Author Topic: EM Drive Developments - related to space flight applications - Thread 3  (Read 1799680 times)

Offline deuteragenie

  • Member
  • Posts: 71
  • Germany
  • Liked: 22
  • Likes Given: 0

You'll likely need to compile from source so get the latest from
https://github.com/stevengj/meep/blob/master/src/meep.hpp#L169-L172
That way your software will be current with the latest documentation.

Have found meep-mpich2 !  MPI Message Passing Interface, almost a standard parallel processing tool in large installations.  It even has a installer for Ubuntu and .deb systems if anyone else out there is interested.
(if you are  running more than one box with linux/ubuntu/?,  you can use all the machines as one single larger Virtual Machine,  eg a quad desktop and a laptop work together to use all 8 cores for the single program)

Need to sort out a few library, driver and complie conflicts and attempt to install alongside the present system.

http://ftp.univ-nantes.fr/ubuntu/pool/universe/m/meep-mpich2/

Instructions on how to build the latest Meep version:
http://geektimes.ru/post/248514/

Would be good to share the binaries...

Does Meep uses GPUs ?

Online aero

  • Senior Member
  • *****
  • Posts: 2739
  • 92129
  • Liked: 704
  • Likes Given: 237
That looks like useful information, unfortunately it's Greek to me. :)

Need to use a translator I guess.

No, Meep doesn't use GPUs, but of course the source language code is available, written in C++, so someone really, really motivated could make it so.

Same answer re. writing new functions for material characteristics, except Meep does already provide hooks for new user supplied functions in some instances and I think material characteristics is one place where it does.
« Last Edit: 06/19/2015 02:50 PM by aero »
Retired, working interesting problems

Offline rfmwguy

  • EmDrive Builder (retired)
  • Senior Member
  • *****
  • Posts: 2165
  • Liked: 2681
  • Likes Given: 1124
15 second break:

"In my opinion, all previous advances in the various lines of invention will appear totally insignificant when compared with those which the present century will witness. I almost wish that I might live my life over again to see the wonders which are at the threshold." - Charles Holland Duell, Commissioner of the US Patent Office in 1899 (His misquote was "Everything that can be invented has been invented")

Offline JasonAW3

  • Senior Member
  • *****
  • Posts: 2409
  • Claremore, Ok.
  • Liked: 374
  • Likes Given: 10
     This may have been answered before, but, has anyone checked if there had been any mass lost from the device being tested?

     It occured to me that just because we can't SEE and apparent thrust being produced, doesn't mean that there IS no thrust being produced.

     Could there be material erosion going on with these devices on the molecular scale that we aren't seeing?
My God!  It's full of universes!

Offline rfmwguy

  • EmDrive Builder (retired)
  • Senior Member
  • *****
  • Posts: 2165
  • Liked: 2681
  • Likes Given: 1124
Found my emdrive "napkin sketch" from last month. 1 cubic foot and 1.5 kg will be a challenge...

Offline deuteragenie

  • Member
  • Posts: 71
  • Germany
  • Liked: 22
  • Likes Given: 0
[quote
In other words, their data is for the Optical range, much higher frequency than Microwave frequency.

Would this be more useful ?

http://www.mathworks.com/matlabcentral/fileexchange/18040-drude-lorentz-and-debye-lorentz-models-for-the-dielectric-constant-of-metals-and-water

I still cannot find experimental data on copper in that range though.

Offline deltaMass

  • Full Member
  • ****
  • Posts: 955
  • A Brit in California
  • Liked: 671
  • Likes Given: 275
I did a lot of checking around about the permittivity (or dielectric constant or "dielectric function" as it's called because it's complex) of copper and its domain of applicability. I stand by the value I calculated.

Well below the plasma frequency the model is said to be quite reliable. Note that the operating frequencies that interest us here are six orders of magnitude down on the plasma frequency. Down at these lower frequencies, which is the case here, several sources have told me that the free electron gas model is perfectly fine for copper. The free electron model is what I used.

