v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization v2009.01.01 - Convex Optimization

convexoptimization.com
from convexoptimization.com More from this publisher
10.03.2015 Views

∇ ∂g mn(X) ∂X kl 622 APPENDIX D. MATRIX CALCULUS x = a . A vector −υ based anywhere in domf × R pointing toward the unique bowl-bottom is specified: [ ] x − a υ ∝ ∈ R K × R (1689) f(x) + b Such a vector is since the gradient is υ = ⎡ ⎢ ⎣ ∇ x f(x) →∇ xf(x) 1 df(x) 2 ⎤ ⎥ ⎦ (1690) ∇ x f(x) = 2(x − a) (1691) and the directional derivative in the direction of the gradient is (1711) D.1.5 →∇ xf(x) df(x) = ∇ x f(x) T ∇ x f(x) = 4(x − a) T (x − a) = 4(f(x) + b) (1692) Second directional derivative By similar argument, it so happens: the second directional derivative is equally simple. Given g(X) : R K×L →R M×N on open domain, ⎡ ∂ 2 g mn(X) ⎤ ∂X kl ∂X 12 · · · = ∂∇g mn(X) ∂X kl = ⎡ ∇ 2 g mn (X) = ⎢ ⎣ ⎡ = ⎢ ⎣ ⎢ ⎣ ∇ ∂gmn(X) ∂X 11 ∇ ∂gmn(X) ∂X 21 . ∇ ∂gmn(X) ∂X K1 ∂∇g mn(X) ∂X 11 ∂∇g mn(X) ∂X 21 . ∂∇g mn(X) ∂X K1 ∂ 2 g mn(X) ∂X kl ∂X 11 ∂ 2 g mn(X) ∂X kl ∂X 21 ∂ 2 g mn(X) ∂X kl ∂X K1 ∂ 2 g mn(X) ∂X kl ∂X 1L ∂ 2 g mn(X) ∂ ∂X kl ∂X 22 · · · 2 g mn(X) ∂X kl ∂X 2L ∈ R K×L (1693) ⎥ . . . ⎦ ∂ 2 g mn(X) ∂ ∂X kl ∂X K2 · · · 2 g mn(X) ∂X kl ∂X KL ⎤ ∇ ∂gmn(X) ∂X 12 · · · ∇ ∂gmn(X) ∂X 1L ∇ ∂gmn(X) ∂X 22 · · · ∇ ∂gmn(X) ∂X 2L ∈ R K×L×K×L ⎥ . . ⎦ ∇ ∂gmn(X) ∂X K2 · · · ∇ ∂gmn(X) ∂X KL (1694) ∂∇g mn(X) ⎤ ∂X 12 · · · ∂∇g mn(X) ∂X 22 . · · · ∂∇g mn(X) ∂X K2 · · · ∂∇g mn(X) ∂X 1L ∂∇g mn(X) ∂X 2L . ∂∇g mn(X) ∂X KL ⎥ ⎦

D.1. DIRECTIONAL DERIVATIVE, TAYLOR SERIES 623 Rotating our perspective, we get several views of the second-order gradient: ⎡ ∇ 2 g(X) = ⎢ ⎣ ∇ 2 g 11 (X) ∇ 2 g 12 (X) · · · ∇ 2 g 1N (X) ∇ 2 g 21 (X) ∇ 2 g 22 (X) · · · ∇ 2 g 2N (X) . . ∇ 2 g M1 (X) ∇ 2 g M2 (X) · · · . ∇ 2 g MN (X) ⎤ ⎥ ⎦ ∈ RM×N×K×L×K×L (1695) ⎡ ∇ 2 g(X) T 1 = ⎢ ⎣ ∇ ∂g(X) ∂X 11 ∇ ∂g(X) ∂X 21 . ∇ ∂g(X) ∂X K1 ∇ ∂g(X) ∂X 12 · · · ∇ ∂g(X) ∂X 1L ∇ ∂g(X) ∂X 22 . · · · ∇ ∂g(X) ∂X 2L . ∇ ∂g(X) ∂X K2 · · · ∇ ∂g(X) ∂X KL ⎤ ⎥ ⎦ ∈ RK×L×M×N×K×L (1696) ⎡ ∇ 2 g(X) T 2 = ⎢ ⎣ ∂∇g(X) ∂X 11 ∂∇g(X) ∂X 21 . ∂∇g(X) ∂X K1 ∂∇g(X) ∂X 12 · · · ∂∇g(X) ∂X 22 . · · · ∂∇g(X) ∂X K2 · · · ∂∇g(X) ∂X 1L ∂∇g(X) ∂X 2L . ∂∇g(X) ∂X KL ⎤ ⎥ ⎦ ∈ RK×L×K×L×M×N (1697) Assuming the limits exist, we may state the partial derivative of the mn th entry of g with respect to the kl th and ij th entries of X ; ∂ 2 g mn(X) ∂X kl ∂X ij = g mn(X+∆t e lim k e T l +∆τ e i eT j )−gmn(X+∆t e k eT l )−(gmn(X+∆τ e i eT)−gmn(X)) j ∆τ,∆t→0 ∆τ ∆t (1698) Differentiating (1678) and then scaling by Y ij ∂ 2 g mn(X) ∂X kl ∂X ij Y kl Y ij = lim ∆t→0 ∂g mn(X+∆t Y kl e k e T l )−∂gmn(X) ∂X ij Y ∆t ij (1699) g mn(X+∆t Y kl e = lim k e T l +∆τ Y ij e i e T j )−gmn(X+∆t Y kl e k e T l )−(gmn(X+∆τ Y ij e i e T)−gmn(X)) j ∆τ,∆t→0 ∆τ ∆t

