In the early 1950s, mathematician Ernst Straus came up with an interesting problem. He wondered whether it would be possible to design a room completely lined with mirrors, but designed in such a way that it would not be 100% illuminated by a single match. For this thought experiment, he assumed that the light from the match could travel infinitely far, but only in straight lines and at mathematically precise angles, with no scattering.
It turns out that there are indeed theoretically unilluminable rooms. The first one was discovered in 1958, by Roger Penrose. It uses the magic of ellipses to guarantee that no matter where the light is ignited, there will be somewhere that remains in shadow. The large ellipses focus light on the O's (located on the lateral surfaces of the apex of the cornua), leaving dark spaces in at least two of the pockets. If the light-bearer stands in one of the pockets, more than half of the room is left in darkness; if she stands in the center of the room, all four pockets are dark. (further illustration). In the illustration below, the light source is marked by the X, and the shadows are marked by dots.
MMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMM
MMMMMMMMMMMMMMMMMMMMMM MMMMMMMMMMMMMMMMMMMMMM
MMMMMMMMMMMMMMMMM MMMMMMMMMMMMMMMMMM
MMMMMMMMMMMMMM MMMMMMMMMMMMMM
MMMMMMMMMMM MMMMMMMMMMMM
MMMMMMMMM MMMMMMMM
MMMMMMM MMMMMMM
MMMMM MMMMM
MMMM X MMMM
MMM MMM
MM MM
MM MM
MM OMMM MMMO MM
MM MMMMM MMMMM MM
MM MMMMMM MMMMMM MM
MMMMMMMMMMMMMM MMMMMMMMMMMMMM
MMMMMMMMMMMMMM MMMMMMMMMMMMMM
MMMMMMMMMMMMMM MMMMMMMMMMMMMM
MMMMMMMMMMMMMM MMMMMMMMMMMMMM
MM .... MMMMMM MMMMMM .... MM
MM .... MMMMMM MMMMMM .... MM
MM .... MMMMM MMMMM .... MM
MM .... OMMM MMMO .... MM
MM MM
MM MM
MMM MMM
MMMMM MMMM
MMMMMM MMMMM
MMMMMMM MMMMMMM
MMMMMMMMM MMMMMMMMM
MMMMMMMMMMM MMMMMMMMMMM
MMMMMMMMMMMMMM MMMMMMMMMMMMMM
MMMMMMMMMMMMMMMMMM MMMMMMMMMMMMMMMMMM
MMMMMMMMMMMMMMMMMMMMMMMM MMMMMMMMMMMMMMMMMMMMMMM
MMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMMM
In 1995 the first polygonal unilluminable room was designed by George Tokarsky, who proved that a light ray starting from one corner of a mirrored right isosceles triangle would never return to that original corner. Armed with this information and some topological trickery, he created a very pointy 26-sided room in which a single light source, placed in just the right spot, leaves a corner in darkness -- but unfortunately this did not hold true for every spot in the room; you could find many places to stand in which you could light to whole room.
ZZZZZZZZZZ +
ZZZZZZZZZZ ZZ+
ZZZZZZZZZZ ZZZ?
ZZZZZZZZZZ ZZZZZ?
ZZZZZZZZZZ ZZZZZZZ
ZZZZZZZZZZ ZZZZZZZZN
?ZZZZZZZZZZZZZZZZZZZZ ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZ ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
+ZZZZZZZZZZZZZZZZZZZZZZZ ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZ ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
NZZZZZZZZZZZZZZZZZZZZZZZZZZ ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZ ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
8ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZD
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ZZZZZZZZM ZZZZZZZZZ ZZZZZZZZ+
ZZZZZZZ ZZZZZZZN ZZZZZZZ
ZZZZZN ZZZZZZ ZZZZZN
ZZZZ ZZZZM ZZZZ
ZZ+ ZZZ ZZN
+ Z? Z
These remain the most well-known of the unilluminable rooms, although in 1997 D. Castro managed to create a polygonal room with only 24 sides (illustration). It also is only 'unilluminable' given certain light source placements, but the resulting areas of shadow are much larger. To the best of my knowledge these rooms remain mathematical constructs, never having been built. Granted, due to the sad absence of perfectly flat mirrors and a scatter-free atmosphere, a real-world room would not have truly unilluminable spaces (given the hypothesized infinitely-strong light source, anyway). Perhaps even more disappointing, I can find no work being done on the smallest possible unilluminable room given two light sources; perhaps thesis material for an up-and-coming topologist?
Other References:
http://plus.maths.org/issue8/xfile/index.html
http://mathworld.wolfram.com/IlluminationProblem.html