# Multiplication of matrices in GF(2) and R

$H$ is an $n \times n$ matrix with elements in $\{ -1,0,1 \}$

$G$ is an $n \times k$ matrix with elements in $GF(2)$ and also upper triangular, invertable

$m$ is an $k \times 1$ vector with elements in $GF(2)$

How can we perceive the output of $HGm$ where $Gm$ multiplication is in $GF(2)$ and $H$ multiplication is a normal real multiplication. Actually I want to combine $HG$ transformation into one $P$ transformation. How can I multiply two matrices while elements in one are in $GF(2)$ and other is in $R$ ? (We can also restrict the entries in $H$ to be one of $-1$ and $1$ but the output can be in $R$).

Motivation: It is a digital communication problem. $Gm$ is output codeword with 1 being mapped to -1 and 0 bit being mapped to +1. This codeword is multiplied to a channel convolution matrix $H$ e.g.

write in MATLAB

H = [1 -1 0 0 0;0 1 -1 0 0; 0 0 1 -1 0; 0 0 0 1 -1; 0 0 0 0 1]'

Now I have the following questions?

1. I want to know whether the whole transformation P = H(G) can be represented by an equivalent transformation P' in GF(2) ? In other words, does there exist a $M \in [GF(2)]^{n \times n}$ such that $P'=G; P=H(MG)$

2. If above is not generally possible, do there exist sufficient conditions for $H$ such that above is even possible?

3. Which branch of mathematics deals with the interaction of real matrices with matrices on finite fields?

-
$HGm$ makes little sense when $H$ is a real matrix and each entry of $Gm$ is an element of $\mathbb F_2$, that is, a residue class of $\mathbb Z$ modulo 2. Each of these residue classes is unchanged under multiplication by $1$ or $-1$. –  Henning Makholm Nov 10 '11 at 20:16
Is $H$ a matrix with entries in $\{+1,-1\}$ or in $\{0,+1,-1\}$? You state the former while describing the set-up but your example uses the latter alphabet. Also, $Gm$ is a vector over $\mathbb F_2$. Are you mapping it to the alphabet $\{+1,-1\}$ using $x \to (-1)^x$? –  Dilip Sarwate Nov 10 '11 at 20:32
I believe you simply take the matrix G and map its entries one by one: 0→−1 and 1→1. I've only seen 0→0 and 1→1, which is a very simple embedding used in decoding algorithms. Neither of these is a homomorphism, but they still end up being helpful in decoding. –  Jack Schmidt Nov 10 '11 at 21:15
Also posted to MO, mathoverflow.net/questions/79427/… –  Gerry Myerson Nov 10 '11 at 21:31
you are right $H \in \{-1,0,+1\}^{n \times n}$ and No I am not mapping it to ${+1, -1}$. Instead, I am mapping it to $\{+1, 0\}$. –  ubaabd Nov 10 '11 at 21:34

The vector ${\mathbf x} = Gm \in \mathbb F_2^n$ is transformed into a vector $\mathbf y$ over the alphabet $\{-1, +1\}$ by mapping each component $x_i$ via $x_i\to (-1)^{x_i}$ since the OP says "output codeword with $1$ being mapped to $-1$ and $0$ bit being mapped to $+1$." But the product ${\mathbf z} = H{\mathbf y}$ is a $n$-tuple of real numbers from the set $\{-n, -n+1, \ldots, n-1, n\}$. So I suppose that the question is how to describe the map from $m \in \mathbb F_2^k$ to $\mathbf z$. One way would be to treat the elements of $\mathbb F_2$ as elements of $\mathbb Z$ and replace all additions $a\oplus b$ in $\mathbb F_2$ with $a + b - 2ab$ while computing the product $Gm$, map $\mathbf x$ to $\mathbf y$ as described above, and finally write the matrix multiplication.
All this might be describable purely in terms of operations in $\mathbb F_2$ with a final map from some $\mathbb F_2^N$ to $\{-n, -n+1, \ldots, n-1, n\}^n$, but I don't see the point of doing so.