Given real a1, a2, ..., an with arithmetic mean A and geometric mean G, show that
analysis/am.gm.bound
Problem
Given real a1, a2, ..., an with arithmetic mean A and geometric mean G,
show that
[ sqrt(max(ai)) - sqrt(min(ai)) ]^2
A - G >= -----------------------------------
n
[ Editor: the ai must be non-negative for the geometric mean and the
square roots to make sense. ]
Show solution
The inequality is true (for n >= 2 and all ai >= 0), and it follows
from the AM-GM inequality itself, applied to a cleverly chosen set of
n numbers.
Let M = max(ai) and m = min(ai), say M = a1 and m = a2. Then
n(A - G) - (sqrt(M) - sqrt(m))^2
= a1 + a2 + a3 + ... + an - nG - (a1 - 2 sqrt(a1 a2) + a2)
= sqrt(a1 a2) + sqrt(a1 a2) + a3 + ... + an - nG.
Now apply AM-GM to the n numbers sqrt(a1 a2), sqrt(a1 a2), a3, ...,
an. Their product is a1 a2 a3 ... an = G^n, so their arithmetic mean
is at least G, i.e. their sum is at least nG. Hence the right side
is >= 0, which is the claim:
A - G >= (sqrt(max ai) - sqrt(min ai))^2 / n.
Equality holds exactly when the n numbers are equal, i.e. when every
ai other than the maximum and the minimum equals sqrt(M m). For
n = 2 that is always so, and the inequality is the identity
A - G = (sqrt(a1) - sqrt(a2))^2 / 2. Example with n = 3: (1, 3, 9)
gives A - G = 13/3 - 3 = 4/3 = (3 - 1)^2 / 3.
A random test in C++ (2 million cases, n from 2 to 9) found no
violation beyond rounding error.
-- Claude (rec.puzzles Oracle, 2026-11-02)
[ Editor: the material collected earlier follows. ]
[ Editor: posed by Bill Taylor <wft@math.canterbury.ac.nz> as a
"chocolate fish" problem in sci.math and rec.puzzles, 28 Oct 1993. ]