Indeed matters become much more complex at optical frequencies, because the plasma frequency is being approached and indeed exceeded. But we need not worry about that.

If it helps you to trust me on this, I should perhaps mention that I have a Masters in Physics with Honours from Oxford University, and that I gained my place there at age 16. Although that degree is now long in the tooth, I did actually study the physics of free electron gasses back then as it was part of the curriculum.
« Last Edit: 06/19/2015 05:57 PM by deltaMass »

Offline dustinthewind

  • Full Member
  • ****
  • Posts: 585
  • U.S. of A.
  • Liked: 226
  • Likes Given: 257
@dustinthewind has linked to this paper many times but it didn't get discussed much. It has lots of good pertinent info in there including some surprising info for you near field fans:

http://arxiv.org/pdf/1502.06288v1.pdf (note that ref 6 is our favorite anomalous thrust production...paper)

Also pulled out reference 15 as it is of interest:
http://www.asps.it/article2.pdf


Edit:
I was thinking, instead of spending thousands of man hours reading, researching, building stuff and generating hundreds of pages of thread content....how bout we just ask Watson?
http://www.ibm.com/smarterplanet/us/en/ibmwatson/what-is-watson.html

He better not say 42.

The paper in bold and underlined, at the end they are also stating that it appears the static part of the current is working against the magnetic part in electromagnetic propulsion.  "The origin of the propulsion is related to the nearfields, such as the static electric field and the inductive electromagnetic field.".  It appears they are talking here about the same thing WarpTech and I were discussing how the static field opposes the magnetic but still leads to some propulsion.  Here they imply that the near fields are a part of the propulsion and they are using two dipole antennas.

Something interesting to note is that in dipole antennas the current alternates between kinetic energy (magnetic) and potential energy (separation of charge) and these two fields are what oppose each other (w.r.t. propulsion) but also provide some propulsion (they don't balance out).  I also suspect the current passing through the magnetic fields of radiation also provide some propulsion and I think they mention this also. 

Now in the case of two circular cavities in TE011 mode but with one cavity out of phase by 90 degrees and taking into account time retardation we notice the current around the axis of the cavity doesn't allow for this charge separation and so we don't have the opposing static fields.  This is because the energy alternates from (kinetic current) to being stored in the light (also kinetic).  In the case that cavities are brought close to each other we only get magnetic interaction I think...  but there may/may-not be a draw back. 

Remember those traveling modes I suggested may exist as energy starts flowing from one cavity to the next?  I am a little divided as to which way these modes should be traveling (if they oppose propulsion or work with it). 

In figure "Fig1 Simple.png" if the force is up as in the depiction of the cavities then the top cavity appears to be working with the force of the evanescent electric field from the lower cavity.  I would think this amplifies the energy in the top cavity and the reverse for the lower cavity.  If this is true then the modes are traveling up and so they are working with the propulsion from the evanescent currents.  If I am wrong please correct me. 

I highly suspect I was wrong and that the modes are actually traveling down.

If we have eliminated the opposing static E-field opposition and only the magnetic provides propulsion and the traveling modes are also dragging the two cavities in the same direction then I suspect this could provide some effective propulsion.  However, if I am wrong and the traveling modes of light are opposing the force from the plates then I am not sure if we would get propulsion at all.  suspect modes are opposing propulsion

One thing I was suspecting was what if the traveling modes are providing propulsion in the frustum. Figure "Moving mode push.png"

Warning Edited Moving mode push.png and changed to imply I think the modes are opposing propulsion
« Last Edit: 06/19/2015 06:42 PM by dustinthewind »

Offline deltaMass

  • Full Member
  • ****
  • Posts: 955
  • A Brit in California
  • Liked: 671
  • Likes Given: 275
I need to tweak that value for the permittivity of copper. I had not used the effective mass of the electron. For copper, this turns out to be 1.38*me, and so the permittivity is correspondingly reduced.
So from
Epsilon = i / (Rho (meff/me) w)
we get at 2.4 GHz

