., isospectral does not imply isometric), and some spectacular
counterexamples constructed, there remain interesting unanswered questions.
We will end with one such question: is there a sense in which for most manifolds,
the spectrum does determine the geometry?

References
[1] I. Alexandrova. A survey of inverse spectral results. Available from
http://math.berkeley.edu/~alanw/240papers00.html, 2000.
[2] I. Chavel. Eigenvalues in Riemannian Geometry. Pure and applied math-
ematics. Academic Press, Orlando, 1994.
[3] S. Y. Cheng. Eigenvalue comparison theorems and its geometric applica-
tions. Mathematische Zeitschrift, 143(3):289–297, 1975.
[4] Y. Colin de Verdière. Spectre du Laplacian et longeur des géodesiques
periodiques II. Compositio Mathematica, 27:159–184, 1973.
[5] Y. Colin de Verdière. Spectre de variétés riemanniens et spectres des
graphes. Proceedings of the ICM, 1:522–530, 1986.
[6] Y. Colin de Verdière. Construction de laplaciens dont une partie finie du
spectre est donée. Annales Scientifiques de l’École Normale Supérieure,
20:599–615, 1987.

8
 [7] M. Craioveanu, M. Puta, and T. Rassias. Old and new aspects in spectral
geometry. Mathematics and applications. Kluwer Academic, Dordrecht;
London, 2001.
[8] P. Gilkey. Invariance theory, the heat equation, and the Atiyah-Singer index
theorem. Publish or Perish, Boston, 1984.
[9] C. Gordon. Survey of isospectral manifolds. In Handbook of Differential
Geometry, volume 1. Elsevier, Amsterdam; New York, 2000.
[10] C. Gordon. Isospectral deformations of metrics on spheres. Inventiones
Mathematicae, 145(2):317–331, 2001.

[11] C. Gordon, R. Gornet, D. Schüth, D. Webb, and E. Wilson. Isospectral de-

formations of closed Riemannian manifolds with different scalar curvature.
Annales de l’Institut Fourier (Grenoble), 48:593–607, 1998.
[12] C. Gordon, D. Webb, and S. Wolpert. Isospectral plane domains and sur-
faces via Riemannian orbifolds. Inventiones Mathematicae, 110:1–22, 1992.
[13] C. Gordon and E. Wilson. Continuous families of isospectral Riemannian
metrics which are not locally isometric. Journal of Differential Geometry,
47(3):504–529, 1997.
[14] A. Ikeda. On lens spaces which are isospectral but not isometric. Annales
Scientifiques de l’École Normale Supérieure, 13:303–315, 1980.
[15] P. Li and S. T. Yau. Estimate of eigenvalues of a compact Riemannian man-
ifold. In Proceedings of the Symposium in Pure Mathematics, volume 36,
pages 205–235, Providence, R.I., 1980. AMS.
[16] A. Lichnerowicz. Geométrie des Groupes des Transformations. Dunod,
Paris, 1958.
[17] S. Minakshisundaram and A. Pleijel. Some properties of the eigenfunctions
of the Laplace-operator on Riemannian manifolds. Canadian Journal of
Mathematics, 1:242–256, 1949.

[18] N. Nadirashvili. Multiple eigenvalues of the Laplace operator. Math USSR

Sbornik, 61(1):225–238, 1988.
[19] D. Schüth. Continuous families of isospectral metrics on simply connected
manifolds. Annals of Mathematics, 149(1):287–308, 1999.
[20] D. Schüth. Isospectral manifolds with different local geometries. Journal
für die reine und angewandte Mathematik, 534:41–94, 2001.
[21] D. Schüth. Isospectral metrics on five-dimensional spheres. Journal of
Differential Geometry, 58(1):87–111, 2001.

9
[22] Z. Szabó. Isospectral pairs of metrics on balls, spheres, and other manifolds
with different local geometries. Annals of Mathematics, 154(2):437–475,
2001.
[23] D. Yang. Lower bound estimates of the first eigenvalue for compact
manifolds with positive Ricci curvature. Pacific Journal of Mathematics,
190(2):383–398, 1999.
[24] J. Q. Zhang and H. C. Yang. On the estimate of the first eigenvalue of
a compact Riemannian manifold. Scientia Sinica Series A, 27:1265–1273,
1984.

10