USQCD project July, 1 2006---June, 30, 2009

"B_K, B_7 and B_8 with unquenched, improved staggered fermions"

Dr. Chulwoo Jung (BNL)

Prof. Weonjong Lee (Seoul National University)

Prof. Stephen R. Sharpe (UW)

Background || Updates || Links

Updates (most recent first)

3-11-2011 Proposal submitted to USQCD for 100 K hours of GPU time during the 2011-12 allocation year. Proposal awarded 100 K GPU-hours

11-12-2010 Proposal submitted for GPU time during first 6 months of 2011. Requested 88K GPU-hours; awarded 81K GPU-hours on the Jlab cluster.

11-10-2010 Running on QCDOC: 667 ultrafine lattices now analyzed, more-or-less on schedule laid out in proposal.
Running on GPUs: double-precision CG code attained 13 GFlop/s/GPU including communication (55 GFlop/s/GPU without communications).

10-2010 Five conference reports based on simulations done with present and previous allocations: "BK with improved staggered fermions: analysis using SU(2) staggered chiral perturbation theory" , "BK with improved staggered fermions: analysis using SU(3) staggered chiral perturbation theory" , "stematic Uncertainties in BK with Improved Staggered Fermions" , "First results for BK on the ultrafine (a=0.045 fm) ensemble" .

August 2010 Long article submitted to Physical Review D (now accepted for publication). Arxiv link here . Result using 3 lattice spacings: BK(NDR,2 GeV)=0.529 +/- 0.009 (statistical) +/- 0.032 (systematic).

7-1-2010 Allocation of 4.8 M J/psi-equivalent core-hours on the QCDOC (equivalent to 20 M QCDOC-node hours) and 75 K GPU-hours on the 9g cluster at JLab begins.

5-1-2010 Presentation on proposal at USQCD All Hands meeting, April 16th 2010.

3-1-2010 Renewal proposal submitted.

1-21-2010 Update on running (now on 4096 node partition): 40^3x96 fine lattices---completed; 48^3x144 superfine lattices---740x2 measurements completed; 64^3x192 ultrafine lattices---142 so far.

10-2009 Four conference reports based on simulations done with the present allocation: "Determination of B_K using improved staggered fermions " "(I) SU(3) chiral perturbation theory fit" , "(II) SU(2) chiral perturbation theory fit" , "(III) Finite-volume effects" , "(IV) One-loop matching" .

3-9-2009 See 2009-10 proposal above for some analysis and for running plans. At present the 512 node partition is calculating BK on the 40^3x96 fine lattice (102 configurations completed), while the two 2048 node partitions are calculating BK on the 48^3x144 superfine lattices (457 configurations are completed).

1-23-2009 After the increase in allocation, we requested move from 512 nodes to 4096 node partition. Initially, this partition had hardware problems. It was then made into two 2048 node partitions, one of which has worked reliably, the other of which has been somewhat unreliable.
At present the 512 node partition is calculating BK on the 40^3x96 fine lattice (< 100 configurations completed), while the 1-2 2048 node partitions are calculating BK on the 48^3x144 superfine lattices (250 configurations are completed).

July 2008 N.B. Original updates from 2008 lost due to accidental deletion of file.
Status as of July 2008 presented at LATTICE 2008. Write-ups available for BK and spectrum . These are based on results from 5 coarse and a single fine lattice. Fitting for BK is done using SU(3) staggered chiral perturbation theory with Bayesian priors.

1-22-08: Matrix element and spectrum calculation on 28^3 x 96 fine lattices (m_l/m_s=0.0062/0.0031) continue: about 400 (out of 512) configurations completed, using both cubic U(1) and cubic wall sources for spectrum calculation. Analysis underway to determine size of taste-breaking in spectrum.
Some coarse lattice data was missing and has been recalculated.
Publication on taste-splitting reduction by HYP smearing submitted to the "arXiv". Published in Physical Review D.

10-31-07: Matrix element and spectrum measurements completed on desired coarse MILC lattice ensembles (m_l/m_s = 0.010/.050, .010/.030, 0.007/.050, 0.020/.050, 0.030/0.050). Analysis underway.
Matrix element and spectrum calculation on 28^3 x 96 fine lattices (m_l/m_s=0.0062/0.0031) underway: about 200 (out of 512) configurations completed. Using both cubic U(1) and cubic wall sources for spectrum calculation.