Epsilon = i0.00288

Thanks so much for all your work, but since now I don't have a real part of permittivity, how will I impliment this model?

http://ab-initio.mit.edu/wiki/index.php/Material_dispersion_in_Meep#Conductivity_and_complex_.CE.B5

Sorry to send you down a link, but that provides a much better explaination of the model than I could. But just in case you miss it, the example model is given in italics in Scheme code as:
(make medium (epsilon 3.4) (D-conductivity (/ (* 2 pi 0.42 0.101) 3.4))

If you are concerned with expressing the relative permittivity only, then use

Epsilonr = 1 + i0.00288/Epsilon0

from which you can see how much bigger is the imaginary part - about a billion times larger than the real part, since Epsilon0 = 8.85*10-12. The real part of the relative permittivity is almost exactly = 1 at these frequencies, for copper.

From that you can verify the expression for absolute permittivity that I've been using:

Epsilon = Epsilon0 * Epsilonr ~= 10-11 + i0.00288
This result is essentially correct, the known result for a conductive metal like copper that:

The Real part of the relative permittivity is one

The Imaginary part of the relative permittivity approaches + Infinity
                                                                                                (+3.25*10^8)

p. 29 and 30 of:
http://www.phys.ufl.edu/~tanner/notes.pdf

One GHz corresponds to 0.033 1/cm frequency, or 30 cm wavelength (and to 4 μeV photon energy), which is way off to the left outside the range of the image below (observe how the Imaginary part of permittivity goes to +Infinity for low frequencies, and what a huge difference in the value of the imaginary permittivity frequency makes ), since the Imaginary part of permittivity goes to Infinity as 1/ω ,  this behavior makes the Imaginary permittivity of a metal a not a very useful function for conducting materials at microwave frequencies, also notice that the (much smaller) real part is negative:


(*IMHO The Drude model is NOT a useful model to model Copper in the GHz range*)


The Tanner notes PDF you reference states on page 1 that it applies to optical frequencies.
So not what we want.

Offline Rodal

  • Senior Member
  • *****
  • Posts: 5835
  • USA
  • Liked: 5915
  • Likes Given: 5253
I did a lot of checking around about the permittivity (or dielectric constant or "dielectric function" as it's called because it's complex) of copper and its domain of applicability. I stand by the value I calculated.

Well below the plasma frequency the model is said to be quite reliable. Note that the operating frequencies that interest us here are six orders of magnitude down on the plasma frequency. Down at these lower frequencies, which is the case here, several sources have told me that the free electron gas model is perfectly fine for copper. The free electron model is what I used.

Indeed matters become much more complex at optical frequencies, because the plasma frequency is being approached and indeed exceeded. But we need not worry about that.

If it helps you to trust me on this, I should perhaps mention that I have a Masters in Physics with Honours from Oxford University, and that I gained my place there at age 16. Although that degree is now long in the tooth, I did actually study the physics of free electron gasses back then as it was part of the curriculum.

As I wrote here: http://forum.nasaspaceflight.com/index.php?topic=37642.msg1391527#msg1391527

I agree with you that the value is essentially theoretically correct. 

However the issue in inputting these values in MEEP is one of numerical correctness due to numerical implementation in the code.

As the Imaginary part of the relative permittivity approaches + Infinity as 1/ω, the value of  (+3.25*10^8) being so large may result in numerical issues in MEEP's numerical implementation of the Drude model, if so it would not be a useful numerical model.  There is no harm in trying these values in MEEP and seeing how MEEP handles it numerically...

The Drude model in MEEP was written and used mainly for optical applications, in which range the value of the Imaginary part of pemittivity is much lower.

The issue is how will MEEP handle these values (that's why the numbers used in machine precision are important in numerical implementations).

