introduction to combinatorial matrix theory
TRANSCRIPT
-
7/31/2019 Introduction to Combinatorial Matrix Theory
1/69
-
7/31/2019 Introduction to Combinatorial Matrix Theory
2/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Combinatorial Matrix Theory
ApplicationsMatricesGraph Terminology
Minimum Rank ProblemsMinimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Matrix Completion ProblemsDefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
References
-
7/31/2019 Introduction to Combinatorial Matrix Theory
3/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Combinatorial Matrix Theory
- Studies patterns of entries in a matrix rather than values.
- In some applications, only the sign of the entry (or whether itis nonzero) is known, not the numerical value.
- In other cases, some entries are missing.
- Uses graphs or digraphs to describe patterns.
-
7/31/2019 Introduction to Combinatorial Matrix Theory
4/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Applications: Sign Patterns
- Study of sign patterns arose in economics and has applicationsto ecology.
- Behavior of supply and demand equilibrium or predator-prey
systems under perturbation can be predicted by a sign patternmatrix of partial derivatives whose signs are known.
- Sign stable patterns guarantee stability.
- Sign nonsingular (SNS) patterns guarantee invertibility andunique solution of a system.
- SNS patterns were first studied in the Dimer Problem instatistical mechanics involving the bonding of atoms in amolecule.
-
7/31/2019 Introduction to Combinatorial Matrix Theory
5/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Applications: Matrix Completions
- Useful when some data unavailable but full matrix has certaintype.
- Used in seismology and geophysics (positive definite
matrices).
- Biomolecular modeling (Euclidean distance matrices).
- Image enhancement.
-
7/31/2019 Introduction to Combinatorial Matrix Theory
6/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
All (numerical) matrices discussed are real and symmetric, but the
ideas can be generalized to nonsymmetric matrices.Submatrices
- The principal submatrix A[] consists of the entries in rowsand columns in (i.e., delete rows and columns not in ).
- The principal submatrix obtained by deleting row and columnk is denoted A(k).
Example:
A = 3 2 0 0
2 1 1 10 1 1 30 1 3 1
A(2) = 3 0 00 1 30 3 1
A[{1,
2}] = 3 2
2 1
-
7/31/2019 Introduction to Combinatorial Matrix Theory
7/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Eigenvalue Interlacing
Theorem
If A is a real symmetric matrix, then the eigenvalues of A(k)interlace the eigenvalues of A.
Example:
A =
3 2 0 02 1 1 10 1 1 3
0 1 3 1
A(2) =
3 0 00 1 3
0 3 1
(A) = {5, 3.56155,0.561553,2} (A(2)) = {4, 3,2}
5
4
3.56155
3
0.561553
2
2
-
7/31/2019 Introduction to Combinatorial Matrix Theory
8/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Graph Terminology
- A simple graph G does not allow loops or multiple edges.- A graph G allows loops but does not allow multiple edges.
- The simple graph associated with G, denoted
G, is obtained
from G by suppressing loops.
Example:
G G
-
7/31/2019 Introduction to Combinatorial Matrix Theory
9/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
- A path is a sequence of distinct vertices v0, v1, v2, . . . , vk such
that for i = 1, . . . , k, vi1vi is an edge.- A cycle is a sequence of vertices v1, v2, . . . , vk, v1 such that
i = j vi = vj and vkv1 is an edge, vi1vi is an edgei = 2, . . . , k.
Example:A path and a cycle.
C bi i l M i Th
-
7/31/2019 Introduction to Combinatorial Matrix Theory
10/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
- A graph is connected if there is a path from any vertex to any
other vertex.- A graph is acyclic if it has no cycles.
- A graph T is a tree if T is connected and T is acyclic.
Example:
A tree.
C bi t i l M t i Th
-
7/31/2019 Introduction to Combinatorial Matrix Theory
11/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Matrices and graphs
- We will associate matrices with graphs in several ways.
- With an n
n matrix we associate a graph having vertices
{1, . . . , n}.- The i,j entry will be associated with the edge (or non-edge)
between i and j.
