Dense matrices over \(\GF{2^e}\) for \(2 \leq e \leq 16\) using the M4RIE library#

The M4RIE library offers two matrix representations:

  1. mzed_t

m x n matrices over \(\GF{2^e}\) are internally represented roughly as m x (en) matrices over \(\GF{2}\). Several elements are packed into words such that each element is filled with zeroes until the next power of two. Thus, for example, elements of \(\GF{2^3}\) are represented as [0xxx|0xxx|0xxx|0xxx|...]. This representation is wrapped as Matrix_gf2e_dense in Sage.

Multiplication and elimination both use “Newton-John” tables. These tables are simply all possible multiples of a given row in a matrix such that a scale+add operation is reduced to a table lookup + add. On top of Newton-John multiplication M4RIE implements asymptotically fast Strassen-Winograd multiplication. Elimination uses simple Gaussian elimination which requires \(O(n^3)\) additions but only \(O(n^2 * 2^e)\) multiplications.

  1. mzd_slice_t

m x n matrices over \(\GF{2^e}\) are internally represented as slices of m x n matrices over \(\GF{2}\). This representation allows for very fast matrix times matrix products using Karatsuba’s polynomial multiplication for polynomials over matrices. However, it is not feature complete yet and hence not wrapped in Sage for now.

See http://m4ri.sagemath.org for more details on the M4RIE library.

EXAMPLES:

sage: K.<a> = GF(2^8)
sage: A = random_matrix(K, 3,4)
sage: E = A.echelon_form()
sage: A.row_space() == E.row_space()
True
sage: all(r[r.nonzero_positions()[0]] == 1 for r in E.rows() if r)
True

AUTHOR:

Todo

Wrap mzd_slice_t.

REFERENCES:

class sage.matrix.matrix_gf2e_dense.M4RIE_finite_field#

Bases: object

A thin wrapper around the M4RIE finite field class such that we can put it in a hash table. This class is not meant for public consumption.

class sage.matrix.matrix_gf2e_dense.Matrix_gf2e_dense#

Bases: Matrix_dense

Create new matrix over \(GF(2^e)\) for \(2 \leq e \leq 16\).

INPUT:

  • parent – a matrix space over GF(2^e)

  • entries – see matrix()

  • copy – ignored (for backwards compatibility)

  • coerce – if False, assume without checking that the entries lie in the base ring

EXAMPLES:

sage: K.<a> = GF(2^4)
sage: l = [K.random_element() for _ in range(3*4)]

sage: A = Matrix(K, 3, 4, l)
sage: l == A.list()
True

sage: l[0] == A[0,0]
True

sage: A = Matrix(K, 3, 3, a); A
[a 0 0]
[0 a 0]
[0 0 a]
augment(right)#

Augments self with right.

INPUT:

  • right - a matrix

EXAMPLES:

sage: K.<a> = GF(2^4)
sage: MS = MatrixSpace(K,3,3)
sage: A = random_matrix(K,3,3)
sage: B = A.augment(MS(1))
sage: B.echelonize()
sage: C = B.matrix_from_columns([3,4,5])
sage: A.rank() < 3 or C == ~A
True
sage: A.rank() < 3 or C*A == MS(1)
True
cling(*C)#

Pack the matrices over \(\GF{2}\) into this matrix over \(\GF{2^e}\).

Elements in \(\GF{2^e}\) can be represented as \(\sum c_i a^i\) where \(a\) is a root the minimal polynomial. If this matrix is \(A\) then this function writes \(c_i a^i\) to the entry \(A[x,y]\) where \(c_i\) is the entry \(C_i[x,y]\).

INPUT:

  • C - a list of matrices over GF(2)

EXAMPLES:

sage: K.<a> = GF(2^2)
sage: A = matrix(K, 5, 5)
sage: A0 = random_matrix(GF(2), 5, 5)
sage: A1 = random_matrix(GF(2), 5, 5)
sage: A.cling(A0, A1)
sage: all(A.list()[i] == A0.list()[i] + a*A1.list()[i] for i in range(25))
True

Slicing and clinging are inverse operations:

sage: B0, B1 = A.slice()
sage: B0 == A0 and B1 == A1
True
echelonize(algorithm='heuristic', reduced=True, **kwds)#

Compute the row echelon form of self in place.

INPUT:

  • algorithm - one of the following - heuristic - let M4RIE decide (default) - newton_john - use newton_john table based algorithm - ple - use PLE decomposition - naive - use naive cubic Gaussian elimination (M4RIE implementation) - builtin - use naive cubic Gaussian elimination (Sage implementation)

  • reduced - if True return reduced echelon form. No guarantee is given that the matrix is not reduced if False (default: True)

EXAMPLES:

sage: K.<a> = GF(2^4)
sage: m,n  = 3, 5
sage: A = random_matrix(K, 3, 5)
sage: R = A.row_space()
sage: A.echelonize()
sage: all(r[r.nonzero_positions()[0]] == 1 for r in A.rows() if r)
True
sage: A.row_space() == R
True

sage: K.<a> = GF(2^3)
sage: m,n  = 3, 5
sage: MS = MatrixSpace(K,m,n)
sage: A = random_matrix(K, 3, 5)
sage: B = copy(A).echelon_form('newton_john')
sage: C = copy(A).echelon_form('naive')
sage: D = copy(A).echelon_form('builtin')
sage: B == C == D
True
sage: all(r[r.nonzero_positions()[0]] == 1 for r in B.rows() if r)
True
randomize(density=1, nonzero=False, *args, **kwds)#

Randomize density proportion of the entries of this matrix, leaving the rest unchanged.

