Size bounds for Tseitin splitting #
The typed transformation is linear in source clauses and literal occurrences. Because the concrete SAT codec writes variable indices in unary, newly allocated indices make the honest bit-level bound quadratic in the source encoding length.
The transformation consumes at most twice as many fresh variables as there are bits in the source encoding.
The transformed formula has at most three times as many clauses as source encoding bits, independently of the initial fresh counter.
Every variable emitted from a compact fresh counter is at most three times the source encoding length.
Concrete quadratic size bound for a compact fresh counter. If all
source variables precede next and next ≤ |encode φ| + 1, then the exact-3
output uses at most 96 * (|encode φ| + 1)^2 bits.
The transformed formula has at most three times as many clauses as source encoding bits.