103
Perturbation analysis of matrix optimization Chao Ding Institute of Applied Mathematics Academy of Mathematics and Systems Science, CAS ICCOPT2019, Berlin 2019.08.06

Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

  • Upload
    others

  • View
    8

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Perturbation analysis of matrix optimization

Chao Ding

Institute of Applied Mathematics

Academy of Mathematics and Systems Science, CAS

ICCOPT2019, Berlin

2019.08.06

Page 2: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Acknowledgements

Based on the joint work with Ying Cui at USC:

• Nonsmooth composite matrix optimizations: strong regularity,

constraint nondegeneracy and beyond, arXiv:1907.13253 (July,

2019).

Page 3: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc

1

Page 4: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc

1

Page 5: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc

1

Page 6: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc

1

Page 7: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc

1

Page 8: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc

1

Page 9: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc

1

Page 10: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Nonsmooth Composite Matrix Optimization Problem

CMatOP:

minimizex∈X

Φ(x) , f(x) + φ ◦ λ(g(x))

subject to h(x) = 0,

• X and Y: two given finite dimensional Euclidean spaces

• f : X→ R, g : X→ Sn and h : X→ Y: twice continuously

differentiable functions

• φ : Rn → (−∞,+∞]: a symmetric function, i.e., for any u ∈ Rn

and any n× n permutation matrix P ,

φ(Pu) = φ(u)

• λ: the vector of eigenvalues for any symmetric matrix

F We focus on the symmetric case just for simplicity;

F The obtained results can be extended to non-symmetric cases;

F This is a general model which includes many “non-polyhedral” OPs:

SDP, Eigenvalue optimization, etc 1

Page 11: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

More applications

• Fastest mixing Markov chain problem (fast load balancing of

paralleled systems)

• Fastest distributed linear averaging problem

• The reduced rank approximations of transition matrices

• The low rank approximations of doubly stochastic matrices

• Low-rank approximation of matrices with linear structures

• Unsupervised learning

• ......

2

Page 12: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Spectral functions

φ ◦ λ: spectral function (Friedland, 1981)

• φ : Rn → (−∞,+∞] is a symmetric convex piecewise linear

function

• a convex piecewise linear function: a polyhedral convex

function (Rockafellar, 1970)

3

Page 13: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Spectral functions

φ ◦ λ: spectral function (Friedland, 1981)

• φ : Rn → (−∞,+∞] is a symmetric convex piecewise linear

function

• a convex piecewise linear function: a polyhedral convex

function (Rockafellar, 1970)

3

Page 14: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Spectral functions

φ ◦ λ: spectral function (Friedland, 1981)

• φ : Rn → (−∞,+∞] is a symmetric convex piecewise linear

function

• a convex piecewise linear function: a polyhedral convex

function (Rockafellar, 1970)

3

Page 15: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear functions

Theorem (Rockafellar & Wets, 1998)

φ can be expressed in the form of

φ(x) = φ1(x) + φ2(x), x ∈ Rn,

with φ1 : Rn → R and φ2 : Rn → (−∞,+∞] are defined by

φ1(x) := max1≤i≤p

{〈ai,x〉 − ci

}and φ2(x) := δdomφ(x),

• a1, . . . ,ap ∈ Rn, c1, . . . , cp ∈ R with some positive integer p ≥ 1;

• domφ is a polyhedral set:

domφ :=

{x ∈ Rn | max

1≤i≤q{〈bi,x〉 − di} ≤ 0

}• b1, . . . ,bq ∈ Rn and d1, . . . , dq ∈ R for some positive integer q ≥ 1.

4

Page 16: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Examples

SDP:

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• ei ∈ Rn: the canonical basis of Rn

g(x) ∈ Sn− ⇐⇒ φ2 ◦ λ(g(x)) = δdomφ(g(x))

Eigenvalue optimizations:

sk(X) =

k∑i=1

λi(X) = max1≤i≤p

{〈ai, λ(X)〉

}

• ai ∈ Rn: the vector contains k ones and n− k zeros

5

Page 17: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Examples

SDP:

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• ei ∈ Rn: the canonical basis of Rn

g(x) ∈ Sn− ⇐⇒ φ2 ◦ λ(g(x)) = δdomφ(g(x))

Eigenvalue optimizations:

sk(X) =

k∑i=1

λi(X) = max1≤i≤p

{〈ai, λ(X)〉

}

• ai ∈ Rn: the vector contains k ones and n− k zeros

5

Page 18: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Examples

SDP:

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• ei ∈ Rn: the canonical basis of Rn

g(x) ∈ Sn− ⇐⇒ φ2 ◦ λ(g(x)) = δdomφ(g(x))

Eigenvalue optimizations:

sk(X) =

k∑i=1

λi(X) = max1≤i≤p

{〈ai, λ(X)〉

}

• ai ∈ Rn: the vector contains k ones and n− k zeros

5

Page 19: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Examples

SDP:

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• ei ∈ Rn: the canonical basis of Rn

g(x) ∈ Sn− ⇐⇒ φ2 ◦ λ(g(x)) = δdomφ(g(x))

Eigenvalue optimizations:

sk(X) =

k∑i=1

λi(X) = max1≤i≤p

{〈ai, λ(X)〉

}

• ai ∈ Rn: the vector contains k ones and n− k zeros

5

Page 20: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Examples

SDP:

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• ei ∈ Rn: the canonical basis of Rn

g(x) ∈ Sn− ⇐⇒ φ2 ◦ λ(g(x)) = δdomφ(g(x))

Eigenvalue optimizations:

sk(X) =

k∑i=1

λi(X) = max1≤i≤p

{〈ai, λ(X)〉

}

• ai ∈ Rn: the vector contains k ones and n− k zeros

5

Page 21: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Perturbation analysis of CMatOPs

Canonically perturbed CMatOPs with parameters (a,b, c) ∈ X×Y× Sn:

minimizex∈X

f(x)− 〈a,x〉+ φ ◦ λ(g(x) + c)

subject to h(x) + b = 0

The Karush-Kuhn-Tucker (KKT) optimality condition for perturbed

problem takes the following form:a = ∇f(x) + h′(x)∗y + g′(x)∗Y + g′(x)∗Z

b = −h(x)

c ∈ −g(x) + ∂θ∗1(Y )

c ∈ −g(x) + ∂θ∗2(Z)

with θ1 = φ1 ◦ λ and θ2 = φ2 ◦ λ are two spectral functions

Strong regularity:

When the solution mapping SKKT(a,b, c) is Lipschitz continuous?

