Minesweeper-type puzzle

Question

The premise is simple. We get a n*n matrix with numbers ranging from 0 to 16. The matrix is the result of a minesweeper-kind of addition whereby we have an original matrix containing numbers ranging from 0 to 3 and the second matrix is none else than the sum of the surrounding cells od the first matrix, wrapping around (i.e. $c_{0\_0}$ = $c_{0\_1}$ + $c_{1\_1}$ $c_{1\_0}$ + $c_{1,2\_0}$ + $c_{0\_2}$ + $c_{2\_2}$ + $c_{0\_0}$ + $c_{2\_0}$

Example:

1 2 3
0 1 3
3 2 1

yields

15 14 13
16 15 13
13 14 15

How would you go about solving this mathematically for squares of arbitrary size 0 <= n < 100?

To clarify: the goal is to get the original matrix (shown first, containing only $0,1,2,3$) from the second matrix.

Answer 1

Partial answer

Edit: Jaap Scherphuis made a insightful comment to show that

Solving this matrix uniquely is not always possible when $n$ is even

Reasoning

Consider the following matrix.

 12 12 12 12
 12 12 12 12
 12 12 12 12
 12 12 12 12
 

One solution for this is as follows

 0 1 2 3
 1 2 3 0
 2 3 0 1
 3 0 1 2
 

However, another perfectly valid solution is this

 1 3 0 2
 3 0 2 1
 0 2 1 3
 2 1 3 0
 

This is due to the fact that the set of simultaneous equations governing the $n=4$ solution are not linearly independent (the rank of the matrix associated to the system is less than $16=n^2$).

This is the case whenever $n$ is even. As Jaap Scherpuis pointed out you could always use a checkerboard pattern using any distinct values for the two colours. For example, in the case $n=6$, we could have

 0 3 0 3 0 3
 3 0 3 0 3 0
 0 3 0 3 0 3
 3 0 3 0 3 0
 0 3 0 3 0 3
 3 0 3 0 3 0
 

or we could have

 1 2 1 2 1 2
 2 1 2 1 2 1
 1 2 1 2 1 2
 2 1 2 1 2 1
 1 2 1 2 1 2
 2 1 2 1 2 1
 

and both give rise to the grid

 12 12 12 12 12 12
 12 12 12 12 12 12
 12 12 12 12 12 12
 12 12 12 12 12 12
 12 12 12 12 12 12
 12 12 12 12 12 12
 

so this result grid does not have a unique solution.

For the other cases

There is an obvious strategy of inverting the system of simultaneous equations but this is quite difficult when $n$ is large (perhaps there is a clever shortcut to find the inverse of such large matrices). In any case, the matrix associated to the system of simultaneous equations seems to always be invertible when $n$ is odd (checked for a few small cases) so the solution is always uniquely determined (even among the reals).

For the case $n=3$

We find that each entry is the sum of all other elements except for the entry in the same position. Hence, if are given a matrix, the strategy would be to take each entry away from its maximum value. Then, check if this is the solution and, if not, add 1 to each entry and check again until the solution is obtained.

For example

In the case given in the question the given matrix is

 15 14 13
 16 15 13
 13 14 15
 

The maximum value is 16, and if we take all entries in the matrix away from 16, we get

 1 2 3
 0 1 3
 3 2 1
 

which is the solution.
Alternatively, if the given matrix were

 16 16 16
 16 16 16
 16 16 16
 

Then taking each entry from the maximum value yields

 0 0 0
 0 0 0
 0 0 0
 

which is not the solution, so we add 1 to each entry to get

 1 1 1
 1 1 1 
 1 1 1
 

which is again not the solution, but if we add 1 more we get

 2 2 2
 2 2 2 
 2 2 2
 

which is the solution.