Means

⚡ Key Takeaways

A mean is a single number that summarizes a set of values by representing their "center" or "typical" magnitude. The three classical Pythagorean means — arithmetic, geometric, and harmonic — each capture different notions of average, and are ordered $HM \le GM \le AM$ for positive numbers. Means appear everywhere from statistics and finance to physics and music theory; they were studied by the ancient Greeks (Pythagoreans, ~500 BCE) and remain foundational across all quantitative fields.

💡 The Core Insight

A mean is not one thing but a family of "centering" operations. Each mean answers a specific question: the arithmetic mean answers "what value, repeated, gives the same sum?"; the geometric mean answers "what value, repeated, gives the same product?"; the harmonic mean answers "what value, repeated, gives the same sum of reciprocals?" The correct mean is dictated by which combination of the data is physically or mathematically meaningful.

DefinitionArithmetic Mean

For real numbers $x_1, x_2, \dots, x_n$ , the arithmetic mean is their sum divided by their count:

1

\quad A = \bar{x} = \frac{1}{n}\sum_{i=1}^{n} x_i = \frac{x_1 + x_2 + \cdots + x_n}{n}

Derivation hint: It is the unique value $A$ such that replacing every $x_i$ by $A$ leaves the total sum $\sum x_i$ unchanged, i.e. $nA = \sum x_i$ . Here $n$ is the number of values (a positive integer) and each $x_i$ is a real number.

DefinitionGeometric Mean

For positive real numbers $x_1, \dots, x_n$ , the geometric mean is the $n$ -th root of their product:

2

\quad G = \left(\prod_{i=1}^{n} x_i\right)^{1/n} = \sqrt[n]{x_1 \cdot x_2 \cdots x_n}

Derivation hint: It is the unique positive value $G$ such that replacing every $x_i$ by $G$ leaves the total product $\prod x_i$ unchanged, i.e. $G^n = \prod x_i$ . All $x_i > 0$ to keep the root real and single-valued.

DefinitionHarmonic Mean

For positive real numbers $x_1, \dots, x_n$ , the harmonic mean is the reciprocal of the arithmetic mean of the reciprocals:

3

\quad H = \frac{n}{\displaystyle\sum_{i=1}^{n} \frac{1}{x_i}} = \frac{n}{\frac{1}{x_1} + \frac{1}{x_2} + \cdots + \frac{1}{x_n}}

Derivation hint: It is the value $H$ such that replacing every $x_i$ by $H$ leaves the total $\sum 1/x_i$ unchanged, i.e. $n/H = \sum 1/x_i$ .

DefinitionWeighted and Generalized (Power) Mean

Attaching weights $w_i > 0$ with $\sum w_i = 1$ , and a real parameter $p \ne 0$ , the generalized (power) mean unifies them all:

4

\quad M_p = \left(\sum_{i=1}^{n} w_i\, x_i^{\,p}\right)^{1/p}

Derivation hint: Substitute $p=1$ for arithmetic, $p=-1$ for harmonic; the geometric mean is the limit as $p \to 0$ , obtained by taking logarithms and applying L'Hôpital's rule. The unweighted versions use $w_i = 1/n$ . As special cases: $M_1 = A$ , $M_{-1} = H$ , and $\lim_{p\to 0} M_p = G$ .

TheoremAM–GM–HM Inequality

For any positive real numbers $x_1, \dots, x_n$ ,

5

\quad H \le G \le A

with equality holding simultaneously throughout if and only if $x_1 = x_2 = \cdots = x_n$ .

ProofAM ≥ GM for two numbers, then the chain

Step 1 — AM ≥ GM for two positive numbers. Let $a, b > 0$ . Since the square of any real number is non-negative,

6

\quad \left(\sqrt{a} - \sqrt{b}\right)^2 \ge 0

Derivation hint: Any squared real quantity is $\ge 0$ , with equality iff the quantity is zero.

