Tensor products of Hilbert spaces

[18/78%]
Show
LaTeX

In Lemma 4.8, show that if L2(X,Ω,μ)L^{2}(X, \Omega , \mu ) is separable, then {ϕij}\left\{ \phi_{ij}\right\} is a basis for L2(X×X,Ω×Ω,μ×μ)L^{2}(X \times X, \Omega \times \Omega , \mu \times \mu ). What if L2(X,Ω,μ)L^{2}(X, \Omega , \mu ) is not separable?

Lemma 4.8: If {ei:i∈I}\left\{ e_{i}: i \in I\right\} is a basis for L2(X,Ω,μ)L^{2}(X, \Omega , \mu ) and

ϕij(x,y)=ej(x)ei(y)‾ \phi _{ij}(x,y) = e_{j}(x) \overline{e_{i}(y)}

for i,ji,j in II and x,yx,y in XX, then {ϕij:i,j∈I}\left\{ \phi_{ij}: i,j \in I\right\} is an orthonormal set in L2(X×X,Ω×Ω,μ×μ)L^{2}(X \times X, \Omega \times \Omega , \mu \times \mu ). If kk and KK are as in Proposition 4.7, then ⟨k,ϕij⟩=⟨Kej,ei⟩\left\langle k, \phi_{ij} \right\rangle = \left\langle Ke_{j}, e_{i} \right\rangle.

?

Let {xn}\left\{ x_{n}\right\} be a basis for a Hilbert space HH and {yn}\left\{ y_{n}\right\} be a basis for a Hilbert space KK. This exercise will show that the tensor product sequence {xm⊗yn}m,n∈N\left\{ x_{m} \otimes y_{n}\right\}_{m, n \in \mathbb {N}} is a basis for the tensor product space H⊗K=B2(H,K)H \otimes K = \mathcal{B}_{2}(H, K) (see Appendix B for definitions).

Let {an},{bn}\left\{ a_{n}\right\} ,\left\{ b_{n}\right\} and CX,CY\mathcal{C}_{X}, \mathcal{C}_{Y} denote the biorthogonal systems and basis constants for {xn},{yn}\left\{ x_{n}\right\} ,\left\{ y_{n}\right\}, respectively. Order N×N\mathbb {N} \times \mathbb {N} as follows:

\begin{aligned} \mathbb {N} \times \mathbb {N} = & \left\{ (1,1), \\ & (2,1),(2,2),(1,2), \\ & (3,1),(3,2),(3,3),(2,3),(1,3), \\ & (4,1),(4,2),(4,3),(4,4),(3,4),(2,4),(1,4), \\ & \quad \cdots \right\} , \end{aligned}

and let {zk}k∈N\left\{ z_{k}\right\}_{k \in \mathbb {N}} denote {xm⊗yn}m,n∈N\left\{ x_{m} \otimes y_{n}\right\}_{m, n \in \mathbb {N}} arranged according to the above ordering of N×N\mathbb {N} \times \mathbb {N}. Let SNX,SNYS_{N}^{X}, S_{N}^{Y}, and SNZS_{N}^{Z} denote the partial sum operators associated with {xm},{yn}\left\{ x_{m}\right\} ,\left\{ y_{n}\right\}, and {zk}\left\{ z_{k}\right\}, respectively.

?
(a)

Show that {xm⊗yn}m,n∈N\left\{ x_{m} \otimes y_{n}\right\}_{m, n \in \mathbb {N}} is exact in H⊗KH \otimes K, and its biorthogonal system is {am⊗bn}m,n∈N\left\{ a_{m} \otimes b_{n}\right\}_{m, n \in \mathbb {N}}.

(b)

Prove the following relationships among the partial sum operators (where we let S0XS_{0}^{X} and S0YS_{0}^{Y} be the zero operators):

SN2Z=SNX⊗SNY,SN2+ℓZ=SNX⊗SNY+(SN+1X−SNX)⊗SℓY,ℓ=1,…,N+1,SN2+N+1+ℓZ=SN+1X⊗SN+1Y−SℓX⊗(SN+1Y−SNY),ℓ=1,…,N. \begin{aligned} S_{N^{2}}^{Z} & = S_{N}^{X} \otimes S_{N}^{Y}, \\ S_{N^{2}+\ell }^{Z} & = S_{N}^{X} \otimes S_{N}^{Y}+\left(S_{N+1}^{X}-S_{N}^{X}\right) \otimes S_{\ell }^{Y}, \quad \ell = 1, \ldots , N+1, \\ S_{N^{2}+N+1+\ell }^{Z} & = S_{N+1}^{X} \otimes S_{N+1}^{Y}-S_{\ell }^{X} \otimes \left(S_{N+1}^{Y}-S_{N}^{Y}\right), \quad \ell = 1, \ldots , N. \end{aligned}
(c)

Show that ∥SNZ∥≤3CXCY\left\| S_{N}^{Z}\right\| \leq 3 \mathcal{C}_{X} \mathcal{C}_{Y} for each N∈NN \in \mathbb {N}, and conclude that {zk}\left\{ z_{k}\right\} is a basis for H⊗KH \otimes K.

