IE3D 10 & FIDELITY 4
1) "IT IS BASED ON CONFORMAL FDTD"
(Warning for CST, as they have just appreciated conformal FIT and are going to implement it. They had better be fast.)
2) IT IS COMPLETELY REWRITTEN.
3) IT SUPPORTS MULTIDOMAIN FDTD.
I hope they have implemented the best algorithm in literature. If so, it will be a TOP GUN.
Regards
wave-maniac
which conformal FDTD algorithm is the best, do you think? I wouldn't expect too much from fidelity.
I don't think it 's worth for wait
Dear friends;
To the best of my knowledge, the method by Dey&Mittra is the best. There is a paper in IEEE which has compared several Conformal FDTD (CFDTD). The authors had concluded the above mentioned method is the best as regards to accuracy and stability. I can provide the address if necessary.
Even CST main brain, i.e. Thomas Weiland, has admitted the shortcomings of their PBA (Perfect Boundary Approximation). Have a look at the following abstract:
http://www.win.tue.nl/scee2002/Schuhmann_abstract.pdf
It clearly indicates that they have selected a method similar to that of Dey&Mittra. Of course, they claim they are going to improve it, as to maximum stable time step.
For additional info, have a look at the following webpage:
http://www.tu-darmstadt.de/fb/et/tem...g/HF/NFIT.html
As to Fidelity, let's not prejudge, but hope Zeland has gone for the same thing. The good thing is that the guys in Zeland have accepted the truth about the weakness of Fidelity.
Best regards
wave-maniac
Dey-Mittra algorithm works fine for some curved PEC boundary problems. The situation is different for irregular dielectric body. The problem (which one is best) is still open.
It seems to me your statement concerning the accuracy and stability is misleading. I am sure the conclusion is with respect to some particular problem (I've read some analysis on reflection from a planar PEC surface not conforming to the grid).
One particular question on the Dey-Mittra procedure has been bothering me. In the guidelines for setting up conformal mesh, use is made of a ratio between area and length. This number, unfortunately is dependent on the unit used, it could be 1e-3 or 1e6. So the guideline is not scientific. I look forward to your comments on this issue.
Dear loucy;
To make sure you're not kidding about Mittra's team (i.e. Dey and Yu) works, please mention the complete address of the papers which have led you to those conclusions. Particularly reveal who has or have claimed:
1) Mittra's group method (aka. D-FDTD) has not been generalized to irregular dielectric surfaces.
2) D-FDTD has generally problems as to stability and accuracy.
My comments will follow the address of the papers.
Best regards
wave-maniac
I'm very interested too.
Thanks.
C. J. Railton "an Analytical and Numerical Analysis of Several Locally Conformal FDTD Schemes" MTT Vol.47, No.1, Jan 1999. In particular, look at the last (lower right) paragraph on the first page regarding the dielectric boundary. (Of course, we can't just take their words, bur first let's see your comments.) BTW, what is the paper wave-maniac is referring to concerning the accuraccy of conformal FDTD?
Let's put aside the issue of curved dielectric boundary for a moment and study the Dey-Mittra procedure. For convenience, the guidelines for setting up the conformal mesh are cited here: (IEEE microwave and guided wave letters", Vol7 No.9, 1997, p.274)
"Numerical experiments have shown that a time step of 50 to 70% of the courant limit associated with the undistorted cell is adequate to ensure the stability of the algorithm, provided that the following conditions are met.
1) The area of the distorted cell (partially filled) is greater than 1.5% (for 50% of Courant limit) and 2.5% (for 70% of limit) of the area of the undistorted cell area.
2) The ratio between the maximum length of the side of a cell and its area is less than 15 (for 50% of Courant limit) and 10 (for 70% of limit). "
I am having trouble understanding the second criteria (length/area), which depends on the unit used. Right now I think this is absolutely not scientific.
Please note that I am not claiming D-FDTD has stability problem. Instead, I am looking for clues related to the stability and, more importantly, the accuracy.
Dear loucy,
The confusion is due to not updating your list of references on conformal FDTD. You can find my comments on Dey-Mittra CFDTD, or D-FDTD as known in literature, on this page (Already mentioned as an attached file). Note that you had mentioned small parts of the references [2] and [6]. I have used [6] in favor of D-FDTD.
