mjo/cone/cone.py

   1 from sage.all import *
   2
   3 def is_lyapunov_like(L,K):
   4     r"""
   5     Determine whether or not ``L`` is Lyapunov-like on ``K``.
   6
   7     We say that ``L`` is Lyapunov-like on ``K`` if `\left\langle
   8     L\left\lparenx\right\rparen,s\right\rangle = 0` for all pairs
   9     `\left\langle x,s \right\rangle` in the complementarity set of
  10     ``K``. It is known [Orlitzky]_ that this property need only be
  11     checked for generators of ``K`` and its dual.
  12
  13     INPUT:
  14
  15     - ``L`` -- A linear transformation or matrix.
  16
  17     - ``K`` -- A polyhedral closed convex cone.
  18
  19     OUTPUT:
  20
  21     ``True`` if it can be proven that ``L`` is Lyapunov-like on ``K``,
  22     and ``False`` otherwise.
  23
  24     .. WARNING::
  25
  26         If this function returns ``True``, then ``L`` is Lyapunov-like
  27         on ``K``. However, if ``False`` is returned, that could mean one
  28         of two things. The first is that ``L`` is definitely not
  29         Lyapunov-like on ``K``. The second is more of an "I don't know"
  30         answer, returned (for example) if we cannot prove that an inner
  31         product is zero.
  32
  33     REFERENCES:
  34
  35     M. Orlitzky. The Lyapunov rank of an improper cone.
  36     http://www.optimization-online.org/DB_HTML/2015/10/5135.html
  37
  38     EXAMPLES:
  39
  40     The identity is always Lyapunov-like in a nontrivial space::
  41
  42         sage: set_random_seed()
  43         sage: K = random_cone(min_ambient_dim=1, max_ambient_dim=8)
  44         sage: L = identity_matrix(K.lattice_dim())
  45         sage: is_lyapunov_like(L,K)
  46         True
  47
  48     As is the "zero" transformation::
  49
  50         sage: K = random_cone(min_ambient_dim=1, max_ambient_dim=8)
  51         sage: R = K.lattice().vector_space().base_ring()
  52         sage: L = zero_matrix(R, K.lattice_dim())
  53         sage: is_lyapunov_like(L,K)
  54         True
  55
  56         Everything in ``K.lyapunov_like_basis()`` should be Lyapunov-like
  57         on ``K``::
  58
  59         sage: K = random_cone(min_ambient_dim=1, max_ambient_dim=6)
  60         sage: all([ is_lyapunov_like(L,K) for L in K.lyapunov_like_basis() ])
  61         True
  62
  63     """
  64     return all([(L*x).inner_product(s) == 0
  65                 for (x,s) in K.discrete_complementarity_set()])
  66
  67
  68 def motzkin_decomposition(K):
  69     r"""
  70     Return the pair of components in the Motzkin decomposition of this cone.
  71
  72     Every convex cone is the direct sum of a strictly convex cone and a
  73     linear subspace [Stoer-Witzgall]_. Return a pair ``(P,S)`` of cones
  74     such that ``P`` is strictly convex, ``S`` is a subspace, and ``K``
  75     is the direct sum of ``P`` and ``S``.
  76
  77     OUTPUT:
  78
  79     An ordered pair ``(P,S)`` of closed convex polyhedral cones where
  80     ``P`` is strictly convex, ``S`` is a subspace, and ``K`` is the
  81     direct sum of ``P`` and ``S``.
