You are on page 1of 90

Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A polyhedral approach to
computing border bases

Sebastian Pokutta

Technische Universität Darmstadt


Department of Mathematics

(joint work with Gábor Braun)

Feb 2010 / Darmstadt

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 1 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Why polynomial systems?

Gian-Carlo Rota (Summer 1985): ‘The one contribution


of mine that I hope will be remembered has consisted in
just pointing out that all sorts of problems of combinatorics
can be viewed as problems of location of the zeros of
certain polynomials and in giving these zeros a
combinatorial interpretation. This is now called the critical
problem.’

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 2 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Why polynomial systems?

Polynomial systems naturally occur in discrete optimization.

0/1 solutions to a linear system of equations Ax = b are given by

{x ∈ {0, 1}n | Ax = b, xi2 − xi = 0 ∀i ∈ [n]}.

⇒ System of polynomial equations.

• Graph isomorphism, graph coloring


• Modeling dynamical systems
(Very successful application in oil exploration)

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 3 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Why polynomial systems?

Polynomial systems naturally occur in discrete optimization.

0/1 solutions to a linear system of equations Ax = b are given by

{x ∈ {0, 1}n | Ax = b, xi2 − xi = 0 ∀i ∈ [n]}.

⇒ System of polynomial equations.

• Graph isomorphism, graph coloring


• Modeling dynamical systems
(Very successful application in oil exploration)

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 3 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Why polynomial systems?

Polynomial systems naturally occur in discrete optimization.

0/1 solutions to a linear system of equations Ax = b are given by

{x ∈ {0, 1}n | Ax = b, xi2 − xi = 0 ∀i ∈ [n]}.

⇒ System of polynomial equations.

• Graph isomorphism, graph coloring


• Modeling dynamical systems
(Very successful application in oil exploration)

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 3 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Why polynomial systems?

Polynomial systems naturally occur in discrete optimization.

0/1 solutions to a linear system of equations Ax = b are given by

{x ∈ {0, 1}n | Ax = b, xi2 − xi = 0 ∀i ∈ [n]}.

⇒ System of polynomial equations.

• Graph isomorphism, graph coloring


• Modeling dynamical systems
(Very successful application in oil exploration)

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 3 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Polynomials

Our objects of interest: polynomials


X
p(X) = am x m
m∈Nn
finite support

Set of indeterminates: X := {x1 , . . . , xn }


Set of monomials: Tn := {x m | m ∈ Nn }
Degree of p: deg(p) := max ||m||1
m∈Nn ,am 6=0
Leading form of p: maximal-degree terms
Polynomials with coefficients in K : K [X]

Example: p(X) = 3x12 x21 x57 + 1x19 x21 − 4x33 x42 − 1


⇒ deg(p) = 10 and LF(p) = 3x12 x21 x57 + 1x19 x21
Sebastian Pokutta Polyhedral border bases Darmstadt 2010 4 / 30
Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Vector spaces vs. rings

Polynomials generate a vector space (isomorphic to K ω ):

p, q ∈ K [X], α ∈ K ⇒ p + q ∈ K [X] and αp ∈ K [X]

Polynomials generate a ring:

p, q ∈ K [X] ⇒ p + q ∈ K [X] and pq ∈ K [X]

⇒ K [X] is an associative K -algebra.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 5 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Vector spaces vs. rings

Polynomials generate a vector space (isomorphic to K ω ):

p, q ∈ K [X], α ∈ K ⇒ p + q ∈ K [X] and αp ∈ K [X]

Polynomials generate a ring:

p, q ∈ K [X] ⇒ p + q ∈ K [X] and pq ∈ K [X]

⇒ K [X] is an associative K -algebra.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 5 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Vector spaces vs. rings

Polynomials generate a vector space (isomorphic to K ω ):

p, q ∈ K [X], α ∈ K ⇒ p + q ∈ K [X] and αp ∈ K [X]

Polynomials generate a ring:

p, q ∈ K [X] ⇒ p + q ∈ K [X] and pq ∈ K [X]

⇒ K [X] is an associative K -algebra.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 5 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Vector spaces vs. rings

Polynomials generate a vector space (isomorphic to K ω ):

p, q ∈ K [X], α ∈ K ⇒ p + q ∈ K [X] and αp ∈ K [X]

Polynomials generate a ring:

p, q ∈ K [X] ⇒ p + q ∈ K [X] and pq ∈ K [X]

⇒ K [X] is an associative K -algebra.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 5 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Vector spaces vs. rings

Polynomials generate a vector space (isomorphic to K ω ):

p, q ∈ K [X], α ∈ K ⇒ p + q ∈ K [X] and αp ∈ K [X]

Polynomials generate a ring:

p, q ∈ K [X] ⇒ p + q ∈ K [X] and pq ∈ K [X]

⇒ K [X] is an associative K -algebra.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 5 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

Ideal :

I ⊆ K [X] subring with p ∈ K [X], q ∈ I then pq ∈ I .