Just because the value is theoretically correct, does not necessarily mean that MEEP will handle it correctly, particularly if the people that wrote the MEEP code had in mind people using the model for optical applications and not for GHz applications.

I will also be interested in finding out how MEEP handles this input for the microwave range.
« Last Edit: 06/19/2015 06:35 PM by Rodal »

Offline SeeShells

  • Senior Member
  • *****
  • Posts: 2326
  • Every action there's a reaction we try to grasp.
  • United States
  • Liked: 2956
  • Likes Given: 2588
That looks like useful information, unfortunately it's Greek to me. :)

Need to use a translator I guess.

No, Meep doesn't use GPUs, but of course the source language code is available, written in C++, so someone really, really motivated could make it so.

Same answer re. writing new functions for material characteristics, except Meep does already provide hooks for new user supplied functions in some instances and I think material characteristics is one place where it does.
Digging I found this.
http://www.ajol.info/index.php/ijest/article/viewFile/83885/73892
Simulation and analysis of microwave heating while joining bulk copper
M. S. Srinath 1*, P. Suryanarayana Murthy2
, Apurbba Kumar Sharma3
,
Pradeep Kumar4
, M. V. Kartikeyan5
1*Department of Mechanical Engineering, Malnad College of Engineering, Hassan-573201, INDIA
2,3,4Department of Mechanical and Industrial Engineering, Indian Institute of Technology Roorkee, Roorkee 247 667, INDIA.
5Department of Electronics and Computer Science Engineering, Indian Institute of Technology Roorkee, Roorkee

I believe you will find what you're looking for here.
Shell

Offline deuteragenie

  • Member
  • Posts: 71
  • Germany
  • Liked: 22
  • Likes Given: 0
I did a lot of checking around about the permittivity (or dielectric constant or "dielectric function" as it's called because it's complex) of copper and its domain of applicability. I stand by the value I calculated.

Well below the plasma frequency the model is said to be quite reliable. Note that the operating frequencies that interest us here are six orders of magnitude down on the plasma frequency. Down at these lower frequencies, which is the case here, several sources have told me that the free electron gas model is perfectly fine for copper. The free electron model is what I used.

Indeed matters become much more complex at optical frequencies, because the plasma frequency is being approached and indeed exceeded. But we need not worry about that.

If it helps you to trust me on this, I should perhaps mention that I have a Masters in Physics with Honours from Oxford University, and that I gained my place there at age 16. Although that degree is now long in the tooth, I did actually study the physics of free electron gasses back then as it was part of the curriculum.

As I wrote here: http://forum.nasaspaceflight.com/index.php?topic=37642.msg1391527#msg1391527

I agree with you that the value is essentially theoretically correct. 

However the issue in inputting these values in MEEP is one of numerical correctness due to numerical implementation in the code.

As the Imaginary part of the relative permittivity approaches + Infinity as 1/ω, the value of  (+3.25*10^8) being so large may result in numerical issues in MEEP's numerical implementation of the Drude model, if so it would not be a useful numerical model.  There is no harm in trying these values in MEEP and seeing how MEEP handles it numerically...

The Drude model in MEEP was written and used mainly for optical applications, in which range the value of the Imaginary part of pemittivity is much lower.

The issue is how will MEEP handle these values (that's why the numbers used in machine precision are important in numerical implementations).

Just because the value is theoretically correct, does not necessarily mean that MEEP will handle it correctly, if the person that wrote the MEEP code had in mind people using the model for optical applications and not for GHz applications.

I will also be interested in finding out how MEEP handles this input for the microwave range.

Well... if it all boils to the fact that copper in these ranges behaves very closely to a perfect metal, than Meep has a predefined material type for this: "perfect-metal".

A predefined material type corresponding to a perfect electric conductor (at the boundary of which the parallel electric field is zero), technically epsilon = -infinity. 

That will simplify things and prevent numerical issues.

