entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 ›...

135
Entropy functions and polymatroids Matroids and Shannon entropy Convolutions and expansions Information inequalities Entropy functions, information inequalities, and polymatroids Frantiˇ sek Mat´ s Institute of Information Theory and Automation Academy of Sciences of the Czech Republic E-mail: [email protected] Applications of Matroid Theory and Combinatorial Optimization Banff, August 6, 2009

Upload: others

Post on 25-Jun-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Entropy functions, information inequalities,and polymatroids

Frantisek Matus

Institute of Information Theory and AutomationAcademy of Sciences of the Czech Republic

E-mail: [email protected]

Applications of Matroid Theory and Combinatorial Optimization

Banff, August 6, 2009

Page 2: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

ξ = (ξi )i∈N ... a random vector indexed by a finite set N

ξI = (ξi )i∈I ... a subvector of ξ, I ⊆ N

The entropy function hξ of ξ maps each subset I of Nto the Shannon entropy of ξI .

(hξ(I ))I⊆N ... an entropic point of Euclidean space RP(N),provided ξ takes a finite number of values

HentN ... the set of entropic points

Page 3: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

ξ = (ξi )i∈N ... a random vector indexed by a finite set N

ξI = (ξi )i∈I ... a subvector of ξ, I ⊆ N

The entropy function hξ of ξ maps each subset I of Nto the Shannon entropy of ξI .

(hξ(I ))I⊆N ... an entropic point of Euclidean space RP(N),provided ξ takes a finite number of values

HentN ... the set of entropic points

Page 4: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

ξ = (ξi )i∈N ... a random vector indexed by a finite set N

ξI = (ξi )i∈I ... a subvector of ξ, I ⊆ N

The entropy function hξ of ξ maps each subset I of Nto the Shannon entropy of ξI .

(hξ(I ))I⊆N ... an entropic point of Euclidean space RP(N),provided ξ takes a finite number of values

HentN ... the set of entropic points

Page 5: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

ξ = (ξi )i∈N ... a random vector indexed by a finite set N

ξI = (ξi )i∈I ... a subvector of ξ, I ⊆ N

The entropy function hξ of ξ maps each subset I of Nto the Shannon entropy of ξI .

(hξ(I ))I⊆N ... an entropic point of Euclidean space RP(N),provided ξ takes a finite number of values

HentN ... the set of entropic points

Page 6: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

ξ = (ξi )i∈N ... a random vector indexed by a finite set N

ξI = (ξi )i∈I ... a subvector of ξ, I ⊆ N

The entropy function hξ of ξ maps each subset I of Nto the Shannon entropy of ξI .

(hξ(I ))I⊆N ... an entropic point of Euclidean space RP(N),provided ξ takes a finite number of values

HentN ... the set of entropic points

Page 7: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

HentN ... a sophisticated set, unknown if |N| > 3

... a supporting hyperplane may intersects HentN in the set

-

6

a

b

. . .@

@@

@@

@

@@

@@

@@@

@@

@@ @@ @@

a + b > lnde a e (FM, Jan 2006, Trans. IT IE3)

Page 8: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

HentN ... a sophisticated set, unknown if |N| > 3

... a supporting hyperplane may intersects HentN in the set

-

6

a

b

. . .@

@@

@@

@

@@

@@

@@@

@@

@@ @@ @@

a + b > lnde a e (FM, Jan 2006, Trans. IT IE3)

Page 9: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

HentN ... a sophisticated set, unknown if |N| > 3

... a supporting hyperplane may intersects HentN in the set

-

6

a

b

. . .@

@@

@@

@

@@

@@

@@@

@@

@@ @@ @@

a + b > lnde a e (FM, Jan 2006, Trans. IT IE3)

Page 10: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

HentN ... a sophisticated set, unknown if |N| > 3

... a supporting hyperplane may intersects HentN in the set

-

6

a

b

. . .@

@@

@@

@

@@

@@

@@@

@@

@@ @@ @@

a + b > lnde a e (FM, Jan 2006, Trans. IT IE3)

Page 11: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

cl(HentN ) ... the closure of Hent

N

consists of almost entropic points

cl(HentN ) is a convex cone (Zhang & Yeung, Nov 1997, Trans. IT IE3)

... not depending on the base of logarithms in −P

x p(x) ln p(x)

ri(cl(HentN )) ⊆ Hent

N (FM, Jan 07, Trans. IT IE3)

... HentN and cl(Hent

N ) differ only on the boundary of the cone

Page 12: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

cl(HentN ) ... the closure of Hent

N

consists of almost entropic points

cl(HentN ) is a convex cone (Zhang & Yeung, Nov 1997, Trans. IT IE3)

... not depending on the base of logarithms in −P

x p(x) ln p(x)

ri(cl(HentN )) ⊆ Hent

N (FM, Jan 07, Trans. IT IE3)

... HentN and cl(Hent

N ) differ only on the boundary of the cone

Page 13: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

cl(HentN ) ... the closure of Hent

N

consists of almost entropic points

cl(HentN ) is a convex cone (Zhang & Yeung, Nov 1997, Trans. IT IE3)

... not depending on the base of logarithms in −P

x p(x) ln p(x)

ri(cl(HentN )) ⊆ Hent

N (FM, Jan 07, Trans. IT IE3)

... HentN and cl(Hent

N ) differ only on the boundary of the cone

Page 14: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

cl(HentN ) ... the closure of Hent

N

consists of almost entropic points

cl(HentN ) is a convex cone (Zhang & Yeung, Nov 1997, Trans. IT IE3)

... not depending on the base of logarithms in −P

x p(x) ln p(x)

ri(cl(HentN )) ⊆ Hent

N (FM, Jan 07, Trans. IT IE3)

... HentN and cl(Hent

N ) differ only on the boundary of the cone

Page 15: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

cl(HentN ) ... the closure of Hent

N

consists of almost entropic points

cl(HentN ) is a convex cone (Zhang & Yeung, Nov 1997, Trans. IT IE3)

... not depending on the base of logarithms in −P

x p(x) ln p(x)

ri(cl(HentN )) ⊆ Hent

N (FM, Jan 07, Trans. IT IE3)

... HentN and cl(Hent

N ) differ only on the boundary of the cone

Page 16: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

cl(HentN ) ... the closure of Hent

N

consists of almost entropic points

cl(HentN ) is a convex cone (Zhang & Yeung, Nov 1997, Trans. IT IE3)

... not depending on the base of logarithms in −P

x p(x) ln p(x)

ri(cl(HentN )) ⊆ Hent

N (FM, Jan 07, Trans. IT IE3)

... HentN and cl(Hent

N ) differ only on the boundary of the cone

Page 17: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 18: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 19: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0

g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 20: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ J

g(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 21: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 22: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 23: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 24: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 25: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 26: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 27: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionBasic observationsLimits of entropic functionsPolymatroids

(N, g) ... polymatroid, sits on the ground set N and has

a rank function g : P(N) → [0,+∞) satisfying

g(∅) = 0g(I ) 6 g(J) for I ⊆ Jg(I ) + g(J) > g(I ∪ J) + g(I ∩ J) for I , J ⊆ N.

