1. Hodge Decomposition Theorem (Finite Dimensional Case)
A cochain complex of vector spaces (or simply a cochain complex) over a field k is a sequence of k-vector spaces {Ci: i ∈ Z} together with a sequence of k-linear maps {di : Ci→ Ci+1: i ∈ Z} such that di◦ di−1= 0 for any i ∈ Z. A cochain complex over k is denoted by C• = (Ci, di)i∈Z. Since di◦ di−1= 0 for all i ∈ Z, Im di−1is a vector subspace of ker di. The i-th cohomology of a cochain complex C• is defined to be the quotient vector space
Hi(C•) = ker di/ Im di−1, i ∈ Z.
We denote H∗(C•) =L
i∈ZHi(C•) and d =L
i∈Zdi.
In this note, all the vector spaces are assumed to be over the real field k = R or over the complex field k = C.
Definition 1.1. A (Riemannian or Hermitian) metric hC on a cochain complex C is a sequence of functions {hi: Ci× Ci→ k} such that hi is an inner product on Ci for each i ∈ Z.
If hC = {hi} is a metric on C, the inner product hi is also denoted by h·, ·iCi for each i ∈ Z.
Let hC be a metric on C. For each i ∈ Z, we define a linear map (di+1)∗: Ci+1 → Ciby hdix, yiCi+1 = hx, (di+1)∗yiCi,
for any x ∈ Ci and for any y ∈ Ci+1. We define a linear operator ∆ion Ci by
∆i= di−1◦ (di)∗+ (di+1)∗◦ di. We call ∆i the Laplace operator on C with respect to hC.
Lemma 1.1. Let C be a cochain complex of vector spaces and hC be a metric on C. Then ker ∆i = ker di∩ ker(di)∗.
Proof. If x ∈ ker di ∩ ker(di)∗, then dix = (di)∗x = 0 which implies that ∆ix = 0. Therefore x ∈ ker ∆i. This proves that
ker ∆i ⊇ ker di∩ ker(di)∗. Suppose x ∈ ker ∆i. Then ∆ix = 0. Hence
h∆ix, xiCi= hdi−1◦ (di)∗x + (di+1)∗◦ dix, xiCi
= h(di)∗x, (di)∗xiCi+ hdix, dixiCi
= k(di)∗xk2Ci+ kdixk2Ci
= 0.
Since k(di)∗xk2C
i ≥ 0 and kdixk2C
i ≥ 0, we find that k(di)∗xk2C
i = kdixk2C
i = 0
which implies that (di)∗x = dix = 0. Therefore x ∈ ker di∩ ker(di)∗. This proves that ker ∆i ⊆ ker di∩ ker(di)∗.
We complete the proof of our assertion.
Theorem 1.1. (Hodge Decomposition Theorem) Let C be a cochain complex of finite dimensional vector spaces (i.e. each Ci is finite dimensional) and hC be a metric on C. For each i ∈ Z, we have the orthogonal decomposition of vector spaces
Ci= ker ∆i⊕ Im di−1⊕ Im(di+1)∗.
1
2
Proof. Let us prove the orthogonality of these subspaces of Ci.
Let x ∈ ker ∆i and y ∈ Im di−1. Since x ∈ ker ∆i, dix = (di)∗x = 0. Since y ∈ Im di−1, we write y = di−1z for z ∈ Ci−1. Therefore
hx, yiCi = hx, di−1ziCi= h(di)∗x, ziCi−1 = h0, ziCi−1 = 0.
We prove that ker ∆i ⊥ Im di−1. Assume that y0 ∈ Im(di+1)∗. We write y0 = (di+1)∗z0 for some z0∈ Ci+1. Then
hx, y0iCi= hx, (di+1)∗z0iCi = hdix, z0iCi+1= h0, z0iCi+1= 0.
We prove that ker ∆i⊥ Im(di+1)∗. Let us compute hy, y0iCi:
hy, y0iCi = hdi−1z, (di+1)∗z0iCi = hdi◦ di−1z, z0iCi+1 = h0, z0iCi+1 = 0 by di◦ di−1= 0. We prove that Im di−1⊥ Im(di+1)∗. We obtain that
ker ∆i⊕ Im di−1⊕ Im(di+1)∗⊆ Ci.
Let us prove the inverse inclusion. It follows from the definition that ∆i: Ci→ Ci is a self-adjoint operator. By spectral theorem, Ci can be decomposed into orthogonal direct sum of eigenspaces of
∆i, i.e.