Combinatorial Matrix Theory
-
7/31/2019 Introduction to Combinatorial Matrix Theory
12/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Permutation Similarity and Vertex Numbering
- Applying a permutation similarity to a matrix does not changeits eigenvalues or its rank.
- Applying a permutation similarity to a symmetric matrix Acorresponds to renumbering the vertices of the graph of A
- Unlabeled graph diagrams can be used.
Combinatorial Matrix Theory
-
7/31/2019 Introduction to Combinatorial Matrix Theory
13/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
ApplicationsMatricesGraph Terminology
Block Diagonal Matrices and Graph Components
- If A = A1 A2 Ak, then the spectrum of A is theunion of the spectra of A1,A2, . . . , Ak andrank(A) =
ki=1 rank(Ai).
- Any symmetric matrix is permutation similar to a direct sumof irreducible matrices (block diagonal matrix).
- The irreducible summands of A correspond to the connectedcomponents of the graph of A.
- Only connected graphs are studied.
Combinatorial Matrix Theory Minimum Rank Problems for Simple Graphsf S
-
7/31/2019 Introduction to Combinatorial Matrix Theory
14/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Let A be a symmetric n n matrix.- The simple graph G(A) of A is the graph having
- vertices 1, . . . , n- edge ij of G(A) if and only if i = j and aij = 0.
- The diagonal is ignored.Example: G(A)
A =
2 1 3 51 0 0 03 0 3 05 0 0 0
1 2
34
Combinatorial Matrix Theory Minimum Rank Problems for Simple GraphsMi i R k P bl f Si l T
-
7/31/2019 Introduction to Combinatorial Matrix Theory
15/69
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Minimum Rank Problems for Simple Graphs
Let A be a symmetric n n matrix and G a simple graph.-
S(
G) = {A : A is a symmetric matrix and
G(A) =
G}.
- The minimum rank of G ismr(G) = min
AbS( bG){rank(A)}.
Combinatorial Matrix Theory Minimum Rank Problems for Simple GraphsMi i R k P bl s f Si l T s
-
7/31/2019 Introduction to Combinatorial Matrix Theory
16/69
yMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Let A be a symmetric n
n matrix and G a simple graph.
- mult(A) denotes the multiplicity of eigenvalue of A.
- The maximum multiplicity of an eigenvalue in
S(
G) is
M(G) = maxAbS( bG)
{mult(A)}.
-
mr(
G) +
M(
G) = the number of vertices of
G.
Combinatorial Matrix Theory Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple Trees
-
7/31/2019 Introduction to Combinatorial Matrix Theory
17/69
yMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Let
G have n vertices.
- It is easy to obtain a matrix A S(G) with rank A = n 1(translate).
- It is easy to have all eigenvalues distinct (i.e., rank A n 1).- For a path,
mr(
Pn) = n 1.
Example: # is nonzero, is indefinite
A =
# 0 . . . 0 0# # . . . 0 00 # . . . 0 0...
..
.
..
. . . .
..
.
..
.0 0 0 . . . #0 0 0 . . . #
-
7/31/2019 Introduction to Combinatorial Matrix Theory
18/69
Combinatorial Matrix TheoryMi i R k P bl
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple Trees
-
7/31/2019 Introduction to Combinatorial Matrix Theory
19/69
Minimum Rank ProblemsMatrix Completion Problems
References
Minimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Algorithm (for computing , and thus M and mr)Let T be a simple tree.Set Q = (set of vertices to be deleted).Set H = {vertices of
T with degree 3} (candidates for deletion).
While H
=
:
1. Set TH = the unique component of T Q that contains anH-vertex.
2. Set W = {w H:at most 1 component of
TH
w is not H-free
}.
