No negative radius spheres #1428
Conversation
In order to avoid confusion between constructor parameter names and
member variables, we either used alternate names for the parameters or
prefixed the parameter names with an underscore. For example:
class foo {
public:
foo(int _i, bool semaphore, short bar) : i(_i), flag(semaphore)
{
baz = bar;
}
private:
int i;
bool flag;
short baz;
};
However, this is unnecessary. Constructor initializer lists are
unambiguous, as the assigned item always refers to the class member, and
the assigned value is always from the parameter (unless prefixed with
"this->").
Inside the constructor body, parameter names override member variables,
and "this->" can be used to indicate the class member in the case of
name collision.
Thus, the above code can be rewritten as this:
class foo {
public:
foo(int i, bool flag, short baz) : i(i), flag(flag)
{
this->baz = baz;
}
private:
int i;
bool flag;
short baz;
};
I've taken advantage of this to clarify names used in constructors
throughout the codebase. Mostly for constructor parameter names,
sometimes for member variable names.
We were using negative radii on spheres to invert the sense of front vs back faces of spheres, and using this to get an inverted index of refraction (IOR) for bubbles. This method was used to model a hollow glass sphere. Over the past couple of years we've encountered a number of bugs induced by negative radii spheres (see the list in the description for issue 1420). Instead, this change interprets the dielectric parameter not as an IOR relative to surrounding air, but instead as the ratio of the IOR of the object material over the IOR of the surrounding medium. Interpreting the parameter this way lets you model a sphere of air enclosed in glass -- just what we need to model a hollow glass sphere purely with the proper IORs, and without resorting to inverted geometry with negative radii spheres. We also end up using this method to properly demonstrate total internal/external reflection in another section. Resolves #1420
books/RayTracingInOneWeekend.html
Outdated
| However, properly handling negative radii can be tricky. Recall the line from `sphere::hit()` in | ||
| listing [sphere-material] that calculates the outward normal: | ||
| The outer sphere is just modeled with a standard glass sphere, with an index of refraction of around | ||
| 1.50. The inner sphere is a bit different, because _its_ index of refraction will now be relative to |
There was a problem hiding this comment.
Some optional small suggestions for this paragraph in bold:
The outer sphere is just modeled with a standard glass sphere, with an index of refraction of around 1.50 relative to the surrounding air. The inner sphere is a bit different because its index of refraction should be relative to the surrounding glass outer sphere and model a transition back to air. This is actually simple ...
There was a problem hiding this comment.
I'll tweak the wording.
There was a problem hiding this comment.
Hmmm, I barely tweaked it:
The outer sphere is just modeled with a standard glass sphere, with an index of refraction of around
1.50. The inner sphere is a bit different, because its index of refraction should be relative to
the "glass atmosphere" of the surrounding outer sphere. This is actually simple to specify, as our
I don't want to add "relative to the surrounding air" since it's really a surrounding vacuum, (air = 1.000293), and most refractive indices are given with an implicit understanding that they're already relative to vacuum. Also, for the second change it's actually transitioning back into glass.
Either way, I'm going to add a bit more explanation earlier where we introduce refraction.
There was a problem hiding this comment.
I agree, vacuum is more correct. I only suggested air because it was used elsewhere in the book. My main point was to highlight that the refractive index is always relative to the surrounding medium (while a reader may assume its absolute meaning, i.e. relative to vacuum). I think mentioning "relative to the surrounding air/vacuum" in the first sentence sets things up for the second sentence but please use whichever phrasing you prefer.
... Also, for the second change it's actually transitioning back into glass.
Doesn't a transition from the inner sphere's exterior (as defined by the outward facing normals) to its interior represent a transition from glass to air/vacuum (or whatever interior/exterior medium pair that happens to have the same refraction ratio)? That's really what I meant but I may have misunderstood something.
Either way, I'm going to add a bit more explanation earlier where we introduce refraction.
Sounds good. As I said I think the main concern was to highlight that the refractive index is always relative. Even the absolute one is relative but readers may make assumptions about the definition of IOR and get confused about the meaning of dialectric::ref_index.
There was a problem hiding this comment.
See the next commit and let me know what you think.
Doesn't a transition from the inner sphere's exterior (as defined by the outward facing normals) to its interior represent a transition from glass to air/vacuum (or whatever interior/exterior medium pair that happens to have the same refraction ratio)? That's really what I meant but I may have misunderstood something.
Ah, yes, I misunderstood your suggested reading. I think it was the "back to air" that confused me. I'd rather explain this as a transition into a third medium that just happens in this example to be the same as the first medium. Anyway, I think the paragraph immediately preceding this explains this all in detail, and we don't need to repeat it in this paragraph.
There was a problem hiding this comment.
This?
The outer sphere is just modeled with a standard glass sphere, with an index of refraction of around 1.50 relative to the surrounding air, modeling a refraction from air into glass. The inner sphere is a bit different because its index of refraction should be relative to the surrounding glass outer sphere and model a transition from glass into air. This is actually simple ...
There was a problem hiding this comment.
Kind of wish we had three different materials to keep things straight, but the model itself would then be rather needlessly complicated. Perhaps use the terms "outside air" and "inside air"? ¯\_(ツ)_/¯
|
Ugh. Too many PRs piled up — now this one has a bunch of conflicts. I'll have to create a second one for this, rebased. |
|
Superseded by #1435. |
[This change is stacked on top of the outstanding PR to normalize constructor parameter names. Ignore that commit, as I'll rebase this onto dev when that PR is merged.]
Implement bubbles with IOR, not negative radii
We were using negative radii on spheres to invert the sense of front vs
back faces of spheres, and using this to get an inverted index of
refraction (IOR) for bubbles. This method was used to model a hollow
glass sphere.
Over the past couple of years we've encountered a number of bugs induced
by negative radii spheres (see the list in the description for issue
1420).
Instead, this change interprets the dielectric parameter not as an IOR
relative to surrounding air, but instead as the ratio of the IOR of the
object material over the IOR of the surrounding medium. Interpreting the
parameter this way lets you model a sphere of air enclosed in glass --
just what we need to model a hollow glass sphere purely with the proper
IORs, and without resorting to inverted geometry with negative radii
spheres.
We also end up using this method to properly demonstrate total
internal/external reflection in another section.
Resolves #1420