Ultramagic Squares of Order 7 
Transpositions 
Let M_{7} be the set of all 7x7 matrices containing the first 7^{2} natural numbers,
respectively the whole numbers A (matrix) transformation T is an injective (onetoone) function T: M_{7} > M_{7}. A transposition P is a transformation that is given by a definite permutation of the matrix elements. For instance P(x,y)>(8x,y) describes the reflection of the squares (matrices) with respect to their middle column. Let s be a magic square (matrix) then P(s) is the reflected square. Two transpositions U and V build a new transposition U·V by first applying V and then U to the squares. The set of all transpositions is a noncommutative group with respect to the described connection. The neutral element is the idendity I. As transpositions are injective functions, for each T exists a T^{1} with T^{1}·T = I. 
The 3 rotations and the 4 reflections of the squares build together with the idendity I
a subgroup RS with 8 elements. The 90° clockwise rotation shall be called R. The reflection at the middle row shall be called S. R and S generate the subgroup RS. Two squares that could be transformed to each other by a transposition from RS are meant to be equivalent. A set of squares is called unique, if its elements are pairwise nonequivalent. 
There is another subgroup containing 6 transpositions.
This subgroup could be generated by a single transpostion, that I simply call K. K does not desturb the panmagic and the selfcomplementary property of a 7x7square. K interchanges the rows with the upwards diagonals and the columns with the downwards diagonals. Therefore the transposed square is still panmagic. 
Examine square 0 below. The rows and columms are numbered from 1 to 7. The cell with the adress xy is in column x and row y. Each cell is filled with the number that equals the adress of the cell. Of course these numbers do not produce a magic square. 
By the transposition K the numbers are placed in other cells. The new arrangement could be find in square 1. The numbers in two cells that are positioned symmetrically to the centrecell always sum up to 88, just as in square 0. Thus the square is still selfcomplementary. 
Another apply of K leads to square 2.
It shows that K^{2} is a noncyclic permutation of rows and columns in the order 5724613. Examinations of square 3 show that K^{3} is an involutoric transposition, i.e. K^{3}·K^{3} = I. The values of the original square stay in their cell or are pairwise interchanged. As you can see in square 4, K^{4} is another noncyclic permutation of rows and columns in the order 6374152. K^{5} is equal to K^{1}, the invers transposition of K. 
> K 
> K 
> K 

Square 0  Square 1  Square 2  
> K 
> K 

Square 3  Square 4  Square 5 
The transpositions R, S and K generate a group of 48 different transpositions. One known order7 square could be transposed into 47 other squares. If e is the total number of selfcomplementary panmagic 7x7squares then d = e / 8 is the number of unique squares (i.e. without rotated and reflected squares). But it is enough to count all essentially different squares c = e / 48. All other squares could be derived by applying a suitable transposition. 
K transposes each of the 24 yellow axiscells to another axiscell.
Now look at the four numbers (11, 71, 17, 77) in the cornercells of square 0.
Notice that these numbers move through all yellow axiscells by continued applying of K. Assume the number 1 is placed in any axiscell. Then you always are able to move the number 1 to one of the cornercells. If necessary an additional rotation brings the number 1 into cell (1,1). You need no reflection for this process. The square could be reflected at the downwards main diagonal to interchange n(2,1) and n(1,2) without changing n(1,1). Thus counting all squares with n(1,1) = 1 and n(2,1) < n(1,2) is enough to get the total number of all squares with 1 in any axiscell. 
K transposes each of the 24 green fillingcells to another fillingcell.
Look at the number 21 in the squares 0, 2 and 4 that were produced by
applying K^{2} once, twice or not at all.
With rotations and reflections number 21 could be moved from these positions to all other fillingcells. Thus n(2,1) could be moved to any fillingcell by applying one of the 24 transpositions of the subgroup that is generated by R, S and K^{2}. As you see in square 3, K^{3} interchanges n(1,1) and n(4,6), without changing n(2,1). Thus counting all squares with n(2,1) = 1 and n(1,1) < n(4,6) is enough to get the total number of all squares with 1 in any fillingcell. 
K(x,y)=(v,w) means that the transposition K moves the number from cell (v,w) to cell (x,y). In the first formula x,y ranges from 1 to 7.  
 
The formula could be shortened by using coordinates x, y from 0 to 6:  
 
Reducing all calculations to the cyclic group (0, 1, ... 6), i.e. modulo 7, you get:  
 
This formula could be easily extended to other squares with odd order N = 2m + 1 and x, y from 0 to N1, calculating modulo N:  

In June, 2004 the French engineer Francis Gaspalou reported another transposition for ultramagic squares of order 7. See improvements. 
Transposition for panmagic squares that are similar to K^{2}, K^{3} and K^{4}
are mentioned in the following very interesting book: 'New Recreations with Magic Squares', William H. Benson and Oswald Jacoby, Dover Publications, Inc., New York 1976 
summary  cells  equations  transpositions  improvements 
results  programs  files  conclusions  samples 