AM–GM Inequality¶

For any nonempty finite list of nonnegative real numbers, its geometric mean does not exceed its arithmetic mean, with equality exactly when all entries are equal.

Version: v1 · 2026-10-03 · History
Domain-specific #: 12978
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Inequalities → Mathematics
Aliases: Inequality of arithmetic and geometric means

Core Idea¶

The arithmetic–geometric mean inequality compares two summaries of the same finite list of nonnegative real numbers. For $n\geq1$ and $a_1,\ldots,a_n\geq0$, set $$ A=\frac{a_1+\cdots+a_n}{n},\qquad G=(a_1\cdots a_n)^{1/n}. $$ Then $G\leq A$, and equality holds if and only if every $a_i$ is equal. In particular, a zero entry does not take the theorem outside its domain: it makes $G=0$, with equality only if every entry is zero. Eberl's original formalization proves the weighted statement for nonnegative reals and gives this equal-weight result as a corollary.^[1]

The inequality is an order theorem with a sharp equality case, not a preference for uniformity in every application. It converts a relation between an additive average and a product-root average into conditional deductions: with a fixed nonnegative sum $S$, the product is at most $(S/n)^n$; with a fixed positive product $P$, the sum is at least $nP^{1/n}$. Both conclusions require their stated feasible domains. A logarithmic proof via concavity of $\log$ applies only when all $a_i>0$; the zero-inclusive theorem needs a separate zero case or a limiting argument, not $\log0$.^[1]

Structural Signature¶

Sig role-phrases: finite nonnegative inputs → arithmetic aggregation → geometric aggregation → order and equality criterion.

Finite nonnegative inputs. A nonempty list and its length $n$ define the common objects being summarized. Nonnegativity gives a real nonnegative $n$th-root product for all $n$. The unweighted relation is not a claim about arbitrary signed inputs.^[1]
Arithmetic aggregation. $A$ divides the sum of those entries by $n$. It is the additive comparator, not an independently sampled or differently weighted benchmark.^[1]
Geometric aggregation. $G$ takes the $n$th root of the product of exactly the same entries. A zero factor gives $G=0$ even though logarithmic expressions cease to be defined.^[1]
Order and equality criterion. The theorem is $G\leq A$, with $G=A$ exactly when the list is constant. The equality condition is what makes fixed-sum and fixed-positive-product extremal deductions sharp.^[1]

What It Is Not¶

It is not a universal comparison of any two quantities called means. Weighted AM–GM uses weights $w_i\geq0$ summing to one; it is a generalization, whereas this identity fixes weights at $1/n$. Quasi-arithmetic means are generated by a strictly monotone transform and do not, as such, state this two-mean order relation. The nearby Log-Sum Inequality has logarithmic ratio terms and different hypotheses; an application may connect these theorems without making them duplicates.^[1]

It is not a theorem about the geometric mean of signed net investment returns. A net rate $r=-50\%$ is negative, but its gross factor $1+r=0.5$ is nonnegative. Compounding multiplies gross factors. If a factor is zero, the theorem still works although a log-return proof does not; if a proposed factor is negative, the real nonnegative-input statement is not licensed. Likewise, $A\geq G$ does not say a higher arithmetic return forecasts a higher compound return for another sequence.^[1]^[2]

Nor does the equality case prove that equal physical allocations are optimal under arbitrary side constraints, heterogeneous values, or different objectives. The fixed-sum product bound applies when the relevant entries really have a common nonnegative sum and the equal point is feasible. A rectangle area argument satisfies those conditions; a constrained engineering or welfare problem may not.^[1]^[3]

Scope of Application¶

In elementary geometry, two-variable AM–GM makes the equal-sided rectangle the greatest-area rectangle among those with a given positive perimeter. The perimeter fixes the side sum; the area is their product. This is a direct derivation from Björner and Stanley's two-variable inequality, not a claim that their chapter studies that rectangle optimization.^[3]

In financial-return description, the arithmetic mean of yearly net rates and the annualized compound net rate are distinct summaries. CFA Institute Research Foundation's long-run tables place arithmetic and geometric return summaries side by side; its US real-equity series for 1900–2015 reports 8.3% arithmetic and 6.4% geometric. The mathematical bridge is to use nonnegative gross factors $g_i=1+r_i$, apply AM–GM to them, then subtract one from each mean. This is a retrospective measurement relation, not investment advice or a prediction of future performance.^[2]^[1]

The same theorem can bound products of any nonnegative list with a known sum or sums with a known positive product. Its content is unchanged across settings, but neither an optimization objective nor a domain-specific interpretation of the entries is built into the theorem.^[1]

