Section 5.15

Angles between Subspaces

[11/100%]
Show
LaTeX
Problem 5.15.1

Determine the angles θmin⁡\theta_{\min } and θmax⁡\theta_{\max } between the following subspaces of R3\mathbb {R}^{3}.

?
(a)

M=xy-planeM = xy\text{-plane}, N=⟨(1,0,0),(0,1,1)⟩N = \left\langle (1,0,0),(0,1,1)\right\rangle.

(b)

M=xy-planeM = xy\text{-plane}, N=⟨(0,1,1)⟩N = \left\langle (0,1,1)\right\rangle.

Problem 5.15.2

Determine the principal angles between the following subspaces of R3\mathbb {R}^{3}.

?
(a)

M=xy-planeM = xy\text{-plane}, N=⟨(1,0,0),(0,1,1)⟩N = \left\langle (1,0,0),(0,1,1)\right\rangle.

(b)

M=xy-planeM = xy\text{-plane}, N=⟨(0,1,1)⟩N = \left\langle (0,1,1)\right\rangle.

Remark.
?

The principal angles between MM and NN are defined recursively: θ1=θmin⁡\theta_{1} = \theta_{\min }, with u1,v1u_{1}, v_{1} the unit vectors attaining the maximum in cos⁡θmin⁡=v1Tu1\cos \theta_{\min } = v_{1}^{T}u_{1}; then M2=u1⊥∩MM_{2} = u_{1}^{\perp } \cap M, N2=v1⊥∩NN_{2} = v_{1}^{\perp } \cap N, and θ2\theta_{2} is the minimal angle between M2M_{2} and N2N_{2}, and so on.

Problem 5.15.3

Let θmin⁡\theta_{\min } be the minimal angle between nonzero subspaces M,N⊆RnM, N \subseteq \mathbb {R}^{n}.

?
(a)

Explain why θmax⁡=0\theta_{\max }=0 if and only if M=NM=N.

(b)

Explain why θmin⁡=0\theta_{\min }=0 if and only if M∩N≠0M\cap N \neq 0.

(c)

Explain why θmin⁡=π/2\theta_{\min }=\pi /2 if and only if M⊥NM\perp N.

Remark.
?

Equation (5.15.16): the maximal angle satisfies sin⁡θmax⁡=gap⁡(M,N)=∥PM−PN∥2\sin \theta_{\max } = \operatorname {gap}(M,N) = \left\| P_{M}-P_{N}\right\|_{2}.

Problem 5.15.4

Let θmin⁡\theta_{\min } be the minimal angle between nonzero subspaces M,N⊂RnM, N \subset \mathbb {R}^{n}, and let θmin⁡⊥\theta_{\min }^{\perp } denote the minimal angle between M⊥M^{\perp } and N⊥N^{\perp }. Prove that if M⊕N=RnM\oplus N = \mathbb {R}^{n}, then θmin⁡=θmin⁡⊥\theta_{\min } = \theta_{\min }^{\perp }.

?
Remark.
?

Equation (5.15.4): if M,NM, N are complementary, sin⁡θmin⁡=1/∥(PM−PN)−1∥2\sin \theta_{\min } = 1/\left\| (P_{M} - P_{N})^{-1}\right\|_{2}.

Problem 5.15.5

For nonzero subspaces M,N⊂RnM, N \subset \mathbb {R}^{n}, let θ~min⁡\tilde{\theta }_{\min } denote the minimal angle between MM and N⊥N^{\perp }, and let θmax⁡\theta_{\max } be the maximal angle between MM and NN. Prove that if M⊕N⊥=RnM \oplus N^{\perp } = \mathbb {R}^{n}, then cos⁡θ~min⁡=sin⁡θmax⁡\cos \tilde{\theta }_{\min } = \sin \theta_{\max }.

?
Remark.
?
Problem 5.15.6

For subspaces M,N⊆RnM, N \subseteq \mathbb {R}^{n}, prove that PM−PNP_{M}-P_{N} is nonsingular if and only if MM and NN are complementary.

?
Remark.
?

Equation (5.15.7): with U = [U1 | U2], V = [V1 | V2] orthogonal matrices whose columns give orthonormal bases for M, M-perp, N-perp, N respectively (so that P_M = U1 U1^T, P_N = V2 V2^T), U^T (P_M - P_N) V = block-diag(U1^T V1, -U2^T V2).

