\bf Puzzles\hskip1in

Puzzles                    






Puzzle: Why is 1/3 ·3 = 1 on most computers nowdays? for lecture on Jan 18

Concept: IEEE 754 floating point
The representation of the fraction [1/3] is a repeating decimal or binary number, and we know that a computer or calculator is going to truncate after some number of digits or bits. Yet on almost all computers and most calculators 1/3 ·3 is exactly 1. Why?

On a cheap pocket calculator, an expensive calculator, in your favorite programming language, and in your favorite spreadsheet, determine what is the smallest integer i > 0 such that the floating point computation 1/i ·i ¹ 1. (If you do this in a program, try both single and double precision.)

Try to determine the machine e for your devices, which is the smallest number such that 1+e > 1.



Puzzle: Lengthening a rail for lecture on Jan 20

Concept: Arranging computation to avoid subtraction
A one-mile-long straight rail, fixed at its ends, is lengthened by one foot (to 5281 feet) so that it bows up in a circular arc. Find the maximum height, d, without losing 4 or 5 significant digits. (A classic, according to F. S. Acton)



Puzzle: A simple game of ``Life'' for lecture on Jan 25

Concept: The power of invariants
Consider a game board with an n ×n array of squares, k of which are black and the rest white. Two squares are considered neighbors if they common side. (But not if they share a single corner.)

Whenever you find a white square with two or more black neighbors, color it black. When no more squares can be colored, the game is done. (You can consider this a simplified form of Conway's "Game of Life" with no death rule, and 4 neighbors instead of 8.)

What is the smallest value of k for which the game can end with no white squares? Prove it. (I heard this one from Subhash Suri.)



Puzzle: Embedding a given tree on a given set of points for lecture on Feb 1

Concept: An application of convex hulls
Suppose that you are given a set P of n points in the plane, no two of which have the same y coordinates and no three of which lie on a common line. You are also given an abstract tree T with n nodes, one of which is the root, and n-1 edges, all directed away from the root. Show that you can always assign each node of T to a point in P so that the tree can be drawn with all edges directed downward and no edges crossing.



Puzzle: Untangling a polygon for lecture on Feb 3

Concept: An example of geometric duality
Suppose you are given a circular sequence of n points P = (p0,p1,¼,pn), with p0 = pn, which can be interpretted as a (possibly self-intersecting) polygon. Suppose that we are asked to untangle it by using the following uncrossing operation: Find any two (non-adjacent) segments that intersect, say e and f. Remove e and f, and reconnect the polyline so that it remains a single loop. This removes the crossing between e and f, although it may introduce other crossings.

Show that repeated application of the uncrossing operation will produce a simple polygon on the points of P. How many uncrossings are necessary?



Puzzle: Connections and completions? for lecture on Feb 8

Concept: Polygons and triangulations
Three questions on connections and completions:
1)
Can any finite set of points be connected into a simple polygon (with no edges crossing)?
2)
Can any finite set of non-intersecting line segments be connected into a simple polygon?
3)
Can any finite set of non-crossing line segments be completed to a triangulation?



Puzzle: What about the top part of the convex hull? for lecture on Feb 10

Concept: Lifting maps for the Delaunay
If we ``lift'' a set of points p_1,p_2,...,p_n in the plane to the paraboloid by sending (xi,yi) to (xi,yi, xi2+yi2), then take the lower part of the convex hull and project back down to the plane, we get the Delaunay triangulation. What if we take the upper part of the hull?



Puzzle: How many changes in a dynamic triangulation? for lecture on Feb 15

Concept: An open problem
There are open questions about how many times you need to change the triangulation as it moves, whether it be a Delaunay or a lazy triangulation. (See problem set #3.) The answer in the worst case is known only to be somewhere between W(n2) and O(n3). Many people have tried to prove O(n2) for the Delaunay; it might be easier to prove for a lazy triangulation.

For your algorithm, what is the worst configuration you can design. Design configurations that are as bad as possible for your algorithms; how many changes can you get as a function of n?


File translated from TEX by TTH, version 2.60.
On 3 Feb 2000, 05:15.