08.10.2016 Views

Foundations of Data Science

2dLYwbK

2dLYwbK

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Lemma 12.23 Let A be a symmetric matrix. Then ‖A‖ 2<br />

= max ∣ x T Ax ∣ .<br />

|x|=1<br />

Pro<strong>of</strong>: By definition, the 2-norm <strong>of</strong> A is ‖A‖ 2<br />

= max |Ax|. Thus,<br />

|x|=1<br />

‖A‖ 2<br />

= max<br />

|x|=1<br />

|Ax| = max<br />

|x|=1<br />

√<br />

xT A T Ax = √ λ 2 1 = λ 1 = max<br />

|x|=1<br />

∣ x T Ax ∣ The two norm <strong>of</strong> a matrix A is greater than or equal to the 2-norm <strong>of</strong> any <strong>of</strong> its<br />

columns. Let a u be a column <strong>of</strong> A.<br />

Lemma 12.24 |a u | ≤ ‖A‖ 2<br />

Pro<strong>of</strong>: Let e u be the unit vector with a 1 in position u and all other entries zero. Note<br />

λ = max |Ax|. Let x = e u where a u is row u. Then |a u | = |Ae u | ≤ max |Ax| = λ<br />

|x|=1 |x|=1<br />

12.7.7 Linear Algebra<br />

Lemma 12.25 Let A be an n × n symmetric matrix. Then det(A) = λ 1 λ 2 · · · λ n .<br />

Pro<strong>of</strong>: The det (A − λI) is a polynomial in λ <strong>of</strong> degree n. The coefficient <strong>of</strong> λ n will be ±1<br />

depending on whether n is odd or even. Let the roots <strong>of</strong> this polynomial be λ 1 , λ 2 , . . . , λ n .<br />

Then det(A − λI) = (−1) n<br />

n ∏<br />

i=1<br />

(λ − λ i ). Thus<br />

det(A) = det(A − λI)| λ=0<br />

= (−1) n<br />

n<br />

∏<br />

i=1<br />

(λ − λ i )<br />

= λ 1 λ 2 · · · λ n<br />

∣<br />

λ=0<br />

The trace <strong>of</strong> a matrix is defined to be the sum <strong>of</strong> its diagonal elements.<br />

tr (A) = a 11 + a 22 + · · · + a nn .<br />

That is,<br />

Lemma 12.26 tr(A) = λ 1 + λ 2 + · · · + λ n .<br />

Pro<strong>of</strong>: Consider the coefficient <strong>of</strong> λ n−1 in det(A − λI) = (−1) n<br />

A − λI =<br />

⎛<br />

⎜<br />

⎝<br />

a 11 − λ a 12 · · ·<br />

a 21 a 22 − λ · · ·<br />

.<br />

.<br />

.<br />

⎞<br />

⎟<br />

⎠ .<br />

n ∏<br />

i=1<br />

(λ − λ i ). Write<br />

Calculate det(A − λI) by expanding along the first row. Each term in the expansion<br />

involves a determinant <strong>of</strong> size n − 1 which is a polynomial in λ <strong>of</strong> deg n − 2 except for<br />

the principal minor which is <strong>of</strong> deg n − 1. Thus the term <strong>of</strong> deg n − 1 comes from<br />

(a 11 − λ) (a 22 − λ) · · · (a nn − λ)<br />

415

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

Saved successfully!

Ooh no, something went wrong!