Let {xn},{yn}\left\{ x_{n}\right\} ,\left\{ y_{n}\right\} be Bessel sequences in separable Hilbert spaces H,KH, K, respectively. Show that the tensor product sequence {xm⊗yn}m,n∈N\left\{ x_{m} \otimes y_{n}\right\}_{m, n \in \mathbb {N}} is a Bessel sequence in H⊗K=B2(H,K)H \otimes K = \mathcal{B}_{2}(H, K) (see Appendix B for definitions).

?

Let {xn},{yn}\left\{ x_{n}\right\} ,\left\{ y_{n}\right\} be Riesz bases for Hilbert spaces H,KH, K, respectively. Show that the tensor product sequence {xm⊗yn}m,n∈N\left\{ x_{m} \otimes y_{n}\right\}_{m, n \in \mathbb {N}} is a Riesz basis for H⊗K=B2(H,K)H \otimes K = \mathcal{B}_{2}(H, K) (see Appendix B for definitions).

?

This exercise gives another way to construct frames for "larger" spaces from "smaller" ones.

?
(a)

Let {xn},{yn}\left\{ x_{n}\right\} ,\left\{ y_{n}\right\} be frames for Hilbert spaces H,KH, K, respectively. Show that the tensor product sequence {xm⊗yn}m,n∈N\left\{ x_{m} \otimes y_{n}\right\}_{m, n \in \mathbb {N}} is a frame for H⊗K=B2(H,K)H \otimes K = \mathcal{B}_{2}(H, K) (see Appendix B for definitions).

(b)

Let E,FE, F be measurable subsets of R\mathbb {R}. Show that if {fn}\left\{ f_{n}\right\} is a frame for L2(E)L^{2}(E) and {gn}\left\{ g_{n}\right\} is a frame for L2(F)L^{2}(F), then {fm(x)‾gn(y)}m,n∈N\left\{ \overline{f_{m}(x)} g_{n}(y)\right\}_{m, n \in \mathbb {N}} is a frame for L2(E×F)L^{2}(E \times F).

Complete the proof of Theorem B.10.

Theorem B.10: Let H,KH, K be separable Hilbert spaces.

?
(a)

T∈B(H,K)T \in \mathcal{B}(H, K) has rank one if and only if T=x⊗yT = x \otimes y for some x∈H,y∈Kx \in H, y \in K not both zero.

(b)

T∈B(H,K)T \in \mathcal{B}(H, K) has finite rank if and only if T=∑k=1Nxk⊗ykT = \sum_{k = 1}^{N} x_{k} \otimes y_{k} for some N∈NN \in \mathbb {N}, xk∈Hx_{k} \in H, and yk∈Ky_{k} \in K.

(c)

{x⊗y:x∈H,y∈K}\left\{ x \otimes y: x \in H, y \in K\right\} is complete in B2(H,K)\mathcal{B}_{2}(H, K).

(d)

If x1,x2∈Hx_{1}, x_{2} \in H and y1,y2∈Ky_{1}, y_{2} \in K, then ⟨x1⊗y1,x2⊗y2⟩HS=⟨x2,x1⟩H⟨y1,y2⟩K\left\langle x_{1} \otimes y_{1}, x_{2} \otimes y_{2} \right\rangle_{\mathrm{HS}} = \left\langle x_{2}, x_{1} \right\rangle_{H} \left\langle y_{1}, y_{2} \right\rangle_{K}.

(e)

If {en}\left\{ e_{n}\right\} is an orthonormal basis for HH and {fn}\left\{ f_{n}\right\} is an orthonormal basis for KK, then {em⊗fn}m,n∈N\left\{ e_{m} \otimes f_{n}\right\}_{m, n \in \mathbb {N}} is an orthonormal basis for B2(H,K)\mathcal{B}_{2}(H, K).

