adda
adda copied to clipboard
adda-team
Radiation forces
Completely rewrite calculation of radiation forces using FFT. Add its
memory requirements to -prognose estimates, and perform other integration
into the code.
Original issue reported on code.google.com by yurkin on 26 Nov 2008 at 8:18
- Blocking: #178
Currently calculation of radiation forces is very unstable (crashes adda on
Win32
from time to time). Further investigation is required to understand why (and
from
which version) this happens. Then a fix is required (without rewriting the
code).
Original comment by yurkin on 25 May 2010 at 2:54
Original comment by yurkin on 7 Jun 2010 at 5:47
- Added labels: Priority-High
- Removed labels: Priority-Medium
Random crashes, described in comment 1, were fixed by r972.
r972 - 17de61acb040087c9cc4c940cd6f9290bb32b20d
Original comment by yurkin on 10 Sep 2010 at 9:43
It is also desirable to make output of total optical torques (additionally to
the total force). It was proposed in
http://groups.google.com/group/adda-discuss/browse_thread/thread/620b7f91e053678
1# .
for torques see #181
Original comment by yurkin on 12 Jan 2011 at 8:18
Original comment by yurkin on 12 Jan 2011 at 8:18
- Added labels: Usability
A workaround for computing Cpr (if only it is needed) without computing the
radiation force at all is discussed in
http://groups.google.com/group/adda-discuss/browse_thread/thread/f2de36035198e3c
a/6f353589944a0518
Original comment by yurkin on 29 Jun 2011 at 5:11
Can any one tell me the unit of the force on each dipole found at the file
VisFrp?
Original comment by [email protected] on 15 Sep 2011 at 4:18
ADDA internally uses Gaussian-CGS system. So if incident field amplitude is 1
in this system and all length variables are measured in cm, the unit of the
force (for values in file VisFrp) will be dyne. However, if particle dimensions
are specified in um (typical) then the units of force will be 10^-4 dyne.
If you prefer SI units, then for incident field amplitude equal to 1 V/m and
particle dimensions specified in um, the unit of force is approximately
1.11*10^-18 N.
Finally, I encourage one to check the above statements for some test problem,
since this feature of ADDA (in its current form) was implemented long ago and
was not well documented. And I personally have almost no experience with it.
Original comment by yurkin on 17 Sep 2011 at 8:16
thanks Yurkin,
I have one more question, how is the incident field amplitude defined in ADDA?
Original comment by [email protected] on 19 Sep 2011 at 7:30
For a plane wave, the incident field amplitude is given by |e^0| in Eq.(12) in
Section 9 ("DDA formulation). It is assumed to be equal to one. For Gaussian
beam the amplitude is its value in the beam focus - see Section 8.2 ("Beam
types) of the manual.
Original comment by yurkin on 19 Sep 2011 at 11:23
Hallo,
I did not get how you got those numbers because, the force is proportional to
the product of square of electric field and Cpr, thus if every thing is in cgs
and the dimension is in µm the force will be 10^-8 dyne (because Cpr is
expressd in µm^2)similarly the force will be 4 X 10^-15 N (where every thing
is defined in cgs).
besira
Original comment by [email protected] on 27 Sep 2011 at 3:47
Besira, sorry for delay with answer.
I agree that I made a mistake in comment 8. I thought that electric field is
not affected by change of linear dimension, while it does (as cm^-0.5). This
can be easily checked by running ADDA with "-size 1 -lambda 1" and "-size
0.0001 -lambda 0.0001" and comparing the output. The latter is 10^8 times
smaller. So indeed the unit of force for unity amplitude of electric field (in
Gaussian-CGS system, 1 statV/cm) and µm dimensions is 10^-8 dyne.
Concerning the SI units, 1 V/m corresponds to 0.33*10^-4 statV/cm (see e.g.
http://en.wikipedia.org/wiki/Gaussian_units ) and force is proportional to
electric field squared. So the latter coefficient should be squared, multiplied
by 10^-8 and 10^-5 (from dyne to N). Finally, for incident field amplitude
equal to 1 V/m and particle dimensions specified in µm, the unit of force is
approximately 1.11*10^-22 N.
Please let me know if you agree with the last paragraph.
Original comment by yurkin on 9 Oct 2011 at 5:57
Hallo,
nice to see your reply.
