1 of 29

Weighted basic parallel processes

from commutative to noncommutative series

L. Clemente (University of Warsaw)

Highlights @ Bordeaux, September 2024

2 of 29

Motivation and context

Theory of power series in commuting variables Q[[x1, …, xd]]

Chomsky-like hierarchy

polynomials

rational

algebraic

D-finite�(holonomic)

D-algebraic

CDA

Typical questions

  • Theory (thick books):
    • closure properties

(+, *, derivation, inverse, composition, …)

    • ring theory
  • Algorithms:
    • Equality: f = g?
    • Membership�e.g.: is a D-finite power series even algebraic?
    • Solvability of systems equations

3 of 29

Motivation and context

Theory of power series in commuting variables Q[[x1, …, xd]]

Chomsky-like hierarchy

polynomials

rational

algebraic

D-finite�(holonomic)

D-algebraic

CDA

Typical questions

  • Theory (thick books):
    • closure properties

(+, *, derivation, inverse, composition, …)

    • ring theory
  • Algorithms:
    • Equality: f = g?
    • Membership�e.g.: is a D-finite power series even algebraic?
    • Solvability of systems equations

What is a good generalisation of power series in noncommuting variables?

4 of 29

Commuting → noncommuting variables

commuting variables x, y, z

noncommuting variables a, b, c

domain

power series

series over Σ := {a, b, c}

notation

[[x, y, z]] := N³ → ℚ

ℚ《Σ》:= Σ* → ℚ

5 of 29

Commuting → noncommuting variables

commuting variables x, y, z

noncommuting variables a, b, c

domain

power series

series over Σ := {a, b, c}

notation

[[x, y, z]] := N³ → ℚ

ℚ《Σ》:= Σ* → ℚ

sum

pointwise

pointwise

6 of 29

Commuting → noncommuting variables

commuting variables x, y, z

noncommuting variables a, b, c

domain

power series

series over Σ := {a, b, c}

notation

[[x, y, z]] := N³ → ℚ

ℚ《Σ》:= Σ* → ℚ

sum

pointwise

pointwise

product

high-school multiplication

xy * z = xyz

7 of 29

Commuting → noncommuting variables

commuting variables x, y, z

noncommuting variables a, b, c

domain

power series

series over Σ := {a, b, c}

notation

[[x, y, z]] := N³ → ℚ

ℚ《Σ》:= Σ* → ℚ

sum

pointwise

pointwise

product

high-school multiplication

xy * z = xyz

shuffle product

ab ⧢ c = cab +acb + abc

8 of 29

Commuting → noncommuting variables

commuting variables x, y, z

noncommuting variables a, b, c

domain

power series

series over Σ := {a, b, c}

notation

[[x, y, z]] := N³ → ℚ

ℚ《Σ》:= Σ* → ℚ

sum

pointwise

pointwise

product

high-school multiplication

xy * z = xyz

shuffle product

ab ⧢ c = cab +acb + abc

derivation

high-school partial derivative

𝛛x(xy) = y

9 of 29

Commuting → noncommuting variables

commuting variables x, y, z

noncommuting variables a, b, c

domain

power series

series over Σ := {a, b, c}

notation

[[x, y, z]] := N³ → ℚ

ℚ《Σ》:= Σ* → ℚ

sum

pointwise

pointwise

product

high-school multiplication

xy * z = xyz

shuffle product

ab ⧢ c = cab +acb + abc

derivation

high-school partial derivative

𝛛x(xy) = y

left derivation

δa(ab) = b, δa(c) = 0

10 of 29

Mantra:

Power series = Series which are commutative

xy

ab + ba

power series in

commuting variables

series in

noncommuting variables

isomorphism of differential rings

11 of 29

Mantra:

Power series = Series which are commutative

xy

ab + ba

power series in

commuting variables

series in

noncommuting variables

isomorphism of differential rings

xyz

xy

abc + … + cab

z

*

ab + ba

c

12 of 29

Mantra:

Power series = Series which are commutative

xy

ab + ba

power series in

commuting variables

series in

noncommuting variables

isomorphism of differential rings

xyz

xy

abc + … + cab

z

*

ab + ba

c

y

b

𝛛x

δa

ab + ba

xy

13 of 29

Mantra:

