Coding Theory 13: Parameters of Codes on Surfaces

Coding theory / 13

Parameters of codes on surfaces

For surface codes, distance is controlled by rational points on zero curves, not by zeros of a function alone.

Let \(X/\mathbb F_q\) be a surface, \(P\subset X(\mathbb F_q)\) a finite evaluation set, and \(\mathcal L=\mathcal L(G)\). The code is the image of

\[H^0(X,\mathcal L)\to\mathbb F_q^{|P|}.\]

Dimension is bounded by \(h^0(X,\mathcal L)\). If the evaluation map is injective, it is equal to that number. Distance depends on how many evaluation points can lie on a zero divisor \(Z(s)\) for nonzero sections \(s\).

Component-count strategy

If every zero divisor has arithmetic genus \(g_Z\) and at most \(r\) irreducible components, then Hasse-Weil type bounds give an upper bound for \(\#Z(s)(\mathbb F_q)\). This component-count strategy leads naturally to the Hansen approach.

Example 1: plane curves

For \(X=\mathbb P^2\) and \(G=mH\), a nonzero section vanishes on a plane curve of degree \(m\). If the curve is irreducible and smooth, its genus is \((m-1)(m-2)/2\). If it splits into many rational lines, the rational point count can be much larger.

Example 2: why components matter

A degree \(m\) plane curve that is a union of \(m\) rational lines can contain roughly \(m(q+1)\) rational points before intersections are subtracted. A smooth irreducible curve of the same degree is controlled by Hasse-Weil with genus about \(m^2/2\). The same divisor class can therefore produce different zero counts.

Hansen bound

Suppose rational points lie on irreducible curves \(C_1,\ldots,C_r\), with \(\#C_i(\mathbb F_q)\le N\). Suppose \(G\cdot C_i\ge0\) and at most \(l\) of the curves \(C_i\) can be contained in a zero divisor. Then

\[d\ge n-lN-\sum_i G\cdot C_i.\]

If every \(G\cdot C_i=\eta\le N\), this becomes

\[d\ge n-lN-(r-l)\eta.\]

Example 3: product surface fibers

On \(\mathbb P^1\times\mathbb P^1\), fix one ruling as the evaluation fibers and use the convention that a divisor of bidegree \((a,b)\) restricts to degree \(a\) on those fibers. For \(0\le a,b<q+1\), such a divisor can contain at most \(b\) fibers of that ruling and intersects each remaining fiber in at most \(a\) rational points. The bound gives \(d\ge(q+1-a)(q+1-b)\).

Example 4: evaluation set affects distance

If evaluation points are spread evenly along ruling fibers, Hansen bound is sharp and structural. If the same number of points is concentrated on a small number of curves, zero divisors containing those curves can destroy distance. Surface codes are sensitive to the geometry of the chosen point set.

Zarzar and Voloch context

Zarzar and Voloch studied surface-code bounds and decoding ideas. Their work is part of the surface-code context, but the source text repeatedly stresses that surface decoding is far less mature than curve decoding. A surface section vanishes on curves, so an error locator would have to control curve components rather than finite sets alone.