7-16-07: Matrix element and spectrum calculation updates: 20^3 x 64 m_l/m_s = 0.01/0.03 ensemble complete; 20^3 x 64 0.03/0.05 ensemble almost complete (475 out of 564 configurations); 246 configurations of 24^3 x 64 m_l/m_s = 0.005/0.05 lattices completed.
Coulomb gauge fixing on 28^3 x 96 lattices progressing: 273 (out of 531) 0.0124/0.031 configurations and 260 (out of 573) .0062/0.031 configurations are fixed.

4-19-07: Running on 20^3 x 64 m_l/m_s = 0.01/0.03 ensemble almost complete; starting on 24^3 x 64 m_l/m_s = 0.005/0.05 lattices.

3-22-07: Running on 20^3 x 64 m_l/m_s = 0.01/0.05 and 0.02/0.05 ensembles complete. Speeded up code by dropping one of sink operators in spectrum calculation (as gave redundant information).

2-07: Tests complete. Production begun on five 128 node partitions. Code speeded up by moving propagators into memory to reduce I/O.

1-07: Spectrum test complete. B_K test indicates sporadic corruption by hardware failure---solved by restarting calculation on given configuration if failure. Gauge fixing ongoing.

8-9-06: Production test underway. Calculation of full spectrum and matrix elements, with 55 pairs of quark masses, on a subset of 20^3 x 64 MILC lattices (m_l/m_s = 0.01/0.05) underway on both Seoul National University (SNU) cluster and on QCDOC. If they agree, we are ready to move to a production run on QCDOC. Choice of wall source turned out, in preliminary work at SNU, to be unimportant, and we have chosen a U(1) cubic source.
8-9-06: 210 gauge configurations from MILC ensemble MILC_2064f21b676m010m050 (coarse lattices of size 20^3 x 64 with m_s=0.05 and m_ud=0.01) have been fixed to Coulomb gauge. This was done on 1 mother board (64 node) of QCDOC, run as a test job.Timing 1-3 hrs/lattice.

Background:

2010-11 Proposal submitted to the US lattice QCD Scientific Program Committee, asking for time on the QCDOC machine at BNL, and the Infiniband GPU cluster at JLab. Allocated 4.8 M J/psi-equivalent core-hours on the QCDOC and 75 K GPU-hours on the 9g cluster at JLab begins.

Abstract: This is a class A proposal, requesting time on the QCDOC, and on the Infiniband GPU cluster at Jefferson Lab, to continue our calculation of kaon matrix elements using HYP-smeared improved staggered quarks. Our major focus will be on a calculation of B_K, which is of particular interest for constraining elements of the CKM matrix. Using MILC coarse, fine and superfine lattices, we have a preliminary result, using SU(2) staggered chiral perturbation theory to do the chiral extrapolation and remove some lattice artifacts. The dominant sources of errors are the matching factors (calculated at one-loop) and statistics. The present proposal aims to reduce both of these errors.
The QCDOC component will reduce the error due to matching factors (as well as other errors) by adding a further data point at smaller lattice spacing. We are presently running on the MILC ultrafine lattices (a= 0.045fm, m_l/m_s=0.2), but will not have completed this calculation by the end of the present allocation. We propose to continue using 4096 nodes on the QCDOC until the end of calendar year 2010, in order to complete the running on ultrafine lattices.
The GPU component will substantially reduce the statistical errors on one of the superfine lattices and add a second ensemble with smaller light sea-quark mass. Our proposed calculations run on the Infiniband GPU cluster.
Graduate students and postdoctoral associates at both the University of Washington and Seoul National University will contribute to this project.

We requesting 14.7 Mnode-hours (3.53M Jpsi-equivalent core-hours) on the QCDOC, 120,000 GPU-hours on the GPU cluster with Infiniband, and 540,000 Jpsi-equivalent core hours for storage.

2009-10 Proposal submitted to the US lattice QCD Scientific Program Committee, asking for time on the QCDOC machine at BNL. Proposal granted 3.86 Million 6n-equivalent node-hours, deployed as 25.08 M node hours on the QCDOC and 0.80 M 6n node hours on the Fermilab clusters.

Abstract: This is a class A proposal, requesting time on the QCDOC to continue our calculation of kaon matrix elements using HYP-smeared improved staggered quarks. Our major focus will be on a calculation of B(K), which is of particular interest for constraining elements of the CKM matrix. The present (2008-9) allocation should allow completion of work on the MILC fine lattices (a=0.09 fm) and the m_l/m_s=0.2 superfine ensemble (a=0.06 fm). We give a fairly extensive status report on our analysis, showing that SU(2) chiral fitting gives results with 1-2% errors. Encouraged by this, we propose to extend this work by adding the lighter sea quark superfine lattice ensemble (m_l/m_s=0.1) and by extending the calculation to the m_l/m_s=0.02 ultrafine lattices (a=0.045 fm).
Graduate students and postdoctoral associates at both the University of Washington and Seoul National University will contribute to this project.
We are requesting 29.5 Mnode-hours on the QCDOC, but are also able to move some of the running to the clusters if needed.