It is a matroid if g(I ) ∈ Z and g(I ) 6 |I | for I ⊆ N.

HN ⊆ RP(N) ... the rank functions sitting on N form

a polyhedral cone (finite intersection of closed halfspaces).

(N, hξ) is a polymatroid if hξ ∈ RP(N) (Fujishige, 1978, Inf. Contr.)

... a polymatroid (N, g) is (a)ent if g is an (a)ent point

HN ⊇ cl(HentN ) ⊇ Hent

N ⊇ ri(cl(HentN ))

Page 28: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

For J ⊆ N and the matroid (N, rJ) with rJ(I ) =

1, I ∩ J 6= ∅0, I ∩ J = ∅

t rJ is entropic for all t > 0.

If N partitions into I ∪ J and r(N) = r(I ) + r(J) in a matroid (N, r)

then t r ∈ HentN iff t r |P(I ) ∈ Hent

I and t r |P(J) ∈ HentJ .

If a matroid (N, r) is connected

then r belongs to en extreme ray of HN .

Page 29: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

For J ⊆ N and the matroid (N, rJ) with rJ(I ) =

1, I ∩ J 6= ∅0, I ∩ J = ∅

t rJ is entropic for all t > 0.

If N partitions into I ∪ J and r(N) = r(I ) + r(J) in a matroid (N, r)

then t r ∈ HentN iff t r |P(I ) ∈ Hent

I and t r |P(J) ∈ HentJ .

If a matroid (N, r) is connected

then r belongs to en extreme ray of HN .

Page 30: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

For J ⊆ N and the matroid (N, rJ) with rJ(I ) =

1, I ∩ J 6= ∅0, I ∩ J = ∅

t rJ is entropic for all t > 0.

If N partitions into I ∪ J and r(N) = r(I ) + r(J) in a matroid (N, r)

then t r ∈ HentN iff t r |P(I ) ∈ Hent

I and t r |P(J) ∈ HentJ .

If a matroid (N, r) is connected

then r belongs to en extreme ray of HN .

Page 31: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

For J ⊆ N and the matroid (N, rJ) with rJ(I ) =

1, I ∩ J 6= ∅0, I ∩ J = ∅

t rJ is entropic for all t > 0.

If N partitions into I ∪ J and r(N) = r(I ) + r(J) in a matroid (N, r)

then t r ∈ HentN iff t r |P(I ) ∈ Hent

I and t r |P(J) ∈ HentJ .

If a matroid (N, r) is connected

then r belongs to en extreme ray of HN .

Page 32: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

For J ⊆ N and the matroid (N, rJ) with rJ(I ) =

1, I ∩ J 6= ∅0, I ∩ J = ∅

t rJ is entropic for all t > 0.

If N partitions into I ∪ J and r(N) = r(I ) + r(J) in a matroid (N, r)

then t r ∈ HentN iff t r |P(I ) ∈ Hent

I and t r |P(J) ∈ HentJ .

If a matroid (N, r) is connected

then r belongs to en extreme ray of HN .

Page 33: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

For J ⊆ N and the matroid (N, rJ) with rJ(I ) =

1, I ∩ J 6= ∅0, I ∩ J = ∅

t rJ is entropic for all t > 0.

If N partitions into I ∪ J and r(N) = r(I ) + r(J) in a matroid (N, r)

then t r ∈ HentN iff t r |P(I ) ∈ Hent

I and t r |P(J) ∈ HentJ .

If a matroid (N, r) is connected

then r belongs to en extreme ray of HN .

Page 34: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

From now on, a matroid (N, r) is connected and r(N) > 2.

For (N, r), t > 0 and random vector ξ = (ξi )i∈N ,t r ∈ Hent

N implies t ∈ ln d : d = 1, 2, . . ., and(ln d) r = hξ implies that ξI takes d r(I ) values

with the same probability, I ⊆ N. (FM, 1994, Int. J. Gen. Syst.)

A matroid (N, r) is partition (p-) representable of the degree d > 2if there exist partitions πi , i ∈ N, of a finite set Ωwith d r(N) elements such that for all I ⊆ N the meetof πi , i ∈ I , has d r(I ) blocks of the same cardinality.

... secret sharing matroids, almost affine codes, ...

Page 35: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

From now on, a matroid (N, r) is connected and r(N) > 2.

For (N, r), t > 0 and random vector ξ = (ξi )i∈N ,t r ∈ Hent

N implies t ∈ ln d : d = 1, 2, . . ., and(ln d) r = hξ implies that ξI takes d r(I ) values

with the same probability, I ⊆ N. (FM, 1994, Int. J. Gen. Syst.)

A matroid (N, r) is partition (p-) representable of the degree d > 2if there exist partitions πi , i ∈ N, of a finite set Ωwith d r(N) elements such that for all I ⊆ N the meetof πi , i ∈ I , has d r(I ) blocks of the same cardinality.

... secret sharing matroids, almost affine codes, ...

Page 36: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

From now on, a matroid (N, r) is connected and r(N) > 2.

For (N, r), t > 0 and random vector ξ = (ξi )i∈N ,t r ∈ Hent

N implies t ∈ ln d : d = 1, 2, . . ., and(ln d) r = hξ implies that ξI takes d r(I ) values

with the same probability, I ⊆ N. (FM, 1994, Int. J. Gen. Syst.)

A matroid (N, r) is partition (p-) representable of the degree d > 2if there exist partitions πi , i ∈ N, of a finite set Ωwith d r(N) elements such that for all I ⊆ N the meetof πi , i ∈ I , has d r(I ) blocks of the same cardinality.

... secret sharing matroids, almost affine codes, ...

Page 37: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

From now on, a matroid (N, r) is connected and r(N) > 2.

For (N, r), t > 0 and random vector ξ = (ξi )i∈N ,t r ∈ Hent

N implies t ∈ ln d : d = 1, 2, . . ., and(ln d) r = hξ implies that ξI takes d r(I ) values

with the same probability, I ⊆ N. (FM, 1994, Int. J. Gen. Syst.)

A matroid (N, r) is partition (p-) representable of the degree d > 2if there exist partitions πi , i ∈ N, of a finite set Ωwith d r(N) elements such that for all I ⊆ N the meetof πi , i ∈ I , has d r(I ) blocks of the same cardinality.

... secret sharing matroids, almost affine codes, ...

Page 38: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

|Ω| = 32 = d r(N), |N| = 4, the four partitions

q q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q q

........................