3. Q = Q W .4. H = H W .5. For each v H, ifdeg bTQv 2, remove v from H.
Combinatorial Matrix TheoryMi i R k P bl
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple Trees
-
7/31/2019 Introduction to Combinatorial Matrix Theory
20/69
Minimum Rank ProblemsMatrix Completion Problems
References
Minimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the tree
T by
computing (T) = maximum multiplicity of eigenvalue 0:
1
2
3
4 5 6
8 7
Combinatorial Matrix TheoryMinimum Rank Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple Trees
-
7/31/2019 Introduction to Combinatorial Matrix Theory
21/69
Minimum Rank ProblemsMatrix Completion Problems
References
pSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the tree
T by
computing (T) = maximum multiplicity of eigenvalue 0:
1
2
3
45 6
8 7
Combinatorial Matrix TheoryMinimum Rank Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple Trees
-
7/31/2019 Introduction to Combinatorial Matrix Theory
22/69
Minimum Rank ProblemsMatrix Completion Problems
References
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Algorithm (for computing , and thus M and mr)While H = :
1. Set TH = the unique component of T Q that contains anH-vertex.
2. Set W = {w H:at most 1 component of TH w is not H-free}.3. Q = Q W .4. H = H W .
5. For each v H, ifdeg bTQv 2, remove v from H.
Combinatorial Matrix TheoryMinimum Rank Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple Trees
-
7/31/2019 Introduction to Combinatorial Matrix Theory
23/69
Minimum Rank ProblemsMatrix Completion Problems
References
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the tree
T by
computing (T) = maximum multiplicity of eigenvalue 0:
1
2
3
4 56
8 7
Combinatorial Matrix TheoryMinimum Rank Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSi l G h
-
7/31/2019 Introduction to Combinatorial Matrix Theory
24/69
Minimum Rank ProblemsMatrix Completion Problems
References
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Algorithm (for computing , and thus M and mr)While H = :
1. Set TH = the unique component of T Q that contains anH-vertex.
2. Set W = {w H:at most 1 component of TH w is not H-free}.3. Q = Q W .4. H = H W .
5. For each v H, ifdeg bTQv 2, remove v from H.
Combinatorial Matrix TheoryMinimum Rank Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSi l G h
-
7/31/2019 Introduction to Combinatorial Matrix Theory
25/69
Matrix Completion ProblemsReferences
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the tree
T by
computing (T) = maximum multiplicity of eigenvalue 0:
1
2
3
45 6
8 7
Combinatorial Matrix TheoryMinimum Rank Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple Graphs
-
7/31/2019 Introduction to Combinatorial Matrix Theory
26/69
Matrix Completion ProblemsReferences
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
Second Iteration:
Algorithm (for computing , and thus M and mr)While H = :
1. Set
TH = the unique component of
T Q that contains an
H-vertex.
2. Set W = {w H:at most 1 component of TH w is not H-free}.
3. Q = Q W .4. H = H
W .
5. For each v H, ifdeg bTQv 2, remove v from H.
Combinatorial Matrix TheoryMinimum Rank Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple Graphs
-
7/31/2019 Introduction to Combinatorial Matrix Theory
27/69
Matrix Completion ProblemsReferences
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the tree
T by
computing (T) = maximum multiplicity of eigenvalue 0:
1
2
3
45 6
7
Combinatorial Matrix TheoryMinimum Rank Problems
C
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple Graphs
-
7/31/2019 Introduction to Combinatorial Matrix Theory
28/69
Matrix Completion ProblemsReferences
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the tree
T by
computing (T) = maximum multiplicity of eigenvalue 0:
1
2
3
5 6
7
Combinatorial Matrix TheoryMinimum Rank Problems
M i C l i P bl
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple Graphs
-
7/31/2019 Introduction to Combinatorial Matrix Theory
29/69
Matrix Completion ProblemsReferences
Simple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the simple tree
T by
computing maximum multiplicity of eigenvalue 0:
- Q = {1, 2, 3, 5, 6, 7}, i.e., 6 vertices were deleted.- There are 18 paths.
- M(T) = 18 6 = 12.- mr(T) = 35 12 = 23.
Combinatorial Matrix TheoryMinimum Rank Problems
M t i C l ti P bl
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple Graphs
-
7/31/2019 Introduction to Combinatorial Matrix Theory
30/69
Matrix Completion ProblemsReferences
p pSign PatternsMinimum Rank Problems for Tree Sign Patterns
Trees are done, but work continues on
- minimum rank of arbitrary graphs-
- if graph has a cut-vertex, reduce the problem to the pieces[BFH04].