6

Page 22: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Perturbation analysis of CMatOPs

Canonically perturbed CMatOPs with parameters (a,b, c) ∈ X×Y× Sn:

minimizex∈X

f(x)− 〈a,x〉+ φ ◦ λ(g(x) + c)

subject to h(x) + b = 0

The Karush-Kuhn-Tucker (KKT) optimality condition for perturbed

problem takes the following form:a = ∇f(x) + h′(x)∗y + g′(x)∗Y + g′(x)∗Z

b = −h(x)

c ∈ −g(x) + ∂θ∗1(Y )

c ∈ −g(x) + ∂θ∗2(Z)

with θ1 = φ1 ◦ λ and θ2 = φ2 ◦ λ are two spectral functions

Strong regularity:

When the solution mapping SKKT(a,b, c) is Lipschitz continuous?

6

Page 23: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Perturbation analysis of CMatOPs

Canonically perturbed CMatOPs with parameters (a,b, c) ∈ X×Y× Sn:

minimizex∈X

f(x)− 〈a,x〉+ φ ◦ λ(g(x) + c)

subject to h(x) + b = 0

The Karush-Kuhn-Tucker (KKT) optimality condition for perturbed

problem takes the following form:a = ∇f(x) + h′(x)∗y + g′(x)∗Y + g′(x)∗Z

b = −h(x)

c ∈ −g(x) + ∂θ∗1(Y )

c ∈ −g(x) + ∂θ∗2(Z)

with θ1 = φ1 ◦ λ and θ2 = φ2 ◦ λ are two spectral functions

Strong regularity:

When the solution mapping SKKT(a,b, c) is Lipschitz continuous?6

Page 24: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Why it matters?

• Perturbation theory

• Algorithm

7

Page 25: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Why it matters?

• Perturbation theory

• Algorithm

7

Page 26: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Why it matters?

• Perturbation theory

• Algorithm

7

Page 27: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis

but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”: the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 28: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”: the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 29: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”: the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 30: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”: the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 31: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”: the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 32: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”: the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 33: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”:

the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 34: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

How?

Variational analysis but in slightly different way:

Variational analysis of spectral functions

• Tangent sets

• Critical cones

• Second-order tangent sets

• The “σ-term”: the key difference between NLPs (polyhedral) and

CMatOPs (non-polyhedral)

8

Page 35: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral

9

Page 36: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

Metric projection operator ΠK:

A := ΠK(C) := argmin

{1

2‖Y − C‖2 | Y ∈ K

}

If K is a polyhedral closed convex set,

• ΠK is directional differentiable (Facchinei & Pang, 2003)1

ΠK(C +H)−ΠK(C) = ΠCK(C)(H) =: Π′K(C;H) ∀H

• CK(C) is the critical cone of K at C

1F. Facchinei and J. S. Pang. Finite-Dimensional Variational Inequalities and

Complementarity Problems: Volume I, Springer-Verlag, New York, 2003.

10

Page 37: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

Metric projection operator ΠK:

A := ΠK(C) := argmin

{1

2‖Y − C‖2 | Y ∈ K

}If K is a polyhedral closed convex set,

• ΠK is directional differentiable (Facchinei & Pang, 2003)1

ΠK(C +H)−ΠK(C) = ΠCK(C)(H) =: Π′K(C;H) ∀H

• CK(C) is the critical cone of K at C

1F. Facchinei and J. S. Pang. Finite-Dimensional Variational Inequalities and

Complementarity Problems: Volume I, Springer-Verlag, New York, 2003.

10

Page 38: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

Metric projection operator ΠK:

A := ΠK(C) := argmin

{1

2‖Y − C‖2 | Y ∈ K

}If K is a polyhedral closed convex set,

• ΠK is directional differentiable (Facchinei & Pang, 2003)1

ΠK(C +H)−ΠK(C) = ΠCK(C)(H) =: Π′K(C;H) ∀H

• CK(C) is the critical cone of K at C

1F. Facchinei and J. S. Pang. Finite-Dimensional Variational Inequalities and

Complementarity Problems: Volume I, Springer-Verlag, New York, 2003.

10

Page 39: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

Metric projection operator ΠK:

A := ΠK(C) := argmin

{1

2‖Y − C‖2 | Y ∈ K

}If K is a polyhedral closed convex set,

• ΠK is directional differentiable (Facchinei & Pang, 2003)1

ΠK(C +H)−ΠK(C) = ΠCK(C)(H) =: Π′K(C;H) ∀H

• CK(C) is the critical cone of K at C

1F. Facchinei and J. S. Pang. Finite-Dimensional Variational Inequalities and

Complementarity Problems: Volume I, Springer-Verlag, New York, 2003.

10

Page 40: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

If K is a non-polyhedral closed convex set

but C2-cone reducible,

• ΠK is directional differentiable and Π′K(C;H) is the unique optimal

solution to (Bonnans et al., 1998)2:

min{‖D −H‖2 − σ(B, T 2

K(A,D)) | D ∈ CK(C)}

• B := C −A and σ(B, T 2K(A,D)) is the “σ-term” of K

polyhedral:

min ‖D −H‖2

s.t. D ∈ CK(C)

non-polyhedral:

min ‖D −H‖2 − σ(B, T 2K(A,D))

s.t. D ∈ CK(C)

2J.F. Bonnans, R. Cominetti and A. Shapiro. Sensitivity analysis of optimization problems

under second order regular constraints. Mathematics of Operations Research 23 (1998) 806–831.

11

Page 41: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

If K is a non-polyhedral closed convex set but C2-cone reducible,

• ΠK is directional differentiable and Π′K(C;H) is the unique optimal

solution to (Bonnans et al., 1998)2:

min{‖D −H‖2 − σ(B, T 2

K(A,D)) | D ∈ CK(C)}

• B := C −A and σ(B, T 2K(A,D)) is the “σ-term” of K

polyhedral:

min ‖D −H‖2

s.t. D ∈ CK(C)

non-polyhedral:

min ‖D −H‖2 − σ(B, T 2K(A,D))

s.t. D ∈ CK(C)

2J.F. Bonnans, R. Cominetti and A. Shapiro. Sensitivity analysis of optimization problems

under second order regular constraints. Mathematics of Operations Research 23 (1998) 806–831.

11

Page 42: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

If K is a non-polyhedral closed convex set but C2-cone reducible,

• ΠK is directional differentiable and Π′K(C;H) is the unique optimal

solution to (Bonnans et al., 1998)2:

min{‖D −H‖2 − σ(B, T 2

K(A,D)) | D ∈ CK(C)}

• B := C −A and σ(B, T 2K(A,D)) is the “σ-term” of K

polyhedral:

min ‖D −H‖2

s.t. D ∈ CK(C)

non-polyhedral:

min ‖D −H‖2 − σ(B, T 2K(A,D))

s.t. D ∈ CK(C)

2J.F. Bonnans, R. Cominetti and A. Shapiro. Sensitivity analysis of optimization problems

under second order regular constraints. Mathematics of Operations Research 23 (1998) 806–831.