⇒ Ideals are ‘absorbing’.

Factor/epimorphic images: I ⊆ K [X] ideal, then:

K [X]/I := {p + I | p ∈ K [X]}.

A fundamental result that (almost always) holds:

‘Every object is an epimorphic image of a free one.’

Question: What does that mean?

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 6 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

Ideal :

I ⊆ K [X] subring with p ∈ K [X], q ∈ I then pq ∈ I .

⇒ Ideals are ‘absorbing’.

Factor/epimorphic images: I ⊆ K [X] ideal, then:

K [X]/I := {p + I | p ∈ K [X]}.

A fundamental result that (almost always) holds:

‘Every object is an epimorphic image of a free one.’

Question: What does that mean?

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 6 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

Ideal :

I ⊆ K [X] subring with p ∈ K [X], q ∈ I then pq ∈ I .

⇒ Ideals are ‘absorbing’.

Factor/epimorphic images: I ⊆ K [X] ideal, then:

K [X]/I := {p + I | p ∈ K [X]}.

A fundamental result that (almost always) holds:

‘Every object is an epimorphic image of a free one.’

Question: What does that mean?

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 6 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

Ideal :

I ⊆ K [X] subring with p ∈ K [X], q ∈ I then pq ∈ I .

⇒ Ideals are ‘absorbing’.

Factor/epimorphic images: I ⊆ K [X] ideal, then:

K [X]/I := {p + I | p ∈ K [X]}.

A fundamental result that (almost always) holds:

‘Every object is an epimorphic image of a free one.’

Question: What does that mean?

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 6 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

One thing to remember: Considering the factor means adding relations.

The result restated:

‘Every object arises as an unrestricted object plus some relations.’

Example 1:
Consider:
xi2 − xi = 0 if and only if xi ∈ {0, 1}.
Define: I := xi2 − xi | i ∈ [n] K [X] and let V := K [X]/I .

⇒ In V we enforce xi2 − xi = 0 for all i ∈ [n].

(Meta-) Example 2:
Polytopes P ⊆ [0, 1]n arise by taking [0, 1]n and adding inequalities.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 7 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

One thing to remember: Considering the factor means adding relations.

The result restated:

‘Every object arises as an unrestricted object plus some relations.’

Example 1:
Consider:
xi2 − xi = 0 if and only if xi ∈ {0, 1}.
Define: I := xi2 − xi | i ∈ [n] K [X] and let V := K [X]/I .

⇒ In V we enforce xi2 − xi = 0 for all i ∈ [n].

(Meta-) Example 2:
Polytopes P ⊆ [0, 1]n arise by taking [0, 1]n and adding inequalities.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 7 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

One thing to remember: Considering the factor means adding relations.

The result restated:

‘Every object arises as an unrestricted object plus some relations.’

Example 1:
Consider:
xi2 − xi = 0 if and only if xi ∈ {0, 1}.
Define: I := xi2 − xi | i ∈ [n] K [X] and let V := K [X]/I .

⇒ In V we enforce xi2 − xi = 0 for all i ∈ [n].

(Meta-) Example 2:
Polytopes P ⊆ [0, 1]n arise by taking [0, 1]n and adding inequalities.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 7 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

One thing to remember: Considering the factor means adding relations.

The result restated:

‘Every object arises as an unrestricted object plus some relations.’

Example 1:
Consider:
xi2 − xi = 0 if and only if xi ∈ {0, 1}.
Define: I := xi2 − xi | i ∈ [n] K [X] and let V := K [X]/I .

⇒ In V we enforce xi2 − xi = 0 for all i ∈ [n].

(Meta-) Example 2:
Polytopes P ⊆ [0, 1]n arise by taking [0, 1]n and adding inequalities.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 7 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

One thing to remember: Considering the factor means adding relations.

The result restated:

‘Every object arises as an unrestricted object plus some relations.’

Example 1:
Consider:
xi2 − xi = 0 if and only if xi ∈ {0, 1}.
Define: I := xi2 − xi | i ∈ [n] K [X] and let V := K [X]/I .

⇒ In V we enforce xi2 − xi = 0 for all i ∈ [n].

(Meta-) Example 2:
Polytopes P ⊆ [0, 1]n arise by taking [0, 1]n and adding inequalities.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 7 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

One thing to remember: Considering the factor means adding relations.

The result restated:

‘Every object arises as an unrestricted object plus some relations.’

Example 1:
Consider:
xi2 − xi = 0 if and only if xi ∈ {0, 1}.
Define: I := xi2 − xi | i ∈ [n] K [X] and let V := K [X]/I .

⇒ In V we enforce xi2 − xi = 0 for all i ∈ [n].