Clarity¶

For two entries, the inequality is especially transparent: $(x+y)/2\geq\sqrt{xy}$ follows from $(x-y)^2\geq0$. Equality in that square occurs only at $x=y$. Björner and Stanley give this argument for $x,y\geq0$; Eberl's formal proof gives the general finite-list statement and equality condition.^[3]^[1]

The $n$-variable form must keep three qualifiers visible: the entries are nonnegative, both means use those same entries with equal weights, and equality requires all entries to coincide. A phrase such as “the arithmetic mean is larger” is false at equality. A phrase such as “take logs and average” silently loses the zero entries, even though the inequality itself includes them.^[1]

Manages Complexity¶

AM–GM replaces a product constraint with a sum-based bound. For a known $S=\sum_i a_i$, the one-line conclusion $\prod_i a_i\leq(S/n)^n$ avoids a separate search over every distribution of the entries. It also identifies the possible equality configuration. The calculation is valid only for the feasible list under consideration; it does not certify that the equal list satisfies extra constraints one has not modeled.^[1]

In a sequence of nonnegative gross factors, $(\prod_i g_i)^{1/n}$ compresses a multi-period product into a per-period equivalent factor. AM–GM tells exactly how that compound-path summary relates to the ordinary period average. The resulting inequality is a mathematical invariant; financial interpretation still needs a common period, consistent return definitions, and the actual factor sequence.^[1]^[2]

Abstract Reasoning¶

To use the theorem, first identify a nonempty finite list of nonnegative real entries and verify that the putative arithmetic and geometric quantities arise from that same list. Form $A$ and $G$, assert $G\leq A$, and then check whether equality would require an equal-entry list. If an extremum is sought, solve the inequality for the desired sum or product and test whether the equality list belongs to the feasible set.^[1]

For positive entries, concavity of $\log$ provides an alternative route: $n^{-1}\sum_i\log a_i\leq\log(n^{-1}\sum_i a_i)$, followed by exponentiation. But that displayed logarithm is undefined if any $a_i=0$. The proof strategy must not shrink the theorem's domain: in the zero case, the product-root is zero and the arithmetic mean is nonnegative, with equality precisely when all entries are zero. Eberl's formal nonnegative statement supplies the full-domain check.^[1]

Knowledge Transfer¶

The portable mathematical move is to recognize when one task's “additive total” and another task's “multiplicative result” are summaries of identical nonnegative entries. In geometry, side lengths give a fixed sum and an area product. In finance, gross factors give an ordinary period average and a compounded product. The formula transfers, but the interpretations and feasibility questions do not: a side sum is not a financial forecast, and a realized return sequence is not a design variable to equalize retroactively.^[3]^[2]

One can seek a wider, cross-domain abstraction of additive-versus-multiplicative aggregation, but the AM–GM node remains a particular real-inequality theorem with a domain and equality condition. That portable skeleton is a future-prime question, not evidence that a currently live prime already supplies a strict genus or that the equation holds for unlike mathematical types.^[1]

Examples¶

Fixed-perimeter rectangle¶

Take a rectangle of perimeter $20$ with nonnegative side lengths $x,y$, so $x+y=10$. Björner and Stanley's two-variable inequality gives $\sqrt{xy}\leq(x+y)/2=5$; squaring nonnegative sides yields area $xy\leq25$. Equality requires $x=y=5$, the square. A physical nondegenerate rectangle has $x,y>0$; the zero-side boundary is mathematically permitted and has area zero. This numerical geometry case is our derivation from the sourced theorem.^[3]

Mapped back: Finite nonnegative inputs → side lengths $x,y$; arithmetic aggregation → $(x+y)/2=5$ fixed by perimeter; geometric aggregation → $\sqrt{xy}=\sqrt{\text{area}}$; order and equality criterion → $\sqrt{\text{area}}\leq5$, hence area at most $25$, attained only at $x=y=5$.

Two-period gross-return summary¶

As an illustrative calculation, suppose two successive period net rates are $+50\%$ and $-50\%$. Their gross factors are $g_1=1.5$ and $g_2=0.5$. The arithmetic gross mean is $(1.5+0.5)/2=1$, so the arithmetic net mean is $0\%$. The geometric gross mean is $\sqrt{1.5\cdot0.5}=\sqrt{0.75}\approx0.866$, so the equivalent compounded net rate per period is about $-13.4\%$. The strict gap follows because the gross factors differ. CFA's published use of arithmetic and geometric return summaries motivates the setting; this particular two-period pair is constructed, not an historical observation or financial recommendation.^[2]^[1]

Mapped back: Finite nonnegative inputs → gross factors $1.5,0.5$ rather than signed net rates; arithmetic aggregation → gross mean $1$ and net mean $0\%$; geometric aggregation → gross mean $\sqrt{0.75}$ and net compound rate about $-13.4\%$; order and equality criterion → $\sqrt{0.75}<1$, with equality impossible for unequal factors.

