"We further showed that certain variants of DOT’s recursive self types can be integrated successfully while keeping the calculus strongly normalizing. This result is surprising, as traditional recursive types are known to make a language Turing-complete."
I suspect you might find the Turing Completeness & Recursive Types "off hand" result from the following paper of interest:
http://drops.dagstuhl.de/opus/volltexte/2017/7276/pdf/LIPIcs...
To quote:
"We further showed that certain variants of DOT’s recursive self types can be integrated successfully while keeping the calculus strongly normalizing. This result is surprising, as traditional recursive types are known to make a language Turing-complete."