gf_update kernel broken? (Unit Test #326)
At cdf48d884ad7aa9d3bd068382ee0ec00af45b296, gf_update tests 2 and 5 fail, indicating that the kernel is not working correctly!
|blocks CL2QCD - Unit Test #325: test opencl_module_hmc||Done||22 Oct 2012|
With comit:15edbeaadf01ce6961528d336ab64cb6c5e24c93 I worked over the gf_update tests. The kernel was not broken.
I was confused that when updating a cold gf with a cold input gm, the plaquette stayed "1", but the polyakov loop changed. However, in this case one can simply calculate the results of the update.
The plaquette stays the same because the updated links are all the same matrix (exp(i eps * cold ae) -> Pl ~ tr(1) because of the daggered links ), while the polyakov loop (who does not contain daggered links) becomes exp(i eps * cold ae * Ntau) at every spatial position. And this is not a unit matrix, thus the polyakovloop is not one anymore. One can check that by doubling Ntau and halfing eps, which then gives the same value of the polyakov loop.
- % Done changed from 0 to 100
- Estimated time set to 1.00
- Status changed from In Progress to Done