INPUT:

  • density - float; proportion (roughly) to be considered for changes

  • nonzero - Bool (default: False); whether the new entries are forced to be non-zero

OUTPUT:

  • None, the matrix is modified in-place

EXAMPLES:

sage: K.<a> = GF(2^4)
sage: total_count = 0
sage: from collections import defaultdict
sage: dic = defaultdict(Integer)
sage: def add_samples():
....:     global dic, total_count
....:     for _ in range(100):
....:         A = Matrix(K,3,3)
....:         A.randomize()
....:         for a in A.list():
....:             dic[a] += 1
....:             total_count += 1.0
sage: add_samples()
sage: while not all(abs(dic[a]/total_count - 1/16) < 0.01 for a in dic):
....:     add_samples()

sage: def add_sample(density):
....:     global density_sum, total_count
....:     total_count += 1.0
....:     density_sum += random_matrix(K, 1000, 1000, density=density).density()

sage: density_sum = 0.0
sage: total_count = 0.0
sage: add_sample(0.1)
sage: while abs(density_sum/total_count - 0.1) > 0.001:
....:     add_sample(0.1)

sage: density_sum = 0.0
sage: total_count = 0.0
sage: add_sample(1.0)
sage: while abs(density_sum/total_count - 1.0) > 0.001:
....:     add_sample(1.0)

sage: density_sum = 0.0
sage: total_count = 0.0
sage: add_sample(0.5)
sage: while abs(density_sum/total_count - 0.5) > 0.001:
....:     add_sample(0.5)

Note, that the matrix is updated and not zero-ed out before being randomized:

sage: def add_sample(density, nonzero):
....:     global density_sum, total_count
....:     total_count += 1.0
....:     A = matrix(K, 1000, 1000)
....:     A.randomize(nonzero=nonzero, density=density)
....:     A.randomize(nonzero=nonzero, density=density)
....:     density_sum += A.density()

sage: density_sum = 0.0
sage: total_count = 0.0
sage: add_sample(0.1, True)
sage: while abs(density_sum/total_count - (1 - 0.9^2)) > 0.001:
....:     add_sample(0.1, True)

sage: density_sum = 0.0
sage: total_count = 0.0
sage: add_sample(0.1, False)
sage: while abs(density_sum/total_count - (1 - 0.9^2)*15/16) > 0.001:
....:     add_sample(0.1, False)

sage: density_sum = 0.0
sage: total_count = 0.0
sage: add_sample(0.05, True)
sage: while abs(density_sum/total_count - (1 - 0.95^2)) > 0.001:
....:     add_sample(0.05, True)

sage: density_sum = 0.0
sage: total_count = 0.0
sage: add_sample(0.5, True)
sage: while abs(density_sum/total_count - (1 - 0.5^2)) > 0.001:
....:     add_sample(0.5, True)
rank()#

Return the rank of this matrix (cached).

EXAMPLES:

sage: K.<a> = GF(2^4)
sage: A = random_matrix(K, 10, 10, algorithm="unimodular")
sage: A.rank()
10
sage: A = matrix(K, 10, 0)
sage: A.rank()
0
slice()#

Unpack this matrix into matrices over \(\GF{2}\).

Elements in \(\GF{2^e}\) can be represented as \(\sum c_i a^i\) where \(a\) is a root the minimal polynomial. This function returns a tuple of matrices \(C\) whose entry \(C_i[x,y]\) is the coefficient of \(c_i\) in \(A[x,y]\) if this matrix is \(A\).

EXAMPLES:

sage: K.<a> = GF(2^2)
sage: A = random_matrix(K, 5, 5)
sage: A0, A1 = A.slice()
sage: all(A.list()[i] == A0.list()[i] + a*A1.list()[i] for i in range(25))
True

sage: K.<a> = GF(2^3)
sage: A = random_matrix(K, 5, 5)
sage: A0, A1, A2 = A.slice()
sage: all(A.list()[i] == A0.list()[i] + a*A1.list()[i] + a^2*A2.list()[i] for i in range(25))
True

Slicing and clinging are inverse operations:

sage: B = matrix(K, 5, 5)
sage: B.cling(A0, A1, A2)
sage: B == A
True
submatrix(row=0, col=0, nrows=-1, ncols=-1)#

Return submatrix from the index row,col (inclusive) with dimension nrows x ncols.

