Update 10/14

MHD

Believe current problems are related to prolongation of the aux fields. Because we lose information when going from face-centered to cell-centered B fields, we can't just reconstruct the "correct" (uncorrected) aux fields from q. I believe this is the relevant piece of code, in ProlongateParentData: 

DO i=1,nDim
  DO j=0,r**nDim-1
    IF (MOD(j/r**(i-1),r)==1) CYCLE
    DO n=1,nDim
      l(n)=MOD(j/r**(n-1),r)  
    END DO
    ic(1:nDim,1)=1+l(1:nDim)-rmbc; ic(1:nDim,2)=Info%mX(1:nDim)-r+1+l(1:nDim)+rmbc;ic(i,2)=ic(i,2)+2
    ip(1:nDim,1)=mB(1:nDim,1)-mbc; ip(1:nDim,2)=mB(1:nDim,2)+mbc;ip(i,2)=ip(i,2)+1
    Info%aux(ic(1,1):ic(1,2):r,ic(2,1):ic(2,2):r,ic(3,1):ic(3,2):r,i) =     &
    Parent%auxChild(ip(1,1):ip(1,2)  ,ip(2,1):ip(2,2)  ,ip(3,1):ip(3,2)  ,i)
  END DO
END DO

Haven't looked to see if there are any obvious errors yet. Outer ghost zones were the source of the error in last test.

AMR line transfer

Thinking some more about prioritization. Setting up a stack isn't a bad idea - may be the next thing I try, since it should be easy.

In the truly ideal case, we'd know which processors each ray depended on, and we'd prioritize based on number of dependencies. There may be a better way to approximate this than a stack. I'll continue thinking.

Charge Exchange

Still progressing. Here's the latest frame:

http://www.pas.rochester.edu/~adebrech/ChargeExchange/charge_exchange_species0014.png

http://www.pas.rochester.edu/~adebrech/ChargeExchange/charge_exchange0014.png

Comments

No comments.