dolfin team mailing list archive

[DOLFIN <dolfin@xxxxxxxxxx>]

One or more new changesets pushed to the primary DOLFIN repository.
A short summary of the last three changesets is included below.

changeset:   1891:abcc54c4a3f446f20c29cf8eb032bbf5f9fbaa19
tag:         tip
user:        "Anders Logg <logg@xxxxxxxxx>"
date:        Sat Apr 22 15:41:44 2006 -0500
files:       src/bench/ode/reaction/bench-domain.log src/bench/ode/reaction/bench-domain.py src/bench/ode/reaction/bench-tol.log src/bench/ode/reaction/bench-tol.py src/bench/ode/reaction/benchutil.py src/bench/ode/reaction/main.cpp src/bench/ode/reaction/parameters-bench.xml
description:
Recompute benchmarks with improved adaptivity.

Results look very good:

bench-tol: cg/fixed-point
---------------------------------

tol     N       Error           CPU time        Steps   Iterations      Index
--------------------------------------------------------------------------------------
1.0e-06 1000    2.345e-05       28.050          117089 (1)      4.000 (0.000)   1.0
5.0e-07 1000    1.173e-05       39.520          165586 (1)      4.000 (0.000)   1.0
1.0e-07 1000    2.344e-06       71.940          370254 (1)      3.000 (0.000)   1.0
5.0e-08 1000    1.172e-06       101.730         523615 (1)      3.000 (0.000)   1.0

bench-tol: mcg/fixed-point
---------------------------------

tol     N       Error           CPU time        Steps   Iterations      Index
--------------------------------------------------------------------------------------
1.0e-06 1000    1.771e-05       14.240          1922 (5)        3.990 (1.498)   95.266
5.0e-07 1000    1.091e-05       23.300          1912 (9)        4.822 (1.544)   138.240
1.0e-07 1000    1.921e-06       48.050          1929 (7)        4.905 (1.594)   142.619
5.0e-08 1000    8.979e-07       49.770          1917 (7)        4.131 (1.680)   172.388


bench-domain: cg/fixed-point
---------------------------------

tol     N       Error           CPU time        Steps   Iterations      Index
--------------------------------------------------------------------------------------
1.0e-06 1000    2.345e-05       28.100          117089 (1)      4.000 (0.000)   1.0
1.0e-06 2000    2.233e-05       64.840          117091 (1)      4.000 (0.000)   1.0
1.0e-06 4000    2.166e-05       101.290         117090 (1)      4.000 (0.000)   1.0
1.0e-06 8000    2.219e-05       175.110         117089 (1)      4.000 (0.000)   1.0
1.0e-06 16000   2.242e-05       327.680         117089 (1)      4.000 (0.000)   1.0

bench-domain: mcg/fixed-point
---------------------------------

tol     N       Error           CPU time        Steps   Iterations      Index
--------------------------------------------------------------------------------------
1.0e-06 1000    1.771e-05       13.640          1922 (5)        3.990 (1.498)   95.266
1.0e-06 2000    1.680e-05       17.310          1923 (5)        3.991 (1.154)   140.461
1.0e-06 4000    1.643e-05       24.000          1920 (6)        3.991 (1.004)   184.969
1.0e-06 8000    1.691e-05       33.670          1918 (5)        3.991 (1.004)   218.782
1.0e-06 16000   1.697e-05       57.850          1919 (5)        3.990 (1.004)   239.992


Newton benchmarks are temporarily disabled: takes longer
time anyway and there seems to be a bug in the multi-adaptive
Newton solver (takes more iterations than fixed point). Also
changed kmax to 1e-3 in bench-domain (works better with new
choice of threshold). kmax is now same for both benchmars and
set in parameters-bench.xml.


changeset:   1890:56b166d515bb78965ca1a7c677038d00ff274ec4
parent:      1889:fe2b9eb261e0862c2584aff1fdb8fdd51d309cf1
parent:      1888:11dda674298121321251da49a0b9c92821cb748e
user:        "Anders Logg <logg@xxxxxxxxx>"
date:        Sat Apr 22 05:08:28 2006 -0500
files:       
description:
Merge


changeset:   1889:fe2b9eb261e0862c2584aff1fdb8fdd51d309cf1
parent:      1886:3671f5761d18bbbf3c32d87876e41b8ba3ea45d5
user:        "Anders Logg <logg@xxxxxxxxx>"
date:        Sat Apr 22 05:07:03 2006 -0500
files:       src/bench/ode/reaction/bench-tol.log src/bench/ode/reaction/bench-tol.py src/bench/ode/reaction/parameters-bench.xml src/kernel/ode/MultiAdaptivity.cpp src/kernel/ode/dolfin/MultiAdaptiveTimeSlab.h src/kernel/ode/dolfin/MultiAdaptivity.h src/kernel/ode/dolfin/ODE.h
description:
Implement new strategy for conservative modification of time steps.

This is a poor man's dual weighting function. For each component,
we look at all other components that the component in question
depends on and tell these other components to reduce their time
steps to the same level. A similar effect would be accomplished
by weighting with the proper weight from the dual solution, but
until we (re-)implement the solution of the dual, this seems to
be a good solution.

Also changed the partitioning threshold from 0.2 to 0.5 for benchmark.

This further speeds up the multi-adaptive benchmark and restores the
the proportionality e ~ tol.


-------------------------------------------------------
For more details, visit http://www.fenics.org/hg/dolfin