The Cauchy–Schwarz inequality is the most quietly important inequality in mathematics. It says, in its simplest formulation, that for any two real (or complex) vectors u\mathbf{u} and v\mathbf{v}:

∣⟨u,v⟩∣  ≤  ∥u∥⋅∥v∥.|\langle \mathbf{u}, \mathbf{v} \rangle| \;\leq\; \|\mathbf{u}\| \cdot \|\mathbf{v}\|.

The absolute value of an inner product is at most the product of the lengths. In the familiar Euclidean plane, where ⟨u,v⟩=u⋅v=u1v1+u2v2\langle \mathbf{u}, \mathbf{v} \rangle = \mathbf{u} \cdot \mathbf{v} = u_1 v_1 + u_2 v_2 and ∥u∥=u12+u22\|\mathbf{u}\| = \sqrt{u_1^2 + u_2^2}, the inequality is almost obvious: it is just the statement that ∣cos⁡θ∣≤1|\cos\theta| \leq 1, where θ\theta is the angle between the two vectors.

What is not obvious is that the same inequality holds in every inner-product space — including infinite-dimensional ones, where the vectors are functions, sequences, or random variables, and where the geometric intuition of “angles” is no longer literal. This generality, combined with how often the inequality turns up, makes it the workhorse on which a vast number of subsequent results are built. This article is about why the inequality is true, what it looks like, and where it appears.

The geometry

For vectors u,v\mathbf{u}, \mathbf{v} in the plane, the dot product is

u⋅v  =  ∥u∥∥v∥cos⁡θ.\mathbf{u} \cdot \mathbf{v} \;=\; \|\mathbf{u}\| \|\mathbf{v}\| \cos\theta.

Since ∣cos⁡θ∣≤1|\cos\theta| \leq 1, the inequality ∣u⋅v∣≤∥u∥∥v∥|\mathbf{u} \cdot \mathbf{v}| \leq \|\mathbf{u}\| \|\mathbf{v}\| is immediate. Equality holds precisely when θ=0\theta = 0 or θ=π\theta = \pi — that is, when the two vectors are parallel.

Cauchy–Schwarz: |u · v| = |u| |v| |cos θ| ≤ |u| |v| v u |u| cos θ θ the dot product equals the projection length times |v|; both factors are at most 1 × |u| |v|

The projection of u\mathbf{u} onto v\mathbf{v} — the green segment — has length ∥u∥cos⁡θ\|\mathbf{u}\| \cos\theta. The dot product is this projection length times ∥v∥\|\mathbf{v}\|. Since the projection cannot be longer than u\mathbf{u} itself, the dot product cannot exceed ∥u∥∥v∥\|\mathbf{u}\| \|\mathbf{v}\|. That is the entire picture in two dimensions: the Cauchy–Schwarz inequality is, geometrically, the observation that projections shrink (or preserve) lengths.

The remarkable fact is that the same inequality survives in every inner-product space — including infinite-dimensional ones, like the space of square-integrable functions on an interval, where the inner product is ∫fg dx\int f g \, dx and the norm is ∫f2 dx\sqrt{\int f^2 \, dx}. The “geometry” of these spaces is abstract, but Cauchy–Schwarz says that the geometric intuition of bounded projections continues to hold.

A surprisingly clean proof

The standard proof is elegant and uses only one trick: consider a real parameter tt and look at the non-negative quantity ∥u−tv∥2≥0\|\mathbf{u} - t\mathbf{v}\|^2 \geq 0. Expanding,

∥u−tv∥2=∥u∥2−2t ⟨u,v⟩+t2 ∥v∥2  ≥  0.\|\mathbf{u} - t\mathbf{v}\|^2 = \|\mathbf{u}\|^2 - 2t\,\langle \mathbf{u}, \mathbf{v} \rangle + t^2\,\|\mathbf{v}\|^2 \;\geq\; 0.

This is a quadratic in tt that is always non-negative. For a quadratic At2+Bt+CAt^2 + Bt + C to be always non-negative, its discriminant B2−4ACB^2 - 4AC must be at most zero:

4⟨u,v⟩2−4 ∥u∥2 ∥v∥2  ≤  0.4\langle \mathbf{u}, \mathbf{v} \rangle^2 - 4\,\|\mathbf{u}\|^2\,\|\mathbf{v}\|^2 \;\leq\; 0.

