use soft IE for open geometries and WS

[jackhamel16]

Need to get some interesting results (positive time delays) using the soft IE for the paper. Eric wants two interesting geometries. I will start with the pill with a hole in it and no thickness to the walls (open geometry).

[jackhamel16]

I first tried the pill with a hole in it, but I wasn't able to show positive time delays. Attached are the files I ran and results
pill_with_hole1.zip

[jackhamel16]

Finer mesh pill with many harmonics

I ran a pill with no holes with lambda=4 and was not able to see positive time delays. I will do a finer mesh and run overnight a big problem to hopefully see some positive delays with the hole geometry.

GMRES is failing to converge when building the scattering matrix for this test. I will go back to the previous problem, but add more harmonics.

Back to previous mesh with more harmonics

With 256 harmonics used, I was able to see a positive time delay for the 256th harmonic! The delay = 0.0078265 - 1.5475e-8im.
The associated data is here pill_with_hole_first_pos_delay.zip

To ensure this isn't a fluke, I am going to run with 361 harmonics.

The time delay plot for this test is

Only the last delay, number 361, is positive and its value is 0.007827 - 1.548e-8im. All delays prior to this one approach zero.

I plotted the surface sources magnitudes when solving with the excitation corresponding to mode 361 and are shown below

pill_with_hole_first_pos_delay2.zip

Try refining around slot

The refined mesh isnt converging. Try running a simple test with the mesh before bring WS into the fold again.

Hold on refining near slot for now.

Two positive delays

I made the slot one wavelength in height and can see two positive delays. Below is a plot of the surface source magnitudes for mode 256 (the highest mode in the test)

Below is a plot of the WS time delays:

pill_with_hole_first_pos_delay3.zip

more elements, lower lambda

I added more elements to run at lambda=1 and am still seeing two delays that are positive.
pill_with_hole_first_pos_delay4.zip

decrease number of harmonics

Maybe I can get away with less harmonics to speed up run times. I went down to max_l = 10 and number of harmonics = 121 and still saw the same two positive delays. I will try going even lower.

Looks good with 64 harmonics. going to go even lower down to 36.

I should run for this mesh somewhere between 64 and 121 harmonics. The image below shows from left to right the time delays when 36 harmonics were used up to 256 harmonics.

You can see that the last two delays are converged once we hit 121 harmonics. I will use 100 harmonics going forward for this mesh. The associated files are here
pill_with_hole_first_pos_delay5.zip
pill_with_hole_first_pos_delay7.zip
pill_with_hole_first_pos_delay6.zip

[jackhamel16]

ACA Settings test

for big slot with 6735 elements these are the results for solving soft IE with ACA with 3 levels:

Parsing mesh in: open_bigslot_cyl_r1_h2_6735.msh...
Number of elements = 6494
Excitation: Sphericalwave
Equation: Sound Soft IE
Running with ACA...
Filling ACA Matrix...
Matrix fill time: 360.303204101
Filling RHS...
RHS fill time: 1.1527248
Solving with ACA...
GMRES Converged after 1104 iterations.
Solve time: 47.550233899
Displaying ACA Metrics:
Octree Metrics:
Number of elements per node:
Mean = 202.9375
Min = 1
Max = 432
Matrix Metrics:
Matrix Rank:
Mean = 11.531976744186046
Min = 1
Max = 37
Compression Metrics:
Percentage of Matrices Compressed = 67.1875%
Compression Ratio = 0.6053453762583338
Total Runtime = 423.3818763 seconds

And with 4 levels:
Parsing mesh in: open_bigslot_cyl_r1_h2_6735.msh...
Number of elements = 6494
Excitation: Sphericalwave
Equation: Sound Soft IE
Running with ACA...
Filling ACA Matrix...
Matrix fill time: 338.6123349
Filling RHS...
RHS fill time: 1.1492743
Solving with ACA...
GMRES Converged after 1104 iterations.
Solve time: 59.7059353
Displaying ACA Metrics:
Octree Metrics:
Number of elements per node:
Mean = 52.37096774193548
Min = 1
Max = 80
Matrix Metrics:
Matrix Rank:
Mean = 11.50546875
Min = 1
Max = 23
Compression Metrics:
Percentage of Matrices Compressed = 91.57127991675338%
Compression Ratio = 0.5330104289961244
Total Runtime = 413.6697744 seconds

So 4 levels is a littel faster

[jackhamel16]

Open cap pill

I will test a new geometry that will have an opening 2 wavelengths in size. looking like the following:

It is the same pill as before but no slot and with one spherical cap removed and replaced by an opening. so the radius is 1 and height of cylinder is 2. With ~4700 elements, it should support lambda = 1.

I will test this next to see how it performs.

Below are the time delays for the above geometry when lambda=1

This looks great, there are 8 time delays greater than 0 in this example.
pill_onecap1.zip

[jackhamel16]

Parallel Plates

Testing says I should use 4 octree levels.

I ran with lambda=0.7 and 100 harmonics and am seeing multiple positive delays

The surface source magnitudes for mode 100 are

.
parallel_plates1.zip

[jackhamel16]

Which modes to include in paper:
1, 24, (something like 120 but without osciallations), 884, 1280, 1296