(f)

If T∈B(H,K)T \in \mathcal{B}(H, K), then ⟨T,em⊗fn⟩=⟨Tem,fn⟩\left\langle T, e_{m} \otimes f_{n} \right\rangle = \left\langle T e_{m}, f_{n} \right\rangle for every m,n∈Nm, n \in \mathbb {N}.

Remark.
?

The book's own text already proves part (e) of Theorem B.10; this exercise asks for the remaining parts.

Let H,KH, K be separable Hilbert spaces. Prove the following facts about tensor products.

?
(a)

∥x⊗y∥HS=∥x∥H∥y∥K\left\| x \otimes y\right\|_{\mathrm{HS}} = \left\| x\right\|_{H}\left\| y\right\|_{K}.

(b)

(ax)⊗y=aˉ(x⊗y)(a x) \otimes y = \bar{a}(x \otimes y) and x⊗(by)=b(x⊗y)x \otimes (b y) = b(x \otimes y).

(c)

(x+w)⊗y=(x⊗y)+(w⊗y)(x+w) \otimes y = (x \otimes y)+(w \otimes y) and x⊗(y+z)=(x⊗y)+(x⊗z)x \otimes (y+z) = (x \otimes y)+(x \otimes z).

(d)

∥x⊗y−w⊗z∥HS≤∥x−w∥H∥y∥K+∥w∥H∥y−z∥K\left\| x \otimes y-w \otimes z\right\|_{\mathrm{HS}} \leq \left\| x-w\right\|_{H}\left\| y\right\|_{K}+\left\| w\right\|_{H}\left\| y-z\right\|_{K}.

?
(a)

Given xk∈Hx_{k} \in H and yk∈Ky_{k} \in K, show that the adjoint of T=∑k=1Mxk⊗ykT = \sum_{k = 1}^{M} x_{k} \otimes y_{k} is T∗=∑k=1Myk⊗xkT^{*} = \sum_{k = 1}^{M} y_{k} \otimes x_{k}

(b)

Given T∈B(H,K)T \in \mathcal{B}(H, K), show that TT is compact if and only if T∗T^{*} is compact.

Let E,FE, F be measurable subsets of R\mathbb {R}, and let {em}m∈N\left\{ e_{m}\right\}_{m \in \mathbb {N}} and {fn}n∈N\left\{ f_{n}\right\}_{n \in \mathbb {N}} be orthonormal bases for L2(E)L^{2}(E) and L2(F)L^{2}(F), respectively. Let em⊗ene_{m} \otimes e_{n} denote the tensor product function defined in equation (B.3), and show that the family {em⊗fn}m,n∈N\left\{ e_{m} \otimes f_{n}\right\}_{m, n \in \mathbb {N}} is an orthonormal basis for L2(E×F)L^{2}(E \times F).

?

Let H,KH, K be separable Hilbert spaces. Fix U∈B(H)U \in \mathcal{B}(H) and V∈B(K)V \in \mathcal{B}(K) and prove the following statements.

?
(a)

If x∈Hx \in H and y∈Ky \in K then V(x⊗y)U∗=Ux⊗VyV(x \otimes y) U^{*} = U x \otimes V y.

(b)

There exists a unique bounded operator T∈B(H⊗K)T \in \mathcal{B}(H \otimes K) such that ∥T∥≤∥U∥∥V∥\left\| T\right\| \leq \left\| U\right\| \left\| V\right\| and T(g⊗h)=Tg⊗ThT(g \otimes h) = T g \otimes T h for all g∈Hg \in H and h∈Kh \in K. We call TT the tensor product of UU and VV, and write T=U⊗VT = U \otimes V.

(c)

(U1⊗V1)(U2⊗V2)=U1U2⊗V1V2\left(U_{1} \otimes V_{1}\right)\left(U_{2} \otimes V_{2}\right) = U_{1} U_{2} \otimes V_{1} V_{2}.

(d)

If UU and VV are each topological isomorphisms, then so is T=U⊗VT = U \otimes V.

(e)

Specialize to H=L2(E)H = L^{2}(E) and K=L2(F)K = L^{2}(F). In particular, show that if U∈B(L2(E))U \in \mathcal{B}\left(L^{2}(E)\right) and V∈B(L2(R))V \in \mathcal{B}\left(L^{2}(R)\right) then there exists a unique operator U⊗V∈B(L2(E×F))U \otimes V \in \mathcal{B}\left(L^{2}(E \times F)\right) such that (U⊗V)(g⊗h)=Ug⊗Vh(U \otimes V)(g \otimes h) = U g \otimes V h, where g⊗hg \otimes h and Ug⊗VhU g \otimes V h are the functions defined by equation (B.3).