- new Colin de Verdiere type parameter used for computation ofminimum rank [BFH05],[HvdH06]
- matrices over fields other than R
- positive semidefinite minimum rank
- etc.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple Graphs
-
7/31/2019 Introduction to Combinatorial Matrix Theory
31/69
Matrix Completion ProblemsReferences
p pSign PatternsMinimum Rank Problems for Tree Sign Patterns
Sign Patterns
- A sign pattern is an n n matrix Z = [zij] whose entries zijare elements of{+,, 0}.
- The qualitative class or sign pattern class of Z,
Q(Z) =
{A : sgn(aij) = zij, 1
i,j
n
}.
- The graph G(Z) of the symmetric n n sign pattern Z:- vertices 1, . . . , n- ij is an edge ofG(Z) if and only if zij = 0.
- The graph has loops.
- A symmetric sign pattern Z is a tree sign pattern (TSP) ifG(Z) is a tree.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple Graphs
-
7/31/2019 Introduction to Combinatorial Matrix Theory
32/69
Matrix Completion ProblemsReferences
Sign PatternsMinimum Rank Problems for Tree Sign Patterns
Minimum Rank Problems for Tree Sign Patterns
- For symmetric Z, S(Z) ={A : A is a symmetric matrix and for all i,j, sgn(aij) = zij}.
- The diagonal is restricted by the sign pattern.
- The minimum rank of Z is the minimum rank among matrices
having sign pattern Z.
- The symmetric minimum rank of symmetric Z is
mr(Z) = min
AS
(Z){rank(A)
}.
- For a symmetric tree sign pattern, minimum rank equalssymmetric minimum rank.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSi P
-
7/31/2019 Introduction to Combinatorial Matrix Theory
33/69
Matrix Completion ProblemsReferences
Sign PatternsMinimum Rank Problems for Tree Sign Patterns
- The maximum multiplicity of an eigenvalue inS
(Z) is
M(Z) = maxAS(Z)
{mult(A)}.
- All positive (respectively, negative) eigenvalues have the same
maximum multiplicity.
- mr(Z) + M0(Z) = the size of Z.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSi P tt
-
7/31/2019 Introduction to Combinatorial Matrix Theory
34/69
Matrix Completion ProblemsReferences
Sign PatternsMinimum Rank Problems for Tree Sign Patterns
Let Z be a (symmetric) sign pattern.- S(Z) (or G(Z)) allows eigenvalue if there is a matrix
A S(Z) having eigenvalue .- S(Z) allows eigenvalue 0 if and only if det Z is identically 0 or
has both positive and negative terms.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
35/69
pReferences
Sign PatternsMinimum Rank Problems for Tree Sign Patterns
Algorithm (Computation ofM for TSP [DHHHW06] )
Let Z be a TSP and T = G(Z).Set Q = (set of vertices to be deleted).Set H = {vertices of T with degree 3} (candidates for deletion).While H = :
1. Set TH = the unique component of T Q that contains anH-vertex.
2. Set W = {w H:at most 1 component of TH w is not H-free}.
3. For each w W , if there are at least two H-free componentsof TH w that allow eigenvalue, then Q = Q {w}.4. H = H W .5. For each v H, ifdegTQv 2, remove v from H.
M
(Z)=(number of components of T
Q that allow) |
Q|.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
36/69
ReferencesSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Z
by computing maximum multiplicity of eigenvalue 0:
1
2
3
56
4
8
9
7
10
11
12
1314
15
16
17 1819
20
_
_
_
_
_
+
+
_+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion Problems
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
37/69
ReferencesSign PatternsMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Z
by computing maximum multiplicity of eigenvalue 0:
1
2
3
56
4
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
_+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsR f
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
38/69
ReferencesSign PatternsMinimum Rank Problems for Tree Sign Patterns
Algorithm (Computation ofM0)
While H = :1. Set TH = the unique component of T Q that contains an
H-vertex.
2. Set W =
{w
H:
at most 1 component of TH w is not H-free}.3. For each w W , if there are at least two H-free components
of TH w that allow eigenvalue 0, then Q = Q {w}.4. H = H
W .