(Meta-) Example 2:
Polytopes P ⊆ [0, 1]n arise by taking [0, 1]n and adding inequalities.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 7 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Ideals and factors

One thing to remember: Considering the factor means adding relations.

The result restated:

‘Every object arises as an unrestricted object plus some relations.’

Example 1:
Consider:
xi2 − xi = 0 if and only if xi ∈ {0, 1}.
Define: I := xi2 − xi | i ∈ [n] K [X] and let V := K [X]/I .

⇒ In V we enforce xi2 − xi = 0 for all i ∈ [n].

(Meta-) Example 2:
Polytopes P ⊆ [0, 1]n arise by taking [0, 1]n and adding inequalities.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 7 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Degree filtrations and zero-dimensional ideals

Let I ideal and define I ≤i := {p ∈ I | deg(p) ≤ i}.


Degree filtration of I :
[
I = I ≤i
i<ω

Similarly for the whole polynomial ring:


[
K [X] = K [X]≤i
i<ω

An ideal I is zero-dimensional if and only if

dim(K [X]/I ) < ∞.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 8 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Degree filtrations and zero-dimensional ideals

Let I ideal and define I ≤i := {p ∈ I | deg(p) ≤ i}.


Degree filtration of I :
[
I = I ≤i
i<ω

Similarly for the whole polynomial ring:


[
K [X] = K [X]≤i
i<ω

An ideal I is zero-dimensional if and only if

dim(K [X]/I ) < ∞.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 8 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Degree filtrations and zero-dimensional ideals

Let I ideal and define I ≤i := {p ∈ I | deg(p) ≤ i}.


Degree filtration of I :
[
I = I ≤i
i<ω

Similarly for the whole polynomial ring:


[
K [X] = K [X]≤i
i<ω

An ideal I is zero-dimensional if and only if

dim(K [X]/I ) < ∞.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 8 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Degree filtrations and zero-dimensional ideals

Let I ideal and define I ≤i := {p ∈ I | deg(p) ≤ i}.


Degree filtration of I :
[
I = I ≤i
i<ω

Similarly for the whole polynomial ring:


[
K [X] = K [X]≤i
i<ω

An ideal I is zero-dimensional if and only if

dim(K [X]/I ) < ∞.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 8 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Order ideals and borders

Definition (Order ideal)


Let O ⊆ Tn such that whenever m1 ∈ O, m2 ∈ Tn with m2 |m1 it
follows m2 ∈ O. Then O is an order ideal .

Definition (Border of O)
Let O be an order ideal. Then

∂O := {xi m | m ∈ O, i ∈ [n]} \ O

is the border of O.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 9 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Order ideals and borders

Definition (Order ideal)


Let O ⊆ Tn such that whenever m1 ∈ O, m2 ∈ Tn with m2 |m1 it
follows m2 ∈ O. Then O is an order ideal .

Definition (Border of O)
Let O be an order ideal. Then

∂O := {xi m | m ∈ O, i ∈ [n]} \ O

is the border of O.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 9 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Order ideals and borders

An order ideal and its border in T2

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 10 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Border bases

Definition (O-border basis)


Let O = {t1 , . . . , tµ } be an order ideal with ∂O := {b1 , . . . , bν }.
Further, let I ⊆ K [X] be a zero-dimensional ideal and
G = {g1 , . . . , gν } ⊆ K [X] a set of polynomials.
Then G is an O-border basis of I if:
P
(i) G is of the form: gj = bj − i∈[µ] αij ti with αij ∈ K ;
(ii) hGiK [X] = I ;
(iii) K [X] = I ⊕ hOiK as vector spaces.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 11 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Border bases

Definition (O-border basis)


Let O = {t1 , . . . , tµ } be an order ideal with ∂O := {b1 , . . . , bν }.
Further, let I ⊆ K [X] be a zero-dimensional ideal and
G = {g1 , . . . , gν } ⊆ K [X] a set of polynomials.
Then G is an O-border basis of I if:
P
(i) G is of the form: gj = bj − i∈[µ] αij ti with αij ∈ K ;
(ii) hGiK [X] = I ;
(iii) K [X] = I ⊕ hOiK as vector spaces.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 11 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Border bases

Definition (O-border basis)


Let O = {t1 , . . . , tµ } be an order ideal with ∂O := {b1 , . . . , bν }.
Further, let I ⊆ K [X] be a zero-dimensional ideal and
G = {g1 , . . . , gν } ⊆ K [X] a set of polynomials.
Then G is an O-border basis of I if:
P
(i) G is of the form: gj = bj − i∈[µ] αij ti with αij ∈ K ;
(ii) hGiK [X] = I ;
(iii) K [X] = I ⊕ hOiK as vector spaces.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 11 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Border bases

Definition (O-border basis)


