Let A, B, C and D be eight-dimensional vectors. The problem here is to solve A × B × C = D for C, when A, B and D are furnished.
First define these 28 real numbers:
h10 = a3b2 − a2b3 + a5b4 − a4b5 + a7b6 − a6b7
h20 = a1b3 − a3b1 + a5b7 − a7b5 + a6b4 − a4b6
h30 = a2b1 − a1b2 + a4b7 − a7b4 + a5b6 − a6b5
h40 = a1b5 − a5b1 + a2b6 − a6b2 + a7b3 − a3b7
h50 = a4b1 − a1b4 + a6b3 − a3b6 + a7b2 − a2b7
h60 = a1b7 − a7b1 + a3b5 − a5b3 + a4b2 − a2b4
h70 = a2b5 − a5b2 + a3b4 − a4b3 + a6b1 − a1b6
h21 = a3b0 − a0b3 + a6b5 − a5b6 + a7b4 − a4b7
h31 = a0b2 − a2b0 + a5b7 − a7b5 + a6b4 − a4b6
h41 = a2b7 − a7b2 + a3b6 − a6b3 + a5b0 − a0b5
h51 = a0b4 − a4b0 + a2b6 − a6b2 + a7b3 − a3b7
h61 = a4b3 − a3b4 + a5b2 − a2b5 + a7b0 − a0b7
h71 = a0b6 − a6b0 + a3b5 − a5b3 + a4b2 − a2b4
h32 = a1b0 − a0b1 + a4b5 − a5b4 + a6b7 − a7b6
h42 = a5b3 − a3b5 + a6b0 − a0b6 + a7b1 − a1b7
h52 = a0b7 − a7b0 + a3b4 − a4b3 + a6b1 − a1b6
h62 = a0b4 − a4b0 + a1b5 − a5b1 + a7b3 − a3b7
h72 = a1b4 − a4b1 + a3b6 − a6b3 + a5b0 − a0b5
h43 = a0b7 − a7b0 + a2b5 − a5b2 + a6b1 − a1b6
h53 = a0b6 − a6b0 + a1b7 − a7b1 + a4b2 − a2b4
h63 = a1b4 − a4b1 + a2b7 − a7b2 + a5b0 − a0b5
h73 = a4b0 − a0b4 + a5b1 − a1b5 + a6b2 − a2b6
h54 = a1b0 − a0b1 + a2b3 − a3b2 + a6b7 − a7b6
h64 = a2b0 − a0b2 + a3b1 − a1b3 + a7b5 − a5b7
h74 = a0b3 − a3b0 + a2b1 − a1b2 + a5b6 − a6b5
h65 = a0b3 − a3b0 + a2b1 − a1b2 + a4b7 − a7b4
h75 = a0b2 − a2b0 + a1b3 − a3b1 + a6b4 − a4b6
h76 = a1b0 − a0b1 + a2b3 − a3b2 + a4b5 − a5b4
The pattern is that ai bj contributes to hkl when Ui × Uj × Uk = Ul accoring to the baseline cross product. Should hkl not be listed, it equals − hlk. Meanwhile, hkk had it been needed would have equalled zero.
Now form the skew-symmetric matrix H:
H = |
|
Then, with the double dot indicating matrix multiplication, seek values of C that satisfy H ·· C = D. This problem is equivalent to solving a system of eight linear equations in eight variables. Not yet established is whether H is invertible, therefore there may be zero, one or many solutions.
Here is an example where three vectors are given:
A = [ | +1, | 0, | +1, | −1, | 0, | +1, | 0, | −1 ] |
B = [ | −1, | 0, | 0, | +1, | −1, | +1, | −1, | 0 ] |
D = [ | +3, | 0, | −7, | −4, | −4, | +2, | −1, | +2 ] |
and the goal is to find C in A × B × C = D.
The coefficients in the following system correspond to the values in the H matrix above, and the right member of each equation comes from D:
0c0 | − 1c1 | + 1c2 | − 2c3 | − 2c4 | − 1c5 | − 1c6 | + 2c7 = | +3 |
1c0 | + 0c1 | + 2c2 | + 2c3 | − 1c4 | − 3c5 | − 1c6 | − 2c7 = | 0 |
− 1c0 | − 2c1 | + 0c2 | + 0c3 | + 3c4 | + 0c5 | − 2c6 | − 1c7 = | −7 |
2c0 | − 2c1 | + 0c2 | + 0c3 | + 0c4 | + 0c5 | − 2c6 | + 2c7 = | −4 |
2c0 | + 1c1 | − 3c2 | + 0c3 | + 0c4 | + 0c5 | − 2c6 | − 1c7 = | −4 |
1c0 | + 3c1 | + 0c2 | + 0c3 | + 0c4 | + 0c5 | − 1c6 | + 1c7 = | +2 |
1c0 | + 1c1 | + 2c2 | + 2c3 | + 2c4 | + 1c5 | + 0c6 | + 2c7 = | −1 |
− 2c0 | + 2c1 | + 1c2 | − 2c3 | + 1c4 | − 1c5 | − 2c6 | + 0c7 = | +2 |
With application of Gaussian elimination, two equations drop out, leaving an underdetermined system:
c3 + c6 − c7 = | 0 |
c5 + c6 + c7 = | +1 |
c4 − c6 = | −2 |
c2 + c7 = | +1 |
c0 − c6 + c7 = | −1 |
c1 = | +1 |
Among the many parameterizations of C is:
C = [ c6 − c7 − 1, 1, 1 − c7, c7 − c6, c6 − 2, 1 − c6 − c7, c6, c7 ]
which, being linear and having two independent variables, namely c6 and c7, denotes a plane. A particular solution is C = [ 0, +1, +1, −1, −1, 0, +1, 0 ].
If A × B × C = 0, but A and B are non-zero, then C must be a linear combination of A and B, also governed by two independent variables.
We prepared a computer program to evaluate the determinant of matrix H by symbolic manipulation (as opposed to numerical techniques). It turns out that for all complex values of an and bn, the determinant vashishes.
Although evaluating the determinant of an 8-by-8 matrix most generally requires calculating 8! = 40320 products of 8 factors each, we bypassed any product that would have incorporated a zero element on the diagonal, reducing the number of products to 14833 (see derangement). Beyond that, we did not take advantage of the skew symmetry of the matrix, preferring to keep the algorithm simple to facilitate checking.
An example of a product is:
h30h51h12h23h04h45h76h67
which is a polynomial in 214207 terms. (Recall that h67 = −h76) An example of a term in that product is:
−2a1(a2)3(a5)2a6a7b1b2(b3)2(b6)3b7
In all terms of all products, the sum of exponents of all the an is eight, and no exponent is greater than six. The same applies to the bn.
A standard result is that the rank of a skew-symmetric matrix must be an even number.