  82
  83     REFERENCES:
  84
  85     .. [Stoer-Witzgall] J. Stoer and C. Witzgall. Convexity and
  86        Optimization in Finite Dimensions I.  Springer-Verlag, New
  87        York, 1970.
  88
  89     EXAMPLES:
  90
  91     The nonnegative orthant is strictly convex, so it is its own
  92     strictly convex component and its subspace component is trivial::
  93
  94         sage: K = Cone([(1,0,0),(0,1,0),(0,0,1)])
  95         sage: (P,S) = motzkin_decomposition(K)
  96         sage: K.is_equivalent(P)
  97         True
  98         sage: S.is_trivial()
  99         True
 100
 101     Likewise, full spaces are their own subspace components::
 102
 103         sage: K = Cone([(1,0),(-1,0),(0,1),(0,-1)])
 104         sage: K.is_full_space()
 105         True
 106         sage: (P,S) = motzkin_decomposition(K)
 107         sage: K.is_equivalent(S)
 108         True
 109         sage: P.is_trivial()
 110         True
 111
 112     TESTS:
 113
 114     A random point in the cone should belong to either the strictly
 115     convex component or the subspace component. If the point is nonzero,
 116     it cannot be in both::
 117
 118         sage: set_random_seed()
 119         sage: K = random_cone(max_ambient_dim=8)
 120         sage: (P,S) = motzkin_decomposition(K)
 121         sage: x = K.random_element()
 122         sage: P.contains(x) or S.contains(x)
 123         True
 124         sage: x.is_zero() or (P.contains(x) != S.contains(x))
 125         True
 126
 127     The strictly convex component should always be strictly convex, and
 128     the subspace component should always be a subspace::
 129
 130         sage: set_random_seed()
 131         sage: K = random_cone(max_ambient_dim=8)
 132         sage: (P,S) = motzkin_decomposition(K)
 133         sage: P.is_strictly_convex()
 134         True
 135         sage: S.lineality() == S.dim()
 136         True
 137
 138     The generators of the components are obtained from orthogonal
 139     projections of the original generators [Stoer-Witzgall]_::
 140
 141         sage: set_random_seed()
 142         sage: K = random_cone(max_ambient_dim=8)
 143         sage: (P,S) = motzkin_decomposition(K)
 144         sage: A = S.linear_subspace().complement().matrix()
 145         sage: proj_S_perp = A.transpose() * (A*A.transpose()).inverse() * A
 146         sage: expected_P = Cone([ proj_S_perp*g for g in K ], K.lattice())
 147         sage: P.is_equivalent(expected_P)
 148         True
 149         sage: A = S.linear_subspace().matrix()
 150         sage: proj_S = A.transpose() * (A*A.transpose()).inverse() * A
 151         sage: expected_S = Cone([ proj_S*g for g in K ], K.lattice())
 152         sage: S.is_equivalent(expected_S)
 153         True
 154     """
 155     linspace_gens  = [ copy(b) for b in K.linear_subspace().basis() ]
 156     linspace_gens += [ -b for b in linspace_gens ]
 157
 158     S = Cone(linspace_gens, K.lattice())
 159
 160     # Since ``S`` is a subspace, its dual is its orthogonal complement
 161     # (albeit in the wrong lattice).
 162     S_perp = Cone(S.dual(), K.lattice())
 163     P = K.intersection(S_perp)
 164
 165     return (P,S)
 166
 167 def positive_operator_gens(K):
 168     r"""
 169     Compute generators of the cone of positive operators on this cone.
 170
 171     OUTPUT:
 172
 173     A list of `n`-by-``n`` matrices where ``n == K.lattice_dim()``.
 174     Each matrix ``P`` in the list should have the property that ``P*x``
 175     is an element of ``K`` whenever ``x`` is an element of
 176     ``K``. Moreover, any nonnegative linear combination of these
 177     matrices shares the same property.
 178
 179     EXAMPLES:
 180
 181     The trivial cone in a trivial space has no positive operators::
 182
 183         sage: K = Cone([], ToricLattice(0))
 184         sage: positive_operator_gens(K)
 185         []
 186
 187     Positive operators on the nonnegative orthant are nonnegative matrices::
 188
 189         sage: K = Cone([(1,)])
 190         sage: positive_operator_gens(K)
 191         [[1]]
 192
 193         sage: K = Cone([(1,0),(0,1)])
 194         sage: positive_operator_gens(K)
