Hi

For TopHat -r option, the second calculation (250-150*2) is correct. But in your case, your read1 and read2 overlap. It is better that you merge them into single read and follow tophat protocol for single end reads. There are several softwares that do this. you can use PEAR http://sco.h-its.org/exelixis/web/software/pear/.

Also since these are long reads, make sure you trim your reads before merging them as there might be adapter sequences because of read through during sequencing.

if 150 bp is each single end length then fragment size can not be 250  (inner distance can not be negative)

if 2*75 is 150 then insertion size you should provide is 100 bps

As Abhinay says, your reads overlap and one way to deal with that is to merge the paired-end reads. Even though your average insert size is negative, you probably have some proportion of pairs that do have an insert. So if you want to map the reads as pairs with Tophat I would suggest setting -r to zero and increase the mate standard deviation above the default value of 20 to take into account a larger variation of insert size (if you believe that's the case).