Difference between revisions of "Overlap eigensystem tests"

From Gw-qcd-wiki
Jump to: navigation, search
(323×64 DSDR configuration comparison of 1, 2 and 4 time p-HYP smearing kappa = 0.2, B.C=111-1 (Test on Hopper))
(323×64 DSDR configuration comparison of 1, 2 and 4 time p-HYP smearing kappa = 0.2, B.C=111-1 (Test on Hopper))
Line 118: Line 118:
  
 
==32<sup>3</sup>&times;64 DSDR configuration comparison of 1, 2 and 4 time p-HYP smearing kappa = 0.2, B.C=111-1 (Test on Hopper)==
 
==32<sup>3</sup>&times;64 DSDR configuration comparison of 1, 2 and 4 time p-HYP smearing kappa = 0.2, B.C=111-1 (Test on Hopper)==
[[File:1hyp.png|left|410px]] [[File:2hyp.png|center|410px]] [[File:4hyp.png|right|410px]]
+
[[File:1hyp.png|left|400px]] [[File:2hyp.png|center|400px]] [[File:4hyp.png|right|400px]]
 
===eigensystem===
 
===eigensystem===
 
{| class="wikitable" style="text-align: center"
 
{| class="wikitable" style="text-align: center"

Revision as of 16:53, 31 October 2011

Eigensystem tests

Boundary conditions: 0 = Dirichlet, 1 = Periodic, -1 = Antiperiodic

243 × 64 dynamical configuration, \(a = 0.114 {\rm fm}\)

Configuration: rbc_conf_2464_m0.005_0.04_002055_hyp (Domain Wall Action)

Hw 2464 compare.png

Number of Cores: 1024

B.C. (x y z t) 1 1 1 0 0 1 1 0
Hw Eigenvectors # 300 300
Polynomial Order (10-9 error) 226 463
Dov Eigenvectors(pair) # 100 100
Dov Precision 10-7 10-7
# Zero Modes 2 3
# Overlap Iterations 6250 5950


323 × 64 dynamical configuration, \(a = 0.085 {\rm fm}\)

Configuration: rbc_conf_3264_m0.004_0.03_002190_hyp (Domain Wall Action)

Number of Cores: 2048

B.C. (x y z t) 1 1 1 0 1 1 1 -1
Hw Eigenvectors # 300 200
Polynomial Order (10-9 error) 284 118
Dov Eigenvectors(pair) # 100 400
Dov Precision 10-7 10-9
# Zero Modes 7 6
# Overlap Iterations 8500 14200


323 × 64 dynamical configuration (new), \(a = ?{\rm fm}\)

Configuration: rbc_conf_3264_m0.004_0.03_test_hyp (Domain Wall Action)

Number of Cores: 8192 (But one should resort to only 4096 due to scaling)

B.C. (x y z t) 1 1 1 0 1 1 1 -1
Hw Eigenvectors # 400 400
Polynomial Order (10-9 error) 395 215
Dov Eigenvectors(pair) # 100 100
Dov Precision 10-7 10-7
# Zero Modes Not Known 15
# Overlap Iterations Did not converge in 24hrs 34900



283 × 56 pure gauge configuration, \(a = 0.15 {\rm fm}\)

Configuration: g282856iw_beta2.34607_194 (Iwasaki Action)

Number of Cores: 1792

# nyhp smears 1 2 3
B.C. (x y z t) 1 1 1 0 1 1 1 0 1 1 1 0
Hw Eigenvectors # 270 270 270
Polynomial Order (10-9 error) 379 297 256
Dov Eigenvectors(pair) # 100 100 100
Dov Precision 10-7 10-7 10-7
# Zero Modes 10 9 14
# Overlap Iterations 26158* 21700 24250


323×64 DSDR configuration comparison of 1, 2 and 4 time p-HYP smearing kappa = 0.2, B.C=111-1 (Test on Hopper)

1hyp.png
2hyp.png
4hyp.png

eigensystem

4 smearings on 1024 cores 1 smearing on 4096 cores 2 smearing on 4096 cores
# of Hw eigenvectors 100 350 350
largest Hw eigenvalue computed 0.2217 0.1106 0.2012
Hw timings 572 sec 616 sec 547 sec
polynomial order for 10-10 error 126 251 139
# of Dov eigenvectors 10-8 error 613 (pair) 613 (pair) 613 (pair)
Chebyshev order/cut 12/0.01 12/0.01 12/0.01
# Arnoldi iterations 2 2 2
Timings for overlap convergence 12.4 hours 7.2 hours 4.2 hours
Topological charge 17.00 14.52 15.82
# of zero modes 17 15 15
largest eigenvalue of Dov +9.09095117521e-03 +1.64894536348e-0 6.2930786766e-03 +i 1.3725754329e-01 7.0716480399e-03 +i 1.4548173739e-01

323×64 DSDR configuration comparison of 1, 2, and 4 time p-HYP smearing kappa = 0.2, B.C.=111-1 (Test on Hopper)

propagator inversions

ma = 0.0015, 0.0034, 0.0056, 0.01, 0.02, 0.03, 0.06, 0.07, 0.08, 0.09

4 smearings on 1024 cores 1 smearing on 1024 cores 2 smearing on 1024 cores
# of Hw eigenvectors 100 350 350
largest Hw eigenvalue computed 0.2217 0.1106 0.2012
polynomial order for 10-10 error 126 251 139
# of Dov eigenvectors 10-8 error 613 (pair) 613 (pair) 613 (pair)
convergence criteria 10-8 10-8 10-8
avg # of iterations per inversion 163 189 180
total inversions time 0.54 hours 1.26 hours 0.80 hours

(* Locking was on during this run, which might have affected the number of iterations. Locking was off for all other runs) --