Power series = Series which are commutative

xy

ab + ba

power series in

commuting variables

series in

noncommuting variables

isomorphism of differential rings

xyz

xy

abc + … + cab

z

*

ab + ba

c

y

b

𝛛x

δa

ab + ba

xy

Gain: shuffle and left derivation are defined for all series in noncommuting variables

14 of 29

Conservative extension

  • Take a class of power series in commuting variables

X ⊆ ℚ[[x, y, z]]

  • Extend to a larger class of series in noncommuting variables

X’ ⊆ ℚ《a, b, c》

  • The extension is conservative:

The new series which are commutative coincide with X

Examples for X:

rational, algebraic, D-finite, differentially algebraic power series

class X ⊆ [[x, y, z]]

of power series in

commuting variables

15 of 29

Conservative extension

class X’ ⊆ ℚ《a, b, c》

of series in

noncommuting variables

class X ⊆ [[x, y, z]]

of power series in

commuting variables

  • Take a class of power series in commuting variables

X ⊆ ℚ[[x, y, z]]

  • Extend to a larger class of series in noncommuting variables

X’ ⊆ ℚ《a, b, c》

  • The extension is conservative:

The new series which are commutative coincide with X

Examples for X:

rational, algebraic, D-finite, differentially algebraic power series

16 of 29

Conservative extension

class X’ ⊆ ℚ《a, b, c》

of series in

noncommuting variables

class X ⊆ [[x, y, z]]

of power series in

commuting variables

X = X’ ∩ commutative series

  • Take a class of power series in commuting variables

X ⊆ ℚ[[x, y, z]]

  • Extend to a larger class of series in noncommuting variables

X’ ⊆ ℚ《a, b, c》

  • The extension is conservative:

The new series which are commutative coincide with X

Examples for X:

rational, algebraic, D-finite, differentially algebraic power series

17 of 29

Conservative extension

class X’ ⊆ ℚ《a, b, c》

of series in

noncommuting variables

class X ⊆ [[x, y, z]]

of power series in

commuting variables

X = X’ ∩ commutative series

  • Take a class of power series in commuting variables

X ⊆ ℚ[[x, y, z]]

  • Extend to a larger class of series in noncommuting variables

X’ ⊆ ℚ《a, b, c》

  • The extension is conservative:

The new series which are commutative coincide with X

Examples for X:

rational, algebraic, D-finite, differentially algebraic power series

18 of 29

Conservative extension

class X’ ⊆ ℚ《a, b, c》

of series in

noncommuting variables

class X ⊆ [[x, y, z]]

of power series in

commuting variables

X = X’ ∩ commutative series

  • Take a class of power series in commuting variables

X ⊆ ℚ[[x, y, z]]

  • Extend to a larger class of series in noncommuting variables

X’ ⊆ ℚ《a, b, c》

  • The extension is conservative:

The new series which are commutative coincide with X

Examples for X:

rational, algebraic, D-finite, differentially algebraic power series

This talk

X = CDA power series [1]

X’ = WBPP series [2]

[1] Bergeron, Reutenauer ”Combinatorial resolution of systems of differential equations […]”. European Journal of Combinatorics, 11(6):501–512, 1990.

[2] C. ”Weighted basic parallel processes and combinatorial enumeration”. CONCUR’24.

19 of 29

Contributions

  1. new model: weighted basic parallel processes (WBPP)�
  2. technical result:

WBPP equivalence in 2EXPSPACE via effective algebraic geometry

  • (applications: combinatorial enumeration)

[multiplicity equivalence of species of structures]

20 of 29

Weighted basic parallel processes

Simultaneous generalisation of

  • Schützenberger weighted finite automata
  • Christensen basic parallel processes

Configurations are polynomials of nonterminals ℚ[X, Y, Z].

The transition function Δ maps nonterminals to configurations.

The semantics of a nonterminal is a series [[X]].

21 of 29

Weighted basic parallel processes

Example

Δa(X) := Y * Z

Δb(Y) := 1

Δb(Z) := 1

Simultaneous generalisation of

  • Schützenberger weighted finite automata
  • Christensen basic parallel processes

Configurations are polynomials of nonterminals ℚ[X, Y, Z].

The transition function Δ maps nonterminals to configurations.

The semantics of a nonterminal is a series [[X]].

