21.06.2015 Views

Nonlinear Equations - UFRJ

Nonlinear Equations - UFRJ

Nonlinear Equations - UFRJ

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.

[SEC. 9.3: APPROXIMATE ZEROS 127<br />

Here is another useful estimate, that we state for homogeneous<br />

systems only:<br />

Lemma 9.5. Let X ∈ C n+1 and f, g ∈ H d . Assume that v =<br />

µ(f, X) < 1. Then, for all Y ⊥ ker Df(X),<br />

‖f−g‖<br />

‖f‖<br />

‖Y‖<br />

√<br />

1 − v<br />

2<br />

1 + v<br />

≤ ‖Dg(X) † Df(X)Y‖ ≤ ‖Y‖<br />

1 − v .<br />

The rightmost inequality holds unconditionally.<br />

Proof. By Lemma 8.9,<br />

∥ ∥ ∥ ( )∥<br />

Df(X)<br />

† ∥∥∥ g − f ∥∥∥<br />

‖Dg(X) − Df(X)‖ ≤ µ(f, X) L x ≤ v<br />

‖f‖<br />

In particular<br />

∥ Df(X) † Dg(X) ker Df(X) ⊥ − I ker Df(X)<br />

∥<br />

⊥ ≤ v.<br />

By Lemmas 9.2 and 7.8,<br />

∥ Dg(X) † Df(X)Y ∥ ∥ ∥∥Dg(X) ≤<br />

−1<br />

Df(X)Y<br />

ker Df(X) ⊥<br />

The lower bound follows from Lemma 9.3:<br />

∥ Dg(X) † Df(X)Y ∥ ∥ ≥<br />

‖Y ‖ √ 1 − v 2<br />

1 + v<br />

∥ ≤ ‖Y ‖<br />

1 − v<br />

9.3 Approximate zeros<br />

The projective distance is defined in C n+1 by<br />

‖X − λY‖<br />

d proj (X, Y) = inf<br />

.<br />

λ∈C × ‖X‖<br />

Since it is scaling invariant, is defines a metric in projective space<br />

that is related to the Riemannian distance by<br />

d proj (x, y) = sin(d Riem (x, y)) ≤ d Riem (x, y)

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

Saved successfully!

Ooh no, something went wrong!