I still has to disagree with the second part. Because if we want to express the
force in dyne then the electric field has to be expressed in statvolt/cm and
distance in cm. Now F(dyn)= 1/8pi X E(statvol/cm)X Cpr(cm). Since you menitoned
that ADDA by default works with cgs units(1 statvolt/cm of E should be rather
assumed), we can express force in dyne and change it to SI unit at the end,
F(dyn)= (proportional to) 1 statvolt/cm X 10e-8 µm2 = 10e-8dyne= 10e-13 N,
if one include 1/8pi constant F(N)= 4 x 10e-15 N. Thus the force obtained in
calculation has to be multiplied by 10e-13 N to express them in SI unit.
Besira
Original comment by [email protected] on 10 Oct 2011 at 7:49
Dear Yukrin,
I have attached a document with the previous comment and I would be very happy
if you have time to see it.
besira
Original comment by [email protected] on 10 Oct 2011 at 7:51
After discussion with Besira, we concluded that expressions given in comment 12
are correct.
Also I have summarized this discussion and earlier comments in the manual. Now
the latter contains a meaningful section on radiation forces. It will be
available with release 1.1.
Original comment by yurkin on 14 Oct 2011 at 8:35
Bonjour,
I would like to raise question on radiation force on each dipole with focused
laser beam. As expressed in the article of Hoekstra et al
(J.Opt.Soc.Am.A/Vol.18/August 2001), a plane wave is considered and it is
mentioned that it also works for other incident beams but the expression for
<Finc, i> (section 4) need to be adapted. I was wondering how ADDA works when I
choose a gaussian beam and calculate the force, does it use the adapted version
of the <Finc,i>?
Original comment by [email protected] on 28 Oct 2011 at 8:42
Besira, you are right, ADDA works wrong with respect to <Finc,i> for Gaussian
beams - see issue 135. Until this issue is addressed it is hard to get
reasonable results from ADDA for this case. The only possible workaround is by
some manual correction of the produced fields on each dipole.
There is also another relevant problem for Gaussian beams - see issue 134.
#134, #135
Original comment by yurkin on 2 Nov 2011 at 10:32
Hallo,
it is bad news for me!!
Can I ask you to explain how to correct it manulally and whether it address the
problem address in issues 134 or not?
Thanks in advance
#134
Original comment by [email protected] on 2 Nov 2011 at 12:29
Unfortunately, issue 134 and issue 135 should be handled separately, since they
required different properties of the (complex) Gaussian beams. I have added
information about possible workarounds as comments to these issues. Completely
fixing any of this issue requires significant effort, so I am not sure how long
it will take.
#134, #135
Original comment by yurkin on 2 Nov 2011 at 4:40
It seems that MPI version of radiation forces calculation was not working at
all. With r1083, it is almost as good (or bad) as the sequential code. The only
remaining difference is generation of VisFrp files, which is incomplete in
parallel (MPI) mode.
r1083 - 2d45e2be79ea912d377c075862c99e50c4de0222
Original comment by yurkin on 22 Nov 2011 at 6:39
The problem mentioned in previous comment was fixed by r1142.
r1142 - 094d4f3d3c5af895291b167a008fd907a7965c31
Original comment by yurkin on 22 May 2012 at 5:27
Original comment by yurkin on 25 Sep 2013 at 10:28
- Now blocking: #178
As mentioned above, the description of the FFT-accelerated calculation of the radiation forces is available in A. G. Hoekstra, M. Frijlink, L. B. F. M. Waters, and P. M. A. Sloot, “Radiation forces in the discrete-dipole approximation,” J. Opt. Soc. Am. A 18, 1944–1953 (2001).
However the code itself is lost in time. A bit more details can be found in the thesis of Martin Frijlink. It is attached as pdf file, but with missing figures (they are given in a separate archive) figures.zip thesis_Frijlink.pdf
More recently, this issue has been discussed by Chaumet et al.:
- P. C. Chaumet, A. Rahmani, A. Sentenac, and G. W. Bryant, “Efficient computation of optical forces with the coupled dipole method,” Phys. Rev. E 72, 046708 (2005).
- P. C. Chaumet and C. Billaudeau, “Coupled dipole method to compute optical torque: Application to a micropropeller,” J. Appl. Phys. 101, 023106 (2007).
- P. C. Chaumet and A. Rahmani, “Electromagnetic force and torque on magnetic and negative-index scatterers,” Opt. Express 17, 2224–2234 (2009).
Another workaround is to compute the gradients of both incident and internal fields outside of ADDA during post-processing, which then can also be additionally processed to compute the radiative forces. This idea was used in combination with the DDSCAT in Bonessi D., Bonin K., and Walker T. Optical forces on particles of arbitrary shape and size, J. Opt. A 9, S228–S234 (2007). However, there are some issues with such gradient calculations near the interfaces. Still, the error probably decreases with dipole size.