Best regards
wave-maniac
***************** My comments *********************
Prior to Dey-Mittra work, some methods had been published on conformal FDTD by various authors, which were sort of complex. The best ones used the idea of borrowing and interpolation of electric fields from the neighboring cells, at the object boundary. Dey-Mittra presented a novel method, first in [1] and [2] and then modified in [3], whose pros and cons are as follow:
Pros:
1) It’s very simple. This has been admitted by many including in [6].
2) It’s accurate. Please read the conclusion in [6].
3) It doesn’t need the borrowing/interpolation stuff [1] to [4], and [6].
4) The standard update equations for the E-fields are unchanged ([1] to [13]).
5) The inclusion of FDTD formulations such as radiation boundary conditions, total-field/scattered field surfaces, and near-to-far field transformations may be used without alteration [13].
Cons:
1) The time step has to be 50 to 70 percent of Courant limit, which according to the conclusion in [6], “is a small price in view of the much greater saving of resources which result from using the method ”.
2) To use their idea, one has to create a particular FDTD mesh near the PEC surfaces. There were two restrictions, including the “length to area ratio”, which I’ll mention later. However, it’s not a big deal. The FDTD guru, Allen Taflove found the method worth of developing a special mesh generator [13]. Note the paper date is 2002. It is also mentioned in his homepage.
Dealing with arbitrary shaped dielectric objects:
In [4] and [5], Dey-Mittra extended their method to include arbitrary shaped dielectric objects. They actually used sort of effective permittivities of FDTD cells.
Final Modification, Yu-Mittra Method:
In 2000, Yu and Mittra modified the original method to get rid of the first disadvantage, using a simple change in H-field updating ([7], [8], [9] and [10]). They actually used the entire cell area instead of the deformed cell area. Thus they were able to use time steps equal to 99.5 percent of Courant limit. They also extended it to include arbitrary shaped dielectric objects in [11]. There, instead of using a volume average, their conformal dielectric algorithm utilizes a linear average concept. They have developed software based on their algorithm. You can download the demo version (20.3 MB) from the following link:
http://www.ecl.ee.psu.edu/demos_fss_cfdtd/cfdtd.zip
Mittra’s team have recently published a lot of papers (e.g. [12]) on various applications of the method and/or software, indicating their versatility.
Line to Area ratio Criterion:
I think they meant a dimensionless, normalized distorted cell geometry. Otherwise it would be a meaningless criterion. As you might see in [6], it has been mentioned the restriction can be meaningfully interpreted for cells with equal lengths. I really doubt Railton, et al, didn’t realize the criterion to be nonsense. By normalized geometry, I mean normalizing by the maximum, particularly distorted, length of the cell. By the way, don’t you think there is no use to think about it? As was mentioned before, the distorted area is no longer used in H-field updating procedure.
Note and conclusion:
By no means I believe that the Mittra’s team is the best forever. But up to now, no other conformal FDTD method has been reported which is very simple in formulation and yet accurate. As I mentioned before, even FDTD giants such as Taflove and Weiland are either using the algorithm or are going to use “their own version of Mittra’s team algorithm. By the way, some guys, including Kosmanis and Tsiboukis have presented some rigorous methods for conformal FDTD, which seem powerful, but by no means simple. [14] is one of their papers.
Anyway, you can use www.fdtd.org or IEEE Xplore to keep yourself updated.
References:
[1] A Locally Conformal Finite Difference Time Domain (FDTD) Algorithm for Modeling 3-D Objects with Curved Surfaces, S. Dey and R. Mittra, IEEE Antennas and Propagat. Soc. Int. Symp., Montréal, Canada, vol. 4, 2172-2175, July, 1997
[2] A Locally Conformal Finite-Difference Time-Domain (FDTD) Algorithm for Modeling Three-Dimensional Perfectly Conducting Objects, S. Dey and R. Mittra, IEEE Microwave Guided Wave Letters, vol. 7, no. 9, 273-275, September, 1997
[3] A Modified Locally Conformal Finite-Difference Time-Domain Algorithm for Modeling Three-Dimensional Perfectly Conducting Objects, S. Dey and R. Mittra, Microwave and Optical Technology Letters, vol. 17, no. 6, 349-352, April, 1998
[4] A locally conformal finite difference time domain technique for modeling arbitrary shaped objects, S. Dey and R. Mittra, IEEE Antennas Propagat Soc Int Symp, June 1998, vol. 1, pp. 584-587.