5. For each v H, ifdegTQv 2, remove v from H.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsR f
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
39/69
ReferencesS g atte sMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Z
by computing maximum multiplicity of eigenvalue 0:
1
2
3
56
4
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
_+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
40/69
Referencesg
Minimum Rank Problems for Tree Sign Patterns
Algorithm (Computation ofM0)
While H = :1. Set TH = the unique component of T Q that contains an
H-vertex.
2. Set W =
{w
H:
at most 1 component of TH w is not H-free}.3. For each w W , if there are at least two H-free components
of TH w that allow eigenvalue 0, then Q = Q {w}.4. H = H
W .
5. For each v H, ifdegTQv 2, remove v from H.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
41/69
ReferencesMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Z
by computing maximum multiplicity of eigenvalue 0:
1
2
3
56
4
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
_+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
42/69
ReferencesMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Z
by computing maximum multiplicity of eigenvalue 0:
1
2
3
56
4
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
_+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMi i R k P bl f T Si P
-
7/31/2019 Introduction to Combinatorial Matrix Theory
43/69
ReferencesMinimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Z
by computing maximum multiplicity of eigenvalue 0:
1
2
3
56
4
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
_+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMi i R k P bl f T Si P tt
-
7/31/2019 Introduction to Combinatorial Matrix Theory
44/69
Minimum Rank Problems for Tree Sign Patterns
Algorithm (Computation ofM0)
While H = :1. Set TH = the unique component of T Q that contains an
H-vertex.
2. Set W =
{w
H:
at most 1 component of TH w is not H-free}.3. For each w W , if there are at least two H-free components
of TH w that allow eigenvalue 0, then Q = Q {w}.4. H = H
W .
5. For each v H, ifdegTQv 2, remove v from H.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
45/69
Minimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Z
by computing maximum multiplicity of eigenvalue 0:
1
2
3
5
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
46/69
Minimum Rank Problems for Tree Sign Patterns
Algorithm (Computation ofM0)
While H = :1. Set TH = the unique component of T Q that contains an
H-vertex.
2. Set W =
{w
H:
at most 1 component of TH w is not H-free}.3. For each w W , if there are at least two H-free components
of TH w) that allow eigenvalue 0, then Q = Q {w}.4. H = H W .5. For each v H, ifdegTQv 2, remove v from H.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
47/69
Minimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Zby computing maximum multiplicity of eigenvalue 0:
1
2
3
5
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
48/69
Minimum Rank Problems for Tree Sign Patterns
- Example: Compute the minimum rank of the sign pattern Zby computing maximum multiplicity of eigenvalue 0:
1 3
5
8
9
7
10
11
12
1314
15
16
17 18
1920
_
_
_
_
_
+
+
+
+
+
+
_
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
Minimum Rank Problems for Simple GraphsMinimum Rank Problems for Simple TreesSimple GraphsSign PatternsMinimum Rank Problems for Tree Sign Patterns
-
7/31/2019 Introduction to Combinatorial Matrix Theory
49/69
g
- Example: Compute the minimum rank of the sign pattern Zby computing maximum multiplicity of eigenvalue 0:
- Q = {2, 4, 6}, i.e., 3 vertices were deleted.- There are 8 components that allow eigenvalue 0.
- M0(Z) = 8 3 = 5.- mr(Z) = 20 5 = 15.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
50/69
Matrix Completion Problems
- A partial matrix is a square array in which some entries arespecified and others are not.
- A completion of a partial matrix is a choice of values for theunspecified entries.
Example:
B =
2 1 ? 01 2 2 ?
? 2 3 10 ? 1 1
A =
2 1 0 01 2 2 1
0 2 3 10 1 1 1
B is a partial matrix and A is a completion of B.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
51/69
The following are equivalent for real matrices:
- A is symmetric and for all x Rn, x = 0, xTAx > 0(A is positive definite).
- A is symmetric and all eigenvalues are positive.
- A is symmetric and all principal minors are positive.
The following are equivalent for real matrices:
- A is symmetric and for all x Rn, xTAx 0(A is positive semidefinite).
- A is symmetric and all eigenvalues are nonnegative.