Offline deltaMass

  • Full Member
  • ****
  • Posts: 955
  • A Brit in California
  • Liked: 671
  • Likes Given: 275
I did a lot of checking around about the permittivity (or dielectric constant or "dielectric function" as it's called because it's complex) of copper and its domain of applicability. I stand by the value I calculated.

Well below the plasma frequency the model is said to be quite reliable. Note that the operating frequencies that interest us here are six orders of magnitude down on the plasma frequency. Down at these lower frequencies, which is the case here, several sources have told me that the free electron gas model is perfectly fine for copper. The free electron model is what I used.

Indeed matters become much more complex at optical frequencies, because the plasma frequency is being approached and indeed exceeded. But we need not worry about that.

If it helps you to trust me on this, I should perhaps mention that I have a Masters in Physics with Honours from Oxford University, and that I gained my place there at age 16. Although that degree is now long in the tooth, I did actually study the physics of free electron gasses back then as it was part of the curriculum.

As I wrote here: http://forum.nasaspaceflight.com/index.php?topic=37642.msg1391527#msg1391527

I agree with you that the value is essentially theoretically correct. 

However the issue in inputting these values in MEEP is one of numerical correctness due to numerical implementation in the code.

As the Imaginary part of the relative permittivity approaches + Infinity as 1/ω, the value of  (+3.25*10^8) being so large may result in numerical issues in MEEP's numerical implementation of the Drude model, if so it would not be a useful numerical model.  There is no harm in trying these values in MEEP and seeing how MEEP handles it numerically...

The Drude model in MEEP was written and used mainly for optical applications, in which range the value of the Imaginary part of pemittivity is much lower.

The issue is how will MEEP handle these values (that's why the numbers used in machine precision are important in numerical implementations).

Just because the value is theoretically correct, does not necessarily mean that MEEP will handle it correctly, if the person that wrote the MEEP code had in mind people using the model for optical applications and not for GHz applications.

I will also be interested in finding out how MEEP handles this input for the microwave range.
Good points. I did not look at the MEEP side of things. As a software engineer with 5 decades of experience, I can share your concerns. But 10^8 is not a terribly big number, and MEEP certainly uses floating point. I'm not about to go study MEEP though. I suggest aero tries it out and reports back.

Offline deltaMass

  • Full Member
  • ****
  • Posts: 955
  • A Brit in California
  • Liked: 671
  • Likes Given: 275
I did a lot of checking around about the permittivity (or dielectric constant or "dielectric function" as it's called because it's complex) of copper and its domain of applicability. I stand by the value I calculated.

Well below the plasma frequency the model is said to be quite reliable. Note that the operating frequencies that interest us here are six orders of magnitude down on the plasma frequency. Down at these lower frequencies, which is the case here, several sources have told me that the free electron gas model is perfectly fine for copper. The free electron model is what I used.

Indeed matters become much more complex at optical frequencies, because the plasma frequency is being approached and indeed exceeded. But we need not worry about that.

If it helps you to trust me on this, I should perhaps mention that I have a Masters in Physics with Honours from Oxford University, and that I gained my place there at age 16. Although that degree is now long in the tooth, I did actually study the physics of free electron gasses back then as it was part of the curriculum.

As I wrote here: http://forum.nasaspaceflight.com/index.php?topic=37642.msg1391527#msg1391527

I agree with you that the value is essentially theoretically correct. 

However the issue in inputting these values in MEEP is one of numerical correctness due to numerical implementation in the code.

As the Imaginary part of the relative permittivity approaches + Infinity as 1/ω, the value of  (+3.25*10^8) being so large may result in numerical issues in MEEP's numerical implementation of the Drude model, if so it would not be a useful numerical model.  There is no harm in trying these values in MEEP and seeing how MEEP handles it numerically...

The Drude model in MEEP was written and used mainly for optical applications, in which range the value of the Imaginary part of pemittivity is much lower.

The issue is how will MEEP handle these values (that's why the numbers used in machine precision are important in numerical implementations).

