Professional Documents
Culture Documents
Can You Hear The Shape of A Manifold?: Andrejs Treibergs
Can You Hear The Shape of A Manifold?: Andrejs Treibergs
Andrejs Treibergs
University of Utah
The URL for these Beamer Slides: “Can You Hear the Shape of a
Manifold?”
http://www.math.utah.edu/treiberg/HearManifoldSlides.pdf
Differentiable Manifolds.
Laplacian and spectrum.
Rayleigh Quotient – Variational characterization – Basic properties.
Weyl’s Asymptotic Formula
Milnor’s Example
Gordon-Webb-Wolpert Example
5. Differentiable Manifolds.
Γ = a1 Z × a2 Z × · · · × an Z
2
Figure: Fund. Domain of T
is a rectangular lattice and the
identifications are x ∼ x + j for all x ∈ Rn
and all j ∈ Γ.
7. Intrinsic Geometry. The Riemannian nmetric.
γ : [α, β] → M is piecewise C 1 ,
d(P, Q) = inf L(γ) :
γ(α) = P, γ(β) = Q
Theorem
(S, d) is a metric space.
Gradient, divergence and Laplacian are defined so that the usual Green’s
formulas continue to hold on the manifold. If V (x) = (v 1 (x), . . . , v n (x))
is a C 1 vector field in local coordinates x = (x1 , . . . , xn ) on a Riemannian
manifold and u ∈ C 2 (M), then using the inverse matrix g ij = [gij ]−1 ,
n
X ∂
grad u = · · · , g ij u, · · ·
∂xj
j=1
n
1 X ∂ √ i
div V = √ gv
g ∂xj
i,j=1
n
1 X ∂ √ ij ∂
∆ u = div grad u = √ gg u
g ∂xj ∂xi
i,j=1
9. Gradient, Divergence, Laplacian.
Gradient, divergence and Laplacian are defined so that the usual Green’s
formulas continue to hold on the manifold. If V (x) = (v 1 (x), . . . , v n (x))
is a C 1 vector field in local coordinates x = (x1 , . . . , xn ) on a Riemannian
manifold and u ∈ C 2 (M), then using the inverse matrix g ij = [gij ]−1 ,
When M = Tn is the flat torus,
n
X ∂ ∂u
grad u = · · · , g ij u, · · · = · · · , ,··· ;
∂xj ∂ xi
j=1
n n
1 X ∂ √ i X ∂ vj
div V = √ gv = ;
g ∂xj ∂ xj
i,j=1 j=1
n n
∂2 u
X
1 X ∂ √ ij ∂
∆ u = div grad u = √ gg u = .
g ∂xj ∂xi ∂ xj 2
i,j=1 j=1
10. Wave Equation. Separation of Variables.
Suppose that a manifold vibrates according to the wave equation. What
frequencies are heard? Let ρ be the density and τ be the tension. Then
the amount of a small transverse vibration is given by v (x, t) where
x ∈ M and t ≥ 0,
∂2 v τ
2
= ∆ v.
∂t ρ
We seek solutions of the form v (x, t) = T (t)u(x). Thus
τ
T 00 (t)u(x) = T (t) ∆ u(x).
ρ
We can separate variables. The only way a t-expression equals an
x-expression is if both equal λ = const.
ρT 00 (t) ∆ u(x)
= −λ =
τ T (t) u(x)
which results in two equations
∆ u + λu = 0,
ρT 00 + λτ T = 0.
11. First equation: eigenvalues of the Laplacian on the manifold.
ρT 00 + λτ T = 0
cycles per unit time. The frequency increases with the eigenvalue λ and
tension τ and decreases with density ρ.
The lowest frequency corresponds to smallest positive eigenvalue λ1 > 0.
Thus λ1 is called the fundametal eigenvalue.
13. Basic Properties.
Theorem
Let M n be a smooth compact manifold.
1 Let λ be an eigenvalue and u its corresponding eigenfunction. Then
u ∈ C ∞ (M).
2 For all λ ∈ spec(M), the eigenspace Eλ = {u : ∆ u + λu = 0} is
finite dimensional. Its dimension is called the multiplicity mλ .
3 The zero eigenspace is one dimensional m0 = 1.
4 The set of eigenvalues is discrete and tends to infinity. The
eigenvalues can be ordered
spec(M) = {0 = λ0 < λ1 ≤ λ2 ≤ · · · → ∞}
by Green’s formula.
Since eigenfunction uj satisfy on (M, g )
∆ uj + λj uj = 0, (3)
1
eigenvalues scale like . So if we scale the lengths of curves by a
distance2
factor s on the manifold by multiplying the metric, s 2 g , then the
eigenvalue becomes
λj (M, g )
λj (M, s 2 g ) = .
s2
“Bigger manifolds make lower tones.”