....................

.....................

...................

................

.............

.........

............................................................................................................................... .

.......................

....................

.....................

...................

................

.................................. ......... ........... ............. ................ ................... ......................

.........................

@@

@

........................

....................

.....................

................... ................ ............. ......... ....... ......................................

................

...................

......................

.........................

...........................

.......................

....................

.................

.............

......................... ....... ......... ............ ............... .................. .................... .................

................

................

.................

@@

@

represent U2,4, the degree is d = 3.

They correspond to two orthogonal Latin squares.

A p-representation of U2,4 of the degree d = 10 exists.

Page 39: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

|Ω| = 32 = d r(N), |N| = 4, the four partitions

q q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q q

........................

....................

.....................

...................

................

.............

.........

............................................................................................................................... .

.......................

....................

.....................

...................

................

.................................. ......... ........... ............. ................ ................... ......................

.........................

@@

@

........................

....................

.....................

................... ................ ............. ......... ....... ......................................

................

...................

......................

.........................

...........................

.......................

....................

.................

.............

......................... ....... ......... ............ ............... .................. .................... .................

................

................

.................

@@

@

represent U2,4, the degree is d = 3.

They correspond to two orthogonal Latin squares.

A p-representation of U2,4 of the degree d = 10 exists.

Page 40: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

|Ω| = 32 = d r(N), |N| = 4, the four partitions

q q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q q

........................

....................

.....................

...................

................

.............

.........

............................................................................................................................... .

.......................

....................

.....................

...................

................

.................................. ......... ........... ............. ................ ................... ......................

.........................

@@

@

........................

....................

.....................

................... ................ ............. ......... ....... ......................................

................

...................

......................

.........................

...........................

.......................

....................

.................

.............

......................... ....... ......... ............ ............... .................. .................... .................

................

................

.................

@@

@

represent U2,4, the degree is d = 3.

They correspond to two orthogonal Latin squares.

A p-representation of U2,4 of the degree d = 10 exists.

Page 41: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

|Ω| = 32 = d r(N), |N| = 4, the four partitions

q q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q q

........................

....................

.....................

...................

................

.............

.........

............................................................................................................................... .

.......................

....................

.....................

...................

................

.................................. ......... ........... ............. ................ ................... ......................

.........................

@@

@

........................

....................

.....................

................... ................ ............. ......... ....... ......................................

................

...................

......................

.........................

...........................

.......................

....................

.................

.............

......................... ....... ......... ............ ............... .................. .................... .................

................

................

.................

@@

@

represent U2,4, the degree is d = 3.

They correspond to two orthogonal Latin squares.

A p-representation of U2,4 of the degree d = 10 exists.

Page 42: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

|Ω| = 32 = d r(N), |N| = 4, the four partitions

q q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q qq q q q

........................

....................

.....................

...................

................

.............

.........

............................................................................................................................... .

.......................

....................

.....................

...................

................

.................................. ......... ........... ............. ................ ................... ......................

.........................

@@

@

........................

....................

.....................

................... ................ ............. ......... ....... ......................................

................

...................

......................

.........................

...........................

.......................

....................

.................

.............

......................... ....... ......... ............ ............... .................. .................... .................

................

................

.................

@@

@

represent U2,4, the degree is d = 3.

They correspond to two orthogonal Latin squares.

A p-representation of U2,4 of the degree d = 10 exists.

Page 43: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 44: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 45: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 46: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 47: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 48: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 49: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 50: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

If (N, r) is linear over the field with d elements

then it is p-representable of the degree d .

The converse holds when d = 2 or d = 3.

The non-Pappus matroid is p-representable of the degree 9.

(is multilinear, Simonis & Ashikhmin, 1998, Des. Codes and Cryptography)

P-representable of some degree implies almost entropic.

Stick the Fano and non-Fano matroids along a point:

not p-representable of any degree, but almost entropic.

Page 51: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

Open problems

? p-representable of some degree but not multilinear ?

? the class of almost entropic matroids, excluded minors ?

? the critical problem for these classes of matroids ?

Page 52: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

Open problems

? p-representable of some degree but not multilinear ?

? the class of almost entropic matroids, excluded minors ?

? the critical problem for these classes of matroids ?

Page 53: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

Open problems

? p-representable of some degree but not multilinear ?

? the class of almost entropic matroids, excluded minors ?

? the critical problem for these classes of matroids ?

Page 54: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

Multiples of matroidal rank functionsPartition representable matroidsClasses of matroidsOpen problems

Open problems

? p-representable of some degree but not multilinear ?

? the class of almost entropic matroids, excluded minors ?

? the critical problem for these classes of matroids ?

Page 55: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Convolution (N, g ∗ f ) of polymatroids (N, g) and (N, f ), given by

g ∗ f (I ) = minJ⊆I [ g(J) + f (I \ J) ] for I ⊆ N,

is a polymatroid whenever f is modular (Lovasz, 1982)

[ f (I ) =∑

i∈I f (i) for I ⊆ N ].

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

g aent and f modular implies g ∗ f aent. (FM07)

[ cl(HentN ) is closed to the convolutions with modular polymatroids. ]

Page 56: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Convolution (N, g ∗ f ) of polymatroids (N, g) and (N, f ), given by

g ∗ f (I ) = minJ⊆I [ g(J) + f (I \ J) ] for I ⊆ N,

is a polymatroid whenever f is modular (Lovasz, 1982)

[ f (I ) =∑

i∈I f (i) for I ⊆ N ].

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

g aent and f modular implies g ∗ f aent. (FM07)

[ cl(HentN ) is closed to the convolutions with modular polymatroids. ]

Page 57: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Convolution (N, g ∗ f ) of polymatroids (N, g) and (N, f ), given by

g ∗ f (I ) = minJ⊆I [ g(J) + f (I \ J) ] for I ⊆ N,

is a polymatroid whenever f is modular (Lovasz, 1982)

[ f (I ) =∑

i∈I f (i) for I ⊆ N ].

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

g aent and f modular implies g ∗ f aent. (FM07)

[ cl(HentN ) is closed to the convolutions with modular polymatroids. ]

Page 58: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Convolution (N, g ∗ f ) of polymatroids (N, g) and (N, f ), given by

g ∗ f (I ) = minJ⊆I [ g(J) + f (I \ J) ] for I ⊆ N,

is a polymatroid whenever f is modular (Lovasz, 1982)

[ f (I ) =∑

i∈I f (i) for I ⊆ N ].

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

g aent and f modular implies g ∗ f aent. (FM07)

[ cl(HentN ) is closed to the convolutions with modular polymatroids. ]

Page 59: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Convolution (N, g ∗ f ) of polymatroids (N, g) and (N, f ), given by

g ∗ f (I ) = minJ⊆I [ g(J) + f (I \ J) ] for I ⊆ N,

is a polymatroid whenever f is modular (Lovasz, 1982)

[ f (I ) =∑

i∈I f (i) for I ⊆ N ].

