Elsevier

Computers & Graphics

Volume 34, Issue 2, April 2010, Pages 158-166
Computers & Graphics

Chaos and Graphics
Dimensions and the probability of finding odd numbers in Pascal's triangle and its relatives

https://doi.org/10.1016/j.cag.2009.10.002Get rights and content

Abstract

It is well-known that the subsets of Pascal's triangle consisting of numbers not divisible by a prime p are relatives of the Sieroiski gasket. By computing the dimensions of these subsets, we obtain the puzzling result that in the infinite Pascal's of these subsets, we obtain the puzzling result that in the infinite Pascal's triangle, the probability that any number is not divisible by p is 0, for all primes p. By a more delicate analysis using hierarchical iterated function systems, we show the divisibility result is true when the prime p is replaced by any positive integer r. We give examples of similar results for Pascal's triangle generated by other polynomials.

Introduction

What is the probability of finding an odd number among the positive integers? Surely the answer is 1/2. A careful formulation of this result requires a bit of thought: selected uniformly randomly from 1,2,,2n, the probability of finding an odd number is 1/2. An appropriate limit as n gives the result.

We can ask a similar question of Pascal's triangle. In Pascal's triangle with n rows, what fraction of the entries are odd? What is the limit of this fraction as n? Odd numbers are those not divisible by 2, so we can ask similar questions about the limit of the fraction of Pascal's triangle entries not divisible by 3, not divisible by 5,, not divisible by p for each prime p. What about the limit of the fraction not divisible by prime powers, or by products of prime powers? Refined in the right way, we see that all these questions have the same answer. How can this be?

Our approach is to relate each of these limits to a specific fractal and show all these fractals have dimensions <2. From this, we deduce the answers to the questions posed above.

Often Pascal's triangle is encountered in precalculus classes as a simple method for finding the coefficients of the terms xj in (1+x)n. Taking the top to be the 0th row and the left edge the 0th column, the coefficient of xj in (1+x)n is the j th entry of the n th row. The left side of Fig. 1 shows rows 0 through 8 of Pascal's triangle, the right side shows the row and column numbers.

Almost every introduction to fractals includes the observation that darkening the Pascal's triangle squares containing numbers that are not multiples of 2 gives a pattern like the familiar Sierpinski gasket. In fact, appropriately rescaling the sequence of Pascal's triangle with 2n rows gives a sequence of sets that converges to the Sierpinski gasket. In addition, Pascal's triangle can be formulated using cellular automata and iterated unction systems. See Chapter 8 of [1] for an excellent exposition. Many interesting applications of Pascal's triangle for visualizing constructions in group theory are given in [2]. An early study of the aesthetics of Sierpinski gaskets found in Pascal's triangles is found in [3].

This appearance of the Sierpinski gasket suggests the natural question: for every integer n>1, what pattern results from darkening the Pascal's triangle squares containing numbers that are not multiples of n? Fig. 2 shows the patterns in the first 200 rows for n=2,,13. Some of these are simple fractals, relatives of the Sierpinski gasket. Others are more complicated, still fractal, but not built by a simple iterated function system in any obvious way. A moment's observation suggests how to differentiate these two categories: for n prime, the pattern is a relative of the gasket; for n composite, the pattern is more complicated. More careful inspection differentiates the composites. Prime powers appear to give simpler patterns than do products of distinct primes.

One very interesting consequence of the proliferation of computer graphics is that many people have seen that a few lines of code can generate images whose complex structure contains features unanticipated by the programmer. Manual drawings rarely revealed such surprises; with computers, mathematics has blossomed as an experimental science. Excellent examples can be found in [4], [5], and of course in the Journal of Experimental Mathematics and in the Chaos and Graphics section of Computers & Graphics.

Here our goal is modest. What else can we deduce by looking at the images in Fig. 2?

In the top row of Fig. 3 we see the squares of Pascal's triangle colored green if the number of the square is divisible by 6, blue if divisible by 3 but not by 2, red if divisible by 2 but not by 3, and black if not divisible by 2, 3, or 6. Left to right we see Pascal's triangle with 64, 128, 256, and 512 rows. The amount of green appears to increase with the number of rows in Pascal's triangle. This suggested to us that as Pascal's triangle grows, the fraction of entries not divisible by 2 and 3 goes to 0.

In the bottom row of Fig. 3 we use a different coloring scheme, based on divisibility by 2, 3, or 5. Specifically, squares are colored red if their Pascal's triangle entry is divisible by 2 but not 3 or 5, green if by 3 but not 2 or 5, and blue if by 5 but not 2 or 3; yellow if divisible by 2 and 3 but not by 5, purple if divisible by 2 and 5 but not by 3, and aqua if divisible by 3 and 5, but not by 2; white if divisible by 2, 3, and 5; black if divisible by none of 2, 3, and 5. More white area appears in larger Pascal's triangles, so we expect the fraction of entries not divisible by 2, 3, and 5 goes to 0 as Pascal's triangle grows.