Step 2 — Expand. Expanding gives $a - 2\sqrt{ab} + b \ge 0$ , hence

7

\quad \frac{a+b}{2} \ge \sqrt{ab}

Derivation hint: Add $2\sqrt{ab}$ to both sides of the expanded inequality and divide by 2. Equality holds iff $\sqrt{a} = \sqrt{b}$ , i.e. $a = b$ . This is $A \ge G$ for $n=2$ .

Step 3 — GM ≥ HM. Apply the just-proved $A \ge G$ to the reciprocals $1/a, 1/b$ :

8

\quad \frac{\frac{1}{a}+\frac{1}{b}}{2} \ge \sqrt{\frac{1}{ab}} = \frac{1}{\sqrt{ab}}

Derivation hint: Use $A \ge G$ on the data set $\{1/a, 1/b\}$ , then take reciprocals (which reverses the inequality). Taking reciprocals of both sides reverses the inequality, yielding $\sqrt{ab} \ge H$ , i.e. $G \ge H$ .

Step 4 — General $n$ . For general $n$ , $A \ge G$ follows by Cauchy's "forward–backward" induction or by the convexity of the exponential function (Jensen's inequality applied to logarithms). Applying that result to reciprocals again gives $G \ge H$ , completing the chain $H \le G \le A$ . $\blacksquare$

∎

mathematician · Pythagoras of Samos | c. 570 – c. 495 BCE

The three classical means are called Pythagorean because the Pythagorean school studied them in connection with music and proportion. They observed that string-length ratios producing harmonious intervals (the octave 2:1, the fifth 3:2, the fourth 4:3) are linked by these means: 2, the harmonic mean of 1 and 3 ... no — concretely, the numbers 6, 8, 9, 12 form a "musical proportion" where 8 is the harmonic mean and 9 the arithmetic mean of 6 and 12. This fusion of arithmetic, geometry, and music ("the quadrivium") shaped mathematics for two millennia.

🧠 Intuition

Picture the data as physical objects.

Arithmetic mean is the balance point (center of mass) of equal weights placed at each value on a number line. Pile the total amount, level it out — that flat level is $A$ .
Geometric mean is the side of a square (or edge of a cube) with the same area (volume) as the rectangle built from the data. For two numbers $a, b$ , $G = \sqrt{ab}$ is literally the side of a square equal in area to an $a \times b$ rectangle.
Harmonic mean governs rates and parallel combinations. If you drive one mile at speed $a$ and one mile at speed $b$ , your average speed is the harmonic mean — because what's fixed is distance, not time, so the slow leg dominates.

The ordering $H \le G \le A$ reflects that the harmonic mean is "pulled toward small values," the arithmetic mean "pulled toward large values," and the geometric mean sits balanced in the multiplicative middle.

Figure

The three Pythagorean means of two numbers a and b shown on a semicircle

ExampleAverage speed — the harmonic mean

Given: A cyclist rides 12 km uphill at 8 km/h, then the same 12 km back downhill at 24 km/h. Find the average speed for the whole trip.

Step 1 — Total distance: $12 + 12 = 24$ km.

Step 2 — Total time: uphill $\frac{12}{8} = 1.5$ h; downhill $\frac{12}{24} = 0.5$ h; total $= 2.0$ h.

Step 3 — Average speed $= \dfrac{\text{distance}}{\text{time}} = \dfrac{24}{2.0} = 12$ km/h.

Check with the harmonic mean (valid because each leg covers equal distance):

H = \frac{2}{\frac{1}{8} + \frac{1}{24}} = \frac{2}{\frac{3}{24} + \frac{1}{24}} = \frac{2}{\frac{4}{24}} = \frac{2 \cdot 24}{4} = 12 \text{ km/h.}

Note the naive arithmetic mean $\frac{8+24}{2} = 16$ km/h is wrong.

\boxed{\bar v = 12 \text{ km/h}}

ExampleAverage growth rate — the geometric mean

Given: An investment grows by $+100\%$ in year 1 (multiplier $2.0$ ), then loses $50\%$ in year 2 (multiplier $0.5$ ). Find the average annual growth factor.

Step 1 — Overall factor over 2 years: $2.0 \times 0.5 = 1.0$ (you end where you started).

Step 2 — Geometric mean of the factors:

G = \sqrt{2.0 \times 0.5} = \sqrt{1.0} = 1.0.

