A DC optimal power flow ignores losses, so its nodal prices differ across buses only when a line limit binds. The AC answer has both effects. We wanted to know how big each is.
Our earlier solver failed on the 300-bus case in 15 of 18 runs. This time we used PYPOWER's interior-point AC solver and checked it first: it returns 719,725.08 $/h on the unmodified case, the same as MATPOWER's published optimum.
We split the price gap by solving the AC problem with and without line limits. Without limits, prices still spread by 7 to 10 $/MWh from losses alone. With the 3 most loaded lines capped, the congestion part is 1.5 to 5.6 $/MWh, and DC gets it within about 1 $/MWh. With 10 lines capped the congestion part is 25.6 and DC says 6.75.
Caveats: the caps are synthetic (the case's own ratings never bind), and 6 of 18 runs were infeasible. Code, table and log are in research/infra-dcopf.