Topics:

∙ 𝐿2 Betti numbers, Novikov-Shubin invariant, 𝐿2 torsion.

∙ Determinant and approximation conjecture.

1 Notations and Reviews

C[𝐺]: group algebra of 𝐺. 𝑙2 (𝐺): 𝐿2 summable functions on 𝐺, can be seen as a
𝐿2 completion of C[𝐺]. 𝒩 (𝐺): 𝐺-equivariant (with left 𝐺 action) bounded linear
2
maps on 𝑙 (𝐺). A finite dimensional Hilbert 𝒩 (𝐺)-module is a 𝐺-Hilbert space
𝑛 2
that 𝐺-equivariantly isometrically embedded into 𝒞 ⊗ 𝑙 (𝐺). Unless specified
otherwise, all spaces are assumed to be finitely dimensional Hilbert 𝐺 modules.

1.1 Spectral theory of bounded self adjoint operators

Let 𝐻 be a Hilbert space. A bounded operator 𝐴 on 𝐻 is called self adjoint if
𝐴* = 𝐴, positive if (𝐴𝑥, 𝑥) > 0 for all 𝑥 ∈ 𝐻 .
𝐸 is called a spectral measure if it sends Borel sets in R to bounded positive
self adjoint maps on 𝐻 , satisfying the following:

∙ 𝐸(∅) = 0, 𝐸(R) = 𝐼
∙ 𝐸(𝐴)2 = 𝐸(𝐴)
∙ 𝐸(𝐴 ∩ 𝐵) = 𝐸(𝐴)𝐸(𝐵)
∙ If 𝐴 ∩ 𝐵 = ∅, 𝐸(𝐴 ∪ 𝐵) = 𝐸(𝐴) + 𝐸(𝐵)
∙ ∀𝑥, 𝑦 ∈ 𝐻 , (𝐸(·)𝑥, 𝑦) is a C-measure.
Theorem 1
𝐴 is ∫︀bounded self adjoint on 𝐻 then there is some spectral measure 𝐸 such that
𝐴 = 𝜆𝐸(𝜆). If 𝑓 is an analytic function, 𝑓 (𝐴) = 𝑓 (𝜆)𝐸(𝜆)
∫︀

Remark 1. When 𝐻 is finite dimensional, 𝐸 = 𝛿𝜆𝑖 𝑣𝑖* 𝑣𝑖 .

∑︀

2 𝐿2-Betti numbers
2.1 Definitions
∙ 𝐺 is a group, 𝑓 is a 𝐺-equivariant self adjoint∑︀ bounded operator from
C𝑛 ⊗𝑙2 (𝐺). The 𝐿2 trace is defined as 𝑡𝑟𝐺 (𝑓 ) = 𝑖 (𝑓 (𝑒𝑖 ⊗1), 𝑒𝑖 ⊗1). The
𝐿2 -trace of a self map on Hilbert 𝐺 modules are defined as the composition
of projection and this self map.

∙ Let 𝑀 ⊂ C𝑛 ⊗ 𝑙2 (𝐺) be a Hilbert 𝐺 module, 𝑝𝑟𝑀 the orthogonal projec-

2
tion on 𝑀 . Then the 𝐿 -dimension is defined as dim𝐺 𝑀 = 𝑡𝑟𝐺 (𝑝𝑟𝑀 ).
2
(Exercise: prove that the 𝐿 -dimension does not depend on the choice of
the embedding.)

1
 ∙ 𝐿2 -chain complex of finitely dimensional Hilbert 𝐺 modules is a se-
quence · · · → 𝐶𝑘+1 → 𝐶𝑘 → 𝐶𝑘−1 → . . . such that the composition of
(2)
two successive boundary maps is 0. The homology are 𝐻𝑘 = 𝑘𝑒𝑟(𝐶𝑘 →
𝐶𝑘−1 )/𝑖𝑚(𝐶𝑘+1 → 𝐶𝑘 ).
∙ Let 𝑋 be a CW-complex with a free, cellular 𝐺 left action such that 𝐺∖𝑋
is finite. Then 𝐿2 dimension of the homology of the 𝐿2 completion of the
(2) 2 (2)
cellular chain complex (𝐶· ) are called the 𝐿 Betti numbers 𝑏· .

