perplexing bowtie2 repeat alignments

[Mark]

I am seeing something perplexing, can you help? I am trying to call variants in plasmids. To do this I am using bowtie2, freebayes and an assortment of other tools. As a QC I’m initially using plasmids of known sequence and reads derived from them. I then introduce alterations in the reference plasmid sequence, align the reads and call variants. One of these altered plasmids has two 2kb exact repeats separated by ~2kb. Into the center of one of these, a 6bp insertion was introduced. The PE reads I then align (separately) come from 2 libraries, one of ~400bp insertions and the other ~800bp insertions. With the 400bp insert library, bowtie2 (in end-to-end mode) produces no gapped alignments that span the insert. With the 800bp insert library, it does. It is unclear to me why bowtie2 did not align these reads and their pairs to the other repeat, since the insert size is not long enough to anchor the pair outside the repeat and the other repeat would have provided a more perfect match (I am running bowtie2 in the default mode that reports only the best alignment). Speculating that the pairs aligning to this position might have longer than average inserts I examined the alignment with “tablet” but found, to the contrary, many have smaller than 800bp inserts. Can someone kindly explain this?