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

g aent and f modular implies g ∗ f aent. (FM07)

[ cl(HentN ) is closed to the convolutions with modular polymatroids. ]

Page 60: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Convolution (N, g ∗ f ) of polymatroids (N, g) and (N, f ), given by

g ∗ f (I ) = minJ⊆I [ g(J) + f (I \ J) ] for I ⊆ N,

is a polymatroid whenever f is modular (Lovasz, 1982)

[ f (I ) =∑

i∈I f (i) for I ⊆ N ].

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

g aent and f modular implies g ∗ f aent. (FM07)

[ cl(HentN ) is closed to the convolutions with modular polymatroids. ]

Page 61: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Convolution (N, g ∗ f ) of polymatroids (N, g) and (N, f ), given by

g ∗ f (I ) = minJ⊆I [ g(J) + f (I \ J) ] for I ⊆ N,

is a polymatroid whenever f is modular (Lovasz, 1982)

[ f (I ) =∑

i∈I f (i) for I ⊆ N ].

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

g aent and f modular implies g ∗ f aent. (FM07)

[ cl(HentN ) is closed to the convolutions with modular polymatroids. ]

Page 62: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Factor of a polymatroid (N, g)

by a set M of disjoint blocks covering N

is the polymatroid (M, f ) given by f (L) = g(⋃

L) for L ⊆ M.~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

Every integer polymatroid is a factor of a matroid.

Page 63: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Factor of a polymatroid (N, g)

by a set M of disjoint blocks covering N

is the polymatroid (M, f ) given by f (L) = g(⋃

L) for L ⊆ M.~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

Every integer polymatroid is a factor of a matroid.

Page 64: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Factor of a polymatroid (N, g)

by a set M of disjoint blocks covering N

is the polymatroid (M, f ) given by f (L) = g(⋃

L) for L ⊆ M.

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

Every integer polymatroid is a factor of a matroid.

Page 65: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Factor of a polymatroid (N, g)

by a set M of disjoint blocks covering N

is the polymatroid (M, f ) given by f (L) = g(⋃

L) for L ⊆ M.~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

Every integer polymatroid is a factor of a matroid.

Page 66: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Factor of a polymatroid (N, g)

by a set M of disjoint blocks covering N

is the polymatroid (M, f ) given by f (L) = g(⋃

L) for L ⊆ M.~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

2

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

2 2 2 2

2

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

2 2 2 2

2

Every integer polymatroid is a factor of a matroid.

Page 67: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Expansion of an integer polymatroid (M, f )

is a matroid (N, r) that factors back to (M, f )

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

Free expansion can be constructed in two steps

1. make f (m) parallel copies of each m ∈ M

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

2. convolve with the free matroid I 7→ |I |

Page 68: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Expansion of an integer polymatroid (M, f )

is a matroid (N, r) that factors back to (M, f )

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

Free expansion can be constructed in two steps

1. make f (m) parallel copies of each m ∈ M

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

2. convolve with the free matroid I 7→ |I |

Page 69: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Expansion of an integer polymatroid (M, f )

is a matroid (N, r) that factors back to (M, f )

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

Free expansion can be constructed in two steps

1. make f (m) parallel copies of each m ∈ M

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

2. convolve with the free matroid I 7→ |I |

Page 70: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Expansion of an integer polymatroid (M, f )

is a matroid (N, r) that factors back to (M, f )

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

Free expansion can be constructed in two steps

1. make f (m) parallel copies of each m ∈ M

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

2. convolve with the free matroid I 7→ |I |

Page 71: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Expansion of an integer polymatroid (M, f )

is a matroid (N, r) that factors back to (M, f )

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

Free expansion can be constructed in two steps

1. make f (m) parallel copies of each m ∈ M

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

2. convolve with the free matroid I 7→ |I |

Page 72: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Expansion of an integer polymatroid (M, f )

is a matroid (N, r) that factors back to (M, f )

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

Free expansion can be constructed in two steps

1. make f (m) parallel copies of each m ∈ M

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

2. convolve with the free matroid I 7→ |I |

Page 73: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Expansion of an integer polymatroid (M, f )

is a matroid (N, r) that factors back to (M, f )

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

Free expansion can be constructed in two steps

1. make f (m) parallel copies of each m ∈ M

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

~

~

~

~

~

~

~

~

0

1

2

3

1

2

1

2

=

~

~

~

~

~

~

~

~

0

1

2

2

1

2

1

2

~

~

~

~

~

~

~

~

0

1

2

2

1

2

2

2

factors to

~

~~

~

0

2

2

2

~

~~

~

0

2

3

2 expands to

m

m m m m

m m m m m m

m m m m

m

0

1 1 1 1

2 2 2 2 2 2

3 3 3 3

3

~

~~

~

0

2

3

2 parallelizes to

m

m m m m

m m m m m m

m m m m

m

0

2 2 2 2

2 2 2 2 2 2

3 3 3 3

3

2. convolve with the free matroid I 7→ |I |

Page 74: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Every aent integer polymatroid (M, f )is a factor of an aent matroid. (FM, Jan 2007, Trans. IT IE3)

The cone cl(HentN ) can be described in terms of aent matroids:

scalings of factors of aent matroids are dense cl(HentN ).

Page 75: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Every aent integer polymatroid (M, f )is a factor of an aent matroid. (FM, Jan 2007, Trans. IT IE3)

The cone cl(HentN ) can be described in terms of aent matroids:

scalings of factors of aent matroids are dense cl(HentN ).

Page 76: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

ConvolutionsFactors and expansions

Every aent integer polymatroid (M, f )is a factor of an aent matroid. (FM, Jan 2007, Trans. IT IE3)

The cone cl(HentN ) can be described in terms of aent matroids:

scalings of factors of aent matroids are dense cl(HentN ).

Page 77: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A point c = (cI )I⊆N of RP(N) generates

a (linear unconditional) information inequality if∑I⊆N cI · g(I ) 6 0 for the entropic points g = (g(I ))I⊆N .

... 〈c , g〉 6 0 for g ∈ HentN , thus c belongs to the polar of Hent

N .

pol(HN) ⊆ pol(cl(HentN )) = pol(Hent

N )

The inequality is of Shannon type if it is satisfied evenby the rank functions of all polymatroids, thus if c ∈ pol(HN).

If |N| 6 3 then HN = cl(HentN )

whence all info inequalities are of Shannon type.

Page 78: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A point c = (cI )I⊆N of RP(N) generates

a (linear unconditional) information inequality if∑I⊆N cI · g(I ) 6 0 for the entropic points g = (g(I ))I⊆N .

