This preview shows pages 1–5. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Simplices
The volume of a simplex of vertices Y = {y 0 , y 1 , . . . , y n } is deﬁned by
vol(Y ) = det(S (Y ))
n! where
S (Y ) = y 1 − y 0 · · · y n − y 0 . Note that vol(Y ) > 0 (since the vertices are aﬃnely independent).
A measure of geometry for simplices is the normalized volume:
von(Y ) = vol Audet and Vicente (SIOPT 2008) 1
Y
diam(Y ) Unconstrained optimization . 37/109 Simplicial DirectSearch Methods
The NelderMead method:
Considers a simplex at each iteration, trying to replace the worst
vertex by a new one.
For that it performs one of the following simplex operations:
reﬂexion, expansion, outside contraction, inside contraction.
−→ Costs 1 or 2 function evaluations (per iteration).
If they all the above fail the simplex is shrunk.
−→ Additional n function evaluations (per iteration). Audet and Vicente (SIOPT 2008) Unconstrained optimization 38/109 Nelder Mead simplex operations (reﬂections, expansions, outside
contractions, inside contractions)
y1 y2 y ic yc y oc yr ye y0
y c is the centroid of the face opposed to the worse vertex y 2 . Audet and Vicente (SIOPT 2008) Unconstrained optimization 39/109 Nelder Mead simplex operations (shrinks)
y1 y2 y0 Audet and Vicente (SIOPT 2008) Unconstrained optimization 40/109 McKinnon counterexample:
The NelderMead method:
Attempts to capture the curvature of the function.
Is globally convergent when n = 1.
Can fail for n > 1 (e.g. due to repeated inside contractions). Audet and Vicente (SIOPT 2008) Unconstrained optimization 41/109 ...
View Full
Document
 Spring '06
 Tapley
 Finance

Click to edit the document details