I am new to the BPMN topic and worked myself into this a bit.
One can read that the BPMN representation is often converted to a Petri net representation, and only then it will be run by a Petri net machine / automat. I hope I got this right.
So I asked myself why this is done so often (or maybe always).
Would it also make sense to build right away a BPMN automat to spare the conversion to a Petri net?
(For checking the soundness of a BPMN topic, the Petri net conversion makes sense as it is mathematically quite well understood. My question is more about execution.)
in the scientific community, BPMN is often converted to petri nets for formal analysis. That is because petri nets have very clear execution semantics and there is a large amount of existing theory (e.g. detecting deadlocks, reachability, etc.), while the BPMN specification is not so clear in that respect and I guess there are less BPMN-specific theory.