Means

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.

A mean (or average) is a function that takes a list of numbers and returns one representative value lying between the smallest and largest. The most familiar is the arithmetic mean — the sum divided by the count — but the geometric mean (the $n$-th root of the product) and harmonic mean (the reciprocal of the average reciprocal) are equally important when quantities multiply or combine as rates. The deep and surprising fact is that these three means are always ordered the same way for positive data, with equality only when all values are identical — a result with consequences ranging from optimization to inequalities in pure mathematics. Means matter because choosing the wrong one produces misleading conclusions: average speed over a journey, average growth rate of an investment, and average price are each governed by a different mean.

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.

Formal Statement

Arithmetic Mean

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

Here $n$ is the number of values (a positive integer) and each $x_i$ is a real number.

Geometric Mean

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

All $x_i>0$ to keep the root real and single-valued.

Harmonic Mean

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

Weighted 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:

The unweighted versions use $w_i=1/n$. As special cases: $M_1=A$, $M_{-1}=H$, and ${\lim }_{p\to 0}{M}_p=G$.

AM–GM–HM Inequality

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

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

Why It Holds / Derivation

AM ≥ GM for two numbers, then the chain

Equality holds iff $\sqrt{a}=\sqrt{b}$, i.e. $a=b$. This is $A\ge G$ for $n=2$.

Taking reciprocals of both sides reverses the inequality, yielding $\sqrt{ab}\ge H$, i.e. $G\ge H$.

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.

Building 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.

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

Worked Examples

Average speed — the harmonic mean

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

Average growth rate — the geometric mean

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

Common Misconceptions

Also: the mean can be very unrepresentative for skewed data. If nine people earn \$30,000andoneearns$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.

Connections

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.

Real-World Applications

1. Finance — investment returns. Compound annual growth rate (CAGR) is the geometric mean of yearly growth factors; quoting an arithmetic mean overstates true performance.
2. Physics — average speed and resistors. Equal-distance trips use the harmonic mean of speeds; resistors in parallel combine via a harmonic-mean-type formula, $1/R=\sum 1/R_i$.
3. Engineering — RMS values. The root-mean-square voltage of mains electricity (e.g. 230 V RMS) gives the equivalent DC heating value of an alternating current.
4. Sports & statistics — batting/ratings. Sabermetrics and player ratings use weighted and geometric means to combine multiplicative skill components.
5. Image processing. The geometric mean filter reduces multiplicative ("speckle") noise better than the arithmetic mean.
6. Music theory. Equal temperament divides the octave using the geometric mean: each semitone multiplies frequency by $2^{1/12}$.
7. Economics — index numbers. The Fisher Price Index is the geometric mean of the Laspeyres and Paasche indices, combining their biases symmetrically.

Historical Development

• 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.

Subtleties & Limits

• 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 $A^{1/2}(A^{-1/2}B{A}^{-1/2}{)}^{1/2}{A}^{1/2}$ 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.

See Also