... 〈c , g〉 6 0 for g ∈ HentN , thus c belongs to the polar of Hent

N .

pol(HN) ⊆ pol(cl(HentN )) = pol(Hent

N )

The inequality is of Shannon type if it is satisfied evenby the rank functions of all polymatroids, thus if c ∈ pol(HN).

If |N| 6 3 then HN = cl(HentN )

whence all info inequalities are of Shannon type.

Page 79: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A point c = (cI )I⊆N of RP(N) generates

a (linear unconditional) information inequality if∑I⊆N cI · g(I ) 6 0 for the entropic points g = (g(I ))I⊆N .

... 〈c , g〉 6 0 for g ∈ HentN , thus c belongs to the polar of Hent

N .

pol(HN) ⊆ pol(cl(HentN )) = pol(Hent

N )

The inequality is of Shannon type if it is satisfied evenby the rank functions of all polymatroids, thus if c ∈ pol(HN).

If |N| 6 3 then HN = cl(HentN )

whence all info inequalities are of Shannon type.

Page 80: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A point c = (cI )I⊆N of RP(N) generates

a (linear unconditional) information inequality if∑I⊆N cI · g(I ) 6 0 for the entropic points g = (g(I ))I⊆N .

... 〈c , g〉 6 0 for g ∈ HentN , thus c belongs to the polar of Hent

N .

pol(HN) ⊆ pol(cl(HentN )) = pol(Hent

N )

The inequality is of Shannon type if it is satisfied evenby the rank functions of all polymatroids, thus if c ∈ pol(HN).

If |N| 6 3 then HN = cl(HentN )

whence all info inequalities are of Shannon type.

Page 81: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A point c = (cI )I⊆N of RP(N) generates

a (linear unconditional) information inequality if∑I⊆N cI · g(I ) 6 0 for the entropic points g = (g(I ))I⊆N .

... 〈c , g〉 6 0 for g ∈ HentN , thus c belongs to the polar of Hent

N .

pol(HN) ⊆ pol(cl(HentN )) = pol(Hent

N )

The inequality is of Shannon type if it is satisfied evenby the rank functions of all polymatroids, thus if c ∈ pol(HN).

If |N| 6 3 then HN = cl(HentN )

whence all info inequalities are of Shannon type.

Page 82: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For the entropy functions g = hξ over N = i , j , k, l

3[g(ik) + g(il) + g(kl)

]+ g(jk) + g(jl)

> g(i) + 2[g(k) + g(l)

]+ g(ij) + 4g(ikl) + g(jkl)

(Zhang & Yeung 1998, Trans. IT IE3)

i , j , k, l considered for singletons, the signs for union omitted

ZY inequality is violated by the polymatroid

k

k k k k

k k k k k k

k k k k

k

0

2 2 2 2

4 3 3 3 3 3

4 4 4 4

4

1

hence it is not of Shannon type and cl(HentN ) $ HN .

Page 83: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For the entropy functions g = hξ over N = i , j , k, l

3[g(ik) + g(il) + g(kl)

]+ g(jk) + g(jl)

> g(i) + 2[g(k) + g(l)

]+ g(ij) + 4g(ikl) + g(jkl)

(Zhang & Yeung 1998, Trans. IT IE3)

i , j , k, l considered for singletons, the signs for union omitted

ZY inequality is violated by the polymatroid

k

k k k k

k k k k k k

k k k k

k

0

2 2 2 2

4 3 3 3 3 3

4 4 4 4

4

1

hence it is not of Shannon type and cl(HentN ) $ HN .

Page 84: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For the entropy functions g = hξ over N = i , j , k, l

3[g(ik) + g(il) + g(kl)

]+ g(jk) + g(jl)

> g(i) + 2[g(k) + g(l)

]+ g(ij) + 4g(ikl) + g(jkl)

(Zhang & Yeung 1998, Trans. IT IE3)

i , j , k, l considered for singletons, the signs for union omitted

ZY inequality is violated by the polymatroid

k

k k k k

k k k k k k

k k k k

k

0

2 2 2 2

4 3 3 3 3 3

4 4 4 4

4

1

hence it is not of Shannon type and cl(HentN ) $ HN .

Page 85: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For the entropy functions g = hξ over N = i , j , k, l

3[g(ik) + g(il) + g(kl)

]+ g(jk) + g(jl)

> g(i) + 2[g(k) + g(l)

]+ g(ij) + 4g(ikl) + g(jkl)

(Zhang & Yeung 1998, Trans. IT IE3)

i , j , k, l considered for singletons, the signs for union omitted

ZY inequality is violated by the polymatroid

k

k k k k

k k k k k k

k k k k

k

0

2 2 2 2

4 3 3 3 3 3

4 4 4 4

4

1

hence it is not of Shannon type and cl(HentN ) $ HN .

Page 86: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(M, f)

(N, g)...............

...............

..............

..............

................. ................... ...................... ........................ ........................... ............................. ............................. ........................... ..............................................

....................................

............................

...............

...............

...............

...............

............................

....................................

...............................................................................................................................................................................................................................

.............................

................

...............

.........

......

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

If N ⊆ M and g(I ) = f (I ) for I ⊆ N theng is the restriction of f and f is an extension of g .

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

Polymatroids g and h are adhesive ifa polymatroid (M ∪ N, f ) extends both and

f (M) + f (N) = f (M ∪ N) + f (M ∩ N)

(f ... an adhesive extension of g and h).

Page 87: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(M, f)

(N, g)...............

...............

..............

..............

................. ................... ...................... ........................ ........................... ............................. ............................. ........................... ..............................................

....................................

............................

...............

...............

...............

...............

............................

....................................

...............................................................................................................................................................................................................................

.............................

................

...............

.........

......

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

If N ⊆ M and g(I ) = f (I ) for I ⊆ N theng is the restriction of f and f is an extension of g .

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

Polymatroids g and h are adhesive ifa polymatroid (M ∪ N, f ) extends both and

f (M) + f (N) = f (M ∪ N) + f (M ∩ N)

(f ... an adhesive extension of g and h).

Page 88: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(M, f)

(N, g)...............

...............

..............

..............

................. ................... ...................... ........................ ........................... ............................. ............................. ........................... ..............................................

....................................

............................

...............

...............

...............

...............

............................

....................................

...............................................................................................................................................................................................................................

.............................

................

...............

.........

......

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

If N ⊆ M and g(I ) = f (I ) for I ⊆ N theng is the restriction of f and f is an extension of g .

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

Polymatroids g and h are adhesive ifa polymatroid (M ∪ N, f ) extends both and

f (M) + f (N) = f (M ∪ N) + f (M ∩ N)

(f ... an adhesive extension of g and h).

Page 89: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(M, f)

(N, g)...............

...............

..............

..............