- A is symmetric and all principal minors are nonnegative.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
52/69
Classes X of matrices to be discussed:
- positive definite matrices- positive semidefinite matrices
- strictly copositive matrices: A is strictly copositive if A issymmetric and for all x 0, x = 0, xTAx > 0
- copositive matrices: A is copositive if A is symmetric and forall x 0, xTAx 0
Example:
A = 2 1 3
1 1 23 2 2
is strictly copositive but not positive definite (any positive matrix isstrictly copositive).
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
53/69
- A matrix completion problem asks whether a partial matrix, or
family of partial matrices with a given pattern of specifiedentries, has a completion of a specific type X, such as apositive definite matrix.
Example:
B =
2 1 ? 01 2 2 ?
? 2 3 10 ? 1 1
A =
2 1 0 01 2 2 1
0 2 3 10 1 1 1
Matrix A completes B to a positive semidefinite matrix.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
54/69
- All classes X of matrices discussed are hereditary, i.e. if A isan X-matrix then every principal submatrix of A is anX-matrix.
- If X is hereditary, in order for a partial matrix B to have anX-completion, it is necessary that every fully specifiedprincipal submatrix of B is an X-matrix.
- A partial matrix B is a partial X-matrix if every fully specifiedprincipal submatrix of B is an X-matrix.
- a pattern of specified entries has the X-completion property if
every partial X-matrix B described by that pattern can becompleted to an X-matrix.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
55/69
Graph Theoretic Techniques
- Graphs are used to describe the pattern of specified entriesclasses of for symmetric matrices such as positive(semi)definite and (strictly) copositive matrices; otherwisedigraphs are used.
- The specified entries in partial matrix B are represented by
edges in the graph G(B).Example:
B =
2 1 ? 01 2 2 ?
? 2 3 10 ? 1 1
1 2
34
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
56/69
Theorem ([GJSW84])
- A graph having a loop at every vertex has the positive definitecompletion property if and only if it is chordal (any cycle oflength 4 has a chord).
- A graph has the positive definite completion property if and
only if the subgraph induced by the vertices with loops hasthe positive definite completion property.
Example:
has the positive definite completion property.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
57/69
Theorem ([GJSW84])A graph has the positive semidefinite completion property if andonly if each component has all loops or none, and componentswith loops are chordal.
Example:
does not have the positive semidefinite completion property.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
58/69
The (Strictly) Copositive Matrix Completion Problems
- A is strictly copositive if A is symmetric and for allx 0, x = 0, xTAx > 0.
- A is copositive if A is symmetric and for all x 0, xTAx 0.Example:
A =
1 1 1 1
1 1 1 11 1 1 1
1 1 1 1
is copositive but not strictly copositive, and not positivesemidefinite.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
59/69
- The partial matrix B is a partial strictly copositive matrix if
every fully specified principal submatrix of B is a strictlycopositive matrix.
- The partial matrix B is a partial copositive matrix if everyfully specified principal submatrix of B is a copositive matrix.
Example:
B =
3 1 x13 11 1 2 x24
x13 2 1 11 x24 1 2
is a partial strictly copositive matrix.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
60/69
Theorem ([HJR05])
Let B be a partial copositive matrix with every diagonal entryspecified. For each pair of unspecified off-diagonal entries, setxij = xji =
biibjj. The resulting matrix is copositive, and is
strictly copositive if B is a partial strictly copositive matrix.
Example:
B =
3 1 x13 11 1 2 x24
x13
2 1
11 x24 1 2
A =
3 1
3 1
1 1 2
23 2 1
1
1 2 1 2
A completes B to a strictly copositive matrix.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
61/69
Theorem ([H06])
Let B = x11 bTb B1
be a partial strictly copositive n n matrixhaving all entries except the 1,1-entry specified. Let be avector norm. Complete B to a strictly copositive matrix bychoosing a value for x
11as follows:
1. = miny0,y=1 bTy.
2. = miny0,||y||=1 yTB1y.
3. x11 > 2
.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
62/69
Corollary
Every partial strictly copositive matrix can be completed to astrictly copositive matrix.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
63/69
Example: The partial matrix
B =
x11 5 1 x14 x15 x165 1 2 x24 x25 1
1 2 5 1 1 1x14 x24 1 1 x45 1x15 x25
1 x45 x55 x56
x16 1 1 1 x56 3
is a partial strictly copositive matrix.