15. Variational Characterization: Eigenfunctions minimize energy.
R
Since M u0 u1 = 0, the first eigenfunction is orthogonal to Rconstants. We
seek functions v , orthogonal to constants, that have fixed M v 2 = 1 and
minimize the energy M | grad v |2 . Equivalently, we minimimize the
R
Rayleigh Quotient.
Z
| grad v |2
MZ
λ1 = inf = inf R(v )
1 v
vR∈ H (M), v2
M v = 0,
M
v 6≡ 0
For the flat rectangular torus, this is the number of integer points in Zn
within an ellipsiod. (Counting positive and negative integers accounts for
the multiplicity two eigenspaces.)
2 j22 jn2
n j1 s
NTn (s) = ] j ∈ Z : 2 + 2 + · · · + 2 ≤ 2
a1 a2 an 4π
18. S1 .
so n o
4π 2 2
spec(S1a ) = a2
j :j ∈Z
0 = ∆θ u + d(d + n − 1)u.
20. Spherical harmonics.
λd = d(d + n − 1).
http : //www.toroidalsnark.net/som.html
23. Weyl’s Asymptotic Formula.
0 = λ0 < λ1 ≤ λ2 ≤ λ3 ≤ · · · → ∞.
where
∗ j1 j2 jn
Γ = , ,..., : j1 , . . . , jn ∈ Z
a1 a2 an
is the dual lattice.
NTn (s) is the number of Γ∗ points in the sphere
√
n s
U(s) = x ∈ R : |x| ≤
2π
25. Proof of Weyl’s formula for rectagular torii. -
Figure: Lattice points within U(s), polyhedron P and its surrounding spheres.
27. Proof of Weyl’s formula for rectagular torii.- - -
Taking volumes
√ √
|B1n | ( s − 2πR)n NTn (s) |B1n | ( s + 2πR)n
≤ ≤ .
(2π)n a1 · · · an (2π)n
which implies
1 2 1
2 +d + ≤ s + 41 ,
dm + 2 = dm m 4
we get q q
− 21 + s+ 1
4 ≤ dm + 1 ≤ 1
2 + s+ 1
4
so 2 2
q q
− 12 + s+ 1
4 ≤ NS2 (s) ≤ 1
2 + s + 14 .
It follows that
(4π) π s V(S2 ) |B12 | s
NS2 (s) ∼ = as s → ∞.
(2π)2 (2π)2
29. Example: General torus Tn = Rn /Γ.
Γ = {j1 e1 + · · · + jn en : j1 , . . . , jn ∈ Z}.
uγ (x) = e 2πix•γ
Thus
spec(Tn ) = {4π 2 |γ| : γ ∈ Γ∗ }.
31. Milnor’s counterexample.
Lemma
Suppose n ∈ 8N. If y ∈ Γ(n) then |y |2 ∈ 2Z.
|y |2 = h2 ni=1 14 = n4 h2 ∈ 2Z.
P
1 Pn
In case y = x + hω with x ∈ Γ2 then since x • ωn = 2 i=1 xi ∈ Z,
Lemma
Suppose n ∈ 4Z. Then Γ(n) = Γ∗ (n).
y • y 0 = x • x 0 + k 0 x • ωn + k x 0 • ωn + kk 0 |ωn |2 ∈ Z.
Lemma
T16 (Γ) is not isometric to T16 (Γ̃).
√
Γ(8) is generated by a set of vectors all of length 2:
{e1 − e8 , e2 − e8 , . . . , e7 − e8 , e1 + e2 , e3 + e4 , e5 + e6 , ω8 }
Lemma
Suppose both Γ and Γ0 are n = 8, 12, 16 or 20 dimensional lattices. If
both satisfy Γ = Γ∗ and |y |2 ∈ 2Z for all y ∈ Γ then the torii are
isospectral: spec(Tn (Γ)) = spec(Tn (Γ0 )).
2 2
For example, (e −π|x| )∼ = e −π|y | .
37. Milnor’s construction. - - - - -
Using the Poisson Summation Formula,
P 1 P ˜
x∈Λ f (x) = V(Γ) y ∈Λ∗ f (y )
√
applied to the lattice Λ = tΓ and using |y |2 even and Γ = Γ∗ ,
−π|x|2 t = 1 −π|y |2 /t = 1 −π|y |2 /t
P P P
x∈Γ e V(Γ) y ∈Γ∗ e t n/2 y ∈Γ e
which implies
ΘΓ (t) = t −n/2 ΘΓ (1/t). (4)
Also, since |x|2 ∈ 2Z,