2008-9 Proposal submitted to the US lattice QCD Scientific Program Committee, asking for time on the QCDOC machine at BNL. Proposal was initially granted 3.96 Mnode-hours on the QCDOC, which was increased to 12 Mnode-hours after rebalancing due to additional time becoming available to the USQCD collaboration as a whole.

Abstract: This is a class B proposal, requesting time on the QCDOC to continue our calculation of kaon matrix elements using HYP-smeared improved staggered quarks. Our major focus will be on a calculation of BK, which is of particular interest for constraining elements of the CKM matrix. The present (2007-8) allocation should allow completion of work on most of the MILC fine lattices. We give an fairly extensive status report on our results, so as to motivate its continuation. We propose to extend this work in two ways: first, to increase statistics on the MILC fine lattices with the lightest sea quark; and, second, to extend the calculation to superfine lattices.
Graduate students and postdoctoral associates at both the University of Washington and Seoul National University will contribute to this project.
We are requesting 4.1 Mnode-hours on the QCDOC.

2007-8 Proposal submitted to the US lattice QCD Scientific Program Committee, asking for time on the QCDOC machine at BNL. Includes preliminary results showing reduction in taste breaking for HYP-smeared fermions compared to asqtad-fermions. Proposal was granted 3.25 Million node-hours on the QCDOC

Abstract: This is a class B proposal, asking for time on the QCDOC to continue our calculation of kaon matrix elements using improved staggered quarks. Our major focus will be on a calculation of $B_K$, which is of particular interest to constraining elements of the CKM matrix, and on the related matrix elements $B_7^{3/2}$ and $B_8^{3/2}$, which enter into the prediction for $\epsilon'/\epsilon$. The present (2006-7) allocation should allow completion of work on the MILC coarse lattices, and possibly a start on the fine lattices. We propose to extend this work to all the MILC fine configurations. The valence propagators will be HYP-smeared staggered fermions. On each lattice we will calculate with 10 different valence quark masses, allowing 55 kaon masses, so as to provide sufficient data points for fitting to the rather elaborate form predicted by staggered chiral perturbation theory. We propose in parallel to calculate the spectrum determining in particular the masses of flavor non-singlet pions of all tastes. This information is needed as input into the chiral perturbation theory fits to B_K, B_7 and B_8 and also will provide cross-checks on the applicability of staggered chiral perturbation theory. We expect that graduate students, and possibly also postdoctoral associates, at both the University of Washington and Seoul National University will contribute to this project.
We are requesting 3.35 Mnode-hours on the QCDOC.

2006-7 Proposal submitted to the US lattice QCD Scientific Program Committee, asking for time on the QCDOC machine at BNL. Proposal was granted 1.2 Million node-hours on the QCDOC.

Abstract: We propose to calculate kaon matrix elements of four-fermion operators using improved staggered quarks. Our major focus will be on a calculation of B_K, which is of particular interest to constraining elements of the CKM matrix, and on the related matrix elements B_7^(3/2) and B_8^(3/2), which enter into the prediction for epsilon'/epsilon. We propose to use the MILC coarse and fine Asqtad configurations, and calculate valence propagators using HYP-smeared staggered fermions. On each lattice we will calculate with 10 different valence quark masses, allowing 55 kaon masses, so as to provide sufficient data points for fitting to the rather elaborate form predicted by staggered chiral perturbation theory. We also propose a parallel spectrum (and possibly also decay constant) calculation, determining in particular the masses and decay constants of flavor non-singlet pions of all tastes. This information is needed as input into the chiral perturbation theory fits to B_K, B_7 and B_8 and also will provide cross-checks on the applicability of staggered chiral perturbation theory. The output of the proposed calculation will be a result for B_K and other kaon matrix elements with all errors controlled.
We are asking for CPU time on the QCDOC, since our codes are built on the Columbia Physics System, and use the level-3 QCD-API routines for the QCDOC. We are asking for a total of 1.2 Million node-hours, so that this is a ``class B'' proposal.
We expect that graduate students, and possibly also postdoctoral associates, at both the University of Washington and Seoul National University will contribute to this project.