Fig. 2.

Overview of the problem: we assume that the prefix of L and the suffix of R is known to be homologous to the reference. It thus suffices to consider a “core region” indicated by the black frame. In this region we consider alignments in which L and R overlap (above) and alignments in which a gap remains between L and R (below). In the first case, the overlapping, hashed, region is scored as three-way alignment and deletions of a prefix of R and a suffix of L do not contribute to the score. In the second case, the gap between L and R is penalized, however. Dashed lines indicate the non-aligned parts of the sequences L and R, resp