Let O = {t1 , . . . , tµ } be an order ideal with ∂O := {b1 , . . . , bν }.
Further, let I ⊆ K [X] be a zero-dimensional ideal and
G = {g1 , . . . , gν } ⊆ K [X] a set of polynomials.
Then G is an O-border basis of I if:
P
(i) G is of the form: gj = bj − i∈[µ] αij ti with αij ∈ K ;
(ii) hGiK [X] = I ;
(iii) K [X] = I ⊕ hOiK as vector spaces.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 11 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Border bases

Definition (O-border basis)


Let O = {t1 , . . . , tµ } be an order ideal with ∂O := {b1 , . . . , bν }.
Further, let I ⊆ K [X] be a zero-dimensional ideal and
G = {g1 , . . . , gν } ⊆ K [X] a set of polynomials.
Then G is an O-border basis of I if:
P
(i) G is of the form: gj = bj − i∈[µ] αij ti with αij ∈ K ;
(ii) hGiK [X] = I ;
(iii) K [X] = I ⊕ hOiK as vector spaces.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 11 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Border bases

A necessary and sufficient condition.

Lemma
Let O be an order ideal and I a zero-dimensional ideal. Then O
supports a border basis of I if and only if

K [X] = I ⊕ hOiK as vector spaces.

Order ideals with this property are admissible.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 12 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

First, even if O ⊆ Tn has the right cardinality, i.e.,

|O| = dim(K [X]/I )

it does not necessarily support a border basis:

Example
Let R := {(−1, 1); (1, 1); (0, 0); (1, 0); (0; −1)} ⊆ A2 (Q). Let

I := {p ∈ K [X] | p(r ) = 0 ∀r ∈ R}.

Then O := {1, x, x 2 , x 3 , x 4 } cannot support a border basis:

hOiK ∩ I 6= {0} as x 3 − x ∈ I .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 13 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

First, even if O ⊆ Tn has the right cardinality, i.e.,

|O| = dim(K [X]/I )

it does not necessarily support a border basis:

Example
Let R := {(−1, 1); (1, 1); (0, 0); (1, 0); (0; −1)} ⊆ A2 (Q). Let

I := {p ∈ K [X] | p(r ) = 0 ∀r ∈ R}.

Then O := {1, x, x 2 , x 3 , x 4 } cannot support a border basis:

hOiK ∩ I 6= {0} as x 3 − x ∈ I .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 13 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

First, even if O ⊆ Tn has the right cardinality, i.e.,

|O| = dim(K [X]/I )

it does not necessarily support a border basis:

Example
Let R := {(−1, 1); (1, 1); (0, 0); (1, 0); (0; −1)} ⊆ A2 (Q). Let

I := {p ∈ K [X] | p(r ) = 0 ∀r ∈ R}.

Then O := {1, x, x 2 , x 3 , x 4 } cannot support a border basis:

hOiK ∩ I 6= {0} as x 3 − x ∈ I .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 13 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

First, even if O ⊆ Tn has the right cardinality, i.e.,

|O| = dim(K [X]/I )

it does not necessarily support a border basis:

Example
Let R := {(−1, 1); (1, 1); (0, 0); (1, 0); (0; −1)} ⊆ A2 (Q). Let

I := {p ∈ K [X] | p(r ) = 0 ∀r ∈ R}.

Then O := {1, x, x 2 , x 3 , x 4 } cannot support a border basis:

hOiK ∩ I 6= {0} as x 3 − x ∈ I .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 13 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

Let σ be a term-ordering. Then

K [X] = I ⊕ hTn \ LTσ (I )iK .

Question: Is every order ideal induced by a term-ordering?

Answer: No, this isn’t the case either!

Example
Consider the ideal I := x 2 + xy + y 2 , xy 2 , y 4 K [X] . Then

LTDegLex (I ) = {x 2 , xy 2 , y 4 } and ODegLex = {1, x, y , xy , y 2 , y 3 }.


Via substitution: O = {1, x, y , x 2 , y 2 , y 3 }.
If now x <σ y , then xy <σ y 2 and if y <σ x, then xy <σ x 2 .
(Deg-2 Gröbner basis element comes from x 2 + xy + y 2 )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 14 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

Let σ be a term-ordering. Then

K [X] = I ⊕ hTn \ LTσ (I )iK .

Question: Is every order ideal induced by a term-ordering?

Answer: No, this isn’t the case either!

Example
Consider the ideal I := x 2 + xy + y 2 , xy 2 , y 4 K [X] . Then

LTDegLex (I ) = {x 2 , xy 2 , y 4 } and ODegLex = {1, x, y , xy , y 2 , y 3 }.


Via substitution: O = {1, x, y , x 2 , y 2 , y 3 }.
If now x <σ y , then xy <σ y 2 and if y <σ x, then xy <σ x 2 .
(Deg-2 Gröbner basis element comes from x 2 + xy + y 2 )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 14 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

Let σ be a term-ordering. Then

K [X] = I ⊕ hTn \ LTσ (I )iK .

