Register to reply

COMSOL-nanoparticles-PML-scattering formalism in RF module

by Srbasket
Tags: formalism, module
Share this thread:
Srbasket
#1
Apr10-09, 08:45 AM
P: 19
Hi, all,
I am starting with calculating scattering cross sections of metallic nanoparticles. I have some doubt about what type of scattering boundaries conditions and PML to use, since near field shows some surface waves on PMLīs inner side and some reflections which affect my far-field results. Since I am not theoretician, I do a lot of guessing here....
If anyone likes to discuss just reply, and I have tons of questions ready....
Cheers
Phys.Org News Partner Engineering news on Phys.org
Greater safety and security at Europe's train stations
Fingerprints for freight items
On the way to a safe and secure smart home
Srbasket
#2
Apr12-09, 10:53 AM
P: 19
Hi, to make it a little bit clearer:
I have a problem with reflection of scattered field from PML (I guess). I remove scatterer, and just solve problem with geometry I want to use. There is always some kind of interference pattern of scattered field. I can decrease the reflection intensity by making PML thicker (better solution), or moving it further away (not so efficient), or both. But itīs wasting my memory. Patern depends on direction of excitation as well as on wavelength.
My geometry is spherical encapsulated in spherical PML. Pattern is different depending on what parameters i set (cartesian or spherical), although the natural choice must be spherical. Also I define scattering boundary conditions on outer boundaries.
Metallic scatterer is inside. It looks like reflections are independent of presence of scatterer. I know that solving is not selfconsistent procedure, but anyway it looks wierd that itīs independent.
Any tips?
ps. I know there must be always some reflection, but in my case is comparable to the strength of field scattered by object of interest
Attached Thumbnails
case2lambda750nm.jpg   dimers.jpg  
mahinda
#3
May5-09, 11:04 AM
P: 13
i think maybe relocate the PML further from the particle helps. the evanescent tail of the surface plasmon should be avoid from extending into the PML.

Srbasket
#4
May7-09, 02:42 AM
P: 19
COMSOL-nanoparticles-PML-scattering formalism in RF module

Hi,
I have solved some of the problems I had, and decreased the reflections. The problem was definiion of the excitation fields in different media, cause I thought that scaling down of wavelength in different media was automatic in COMSOL, and it's not.
Now, I have the folowing problem- My model consists of substrate, scattering particle lying down on substrate and air. I define PMLs next to different media as materials with ref.indices to be same as adjacent media, and in that case reflection comes from boundary between two PMLs. In order to avoid that, I read in some paper that one should calculate first scattering field in the same model geoetry but without particle first, and then use the solution as excitation in case when particle is present....
So, nw problem is not the COMSOL but MAtlab, and what is the best way to extract solutions from fem.sol.u and to use it is excitation in the next loop. I don'tt have time to spend on that at the moment, but would appreciate if someone might answer few of mine questions:
1) In harmonic scattering propagation in RF module in solver parameters there is a tab with label SOLVE FOR, and there I see variables tExscEyscEzsc10,tExscEyscEzsc20, tExscEyscEzsc11. I nknow that sc means scattered , and don't know what numbers mean and t letter.
2) in fem.sol.u I got array with some solution, and since I don't know answer to question 1) I dont know what is there. Also, If I want to use that as excitation for the next step, I don't know where and how to define it. I am sure that meshing must be the same, but don't know how to overlay former solution in the same mesh element as a new excitation for that element. Maybe sme function might be used like posteval, and similar, but ....

Thanks
mahinda
#5
May7-09, 02:57 AM
P: 13
thanks for the update man!
mahinda
#6
May7-09, 12:07 PM
P: 13
hi dude. i have one question. do you know how to use the dielectric function of metal from the material library to perform the scattering simulation?
mahinda
#7
May7-09, 09:25 PM
P: 13
BTW. do you have any idea about whether PML can be implemented in transient propagation mode? thank you very much.
Srbasket
#8
May8-09, 06:33 AM
P: 19
Well, I use complex refractive index for metals from different sources (Palik, I think), n-j*k for particular wavelength, I am not sure about data in implemented library of COMSOL.
PMLs in transient propagation mode, don't have any idea. I am using occasionally scattering harmonic propagation only, so cannot help with that...
mahinda
#9
May8-09, 08:55 AM
P: 13
oh. my goodness. does that mean each time you do the simulation you get only the result for a single frequency?