2.2 Examples
Example 2. 𝑋 = R2 tiled by unit cubes, 𝐺 = Z2 (with generators 𝑎, 𝑏), 𝐺∖𝑋 =
2 2
𝑇 . The 𝐿 chain complex is

0 → 𝑙2 → (𝑙2 )2 → 𝑙2 → 0

Such that 𝜕2 (𝑥) = ((𝑥𝑎−𝑥), (𝑥−𝑥𝑏)), 𝜕1 (𝑥, 𝑦) = 𝑥𝑏−𝑥+𝑦𝑎−𝑦 . By computation

(2) (2)
we see 𝐻* = 0, hence 𝑏* = 0.

Example 3. 𝑋 a double cover of the 𝜃 graph unwrapping over one of the two
(2)
loops, 𝐺 = Z/2, 𝐺∖𝑋 is the 𝜃-shaped graph. 𝐻1 is of dimension 3, (𝑝𝑟𝐻1 (𝑒), 𝑒)
(2) (2)
can be computed explicitly, and 𝑏1 = 3/2, 𝑏0 = 1/2.

Remark 4.

∙ When 𝑋 𝑝𝑟𝐻1 (𝑒) can be calculated via span-

is a finite simplicial graphs,
ning trees: Let 𝒯 be the set of all spanning trees on 𝑋 . For any 𝑇 ∈ 𝒯 ,
let 𝑝𝑎𝑡ℎ(𝑇, 𝑒− , 𝑒+ ) be the path on 𝑇 from 𝑒− to 𝑒+ .
1 ∑︁
𝑝𝑟𝐻1 (𝑒) = 𝑒 − 𝑝𝑎𝑡ℎ(𝑇, 𝑒− , 𝑒+ )
|𝒯 |
𝑇 ∈𝒯

∙ There is also a physical interpretation of 𝑝𝑟𝐻1 (𝑒) via electrical currents.

(2)
∙ In general, if 𝐺 is a finite group, 𝑏𝑘 (𝑋) = 𝑏𝑘 (𝑋)/|𝐺|.

Example 5. 𝑋 is the universal cover of 𝜃-shaped graph, 𝐺 = 𝐹2 the deck trans-

formation. 𝑝𝑟𝐻1 𝑒 can be explicitly calculated (hint: first show that the element
(2)
in 𝐶1 of a complete binary tree whose 𝜕 is at the root that minimizes the norm
(2) (2)
has norm 1) (𝑝𝑟𝐻1 𝑒, 𝑒) = 1/3, 𝑏1 = 1, 𝑏0 = 0.

Remark 6. There are alternative interpretations of the computation above through

electrical currents and random walks.

2
2.3 Elementary properties of 𝐿2 dimension, 𝐿2 homology
and 𝐿2 Betti numbers
Some elementary properties of 𝐿2 trace:

∙ 𝑓 ≤ 𝑔 =⇒ 𝑡𝑟𝐺 (𝑓 ) ≤ 𝑡𝑟𝐺 (𝑔)

∙ If 𝑓𝑖 is increasing and weakly converging to 𝑓 , 𝑡𝑟𝐺 𝑓 = sup{𝑡𝑟𝐺 (𝑓𝑖 )}.

∙ 𝑓 ≥ 0, 𝑡𝑟𝐺 (𝑓 ) = 0 ⇐⇒ 𝑓 = 0.
∙ 𝑡𝑟𝐺 (𝑓 + 𝜆𝑔) = 𝑡𝑟𝐺 (𝑓 ) + 𝜆𝑡𝑟𝐺 (𝑔)

∙ 𝑓, 𝑔 and ℎ are self adjoint maps compatible with an exact sequence of

Hilbert 𝐺 modules, then 𝑡𝑟𝐺 (𝑔) = 𝑡𝑟𝐺 (𝑓 ) + 𝑡𝑟𝐺 (ℎ).
∙ 𝑓 :𝑈 →𝑉, then 𝑡𝑟𝐺 (𝑓 * 𝑓 ) = 𝑡𝑟𝐺 (𝑓 𝑓 * )
∙ 𝑓 and 𝑔 are maps on Hilbert 𝐺 and 𝐻 modules, then 𝑡𝑟𝐺×𝐻 𝑓 ⊗ 𝑔 =
𝑡𝑟𝐺 𝑓 ⊗ 𝑡𝑟𝐻 𝑔 .

