AoPSWiki
Want to learn how to tackle those tough AMC/AIME/Olympiad algebra problems? Check out Art of Problem Solving's Intermediate Algebra by Richard Rusczyk and Mathew Crawford. Over 1600 problems!
Personal tools

1989 AIME Problems/Problem 11

From AoPSWiki

Problem

A sample of 121 integers is given, each between 1 and 1000 inclusive, with repetitions allowed. The sample has a unique mode (most frequent value). Let D be the difference between the mode and the arithmetic mean of the sample. What is the largest possible value of \lfloor D\rfloor? (For real x, \lfloor x\rfloor is the greatest integer less than or equal to x.)

Solution

Let the mode be x, which we let appear n > 1 times. We let the arithmetic mean be M, and the sum of the numbers \neq x be S. Then

\begin{align*}D &= \left|M-x\right| = \left|\frac{S+xn}{121}-x\right| = \left|\frac{S}{121}-\left(\frac{121-n}{121}\right...

As S is essentially independent of x, it follows that we wish to minimize or maximize x (in other words, x=1,1000). Indeed, D(x) is symmetric about x = 500.5; consider replacing all of numbers x_i in the sample with 1001-x_i, and the value of D remains the same. So, without loss of generality, let x=1. Now, we would like to maximize the quantity

\frac{S}{121}-\left(\frac{121-n}{121}\right)(1) = \frac{S+n}{121}-1

S contains 121-n numbers that may appear at most n-1 times. Therefore, to maximize S, we would have 1000 appear n-1 times, 999 appear n-1 times, and so forth. We can thereby represent S as the sum of n-1 arithmetic series of 1000, 999, \ldots, 1001 - \left\lfloor \frac{121-n}{n-1} \right\rfloor. We let k = \left\lfloor \frac{121-n}{n-1} \right\rfloor, so

S = (n-1)\left[\frac{k(1000 + 1001 - k)}{2}\right] + R(n)

where R(n) denotes the sum of the remaining 121-(n-1)k numbers, namely R(n) = (121-(n-1)k)(1000-k).

At this point, we introduce the crude estimate[1] that k=\frac{121-n}{n-1}, so R(n) = 0 and

\begin{align*}2S+2n &= (121-n)\left(2001-\frac{121-n}{n-1}\right)+2n = (120-(n-1))\left(2002-\frac{120}{n-1}\right)

Expanding (ignoring the constants, as these do not affect which n yields a maximum) and scaling, we wish to minimize the expression 5(n-1) + \frac{36}{n-1}. By AM-GM, we have 5(n-1) + \frac{36}{n-1} \ge 2\sqrt{5(n-1) \cdot \frac{36}{n-1}}, with equality coming when 5(n-1) = \frac{36}{n-1}, so n-1 \approx 3. Substituting this result and some arithmetic gives an answer of \boxed{947}.


In less formal language, it quickly becomes clear after some trial and error that in our sample, there will be n values equal to one and n-1 values each of 1000, 999, 998 \ldots. It is fairly easy to find the maximum. Try n=2, which yields 924, n=3, which yields 942, n=4, which yields 947, and n=5, which yields 944. The maximum difference occurred at n=4, so the answer is 947.

Notes

  • ^ In fact, when n = 2,3,4,5 (which some simple testing shows that the maximum will occur around), it turns out that \frac{121-n}{n-1} is an integer anyway, so indeed k = \left\lfloor \frac{121-n}{n-1} \right\rfloor = \frac{121-n}{n-1}.

See also

1989 AIME (ProblemsResources)
Preceded by
Problem 10
Followed by
Problem 12
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Want to learn how to tackle those tough AMC/AIME/Olympiad counting and probability problems? Check out Art of Problem Solving's Intermediate Counting & Probability by David Patrick.
© Copyright 2008 AoPS Incorporated. All Rights Reserved. • FoundationPrivacyContact Us