So the true average annual growth factor is $1.0$ , i.e. 0% per year — correct, since the money is unchanged.

Contrast: The arithmetic mean of the factors is $\frac{2.0 + 0.5}{2} = 1.25$ , falsely suggesting +25%/year. The geometric mean is the only correct average for multiplicative growth.

\boxed{G = 1.0 \text{ per year } (0\% \text{ growth})}

⚠️ Common mistake

The arithmetic mean is not always the "right" average. Using it for rates (speeds, prices per unit) or for multiplicative growth (returns, ratios) gives systematically wrong answers. Match the mean to the structure: additive quantities → arithmetic; multiplicative quantities → geometric; rates over a fixed numerator → harmonic. Also: the mean can be very unrepresentative for skewed data. If nine people earn $30,000 and one earns $3,000,000, the arithmetic mean income is $327,000 — a figure no one actually receives. Here the median (a different notion of center) is more honest.

Counterexample

"The geometric mean equals the arithmetic mean for typical data." False — they are equal only when all values are identical. Take $\{1, 9\}$ :

A = \frac{1+9}{2} = 5, \qquad G = \sqrt{1 \cdot 9} = 3, \qquad H = \frac{2}{1 + \frac19} = \frac{2}{10/9} = 1.8.

Here $H = 1.8 < G = 3 < A = 5$ , strictly — a single concrete witness that the AM–GM–HM inequalities are generally strict, and that the gaps widen as the data spread.

connection

Means tie together many areas:

The AM-GM Inequality is a cornerstone of optimization and is itself a special case of Jensen's Inequality, which applies to all convex functions.
The Median and Mode are alternative measures of central tendency, more robust to outliers; together with the mean they form the trio taught in Descriptive Statistics.
The geometric mean is the arithmetic mean in logarithmic space: $\ln G = \frac{1}{n}\sum \ln x_i$ , linking means to Logarithms.
The Root Mean Square (quadratic mean, $p=2$ ) sits above the arithmetic mean in the Power Mean Inequality hierarchy and governs the standard deviation and AC electrical signals.
Means underlie the Expected Value of a random variable in Probability Theory.

🔬 Where this appears

⏳ History

- c. 500 BCE — Pythagorean school identifies the arithmetic, geometric, and harmonic (then "subcontrary") means via musical ratios (6, 8, 9, 12).

- c. 350 BCE — Archytas of Tarentum and later commentators define and distinguish the three classical means precisely.

- c. 300 BCE — Euclid's Elements (Book VI) gives the geometric mean as the "mean proportional," constructible with compass and straightedge.

- c. 100 CE — Nicomachus and Pappus of Alexandria catalogue up to ten distinct means, formalizing the Greek theory of proportions.

- 1729 — Colin Maclaurin and contemporaries develop inequalities among symmetric functions, foreshadowing the power-mean hierarchy.

- 1821 — Augustin-Louis Cauchy proves the general AM–GM inequality by his elegant "forward–backward" induction.

- 1906 — Johan Jensen publishes Jensen's inequality, revealing AM–GM as a consequence of convexity.

- 1934 — Hardy, Littlewood, and Pólya's Inequalities gives the definitive modern treatment of power means and their ordering.

open-problem

Subtleties that survive into research mathematics:

The full power mean $M_p$ is monotonically increasing in $p$ : $p < q \Rightarrow M_p \le M_q$ . Proving this in full generality, and characterizing the limiting cases $p \to \pm\infty$ (which give the maximum and minimum), requires convexity arguments.
Means of matrices are subtle: the geometric mean of two positive-definite matrices $A, B$ is not simply $\sqrt{AB}$ (which need not be symmetric); the correct Riemannian/geometric mean $A^{1/2}(A^{-1/2}BA^{-1/2})^{1/2}A^{1/2}$ and its extension to three or more matrices was only resolved in the 2000s.
Generalized means on infinite sets / measures connect to deep questions in functional analysis and ergodic averaging, where "the mean of a function" (its integral average) can fail to exist or depend on the averaging procedure.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also