We introduce some terminologies here to divide the whole set of magic squares into small subsets.
If you replace each element x of a magic series by n2 + 1 - x , the result is also a magic series. We call the resulting set the complement of the original magic series.
In binary representations, the complement of a magic series is obtained by the bit reverse manipulation.
Every row and column of a magic square is always a magic series. We define the representative magic series of a magic square as the largest magic series which forms a row, a column, the complement of a row, or the complement of a column. Note that diagonal magic series are not considered a representative magic series.
Example: The representative magic series of a magic square
16 12 1 5 7 3 14 10 9 13 4 8 2 6 15 11
is { 16, 13, 3, 2 } = 0x9006, which forms the complement of the third column.
We classify magic squares by their representative magic series. This classification is invariant under rotations, reflections, M-transformations, and the complement transformation.