22 of 29

Weighted basic parallel processes

Example

Δa(X) := Y * Z

Δb(Y) := 1

Δb(Z) := 1

Y*Z

X

a

WBPP run example

Simultaneous generalisation of

  • Schützenberger weighted finite automata
  • Christensen basic parallel processes

Configurations are polynomials of nonterminals ℚ[X, Y, Z].

The transition function Δ maps nonterminals to configurations.

The semantics of a nonterminal is a series [[X]].

23 of 29

Weighted basic parallel processes

Example

Δa(X) := Y * Z

Δb(Y) := 1

Δb(Z) := 1

Y*Z

X

Z+Y

a

b

WBPP run example

Leibniz rule from calculus

Δb(Y * Z) = Δb(Y) * Z + Y * Δb(Z)

Simultaneous generalisation of

  • Schützenberger weighted finite automata
  • Christensen basic parallel processes

Configurations are polynomials of nonterminals ℚ[X, Y, Z].

The transition function Δ maps nonterminals to configurations.

The semantics of a nonterminal is a series [[X]].

24 of 29

Weighted basic parallel processes

Example

Δa(X) := Y * Z

Δb(Y) := 1

Δb(Z) := 1

Linearity

Δb(Z + Y) = Δb(Z) + Δb(Y)

Y*Z

X

Z+Y

a

b

2

b

WBPP run example

Leibniz rule from calculus

Δb(Y * Z) = Δb(Y) * Z + Y * Δb(Z)

Simultaneous generalisation of

  • Schützenberger weighted finite automata
  • Christensen basic parallel processes

Configurations are polynomials of nonterminals ℚ[X, Y, Z].

The transition function Δ maps nonterminals to configurations.

The semantics of a nonterminal is a series [[X]].

25 of 29

Technical contribution

Equivalence problem [[X]] = [[Y]]?

Theorem. The equivalence problem for WBPP is in 2-EXPSPACE.

26 of 29

Technical contribution

Equivalence problem [[X]] = [[Y]]?

Theorem. The equivalence problem for WBPP is in 2-EXPSPACE.

Algorithm: Build a chain of polynomial ideals of configurations.

Termination: Hilbert’s finite basis theorem.

27 of 29

Technical contribution

Equivalence problem [[X]] = [[Y]]?

Theorem. The equivalence problem for WBPP is in 2-EXPSPACE.

Algorithm: Build a chain of polynomial ideals of configurations.

Termination: Hilbert’s finite basis theorem.

Complexity: Effective bounds from algebraic geometry.

Theorem 4 of [1]. When Σ = {a}, the length of the chain is 2-EXP.

[1] Novikov, Yakovenko “Trajectories of polynomial vector fields […]”, Annales de l’Institut Fourier, 49(2):563–609, 1999.

Note: [1] is about dynamical systems but its techniques apply to WBPP.

28 of 29

Technical contribution

Equivalence problem [[X]] = [[Y]]?

Theorem. The equivalence problem for WBPP is in 2-EXPSPACE.

Algorithm: Build a chain of polynomial ideals of configurations.

Termination: Hilbert’s finite basis theorem.

Complexity: Effective bounds from algebraic geometry.

Theorem 4 of [1]. When Σ = {a}, the length of the chain is 2-EXP.

[1] Novikov, Yakovenko “Trajectories of polynomial vector fields […]”, Annales de l’Institut Fourier, 49(2):563–609, 1999.

Theorem. The same for arbitrary Σ.

Note: [1] is about dynamical systems but its techniques apply to WBPP.

Note: Elementary bounds on polynomial ideals are rare.

29 of 29

Technical contribution

Equivalence problem [[X]] = [[Y]]?

Theorem. The equivalence problem for WBPP is in 2-EXPSPACE.

Algorithm: Build a chain of polynomial ideals of configurations.

Termination: Hilbert’s finite basis theorem.

Complexity: Effective bounds from algebraic geometry.

Theorem 4 of [1]. When Σ = {a}, the length of the chain is 2-EXP.

[1] Novikov, Yakovenko “Trajectories of polynomial vector fields […]”, Annales de l’Institut Fourier, 49(2):563–609, 1999.

Theorem. The same for arbitrary Σ.

Note: [1] is about dynamical systems but its techniques apply to WBPP.

Note: Elementary bounds on polynomial ideals are rare.

more details on the poster