Question: Is every order ideal induced by a term-ordering?

Answer: No, this isn’t the case either!

Example
Consider the ideal I := x 2 + xy + y 2 , xy 2 , y 4 K [X] . Then

LTDegLex (I ) = {x 2 , xy 2 , y 4 } and ODegLex = {1, x, y , xy , y 2 , y 3 }.


Via substitution: O = {1, x, y , x 2 , y 2 , y 3 }.
If now x <σ y , then xy <σ y 2 and if y <σ x, then xy <σ x 2 .
(Deg-2 Gröbner basis element comes from x 2 + xy + y 2 )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 14 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

Let σ be a term-ordering. Then

K [X] = I ⊕ hTn \ LTσ (I )iK .

Question: Is every order ideal induced by a term-ordering?

Answer: No, this isn’t the case either!

Example
Consider the ideal I := x 2 + xy + y 2 , xy 2 , y 4 K [X] . Then

LTDegLex (I ) = {x 2 , xy 2 , y 4 } and ODegLex = {1, x, y , xy , y 2 , y 3 }.


Via substitution: O = {1, x, y , x 2 , y 2 , y 3 }.
If now x <σ y , then xy <σ y 2 and if y <σ x, then xy <σ x 2 .
(Deg-2 Gröbner basis element comes from x 2 + xy + y 2 )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 14 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

Let σ be a term-ordering. Then

K [X] = I ⊕ hTn \ LTσ (I )iK .

Question: Is every order ideal induced by a term-ordering?

Answer: No, this isn’t the case either!

Example
Consider the ideal I := x 2 + xy + y 2 , xy 2 , y 4 K [X] . Then

LTDegLex (I ) = {x 2 , xy 2 , y 4 } and ODegLex = {1, x, y , xy , y 2 , y 3 }.


Via substitution: O = {1, x, y , x 2 , y 2 , y 3 }.
If now x <σ y , then xy <σ y 2 and if y <σ x, then xy <σ x 2 .
(Deg-2 Gröbner basis element comes from x 2 + xy + y 2 )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 14 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The catch: by far, not every order ideal supports a border basis.

Let σ be a term-ordering. Then

K [X] = I ⊕ hTn \ LTσ (I )iK .

Question: Is every order ideal induced by a term-ordering?

Answer: No, this isn’t the case either!

Example
Consider the ideal I := x 2 + xy + y 2 , xy 2 , y 4 K [X] . Then

LTDegLex (I ) = {x 2 , xy 2 , y 4 } and ODegLex = {1, x, y , xy , y 2 , y 3 }.


Via substitution: O = {1, x, y , x 2 , y 2 , y 3 }.
If now x <σ y , then xy <σ y 2 and if y <σ x, then xy <σ x 2 .
(Deg-2 Gröbner basis element comes from x 2 + xy + y 2 )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 14 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Finding admissible order ideals - a combinatorial problem.

The problem in finding an admissible order ideal is twofold.

(i) We have to choose dim(K [X]/I ) monomials so that we


obtain an order ideal;

(ii) The chosen monomials have to form a basis of K [X]/I .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 15 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Finding admissible order ideals - a combinatorial problem.

The problem in finding an admissible order ideal is twofold.

(i) We have to choose dim(K [X]/I ) monomials so that we


obtain an order ideal;

(ii) The chosen monomials have to form a basis of K [X]/I .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 15 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Finding admissible order ideals - a combinatorial problem.

The problem in finding an admissible order ideal is twofold.

(i) We have to choose dim(K [X]/I ) monomials so that we


obtain an order ideal;

(ii) The chosen monomials have to form a basis of K [X]/I .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 15 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

The order ideal polytope.

Let z denote the characteristic vector of an order ideal O, i.e.,


zm = 1 if and only if m ∈ O.

zm1 ≥ zm2 ∀m1 , m2 ∈ Tn≤d−1 : m1 | m2


X
zm = d
m∈Tn≤d−1
X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U
zm ∈ [0, 1] ∀m ∈ Tn≤d−1

Figure: Order ideal polytope P(I ) with d := dim (K [X]/I )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 16 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

The order ideal polytope.

Let z denote the characteristic vector of an order ideal O, i.e.,


zm = 1 if and only if m ∈ O.

zm1 ≥ zm2 ∀m1 , m2 ∈ Tn≤d−1 : m1 | m2


X
zm = d
m∈Tn≤d−1
X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U
zm ∈ [0, 1] ∀m ∈ Tn≤d−1

Figure: Order ideal polytope P(I ) with d := dim (K [X]/I )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 16 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

The order ideal polytope.