[5] A Conformal Finite-Difference Time-Domain Technique for Modeling Cylindrical Dielectric Resonantors, S. Dey and R. Mittra, IEEE Transactions on Microwave Theory and Techniques, vol. 47, no. 9, 1737-1739, September, 1999
[6] An Analytical and Numerical Analysis of Several Locally Conformal FDTD Schemes, C. J. Railton and J. B. Schneider, IEEE Transactions on Microwave Theory and Techniques, vol. 47, no. 1, 56-66, January, 1999
[7] Novel Conformal FDTD Approach for Modeling Monolithic Microwave Integrated Circuits (MMIC), W. Yu and R. Mittra, IEEE Antennas and Propagat. Soc. Int. Symposium, Salt Lake City, UT, vol. 1, 244-247, July, 2000
[8] A Conformal FDTD Algorithm for Modeling Perfectly Conducting Objects with Curve-Shaped Surfaces and Edges, W. Yu and R. Mittra, Microwave and Optical Technology Letters, vol. 27, no. 2, 136-138, October, 2000
[9] A Conformal FDTD Software Package Modeling Antennas and Microstrip Circuit Components, W. Yu and R. Mittra, IEEE Antennas and Propagation Magazine, vol. 42, no. 5, 28-39, October, 2000
[10] Accurate Modeling of Planar Microwave Circuit Using Conformal FDTD Algorithm, W. Yu and R. Mittra, Electronics Letters, vol. 36, no. 7, 618-619, 2000
[11] A Conformal Finite Difference Time Domain Technique for Modeling Curved Dielectric Surfaces, W. Yu and R. Mittra, IEEE Microwave and Wireless Components Letters, vol. 11, no. 1, 25-27, January, 2001
[12] Application of FDTD Method to Conformal Patch Antennas, W. Yu, N. Farahat, and R. Mittra, IEE Proceedings H: Microwaves, Antennas and Propagation, vol. 148, no. 3, 218-220, June, 2001
[13] Three-Dimensional CAD-Based Mesh Generator for the Dey-Mittra Conformal FDTD Algorithm, G. Waldschmidt and A. Taflove, IEEE Antennas and Propagat. Soc. Int. Symposium, San Antonio, TX, vol. 3, 612-615, June, 2002.
[14] A Systematic Conformal Finite-Difference Time-Domain (FDTD) Technique for the Simulation of Arbitrarily Curved Interfaces between Dielectrics, T. I. Kosmanis and T. D. Tsiboukis, IEEE Transactions on Magnetics, vol. 38, no. 2, 645-648, March, 2002.
A link for some demos from the same team:
http://www.personal.psu.edu/faculty/...tware_wen.html
Wave-maniac,
Since you have "used [6] in favor of D-FDTD", it should be emphasized again that the analysis in [6] of your list (which I cited above) is for PEC surface.
I've read all of the papers in your list except the last one, which you said is more rigorous. In my opinion, the most important information missing from the papers by Yu et al. is a comparison with staircasing approximation to show a clear advantage. I'd like to see at least a careful convergence study with respect to the cell size, if a rigorous mathematical analysis on the stability and accuracy is not feasible. I must say that those papers are not convincing at all.
Going back to the issue of "length/area", I don't see how a normalization helps. If all edge lengths are normalized by the same factor, such factor would still come up after taking length/area.
If any of you likes the software listed in the above link, please share your experience. I think even the several-year-old XFDTD 5.1 is better, which consumes much less memory, runs faster and hence affords a much bigger computation domain.
Dear loucy,
Please study my comments again.
You mentioned:
************************************************** *********
Since you have "used [6] in favor of D-FDTD", it should be emphasized again that the analysis in [6] of your list (which I cited above) is for PEC surface.