11

Page 43: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

If K is a non-polyhedral closed convex set but C2-cone reducible,

• ΠK is directional differentiable and Π′K(C;H) is the unique optimal

solution to (Bonnans et al., 1998)2:

min{‖D −H‖2 − σ(B, T 2

K(A,D)) | D ∈ CK(C)}

• B := C −A and σ(B, T 2K(A,D)) is the “σ-term” of K

polyhedral:

min ‖D −H‖2

s.t. D ∈ CK(C)

non-polyhedral:

min ‖D −H‖2 − σ(B, T 2K(A,D))

s.t. D ∈ CK(C)

2J.F. Bonnans, R. Cominetti and A. Shapiro. Sensitivity analysis of optimization problems

under second order regular constraints. Mathematics of Operations Research 23 (1998) 806–831.

11

Page 44: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: polyhedral =⇒ non-polyhedral (cont’d)

If K is a non-polyhedral closed convex set but C2-cone reducible,

• ΠK is directional differentiable and Π′K(C;H) is the unique optimal

solution to (Bonnans et al., 1998)2:

min{‖D −H‖2 − σ(B, T 2

K(A,D)) | D ∈ CK(C)}

• B := C −A and σ(B, T 2K(A,D)) is the “σ-term” of K

polyhedral:

min ‖D −H‖2

s.t. D ∈ CK(C)

non-polyhedral:

min ‖D −H‖2 − σ(B, T 2K(A,D))

s.t. D ∈ CK(C)

2J.F. Bonnans, R. Cominetti and A. Shapiro. Sensitivity analysis of optimization problems

under second order regular constraints. Mathematics of Operations Research 23 (1998) 806–831.

11

Page 45: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric

(Rockafellar & Wets, 1998): φ = φ1 + φ2 with φ2 = δdomφ

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}, domφ =

{x ∈ Rn | max

1≤i≤q

{〈bi,x〉 − di

}≤ 0}

Proposition

Let φ = φ1 + φ2 : Rn → (−∞,∞] be a given proper convex piecewise

linear function. φ is symmetric over Rn if and only if the functions

φ1 : Rn → R and φ2 : Rn → (−∞,∞] satisfy the following conditions:

for any x ∈ Rn,

φ1(x) = max1≤i≤p

{maxQ∈Pn

{〈Qai,x〉 − ci

}}and φ2(x) = δdomφ(x),

where domφ =

{x ∈ Rn | max

1≤i≤q

{maxQ∈Pn

{〈Qbi,x〉 − di

}}≤ 0

}.

12

Page 46: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric

(Rockafellar & Wets, 1998): φ = φ1 + φ2 with φ2 = δdomφ

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}, domφ =

{x ∈ Rn | max

1≤i≤q

{〈bi,x〉 − di

}≤ 0}

Proposition

Let φ = φ1 + φ2 : Rn → (−∞,∞] be a given proper convex piecewise

linear function. φ is symmetric over Rn if and only if the functions

φ1 : Rn → R and φ2 : Rn → (−∞,∞] satisfy the following conditions:

for any x ∈ Rn,

φ1(x) = max1≤i≤p

{maxQ∈Pn

{〈Qai,x〉 − ci

}}and φ2(x) = δdomφ(x),

where domφ =

{x ∈ Rn | max

1≤i≤q

{maxQ∈Pn

{〈Qbi,x〉 − di

}}≤ 0

}.

12

Page 47: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric

(Rockafellar & Wets, 1998): φ = φ1 + φ2 with φ2 = δdomφ

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}, domφ =

{x ∈ Rn | max

1≤i≤q

{〈bi,x〉 − di

}≤ 0}

Proposition

Let φ = φ1 + φ2 : Rn → (−∞,∞] be a given proper convex piecewise

linear function. φ is symmetric over Rn if and only if the functions

φ1 : Rn → R and φ2 : Rn → (−∞,∞] satisfy the following conditions:

for any x ∈ Rn,

φ1(x) = max1≤i≤p

{maxQ∈Pn

{〈Qai,x〉 − ci

}}and φ2(x) = δdomφ(x),

where domφ =

{x ∈ Rn | max

1≤i≤q

{maxQ∈Pn

{〈Qbi,x〉 − di

}}≤ 0

}.

12

Page 48: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric (cont’d)

• For i = 1, . . . , p, define

Di :={x ∈ domφ | 〈aj ,x〉 − cj ≤ 〈ai,x〉 − ci ∀ j = 1, . . . , p

},

then domφ =⋃

i=1,...,p

Di

• any x ∈ domφ, we have two active index sets:

ι1(x) := {1 ≤ i ≤ p | x ∈ Di}, ι2(x) := {1 ≤ i ≤ q | 〈bi,x〉−di = 0}.

Proposition

For any i ∈ ι1(x), j ∈ ι2(x) and Q ∈ Pnx (i.e., Qx = x), there exist

i′ ∈ ι1(x) and j′ ∈ ι2(x) such that ai′

= Qai and bj′

= Qbj ,

respectively.

13

Page 49: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric (cont’d)

• For i = 1, . . . , p, define

Di :={x ∈ domφ | 〈aj ,x〉 − cj ≤ 〈ai,x〉 − ci ∀ j = 1, . . . , p

},

then domφ =⋃

i=1,...,p

Di

• any x ∈ domφ, we have two active index sets:

ι1(x) := {1 ≤ i ≤ p | x ∈ Di}, ι2(x) := {1 ≤ i ≤ q | 〈bi,x〉−di = 0}.

Proposition

For any i ∈ ι1(x), j ∈ ι2(x) and Q ∈ Pnx (i.e., Qx = x), there exist

i′ ∈ ι1(x) and j′ ∈ ι2(x) such that ai′

= Qai and bj′

= Qbj ,

respectively.

13

Page 50: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric (cont’d)

Rockafellar & Wets, 1998, Mordukhovich & Sarabi, 2016:

• the subgradients:

∂φ1(x) = conv{ai, i ∈ ι1(x)}, ∂φ2(x) = Ndomφ(x) = cone{bi, i ∈ ι2(x)}

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}is finite everywhere,

• φ1 is directionally differentiable

• the directional derivate:

φ′1(x;h) = maxi∈ι1(x)

〈ai,h〉, h ∈ Rn.

Let ψ(x) := max1≤i≤q

{〈bi,x〉 − di

}. Then, domφ =

{x ∈ Rn | ψ(x) ≤ 0

}• ψ is directionally differentiable

• the directional derivate:

ψ′(x;h) = maxi∈ι2(x)

〈bi,h〉, h ∈ Rn.