Ci=M
λ
E∆i(λ) = ker ∆i⊕M
λ6=0
E∆i(λ), where E∆i(λ) = ker(∆i− λI). For each x ∈ Ci, we can write x = x0+P
λ6=0xλ (this is a finite sum because Ci is finite dimensional) where x0 ∈ ker ∆i and xλ ∈ E∆i(λ) with λ 6= 0. Then
∆ix =P
λλxλ∈L
λ6=0E∆i(λ). In other words, Im ∆i⊆L
λ6=0E∆i(λ). For each x = xλ∈ E∆i(λ),
∆ix = λx which implies that x = ∆i(x/λ) ∈ Im ∆i. We find that E∆i(λ) ⊆ Im ∆i for any λ 6= 0.
Since Im ∆i is a vector subspace of Ci,L
λ6=0E∆i(λ) ⊆ Im ∆i. We prove the equation Im ∆i=M
λ6=0
E∆i(λ).
We find that Ci= ker ∆i⊕ Im ∆i. For each x ∈ Ci, let us write x = x1+ x2where x1∈ ker ∆i and x2∈ Im ∆i. By definition, choose y ∈ Ci so that
x2= ∆iy = di−1((di)∗y) + (di+1)∗(diy) ∈ Im di−1⊕ Im(di+1)∗. This implies that x ∈ ker ∆i⊕ Im di−1⊕ Im(di+1)∗ for any x ∈ Ci. We find that
Ci⊆ ker ∆i⊕ Im di−1⊕ Im(di+1)∗.
We complete the proof of our assertion.
Corollary 1.1. Let C be a cochain complex of finite dimensional vector spaces and hCbe a metric on C. Then there is a linear isomorphism of vector spaces
Hi(C) ∼= ker ∆i for any i ∈ Z.
Proof. Since ker ∆i = ker di∩ ker(di)∗, ker ∆i ⊆ ker di. Since di◦ di−1 = 0, Im di−1 ⊆ ker di. We find that ker ∆i⊕ Im di−1⊆ ker di. Let us prove that ker di⊆ ker ∆i⊕ Im di−1. Let x ∈ ker di. By the Hodge decomposition, we can write x = x1+ x2+ x3 for x1∈ ker ∆i and for x2∈ Im di−1 and x3∈ Im(di+1)∗. Choose z ∈ Ci+1 so that x3= (di+1)∗z. Since x ∈ ker di, dix3= 0. Now,
kx3k2Ci = hx3, x3iCi= hx3, (di+1)∗ziCi = hdix3, ziCi+1= h0, ziCi+1 = 0.
This proves that x3 = 0. Therefore x ∈ ker ∆i ⊕ Im di−1 for any x ∈ ker di and hence ker di ⊆ ker ∆i⊕ Im di−1. We conclude that ker di = ker ∆i⊕ Im di−1. By linear algebra, any (orthogonal) direct sum decomposition U = V ⊕ W of vector spaces induces a linear isomorphism U/W ∼= V. We find that
Hi(C) = ker di/ Im di−1∼= ker ∆i.
3
The orthogonal decomposition Ci= ker ∆i⊕ Im ∆i can also be proved using the following basic facts in linear algebra.
Lemma 1.2. Let A : V → W be a linear map between inner product spaces. Then (1) Im A⊥= ker A∗, and
(2) (Im A∗)⊥= ker A.
Proof. Let z ∈ Im A⊥.Then hy, ziW = 0 for any y ∈ Im A. For any x ∈ V, hAx, ziW = hx, A∗ziV = 0.
This implies that A∗z = 0. In other words, z ∈ ker A∗. This proves that Im A⊥ ⊆ ker A∗. Suppose z ∈ ker A∗. Then A∗z = 0. For any y ∈ Im A, we can write y = Ax and hence
hy, ziW = hAx, ziW = hx, A∗ziV = hx, 0iV = 0.
We see that z ∈ Im A⊥. Therefore ker A∗⊆ Im A⊥. We conclude (1).
This lemma implies the following result.
Corollary 1.2. Let A : V → V be a self-adjoint linear operator on a finite dimensional inner product space. Then we have the following orthogonal decomposition:
V = ker A ⊕ Im A.
Proof. Since A is self-adjoint, A = A∗. Hence Im A⊥ = ker A which implies that ker A ∩ Im A = {0}.
Hence the sum of vector spaces ker A + Im A is a direct sum ker A ⊕ Im A. Since ker A ⊕ Im A is a vector subspace of V and by the rank-nullity lemma,
dim V = dim ker A + dim Im A = dim(ker A ⊕ Im A),
we see that V = ker A ⊕ Im A. (Here we use the fact that if W is a vector subspace of a finite dimensional vector space V so that dim V = dim W, then W = V.)