In row 1, let **A** be the sum of the three cells labelled
**a**, and **B** the sum of the three
cells labelled **b**.

If the square is magic, V zigzagA_{3}:

. the sum ofaaain each row and column isA. the sum ofbbbin each row and column isB.A + B = Σ,(magic constant)

There are 33,550,848 order 6 knight move magic squares. Here is a small sample.

These distribution charts are based on squares of numbers 1 .. 36.

The squares were made with program Order6ZigZagA3. Total run time was 14½ hours.

Make 3x6 rectangles:

For 6x6 squares with pattern:a b a b a b b a b a b a a b a b a b b a b a b a a b a b a b b a b a b awhere the sum ofacells in each row and in each column is the same, (A), and the sum ofbcells in each row and in each column is the same, (B = Σ-A): Make all 3x6 rectangles of the values 0 .. 35, corresponding to thea's and/orb's of the 6x6 square:x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x14 x15 x16 x17where each row has the same sum, and alternating column values have this sum, e.g.,x0+x6+x12 = x3+x9+x15 = x0+x1+x2. Check the 18 values not included in the rectangle to determine if they can make any similar rectangle with sumB = Σ-A. This step is crucial, reducing the second phase of the algorithm from many years of computation to only 1½ hours. Ifx17 > x0and \diagonalx0+x3+x7+x10+x14+x17 = Σ, save in file AA. Ifx15 > x2and /diagonalx2+x5+x7+x10+x12+x15 = Σ, save in file BA. For each file, write a 64-bit vector representing the values in the rectangle. For A files, writex0, x1, x3, x4, x6, x7, x9, x10. For B files, writex1, x2, x4, x5, x7, x8, x10, x11. The remaining values can be computed. For efficient I/O write the bit vector and 8 1-byte values as 2 64-bit hexadecimals.

Combine the A**A** and B**B** rectangles:

For each rectangle in each AAfile, for each rectangle in its corresponding BBfile, ifx2 > x0and the intersection of the bit vectors is empty, make the square.

The numbers of squares made by these checks matched the final results.
Here **x0** and **A** sum refer to number range 1 .. 36.

All squares were made for a few big **x0** values
by a backtracking algorithm, (optimized for the special symmetry):

x0 squares time ---- ------- ---------- 26 38,400 816:45:36 27 25,344 446:20:51 28 6,912 251:11:32 29 3,072 132:04:57 30 0 62:22:18 31 0 24:55:32 32 0 7:53:58 33 0 1:30:14

Before the "crucial" step was added to **phase I**, all squares were made for a few small
**A** A**A** file, B**B** file combinations:

A squares time --- --------- ---------- 33 71,424 0:00:05 34 20,736 0:00:21 35 20,736 0:01:06 36 82,944 0:08:46 37 19,712 0:13:08 38 0 0:45:22 39 20,736 3:41:03 40 0 9:35:00 41 0 25:34:47