Just because the value is theoretically correct, does not necessarily mean that MEEP will handle it correctly, if the person that wrote the MEEP code had in mind people using the model for optical applications and not for GHz applications.

I will also be interested in finding out how MEEP handles this input for the microwave range.

Well... if it all boils to the fact that copper in these ranges behaves very closely to a perfect metal, than Meep has a predefined material type for this: "perfect-metal".


A predefined material type corresponding to a perfect electric conductor (at the boundary of which the parallel electric field is zero), technically epsilon = -infinity. 

That will simplify things and prevent numerical issues.

I think you solved the problem.  At least well enough for aero to do useful work.

Offline SeeShells

  • Senior Member
  • *****
  • Posts: 2326
  • Every action there's a reaction we try to grasp.
  • United States
  • Liked: 2956
  • Likes Given: 2588
https://hal.archives-ouvertes.fr/jpa-00246781/document
On the other hand, the 60 mm length of this region corresponds to a delay time of 0.2 nsec of light in vacuum, this is clearly longer than the measured delay in the first 5 nsec duration of the pulse superluminal conditions are present both for the center of gravity and the maximum value of the electromagnetic packet. Furthermore this confirmes the correctness of the frequency domain data and the corresponding Fourier evaluation [2, 3]. The zerc-time traversal described in references [2] and [3] proves to be correct, I-e- there is no additional time delay caused by an additional length of the evanescent region.
<end quote>

Offline Rodal

  • Senior Member
  • *****
  • Posts: 5835
  • USA
  • Liked: 5915
  • Likes Given: 5253
I did a lot of checking around about the permittivity (or dielectric constant or "dielectric function" as it's called because it's complex) of copper and its domain of applicability. I stand by the value I calculated.

Well below the plasma frequency the model is said to be quite reliable. Note that the operating frequencies that interest us here are six orders of magnitude down on the plasma frequency. Down at these lower frequencies, which is the case here, several sources have told me that the free electron gas model is perfectly fine for copper. The free electron model is what I used.

Indeed matters become much more complex at optical frequencies, because the plasma frequency is being approached and indeed exceeded. But we need not worry about that.

If it helps you to trust me on this, I should perhaps mention that I have a Masters in Physics with Honours from Oxford University, and that I gained my place there at age 16. Although that degree is now long in the tooth, I did actually study the physics of free electron gasses back then as it was part of the curriculum.

As I wrote here: http://forum.nasaspaceflight.com/index.php?topic=37642.msg1391527#msg1391527

I agree with you that the value is essentially theoretically correct. 

However the issue in inputting these values in MEEP is one of numerical correctness due to numerical implementation in the code.

As the Imaginary part of the relative permittivity approaches + Infinity as 1/ω, the value of  (+3.25*10^8) being so large may result in numerical issues in MEEP's numerical implementation of the Drude model, if so it would not be a useful numerical model.  There is no harm in trying these values in MEEP and seeing how MEEP handles it numerically...

The Drude model in MEEP was written and used mainly for optical applications, in which range the value of the Imaginary part of pemittivity is much lower.

The issue is how will MEEP handle these values (that's why the numbers used in machine precision are important in numerical implementations).

Just because the value is theoretically correct, does not necessarily mean that MEEP will handle it correctly, if the person that wrote the MEEP code had in mind people using the model for optical applications and not for GHz applications.

I will also be interested in finding out how MEEP handles this input for the microwave range.

Well... if it all boils to the fact that copper in these ranges behaves very closely to a perfect metal, than Meep has a predefined material type for this: "perfect-metal".

A predefined material type corresponding to a perfect electric conductor (at the boundary of which the parallel electric field is zero), technically epsilon = -infinity. 

That will simplify things and prevent numerical issues.

Yes, this is what I advocate:

1) That @aero uses the "perfect metal" or other such boundary condition (perhaps somebody wrote a skin effect BC?).  The issue of  finite Q can be handled as done for example by Greg Egan and in textbooks using the skin effect.  Actually, perhaps somebody wrote such a condition (to obtain a finite Q based on a simple formula based on skin effect) for MEEP?