Let z denote the characteristic vector of an order ideal O, i.e.,


zm = 1 if and only if m ∈ O.

zm1 ≥ zm2 ∀m1 , m2 ∈ Tn≤d−1 : m1 | m2


X
zm = d
m∈Tn≤d−1
X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U
zm ∈ [0, 1] ∀m ∈ Tn≤d−1

Figure: Order ideal polytope P(I ) with d := dim (K [X]/I )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 16 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

The order ideal polytope.

Let z denote the characteristic vector of an order ideal O, i.e.,


zm = 1 if and only if m ∈ O.

zm1 ≥ zm2 ∀m1 , m2 ∈ Tn≤d−1 : m1 | m2


X
zm = d
m∈Tn≤d−1
X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U
zm ∈ [0, 1] ∀m ∈ Tn≤d−1

Figure: Order ideal polytope P(I ) with d := dim (K [X]/I )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 16 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

The order ideal polytope.

Let z denote the characteristic vector of an order ideal O, i.e.,


zm = 1 if and only if m ∈ O.

zm1 ≥ zm2 ∀m1 , m2 ∈ Tn≤d−1 : m1 | m2


X
zm = d
m∈Tn≤d−1
X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U
zm ∈ [0, 1] ∀m ∈ Tn≤d−1

Figure: Order ideal polytope P(I ) with d := dim (K [X]/I )

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 16 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

A polyhedral characterization.

Theorem (Braun, Pokutta 2009)


Let I be a zero-dimensional ideal. There is a bijection between its
order ideals and integral points of the order ideal polytope of I .
The bijection is given by
n
ξ : z ∈ P(I ) ∩ ZT 7→ O(z) := {m ∈ Tn | zm = 1}.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 17 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

First, observe that the condition

zm1 ≥ zm2 ∀m1 , m2 ∈ Tn≤d−1 : m1 | m2

indeed ensures that O(z) is an order ideal.

Second, the condition


X
zm = d
m∈Tn≤d−1

ensures that O(z) is of the right size, i.e., |O(z)| = dim(K [X]/I ).

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 18 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

First, observe that the condition

zm1 ≥ zm2 ∀m1 , m2 ∈ Tn≤d−1 : m1 | m2

indeed ensures that O(z) is an order ideal.

Second, the condition


X
zm = d
m∈Tn≤d−1

ensures that O(z) is of the right size, i.e., |O(z)| = dim(K [X]/I ).

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 18 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

It remains to show that


X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U

is equivalent to
I ∩ hO(z)iK = {0},
i.e., the image of O(z) is linearly independent in the factor
K [X]/I .
⇒ together with |O(z)| = dim(K [X]/I ) it follows

K [X] = I ⊕ hOiK .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 19 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

It remains to show that


X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U

is equivalent to
I ∩ hO(z)iK = {0},
i.e., the image of O(z) is linearly independent in the factor
K [X]/I .
⇒ together with |O(z)| = dim(K [X]/I ) it follows

K [X] = I ⊕ hOiK .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 19 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

With our interpretation of z being the characteristic vector


X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U

can be rewritten as

|U ∩ O(z)| ≤ dim(hU ∪ I iK /I ).

The size of U ∩ O(z) is at most the dimension of the vector space


generated by the image of U in the factor K [X]/I .
⇒ Obviously necessary for directness.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 20 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

With our interpretation of z being the characteristic vector


X
zm ≤ dim hU ∪ I iK /I ∀U ⊆ Tn≤d−1 : |U| = d
m∈U

can be rewritten as

|U ∩ O(z)| ≤ dim(hU ∪ I iK /I ).

The size of U ∩ O(z) is at most the dimension of the vector space


generated by the image of U in the factor K [X]/I .
⇒ Obviously necessary for directness.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 20 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

For sufficiency choose U := O(z) in

|U ∩ O(z)| ≤ dim(hU ∪ I iK /I ),

so that we obtain

|O(z)| ≤ dim(hO(z) ∪ I iK /I ).

The dimension of the vector space generated by


the image of O(z) in the factor K [X]/I is at least |O(z)|.
⇒ Sufficient for directness.


Sebastian Pokutta Polyhedral border bases Darmstadt 2010 21 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

The combinatorial problem and a polyhedral description

Sketch of proof.

For sufficiency choose U := O(z) in

|U ∩ O(z)| ≤ dim(hU ∪ I iK /I ),

so that we obtain

|O(z)| ≤ dim(hO(z) ∪ I iK /I ).

The dimension of the vector space generated by


the image of O(z) in the factor K [X]/I is at least |O(z)|.
⇒ Sufficient for directness.


Sebastian Pokutta Polyhedral border bases Darmstadt 2010 21 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

A common problem.

Given a zero-dimensional ideal I , find a ‘nice’ basis of K [X]/I , i.e.,


one that contains a favorable mix of monomials.