Structural Tensions¶

T1: Additive-period summary versus multiplicative-path summary. Arithmetic averaging is appropriate when the question is the average of separate period rates; treating it as a compounded path can overstate the realized growth factor. Geometric averaging captures the path product, but substituting it for an arithmetic summary changes the question. The choice is not merely stylistic: a constructed sequence of gross factors $(1.5,0.5)$ has arithmetic mean $1$ and geometric mean about $0.866$, while $(0.95,0.95)$ has lower arithmetic mean $0.95$ but higher geometric mean $0.95$. Ranking sequences by one mean need not rank them by the other. Diagnostic: Are the entries being summarized separately or multiplied through time?^[1]^[2]

Structural–Framed Character¶

AM–GM lies near the structural end within a typed real-mathematics frame. Evaluative weight: $G\leq A$ is a theorem, not a preference for equality; the optimality of an equal allocation depends on an application's objective. Human-practice dependence: people choose what entries represent, but the relation is not established by convention once the nonnegative reals and means are fixed. Institutional origin: named proof traditions and formalization archives document the result; no institution makes a tuple satisfy it. Vocabulary travel: “mean” and “balance” appear in many fields, yet only the specified arithmetic and geometric operations on the same list instantiate this node. Import versus recognition: applying it in geometry or finance requires checking nonnegative entries, common weighting and the output interpretation, not merely noticing two quantities called averages.^[1]^[3]^[2]

Its character: a sharply structural mathematical relation whose real-number domain, two means and equality case remain constitutive; its applications are framed by chosen entries and feasible constraints. That makes it a domain-specific formal abstraction, not a substrate-free prime.

Structural Core vs. Domain Accent¶

The core is the finite nonnegative list, its equal-weight additive and multiplicative means, their order, and equality exactly at a constant list. Rectangle perimeter and area, or annual return factors and compound growth, are domain accents: they furnish meanings for inputs and outputs but do not alter the theorem. Fixed-sum maximum-product and fixed-positive-product minimum-sum claims follow by rearrangement; they are consequences, not additional membership roles.^[1]

The more portable idea “compare additive and multiplicative aggregates” may be a future-prime question. Live prime Comparison is too broad to give AM–GM's typed order and equality condition, and prime Ratio does not imply them either. Neither is proposed as a strict parent merely because a proof compares two means. A log-sum relation can derive or be derived in suitable settings, but its different relata do not make it the same node.^[1]

Comparison describes bringing items into a common frame but does not require an arithmetic-versus-geometric ordering; Ratio describes one possible numerical representation, not this theorem. Live Log-Sum Inequality is a distinct mathematical inequality; Isoperimetric Inequality involves planar area and boundary length rather than the general two-mean relation. The rectangle application is not an identity merger with that planar theorem.

Neighborhood in Abstraction Space¶

AM–GM Inequality sits in a sparse region of the domain-specific corpus (71^st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Codes, Matrices & Combinatorial Problems (30 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Weighted AM–GM: assigns general nonnegative weights summing to one; this node is its equal-weight special case, not an alias for the general theorem.^[1]
A logarithmic proof: valid directly only at positive inputs; the theorem also includes zeros.^[1]
An empirical return prediction: the inequality orders summaries of one known nonnegative gross-factor sequence, not two future investments or signed net percentages.^[2]^[1]
A blanket equalization prescription: the fixed-sum equality tuple need not be feasible or desirable under side constraints and alternative objectives.^[1]
Log-Sum or Structural Inequality: respectively a different formal logarithmic bound and an institutional-social phenomenon, neither an exact AM–GM duplicate.

References¶

[1] Manuel Eberl, “Pólya’s Proof of the Weighted Arithmetic–Geometric Mean Inequality”, Archive of Formal Proofs (2022), abstract and unweighted corollary; author-original formal proof. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27

[2] David Chambers and Elroy Dimson, eds., Financial Market History, CFA Institute Research Foundation (2016), Chapter 1 appendix introduction, PDF p. 37, and Appendix 1.2, PDF p. 39; original published analysis of arithmetic versus annualized geometric return summaries. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h

[3] Anders Björner and Richard P. Stanley, A Combinatorial Miscellany (2010), Chapter 6, Eq. (6.3), PDF p. 55; author-original two-variable proof from $(x-y)^2\geq0$. registry ↩a ↩b ↩c ↩d ↩e ↩f