I watched what Tophat does to create the index, and I came up with the following script to do the same thing:

#!/bin/bash

ANNOTATION="$1"
GENOME="$2"
INDEX_BASE="$3"


if [ -f $INDEX_BASE.fa ]; then
  echo "Index already built at $INDEX_BASE"
  exit 0
else
  echo "Building transcriptome fasta file at $INDEX_BASE.fa for $ANNOTATION:($GENOME.fa)"
  gtf_to_fasta $ANNOTATION $GENOME.fa $INDEX_BASE.fa || {
    echo "gtf_to_fasta failed with exit code $?"
    exit 1
  }
  if [ -f $INDEX_BASE.1.ebwt ]; then
    echo "Forcing rebuild of Bowtie 1 index"
    rm -f $INDEX_BASE.*.ebwt
  fi
  if [ -f $INDEX_BASE.1.bt2 ]; then
    echo "Forcing rebuild of Bowtie 2 index"
    rm -f $INDEX_BASE.*.bt2
  fi
fi

if [ -f $INDEX_BASE.1.ebwt ]; then
  echo "Already built Bowtie 1 index"
else
  echo "Building Bowtie 1 index"
  bowtie-build $INDEX_BASE.fa $INDEX_BASE
fi

if [ -f $INDEX_BASE.1.bt2 ]; then
  echo "Already built Bowtie 2 index"
else
  echo "Building Bowtie 2 index"
  bowtie2-build $INDEX_BASE.fa $INDEX_BASE
fi

If your empty fastq file doesn't work, how about you just use a fastq file with one record in it.

Once the transcriptome index is built, move it to a suitable place so that the nodes in your cluster can see it and I think you should be good.

Create the transcritome index for the first tophat run. For the next tophat run, the program will use the same index.