************************************************** *********
So what? Did I say it is on dielectric surfaces? Please read it again. I used it for its conclusion where it said the accuracy of D-FDTD is better than others and the extra processing time necessary for smaller time steps is not important. Actually the authors in [6] had mentioned "this is considered a small price in view of the much greater saving of the resources which will result from using the method", which is against your final conclusion.
------------------------------------------------------
You mentioned:
************************************************** *********
I've read all of the papers in your list except the last one, which you said is more rigorous. In my opinion, the most important information missing from the papers by Yu et al. is a comparison with staircasing approximation to show a clear advantage. I'd like to see at least a careful convergence study with respect to the cell size, if a rigorous mathematical analysis on the stability and accuracy is not feasible.
************************************************** *********
You're either kidding, or used to neglecting parts that you don't like to see.
The comparison has been presented in many papers, including some of those I've mentioned. Is it a problem that others do the same test or you insist that YU should do it? If not, please have a better look at [13], figures 3 and 4.
Also study [5], tables 1 and 2. This actually includes the dielectric surfaces.
Also [6], page 65, "B. Cylindrical Resonant Cavity", and the relevant figures 10 and 11.
Thomas Weiland, the main brain of CST MWS and MAFIA, mentions the following stuff about CFDTD, particularly Dey-Mittra method in the document that I had already presented its address, i.e. http://www.win.tue.nl/scee2002/Schuhmann_abstract.pdf :
The authors say:
*******************
However, there are also several shortcomings of the standard FIT formulation, including:
1) The poor modeling quality for arbitrarily shaped geometric objects, if Cartesian grids are used (?staircase?-problem). As a remedy, FIT has been extended to a variety of generalized grid types, like (2D) triangular or (3D) structured non-orthogonal grids. The most promising approach, however, is the application of conformal techniques as proposed in (6) or (7), where the efficient data-structure of Cartesian grids remains untouched. So far, these techniques require a moderate reduction
of the maximum stable time step.
*******************
It should be noted that (7) in the above-mentioned document, refers to Dey-Mittra original paper which is [2].
-----------------------------------------------------------
You mentioned:
*************************
I must say that those papers are not convincing at all.
*************************
My dear friend, believe me that it's not my problem.
---------------------------------------------
You mentioned:
********************
Going back to the issue of "length/area", I don't see how a normalization helps. If all edge lengths are normalized by the same factor, such factor would still come up after taking length/area.
********************
You think the criterion is not scientific, but the authors in [6] say it can be meaningfully interpreted. Allen Taflove has fully understood and developed a mesh based on that. It you still want to think about that, no problem. I never mind. It's up to you.
--------------------------------------------------------------------------
You mentioned:
***********************************
I think even the several-year-old XFDTD 5.1 is better (than D-FDTD), which consumes much less memory, runs faster and hence affords a much bigger computation domain. "
***********************************
Scientific literature indicates something else. "I think this" or "I say that? without bringing a single paper, which might approve your ideas, is not a scientific method.
Allen Taflove, in [13], in the first page says:
"Locally conformal FDTD methods alter the standard FDTD grid only where the geometry intersects the grid to permit partial edge lengths and areas. Elsewhere, the grid remains untouched. Locally conformal methods have been rigorously analyzed and have been shown to remain stable and give superior results to the standard FDTD method [Ref. 3 which is equal to [6]).
*********************************
As you know, XFDTD uses standard FDTD which has the staircasing problem.
------------------------------------------------
The conclusion:
So you claim you're not convinced by any papers by Mittra, Railton [6], and Taflove and insist the XFDTD 5.1 is much better than any of the CFDTD methods, particularly D-FDTD. I'm dead sure you're kidding.
Good luck
wave-maniac
on "So what? Did I say it is on dielectric surfaces?"
So the D-FDTD is not necessarily accurate for dielectric surfaces, because this case was not analyzed in [6].
-----------------------------------
On "The comparison has been presented in many papers, including some of those I've mentioned. Is it a problem that others do the same test or you insist that YU should do it? If not, please have a better look at [13], figures 3 and 4. ....."