14

Page 51: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric (cont’d)

Rockafellar & Wets, 1998, Mordukhovich & Sarabi, 2016:

• the subgradients:

∂φ1(x) = conv{ai, i ∈ ι1(x)}, ∂φ2(x) = Ndomφ(x) = cone{bi, i ∈ ι2(x)}

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}is finite everywhere,

• φ1 is directionally differentiable

• the directional derivate:

φ′1(x;h) = maxi∈ι1(x)

〈ai,h〉, h ∈ Rn.

Let ψ(x) := max1≤i≤q

{〈bi,x〉 − di

}. Then, domφ =

{x ∈ Rn | ψ(x) ≤ 0

}• ψ is directionally differentiable

• the directional derivate:

ψ′(x;h) = maxi∈ι2(x)

〈bi,h〉, h ∈ Rn.

14

Page 52: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric (cont’d)

Rockafellar & Wets, 1998, Mordukhovich & Sarabi, 2016:

• the subgradients:

∂φ1(x) = conv{ai, i ∈ ι1(x)}, ∂φ2(x) = Ndomφ(x) = cone{bi, i ∈ ι2(x)}

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}is finite everywhere,

• φ1 is directionally differentiable

• the directional derivate:

φ′1(x;h) = maxi∈ι1(x)

〈ai,h〉, h ∈ Rn.

Let ψ(x) := max1≤i≤q

{〈bi,x〉 − di

}. Then, domφ =

{x ∈ Rn | ψ(x) ≤ 0

}• ψ is directionally differentiable

• the directional derivate:

ψ′(x;h) = maxi∈ι2(x)

〈bi,h〉, h ∈ Rn.

14

Page 53: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric (cont’d)

Rockafellar & Wets, 1998, Mordukhovich & Sarabi, 2016:

• the subgradients:

∂φ1(x) = conv{ai, i ∈ ι1(x)}, ∂φ2(x) = Ndomφ(x) = cone{bi, i ∈ ι2(x)}

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}is finite everywhere,

• φ1 is directionally differentiable

• the directional derivate:

φ′1(x;h) = maxi∈ι1(x)

〈ai,h〉, h ∈ Rn.

Let ψ(x) := max1≤i≤q

{〈bi,x〉 − di

}. Then, domφ =

{x ∈ Rn | ψ(x) ≤ 0

}

• ψ is directionally differentiable

• the directional derivate:

ψ′(x;h) = maxi∈ι2(x)

〈bi,h〉, h ∈ Rn.

14

Page 54: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Convex piecewise linear + Symmetric (cont’d)

Rockafellar & Wets, 1998, Mordukhovich & Sarabi, 2016:

• the subgradients:

∂φ1(x) = conv{ai, i ∈ ι1(x)}, ∂φ2(x) = Ndomφ(x) = cone{bi, i ∈ ι2(x)}

φ1(x) = max1≤i≤p

{〈ai,x〉 − ci

}is finite everywhere,

• φ1 is directionally differentiable

• the directional derivate:

φ′1(x;h) = maxi∈ι1(x)

〈ai,h〉, h ∈ Rn.

Let ψ(x) := max1≤i≤q

{〈bi,x〉 − di

}. Then, domφ =

{x ∈ Rn | ψ(x) ≤ 0

}• ψ is directionally differentiable

• the directional derivate:

ψ′(x;h) = maxi∈ι2(x)

〈bi,h〉, h ∈ Rn.

14

Page 55: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets

For θ1 = φ1 ◦ λ:

• Tangent set of epigraph:

Tepi θ1(X, θ(X)) = epi θ′1(X; ·) :={

(H, y) ∈ Sn × R | θ′1(X;H) ≤ y}

• The lineality space:

T linθ1 (X) :=

{H ∈ Sn | θ′1(X;H) = −θ′1(X;−H)

}Proposition

H ∈ T linθ1

(X) if and only if 〈z, λ′(X;H)〉 is a constant for any

z ∈ ∂φ1(λ(X)), i.e.,

〈λ′(X;H),ai − aj〉 = 0 ∀ i, j ∈ ι1(λ(X)).

15

Page 56: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets

For θ1 = φ1 ◦ λ:

• Tangent set of epigraph:

Tepi θ1(X, θ(X)) = epi θ′1(X; ·) :={

(H, y) ∈ Sn × R | θ′1(X;H) ≤ y}

• The lineality space:

T linθ1 (X) :=

{H ∈ Sn | θ′1(X;H) = −θ′1(X;−H)

}Proposition

H ∈ T linθ1

(X) if and only if 〈z, λ′(X;H)〉 is a constant for any

z ∈ ∂φ1(λ(X)), i.e.,

〈λ′(X;H),ai − aj〉 = 0 ∀ i, j ∈ ι1(λ(X)).

15

Page 57: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets

For θ1 = φ1 ◦ λ:

• Tangent set of epigraph:

Tepi θ1(X, θ(X)) = epi θ′1(X; ·) :={

(H, y) ∈ Sn × R | θ′1(X;H) ≤ y}

• The lineality space:

T linθ1 (X) :=

{H ∈ Sn | θ′1(X;H) = −θ′1(X;−H)

}Proposition

H ∈ T linθ1

(X) if and only if 〈z, λ′(X;H)〉 is a constant for any

z ∈ ∂φ1(λ(X)), i.e.,

〈λ′(X;H),ai − aj〉 = 0 ∀ i, j ∈ ι1(λ(X)).

15

Page 58: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets

For θ1 = φ1 ◦ λ:

• Tangent set of epigraph:

Tepi θ1(X, θ(X)) = epi θ′1(X; ·) :={

(H, y) ∈ Sn × R | θ′1(X;H) ≤ y}

• The lineality space:

T linθ1 (X) :=

{H ∈ Sn | θ′1(X;H) = −θ′1(X;−H)

}

Proposition

H ∈ T linθ1

(X) if and only if 〈z, λ′(X;H)〉 is a constant for any

z ∈ ∂φ1(λ(X)), i.e.,

〈λ′(X;H),ai − aj〉 = 0 ∀ i, j ∈ ι1(λ(X)).

15

Page 59: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets

For θ1 = φ1 ◦ λ:

• Tangent set of epigraph:

Tepi θ1(X, θ(X)) = epi θ′1(X; ·) :={

(H, y) ∈ Sn × R | θ′1(X;H) ≤ y}

• The lineality space:

T linθ1 (X) :=

{H ∈ Sn | θ′1(X;H) = −θ′1(X;−H)

}Proposition

H ∈ T linθ1

(X) if and only if 〈z, λ′(X;H)〉 is a constant for any

z ∈ ∂φ1(λ(X)), i.e.,

〈λ′(X;H),ai − aj〉 = 0 ∀ i, j ∈ ι1(λ(X)).