................. ................... ...................... ........................ ........................... ............................. ............................. ........................... ..............................................

....................................

............................

...............

...............

...............

...............

............................

....................................

...............................................................................................................................................................................................................................

.............................

................

...............

.........

......

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

If N ⊆ M and g(I ) = f (I ) for I ⊆ N theng is the restriction of f and f is an extension of g .

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

Polymatroids g and h are adhesive ifa polymatroid (M ∪ N, f ) extends both and

f (M) + f (N) = f (M ∪ N) + f (M ∩ N)

(f ... an adhesive extension of g and h).

Page 90: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(M, f)

(N, g)...............

...............

..............

..............

................. ................... ...................... ........................ ........................... ............................. ............................. ........................... ..............................................

....................................

............................

...............

...............

...............

...............

............................

....................................

...............................................................................................................................................................................................................................

.............................

................

...............

.........

......

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

If N ⊆ M and g(I ) = f (I ) for I ⊆ N theng is the restriction of f and f is an extension of g .

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

1

Polymatroids g and h are adhesive ifa polymatroid (M ∪ N, f ) extends both and

f (M) + f (N) = f (M ∪ N) + f (M ∩ N)

(f ... an adhesive extension of g and h).

Page 91: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A polymatroid (N, g) is selfadhesive at I ⊆ Nif (N, g) and its copy (M, h) along I adhere.

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........ I = N ∩M

.

....................................

...................................

..................................

.................................

...............................................................

............................. ............................ ............................ ............................................................

................................

.................................

................................

..

............................

.......

........................

............· ·ª R

·.........................

....................

..................

...............

.............

........................... ........ ........... ............. ............... ................. ................. ............... ............. ...........

...................................

.............

...............

..................

....................

........................

-

1

A polymatroid is selfadhesiveif it is selfadhesive at each subset of its ground set.

Over the ground set N of cardinality four, a polymatroid isselfadhesive if and only if it satisfies all instances of ZY inequality.

(FM 2007, Discrete Math)

The entropy functions are selfadhesive.

Page 92: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A polymatroid (N, g) is selfadhesive at I ⊆ Nif (N, g) and its copy (M, h) along I adhere.

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........ I = N ∩M

.

....................................

...................................

..................................

.................................

...............................................................

............................. ............................ ............................ ............................................................

................................

.................................

................................

..

............................

.......

........................

............· ·ª R

·.........................

....................

..................

...............

.............

........................... ........ ........... ............. ............... ................. ................. ............... ............. ...........

...................................

.............

...............

..................

....................

........................

-

1

A polymatroid is selfadhesiveif it is selfadhesive at each subset of its ground set.

Over the ground set N of cardinality four, a polymatroid isselfadhesive if and only if it satisfies all instances of ZY inequality.

(FM 2007, Discrete Math)

The entropy functions are selfadhesive.

Page 93: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A polymatroid (N, g) is selfadhesive at I ⊆ Nif (N, g) and its copy (M, h) along I adhere.

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........ I = N ∩M

.

....................................

...................................

..................................

.................................

...............................................................

............................. ............................ ............................ ............................................................

................................

.................................

................................

..

............................

.......

........................

............· ·ª R

·.........................

....................

..................

...............

.............

........................... ........ ........... ............. ............... ................. ................. ............... ............. ...........

...................................

.............

...............

..................

....................

........................

-

1

A polymatroid is selfadhesiveif it is selfadhesive at each subset of its ground set.

Over the ground set N of cardinality four, a polymatroid isselfadhesive if and only if it satisfies all instances of ZY inequality.

(FM 2007, Discrete Math)

The entropy functions are selfadhesive.

Page 94: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A polymatroid (N, g) is selfadhesive at I ⊆ Nif (N, g) and its copy (M, h) along I adhere.

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........ I = N ∩M

.

....................................

...................................

..................................

.................................

...............................................................

............................. ............................ ............................ ............................................................

................................

.................................

................................

..

............................

.......

........................

............· ·ª R

·.........................

....................

..................

...............

.............

........................... ........ ........... ............. ............... ................. ................. ............... ............. ...........

...................................

.............

...............

..................

....................

........................

-

1

A polymatroid is selfadhesiveif it is selfadhesive at each subset of its ground set.

Over the ground set N of cardinality four, a polymatroid isselfadhesive if and only if it satisfies all instances of ZY inequality.

(FM 2007, Discrete Math)

The entropy functions are selfadhesive.

Page 95: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A polymatroid (N, g) is selfadhesive at I ⊆ Nif (N, g) and its copy (M, h) along I adhere.

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........ I = N ∩M

.

....................................

...................................

..................................

.................................

...............................................................

............................. ............................ ............................ ............................................................

................................

.................................

................................

..

............................

.......

........................

............· ·ª R

·.........................

....................

..................

...............

.............

........................... ........ ........... ............. ............... ................. ................. ............... ............. ...........

...................................

.............

...............

..................

....................

........................

-

1

A polymatroid is selfadhesiveif it is selfadhesive at each subset of its ground set.

Over the ground set N of cardinality four, a polymatroid isselfadhesive if and only if it satisfies all instances of ZY inequality.

(FM 2007, Discrete Math)

The entropy functions are selfadhesive.

Page 96: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A polymatroid (N, g) is selfadhesive at I ⊆ Nif (N, g) and its copy (M, h) along I adhere.

(M, h) (N, g)

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........

.........

........

................

................

................

................

...................

.....................

....................... ......................... ............................ .............................. ................................ ................................... ..................................... ..................................... ................................... ................................ ..........................................................

................................................

........................................

................

................

................

................

.................

.................

................

................

................

...................................

............................................

.....................................................

............................................................................................................................................................................................................................................................................................................................................................................

.................

...............

....

................

................

................

.........

........ I = N ∩M

.

....................................

...................................

..................................

.................................

...............................................................

............................. ............................ ............................ ............................................................

................................

.................................

................................

..

............................

.......

........................

............· ·ª R

·.........................

....................

..................

...............

.............

........................... ........ ........... ............. ............... ................. ................. ............... ............. ...........

...................................

.............

...............

..................

....................

........................

-

1

A polymatroid is selfadhesiveif it is selfadhesive at each subset of its ground set.

Over the ground set N of cardinality four, a polymatroid isselfadhesive if and only if it satisfies all instances of ZY inequality.

(FM 2007, Discrete Math)

The entropy functions are selfadhesive.

Page 97: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 98: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 99: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 100: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 101: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 102: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 103: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 104: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 105: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Iterate adhesive extensions and restrictions:

a sequence HarN (s), s > 0, starting at Har

N (0) = HN is defined by

g ∈ HarN (s + 1) iff g ∈ HN and

the restrictions of g to any I , J ⊆ N with I ∪ J = N

have an adhesive extension in HarN (s).

