To help visualize how extremely wrong this is, I did another graph. Also, in this graph I've changed the notes to use h instead of b/2.

So again we have the case of h=3. In addition, I've graphed the case where h=1.

In that case, when x = 2, the linear attenuation has already reached 0 - no light is transmitted to the camera.

Whereas, the exponential decay can be seen to get close to 0, but not until the distance is at least 8 times the half distance.

That's a huge difference. With the standard use of linear attenuation, you're completely unable to see anything from x = 2 to x = 8, things you should be seeing in the render.

Note I haven't talked about units here. The units really don't matter, but let's use some units that mean something to us. Suppose h = 100 feet. Then the linear attenuation makes everything invisible after 200 feet. But with the correct attenuation, we should still be able to easily see things out to 600 feet, and if we look hard, we should be able to detect objects out to 800 feet.

If you're used to metric, then imagine 200 meters versus 800 meters.

Whatever units you're used to thinking in, it should be intuitively obvious that this is a serious departure from reality.

So what do we do about it?

---

Renderosity forum reply notifications are wonky. If I read a follow-up in a thread, but I don't myself reply, then notifications no longer happen AT ALL on that thread. So if I seem to be ignoring a question, that's why. (Updated September 23, 2019)