15

Page 60: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets (cont’d)

For θ2 = φ2 ◦ λ:

• θ2 = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ• Tangent set of K:

TK(X) ={H ∈ Sn | ζ ′(X;H) ≤ 0

}=

{H ∈ Sn | 〈bi, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}• The lineality space:

lin(TK(X)) ={H ∈ Sn | ζ ′(X;H) = −ζ ′(X;−H) = 0

}Proposition

H ∈ lin(TK(X)) if and only if 〈bi, λ′(X;H)〉 = 0 for any i ∈ ι2(λ(X)).

16

Page 61: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets (cont’d)

For θ2 = φ2 ◦ λ:

• θ2 = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ

• Tangent set of K:

TK(X) ={H ∈ Sn | ζ ′(X;H) ≤ 0

}=

{H ∈ Sn | 〈bi, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}• The lineality space:

lin(TK(X)) ={H ∈ Sn | ζ ′(X;H) = −ζ ′(X;−H) = 0

}Proposition

H ∈ lin(TK(X)) if and only if 〈bi, λ′(X;H)〉 = 0 for any i ∈ ι2(λ(X)).

16

Page 62: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets (cont’d)

For θ2 = φ2 ◦ λ:

• θ2 = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ• Tangent set of K:

TK(X) ={H ∈ Sn | ζ ′(X;H) ≤ 0

}=

{H ∈ Sn | 〈bi, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}

• The lineality space:

lin(TK(X)) ={H ∈ Sn | ζ ′(X;H) = −ζ ′(X;−H) = 0

}Proposition

H ∈ lin(TK(X)) if and only if 〈bi, λ′(X;H)〉 = 0 for any i ∈ ι2(λ(X)).

16

Page 63: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets (cont’d)

For θ2 = φ2 ◦ λ:

• θ2 = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ• Tangent set of K:

TK(X) ={H ∈ Sn | ζ ′(X;H) ≤ 0

}=

{H ∈ Sn | 〈bi, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}• The lineality space:

lin(TK(X)) ={H ∈ Sn | ζ ′(X;H) = −ζ ′(X;−H) = 0

}

Proposition

H ∈ lin(TK(X)) if and only if 〈bi, λ′(X;H)〉 = 0 for any i ∈ ι2(λ(X)).

16

Page 64: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets (cont’d)

For θ2 = φ2 ◦ λ:

• θ2 = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ• Tangent set of K:

TK(X) ={H ∈ Sn | ζ ′(X;H) ≤ 0

}=

{H ∈ Sn | 〈bi, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}• The lineality space:

lin(TK(X)) ={H ∈ Sn | ζ ′(X;H) = −ζ ′(X;−H) = 0

}Proposition

H ∈ lin(TK(X)) if and only if 〈bi, λ′(X;H)〉 = 0 for any i ∈ ι2(λ(X)).

16

Page 65: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets: SDP

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

X = V

0α · · · 0... 0β

...

0 · · · Λγ(X)

V T , ι2(λ(X)) = α ∪ β = γ

TSn−(X) ={H ∈ Sn | 〈ei, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ � 0

}

lin(TSn−(X)) ={H ∈ Sn | 〈ei, λ′(X;H)〉 = 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ = 0

}

17

Page 66: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets: SDP

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

X = V

0α · · · 0... 0β

...

0 · · · Λγ(X)

V T , ι2(λ(X)) = α ∪ β = γ

TSn−(X) ={H ∈ Sn | 〈ei, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ � 0

}

lin(TSn−(X)) ={H ∈ Sn | 〈ei, λ′(X;H)〉 = 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ = 0

}

17

Page 67: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets: SDP

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

X = V

0α · · · 0... 0β

...

0 · · · Λγ(X)

V T , ι2(λ(X)) = α ∪ β = γ

TSn−(X) ={H ∈ Sn | 〈ei, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ � 0

}

lin(TSn−(X)) ={H ∈ Sn | 〈ei, λ′(X;H)〉 = 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ = 0

}

17

Page 68: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Tangent sets: SDP

Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

X = V

0α · · · 0... 0β

...

0 · · · Λγ(X)

V T , ι2(λ(X)) = α ∪ β = γ

TSn−(X) ={H ∈ Sn | 〈ei, λ′(X;H)〉 ≤ 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ � 0

}

lin(TSn−(X)) ={H ∈ Sn | 〈ei, λ′(X;H)〉 = 0 ∀ i ∈ ι2(λ(X))

}=

{H ∈ Sn | V TγHV γ = 0

}17

Page 69: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y .

• Critical cone:

C(A; ∂θ1(X)) :={H ∈ Sn | θ′1(X;H) ≤ 〈Y ,H〉

}=

{H ∈ Sn | θ′1(X;H) = 〈Y ,H〉

}Proposition

H ∈ C(A; ∂θ1(X)) if and only if H ∈ Sn satisfies for any i, j ∈ η1(x,y),

〈diag(UTHU),ai〉 = 〈diag(U

THU),aj〉 = max

i∈ι1(x)〈λ′(X;H),ai〉,

where the index set η1(x,y) ⊆ ι1(x):

η1(x,y) :={i ∈ ι1(x) |

∑i∈ι1(x)

uiai = y,∑

i∈ι1(x)

ui = 1, 0 < ui ≤ 1}

with x := λ(X) and y := λ(Y ).

18

Page 70: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y .

• Critical cone:

C(A; ∂θ1(X)) :={H ∈ Sn | θ′1(X;H) ≤ 〈Y ,H〉

}=

{H ∈ Sn | θ′1(X;H) = 〈Y ,H〉

}

Proposition

H ∈ C(A; ∂θ1(X)) if and only if H ∈ Sn satisfies for any i, j ∈ η1(x,y),

〈diag(UTHU),ai〉 = 〈diag(U

THU),aj〉 = max

i∈ι1(x)〈λ′(X;H),ai〉,

where the index set η1(x,y) ⊆ ι1(x):

η1(x,y) :={i ∈ ι1(x) |

∑i∈ι1(x)

uiai = y,∑

i∈ι1(x)

ui = 1, 0 < ui ≤ 1}

with x := λ(X) and y := λ(Y ).

18

Page 71: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y .

• Critical cone:

C(A; ∂θ1(X)) :={H ∈ Sn | θ′1(X;H) ≤ 〈Y ,H〉

}=

{H ∈ Sn | θ′1(X;H) = 〈Y ,H〉

}Proposition

H ∈ C(A; ∂θ1(X)) if and only if H ∈ Sn satisfies for any i, j ∈ η1(x,y),

〈diag(UTHU),ai〉 = 〈diag(U

THU),aj〉 = max

i∈ι1(x)〈λ′(X;H),ai〉,

where the index set η1(x,y) ⊆ ι1(x):

η1(x,y) :={i ∈ ι1(x) |

∑i∈ι1(x)

uiai = y,∑

i∈ι1(x)

ui = 1, 0 < ui ≤ 1}

with x := λ(X) and y := λ(Y ). 18

Page 72: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone (cont’d)

For θ2 = φ2 ◦ λ:

• Let Z ∈ NK(X). Denote B = X + Z.