I specifically mentioned those "papers by Yu et al" in the relevant comment. Tell me which paper by Yu and coauthors has convergence study reported.
--------------------
on "My dear friend, believe me that it's not my problem."
No comment.
---------------------------------------------
on "You think the criterion is not scientific, but the authors in [6] say it can be meaningfully interpreted. Allen Taflove has fully understood and developed a mesh based on that."
They didn't intergret it, even for the case of dx=dy. They used it in a 1-D situation.
I am sure Allen Taflove and many others understood it. They just won't say.
--------------------------------------------------------------------------
On 'Scientific literature indicates something else. "I think this" or "I say that” without bringing a single paper, which might approve your ideas, is not a scientific method. '
No Comment at all.
---------------------------------------------------------------------------
On " As you know, XFDTD uses standard FDTD which has the staircasing problem."
The concept of equivalent dielectric parameter for the conformal cells in D-FDTD series (Yu) is actually very similar to the "Fuzzy cell" in XFDTD. I'd like to see a comparison of DFDTD and "staicasing+fuzzy cell".
(in my opinion, XFDTD is not as good as the CST and EMpire.).
------------------------------------------------
Dear loucy;
1) You mentioned:
*****************************************
So the D-FDTD is not necessarily accurate for dielectric surfaces, because this case was not analyzed in [6].
*****************************************
This means you have accepted that D-FDTD is accurate for PEC. Don't say no, otherwise, the above mentioned sentense would be meaningless.
Also the above sentence means you would accept D-FDTD accuracy for dielectric surfaces if it was mentioned in [6]. So you believe in Railton, et al paper [6] that much.
Also, it seems that you think it's possible, but not necessarily, that D-FDTD is accurate for dielectric surfaces. It's much better than what you had calimed "decisively" before.
BTW, [6] was published in 1999, but the final form of D-FDTD, presented by YU-Mittra, which considers dielectric surfaces [11], was published in 2001. So if you just believe in Railton proficiency in FDTD, but
neither Taflove nor Weiland, unless you ask him directly, you can not get rid of doubts.
----------------------------------------------------------------------
2) You mentioned:
*******************************
I specifically mentioned those "papers by Yu et al" in the relevant comment. Tell me which paper by Yu and coauthors has convergence study reported
******************************
So, you claim unless YU provides data on convergence report, you're not going to accept it from anyone else, even Taflove[13] or Railton [6]. It seems you don't like or believe selectively that part of [6] on the convergence study. So you even don't accept the comparison of D-FDTD and normal FDTD, which uses staircasing, presented in any of those papers, even by Yu, et al. Didn't I say you're kidding?:D
I recommend that you have a look at the presentation by Taflove, which you can find in his website.
http://www.ece.northwestern.edu/ecef...esentation.pdf, pages 36-39
---------------------------------------------------
3) You mentioned:
************************************
They didn't interpret it, even for the case of dx=dy. They used it in a 1-D situation.
I am sure Allen Taflove and many others understood it. They just won't say
************************************
Did I claim they have presented their understandings? Presentation of that is another story. I mentioned as they have had no problem with that and/or consider that meaningfully interpretable, the "length to area ratio" can be neither nonsense nor non-scientific.
-------------------------------------------------
4) You mentioned:
********************
The concept of equivalent dielectric parameter for the conformal cells in D-FDTD series (Yu) is actually very similar to the "Fuzzy cell" in XFDTD. I'd like to see a comparison of DFDTD and "staicasing+fuzzy cell".
********************
Thanks for bringing my attention to fuzzy cell in XFDTD. I read its manual. You're wrong. It uses average volume concept which is exactly that of Dey-Mittra extension method to dielectric surfaces [4] and [5].
It is zero-order accuracy. Yu-Mittra improved the accuracy by considering a first order accuracy [11] . Please study sections A and B of [11] more carefully.