Select index 5. The only principal submatrix completed by a choice
of b55 is B[{3, 5}].Any value that makes 5x55 > (1)2 will work.Choose x55 = 1.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
64/69
Example: The partial matrix
B =
x11 5 1 x14 x15 x165 1 2 x24 x25 1
1 2 5 1 1 1x14 x24 1 1 x45 1x15 x25
1 x45 x55 x56
x16 1 1 1 x56 3
is a partial strictly copositive matrix.Select index 1. The only principal submatrices completed by achoice of b11 are principal submatrices of
B[{1, 2, 3}] = x11 5 15 1 2
1 2 5
.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
65/69
B[{1, 2, 3}] = x11
5 1
5 1 21 2 5
Using 1:
1. = min||y||1=1 bTy = 5.
2. = min||y||1=1 yTB[{2, 3}]y = 110 .
3. Choose x11 >2
; choose b11 = 256.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
6
-
7/31/2019 Introduction to Combinatorial Matrix Theory
66/69
b11 = 256, b55 = 1.
256 5 1 x14 x15 x16
5 1
2 x24 x25 1
1 2 5 1 1 1x14 x24 1 1 x45 1x15 x25 1 x45 1 x56x16 1 1 1 x56 3
Set xij = xji = biibjj.
256 5 1 16 16 16 35 1 2 1 1 1
1 2 5 1 1 1
16 1 1 1 1 116 1 1 1 1 316
3 1 1 1 3 3
is a strictly copositive matrix.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
DefinitionsGraph Theoretic TechniquesThe Positive (Semi)Definite Matrix Completion ProblemsThe (Strictly) Copositive Matrix Completion Problems
-
7/31/2019 Introduction to Combinatorial Matrix Theory
67/69
It is not true that every partial copositive matrix can be completedto a copositive matrix.
Example: B =
x11 11 0
is a partial copositive matrix that
cannot be completed to a copositive matrix.
Choose a value for x11 (> 0).If x11 > 0, then for the vector x = [1, x11]
T, xTBx = x11.
Combinatorial Matrix TheoryMinimum Rank ProblemsMatrix Completion Problems
References
-
7/31/2019 Introduction to Combinatorial Matrix Theory
68/69
[BFH04] F. Barioli, S. Fallat, and L. Hogben. Computation ofminimal rank and path cover number for graphs. Linear
Algebra and Its Applications, 392: 289303, 2004.
[BFH05] F. Barioli, S. Fallat, and L. Hogben. A variant on the graphparameters of Colin de Verdiere: Implications to theminimum rank of graphs. Electronic Journal of Linear
Algebra, 13: 387404, 2005.
[DHHHW06] L.M. DeAlba, T.L. Hardy, I.R. Hentzel, L. Hogben, A.Wangsness. Minimum Rank and Maximum EigenvalueMultiplicity of Symmetric Tree Sign Patterns. To appear inLinear Algebra and Its Applications.
[GJSW84] R. Grone, C.R. Johnson, E.M. Sa, and H. Wolkowicz.Positive definite completions of partial Hermitian matrices.Linear Algebra and Its Applications, 58:109124, 1984.
Combinatorial Matrix TheoryMinimum Rank Problems
Matrix Completion ProblemsReferences
-
7/31/2019 Introduction to Combinatorial Matrix Theory
69/69
[H06] L. Hogben. Completions of partial strictly copositivematrices omitting some diagonal entries. To appear in
Linear Algebra and Its Applications.
[HvdH06] L. Hogben and H. van der Holst. Forbidden minors for theclass of graphs G with (G) 2. To appear in LinearAlgebra and Its Applications. PDF provided.
[HJR05] L. Hogben, C. R. Johnson and R. Reams. The copositivematrix completion problem. Linear Algebra and ItsApplications, 408:207-211, 2005.
[JLD99] C. R. Johnson and A. Leal Duarte. The maximum
multiplicity of an eigenvalue in a matrix whose graph is atree. Linear and Multilinear Algebra 46: 139144, 1999.