i haven't anticipated such difficulty with comsol simulation. i am actually a beginner doing light scattering by metallic nanoparticles. and i pretty much want to find the spectra of such interactions.
Srbasket
#10
May8-09, 04:22 PM
P: 19
Yes, one frequency at time since COMSOL is FEM. (U may try with some software based on FDTD method, it's faster but with some drawbacks)...
Well, I am doing the same, scattering of metallic particles, and it works fine if u calculate spectra of particles in uniform medium, but if there is some interface, problems come into play, as I have mentioned in posts at the beginning...
stealf
#11
May12-09, 05:20 AM
P: 2
Hi, I have the same problem to solve and I think I have the same problem of reflection of scattered field caused by PML. I tried to remove the scattered element but it doesn't work... I read that you solved the problem..
Can you explain me how to define the excitation fields in different media?
thanks
Srbasket
#12
May12-09, 09:18 AM
P: 19
Iīll try to explain. When I keep my excitation as exp(-j*k0_rfw*z) it works fine for vacuum medium. But in medium with ref. index of n, you have to define it as exp(-j*k0_rfw*n*z). If u have different media in your model, for example substrate and air, you have to define it in each subdomain separately. I use signum function, so if I have glass/air interface (glass for z<0) I define excitation as (1-sign(z))/2*exp(-j*k0_rfw*n*z)+(1+sign(z))/2*exp(-j*k0_rfw*z). U should incorporate also Fresnel formulae in this, but anyway like this it will work better.
Send me an email to srbasket@yahoo.com for more informations
ps. put COMSOL into subject
Cheers
tatarin
#13
Sep8-09, 01:48 PM
P: 5
Quote Quote by Srbasket View Post
Iīll try to explain. When I keep my excitation as exp(-j*k0_rfw*z) it works fine for vacuum medium. But in medium with ref. index of n, you have to define it as exp(-j*k0_rfw*n*z). If u have different media in your model, for example substrate and air, you have to define it in each subdomain separately. I use signum function, so if I have glass/air interface (glass for z<0) I define excitation as (1-sign(z))/2*exp(-j*k0_rfw*n*z)+(1+sign(z))/2*exp(-j*k0_rfw*z). U should incorporate also Fresnel formulae in this, but anyway like this it will work better.
Send me an email to srbasket@yahoo.com for more informations
ps. put COMSOL into subject
Cheers
Thank you for sharing.
Srbasket
#14
Dec20-09, 08:03 AM
P: 19
Hi, all,
folks keep contacting me about different things regarding this subject. I have advanced a bit, but still some issues remaining.
1) In excitation field definition you can forget about using complicated formulae with different expressions for z>0 and z<0. It looks like that if u use k_rfw, COMSOL will automatically consider refractive index of material into k vector in different media.
2) THERE is a big difference if u are solving for SCATTERD or TOTAL field. You should read Jiaming Jin's book to find out more. In short, if u are solving for SCATTERD field your excitation should be defined in the most precize analytical way, using Fresnel coefficients, and in that case you need complicated formulae with z>0 and z<0, since in SCATTERING MODULE you have to define field present everywhere except PML. In total field calculation, you have to insert incident wave on one boundary (port or scattering boundary condition with excitation), and propagation should be calculated... There are some advantages/disadvantages in both cases, depending what you whant to calculatet, how much memory you have, and some FEM issues that you can read in Jim's book...
3) About 2 step procedure: When u create new model name your dependents variables to scEx1,etc, and Application mode name to rfw1. Then you can make your geometry. After, you go to Multyphysics/Module navigator and ADD new scEx2,...,rfw2, and now you have two models on the same geometry. In both models you define separately Subdomain setting, Boundary settings, etc. Meshing is the same, since you have one geometry, and 2 models defioned on the same geometry. So, first you solve situation when your particle is made of air let's say, and later use that solution as excitation in your second model. Thus, in Physics/Scalar Variables you set Eoix... with Fresnel coefficients (or not if u don't want), and for E0ix2 you set Ex, E0iy2 set Ey, etc.... First step is to SOLVER MANAGER/Solve for you highlight rfw1, and for the output rfw1 (not necessary), and when it is done, you go to SOLVER MANAGER/Initial value, there you first Store solution, and check STORED SOLUTION option as initial values, SOLVE FOR you highlight rfw2, and the same for OUTPUT... That should work. Theoretical benefits of this procedure are not so clear to me. IF SOMEONE COMPARES IT; PLEASE PUT YOUR RESULTS AND CONCLUSION HERE....
4) Someone asked me "How did you separate the scattered field from the substrate and from the nanoparticle?" Maybe this 2 step procedure is the answer to that. Maybe you can normalized what you get for particle air and partice made of metal case, I don't know exactely.
5)Scattering cross-section: Read RF module reference guide. There is a Stratton-Chu formula implemented ion COMSOL for calculating far-fileld components. Onece you have them, you need spherical surface (it doesn't have to be the same as one for Stratton-Chu, can be smaller to save you integratin time) and one should do boundary integration of normEfar*normEfar, and that is proportional to scattering cross section. If you want to be more precise calculate this expresion: (normEfar*normEfar)/(Surface of particle*Ein*Ein*R*R) where R is the radius of your integrating spherical surface.
6) IF ANYONE HAVE GOOD RECIPE FOR PML THAT WORKS FINE, please POST IT!

