Bug in matrix multiplication and inversion? I'm using sage 7.1 and this code here returns false.

M = matrix ([[sqrt(1-1/2),1/2],[1,-sqrt(1-1/2)]])
N = matrix ([[sqrt(1-1/3),1],[1/3,-sqrt(1-1/3)]])
(M*N).inverse() == N.inverse()*M.inverse()

What is going on?

edit retag close merge delete

Sort by » oldest newest most voted

If E denotes the difference;

sage: E = (M*N).inverse() - N.inverse()*M.inverse()
sage: E
[-sqrt(2/3)*sqrt(1/2) + 6/(6*sqrt(2/3)*sqrt(1/2) + 1) - 6*(3*sqrt(2/3) - sqrt(1/2))*(sqrt(2/3) - 2*sqrt(1/2))/((6*sqrt(2/3)*sqrt(1/2) + 1)^2*(sqrt(2/3)*sqrt(1/2) + (3*sqrt(2/3) - sqrt(1/2))*(sqrt(2/3) - 2*sqrt(1/2))/(6*sqrt(2/3)*sqrt(1/2) + 1) + 1)) - 1                                                           -1/2*sqrt(2/3) + sqrt(1/2) + 3*(sqrt(2/3) - 2*sqrt(1/2))/((6*sqrt(2/3)*sqrt(1/2) + 1)*(sqrt(2/3)*sqrt(1/2) + (3*sqrt(2/3) - sqrt(1/2))*(sqrt(2/3) - 2*sqrt(1/2))/(6*sqrt(2/3)*sqrt(1/2) + 1) + 1))]
[                                                           sqrt(2/3) - 1/3*sqrt(1/2) - 2*(3*sqrt(2/3) - sqrt(1/2))/((6*sqrt(2/3)*sqrt(1/2) + 1)*(sqrt(2/3)*sqrt(1/2) + (3*sqrt(2/3) - sqrt(1/2))*(sqrt(2/3) - 2*sqrt(1/2))/(6*sqrt(2/3)*sqrt(1/2) + 1) + 1))                                                                                                                   -sqrt(2/3)*sqrt(1/2) + 1/(sqrt(2/3)*sqrt(1/2) + (3*sqrt(2/3) - sqrt(1/2))*(sqrt(2/3) - 2*sqrt(1/2))/(6*sqrt(2/3)*sqrt(1/2) + 1) + 1) - 1/6]

You get a matrix whose entries are symbolic expressions. If we write:

sage: E == 0
False

We get false because the entries of E are not all zero, as symbolic expressions, but they are actually zero if we simplify all entries one by one:

sage: E
-sqrt(2/3)*sqrt(1/2) + 6/(6*sqrt(2/3)*sqrt(1/2) + 1) - 6*(3*sqrt(2/3) - sqrt(1/2))*(sqrt(2/3) - 2*sqrt(1/2))/((6*sqrt(2/3)*sqrt(1/2) + 1)^2*(sqrt(2/3)*sqrt(1/2) + (3*sqrt(2/3) - sqrt(1/2))*(sqrt(2/3) - 2*sqrt(1/2))/(6*sqrt(2/3)*sqrt(1/2) + 1) + 1)) - 1
sage: E.full_simplify()
0
sage: E.full_simplify()
0
sage: E.full_simplify()
0
sage: E.full_simplify()
0

Note that checking whether a symbolic expression is equal to zero is undecidable in general (though we should admit that in the present case, this is just because symbolics is weak in Sage). Since your symbolic expressions represent algebraic numbers, you can change the ring over which E is defined to be the field of algebraic numbers. In this field, the equalty is decidable, and Sage knows how to handle it well:

sage: E.change_ring(QQbar) == 0
True
more

2

Not that one can also simplify the matrix E directly:

sage: E.simplify_full()
[0 0]
[0 0]

In terms of weakness of symbolics in Sage, is there a reason for not trying to apply simplify_full when one tests an expression for zero? (Note that I am very much illiterate regarding Sage's symbolics!)

Damn, i was looking for full_simplify() but tab completion did not gave anything !

OK thanks. I did actually try simplify on the difference, but not simplify_full()

2

As a general rule, when you are doing purely algebraic computation, an advice is to tell Sage that you're doing algebra, and not deal with symbolics. Here, you can do:

sage: M = matrix (QQbar,[[sqrt(1-1/2),1/2],[1,-sqrt(1-1/2)]])
sage: N = matrix (QQbar,[[sqrt(1-1/3),1],[1/3,-sqrt(1-1/3)]])
sage: (M*N).inverse() == N.inverse()*M.inverse()
True

And another solution is not to work on QQbar but staying in the symbolic world and using the much more powerful method is_zero:

sage: M = matrix ([[sqrt(1-1/2),1/2],[1,-sqrt(1-1/2)]])
sage: N = matrix ([[sqrt(1-1/3),1],[1/3,-sqrt(1-1/3)]])
sage: ((M*N).inverse() - N.inverse()*M.inverse()).is_zero()
True