Check on Inversion with even-odd-preconditiong (Defect #389)


Added by Christopher Pinke over 6 years ago. Updated over 6 years ago.


Status:In Progress Start date:12 Jan 2013
Priority:High Due date:30 Apr 2013
Assignee:Christopher Pinke % Done:

70%

Category:-
Target version:-

Description

Using eo-prec. in the inversion requires the reconstruction of the odd solution by means of the even solution.
Apparently, the analytic expression

x_o = M_inv D x_e - M_inv b_o

has to be replaced by
x_o = -(M_inv D x_e + M_inv b_o)

to give exaclty the same results, which has to be!
The change in the code was applied in c9b6ede9.
Why is that?


Related issues

blocked by CL2QCD - Defect #387: Broken geometric conventions in dslash_eo? New 09 Jan 2013

Associated revisions

Revision 5acc4321
Added by Christopher Pinke over 6 years ago

updated eo inverter comments (refs #389)

Revision 32872301
Added by Christopher Pinke over 6 years ago

fixed sign in pbp in case of eo usage
refs #341, #389

History

Updated by Christopher Pinke over 6 years ago

I recalculated the analytic expression, and it is actually

x_o = - M_inv D x_e + M_inv b_o

However, this does not solve the problem that one has to calculate
x_o = - M_inv D x_e - M_inv b_o

This is true for tm as well as pure wilson fermions.
It seems as if the odd source has to be taken negative, therefore I think this could be related to #387.

However, I checked that with this workaround the correlators and pbp (with 32872301) are the same with and without eo.
Thus, one can use the program and check on this issue again if #387 is fixed.

  • % Done changed from 0 to 70
  • Status changed from New to In Progress

Updated by Christopher Pinke over 6 years ago

I also checked with the tmlqcd program. There, the analytic expressions from above are used, which again indicates the problem is related to #387.

Updated by Christopher Pinke over 6 years ago

  • Due date set to 30 Apr 2013

Also available in: Atom PDF