HarN (s) ⊇ Har

N (s + 1) ... polyhedral cones

H iaN =

⋂s>0 Har

N (s) ... the polymatroids with inner adhesivity.

cl(HentN ) ⊆ H ia

N

... because two restrictions of an entropic polymatroid havean adhesive extension that is entropic.

Page 106: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For N = 1 , 2 , 3 , 4 , 5 and g ∈ HarN (s), s > 1,

s[12 ,34 g + ∆34 |5 g + ∆45 |3 g

]+∆35 |4 g + s(s−1)

2

[∆24 |3 g + ∆34 |2 g

]> 0.

where ∆34 |5 g = g(35) + g(45)− g(5)− g(345)

and 12 ,34 g = ∆34 |1 g + ∆34 |2 g + ∆12 |∅ g −∆34 |∅ g

A proof is by induction on s, proving eventhree sequences of such inequalities simultaneously.

Hence, all the inequalities hold for the almost entropic points.

Page 107: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For N = 1 , 2 , 3 , 4 , 5 and g ∈ HarN (s), s > 1,

s[12 ,34 g + ∆34 |5 g + ∆45 |3 g

]+∆35 |4 g + s(s−1)

2

[∆24 |3 g + ∆34 |2 g

]> 0.

where ∆34 |5 g = g(35) + g(45)− g(5)− g(345)

and 12 ,34 g = ∆34 |1 g + ∆34 |2 g + ∆12 |∅ g −∆34 |∅ g

A proof is by induction on s, proving eventhree sequences of such inequalities simultaneously.

Hence, all the inequalities hold for the almost entropic points.

Page 108: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For N = 1 , 2 , 3 , 4 , 5 and g ∈ HarN (s), s > 1,

s[12 ,34 g + ∆34 |5 g + ∆45 |3 g

]+∆35 |4 g + s(s−1)

2

[∆24 |3 g + ∆34 |2 g

]> 0.

where ∆34 |5 g = g(35) + g(45)− g(5)− g(345)

and 12 ,34 g = ∆34 |1 g + ∆34 |2 g + ∆12 |∅ g −∆34 |∅ g

A proof is by induction on s, proving eventhree sequences of such inequalities simultaneously.

Hence, all the inequalities hold for the almost entropic points.

Page 109: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For N = 1 , 2 , 3 , 4 , 5 and g ∈ HarN (s), s > 1,

s[12 ,34 g + ∆34 |5 g + ∆45 |3 g

]+∆35 |4 g + s(s−1)

2

[∆24 |3 g + ∆34 |2 g

]> 0.

where ∆34 |5 g = g(35) + g(45)− g(5)− g(345)

and 12 ,34 g = ∆34 |1 g + ∆34 |2 g + ∆12 |∅ g −∆34 |∅ g

A proof is by induction on s, proving eventhree sequences of such inequalities simultaneously.

Hence, all the inequalities hold for the almost entropic points.

Page 110: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

For N = 1 , 2 , 3 , 4 , 5 and g ∈ HarN (s), s > 1,

s[12 ,34 g + ∆34 |5 g + ∆45 |3 g

]+∆35 |4 g + s(s−1)

2

[∆24 |3 g + ∆34 |2 g

]> 0.

where ∆34 |5 g = g(35) + g(45)− g(5)− g(345)

and 12 ,34 g = ∆34 |1 g + ∆34 |2 g + ∆12 |∅ g −∆34 |∅ g

A proof is by induction on s, proving eventhree sequences of such inequalities simultaneously.

Hence, all the inequalities hold for the almost entropic points.

Page 111: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(ξ5 = ξ1 )

For s > 1 and the entropic function f of (ξ1 , ξ2 , ξ3 , ξ4 )

s 12 ,34 f + ∆24 ,34 f + s(s+1)2

[∆23 ,34 f + ∆23 ,24 f

]> 0

In terms of mutual information,

s[I (ξ3 ; ξ4 |ξ1 ) + I (ξ3 ; ξ4 |ξ2 ) + I (ξ1 ; ξ2 )− I (ξ3 ; ξ4 )

]+I (ξ2 ; ξ3 |ξ4 ) + s(s+1)

2

[I (ξ2 ; ξ4 |ξ3 ) + I (ξ3 ; ξ4 |ξ2 )

]> 0

s = 1: ZY inequality.

s = 2: 212 ,34 f + ∆24 ,34 f + 3∆23 ,34 f + 3∆23 ,24 f > 0

(Dougherty, Freiling & Zeger, ISIT 2006)

Page 112: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(ξ5 = ξ1 )

For s > 1 and the entropic function f of (ξ1 , ξ2 , ξ3 , ξ4 )

s 12 ,34 f + ∆24 ,34 f + s(s+1)2

[∆23 ,34 f + ∆23 ,24 f

]> 0

In terms of mutual information,

s[I (ξ3 ; ξ4 |ξ1 ) + I (ξ3 ; ξ4 |ξ2 ) + I (ξ1 ; ξ2 )− I (ξ3 ; ξ4 )

]+I (ξ2 ; ξ3 |ξ4 ) + s(s+1)

2

[I (ξ2 ; ξ4 |ξ3 ) + I (ξ3 ; ξ4 |ξ2 )

]> 0

s = 1: ZY inequality.

s = 2: 212 ,34 f + ∆24 ,34 f + 3∆23 ,34 f + 3∆23 ,24 f > 0

(Dougherty, Freiling & Zeger, ISIT 2006)

Page 113: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(ξ5 = ξ1 )

For s > 1 and the entropic function f of (ξ1 , ξ2 , ξ3 , ξ4 )

s 12 ,34 f + ∆24 ,34 f + s(s+1)2

[∆23 ,34 f + ∆23 ,24 f

]> 0

In terms of mutual information,

s[I (ξ3 ; ξ4 |ξ1 ) + I (ξ3 ; ξ4 |ξ2 ) + I (ξ1 ; ξ2 )− I (ξ3 ; ξ4 )

]+I (ξ2 ; ξ3 |ξ4 ) + s(s+1)

2

[I (ξ2 ; ξ4 |ξ3 ) + I (ξ3 ; ξ4 |ξ2 )

]> 0

s = 1: ZY inequality.

s = 2: 212 ,34 f + ∆24 ,34 f + 3∆23 ,34 f + 3∆23 ,24 f > 0

(Dougherty, Freiling & Zeger, ISIT 2006)

Page 114: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(ξ5 = ξ1 )

For s > 1 and the entropic function f of (ξ1 , ξ2 , ξ3 , ξ4 )

s 12 ,34 f + ∆24 ,34 f + s(s+1)2

[∆23 ,34 f + ∆23 ,24 f

]> 0

In terms of mutual information,

s[I (ξ3 ; ξ4 |ξ1 ) + I (ξ3 ; ξ4 |ξ2 ) + I (ξ1 ; ξ2 )− I (ξ3 ; ξ4 )

