DIRECT SUM AND DIRECT INTEGRAL OF HILBERT SPACES
1. Direct Sum
Let H1 and H2 be two complex Hilbert spaces. On H1× H2, we define h(h1, h2), (k1, k2)i = hh1, k1iH1+ hh2, k2iH2.
Here h1, k1 ∈ H1 and h2, k2 ∈ H2. Then we can verify that H1 × H2 together with the sesqulinear form defined above is a Hilbert space. We denote this Hilbert space by H1⊕ H2. Similarly, given a finite sequence of Hilbert spaces H1, · · · , Hn, we can define their direct sum H1⊕ · · · ⊕ Hn.
Let {Hn: n ≥ 1} be a sequence of Hilbert spaces over C. Let H be the space of functions ϕ : N →S
nHn so that (1) ϕ(n) ∈ Hn for each n, (2) P∞
n=1kϕ(n)k2n< ∞.
The vector space structure of H is defined by
(aψ)(n) = aψ(n), (ψ + ϕ)(n) = ψ(n) + ϕ(n)
for a ∈ C and ψ, ϕ ∈ H. The space of functions H can be equipped with an inner product by
(1.1) hϕ, ψi =
∞
X
n=1
hϕ(n), ψ(n)iHn.
Theorem 1.1. H together with the inner product defined in (1.1) is a Hilbert space.
Proof. It is routine to check that (1.1) is an inner product on H.
Let {ψk} be Cauchy sequence in H. For each n ≥ 1,
kψk(n) − ψl(n)kHn ≤ kψk− ψlkH
implies that (ψk(n)) is a Cauchy sequence in Hn. Since Hn is complete, we can choose ψ(n) so that ψ(n) = limk→∞ψk(n). We obtain a function ψ : N →S
nHn. Since a Cauchy sequence in a normed space is bounded, we can choose M ≥ 0 so that kψkkH ≤ M for all k. In other words,P∞
n=1kψk(n)k2 ≤ M for all k. Taking lim supk→∞ of this inequality, we obtain P∞
n=1kψ(n)k2≤ M. This implies that ψ ∈ H.
For any > 0, choose N so that for any k, l ≥ N, kψk− ψlkH < /2. Again, taking lim supl→∞, we find kψ − ψkkH≤ /2 < for all k ≥ N. This implies that
k→∞lim ψk= ψ.
Hence the sequence (ψk) in H is convergent to ψ in H. We complete the proof of our
assertion.
We will denote H by L∞
n=1Hn and call H the direct sum of {Hn}. In this case, we identify Hn with the closed subspace of H consisting of functions ϕ ∈ H so that ϕ(m) = 0 for m 6= n. Then we find that Hn⊥ Hm for n 6= m.
1
2 DIRECT SUM AND DIRECT INTEGRAL OF HILBERT SPACES
If H is a Hilbert space and {Hn} is a sequence of closed subspaces so that Hn⊥ Hm and for each h ∈ H, there is a unique hn∈ Hnfor all n so that h =P∞
n=1hn. We can identify H withL∞
n=1Hn via h 7→ ϕh, where ϕh(n) = hn. In this case, we also denote H byL∞ n=1Hn. 2. Direct Integral
Let X be a σ-compact1locally compact Hausdorff space and µ is a complete Borel measure on X. Let {Hx : x ∈ X} be a family of Hilbert spaces parametrized by (points of) X. Let H be the space of functions f : X →S
x∈XHx so that (1) f (x) ∈ Hx for each x ∈ X
(2) the function x 7→ hf (x), g(x)iHx is µ-integrable,
(3) if hx ∈ Hx for all x ∈ X and x 7→ hhx, g(x)i is integrable for each g ∈ H, then there is f ∈ H so that f (x) = hx for almost all x ∈ X.
On H, we define
hf, gi = Z
X
hf (x), g(x)ixdµ(x).
Theorem 2.1. The space H together with the inner product defined above is a Hilbert space.
Proof. The proof is similar as above.
We denote H by
H = Z L
X
Hxdµ(x) and call H the direct integral of {Hx} over (X, µ).
1A space is σ-compact if it is a countable union of compact sets.