Christoph Ortner

The problem using n buffers for n threads is it will only work with static scheduling. It needs a different pattern, or it needs to be managed at the user...

I recently started playing recently with hand-coding laplacians for chains. Very painful... I think this would be a fantastic addition for ForwardDiff. Closely related is of course the divergence operator....

After teaching myself about Dual and HyperDual numbers, I wrote this little test: ```julia using ForwardDiff, LinearAlgebra, HyperDualNumbers F(xx) = sum( (xx[i] - xx[j])^2 / (1 + xx[k]^2) for i...

Thanks for flagging this. I'm surprised this hasn't come up in the past, but clearly a bug. EDIT ... I see it's not about `SumIP` but about `OneBody` ... The...

maybe this is now fixed in #158 - but leaving open until can test this thoroughly

@tjjarvinen -- do you agree that this was likely fixed by switching to a static scheduleer?

thank you

sorry for the delay on this. It took forever to register the neighbourlist (and then I was just too busy and forgot). I'll try to do the same for JuLIP...

same problem here

I can confirm that @maximerischard 's suggestion worked for me as well. (though some Cairo tests are failing now) Thank you!