Opened 4 years ago

Closed 4 months ago

#20755 closed defect (fixed)

Bug in solve due to a bug in symbolic_expression_from_maxima_string

Reported by: bruno Owned by:
Priority: major Milestone: sage-8.9
Component: symbolics Keywords: maxima, solve
Cc: Merged in:
Authors: Bruno Grenet Reviewers: Vincent Delecroix
Report Upstream: N/A Work issues:
Branch: eb0ad68 (Commits) Commit: eb0ad68ae007807828945e13e58e9865c67adf77
Dependencies: Stopgaps:

Description

A bug in sage.calculus.calculus.symbolic_expression_from_maxima_string implies bugs in solve and roots for symbolic expressions.

Symptoms

The method solve for symbolic expressions is buggy (the list of multiplicities has size 2 instead of 4), as well as roots as a (serious!) consequence:

sage: w = x^4 - (1+3*i)*x^3 - (2-4*i)*x^2 + (6-2*i)*x - 4 - 4*i
sage: w.solve(x,multiplicities=True)
([x == -1/2*sqrt(2*I) + 3/2*I - 1/2, x == 1/2*sqrt(2*I) + 3/2*I - 1/2, x == (-I + 1), x == (I + 1)],
 [1, 1])
sage: w.roots() # should be 4 roots
[(-1/2*sqrt(2*I) + 3/2*I - 1/2, 1), (1/2*sqrt(2*I) + 3/2*I - 1/2, 1)]

Diagnosis

  1. The behavior of roots is easily explained by the behavior of solve, since roots assume that the length of the list of solutions is the same as the length of the list of multiplicities. This should clearly be the case so the bug is not in roots.
  1. Given the parameter in the example, solve calls Maxima and parses the result. The multiplicities are obtained by invoking P.get('multiplicities'). This is the right invocation to Maxima, no bug there.
  1. To parse the solutions returned by Maxima, solve calls sage.symbolic.relation.string_to_list_of_solutions, which itself calls sage.calculus.symbolic_expression_from_maxima_string. The bug occurs in this last function: Indeed, while there is no apparent reason for this, invoking this function changes the variable multiplicities of Maxima. Here is an example:
sage: w = x^4 - (1+3*i)*x^3 - (2-4*i)*x^2 + (6-2*i)*x - 4 - 4*i
sage: m = (w == 0)._maxima_()
sage: P = m.parent()
sage: s = m.solve(x).str()
sage: s # the list of solutions returned by Maxima
'[_SAGE_VAR_x=-(sqrt(2*%i)-3*%i+1)/2,_SAGE_VAR_x=(sqrt(2*%i)+3*%i-1)/2,_SAGE_VAR_x=1-%i,_SAGE_VAR_x=%i+1]'
sage: P.get('multiplicities') # correct!
'[1,1,1,1]'
sage: l = sage.calculus.calculus.symbolic_expression_from_maxima_string(s)
sage: l
[x == -1/2*sqrt(2*I) + 3/2*I - 1/2,
 x == 1/2*sqrt(2*I) + 3/2*I - 1/2,
 x == (-I + 1),
 x == (I + 1)]
sage: P.get('multiplicities') # WTF???
'[1,1]'

Change History (6)

comment:1 Changed 4 years ago by bruno

  • Type changed from PLEASE CHANGE to defect

comment:2 Changed 4 years ago by nbruin

Well, multiplicities is a system variable. See http://maxima.sourceforge.net/docs/manual/maxima_20.html. Who knows when it gets updated? In particular, the reconstruction of symbolic expressions from strings might be calling into maxima again.

The quick fix is to get the correct value of multiplicities out as soon as we can, before conversion back.

In the longer run, it would be preferable to do the conversion while avoiding strings altogether. See sr_sum and sr_limit etc. for examples how one can use sr_to_max and max_to_sr to convert.

comment:3 Changed 5 months ago by chapoton

  • Milestone changed from sage-7.3 to sage-8.9

seems to work in 8.9.b1

one needs to check and add a doctest

comment:4 Changed 4 months ago by bruno

  • Authors set to Bruno Grenet
  • Branch set to u/bruno/20755_bug_in_solve
  • Commit set to eb0ad68ae007807828945e13e58e9865c67adf77
  • Status changed from new to needs_review

The bug seems indeed fixed. I've add a doctest.


New commits:

eb0ad6820755: Add doctest

comment:5 Changed 4 months ago by vdelecroix

  • Reviewers set to Vincent Delecroix
  • Status changed from needs_review to positive_review

comment:6 Changed 4 months ago by vbraun

  • Branch changed from u/bruno/20755_bug_in_solve to eb0ad68ae007807828945e13e58e9865c67adf77
  • Resolution set to fixed
  • Status changed from positive_review to closed
Note: See TracTickets for help on using tickets.