-------------------------------------------
5) You mentioned:
****************************
in my opinion, XFDTD is not as good as the CST and EMpire
****************************
So, you accept CST is better than XFDTD. Also you claimed XFDTD is better than CFDTD. So the logical result is that you believe CST method (i.e. FIT) is better than CFDTD, particularly D-FDTD. BUT, Weiland of CST says FIT, as used in their products has short comings and they have targeted CFDTD methods to solve the problem, and they consider D-FDTD as a promising one. What an idea? :D
--------------------------------------------------
BTW, as to resource consuming of D-FDTD, here is a part of the above mentioned presentation by Taflove:
###################
"• Twisted waveguide was designed with ProE and
imported into the D-FDTD mesh generator.
• Typical mesh for a 4-period twisted waveguide
included 50,000 modified FDTD grid edges, and
was created in 5 minutes.
• Provided error detection for meshing irregularities,
and a C++ visualization tool.
• HFSS™ required 500 MB of memory and 4 hours
for the solution of a 3-period twisted waveguide.
• D-FDTD required 20 MB of memory and
30 minutes for the same solution."
########################
Truly yours
wave-maniac
On "This means you have accepted that D-FDTD is accurate for PEC."
Yes, I accept. The issue is the relative cell size used compared with a staircasing analysis. I don't know of any geometry for which staircasing gives un-acceptable error even with very small cell size, while DFDTD performs much better. I am aware of the comment that there is such a possibility.
----------------------------
On 'it seems that you think it's possible, but not necessarily, that D-FDTD is accurate for dielectric surfaces. It's much better than what you had calimed "decisively" before.'
To let everyone see how "decisively" I was, I would simply copy my "claim" from above: Dey-Mittra algorithm works fine for some curved PEC boundary problems. The situation is different for irregular dielectric body. The problem (which one is best) is still open.
I am sure if fine enough discretization is used in D-FDTD, the result would be accurate for almost all of the practical problems.
----------------------------
On " [6] was published in 1999, but the final form of D-FDTD, presented by YU-Mittra, which considers dielectric surfaces [11], was published in 2001. "
One difference is that in Dey-Mittra's papers, they addressed the issue of stability by giving out some guidelines and recognized the need for reduced time step. On the contrary, in Yu's scheme, timestep is relaxed to 0.995 of CFL limit and everything just works without any penalty!
It seems to me the authors of [6] are more concerned with the stability and accuracy issue, I will not judge the quality of a research simply by the date it is published.
-----------------------------------
On "So if you just believe in Railton proficiency in FDTD, but
neither Taflove nor Weiland, unless you ask him directly, you can not get rid of doubts."
The best way is test the algorithm on different problems. My comments above are largely based upon my experience testing different conformal algorithms.
----------------------------------------------------------------------
On "So, you claim unless YU provides data on convergence report, you're not going to accept it from anyone else, even Taflove[13] or Railton [6]. It seems you don't like or believe selectively that part of [6] on the convergence study. So you even don't accept the comparison of D-FDTD and normal FDTD, which uses staircasing, presented in any of those papers, even by Yu, et al."
I guess you have not tested the software provided by Yu et al.
-------------------------------------------------
On "Thanks for bringing my attention to fuzzy cell in XFDTD. I read its manual. You're wrong. It uses average volume concept which is exactly that of Dey-Mittra extension method to dielectric surfaces [4] and [5].
It is zero-order accuracy. Yu-Mittra improved the accuracy by considering a first order accuracy [11] ."
Well, I was not wrong when I said the concepts are similar. What do you mean by "considering a first order accuracy"? The Yu et. al. formula for effective dielectric constant for the cell is first order accurate, or more accurate than the one based on volume average? It is not.
-------------------------------------------
On "So, you accept CST is better than XFDTD. Also you claimed XFDTD is better than CFDTD. So the logical result is that you believe CST method (i.e. FIT) is better than CFDTD, particularly D-FDTD. BUT, Weiland of CST says FIT, as used in their products has short comings and they have targeted CFDTD methods to solve the problem, and they consider D-FDTD as a promising one."
Download the CFDTD and try it for yourself.
--------------------------------------------------
on "as to resource consuming of D-FDTD....."
Recall that, when making the comparison with XFDTD, I was talking about the software downloaded from the link Cross provided (also in your comment).