These pictures led to the conjecture that for every integer r>1, as Pascal's triangle grows, the fraction of entries not divisible by r goes to 0. After completing our analysis, we found that this result had been obtained in [6] using algebraic methods. Our approach was motivated by detail found in the images, and is straightforward, given some elementary properties of dimension and measure, reviewed in Section 2.

Section snippets

Review of dimension and measure

For a self-similar set G with similarity transformations Ti satisfying G=T1(G)TN(G)if the Ti(G) are sufficiently disjoint, the Hausdorff dimension dim(G) of G is the unique solution d of the Moran equationr1d++rNd=1where ri is the contraction factor of Ti. That the Ti are sufficiently disjoint is guaranteed by the open set condition: there is a nonempty bounded open set O for which OT1(O)TN(O)with the unions disjoint. For the gasket, O can be taken to be the interior of its convex hull.

Rescaling Pascal's triangles

We consider a sequence of Pascal's triangles, those having 2,4,8,16,,2n rows, for example, ask what fraction of the entries of these Pascal's triangles are odd, and look for limiting behavior of this fraction as n. Can we use this to deduce the general behavior of the proportion of odd entries in Pascal's triangles? Thinking of odd numbers as those 0(mod2), we can ask what proportion of the entries of Pascal's triangles are 0(mod3)? What proportion are 0(mod5)? . For every prime p, what

Why is the limit the gasket?

That filling in the Pascal's triangle squares containing odd numbers has limiting shape the Sierpinski gasket is a straightforward consequence of a result of Kummer. See Chapter 8 of [1]. With the row and column indexing introduced in Fig. 1, the Pascal's triangle entry in row n and column k is nk. Kummer [11] showed that for each prime p, the number of factors of p in nk is the number of carries that occur when the base p representations of n-k and k are added. For example, 8-2 and 2 have base

The probability of finding odd numbers, and other surprises

We have seen that P(2) is the Sierpinski gasket of dimension log(3)/log(2), hence having area 0. From this, we deduce that the probability of finding an odd number in P(2n) goes to 0 as n. In this section we address how to extend this result to other primes, and how to replace the 2n limit with the m limit.

Odd numbers are just those 0(mod2). For every prime number p, we can look for similar patterns in sPm(p), the squares in the rescaled m-row Pascal's triangle that contain numbers 0(

Prime powers and composite numbers

We have seen that in the infinite rescaled Pascal's triangle, for each prime p the subset corresponding to those numbers not divisible by p has dimension less than 2. Based on this, we say that the probability of finding a number not divisible by p is 0. What about the probability of finding a number divisible by p but not by p2, or divisible by p2 but not by p3, and so on? Or for that matter, what about the probability of finding a number not divisible by r=p1s1pksk? Fig. 3 suggests that all

Some generalizations: other polynomials

One familiar motivation for studying Pascal's triangle is that the coefficients of 1,x,,xn in (1+x)n are the entries of the n th row of Pascal's triangle, taking the apex of the triangle to be the 0th row. Motivated by [15], similar arguments can be carried out for powers of any polynomial. For example, in Fig. 12 we see the first 200 rows of the Pascal's triangle for (1+x+x2)n, shading the squares containing numbers 0(mod2) (left) and 0(mod3) (right). To adapt our arguments to these

Conclusion

While some effort is required to compute the dimensions of the P(r) for any integer r>1 using the hIFS formalism, that computing the dimension is worthwhile is immediately apparent from looking at the pictures of Pascal's triangle, filling all squares containing numbers that are not multiples of r. Common experience with summing the areas of the empty triangles in the base 2 Sierpinski gasket suggests that the empty triangles in P(r) sum to 1/2, so the numbers 0(modr) form a subset of

References (16)

  • S. Willson

    Computing fractal dimensions for additive cellular automata

    Physica D

    (1987)
  • H.-O. Peitgen et al.

    Chaos and fractals. New frontiers of science

    (2004)
  • Shannon K, Bardzell M, Bergner J, Evans T, Fisher M, Ludwick K, et al. PascGalois project....
  • C. Pickover

    On the aesthetics of Sierpinski gaskets formed from large Pascal's triangles

    Leonardo

    (1990)
  • Borwein J, Bailey D. Mathematics by experiment. Plausible reasoning in the 21st century. Natic: A. K. Peters;...
  • Borwein J, Bailey D, Girgensohn R. Experimentation in mathematics. Computational paths to discovery. Natic: A. K....
  • N. Fine

    Binomial coefficients modulo a prime

    Amer Math Monthly

    (1947)
  • K. Falconer

    Fractal geometry. Mathematical foundations and applications

    (1990)
There are more references available in the full text version of this article.

Cited by (0)

1

Partially supported by NSF DMS 0203203.

View full text