 195         [
 196         [1 0]  [0 1]  [0 0]  [0 0]
 197         [0 0], [0 0], [1 0], [0 1]
 198         ]
 199
 200     Every operator is positive on the ambient vector space::
 201
 202         sage: K = Cone([(1,),(-1,)])
 203         sage: K.is_full_space()
 204         True
 205         sage: positive_operator_gens(K)
 206         [[1], [-1]]
 207
 208         sage: K = Cone([(1,0),(-1,0),(0,1),(0,-1)])
 209         sage: K.is_full_space()
 210         True
 211         sage: positive_operator_gens(K)
 212         [
 213         [1 0]  [-1  0]  [0 1]  [ 0 -1]  [0 0]  [ 0  0]  [0 0]  [ 0  0]
 214         [0 0], [ 0  0], [0 0], [ 0  0], [1 0], [-1  0], [0 1], [ 0 -1]
 215         ]
 216
 217     TESTS:
 218
 219     A positive operator on a cone should send its generators into the cone::
 220
 221         sage: set_random_seed()
 222         sage: K = random_cone(max_ambient_dim=5)
 223         sage: pi_of_K = positive_operator_gens(K)
 224         sage: all([K.contains(p*x) for p in pi_of_K for x in K.rays()])
 225         True
 226
 227     The dimension of the cone of positive operators is given by the
 228     corollary in my paper::
 229
 230         sage: set_random_seed()
 231         sage: K = random_cone(max_ambient_dim=5)
 232         sage: n = K.lattice_dim()
 233         sage: m = K.dim()
 234         sage: l = K.lineality()
 235         sage: pi_of_K = positive_operator_gens(K)
 236         sage: L = ToricLattice(n**2)
 237         sage: actual = Cone([p.list() for p in pi_of_K], lattice=L).dim()
 238         sage: expected = n**2 - l*(m - l) - (n - m)*m
 239         sage: actual == expected
 240         True
 241
 242     The lineality of the cone of positive operators is given by the
 243     corollary in my paper::
 244
 245         sage: set_random_seed()
 246         sage: K = random_cone(max_ambient_dim=5)
 247         sage: n = K.lattice_dim()
 248         sage: pi_of_K = positive_operator_gens(K)
 249         sage: L = ToricLattice(n**2)
 250         sage: actual = Cone([p.list() for p in pi_of_K], lattice=L).lineality()
 251         sage: expected = n**2 - K.dim()*K.dual().dim()
 252         sage: actual == expected
 253         True
 254
 255     The cone ``K`` is proper if and only if the cone of positive
 256     operators on ``K`` is proper::
 257
 258         sage: set_random_seed()
 259         sage: K = random_cone(max_ambient_dim=5)
 260         sage: pi_of_K = positive_operator_gens(K)
 261         sage: L = ToricLattice(K.lattice_dim()**2)
 262         sage: pi_cone = Cone([p.list() for p in pi_of_K], lattice=L)
 263         sage: K.is_proper() == pi_cone.is_proper()
 264         True
 265     """
 266     # Matrices are not vectors in Sage, so we have to convert them
 267     # to vectors explicitly before we can find a basis. We need these
 268     # two values to construct the appropriate "long vector" space.
 269     F = K.lattice().base_field()
 270     n = K.lattice_dim()
 271
 272     tensor_products = [ s.tensor_product(x) for x in K for s in K.dual() ]
 273
 274     # Convert those tensor products to long vectors.
 275     W = VectorSpace(F, n**2)
 276     vectors = [ W(tp.list()) for tp in tensor_products ]
 277
 278     # Create the *dual* cone of the positive operators, expressed as
 279     # long vectors..
 280     pi_dual = Cone(vectors, ToricLattice(W.dimension()))
 281
 282     # Now compute the desired cone from its dual...
 283     pi_cone = pi_dual.dual()
 284
 285     # And finally convert its rays back to matrix representations.
 286     M = MatrixSpace(F, n)
 287     return [ M(v.list()) for v in pi_cone.rays() ]
 288
 289
 290 def Z_transformation_gens(K):
 291     r"""
 292     Compute generators of the cone of Z-transformations on this cone.
 293
 294     OUTPUT:
 295
 296     A list of `n`-by-``n`` matrices where ``n == K.lattice_dim()``.
 297     Each matrix ``L`` in the list should have the property that
 298     ``(L*x).inner_product(s) <= 0`` whenever ``(x,s)`` is an element the
 299     discrete complementarity set of ``K``. Moreover, any nonnegative
 300     linear combination of these matrices shares the same property.