An operator AA is called a Hilbert–Schmidt operator if the series ∑β∈B∥Axβ∥2\sum_{\beta \in B}\left\| Ax_\beta \right\|^2 converges for some Hilbert basis {xβ}β∈B\left\{ x_\beta \right\}_{\beta \in B}.

?
(a)

Prove that if AA is a Hilbert--Schmidt operator, then the quantity ∥A∥2=(∑β∈B∥Axβ∥)1/2\left\| A\right\|_{2}=(\sum_{\beta \in B}\left\| Ax_\beta \right\| )^{1/2} does not depend on the choice of basis and defines in the space L2(H)\mathcal L_2(H) of Hilbert--Schmidt operators a norm which majorizes the usual operator norm.

(b)

Prove that the norm ∥⋅∥2\left\| \cdot \right\|_{2} is generated by the scalar product ⟨A,B⟩L2(H)=∑γ∈Γ(Ayγ,Byγ)H\langle A,B\rangle_{\mathcal L_2(H)}=\sum_{\gamma \in \Gamma }(Ay_\gamma ,By_\gamma )_H, where {yγ}γ∈Γ\left\{ y_\gamma \right\}_{\gamma \in \Gamma } is any Hilbert basis in HH.

(c)

Prove that every Hilbert--Schmidt operator is compact.

(d)

Construct an isomorphism between L2(H)\mathcal L_2(H) and the Hilbert tensor product of HH and H′H'.

(e)

Prove that every Hilbert--Schmidt operator on L2(X,μ)L^{2}\left({X,\mu }\right) is an integral operator with kernel K\inL2(X×X,μ×μ)K\inL^{2}\left({X\times X,\mu \times \mu }\right).

Suppose that H1H_1 and H2H_2 are Hilbert spaces, and H1⊗H2H_1\otimes H_2 their Hilbert tensor product. Prove that if A1A_1 and A2A_2 are selfadjoint operators in H1H_1 and H2H_2, respectively, then the operator A1⊗1+1⊗A2A_1\otimes 1+1\otimes A_2, with domain DA1⊗DA2D_{A_1}\otimes D_{A_2}, is essentially selfadjoint in HH.

?

Suppose that V1,…,VmV_{1}, \ldots , V_{m} are finite-dimensional inner product spaces. Prove that there is a unique inner product on V1⊗⋯⊗VmV_{1} \otimes \cdots \otimes V_{m} such that

⟨v1⊗⋯⊗vm,u1⊗⋯⊗um⟩=⟨v1,u1⟩⋯⟨vm,um⟩ \left\langle v_{1} \otimes \cdots \otimes v_{m}, u_{1} \otimes \cdots \otimes u_{m} \right\rangle = \left\langle v_{1}, u_{1} \right\rangle \cdots \left\langle v_{m}, u_{m} \right\rangle

for all (v1,…,vm)\left(v_{1}, \ldots , v_{m}\right) and (u1,…,um)\left(u_{1}, \ldots , u_{m}\right) in V1×⋯×VmV_{1} \times \cdots \times V_{m}. Note that the equation above implies that

∥v1⊗⋯⊗vm∥=∥v1∥×⋯×∥vm∥ \left\| v_{1} \otimes \cdots \otimes v_{m}\right\| = \left\| v_{1}\right\| \times \cdots \times \left\| v_{m}\right\|

for all (v1,…,vm)∈V1×⋯×Vm\left(v_{1}, \ldots , v_{m}\right) \in V_{1} \times \cdots \times V_{m}.

?

Suppose that V1,…,VmV_{1}, \ldots , V_{m} are finite-dimensional inner product spaces and V1⊗⋯⊗VmV_{1} \otimes \cdots \otimes V_{m} is made into an inner product space using the inner product from Exercise 12. Suppose e1k,…,enkke_{1}^{k}, \ldots , e_{n_{k}}^{k} is an orthonormal basis of VkV_{k} for each k=1,…,mk = 1, \ldots , m. Show that the list

{ej11⊗⋯⊗ejmm}j1=1,…,n1;⋯ ;jm=1,…,nm \left\{ e_{j_{1}}^{1} \otimes \cdots \otimes e_{j_{m}}^{m}\right\} _{j_{1} = 1, \ldots , n_{1} ; \cdots ; j_{m} = 1, \ldots , n_{m}}

is an orthonormal basis of V1⊗⋯⊗VmV_{1} \otimes \cdots \otimes V_{m}.

?
Remark.
?