Problem 5.15.7

For complementary spaces M,N⊂RnM, N \subset \mathbb {R}^{n}, let P=PMNP = P_{MN} be the oblique projector onto MM along NN, and let Q=PM⊥N⊥Q = P_{M^{\perp }N^{\perp }} be the oblique projector onto M⊥M^{\perp } along N⊥N^{\perp }.

?
(a)

Prove that (PM−PN)−1=P−Q(P_{M}-P_{N})^{-1} = P-Q.

(b)

If θmin⁡\theta_{\min } is the minimal angle between MM and NN, explain why sin⁡θmin⁡=1/∥P−Q∥2\sin \theta_{\min } = 1/\left\| P-Q\right\|_{2}.

(c)

Explain why ∥P−Q∥2=∥P∥2\left\| P-Q\right\|_{2} = \left\| P\right\|_{2}.

Problem 5.15.8

Prove that if f:V→Rf: V \to \mathbb {R} is a function defined on a space VV such that f(αx)=αf(x)f(\alpha x) = \alpha f(x) for scalars α≥0\alpha \geq 0, then max⁡∥x∥=1f(x)=max⁡∥x∥≤1f(x)\max_{\left\| x\right\| =1} f(x) = \max_{\left\| x\right\| \leq 1} f(x).

?
Problem 5.15.9

Let MM and NN be nonzero complementary subspaces of Rn\mathbb {R}^{n}.

?
(a)

Explain why PMN=[(I−PN)PM]†P_{MN} = \left[(I-P_{N})P_{M}\right]^{\dagger }, where PMP_{M} and PNP_{N} are the orthogonal projectors onto MM and NN, respectively, and PMNP_{MN} is the oblique projector onto MM along NN.

(b)

If θmin⁡\theta_{\min } is the minimal angle between MM and NN, explain why

sin⁡θmin⁡=∥[(I−PN)PM]†∥2−1=∥[PM(I−PN)]†∥2−1=∥[(I−PM)PN]†∥2−1=∥[PN(I−PM)]†∥2−1. \sin \theta _{\min } = \left\| \left[(I-P_{N})P_{M}\right]^{\dagger }\right\| _{2}^{-1} = \left\| \left[P_{M}(I-P_{N})\right]^{\dagger }\right\| _{2}^{-1} = \left\| \left[(I-P_{M})P_{N}\right]^{\dagger }\right\| _{2}^{-1} = \left\| \left[P_{N}(I-P_{M})\right]^{\dagger }\right\| _{2}^{-1}.
Problem 5.15.10

For complementary subspaces M,N⊂RnM, N \subset \mathbb {R}^{n}, let θmin⁡\theta_{\min } be the minimal angle between MM and NN, and let θˉmin⁡\bar{\theta }_{\min } denote the minimal angle between MM and N⊥N^{\perp }.

?
(a)

If PMNP_{MN} is the oblique projector onto MM along NN, prove that cos⁡θˉmin⁡=∥PMN†∥2\cos \bar{\theta }_{\min } = \left\| P_{MN}^{\dagger }\right\|_{2}.

(b)

Explain why sin⁡θmin⁡≤cos⁡θˉmin⁡\sin \theta_{\min } \leq \cos \bar{\theta }_{\min }.

Problem 5.15.11

Let U=[U1∣U2]U = \left[U_{1}\mid U_{2}\right] and V=[V1∣V2]V = \left[V_{1}\mid V_{2}\right] be the orthogonal matrices whose columns give orthonormal bases for M,M⊥,N⊥,NM, M^{\perp }, N^{\perp }, N respectively (as in the proof of (5.15.3)-(5.15.4)).

?
(a)

Prove that if U2TV2U_{2}^{T}V_{2} is nonsingular, then 1∥(U2TV2)−1∥22=1−∥U2TV1∥22\dfrac {1}{\left\| (U_{2}^{T}V_{2})^{-1}\right\|_{2}^{2}} = 1 - \left\| U_{2}^{T}V_{1}\right\|_{2}^{2}.

(b)

Prove that if U2TV1U_{2}^{T}V_{1} is nonsingular, then ∥U2TV2∥22=1−1∥(U2TV1)−1∥22\left\| U_{2}^{T}V_{2}\right\|_{2}^{2} = 1 - \dfrac {1}{\left\| (U_{2}^{T}V_{1})^{-1}\right\|_{2}^{2}}.