Dear loucy,
-------------------------- IMPORTANT STUFF ------------------------------------
You mentioned:
*****************************
1) Yes, I accept. The issue is the relative cell size used compared with a staircasing analysis. I don't know of any geometry for which staircasing gives un-acceptable error even with very small cell size, while DFDTD performs much better. I am aware of the comment that there is such a possibility.
*****************************
Thanks for your honesty in our scientific discussion.
You mentioned:
*****************************
2) I am sure if fine enough discretization is used in D-FDTD, the result would be accurate for almost all of the practical problems.
*****************************
Thanks again for your honesty, even for this weak-form approval.
You mentioned:
*****************************
3) One difference is that in Dey-Mittra's papers, they addressed the issue of stability by giving out some guidelines and recognized the need for reduced time step. On the contrary, in Yu's scheme, timestep is relaxed to 0.995 of CFL limit and everything just works without any penalty!
*****************************
Cool. You got it. These are the stuff I had mentioned before. So as you see, even that very small penalty, as mentioned by Railton in [6], has been resolved.
-------------------------- NOT IMPORTANT STUFF --------------------------------
You mentioned:
*****************************
4) It seems to me the authors of [6] are more concerned with the stability and accuracy issue, I will not judge the quality of a research simply by the date it is published.
*****************************
Another misunderstanding by you again. My dear friend, who is stupid enough to do that? What I mentioned was in reply to what you said, i.e.
?So the D-FDTD is not necessarily accurate for dielectric surfaces, because this case was not analyzed in [6].?
I meant, how do you expect that the final extension of D-FDTD to dielectric surfaces should have been mentioned in [6], whereas it was published two years later.
You mentioned:
*****************************
5) The best way is test the algorithm on different problems. My comments above are largely based upon my experience testing different conformal algorithms.
*****************************
If you mean developing a code based on that algorithm is your best method, I should say it?s not mine. But if you mean testing accessible codes developed by others using the same algorithm, it?s cool. Actually, as it?s not possible for me to assess the validity, accuracy and versatility of all the algorithms by developing codes, if possible, I first try to dig into the literature to find papers by other guys, particularly well-known experts, about the same subject. Also I try to find various implementations of that algorithm to test, particularly commercial ones, or at least gather info about those implementations. Then I try to judge about that. For this specific case, it seems my method was more reliable than yours, considering your initial comments in this thread.
The problem with in-house developed codes is that the possibility of hidden-bugs (causing inaccuracy) or non-optimality as regards to resource consumption is much more than those by others who have some guys to do their task. One should be careful not to consider the shortcomings due to those algorithms.
In three parts, in reply to my comments, You mentioned:
*****************************
6-1) I guess you have not tested the software provided by Yu et al.
6-2) Download the CFDTD and try it for yourself.
6-3) Recall that, when making the comparison with XFDTD, I was talking about the software downloaded from the link Cross provided (also in your comment).
*****************************
I mentioned the software, not as the best implementation of D-FDTD method, but to show you at least there is an implementation of the Dey-Mittra algorithm, which takes care of both PEC and dielectric surfaces. Thus you could actually test it. By no means, I consider it as the best, or even a good implementation. I would be glad to give you the link to download Taflove code on D-FDTD if I had it. Taflove presentation easily indicates there is at least a much more efficient way of implementing the same algorithm. That?s very important for me. It?s not impossible that if some expert code developers use the algorithm in a hopefully commercial product, it would be even more resource and user friendly. There is a hope that CST and/or Zeland might do that.
BTW, WIPL-D has sort of the same weakness as to implementation. However I believe it could be dramatically improved and be one of the best moment method codes.
You mentioned:
*****************************
7) Well, I was not wrong when I said the concepts are similar. What do you mean by "considering a first order accuracy"? The Yu et. al. formula for effective dielectric constant for the cell is first order accurate, or more accurate than the one based on volume average? It is not.
*****************************
Please have another look at those papers: When using the average volume concept, such as in XFDTD, the effective permittivity is used for updating all the E-fields assigned to that cell. This is a very basic, hence zero-order accuracy method as it does not consider the geometry of the fillings completely. Yu-Mittra modified it, i.e. the effective permittivities which are defined based on edges which have intersected dielectric filling surfaces, are used for updating just those E-fields assigned to the same edges not the non-intersecting edges.
