Tool for conversion of ANF to 2-XOR-OR-AND normal form (2-XNF) implemented in Python 3.

python3 anf_to_2xnf.py input.anf -xp output.xnf

input.anf is a text file with the following structure:

• The first line contains all indeterminates (= unicode strings not containing spaces and not being equal to 1 or 0) separated with a comma and AT LEAST ONE SPACE (important for reading in indeterminates of the form x[1,1]),
• All other lines contain each exactly one polynomial,
• Polynomials are sums (+) of terms and a term is a product (*) of indeterminates or simply 1,
• Lines marked with # at the start are treated as comment lines.

For example, the system of ANFs ${x_1x_2x_3+x_1x_2+1, x_2+x_3+1}$ in the indeterminates $x_1,x_2,x_3$ is encoded as

x[1], x[2], x[3]
x[1]*x[2]*x[3] + x[1]*x[2] + 1
x[2] + x[3] + 1

output.xnf is a text file containing a DIMACS-like description of a 2-XNF representation of the system in input.anf:

• Lines starting with c are ignored as comment lines,
• The header is of the form p xnf n_vars n_cls,
• Linerals are represented as sums of literals (e.g. -1+2+-3 for the lineral $(\neg X_1 \oplus X_2 \oplus \neg X_3)$,
• XNF literals are lists of linerals, separated using spaces and terminated with 0 (e.g. -1+2 3 0 represents $(\neg X_1 \oplus X_2) \lor X_3$).

For example, the XNF $((X_1 \oplus X_2) \lor X_3) \land (X_1 \lor \neg(X_1 \oplus X_3)) \land (X_2 \oplus X_1) \land X_2$ is encoded as

p xnf 3 4
1+2 3 0
1 -1+3 0
2+1 0
2 0

Run python3 anf_to_2xnf.py -h to get detailed informations about all options.

• Python 3.10.12 or newer
• Python package galois (see https://github.com/mhostetter/galois)
• Python package numpy (see https://numpy.org/)

Optional:

• Python package pysat (for option --maxsat) (see https://pysathq.github.io/installation/)