• Critical cone:

C(B;NK(X)) := TK(X)∩Z⊥ ={H ∈ Sn | ζ ′(X;H) ≤ 0, 〈Z,H〉 = 0

}Proposition

H ∈ C(B;NK(X)) if and only if H ∈ Sn satisfies for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉,

where the index set η2(x, z) ⊆ ι2(x):

η2(x, z) :={i ∈ ι2(x) |

∑i∈ι2(x)

uibi = z, ui > 0}

with x := λ(X) and z := λ(Z).

19

Page 73: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone (cont’d)

For θ2 = φ2 ◦ λ:

• Let Z ∈ NK(X). Denote B = X + Z.

• Critical cone:

C(B;NK(X)) := TK(X)∩Z⊥ ={H ∈ Sn | ζ ′(X;H) ≤ 0, 〈Z,H〉 = 0

}Proposition

H ∈ C(B;NK(X)) if and only if H ∈ Sn satisfies for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉,

where the index set η2(x, z) ⊆ ι2(x):

η2(x, z) :={i ∈ ι2(x) |

∑i∈ι2(x)

uibi = z, ui > 0}

with x := λ(X) and z := λ(Z).

19

Page 74: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• Z ∈ NSn−(X)

X + Z = V

Λα(Z) · · · 0... 0β

...

0 · · · Λγ(X)

V T ,{

ι2(x) = α ∪ β = γ

η2(x, z) = α

H ∈ C(B;NSn−(X)) if and only if for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉

i.e.,

VTHV =

diag = 0 � 0 ×

� 0 �... ×

× × ×

⇐⇒ VTHV =

0 0 ×0 � 0 ×× × ×

20

Page 75: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• Z ∈ NSn−(X)

X + Z = V

Λα(Z) · · · 0... 0β

...

0 · · · Λγ(X)

V T ,{

ι2(x) = α ∪ β = γ

η2(x, z) = α

H ∈ C(B;NSn−(X)) if and only if for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉

i.e.,

VTHV =

diag = 0 � 0 ×

� 0 �... ×

× × ×

⇐⇒ VTHV =

0 0 ×0 � 0 ×× × ×

20

Page 76: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• Z ∈ NSn−(X)

X + Z = V

Λα(Z) · · · 0... 0β

...

0 · · · Λγ(X)

V T ,{

ι2(x) = α ∪ β = γ

η2(x, z) = α

H ∈ C(B;NSn−(X)) if and only if for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉

i.e.,

VTHV =

diag = 0 � 0 ×

� 0 �... ×

× × ×

⇐⇒ VTHV =

0 0 ×0 � 0 ×× × ×

20

Page 77: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• Z ∈ NSn−(X)

X + Z = V

Λα(Z) · · · 0... 0β

...

0 · · · Λγ(X)

V T ,{

ι2(x) = α ∪ β = γ

η2(x, z) = α

H ∈ C(B;NSn−(X)) if and only if for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉

i.e.,

VTHV =

diag = 0 � 0 ×

� 0 �... ×

× × ×

⇐⇒ VTHV =

0 0 ×0 � 0 ×× × ×

20

Page 78: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• Z ∈ NSn−(X)

X + Z = V

Λα(Z) · · · 0... 0β

...

0 · · · Λγ(X)

V T ,{

ι2(x) = α ∪ β = γ

η2(x, z) = α

H ∈ C(B;NSn−(X)) if and only if for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉

i.e.,

VTHV =

diag = 0 � 0 ×

� 0 �... ×

× × ×

⇐⇒ VTHV =

0 0 ×0 � 0 ×× × ×

20

Page 79: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Critical cone: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0} = {X ∈ Sn | max1≤i≤n

{〈ei, λ(X)〉} ≤ 0}

• Z ∈ NSn−(X)

X + Z = V

Λα(Z) · · · 0... 0β

...

0 · · · Λγ(X)

V T ,{

ι2(x) = α ∪ β = γ

η2(x, z) = α

H ∈ C(B;NSn−(X)) if and only if for any i ∈ η2(x, z),

0 = 〈diag(VTHV ),bi〉 = max

i∈ι2(x)〈λ′(X;H),bi〉

i.e.,

VTHV =

diag = 0 � 0 ×

� 0 �... ×

× × ×

⇐⇒ VTHV =

0 0 ×0 � 0 ×× × ×

20

Page 80: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y and H ∈ C(A; ∂θ1(X)).

• θ1 is (parabolic) second-order directionally differentiable:

z(W ) := θ′′1 (X;H,W ) = φ′′1(λ(X);λ′(X;H), λ′′(X;H,W ))

The σ-term of θ1 , the conjugate function z∗(Y )

Moreover,

z∗(Y ) = 2

r∑l=1

〈Λ(Y )αlαl , UTαlH(X − vlI)†HUαl〉 := Υ1

X

(Y ,H

)

Υ1X

(Y ,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Y )− λj(Y )

λi(X)− λj(X)(U

T

αlHUαl′ )2ij

21

Page 81: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y and H ∈ C(A; ∂θ1(X)).

• θ1 is (parabolic) second-order directionally differentiable:

z(W ) := θ′′1 (X;H,W ) = φ′′1(λ(X);λ′(X;H), λ′′(X;H,W ))

The σ-term of θ1 , the conjugate function z∗(Y )

Moreover,

z∗(Y ) = 2

r∑l=1

〈Λ(Y )αlαl , UTαlH(X − vlI)†HUαl〉 := Υ1

X

(Y ,H

)

Υ1X

(Y ,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Y )− λj(Y )

λi(X)− λj(X)(U

T

αlHUαl′ )2ij

21

Page 82: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y and H ∈ C(A; ∂θ1(X)).

• θ1 is (parabolic) second-order directionally differentiable:

z(W ) := θ′′1 (X;H,W ) = φ′′1(λ(X);λ′(X;H), λ′′(X;H,W ))

The σ-term of θ1 , the conjugate function z∗(Y )

Moreover,

z∗(Y ) = 2

r∑l=1

〈Λ(Y )αlαl , UTαlH(X − vlI)†HUαl〉 := Υ1

X

(Y ,H

)

Υ1X

(Y ,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Y )− λj(Y )

λi(X)− λj(X)(U

T

αlHUαl′ )2ij

21

Page 83: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y and H ∈ C(A; ∂θ1(X)).

