k-Wave

1. magnus
   

   Member
   Joined: Oct '20
   Posts: 2

   Hi,

   I'm using pstdElastic2D for some simulations of a lamb wave propagating in a plate. I'm a bit puzzled by the effect of the 'PMLSize' input argument. If explicitly stated to a value eg 10, I do get a reasonable result. However, if I don't include this argument, I get a completely different behavior of the propagating lamb wave: higher propagation velocity and lower amplitudes (more damping with longer distance traveled.) According to the documentation, the default value of 'PMLSize' is also 10, so results should be identical. I have not found a way to read back the 'PMLSize' parameter so I'm not 100% sure what is actually used in the simulation if not stated explicitly.
   Is there a way to read back all the parameters used in the simulation?
   Is there any explanation for the behavior observed?

   Best regards
   Magnus

   Posted 3 years ago #

2. Andrea Addison
   

   Member
   Joined: Jan '20
   Posts: 10

   Hi,

   I want to add another question that I believe related to this topic.
   How we should decide the "PML_size" grid size in pstdElastic2D? In general, is 5 points per wavelength will be sufficient for full absorption in PML?

   Thank you for your time and effort,
   Andrea

   Posted 3 years ago #

3. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,643

   @Magnus: In 2D, the default size is 20 grid points. Worth checking that things you care about are not within the PML. There is no straightforward way to see the parameters, but you can see the defaults in /private/kspaceFirstOrder_setDefaults.

   @Andrea: 5 points seems a little slim to me. I normally use 10 as a minimum, preferably more. The performance depends on the frequency content relevant to the grid spacing, and also the angle of incidence. There are some details for the fluid code in this paper that might be interesting, as well as some plots in the k-Wave user manual.

   Brad

   Posted 3 years ago #

4. Andrea Addison
   

   Member
   Joined: Jan '20
   Posts: 10

   Hi Bradley,

   Thank you for your answer.

   Best regards
   Andrea

   Posted 3 years ago #

5. Chris Wang
   

   Member
   Joined: Jun '21
   Posts: 1

   Hi Brad,

   Recently I started to simulate shear wave propagation in tissue-mimicking phantoms by using the "pstdElastic2D" function. I am simulating a uniform phantom (100 mm x 100 mm) with a sinusoidal velocity source on the left side. I thought that in this situation the shear wave only propagates from left to right and therefore PMLs are only needed on the left and right sides but not needed on the top and bottom sides. (a) When I set the PMLSize to be zero on the top and bottom sides, I got uniform vibration amplitudes in the entire domain, which is expected. On the other hand, it is written in the k-Wave manual that "To remove the PML, set the appropraite PMLAlpha to zero rather than forcing the PML to be of zero size. (Page 63)" (b) When I set the PMLSize to be nonzero but PMLAlpha = 0 on the top and bottom sides, I got different results from those in (a). This leads me to the following question: How should the PMLs be set up in k-Wave? For example, should PMLs always be set up on all sides or only on the sides in the wave propagation direction? Why "PMLSize = 0" is not equivalent to "PMLSize=/=0 & PMLAlpha = 0"?

   Additionally, I would like to ask whether it is correct to set the compression wave speed (c_p) to be a small (nonphysical) value (e.g., 0.001 m/s) to reduce the compute time, if the focus is on the shear wave (I used c_s = 4.59 m/s). I tried simulations with different c_p values and found that the wave image results with c_p = 1000 m/s and c_p = 0.001 m/s seems nearly the same (for shear wave propagation in a uniform phantom). As far as I know, c_p should be quite different (either much larger or much smaller) from c_s to avoid numerical errors. I also noticed that the PMLs worked fine in attenuating the shear wave when c_p = 1000 m/s but did not work when c_p = 0.001 m/s. Is this behavior of the PMLs expected?

   Thank you very much and I look forward to your reply.

   Best regards,

   Chris

   Posted 3 years ago #

6. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,643

   Hi Chris,

   Regarding the PMLs, the elastic code uses a multi-axial PML, where the PML is also applied a small amount in the other direction. The "amount" is set by the flag 'MultiAxialPMLRatio' (set it to zero to turn it off). I suspect this might the reason the two are different. This is different to the PML used in the fluid code.

   If the medium is homogeneous, there is no conversion between the compressional and shear waves, so in principle the equations modelling the two wave types are completely decoupled. In practice though, they are being modelled by the same code, and making cp smaller than cs (which is not physical) might lead to strange effects.

   Hope that helps,

   Brad.

   Posted 3 years ago #

7. songrl
   

   Member
   Joined: May '22
   Posts: 1

   Hi Bradley,

   Why multi-axial PML is used in elastic code? What are the advantages over the PML used in the fluid code?
   Would you please recommend some publications that discussed this problem?

   Best Regards,

   Ruolong.

   Posted 2 years ago #

RSS feed for this topic

Reply

You must log in to post.