Monday, March 1, 2010

Phase 5: Poking holes in a square

Man, trying to keep myself to these Sunday night deadlines is a hassle. It feels like it's always Sunday.

"But Kevin, it 330am. Doesn't that mean it's Monday?"

Don't be ridiculous, silly reader. Don't be ridiculous.

This week, we're going to take a little excursion. Instead of talking about some more set theory or abstract algebra, let's play a little game.

Say we start out with a square of side-length L, then inscribe a circle and cut out the circle's area. Next we inscribe a square into that circle hole and add that area back to the total. We have a square again, so let's cut out another circle. Then add another square. Then subtract yet another circle. Then add a smaller square... On and on, til we have something that looks like this:



The question is: If the process of subtracting circles and adding squares continues to itsy, bitsy, infinitesimal, microscopic mothaflankin' Planck-like lengths, what total area does our final figure assume?

Let's define an iteration of our process as "add 1 square, subtract its inscribed circle."

So, for example, iteration 0 is to begin with a square of side-length L and subtract its inscribed circle of radius r=L/2.

I hate to say it, but: just in case I lost you on that radius, check out this figure:



So the first area, let's denote is A0, we find to be:
A0 = square(0) - circle(0)
= L2 - πr2
= L2 -  π(L/2)2
= L2 - (π/4)L2
= (1 - π/4)L2


What is the area after a second iteration?

To answer this, we first must figure out what the side-length of the new inscribed square should be. Just note that its corner points are touching the circle: this gives us the length of its diagonal! It is two times the radius of the circle--and since the circle's radius is L/2, that makes the new square's diagonal equal to L.

We can now find the side length of square(1) using the good ol' Pythagorean Theorem, which is formally:
c2 = a2 + b2


...where c2 is the hypotenuse, while a and b are the triangle's legs. In our case, c = L and a = b:
L2 = a2 + a2 = 2a2


So the side length of square(1) is L/√2.



What will circle(1)'s radius be? Again, it's half the length of the square it's inscribed in, so it is a/2 = L/(2√2).



The area upon the second iteration then will be
A1 = A0 + Square(1) - circle(1)
= A0 + a2 - π(a/2)2
= A0 + (1/2)L2 - (π/8)L2
= (1 - π/4 + 1/2 - π/8)L2


For now on, let's just say that L = 1, then we don't have to worry about carrying it around.

Upon iteration 2, we would find:
A2 = A1 + 1/4 - π/16
= 1 - π/4 + 1/2 - π/8 + 1/4 - π/16


Similarly,
A3 = A2 + 1/8 - π/32
= 1 - π/4 + 1/2 - π/8 + 1/4 - π/16 + 1/8 - π/32


There's a pattern emerging here. We really don't have to do much more work to extrapolate this trend towards infinity. Let's regroup the terms of A3:

A3 = (1 + 1/2 + 1/4 + 1/8) - (π/4 + π/8 + π/16 + π/32)
= (1 + 1/2 + 1/4 + 1/8) - (π/4)( 1 + 1/2 + 1/4 + 1/8)
= (1-π/4)(1 + 1/2 + 1/4 + 1/8)


Observe:





(1/2)0 = 1
(1/2)1 = 1/2
(1/2)2 = 1/4
(1/2)3 = 1/8

So, we can simplify the sum using "sigma summation notation":
A3 = (1-π/4)∑(1/2)n,

where the sum runs from n=0 to n=3. (The capital sigma let's us know to SUM over n; n is called the summation index.)


Notice that it will now have this same form for every iteration. For example, the nth iteration will have total area given by:

An = (1-π/4)∑(1/2)n (where the sum runs from j=0 to j=n)


Does this sum converge as n goes to infinity? The answer is yes. You can prove it using a technique from analysis called the "root test." Or, perhaps you know that this is a well-known type of series called a "geometric series," which is convergent as n goes off into the distant lands, ever-seeking that holy grail.

So what is the area of that object with all the weird holes in it? Here's a little table of some numerical values for the total area upon successive iterations:








A00.214601837
A10.321902755
A2 0.375553214
A30.402378444
A4 0.415791058
A5 0.422497366
A6 0.42585052
A7 0.427527096
A8 0.428365385

We can pretty much see that it's going to end up somewhere around 0.43 --- that is, 43% the area of the original square.

Using some knowledge about the geometric series, I can tell you that the series ∑(1/2)n approaches 2, so our series (1-π/4)∑(1/2)n approaches (1-π/4)(2) = 2 - π/2, which equals 0.429203673 out to nine decimal places.

So we were pretty much right.

We rule.