• θ1 is (parabolic) second-order directionally differentiable:

z(W ) := θ′′1 (X;H,W ) = φ′′1(λ(X);λ′(X;H), λ′′(X;H,W ))

The σ-term of θ1 , the conjugate function z∗(Y )

Moreover,

z∗(Y ) = 2

r∑l=1

〈Λ(Y )αlαl , UTαlH(X − vlI)†HUαl〉

:= Υ1X

(Y ,H

)

Υ1X

(Y ,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Y )− λj(Y )

λi(X)− λj(X)(U

T

αlHUαl′ )2ij

21

Page 84: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y and H ∈ C(A; ∂θ1(X)).

• θ1 is (parabolic) second-order directionally differentiable:

z(W ) := θ′′1 (X;H,W ) = φ′′1(λ(X);λ′(X;H), λ′′(X;H,W ))

The σ-term of θ1 , the conjugate function z∗(Y )

Moreover,

z∗(Y ) = 2

r∑l=1

〈Λ(Y )αlαl , UTαlH(X − vlI)†HUαl〉 := Υ1

X

(Y ,H

)

Υ1X

(Y ,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Y )− λj(Y )

λi(X)− λj(X)(U

T

αlHUαl′ )2ij

21

Page 85: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”

For θ1 = φ1 ◦ λ:

• Let Y ∈ ∂θ1(X). Denote A = X + Y and H ∈ C(A; ∂θ1(X)).

• θ1 is (parabolic) second-order directionally differentiable:

z(W ) := θ′′1 (X;H,W ) = φ′′1(λ(X);λ′(X;H), λ′′(X;H,W ))

The σ-term of θ1 , the conjugate function z∗(Y )

Moreover,

z∗(Y ) = 2

r∑l=1

〈Λ(Y )αlαl , UTαlH(X − vlI)†HUαl〉 := Υ1

X

(Y ,H

)

Υ1X

(Y ,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Y )− λj(Y )

λi(X)− λj(X)(U

T

αlHUαl′ )2ij

21

Page 86: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term” (cont’d)

For θ2 = φ2 ◦ λ = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ

Let Z ∈ NK(X). Denote B = X + Z and H ∈ C(B,NK(X))

the “σ-term” of K , the support function of T 2K(X,H)

δ∗T 2K(X,H)

(Z) = 2

r∑l=1

〈Λ(Z)αlαl , VT

αlH(X − vlI)†HV αl〉 := Υ2X

(Z,H

)

Υ2X

(Z,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Z)− λj(Z)

λi(X)− λj(X)(V

T

αlHV αl′ )2ij

22

Page 87: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term” (cont’d)

For θ2 = φ2 ◦ λ = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ

Let Z ∈ NK(X). Denote B = X + Z and H ∈ C(B,NK(X))

the “σ-term” of K , the support function of T 2K(X,H)

δ∗T 2K(X,H)

(Z) = 2

r∑l=1

〈Λ(Z)αlαl , VT

αlH(X − vlI)†HV αl〉

:= Υ2X

(Z,H

)

Υ2X

(Z,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Z)− λj(Z)

λi(X)− λj(X)(V

T

αlHV αl′ )2ij

22

Page 88: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term” (cont’d)

For θ2 = φ2 ◦ λ = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ

Let Z ∈ NK(X). Denote B = X + Z and H ∈ C(B,NK(X))

the “σ-term” of K , the support function of T 2K(X,H)

δ∗T 2K(X,H)

(Z) = 2

r∑l=1

〈Λ(Z)αlαl , VT

αlH(X − vlI)†HV αl〉 := Υ2X

(Z,H

)

Υ2X

(Z,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Z)− λj(Z)

λi(X)− λj(X)(V

T

αlHV αl′ )2ij

22

Page 89: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term” (cont’d)

For θ2 = φ2 ◦ λ = δK with

K = {X ∈ Sn | λ(X) ∈ domφ} = {X ∈ Sn | ζ(X) ≤ 0} ,

where ζ = ψ ◦ λ

Let Z ∈ NK(X). Denote B = X + Z and H ∈ C(B,NK(X))

the “σ-term” of K , the support function of T 2K(X,H)

δ∗T 2K(X,H)

(Z) = 2

r∑l=1

〈Λ(Z)αlαl , VT

αlH(X − vlI)†HV αl〉 := Υ2X

(Z,H

)

Υ2X

(Z,H

)= −2

∑1≤l<l′≤r

∑i∈αl

∑j∈αl′

λi(Z)− λj(Z)

λi(X)− λj(X)(V

T

αlHV αl′ )2ij

22

Page 90: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0}

• Z ∈ NSn−(X), B = X + Z, H ∈ C(B,NK(X))

X + Z = V

Λα(Z) · · · 0... 0β 0

0 · · · Λγ(X)

V TThe “σ-term” of Sn−:

Υ2X

(Z,H

)= 2

∑i∈γ,j∈α

λj(Z)

λi(X)(H)2ij , cf. (Sun, 2006)

where H = VTHV .

23

Page 91: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0}

• Z ∈ NSn−(X), B = X + Z, H ∈ C(B,NK(X))

X + Z = V

Λα(Z) · · · 0... 0β 0

0 · · · Λγ(X)

V TThe “σ-term” of Sn−:

Υ2X

(Z,H

)= 2

∑i∈γ,j∈α

λj(Z)

λi(X)(H)2ij ,

cf. (Sun, 2006)

where H = VTHV .

23

Page 92: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

The “σ-term”: SDP

• Sn− = {X ∈ Sn | λmax(X) ≤ 0}

• Z ∈ NSn−(X), B = X + Z, H ∈ C(B,NK(X))

X + Z = V

Λα(Z) · · · 0... 0β 0

0 · · · Λγ(X)

V TThe “σ-term” of Sn−:

Υ2X

(Z,H

)= 2

∑i∈γ,j∈α

λj(Z)

λi(X)(H)2ij , cf. (Sun, 2006)

where H = VTHV .

23

Page 93: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Robinson CQ

CMatOP:minimize

x∈Xf(x) + θ1(g(x))

subject to h(x) = 0,

g(x) ∈ K

Proposition

Let x ∈ X be a feasible point of the CMatOP. We say that the

Robinson CQ (RCQ) holds at x if[h′(x)

g′(x)

]X +

[{0}

TK(g(x))

]=

[Y

Sn

].

Thus, the set of Lagrange multipliers M(x) is a non-empty, convex,

bounded and compact subset if and only if the RCQ holds at x.

24

Page 94: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Robinson CQ

CMatOP:minimize

x∈Xf(x) + θ1(g(x))

subject to h(x) = 0,

g(x) ∈ K

Proposition

Let x ∈ X be a feasible point of the CMatOP. We say that the

Robinson CQ (RCQ) holds at x if[h′(x)

g′(x)

]X +

[{0}

TK(g(x))

]=

[Y

Sn

].