Question: How hard is it to choose an order ideal with respect to a


(linear) preference?

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 22 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

A common problem.

Given a zero-dimensional ideal I , find a ‘nice’ basis of K [X]/I , i.e.,


one that contains a favorable mix of monomials.

Question: How hard is it to choose an order ideal with respect to a


(linear) preference?

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 22 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Choosing nice bases is NP-hard

Theorem (Braun, Pokutta 2009)


n
Let I ⊆ K [X] be a zero-dimensional ideal and let c ∈ ZT .
Computing an order ideal O that is admissible for I and maximizes
c is NP-hard.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 23 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

A zero-dimensional ideal that solves k-Clique.

We will establish hardness by a reduction from k-Clique.

For this, for any graph Γ, we will construct a zero-dimensional ideal


IΓ whose order ideals are in one-to-one with the graph’s cliques.

We define
Fn,k := {vj | j ∈ [n − k]} ∪ Tn=3
jx
P
with vj := i∈[n] i i and define In,k := hFn,k iK [X] .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 24 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

A zero-dimensional ideal that solves k-Clique.

We will establish hardness by a reduction from k-Clique.

For this, for any graph Γ, we will construct a zero-dimensional ideal


IΓ whose order ideals are in one-to-one with the graph’s cliques.

We define
Fn,k := {vj | j ∈ [n − k]} ∪ Tn=3
jx
P
with vj := i∈[n] i i and define In,k := hFn,k iK [X] .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 24 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

A zero-dimensional ideal that solves k-Clique.

We will establish hardness by a reduction from k-Clique.

For this, for any graph Γ, we will construct a zero-dimensional ideal


IΓ whose order ideals are in one-to-one with the graph’s cliques.

We define
Fn,k := {vj | j ∈ [n − k]} ∪ Tn=3
jx
P
with vj := i∈[n] i i and define In,k := hFn,k iK [X] .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 24 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Characterizing the order ideals of In,k .

Lemma
Let n ∈ N and k ∈ [n]. Then In,k is a zero-dimensional ideal such
that an order ideal O is admissible if and only if
(i) O=1 ⊆ Tn=1 with |O=1 | = k;
(ii) O=2 = {xy | x, y ∈ O=1 };
(iii) O=` = ∅ for ` ≥ 3.

Therefore, the order ideals of In,k are in one-to-one correspondence


with the choices of k vertices and the associated edges on a
complete graph Kn .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 25 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Characterizing the order ideals of In,k .

Lemma
Let n ∈ N and k ∈ [n]. Then In,k is a zero-dimensional ideal such
that an order ideal O is admissible if and only if
(i) O=1 ⊆ Tn=1 with |O=1 | = k;
(ii) O=2 = {xy | x, y ∈ O=1 };
(iii) O=` = ∅ for ` ≥ 3.

Therefore, the order ideals of In,k are in one-to-one correspondence


with the choices of k vertices and the associated edges on a
complete graph Kn .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 25 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Characterizing the order ideals of In,k .

Lemma
Let n ∈ N and k ∈ [n]. Then In,k is a zero-dimensional ideal such
that an order ideal O is admissible if and only if
(i) O=1 ⊆ Tn=1 with |O=1 | = k;
(ii) O=2 = {xy | x, y ∈ O=1 };
(iii) O=` = ∅ for ` ≥ 3.

Therefore, the order ideals of In,k are in one-to-one correspondence


with the choices of k vertices and the associated edges on a
complete graph Kn .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 25 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Characterizing the order ideals of In,k .

Lemma
Let n ∈ N and k ∈ [n]. Then In,k is a zero-dimensional ideal such
that an order ideal O is admissible if and only if
(i) O=1 ⊆ Tn=1 with |O=1 | = k;
(ii) O=2 = {xy | x, y ∈ O=1 };
(iii) O=` = ∅ for ` ≥ 3.

Therefore, the order ideals of In,k are in one-to-one correspondence


with the choices of k vertices and the associated edges on a
complete graph Kn .

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 25 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Sketch of proof.

Observe, that the polynomials vj are homogeneous of degree one


and the coefficient matrix A := (vj )j∈[n−k] is a Vandermonde
matrix, i.e., every square submatrix is invertible.

If we thus remove any k columns (belonging to k variables of


{x1 , . . . , xn }) from A we obtain an invertible matrix. (without loss
of generality we choose {x1 , . . . , xk }).

Therefore, {x1 , . . . , xk , v1 , . . . , vn−k } is a basis for the


homogeneous polynomials of degree one. So K [X] is also a
polynomial ring in those variables.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 26 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Sketch of proof.

Observe, that the polynomials vj are homogeneous of degree one


and the coefficient matrix A := (vj )j∈[n−k] is a Vandermonde
matrix, i.e., every square submatrix is invertible.