]+I (ξ2 ; ξ3 |ξ4 ) + s(s+1)

2

[I (ξ2 ; ξ4 |ξ3 ) + I (ξ3 ; ξ4 |ξ2 )

]> 0

s = 1: ZY inequality.

s = 2: 212 ,34 f + ∆24 ,34 f + 3∆23 ,34 f + 3∆23 ,24 f > 0

(Dougherty, Freiling & Zeger, ISIT 2006)

Page 115: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(ξ5 = ξ1 )

For s > 1 and the entropic function f of (ξ1 , ξ2 , ξ3 , ξ4 )

s 12 ,34 f + ∆24 ,34 f + s(s+1)2

[∆23 ,34 f + ∆23 ,24 f

]> 0

In terms of mutual information,

s[I (ξ3 ; ξ4 |ξ1 ) + I (ξ3 ; ξ4 |ξ2 ) + I (ξ1 ; ξ2 )− I (ξ3 ; ξ4 )

]+I (ξ2 ; ξ3 |ξ4 ) + s(s+1)

2

[I (ξ2 ; ξ4 |ξ3 ) + I (ξ3 ; ξ4 |ξ2 )

]> 0

s = 1: ZY inequality.

s = 2: 212 ,34 f + ∆24 ,34 f + 3∆23 ,34 f + 3∆23 ,24 f > 0

(Dougherty, Freiling & Zeger, ISIT 2006)

Page 116: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

(ξ5 = ξ1 )

For s > 1 and the entropic function f of (ξ1 , ξ2 , ξ3 , ξ4 )

s 12 ,34 f + ∆24 ,34 f + s(s+1)2

[∆23 ,34 f + ∆23 ,24 f

]> 0

In terms of mutual information,

s[I (ξ3 ; ξ4 |ξ1 ) + I (ξ3 ; ξ4 |ξ2 ) + I (ξ1 ; ξ2 )− I (ξ3 ; ξ4 )

]+I (ξ2 ; ξ3 |ξ4 ) + s(s+1)

2

[I (ξ2 ; ξ4 |ξ3 ) + I (ξ3 ; ξ4 |ξ2 )

]> 0

s = 1: ZY inequality.

s = 2: 212 ,34 f + ∆24 ,34 f + 3∆23 ,34 f + 3∆23 ,24 f > 0

(Dougherty, Freiling & Zeger, ISIT 2006)

Page 117: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

cl(HentN ) is not polyhedral if and only if |N| > 4.

(FM, ISIT 2007)

For a proof it suffices to consider |N| = 4.

The latter sequence of inequalities is used to arrive atcontradiction with a geometrical lemma.

Page 118: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

cl(HentN ) is not polyhedral if and only if |N| > 4.

(FM, ISIT 2007)

For a proof it suffices to consider |N| = 4.

The latter sequence of inequalities is used to arrive atcontradiction with a geometrical lemma.

Page 119: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

cl(HentN ) is not polyhedral if and only if |N| > 4.

(FM, ISIT 2007)

For a proof it suffices to consider |N| = 4.

The latter sequence of inequalities is used to arrive atcontradiction with a geometrical lemma.

Page 120: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

cl(HentN ) is not polyhedral if and only if |N| > 4.

(FM, ISIT 2007)

For a proof it suffices to consider |N| = 4.

The latter sequence of inequalities is used to arrive atcontradiction with a geometrical lemma.

Page 121: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

If C ⊆ Rd is polyhedral and a curve ϕ : [0, 1] → C has a tangentϕ(0+) then C contains the segment with endpoints ϕ(0) andϕ(0) + εϕ(0+) for some ε > 0.

ZZ

ZZ

ZZ. ............. ............. .............. .............. ................ .............

...............................................................

....................

.......................

.........................

............................

s sϕ(0) ϕ(0) + ε ϕ(0+)

p 7→ ϕ(p)

C

Page 122: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

If C ⊆ Rd is polyhedral and a curve ϕ : [0, 1] → C has a tangentϕ(0+) then C contains the segment with endpoints ϕ(0) andϕ(0) + εϕ(0+) for some ε > 0.

ZZ

ZZ

ZZ. ............. ............. .............. .............. ................ .............

...............................................................

....................

.......................

.........................

............................

s sϕ(0) ϕ(0) + ε ϕ(0+)

p 7→ ϕ(p)

C

Page 123: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 124: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 125: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 126: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 127: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 128: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 129: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 130: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 131: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 132: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

An appropriate curve in cl(HentN ) is constructed

from four 0, 1 -valued variables:

ξ1 = 0 with the probability 2p

ξ2 = 0 with the probability 12

ξ1 independent of ξ2

ξ3 = ξ1 · ξ2

ξ4 = (1− ξ1 )(1− ξ2 ).

h(p)ξ ... the entropy function of (ξ1, ξ2, ξ3, ξ4)

ln 2 ·ϕ(p) = h(p)ξ + β(p) r14

1 + [ ln 2 + 2p ln 2− 12β(2p) ] [ r23

1 + r42 ]

where β(p) = −p ln p − (1− p) ln(1− p)

and r141 , r23

1 , r42 are linear matroids.

Page 133: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

A projection to R3 of the halfspaces given by the new informationinequalities and the curve p 7→ ϕ(p).

Page 134: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Added after the discussion: Classes of matroids

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

binary ternaryF7

F−7

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

linear......................... ....... ........ ......... . ......... ........ ....... .......

........

.........

......................... ............... ......... . ......... ........ ....... ........................

multilinearnon-Pappus

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

partition representable?......................... ....... ........ ......... . ......... ........ ....... .......

........

.........

......................... ............... ......... . ......... ........ ....... ........................

almost entropicF7 ⊕ F−7

Vamos

Page 135: Entropy functions, information inequalities, and polymatroids › workshops › 2009 › 09w5103 › files › matus_09w… · Entropy functions and polymatroids Matroids and Shannon

Entropy functions and polymatroidsMatroids and Shannon entropy

Convolutions and expansionsInformation inequalities

DefinitionZhang-Yeung inequalityAdhesivity of polymatroidsInner adhesivity, also in recurrence

Added after the discussion: Classes of matroids

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

binary ternaryF7

F−7

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

linear......................... ....... ........ ......... . ......... ........ ....... .......

........

.........

......................... ............... ......... . ......... ........ ....... ........................

multilinearnon-Pappus

......................... ....... ........ ......... . ......... ........ ....... ........................

......................... ............... ......... . ......... ........ ....... ........................

partition representable?......................... ....... ........ ......... . ......... ........ ....... .......

........

.........

......................... ............... ......... . ......... ........ ....... ........................

almost entropicF7 ⊕ F−7

Vamos