Thus, the set of Lagrange multipliers M(x) is a non-empty, convex,

bounded and compact subset if and only if the RCQ holds at x.

24

Page 95: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Second-order optimality conditions

Critical cone of CMatOP:

C(x) := {d ∈ X | h′(x)d = 0, g′(x)d ∈ C(A; ∂θ1(g(x))), g′(x)d ∈ C(B;NK(g(x)))}

Theorem (“no gap” second-order optimality conditions)

Suppose that x ∈ X is a locally optimal solution of the CMatOP andthe RCQ holds. Then, the following inequality holds: for any d ∈ C(x),

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}≥ 0.

Conversely, let x be a feasible point of the CMatOP such that M(x) isnonempty. Suppose that the RCQ holds at x. Then the followingcondition: for any d ∈ C(x) \ {0},

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}> 0

is necessary and sufficient for the quadratic growth condition at thepoint x: for any x ∈ N such that h(x) = 0 and g(x) ∈ K,

f(x) + φ1 ◦ λ(g(x)) ≥ f(x) + φ1 ◦ λ(g(x)) + ρ‖x− x‖2.

25

Page 96: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Second-order optimality conditions

Critical cone of CMatOP:

C(x) := {d ∈ X | h′(x)d = 0, g′(x)d ∈ C(A; ∂θ1(g(x))), g′(x)d ∈ C(B;NK(g(x)))}

Theorem (“no gap” second-order optimality conditions)

Suppose that x ∈ X is a locally optimal solution of the CMatOP andthe RCQ holds. Then, the following inequality holds: for any d ∈ C(x),

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}≥ 0.

Conversely, let x be a feasible point of the CMatOP such that M(x) isnonempty. Suppose that the RCQ holds at x. Then the followingcondition: for any d ∈ C(x) \ {0},

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}> 0

is necessary and sufficient for the quadratic growth condition at thepoint x: for any x ∈ N such that h(x) = 0 and g(x) ∈ K,

f(x) + φ1 ◦ λ(g(x)) ≥ f(x) + φ1 ◦ λ(g(x)) + ρ‖x− x‖2.

25

Page 97: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Second-order optimality conditions

Critical cone of CMatOP:

C(x) := {d ∈ X | h′(x)d = 0, g′(x)d ∈ C(A; ∂θ1(g(x))), g′(x)d ∈ C(B;NK(g(x)))}

Theorem (“no gap” second-order optimality conditions)

Suppose that x ∈ X is a locally optimal solution of the CMatOP andthe RCQ holds. Then, the following inequality holds: for any d ∈ C(x),

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}≥ 0.

Conversely, let x be a feasible point of the CMatOP such that M(x) isnonempty. Suppose that the RCQ holds at x. Then the followingcondition: for any d ∈ C(x) \ {0},

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}> 0

is necessary and sufficient for the quadratic growth condition at thepoint x: for any x ∈ N such that h(x) = 0 and g(x) ∈ K,

f(x) + φ1 ◦ λ(g(x)) ≥ f(x) + φ1 ◦ λ(g(x)) + ρ‖x− x‖2.

25

Page 98: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Second-order optimality conditions

Critical cone of CMatOP:

C(x) := {d ∈ X | h′(x)d = 0, g′(x)d ∈ C(A; ∂θ1(g(x))), g′(x)d ∈ C(B;NK(g(x)))}

Theorem (“no gap” second-order optimality conditions)

Suppose that x ∈ X is a locally optimal solution of the CMatOP andthe RCQ holds. Then, the following inequality holds: for any d ∈ C(x),

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}≥ 0.

Conversely, let x be a feasible point of the CMatOP such that M(x) isnonempty. Suppose that the RCQ holds at x. Then the followingcondition: for any d ∈ C(x) \ {0},

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}> 0

is necessary and sufficient for the quadratic growth condition at thepoint x: for any x ∈ N such that h(x) = 0 and g(x) ∈ K,

f(x) + φ1 ◦ λ(g(x)) ≥ f(x) + φ1 ◦ λ(g(x)) + ρ‖x− x‖2.

25

Page 99: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Strong second-order sufficient condition

Definition

Let x ∈ X be a stationary point of the CMatOP. We say the strong

second-order sufficient condition holds at x if for any d ∈ C(x) \ {0},

sup(y,Y,Z)∈M(x)

{〈d,L′′xx(x,y, Y, Z)d〉 −Υ1

g(x)

(Y, g′(x)d

)−Υ2

g(x)

(Z, g′(x)d

)}> 0

with

C(x) :=⋂

(y,Y,Z)∈M(x)

app(y, Y, Z),

where for any (y, Y, Z) ∈M(x), the set app(y, Y, Z) is given by

app(y, Y , Z) := {d ∈ X | h′(x)d = 0, g′(x)d ∈ aff(C(A; ∂θ1(g(x)))),

g′(x)d ∈ aff(C(B;NK(g(x))))}.

26

Page 100: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Constraint nondegeneracy (LICQ)

The constraint nondegeneracy for the CMatOP is defined as followsh′(x)

g′(x)

g′(x)

X +

{0}

T linθ1

(g(x))

lin (TK(g(x)))

=

Y

Sn

Sn

.

27

Page 101: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Strong regularity of CMatOPs

Theorem

Let x ∈ X be a stationary point of CMatOP with multipliers (y, Y , Z):

(i) the strong second order sufficient condition and constraint

nondegeneracy hold at x;

(ii) every element in ∂F (x,y, Y , Z) is nonsingular;

(iii) (x,y, Y , Z) is a strongly regular solution of the KKT system.

It holds that (i) =⇒ (ii) =⇒ (iii).

(iii) =⇒ (i) can be established for particular CMatOPs:

• NLSDP (Sun, MOR 2006)

• CMatOPs with the sum of k-largest eigenvalues, etc (in our work)

28

Page 102: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Strong regularity of CMatOPs

Theorem

Let x ∈ X be a stationary point of CMatOP with multipliers (y, Y , Z):

(i) the strong second order sufficient condition and constraint

nondegeneracy hold at x;

(ii) every element in ∂F (x,y, Y , Z) is nonsingular;

(iii) (x,y, Y , Z) is a strongly regular solution of the KKT system.

It holds that (i) =⇒ (ii) =⇒ (iii).

(iii) =⇒ (i) can be established for particular CMatOPs:

• NLSDP (Sun, MOR 2006)

• CMatOPs with the sum of k-largest eigenvalues, etc (in our work)

28

Page 103: Perturbation analysis of matrix optimization · 2019-08-13 · Perturbation analysis of matrix optimization ... a symmetric function, i.e., for any u2Rn and any n npermutation matrix

Thank you!