Title: Early experiences with DWF on QCDOC
1Early experiences with DWF on QCDOC
This is a physics talk
Only possible due to a huge amount of work from
all the people on the QCDOC project
Physics contributions from many RBC and UKQCD
people
All results are very preliminary
2Introduction
- Chirally symmetric fermions
- Domain Wall fermions (DWF)
- Both chiral and flavour symmetry at finite a
- Continuum-like chiral perturbation theory
- No flavour mixing, wrong chirality operating
mixing small - Dynamical DWF is QFT
Kaplan 92, Shamir 93, Furman-Shamir 95
- Computationally expensive
- Residual chiral symmetry breaking mres
3Outline
- Chiral symmetry breaking mres
- Preliminary results (statistics limited) for
- The effect of the gauge action on mres
- Valence M5 and LS. The potential
- The effect of the gauge coupling on mres
- The effect of the quark mass on mres
- The effect of LS on mres
- Other talks
- Algorithm (MAC) and Machine (PAB)
- Summary and outlook
4Chiral symmetry breaking and mres
- 4d quark field constructed from left (right)
projections of on boundary
Quark mass is
LS not infinite ? L-R coupling Define J5 current
in terms of fields at LS/2 - (1,LS) Axial
Ward-Takahashi Identity and thus mres
5Small is beautiful
- How to reduce chiral symmetry breaking
- Desire mres as small as possible
- LS ? infinity is expensive
- Overlap Ultimate operator. Expensive
- See BJ talk GWF and cost
- We are keen to adopt these improvements
- How do simulation parameters effect mres
- Gauge action, M5, LS, gauge coupling mf
- Inform decision on Big Run
6Gauge Action
Measure dependence of mres on rectangle
term Quenched studies show lines of constant a
approx quadratic in b,c1/1-8c1 plane Zoom in
to top left
Keep a constant vary c1 and b
7Ensembles RBC
- Nf3 amf0.04, t0.5 dt 0.01 aM51.8
b c1 Mud/ms LS Alg Traj/cfg
0.72 -1.4069 1 8 R 1500 / 96
0.48 -2.300 1 8 R 1605 / 101
0.32 -3.570 1 8 R 1600 / 101
0.16 -7.470 1 8 R 1520 / 102
0.80 -1.4069 1 8 R 1900 / 126
0.53 -2.300 1 8 R 1400 / 81
8Residual mass
Nf 3 amf 0.04 LS8 DBW2
9Mres versus rectangle
Increase rectange, decrease mres No dramatic
decrease in mres Lattice artefacts scaling
behaviour? Conservative opinion Maybe not worth
it
10The potential
- Compare lattice spacing from mr and r0
- Look for lattice artefacts
- ms0.04 mv-mres
- No sea extrapolation
- No strong sea quark mass dependence
11NF2 sea quark mass dependence
12Scaled potential
- Rescaled potential
- NF0,NF2 different short distance behaviour
- No obvious differences for c1
13Lattice artefacts
- Some variation with c1
- Approx 10
- This suggests it is not an extreme effect
14Valence LS
- LSsea8
- mres decreases as LS increases
- Not exponetial
- Not surprising as PQ
- Same as quenched
15Ensembles UKQCD
b c1 Mud/ms LS Alg Traj/cfg
0.72 -1.4069 1 8 R 1500 / 70
0.72 -1.4069 0.5 8 R 1345 / 51
0.72 -1.4069 1 8 RHMC 1200
0.72 -1.4069 1 12 RHMC lt500
0.72 -1.4069 0.5 12 RHMC 1200
0.72 -1.4069 0.3 12 RHMC lt500
0.88 -1.4069 1 8 R 1500 / 75
0.88 -1.4069 0.5 12 RHMC lt500
16mres vs mf
b 0.72 LS 8 R algorithm mval msea No
strong mq dependence No fit. Straight line
through two points
17(amPS)2 vs amq
- Intercept not equal to mres
- Straight line through two points
- Autocorrelation?
- Statistics (50)
18amr vs amq
- msea mval
- Straight line through two points
- amr at mq - mres
- a-1 1.4(3) GeV
19amr vs amf valence
- Valence only extrapolation
- a-1 1.45 GeV
20mres vs a2
- Lattice spacing PQ r mass
- mres PQ extrapolation
- Sea quark mass not same
- Approx exponential behaviour
Preliminary
21Summary and Outlook
- First physics from QCDOC!
- Very preliminary
- We are exploring the effects of simulation
parameters on residual chiral symmetry breaking - Accumulating more data
- Some conclusions soon!