Incosistent number of paired end reads (BAM file generated by STAR using RSEM)

[Dolev Rahat]

I have used RSEM to quantify paired-end RNA-Seq data. The alignment was done
using STAR which was called internally by RSEM, using a reference prepared using the rsem-prepare-reference script, with a GTF that I generated by downloading the GRCh38.90 gtf from Ensembl and then excluding from the GTF transcripts that are not relevant for my downstream analysis.

I then run samtools flagstat on the genome BAM file:

samtools flagstat SRR627536.genome.sorted.bam

41323930 + 0 in total (QC-passed reads + QC-failed reads)
18115150 + 0 secondary
0 + 0 supplementary
0 + 0 duplicates
39343574 + 0 mapped (95.21% : N/A)
23208780 + 0 paired in sequencing
11604390 + 0 read1
11604390 + 0 read2
21228424 + 0 properly paired (91.47% : N/A)
21228424 + 0 with itself and mate mapped
0 + 0 singletons (0.00% : N/A)
0 + 0 with mate mapped to a different chr
0 + 0 with mate mapped to a different chr (mapQ>=5)

The number of mapped reads is greater then the number of paired reads. So
if I understand correctly, I should expect that have reads for which neither the 0x1 nor the 0x4 bit are set.
However when I run:

samtools view -F 1 SRR627536.genome.sorted.bam

I get a BAM file with 0 rows.

How can this discrepancy be explained? Does STAR/RSEM interpret SAM flags in a way that is different from samtools in some way?

Thanks in advance