2) That @aero only meshes the interior of the cavity with a Finite Difference grid (not the outside)

3) That @aero conducts a Time-Marching Finite-Difference TIme-Domain scheme to study the possible presence of evanescent waves and time-domain effects that have been brought up in this thread

4) After that we could examine the issues of evanescent wave leaking, the non-perfect metal, etc.
« Last Edit: 06/19/2015 06:55 PM by Rodal »

Offline rfmwguy

  • EmDrive Builder (retired)
  • Senior Member
  • *****
  • Posts: 2165
  • Liked: 2681
  • Likes Given: 1124
Future considerations:

Should the emdrive become a reality, I can envision a next-generation of high-power, miniaturized frustums (from my old filter days - when these were the newest disruptive innovation/technology introduced)

Below is a miniature rectangular bandpass cavity resonator and an old round air cavity. Its outer walls and inner conductor (hole) are silver plated. Underneath the ceramic is a somewhat proprietary ceramic material. Simply think of this as a straight-sided frustum, a common rectangular cavity filter. Ceramic dielectric is used to miniaturize the assembly, replacing the air dielectric (K=1). Power handling is somewhat less than an air cavity.

Now, shape the frustum with ceramic and it will significantly shrink the form factor at any frequency chosen. Another benefit would be a molded (repeatable) design that is lower cost to manufacture. The weight will be much less another befit.

The ceramic material is a powder, thermally cured and typically not milled.

Online aero

  • Senior Member
  • *****
  • Posts: 2739
  • 92129
  • Liked: 704
  • Likes Given: 237
So the consensus is that I should use e = 1 +i 3.25E+8, correct?

I'll try that, first to see what effect it has on the resonant frequency. I'll let you know.

I don't anticipate that this will introduce numerical problems in meep but we will see.
Retired, working interesting problems

Offline jmossman

  • Member
  • Posts: 72
  • San Jose, CA
  • Liked: 58
  • Likes Given: 171
...
No, Meep doesn't use GPUs, but of course the source language code is available, written in C++, so someone really, really motivated could make it so.
...

Meep utilizes BLAS and LAPACK libraries, which can be optionally replaced (at compile time) with GPU optimized versions.  How much speedup would be available is tough to say given that both Nvidia and AMD/ATI like to play marketing games.  (and exact performance will depend upon GPU version;  consumer GPUs won't provide nearly the boost of a $$$ professional-level GPU product)

http://ab-initio.mit.edu/wiki/index.php/Meep_Installation#BLAS_and_LAPACK_.28recommended.29

https://developer.nvidia.com/cublas
https://developer.nvidia.com/magma
http://developer.amd.com/tools-and-sdks/cpu-development/amd-core-math-library-acml/
http://developer.amd.com/tools-and-sdks/opencl-zone/amd-accelerated-parallel-processing-math-libraries/

Offline Rodal

  • Senior Member
  • *****
  • Posts: 5835
  • USA
  • Liked: 5915
  • Likes Given: 5253
So the consensus is that I should use e = 1 +i 3.25E+8, correct?

I'll try that, first to see what effect it has on the resonant frequency. I'll let you know.

I don't anticipate that this will introduce numerical problems in meep but we will see.
By numerical problems I don't mean that the code will halt, I meant the accuracy of the solution due to finite precision.  It depends on how the MEEP coder handled the numerical operations (whether they took into account that the input can differ by so many orders of magnitude and how these impacts numerical handling).

To look at the accuracy of this, two runs will be needed:

1) one run with the boundary condition set as "perfect metal"

2) another run with the Drude mode as per input from deltaMass, including the Imaginary part =  +i 3.25E+8

The Finite Difference Mesh for the interior of the cavity should be exactly the same for both runs.
This will be interesting !
« Last Edit: 06/19/2015 07:31 PM by Rodal »

Tags: