Opened 3 years ago

Last modified 2 years ago

#28176 closed defect

Fix few bugs in ClusterAlgebra discovered while working on #26771 — at Version 3

Reported by: etn40ff Owned by:
Priority: major Milestone: sage-8.9
Component: algebra Keywords: ClusterAlgebras
Cc: tscrim, gmoose05, stumpc5, gh-EBanaian, drupel, gh-kelleye Merged in:
Authors: Salvatore Stella Reviewers:
Report Upstream: N/A Work issues:
Branch: u/etn40ff/28176 (Commits, GitHub, GitLab) Commit: d3cd59d86bf16c8414d36cdf5d95d2c5abd7d07f
Dependencies: Stopgaps:

Status badges

Description (last modified by etn40ff)

As pointed out by the people at #26771, ClusterAlgebra has few bugs.

  • Default arguments do not play nicely with UniqueRepresentation

  sage: A1 = ClusterAlgebra(['A',2])
  sage: A2 = ClusterAlgebra(['A',2], cluster_variable_prefix='x')
  sage: A1 is A2
  • mutate_initial keeps the same variable names; In theory this is not a problem but may create confusion. For example suppose we are looking at the Grassmannian of planes in a 4 dimensional vector space. Then the associate cluster algebra is
  sage: A = ClusterAlgebra( matrix(5,[0,1,-1,1,-1]), cluster_variable_names=['p13'], coefficient_names=['p12','p23','p34','p41'], scalars=QQbar); A
  A Cluster Algebra with cluster variable p13 and coefficients p12, p23, p34, p41 over Algebraic Field

Swapping out p13 from the initial seed should give p24 or, at the very least *not* p13. Getting p24 requires some understanding of what the ring in question really is that goes beyond the scope of this class. Currently we get

  sage: A.mutate_initial(0)
  A Cluster Algebra with cluster variable p13 and coefficients p12, p23, p34, p41 over Algebraic Field
  • mutate_initial does not compute all the required F-polynomials and it does not enforce the fact that they are not rational expressions
  sage: A1 = ClusterAlgebra(['A',[2,1],1])
  sage: A2 = A1.mutate_initial([0,1,0])
  sage: len(A2.g_vectors_so_far()) == len(A2.F_polynomials_so_far())
  sage: all(parent(f) == A2._U for f in A2.F_polynomials_so_far())
  • finally mutate_initial does not enforce the fact that initial g-vectors belong to the initial cluster but rather prepend the performed mutation sequence to all the paths already known.
  sage: A2.find_g_vector((0,0,1))
  [0, 1, 0]

This, while technically not false, may generate confusion.

  • _mutated_F is doing / rather than // resulting in slower computations and parent issues.

This ticket deals with all these issues.

Change History (3)

comment:1 Changed 3 years ago by etn40ff

  • Branch set to u/etn40ff/28176
  • Cc tscrim gmoose05 stumpc5 gh-EBanaian drupel gh-kelleye added
  • Commit set to d3cd59d86bf16c8414d36cdf5d95d2c5abd7d07f
  • Component changed from PLEASE CHANGE to algebra
  • Status changed from new to needs_review

New commits:

d3cd59dFix 28176

comment:2 Changed 3 years ago by chapoton

bot is not happy:

sage -t --long src/sage/algebras/  # Tab character found

comment:3 Changed 3 years ago by etn40ff

  • Description modified (diff)

Oops. Fixed, sorry Frédéric!

With the occasion I also got some dramatic speedup in mutate_initial by not mutating F-polynomials a quadratic number of times.

Also I changed a division into a polynomial division in _mutated_F; this enforces F-polynomials to live in the correct ambient and yields a decent speedup in mutate.

Note: See TracTickets for help on using tickets.