Professional Documents
Culture Documents
Bill Helton
UCSD
Department of Mathematics
Numerical Optimization
Unconstrained: Gradient descent, Newton - nds a local
optimum eciently.
Constrained: is harder
Old idea is interior point method.
Example subject to g(x) 0
min f (x1 , . . . , xg )
X
subject to
g(x) 0
Numerical Optimization
Solution use a barrier function.
Example b (x) := ln y
y = g(x).
Set
an unconstrained algorithm.
Rough History:
1. Solution 197? SOL paper said is VERY unreliable,
since f (x ) is badly conditioned if solution x is on
boundary, g(x ) = 0.
2. Kamarkar, Bell labs 1985ish - showed works on
many examples.
3. Nestrov-Nimerovski 1990ish introduced LMIs, and
numerical solution
Sums of Squares
Dimension Dependent
Given -
Find
AX + X AT + C T C X B
0
I
BT X
which is equivalent to the Riccati inequality.
ncDimlessPart
10
DIMENSIONLESS FORMULAS
11
Matrices Whole
AX + XAT + C T C
XB
B X
12
0
(1)
Matrices Whole
AX + XAT + C T C
XB
BT X
12
0
(1)
Lj Xj 0
j=0
Disaggregated Matrices
13
X =
X1
X2
X2
X3
3
Lj Xj 0
j =0
L0 :=
L2 :=
CT C
2a12
a22 + a11
b21
b22
L1 :=
a11 + a22
2a21
b21
b11
b22
b12
b11
b12
L3 :=
2a11
a21
0
b11
0
b12
0
a21
b11
b12
a12
a22
a12
a22
b21
b21
2b22
14
NC Positive Polynomial
NC Convex Polynomials
introNCPolyPart
15
1. POSITIVE NC POLYS
p0
Means What ?
p( x ) = x1 2 + (x1 2 )T + x2 T x2
Dene MATRIX POSITIVE polynomial
Plug in n n matrices Xj for xj in p
always get
p(X1 , X2 ) is a PosSD n n matrix.
16
p( x )
p(X)
x = {x1 , x2 }
2 + x )T
= (3x1
2
(3x2
1 + x2 )
+(x1 5 )T
(x1 5 )
PosSD
PosSD
p( x ) =
c
Lj ( x ) Lj ( x )
j=1
Matrix
positive
non-commutative
polynomials are sums of squares
Can compute SoS, thus CAN TEST Matrix Positivity.
Proof: Operator theory techniques
See 2001 H preprint.
S. McCullough 2000 preprint - when all variables xj
are unitary.
19
NC CONVEX
POLYNOMIALS
20
21
Proof:
1. NC Positive Polynomials
2. NC Second Derivatives
3. Put the two together
Convexity Algorithm
In[1] :=
<< NCAlgebra.m;
In[2] :=
<< NCConvexity.m;
In[3] :=
SetNonCommutative[X, Y];
In[4] :=
F = inv[X inv[Y]]
NCConvexityRegion[F,{X, Y}]
{{2inv[X inv[Y]], 2inv[Y]}}
22
SUMMARY
1.
2.
THM
3.
ALGORTHIM
CHECKS CONVEXITY OF NC
RATIONAL FUNCTION. TRY NCAlgebra
23
ALGORITHMS and
IMPLEMENTATIONS
1. Convexity Checker - Camino, Skelton, H
24
Public
Try NCAlgebra
25
END 2006
END 2006
26