 301
 302     EXAMPLES:
 303
 304     Z-transformations on the nonnegative orthant are just Z-matrices.
 305     That is, matrices whose off-diagonal elements are nonnegative::
 306
 307         sage: K = Cone([(1,0),(0,1)])
 308         sage: Z_transformation_gens(K)
 309         [
 310         [ 0 -1]  [ 0  0]  [-1  0]  [1 0]  [ 0  0]  [0 0]
 311         [ 0  0], [-1  0], [ 0  0], [0 0], [ 0 -1], [0 1]
 312         ]
 313         sage: K = Cone([(1,0,0,0),(0,1,0,0),(0,0,1,0),(0,0,0,1)])
 314         sage: all([ z[i][j] <= 0 for z in Z_transformation_gens(K)
 315         ....:                    for i in range(z.nrows())
 316         ....:                    for j in range(z.ncols())
 317         ....:                    if i != j ])
 318         True
 319
 320     The trivial cone in a trivial space has no Z-transformations::
 321
 322         sage: K = Cone([], ToricLattice(0))
 323         sage: Z_transformation_gens(K)
 324         []
 325
 326     Z-transformations on a subspace are Lyapunov-like and vice-versa::
 327
 328         sage: K = Cone([(1,0),(-1,0),(0,1),(0,-1)])
 329         sage: K.is_full_space()
 330         True
 331         sage: lls = span([ vector(l.list()) for l in K.lyapunov_like_basis() ])
 332         sage: zs  = span([ vector(z.list()) for z in Z_transformation_gens(K) ])
 333         sage: zs == lls
 334         True
 335
 336     TESTS:
 337
 338     The Z-property is possessed by every Z-transformation::
 339
 340         sage: set_random_seed()
 341         sage: K = random_cone(max_ambient_dim=6)
 342         sage: Z_of_K = Z_transformation_gens(K)
 343         sage: dcs = K.discrete_complementarity_set()
 344         sage: all([(z*x).inner_product(s) <= 0 for z in Z_of_K
 345         ....:                                  for (x,s) in dcs])
 346         True
 347
 348     The lineality space of Z is LL::
 349
 350         sage: set_random_seed()
 351         sage: K = random_cone(min_ambient_dim=1, max_ambient_dim=6)
 352         sage: lls = span([ vector(l.list()) for l in K.lyapunov_like_basis() ])
 353         sage: z_cone  = Cone([ z.list() for z in Z_transformation_gens(K) ])
 354         sage: z_cone.linear_subspace() == lls
 355         True
 356
 357     And thus, the lineality of Z is the Lyapunov rank::
 358
 359         sage: set_random_seed()
 360         sage: K = random_cone(max_ambient_dim=6)
 361         sage: Z_of_K = Z_transformation_gens(K)
 362         sage: L = ToricLattice(K.lattice_dim()**2)
 363         sage: z_cone  = Cone([ z.list() for z in Z_of_K ], lattice=L)
 364         sage: z_cone.lineality() == K.lyapunov_rank()
 365         True
 366
 367     The lineality spaces of pi-star and Z-star are equal:
 368
 369         sage: set_random_seed()
 370         sage: K = random_cone(max_ambient_dim=5)
 371         sage: pi_of_K = positive_operator_gens(K)
 372         sage: Z_of_K = Z_transformation_gens(K)
 373         sage: L = ToricLattice(K.lattice_dim()**2)
 374         sage: pi_star = Cone([p.list() for p in pi_of_K], lattice=L).dual()
 375         sage: z_star  = Cone([ z.list() for z in Z_of_K], lattice=L).dual()
 376         sage: pi_star.linear_subspace() == z_star.linear_subspace()
 377         True
 378     """
 379     # Matrices are not vectors in Sage, so we have to convert them
 380     # to vectors explicitly before we can find a basis. We need these
 381     # two values to construct the appropriate "long vector" space.
 382     F = K.lattice().base_field()
 383     n = K.lattice_dim()
 384
 385     # These tensor products contain generators for the dual cone of
 386     # the cross-positive transformations.
 387     tensor_products = [ s.tensor_product(x)
 388                         for (x,s) in K.discrete_complementarity_set() ]
 389
 390     # Turn our matrices into long vectors...
 391     W = VectorSpace(F, n**2)
 392     vectors = [ W(m.list()) for m in tensor_products ]
 393
 394     # Create the *dual* cone of the cross-positive operators,
 395     # expressed as long vectors..
 396     Sigma_dual = Cone(vectors, lattice=ToricLattice(W.dimension()))
 397
 398     # Now compute the desired cone from its dual...
 399     Sigma_cone = Sigma_dual.dual()
 400
 401     # And finally convert its rays back to matrix representations.
 402     # But first, make them negative, so we get Z-transformations and
 403     # not cross-positive ones.
 404     M = MatrixSpace(F, n)
 405     return [ -M(v.list()) for v in Sigma_cone.rays() ]
 406
 407
 408 def Z_cone(K):
 409     gens = Z_transformation_gens(K)
 410     L = None
 411     if len(gens) == 0:
 412         L = ToricLattice(0)
 413     return Cone([ g.list() for g in gens ], lattice=L)
 414
 415 def pi_cone(K):
 416     gens = positive_operator_gens(K)
 417     L = None
 418     if len(gens) == 0:
 419         L = ToricLattice(0)
 420     return Cone([ g.list() for g in gens ], lattice=L)