SetOptions in parallel computation

Question

I have NIntegrate[] inside a ParallelDo[] loop. Outside the loop I set some options for the integration using SetOptions[].

To give an example:

SetOptions[NIntegrate, MaxRecursion -> 20];
ParallelDo[NIntegrate[Cos[20000 x]/Sqrt[x], {x, 0, 1}], {i, 1, 4}]

results in the error output

NIntegrate failed to converge to prescribed accuracy after 9 recursive bisections in x near {x} = {0.}. NIntegrate obtained 0.005799163549191785` and 0.014875547652988553` for the integral and error estimates.

Thus, not every kernel seems to know about the set options. How to distribute this setting over every kernel?

Answer 1

Use ParallelEvaluate to make sure that SetOptions is evaluated by each kernel:

ParallelEvaluate[SetOptions[NIntegrate, MaxRecursion -> 20]];
ParallelDo[NIntegrate[Cos[20000 x]/Sqrt[x], {x, 0, 1}], {i, 1, 4}]
(* no messages *)