Recent Progress on ISAT Stephen B. Pope January 25, 2007.

34
Recent Progress on ISAT Stephen B. Pope January 25, 2007

Transcript of Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Page 1: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Recent Progress on ISAT

Stephen B. Pope

January 25, 2007

Page 2: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Research Group Meetings

• Practice talks– e.g. for upcoming Combustion Institute

meeting

• Report and discussion of recent research

Page 3: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

In Situ Adaptive Tabulation

• Particle methods– 1010 particle steps– Reaction sub-step: direct integration 104 s– 108 s = 3.2 years– X 1,000 speed-up -> 28 hrs

• ISAT– Tabulate results of ODE integration

Page 4: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

ISAT-CK 5.1• Primary retrieve

– BT, MRU, MFU

• Secondary retrieve– EBT in affine space – can afford to be complete

• Grow– EOA/EOI– EBT

• Add– EOI initialized to exclude existing points

Page 5: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Very Recent Additions

• de_nearby– Location (x) of direct evaluation may be

slightly different from the query location– Needed for ICE-PIC/ISAT

• “Slow progress”– Query time > DE time– Reduce growing– Worst case: query time = 2 x DE time

Page 6: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Preliminary Testing of ISAT-CK 5.1

• Systematic testing to be performed by Liuyan

• Study performance as function of parameters– Error tolerance– Available storage

Page 7: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YF: varying error tolerance

• Skeletal mechanism (16 species)– CPU for DI =5,000 s

• Premixed

• Storage = 1.2 Gbytes

• Idites=1000 (affects performance)

• Q=107

• Etol=2-p, p=1,2,…,32

Page 8: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 9: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 10: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

CDF of error for etol=1.2e-4

Page 11: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YF: observations

• Error well controlled down to 1e-8: confirmed that errors due to ODE

• Timing affected by error testing

• Speed-up on table filling

Page 12: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YH: varying number of leaves

• Skeletal mechanism (16 species)– CPU for DI =5,000 s

• Premixed

• Q=107

• Etol=10-4

• Leaves = 16, 32, 64,…,2048.

Page 13: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 14: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 15: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YH: observations

• As storage decreases, query time tends to DE time

• “Slow progress” avoids excessive CPU time– “slow progress” if query time > DE time– Ret_frac set to ½– Grow_frac set to ½– EOIs not shrunk

Page 16: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YA: varying error tolerance

• GRI 3.0 mechanism (53 species)– CPU for DI =80,000 s (x16 cf skeletal)

• Premixed

• Storage = 1.2 Gbytes

• Idites=1000 (affects performance)

• Q=108

• Etol=2-p, p=1,2,…,19

Page 17: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 18: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 19: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YA: Observations

• Error well controlled

• Even for smallest etol, retrieves dominate in the end; but it takes longer to reach retrieving asymptote

Page 20: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test K: Dim. Of Affine Space

• Skeletal mechanism (16 species)

• Premixed

• Storage = 1.0 Gbytes

• Idites=1000 (affects performance)

• Q=1e7

• Etol=1e-3

• Affine space, na=1,2,3,…15

Page 21: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 22: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test M: Pair Cover

• Skeletal mechanism (16 species)• Premixed• Storage = 2.5 Gbytes• Idites=1000 (affects performance)• Q=1e7• Etol=1e-4• Pair_cover: type of ellipsoid used to cover two

other ellipsoids (in EBT)– 1 – spheroid (no shrinking)– 2 – ellipsoid (covariance algorithm)– 3 – ellipsoid (iterative algorithm)– 4 – spheroid (with shrinking)

Page 23: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 24: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YJ: storage and etol

• Skeletal mechanism (16 species)

• Premixed

• Idites=10,000 Q=1e7

• Etol = 4-p = 1, ¼, 1/16,…,1.5e-5

• Leaves = 4q = 16, 64, 256,…,16,384

Page 25: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 26: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 27: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 28: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 29: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Test YI: varying etol

• Skeletal mechanism (16 species)– CPU for DI =5e3 s

• Non-premixed, variable time step

• Idites=10,000 Q=1e8

• Storage = 1.2 Gbytes

• etola = 1, 1/2,...3e-5

Page 30: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 31: Recent Progress on ISAT Stephen B. Pope January 25, 2007.
Page 32: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Future Plans: ISATAB

• Single precision for ellipsoids

• x2f_mpi

• ISAT/ANN

Page 33: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Future Plans: ISAT-CK

• ICE-PIC

• CEQ in place of Stanjan

• ADIFOR for Jacobians

• LP for realizability

• Conservation correction for sensitivity matrix

Page 34: Recent Progress on ISAT Stephen B. Pope January 25, 2007.

Future Plans: papers

• Pope (2007) – new algorithms

• Lu & Pope (2007) – performance of ISAT

• Lu, Lantz, Ren, Pope (2007) – x2f_mpi

• Ren & Pope (2007) – ICE-PIC/ISAT