How does BWA MEM handle repetitive regions on paired-end reads?

Question

[xianfan.jhu]

If an Illumina read pair was sequenced completely from a repetitive region, i.e., both ends have multiple alignments, would BWA MEM tries to pair up the locations of the two and put them together? What if one of the reads has some sequencing error and have a higher alignment score to another genomic locus? Thanks!