If we thus remove any k columns (belonging to k variables of


{x1 , . . . , xn }) from A we obtain an invertible matrix. (without loss
of generality we choose {x1 , . . . , xk }).

Therefore, {x1 , . . . , xk , v1 , . . . , vn−k } is a basis for the


homogeneous polynomials of degree one. So K [X] is also a
polynomial ring in those variables.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 26 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Sketch of proof.

Observe, that the polynomials vj are homogeneous of degree one


and the coefficient matrix A := (vj )j∈[n−k] is a Vandermonde
matrix, i.e., every square submatrix is invertible.

If we thus remove any k columns (belonging to k variables of


{x1 , . . . , xn }) from A we obtain an invertible matrix. (without loss
of generality we choose {x1 , . . . , xk }).

Therefore, {x1 , . . . , xk , v1 , . . . , vn−k } is a basis for the


homogeneous polynomials of degree one. So K [X] is also a
polynomial ring in those variables.

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 26 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Sketch of proof.

We can therefore conclude


. D E
∼ ∼
hOiK = K [x1 , . . . , xn ] hFn,k iK [X] = K [x1 , . . . , xk ] Tk=3 .
K [X]

Substitution preserves degrees, homogeneity, etc.


Thus for any order ideal: |O=1 | = k, O=2 = {xy | x, y ∈ O=1 },
and O=` = ∅ for all ` ≥ 3.
The other direction follows similary. 

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 27 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Sketch of proof.

We can therefore conclude


. D E
∼ ∼
hOiK = K [x1 , . . . , xn ] hFn,k iK [X] = K [x1 , . . . , xk ] Tk=3 .
K [X]

Substitution preserves degrees, homogeneity, etc.


Thus for any order ideal: |O=1 | = k, O=2 = {xy | x, y ∈ O=1 },
and O=` = ∅ for all ` ≥ 3.
The other direction follows similary. 

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 27 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Proving hardness.

Let Γ = (V , E ) with n := |V | and k ∈ [n] be an instance of


k-Clique.
n
We consider In,k and define c ∈ ZT≤3 via
(
1, if m = xu xv and either (u, v ) ∈ E or u = v ;
cm =
0, otherwise.

Observe that there exists an admissible order ideal O with score


k(k+1)
2 if and only if Γ contains a clique of size k 

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 28 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Proving hardness.

Let Γ = (V , E ) with n := |V | and k ∈ [n] be an instance of


k-Clique.
n
We consider In,k and define c ∈ ZT≤3 via
(
1, if m = xu xv and either (u, v ) ∈ E or u = v ;
cm =
0, otherwise.

Observe that there exists an admissible order ideal O with score


k(k+1)
2 if and only if Γ contains a clique of size k 

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 28 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

A complexity theoretic consideration

Proving hardness.

Let Γ = (V , E ) with n := |V | and k ∈ [n] be an instance of


k-Clique.
n
We consider In,k and define c ∈ ZT≤3 via
(
1, if m = xu xv and either (u, v ) ∈ E or u = v ;
cm =
0, otherwise.

Observe that there exists an admissible order ideal O with score


k(k+1)
2 if and only if Γ contains a clique of size k 

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 28 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

polynomial system order ideal signature optimization [s] counting [s] # order ideals
x 3 , xy 2 + y 3 (1, 3, 1, 1, 1) < 0.01 0.02 3
vanishing ideal of the points (1, 4, 2) < 0.01 0.02 45
(0, 0, 0, 1), (1, 0, 0, 2),
(3, 0, 0, 2), (5, 0, 0, 3),
(−1, 0, 0, 4), (4, 4, 4, 5),
(0, 0, 7, 6)).
x + y + z − u − v , x 2 − x, (1, 4, 5) < 0.01 0.35 1,260
y − y , z − z, u 2 − u, v 2 − v
2 2

x + y + z − u − v , x 3 − x, (1, 4, 7, 6) 0.02 51.50 106,820


y 3 − y , z 2 − z, u 2 − u, v 2 − v
x + y + z − u − v , x 3 − x, (1, 4, 8, 9) 0.02 53.00 108,900
y 3 − y , z 3 − z, u 2 − u, v 2 − v
x + y + z − u − v , x 3 − x, (1, 4, 9, 12, 9) 0.08 300.00* > 1,349,154
y 3 − y , z 3 − z, u 3 − u, v 2 − v
x + y + z − u − v + a, (1, 5, 9) < 0.01 8.68 30,030
x 2 − x, y 2 − y , z 2 − z,
u 2 − u, v 2 − v , a2 − a

Table: Computational results (performed with SCIP and CoCoA).

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 29 / 30


Motivation Algebra 101 - a quick recap Border bases General characterization Hardness Computational results

Thank you!

Sebastian Pokutta Polyhedral border bases Darmstadt 2010 30 / 30

You might also like