1.9

• New: Defining the constraints now also allows using for the symbols <, <=, >=, >. Additionally, the function Eq from SymPy can be used to defined equalities.
• New: The function solve_sdp also accepts solver="cvxopt" to use CVXOPT for solving a relaxation (requires PICOS and CVXOPT).
• New: convert_to_human_readable function returns the objective function and the moment matrix as a string and a matrix of strings to give a symbolic representation of the problem.
• New: get_next_neighbors function retrieves the forward neighbors at a given distance of a site or set of sites in a lattice.
• New: Much faster substitutions if the right-hand side of the substitution never contains variables that are not in the left-hand side.
• New: Non-unique variables are considered only once in each variable set.
• New: When using solve_sdp to solve the relaxation, the solution, its status, and the time it takes to solve are now part of the class SdpRelaxation.
• New: The class Probability provides an intuitive way to define quantum probabilities and Bell inequalities.
• New: The function solve_sdp autodetects available solvers and complains if there is none.
• New: The optional parameter solverparameters to the function solve_sdp can contain a dictionary of options, with a different set for each of the target solvers.
• New: Regression testing framework added.
• Changed: The functions find_rank_loop, sos_decomposition, and get_xmat_value are no longer required an x_mat or y_mat parameter to pass the primal or dual solution. These values are extracted from the solved relaxation. The respective parameters became optional.
• Changed: Constant term in objective function is added to the primal and dual values when using the solve_sdp function.
• Changed: The primal and dual values of the Mosek solution change their signs when using the solve_sdp function.
• Changed: The verbosity parameter also controls the console output of every solver.
• Changed: Faacets relaxations got their own class FaacetsRelaxation.
• Fixed: Localizing matrices are built correctly when substitution rules contain polynomials and when the identity operator is not part of the monomial sets.
• Fixed: The function get_xmat_value also works in Pypy.