Helpful person <rrllff@yahoo.com> wrote in news:3cc0512f-a127-4ca8-8cf3-
ea94cff7ef4b@b29g2000hsa.googlegroups.com:

There was definately an accuracy problem in the old days of double
precision ForTran programs. However, although this should not still
be the case I have found it to be so.

Haven't they heard of arbitray precision? A fractal program I had
from the old days of DOS could do arbitrary precision to mind-numbingly
large values that were only limited by your CPU power and patience.
Back in the days of 386's, you had to be REALLY patient.

Brian

Helpful person wrote:
I believe that thin films with absorbing (metal) layers are not the
same if reversed.

As Prof. S. pointed out, you have to replace loss with gain when you do
the time reversal.

Cheers,

Phil Hobbs

In article
c76cef5f-e264-4f26-be29-e759e9033...@h11g2000prf.googlegroups.com>,
 Helpful person <rrl...@yahoo.com> wrote:

I'm not familiar with Prof S., probably after my time.

I'm afraid I was probably *before* your time . . .

   --Prof S.  (still around, though fading steadily)

Then you must remember using Conrady's formulae for ray tracing using

I'm not familiar with Prof S., probably after my time.

Except that linear gain (e.g., laser gain) has to be turned into linear
loss, and v.v., on time reversal -- which can make for some interesting
and even controversial discussions when one or both are present.

Entropy?

[But despite that, if anyone can point to a good, clear textbook
discussion of time-reversal symmetry in e-m and optics problems, that's
a favorite of theirs, and that specifically includes loss and gain, I'd
be glad to hear of it. I'm pretty sure I understand the topic, but I'd
be glad to check my understanding against someone else.]

Then you must remember using Conrady's formulae for ray tracing using
Monroe and Marchant calculators. (Just a little before my time.)

Skywise wrote:
Phil Hobbs <p...@SpamMeSenseless.pergamos.net> wrote in
news:47BDC86B.4050502@SpamMeSenseless.pergamos.net:

Okay, since we've done the OP's homework for him: Since ray tracing is
time reversal invariant, including scattering and thin film effects, why
doesn't backward ray tracing accurately render the visual image of a
caustic or a colour shift?
snip
In POVRay, the software I've been playing with for years, caustics
used to be simulated and were confined to being within the shadow
of the object.

Now, real caustics can be computed. The software first does a
relatively low resolution forward trace from the light source,
through the object casting the caustic, and onto any surfaces it
then encounters. This information is stored and used to modify the
results of the backward rays later on. Now, when we trace the above
example ray to the surface, there is information about the result
of the caustics. Due to the resolution of the forward ray image,
the result may be interpolated.

This technique can be utilized to simulate quite realistic scenery.
I myself once made a rotating diamond animation that was just
stunning. All the sparkles, reflections, color refraction, caustics,
etc... were calculated with amazing reality.

The price you pay is that the more realistic you want the caustics
there's an exponential cost in CPU time.

Okay, so if I understand this correctly, the problem is not so much the
caustics themselves as it is the diffuse scattering surface that makes
them visible. To model the caustic with the same accuracy as the main
ray trace, you'd need as many rays per unit area on the diffuser as on
the eyeball, which could get expensive. To do it efficiently, you'd
need to trace forwards to get the caustic, and then trace backwards to
see what it looks like to the eye.

Cheers,

Phil Hobbs

Phil Hobbs <pcdh@SpamMeSenseless.pergamos.net> wrote in
news:47BDC86B.4050502@SpamMeSenseless.pergamos.net:

Okay, since we've done the OP's homework for him: Since ray tracing is
time reversal invariant, including scattering and thin film effects, why
doesn't backward ray tracing accurately render the visual image of a
caustic or a colour shift?
snip
In POVRay, the software I've been playing with for years, caustics
used to be simulated and were confined to being within the shadow
of the object.

Now, real caustics can be computed. The software first does a
relatively low resolution forward trace from the light source,
through the object casting the caustic, and onto any surfaces it
then encounters. This information is stored and used to modify the
results of the backward rays later on. Now, when we trace the above
example ray to the surface, there is information about the result
of the caustics. Due to the resolution of the forward ray image,
the result may be interpolated.

This technique can be utilized to simulate quite realistic scenery.
I myself once made a rotating diamond animation that was just
stunning. All the sparkles, reflections, color refraction, caustics,
etc... were calculated with amazing reality.

The price you pay is that the more realistic you want the caustics
there's an exponential cost in CPU time.

Skywise wrote:
Helpful person <rrllff@yahoo.com> wrote in news:3cc0512f-a127-4ca8-8cf3-
ea94cff7ef4b@b29g2000hsa.googlegroups.com:

There was definately an accuracy problem in the old days of double
precision ForTran programs. However, although this should not still
be the case I have found it to be so.

Haven't they heard of arbitray precision? A fractal program I had
from the old days of DOS could do arbitrary precision to mind-numbingly
large values that were only limited by your CPU power and patience.
Back in the days of 386's, you had to be REALLY patient.

Brian

You don't know patience until you tried doing math on an 8080 at 2 MHz!

Dave
.... kids these days. Humph!

Okay, so if I understand this correctly, the problem is not so much the
caustics themselves as it is the diffuse scattering surface that makes
them visible. To model the caustic with the same accuracy as the main
ray trace, you'd need as many rays per unit area on the diffuser as on
the eyeball, which could get expensive. To do it efficiently, you'd
need to trace forwards to get the caustic, and then trace backwards to
see what it looks like to the eye.

can any on explain what this phenomenon actually is and why its better
than forward ray tracing.

In article
1541eaea-7a55-4ec7-bcc8-9a0bf8593...@q33g2000hsh.googlegroups.com>,
 Helpful person <rrl...@yahoo.com> wrote:

Then you must remember using Conrady's formulae for ray tracing using
Monroe and Marchant calculators.  (Just a little before my time.)

Nope -- but only because I've never personally done ray tracing.  

But there was indeed a room upstairs in the lab with half a dozen to a
dozen young women (at least, almost all women) staffers sitting at huge
Monroe and Marchant calculators doing numerical integration,
differential equation solving, and the like for grad student and faculty
researchers.

[It was next door to the "Drafting Room", with the big drawing tables
and draftsmen, making plots of the results .]