INPUT:

  • row – index of start row

  • col – index of start column

  • nrows – number of rows of submatrix

  • ncols – number of columns of submatrix

EXAMPLES:

sage: K.<a> = GF(2^10)
sage: A = random_matrix(K,200,200)
sage: A[0:2,0:2] == A.submatrix(0,0,2,2)
True
sage: A[0:100,0:100] == A.submatrix(0,0,100,100)
True
sage: A == A.submatrix(0,0,200,200)
True

sage: A[1:3,1:3] == A.submatrix(1,1,2,2)
True
sage: A[1:100,1:100] == A.submatrix(1,1,99,99)
True
sage: A[1:200,1:200] == A.submatrix(1,1,199,199)
True

TESTS for handling of default arguments (github issue #18761):

sage: A.submatrix(17,15) == A.submatrix(17,15,183,185)
True
sage: A.submatrix(row=100,col=37,nrows=1,ncols=3) == A.submatrix(100,37,1,3)
True
sage.matrix.matrix_gf2e_dense.unpickle_matrix_gf2e_dense_v0(a, base_ring, nrows, ncols)#

EXAMPLES:

sage: K.<a> = GF(2^2)
sage: A = random_matrix(K,10,10)
sage: f, s= A.__reduce__()
sage: from sage.matrix.matrix_gf2e_dense import unpickle_matrix_gf2e_dense_v0
sage: f == unpickle_matrix_gf2e_dense_v0
True
sage: f(*s) == A
True

We can still unpickle pickles from before github issue #19240:

sage: old_pickle = b'x\x9c\x85RKo\xd3@\x10\xae\xdd$$\xdb&\xe5U\x1e-\x8f\xc2\xc9\x12RD#$\xce\xa0\xb4\x80\x07\xa2\xca\xc2\x07\x0e\xd5\xe2:\x1b\xdb\x8acg\x1c\xa7J\x85*!\xa4\x90\xe6\x07p\xe0\xc4\x01q\xe5\xc4\x19\xf5\xd0?\xc1\x81\xdf\x80\xb8q\x0b\xb3\x8eMS\xa1\x82V;;\xb3\xdf\xce\xf7\xcd\x8e\xe6\xb5j\xf7,GT;V\x1cy\x83\xf4\xe0\x9d\xb0Y\x13\xbc)\x82\x9e`\xfd\xa0\xeb\xd9m_\xf0\xbf1\xbe{\x97\xa1\xa2\x9d\xc6\xf0\x0f\x82,\x7f\x9d\xa1\xaa\x81\n\xb9m\x9c\xd7\xf4\xf1d2\x81-h\xc0#(\x03\x83\x15\xdas\xc9*\xc3\x13x\x0cu0\xd28\x97\x9e*(0\x9f\xfa\x1b\xd0\xd2\x7fH\x82\xb5\xf4\xa2@TO\xe19\x01I\xac\x136\x991\x9f\xa4\xf9&\xcd\x07i\xbe\xcb\xd4ib\t\xba\xa4\xf6\x02zIT\xd1\x8f2(u\x15\xfd\x9d<\xee@\x05V\xd3\x94E*\xb0\x0e\x0fH\xad\xa8\xbf\x97\xa0\r\x03\xfd\xf0\xb8\x1aU\xff\x92\x90\xe8?\xa5\xd6\x814_\xa5\xf9(\xcd\xafc\xe99\xe2\xd9\xa0\x06\xd4\xf5\xcf\xf2\xf2!\xbc\xd4\xdf\x90#\xc0\x8f\r\xccM\x1b\xdd\x8b\xa3\xbe\x1d\xf7#QmYv\x1cF{\xcc\x11\x81\x88<\x9b\xa71\xcf:\xce0\xaf\x9d\x96\xe3\x87a\xbb\xdf\xe5\x8e\x1f\xeeX>\xc3\x82\xb9\xb0\xe9\x05^,6=\xe17\xf1\xcc\xd0\xc0"u\xb0d\xe6wDl\xdd\x1fa)e\x8a\xbc\xc0\xe9U\xbd \x16\x8e\x88X\xc7j\x0b\x9e\x05\xc8L\xe5\x1e%.\x98\x8a5\xc4\xc5\xd9\xf7\xdd\xd0\xdf\x0b\xc2\x8eg\xf93.wZ\xb5\xc1\x94B\xf8\xa2#\x82\x98a\xf9\xffY\x12\xe3v\x18L\xff\x14Fl\xeb\x0ff\x10\xc4\xb0\xa2\xb9y\xcd-\xba%\xcd\xa5\x8ajT\xd1\x92\xa9\x0c\x86x\xb6a\xe6h\xf8\x02<g\xaa\xaf\xf6\xdd%\x89\xae\x13z\xfe \xc6\x0b\xfb1^4p\x99\x1e6\xc6\xd4\xebK\xdbx\xf9\xc4\x8f[Iw\xf8\x89\xef\xcbQf\xcfh\xe3\x95\x8c\xebj&\xb9\xe2.\x8f\x0c\\ui\x89\xf1x\xf4\xd6\xc0kf\xc1\xf1v\xad(\xc4\xeb\x89~\xfa\xf0\x06\xa8\xa4\x7f\x93\xf4\xd7\x0c\xbcE#\xad\x92\xfc\xed\xeao\xefX\\\x03'
sage: loads(old_pickle)
[    0     a]
[a + 1     1]