# Some smaller piece of information that can be derived from a specific type of matrixes?

Given a matrix:

$$A_{m,n} = \begin{pmatrix} a_{1,1} & a_{1,2} & \cdots & a_{1,n} \\ a_{2,1} & a_{2,2} & \cdots & a_{2,n} \\ \vdots & \vdots & \ddots & \vdots \\ a_{m,1} & a_{m,2} & \cdots & a_{m,n} \end{pmatrix}$$

$a_{i,j}$ is a signless integral and bounded. And $b_{i,j}$ is the same. Is there any similarity function between $a_{i,j}$ and $b_{i,j}$. Such that $f(a_{i,j})=f(b_{i,j})$ if and only if $a_{i,j}=b_{i,j}$.

For example, the rank of matrices can identify a class of matrices, not a single matrix.

-
What exactly do you mean by "identity" here? One can see a complex in different ways, e.g. a matrix can be both symmetric and orthogonal, or Hessenberg and banded, or any number of other properties... – J. M. Feb 14 '12 at 3:46
I don't understand. On the simplest interpretation of your question, the way to detect the identity is to see whether $AB=BA=B$ for all matrices $B$. – Gerry Myerson Feb 14 '12 at 3:47
I think by "identity" the OP means "how can I tell if some matrix is identical to a given matrix $A$?" I don't really understand what kind of answer the OP is expecting other than "compare their entries." – Qiaochu Yuan Feb 14 '12 at 3:54
If "$a_{i,j}$ is a signless integral and bounded" means the entries are nonnegative integers with some known bound $B$, then $f(M) = \sum_i \sum_j a_{i,j} B^{i+mj}$ would do. This could conveniently be written as $f(M) = u^T M v$ where $u^T = [B, B^2, \ldots, B^m]$ and $v = [B^m, B^{2m}, \ldots, B^{nm}]^T$. – Robert Israel Feb 14 '12 at 9:13
Following Robert, maybe hashing is really what you are looking for: en.wikipedia.org/wiki/Hash_function ? – Hauke Strasdat Feb 14 '12 at 11:16

In the absence of additional assumptions about the matrix (such as: symmetric, triangular, orthogonal, etc), the matrix form does not really help. The problem is exactly the same as for the list of $mn$ integer values. If each of these integers takes $k$ bits to store, the similarity function $f$ could be their concatenation into a string of $mnk$ bits. This is essentially the function given by Robert Israel: $f(M)=\sum_{ij} a_{ij}B^{i+mj}$ where $B$ is an upper bound on the entries of $M$.
If more is known about the matrix, then some compression is possible: for example, a square antisymmetric matrix of size $n$ is determined by its $n(n-1)/2$ entries lying above the main diagonal.