∙ 𝐻 is a finite index subgroup of 𝐺, then 𝑡𝑟𝐻 𝑓 = [𝐺 : 𝐻]𝑡𝑟𝐺 𝑓

Proof: use definition and functional analysis.
Some elementary properties of 𝐿2 -dimensions:

∙ dim𝐺 (𝑉 ) = 0 ⇐⇒ 𝑉 = 0.
2
∙ 0 → 𝑈 → 𝑉 → 𝑊 → 0 weakly exact (𝐿 homology vanishes), then
dim𝐺 (𝑉 ) = dim𝐺 (𝑈 ) + dim𝐺 (𝑊 ).

∙ 𝑉𝑖 increasing, dim𝐺 ∪𝑖 𝑉𝑖 = sup𝑖 dim𝐺 𝑉𝑖 .

∙ 𝑉𝑖 decreasing, dim𝐺 ∩𝑖 𝑉𝑖 = inf 𝑖 dim𝐺 𝑉𝑖 .
∙ 𝑈, 𝑉 are 𝐺 and 𝐻 modules respectively, then dim𝐺×𝐻 𝑈 ⊗𝑉 = dim𝐺 𝑈 dim𝐻 𝑉 .

∙ [𝐺 : 𝐻] < ∞, 𝑑𝑖𝑚𝐻 (𝑉 ) = [𝐺 : 𝐻]𝑑𝑖𝑚𝐺 𝑉 .

Theorem 2
Let 0 → 𝐶* → 𝐷* → 𝐸* → 0 be an exact sequence of chains of 𝐺-modules, then
there is a long exact sequence which is weakly exact.
Some elementary properties of 𝐿2 Betti numbers

∙ 𝑓 : 𝑋 → 𝑌 cellular 𝐺-equiv maps between free 𝐺 cell complexes, with

induced map on homology isomorphism for 𝑝 < 𝑑 and surjective for 𝑝 = 𝑑,
then so are the induced maps on 𝐿2 homology. (proof: chain homotopy,
then use long exact sequence)

𝑘 (2)
∑︀
∙ 𝑋 free 𝐺 cell complex with 𝐺∖𝑋 finite. Then 𝜒(𝐺∖𝑋) = 𝑘 (−1) 𝑏𝑘 (𝑋).

3
 (2) (2)
∙ 𝑋 is a cocompact 𝑑-dimensional manifold, then 𝑏𝑝 = 𝑏𝑑−𝑝 .

∙ Künneth formula for products, formula for wedges, connected sums for
manifolds of dimension at least 3, Morse inequalities all same as the usual
Betti numbers.

(2)
∙ If 𝑋 is connected, 𝑏0 = 1/|𝐺|.

∙ [𝐺 : 𝐻] < ∞, then 𝑋 seen as 𝐻 complex has 𝐿2 -Betti numbers [𝐺 : 𝐻] of

it seen as 𝐺 complex.

Theorem 3
𝑓 a cellular map of a finite connected complex, 𝑇𝑓 its mapping tori, 𝜋1 (𝑇𝑓 ) →
𝐺 → Z for some 𝐺, then the 𝐺-cover of 𝑇𝑓 , denoted as 𝑇𝑓 and seen as 𝐺-
complex has zero 𝐿2 Betti numbers.
Proof. Let 𝐺𝑛 be the preimage of 𝑛Z in 𝐺 → Z. Then 𝑇𝑓 has 𝑛-times as much
𝐿2 -Betti numbers, however 𝐺𝑛 ∖𝑇𝑓 = 𝑇𝑓 𝑛 has bounded number of cells, hence
all Betti number has to be 0.

3 Approximation for subgroups of finite index

Theorem 4
𝑋 is a cell complex with free cellular 𝐺 action as before, 𝐺∖𝑋 finite. 𝐺 ⊃
𝐺1 . . . normal subgroups such that ∩𝑖 𝐺𝑖 = 1, [𝐺 : 𝐺1 ] < ∞, then 𝑏𝑘 (𝑋) =
(2)

lim𝑖→∞ 𝑏𝑘 (𝐺𝑖 ∖𝑋), the latter as 𝐺/𝐺𝑖 complexes.