Exercise 12: the inner product on V1⊗⋯⊗VmV_{1} \otimes \cdots \otimes V_{m} is the unique one such that ⟨v1⊗⋯⊗vm,u1⊗⋯⊗um⟩=⟨v1,u1⟩⋯⟨vm,um⟩\left\langle v_{1} \otimes \cdots \otimes v_{m}, u_{1} \otimes \cdots \otimes u_{m} \right\rangle = \left\langle v_{1}, u_{1} \right\rangle \cdots \left\langle v_{m}, u_{m} \right\rangle for all (v1,…,vm)\left(v_{1}, \ldots , v_{m}\right) and (u1,…,um)\left(u_{1}, \ldots , u_{m}\right) in V1×⋯×VmV_{1} \times \cdots \times V_{m}.

Let

x→1=12(11),x→2=12(1−1). \overrightarrow {x}_{1} = \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 1 \end{pmatrix}, \qquad \overrightarrow {x}_{2} = \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ -1 \end{pmatrix}.

Thus {x→1,x→2}\left\{ \overrightarrow {x}_{1}, \overrightarrow {x}_{2}\right\} forms an orthonormal basis in R2\mathbb {R}^{2}. Calculate

x→1⊗x→1,x→1⊗x→2,x→2⊗x→1,x→2⊗x→2 \overrightarrow {x}_{1} \otimes \overrightarrow {x}_{1}, \qquad \overrightarrow {x}_{1} \otimes \overrightarrow {x}_{2}, \qquad \overrightarrow {x}_{2} \otimes \overrightarrow {x}_{1}, \qquad \overrightarrow {x}_{2} \otimes \overrightarrow {x}_{2}

and interpret the result.

?

Given the orthonormal basis

x→1=(eiϕcos⁡(θ)sin⁡(θ)),x→2=(−sin⁡(θ)e−iϕcos⁡(θ)) \overrightarrow {x}_{1} = \begin{pmatrix} e^{i \phi } \cos (\theta ) \\ \sin (\theta ) \end{pmatrix}, \qquad \overrightarrow {x}_{2} = \begin{pmatrix} -\sin (\theta ) \\ e^{-i \phi } \cos (\theta ) \end{pmatrix}

in the vector space C2\mathbb {C}^{2}. Use this orthonormal basis to find an orthonormal basis in C4\mathbb {C}^{4}.

?
?
(i)

An orthonormal basis in R3\mathbb {R}^{3} is given by

v→1=12(101),v→2=(010),v→3=12(10−1). \overrightarrow {v}_{1} = \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 0 \\ 1 \end{pmatrix}, \qquad \overrightarrow {v}_{2} = \begin{pmatrix} 0 \\ 1 \\ 0 \end{pmatrix}, \qquad \overrightarrow {v}_{3} = \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 0 \\ -1 \end{pmatrix}.

Use this orthonormal basis and the Kronecker product to find an orthonormal basis in R9\mathbb {R}^{9}.

(ii)

A basis in the vector space M2(C)M_{2}(\mathbb {C}) of 2×22 \times 2 matrices over C\mathbb {C} is given by the Pauli spin matrices (including the 2×22 \times 2 identity matrix) σ0=I2\sigma_{0} = I_{2}, σ1\sigma_{1}, σ2\sigma_{2}, σ3\sigma_{3}. Use this basis and the Kronecker product to find a basis in the vector space M4(C4)M_{4}(\mathbb {C}^{4}) of all 4×44 \times 4 matrices over C\mathbb {C}.

Consider the two normalized vectors

12(1001),12(100−1) \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 0 \\ 0 \\ 1 \end{pmatrix}, \qquad \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 0 \\ 0 \\ -1 \end{pmatrix}

in the Hilbert space R4\mathbb {R}^{4}. Show that the vectors obtained by applying the matrix (I2⊗σ1)(I_{2} \otimes \sigma_{1}) together with the original ones form an orthonormal basis in R4\mathbb {R}^{4}.

?

References

[Con07] Conway J.B. A Course in Functional Analysis. 2nd ed., Springer, 2007.
[Hei11] Heil C. A Basis Theory Primer. Birkhäuser, 2011.
[KiGv88] Kirillov A.A., Gvishiani A.D. Теоремы и задачи функционального анализа. 2-е изд., Наука, 1988.
[Axl24] Axler S. Linear Algebra Done Right. 4th ed., Springer, 2024.
[StHa17] Steeb W., Hardy Y. Problems and Solutions in Introductory and Advanced Matrix Calculus. 2nd ed., World Scientific Publishing, 2017.