# extended linear codes over the field $\mathbb F_q$

Suppose we extend the $[n,k]$ linear code $C$ over the field $\Bbb F_q$ to the code $C'$, where

$$C' = \{(x_1,\ldots ,x_n,x_{n+1})\in \Bbb F_q^{n+1} : (x_1,\ldots,x_n) \in C \text{ and } x_1^2+\ldots+x_{n+1}^2 = 0\}$$

Under what conditions is $C'$ linear?

-
I edited your question using $\LaTeX$. Please check that I didn't alter the meaning of your question. Anyway, $C'$ will be a linear code, if and only if, $C'$ is a subspace of the vector space $\Bbb F_{q^{n+1}}$. Edit: There's something wrong with the question, $C'$ is empty as it is currently defined. – Git Gud Apr 7 '13 at 19:43
I'm relatively certain x12+x22+...+xn+12 was meant to be a sum of squares. ötarcan, can you verify that my edit is indeed what you meant to ask? – Snowball Apr 7 '13 at 23:47
your edit is correct snowball.i meant the sum of squares of n+1 elements is 0 – ötarcan Apr 8 '13 at 6:01

In general, the extended code $C'$ is linear if and only if the map $C \to \mathbb F_q$, mapping a codeword to its extension symbol is linear.

In your case, the extension may or may not be linear, depending on the original code $C$. The problem is that squaring is not linear, normally.

However, there is a large class of cases where we can directly answer the question:

If $q$ is even, then squaring is an automorphism of $\mathbb F_q$ (Frobenius automorphism) and thus, the extension rule is indeed linear. So whenever $q$ is even, $C'$ is linear.

-
+1: The code may also be linear in odd characteristic, if the quadratic polynomial $x_1^2+x_2^2+\cdots+x_n^2$ evaluated at all codewords is either $0$ or a quadratic non-residue modulo $p$. It is not easy to see describe, when this would be the case, which is why I refrained from answering. If the "squared norm" ever takes a non-zero quadratic residue as its value, then there will be two solutions for $x_{n+1}$, and linearity is automatically violated. – Jyrki Lahtonen Apr 12 '13 at 12:39
azimut,do you mean the linearness of C' depends on q is even? – ötarcan Apr 12 '13 at 12:46
@ötarcan: If $q$ is even, $C'$ is linear. If $q$ is odd, $C'$ may or may not be linear, compare the comment of (btw, thanks!) Jyrki Lahtonen. – azimut Apr 12 '13 at 13:20
@ötarcan: For an example of odd characteristic case, when we do geta a linear code: let $p=3$, let $C$ be the code generated by $111$. The code has 3 words: $000,111,222$. The set $C'$ is thus $0000,1110,2220$, which is clearly linear. Alas, I still don't know of a necessary and sufficient condition for the extension like this of an odd $q$ linear code to be linear. Sufficient is easy - norms vanish. Necessary is easy - non-zero norms must all be non-squares. Sufficient and necessary?? – Jyrki Lahtonen Apr 23 '13 at 4:32