EM reflection off Silver-Muller boundary

[alexanderboe]

Hi all,

I am using a Silver-Muller boundary condition in a 2D simulation, to absorb a laser pulse (800 nm, a0 = 0.6), incident at a normal to the boundary surface.

I observe that ~1/1000th of the incident electric field is reflected. This reflection back-propagates through the simulation domain, and seeds high-frequency perturbations of the electron density, that I would like to avoid.

The simulation resolution along the laser propagation is 15 cells per cycle (800 nm), the time-step is set to 0.95*CFL. I can replicate the reflection in 1D, and in an entirely empty simulation domain.

Is it possible to optimise the Silver-Muller boundary condition to reduce the amount of reflected light, possibly by altering parameters in the boundary implementation?
Enabling control of the Silver-Muller BC was discussed here #9, but as of now does not appear to be included in the code documentation.

Any thoughts or advice would be very helpful!

Alexander

[beck-llr]

Hi Alexander

The current Silver-Muller BC implementation is the best one we are aware of. An improvement of these absorbing boundaries would be the "perfectly matched layer" (PML) which are not yet implemented in Smilei due to a lack of manpower. We hope to have them within 18 months.

Some tricks are currently possible to minimize reflections of the laser pulse on the transverse boundaries (where the wave vector of the laser is parallel to the boundary). But if you face problems with the normal incidence I'm afraid there is nothing to do for the moment without changing your global setup.

If the physics of your problem allows it, you can use a longer simulation domain, possibly empty of any particles, in order to let the laser propagate further away and delay the moment at which the reflection impacts your results. In a 2D simulation this should not be too costly. The required additional length depends of course of the total simulation time you need.

Hope this helps.
Arnaud

[alexanderboe]

Hi Arnaud,

Many thanks for clarifying this and the advice. I will see if I can sufficiently delay the reflection by appending an empty region to the simulation.

All the best,

Alexander

[BenK3nobi]

Hello despite the closed issue:
Are there any news about the PML implementation?
I get similar issues as mentioned above and am currently looking for a solution.

[beck-llr]

Hi
The implementation of PML is pretty advanced. We still have to work on it to make it compatible with the dynamic load balancing and the moving window. At that point we will probably make a first release. The results are quite promising so far.
Cheers

[BenK3nobi]

Perfect, thanks a lot for the update.
Does a timeframe for a release already exist?

[beck-llr]

I'm sorry but no. Hopefully before March but we can not guarantee anything.

[Tissot11]

I was wondering if the PML is now implemented in SMILEI?

[beck-llr]

PML are implemented in a developing branch for all geometries. It is not released yet because we are still working on PML for the envelope model which is a pretty tricky problem.
Computational performances of the PML have not been considered yet either.

[Tissot11]

How to get access to this developing branch? I only see four branches available; master, gh-pages,prescribed_fields and time_frozen... I'm interested in absorbing the electromagnetic fluctuations at the boundaries and currently with SM, I sometime see issues in my PIC simulations. So I would be keen to try PML. Also how to enable these in the Namelist?

[beck-llr]

I am sorry but you can not access this branch. It is dedicate to work in progress. When the feature is released, the documentation will be updated and will answer your questions. These things are not well defined yet.
We hope to release an experimental version of the PML pretty soon. At that point we'd be glad to have you do some tests and give us feedback.
Cheers