∇ ∂g mn(X)<br />

∂X kl<br />

622 APPENDIX D. MATRIX CALCULUS<br />

x = a . A vector −υ based anywhere in domf × R pointing toward the<br />

unique bowl-bottom is specified:<br />

[ ] x − a<br />

υ ∝<br />

∈ R K × R (1689)<br />

f(x) + b<br />

Such a vector is<br />

since the gradient is<br />

υ =<br />

⎡<br />

⎢<br />

⎣<br />

∇ x f(x)<br />

→∇ xf(x)<br />

1<br />

df(x)<br />

2<br />

⎤<br />

⎥<br />

⎦ (1690)<br />

∇ x f(x) = 2(x − a) (1691)<br />

and the directional derivative in the direction of the gradient is (1711)<br />

D.1.5<br />

→∇ xf(x)<br />

df(x) = ∇ x f(x) T ∇ x f(x) = 4(x − a) T (x − a) = 4(f(x) + b) (1692)<br />

<br />

Second directional derivative<br />

By similar argument, it so happens: the second directional derivative is<br />

equally simple. Given g(X) : R K×L →R M×N on open domain,<br />

⎡<br />

∂ 2 g mn(X)<br />

⎤<br />

∂X kl ∂X 12<br />

· · ·<br />

= ∂∇g mn(X)<br />

∂X kl<br />

=<br />

⎡<br />

∇ 2 g mn (X) =<br />

⎢<br />

⎣<br />

⎡<br />

=<br />

⎢<br />

⎣<br />

⎢<br />

⎣<br />

∇ ∂gmn(X)<br />

∂X 11<br />

∇ ∂gmn(X)<br />

∂X 21<br />

.<br />

∇ ∂gmn(X)<br />

∂X K1<br />

∂∇g mn(X)<br />

∂X 11<br />

∂∇g mn(X)<br />

∂X 21<br />

.<br />

∂∇g mn(X)<br />

∂X K1<br />

∂ 2 g mn(X)<br />

∂X kl ∂X 11<br />

∂ 2 g mn(X)<br />

∂X kl ∂X 21<br />

∂ 2 g mn(X)<br />

∂X kl ∂X K1<br />

∂ 2 g mn(X)<br />

∂X kl ∂X 1L<br />

∂ 2 g mn(X) ∂<br />

∂X kl ∂X 22<br />

· · ·<br />

2 g mn(X)<br />

∂X kl ∂X 2L<br />

∈ R K×L (1693)<br />

⎥<br />

. . . ⎦<br />

∂ 2 g mn(X) ∂<br />

∂X kl ∂X K2<br />

· · ·<br />

2 g mn(X)<br />

∂X kl ∂X KL<br />

⎤<br />

∇ ∂gmn(X)<br />

∂X 12<br />

· · · ∇ ∂gmn(X)<br />

∂X 1L<br />

∇ ∂gmn(X)<br />

∂X 22<br />

· · · ∇ ∂gmn(X)<br />

∂X 2L<br />

∈ R K×L×K×L<br />

⎥<br />

. . ⎦<br />

∇ ∂gmn(X)<br />

∂X K2<br />

· · · ∇ ∂gmn(X)<br />

∂X KL<br />

(1694)<br />

∂∇g mn(X)<br />

⎤<br />

∂X 12<br />

· · ·<br />

∂∇g mn(X)<br />

∂X 22<br />

.<br />

· · ·<br />

∂∇g mn(X)<br />

∂X K2<br />

· · ·<br />

∂∇g mn(X)<br />

∂X 1L<br />

∂∇g mn(X)<br />

∂X 2L<br />

.<br />

∂∇g mn(X)<br />

∂X KL<br />

⎥<br />

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!