(2)

This can be easily reduced to the following “algebraic” statement:

Proposition 7
Suppose 𝑓 is a positive self adjoint map on C𝑛 ⊗ 𝑙2 (𝐺) induced by a (left) Z[𝐺]
module homomorphism, 𝑓𝑖 be the induced maps on Z[𝐺/𝐺𝑖 ], then 𝑑𝑖𝑚𝐺 ker(𝑓 ) =
lim𝑖→∞ dim𝐺/𝐺𝑖 ker(𝑓𝑖 ).

Proof. Step 1: Let 𝐾 be 𝑛2 of the largest sum of all coeff of an entry in the
matrix representing 𝑓 , then it is larger than the operator norm of both 𝑓 and
𝑓𝑚 .
Step 2: The map can be represented as a right multiplication of a Z[𝐺]-
matrix, hence 𝑡𝑟𝐺 (𝑓 ) = 𝑡𝑟𝐺/𝐺𝑖 (𝑓𝑖 ) for large enough 𝑖. Furthermore, for any
polynomial 𝑝, 𝑡𝑟𝐺 (𝑝(𝑓 )) = 𝑡𝑟𝐺/𝐺𝑖 (𝑝(𝑓𝑖 )) for large enough 𝑖.
Step 3: Let 𝐹 , 𝐹𝑖 be the spectral density function for 𝑓 and 𝑓𝑖 (𝐹 (𝜆) =
dim𝐺 (𝐸([0, 𝜆]), 𝐹 (𝜆) = dim𝐺/𝐺𝑖 (𝐸𝑖 ([0, 𝜆])). Let 𝐹 , 𝐹 be the lim sup and lim
+
inf of 𝐹𝑖 . We shall prove that 𝐹 ≤ 𝐹 ≤ 𝐹 . Let 𝑝𝑛 be polynomials above
𝜒([0, 𝜆]) and below 𝜒([0, 𝜆 + 1/𝑛] + 1/𝑛𝜒([0, 𝐾]) slightly above that. Then

𝐹 (𝜆) ≤ 𝑡𝑟𝐺 (𝑝𝑛 (𝑓 )) ≤ 𝐹 (𝜆 + 1/𝑛) + 1/𝑛

4
And as 𝑛→∞ the middle term converges to 𝐹 (𝜆) due to spectral decomposi-
tion.
Step 4: We now prove that 𝐹𝑖 are uniformly right-continuous at 0. This is
due to a fact in linear algebra:

Lemma 8. 𝑓 : self adjoint positive linear map on C𝑛 . 𝐾 a bound on opera-

tor norm of 𝑓 , 𝐶 a lower bound on the first non-zero term of characteristic
polynomial of 𝑓 . Then for 𝜆 < 1,

num. of roots in (0, 𝜆) − log(𝐶) log(𝐾)

≤ +
𝑛 𝑛(− log(𝜆)) − log(𝜆)

Proof. Count non-zero roots.

Because the matrix is integral 𝐶 can be chosen uniformly as 1, which finishes

the proof.

Example 9. 𝑋 is the universal cover of closed surfaces, 𝐺 the deck group.

Example 10. Let Γ be a finite graph, Γ ← Γ1 ← . . . regular covers, for every

edge 𝑒 ∈ Γ, let𝑑𝑖 (𝑒) be the ratio of spanning trees of Γ𝑖 that doesn’t contain a
specific lift of 𝑒. Then 𝑑𝑖 converges.

4 Other 𝐿2 invariants
4.1 Definition
𝐹 is the spectral density function.

∙ Novikov-Shubin invariants 𝛼(𝐹 ) = lim inf 𝜆→0+ 𝑙𝑜𝑔(𝐹 (𝜆)−𝐹

log 𝜆
(0))

∫︀
∙ Fuglede-Kadison determinant det = exp( log(𝜆)𝑑𝐹 ).

Remark 11.

∙ Determinant conjecture: For any group 𝐺, any Z[𝐺] matrix 𝑓, the F-K
determinant of 𝑓 *𝑓 is at least 1.

∙ Determinant conjecture implies approximation for any sequence of sub-

groups.