Cheers
Srbasket
#15
Dec20-09, 08:52 AM
P: 19
This is how it looks.
Cylindar made of gold, 50nm diameter, 10nm height, laying on glass substrate...
Integration sphere for Straton-Chu was placed on 175nm from the center... That is the one geometry that worked fine in my case....Spectra are in case of air, water and dielectric of 1.517 ref. index....
Attached Thumbnails
example.gif  
Srbasket
#16
Dec21-09, 03:37 AM
P: 19
Quote Quote by Srbasket View Post
Hi, all,
folks keep contacting me about different things regarding this subject. I have advanced a bit, but still some issues remaining.
1) In excitation field definition you can forget about using complicated formulae with different expressions for z>0 and z<0. It looks like that if u use k_rfw, COMSOL will automatically consider refractive index of material into k vector in different media.
2) THERE is a big difference if u are solving for SCATTERD or TOTAL field. You should read Jiaming Jin's book to find out more. In short, if u are solving for SCATTERD field your excitation should be defined in the most precize analytical way, using Fresnel coefficients, and in that case you need complicated formulae with z>0 and z<0, since in SCATTERING MODULE you have to define field present everywhere except PML. In total field calculation, you have to insert incident wave on one boundary (port or scattering boundary condition with excitation), and propagation should be calculated... There are some advantages/disadvantages in both cases, depending what you whant to calculatet, how much memory you have, and some FEM issues that you can read in Jim's book...
3) About 2 step procedure: When u create new model name your dependents variables to scEx1,etc, and Application mode name to rfw1. Then you can make your geometry. After, you go to Multyphysics/Module navigator and ADD new scEx2,...,rfw2, and now you have two models on the same geometry. In both models you define separately Subdomain setting, Boundary settings, etc. Meshing is the same, since you have one geometry, and 2 models defioned on the same geometry. So, first you solve situation when your particle is made of air let's say, and later use that solution as excitation in your second model. Thus, in Physics/Scalar Variables you set Eoix... with Fresnel coefficients (or not if u don't want), and for E0ix2 you set Ex, E0iy2 set Ey, etc.... First step is to SOLVER MANAGER/Solve for you highlight rfw1, and for the output rfw1 (not necessary), and when it is done, you go to SOLVER MANAGER/Initial value, there you first Store solution, and check STORED SOLUTION option as initial values, SOLVE FOR you highlight rfw2, and the same for OUTPUT... That should work. Theoretical benefits of this procedure are not so clear to me. IF SOMEONE COMPARES IT; PLEASE PUT YOUR RESULTS AND CONCLUSION HERE....
4) Someone asked me "How did you separate the scattered field from the substrate and from the nanoparticle?" Maybe this 2 step procedure is the answer to that. Maybe you can normalized what you get for particle air and partice made of metal case, I don't know exactely.
5)Scattering cross-section: Read RF module reference guide. There is a Stratton-Chu formula implemented ion COMSOL for calculating far-fileld components. Onece you have them, you need spherical surface (it doesn't have to be the same as one for Stratton-Chu, can be smaller to save you integratin time) and one should do boundary integration of normEfar*normEfar, and that is proportional to scattering cross section. If you want to be more precise calculate this expresion: (normEfar*normEfar)/(Surface of particle*Ein*Ein*R*R) where R is the radius of your integrating spherical surface.
6) IF ANYONE HAVE GOOD RECIPE FOR PML THAT WORKS FINE, please POST IT!

Cheers
One clarification:
In 3) second time you are solving press RESTART, not the SOLVE
tatarin
#17
Dec21-09, 06:38 PM
P: 5
Quote Quote by Srbasket View Post
This is how it looks.
Cylindar made of gold, 50nm diameter, 10nm height, laying on glass substrate...
Integration sphere for Straton-Chu was placed on 175nm from the center... That is the one geometry that worked fine in my case....Spectra are in case of air, water and dielectric of 1.517 ref. index....
Did you compared it to any experimental data?
thx
Srbasket
#18
Dec22-09, 03:20 AM
P: 19
Not particles of that size, but they behave in expected way. For instance Bulk refractive index sensitivity is more-less expected. I did long time ago for the bigger cylinders that I fabricate, and it fitted well.
I can try again.
Anyway, it looks like one have to optimize PML and meshing to get geometry that will behave fine. I was lucky with this one...


Register to reply

Related Discussions
COMSOL : a problem about self-varying(!) subdomain IDs in Comsol with Matlab Math & Science Software 2
Diffusion Module in Comsol Multiphysics General Engineering 0
Comsol rf module (very very ) Engineering Systems & Design 0
Hamilton formalism & Lagrange Formalism General Physics 3
Need help with Comsol (Heat Transfer Module) Math & Science Software 0