Actually these are sort of similar anti-aliasing in computer graphics.
It was a good discussion. I think that's enough for me. I'm going to design a very compact antenna using GA and FDTD.
Good luck
wave-maniac
Dear Wave-maniac,
It seems that my reply (repeated below) has lead you to conclude that I accept the Yu procedure as a good conformal FDTD algorithm. It is not. The point is that Yu relax the timestep limit without any justification, without a careful stability and accuracy test. The procedure is a little different from those in the Dey-Mittra papers Let's not mix them together simply because of the familiar name of Prof. Mittra.
Let us wait for the fidelity4.0 and see what is implemented there.
------------------
" One difference is that in Dey-Mittra's papers, they addressed the issue of stability by giving out some guidelines and recognized the need for reduced time step. On the contrary, in Yu's scheme, timestep is relaxed to 0.995 of CFL limit and everything just works without any penalty!
*****************************
Cool. You got it. These are the stuff I had mentioned before. So as you see, even that very small penalty, as mentioned by Railton in [6], has been resolved. "
Dear friends!
It has been released now :) :)
Best regards,
Eirp
The major new features of the IE3D 10.0 are: Improved Green's
functions for higher efficiency and accuracy; Improved 3D Modeling
capability; Introduction of the Advanced Extension de-embedding
scheme with the highest flexiblity and ultra wide application
frequency range; Electromagnetic Synthesis of Filters (on FilterSyn);
Much enhanced GUI with new layer system; Robust and complete near
field calculation and visualization; General Pattern Calculation
with Changeable Excitations and Zc; Automatic Creation of Vias on
Import; LEF/DEF (from Cadence) importing and exporting; Improved
bundled circuit simulator MODUA with higher capability in display
and visualization; User Programmable Ie3d Object for Ultimate
Flexibility in Electromagnetic Optimization (More Powerful than
Script Language); Adaptive Distributed Electromagnetic Simulation
on ZDS.
3. The major new features of the FIDELITY 4.0 are: Complete Re-Written
Objected Oriented, Multiple Domain Based Conformal FDTD engine and
GUI; Support of Single and Double Precision; Implementation of
Flexible Source Types including Clock Waveform and User Defined
Waveform; Implementation of ACIS Compatible objects with ACIS
Import/Export Capability; Completely Renovated GUI with full OpenGL
support; Implementation of Polygon Drawing and Importing from IE3D
files; Improved bundled circuit simulator;
4. The MDSPICE 3.2 features 3rd generation s-parameter based SPICE
simulation engine for robust time domain simulation; integrated
schematic editor; frequency-dependent coupled transmission line
SPICE simulation; wide band equivalent cirucit extraction
in SPICE format; Eye-pattern display; Wide-band RLC extraction
from s-parameters.
5. We also release the COCAFIL 1.1, a synthesis program for
Coupled Cavity Rectangular Waveguide Filters.
6. The LineGauge is a TLN calculator for wide variety of TLN. The
LineGuage 9.2 allows automatic creation of TLN segments for IE3D 10.
7. The FilterSyn module for IE3DLIBRARY for synthesis of planar filters.
The FilterSyn 10 features 25 kinds of pre-defined filters and many
coupler and hybrid models. It performs real time synthesis of filters.
The sythesized geometry can be exported into IE3DLIBRARY (or MGRID)
with variables defined for IE3D full-wave simulation and optimization.
who knows which conformal algorithm is used in fidelity 4? I didn't find an answer from the help file.
However, I did notice in summary of chpt 10, there goes:
"In the current Fidelity version, conformal FDTD is implemented only for metal structures, hence we can only use non-conformal meshing and conventional FDTD for dielectric objects. Fortunately, for curved dielectric objects the staircase error is much smaller than in metal cases, and in our conventional FDTD some approximations are used to reduce such an error."
I think fidelity4 won't be a top gun. CST MWS is still ahead in terms of conformal modeling.
I have an old version, but cannot use it , who have the training book for it ?