Dividing by 44 and taking square roots gives exactly the Cauchy–Schwarz inequality. The whole proof fits in three lines and uses only properties of inner products that hold in any inner-product space: linearity, positivity of ∥⋅∥2\|\cdot\|^2, and the symmetry ⟨u,v⟩=⟨v,u⟩\langle \mathbf{u}, \mathbf{v} \rangle = \langle \mathbf{v}, \mathbf{u} \rangle.

The same calculation reveals when equality is attained: the discriminant is exactly zero, which means the quadratic in tt has a single real root, which means u−tv=0\mathbf{u} - t\mathbf{v} = 0 for some real tt — that is, u\mathbf{u} and v\mathbf{v} are linearly dependent. So Cauchy–Schwarz becomes an equality precisely when one vector is a scalar multiple of the other.

Applications spread everywhere

The reach of this one inequality is staggering. A short tour through some of its most important applications:

The triangle inequality. The familiar fact that the length of a sum is at most the sum of the lengths — ∥u+v∥≤∥u∥+∥v∥\|\mathbf{u} + \mathbf{v}\| \leq \|\mathbf{u}\| + \|\mathbf{v}\| — follows immediately from Cauchy–Schwarz. Square the left side: ∥u+v∥2=∥u∥2+2⟨u,v⟩+∥v∥2≤∥u∥2+2∥u∥∥v∥+∥v∥2=(∥u∥+∥v∥)2\|\mathbf{u} + \mathbf{v}\|^2 = \|\mathbf{u}\|^2 + 2\langle \mathbf{u}, \mathbf{v}\rangle + \|\mathbf{v}\|^2 \leq \|\mathbf{u}\|^2 + 2\|\mathbf{u}\|\|\mathbf{v}\| + \|\mathbf{v}\|^2 = (\|\mathbf{u}\| + \|\mathbf{v}\|)^2. Taking square roots gives the triangle inequality. The triangle inequality is the defining property of a norm, so Cauchy–Schwarz is literally what makes inner-product spaces into normed spaces.

Correlation in statistics. For random variables XX and YY with finite variances, define the inner product ⟨X,Y⟩=Cov(X,Y)=E[(X−μX)(Y−μY)]\langle X, Y \rangle = \mathrm{Cov}(X, Y) = \mathbb{E}[(X - \mu_X)(Y - \mu_Y)]. The Cauchy–Schwarz inequality then says

Cov(X,Y)2  ≤  Var(X) Var(Y),\mathrm{Cov}(X, Y)^2 \;\leq\; \mathrm{Var}(X)\,\mathrm{Var}(Y),

which is precisely the statement that the correlation coefficient

ρ(X,Y)=Cov(X,Y)Var(X) Var(Y)\rho(X, Y) = \frac{\mathrm{Cov}(X, Y)}{\sqrt{\mathrm{Var}(X)\,\mathrm{Var}(Y)}}

lies between −1-1 and +1+1. Equality holds when XX and YY are linearly related — exactly the case of perfect correlation.

Heisenberg’s uncertainty principle. In quantum mechanics, the relationship between the standard deviations of position σx\sigma_x and momentum σp\sigma_p for a particle is

σx⋅σp  ≥  ℏ/2.\sigma_x \cdot \sigma_p \;\geq\; \hbar/2.

This celebrated inequality is, mechanically, an application of Cauchy–Schwarz to the position and momentum operators acting on the wave function. The inner product is the Hilbert-space inner product on square-integrable wave functions; the bound comes from a clever choice of vectors involving the position and momentum operators. Heisenberg’s principle — the philosophical centrepiece of quantum mechanics — is, in pure mathematical terms, a Cauchy–Schwarz argument.

Information theory. The Cramér–Rao bound — a lower limit on the variance of any unbiased estimator in statistics — comes from applying Cauchy–Schwarz to the score function (the derivative of the log-likelihood) and the estimator. It is the foundation of the theory of statistical efficiency.

Functional analysis. In any Hilbert space, Cauchy–Schwarz is the inequality that makes the geometry work. It implies the boundedness of linear functionals (the Riesz representation theorem), the continuity of the inner product, and the well-posedness of orthogonal projections. Without Cauchy–Schwarz, functional analysis would not get off the ground.

Linear algebra. Norms induced by inner products satisfy the parallelogram identity ∥u+v∥2+∥u−v∥2=2∥u∥2+2∥v∥2\|\mathbf{u} + \mathbf{v}\|^2 + \|\mathbf{u} - \mathbf{v}\|^2 = 2\|\mathbf{u}\|^2 + 2\|\mathbf{v}\|^2 — a structural fact that distinguishes Hilbert spaces from arbitrary normed spaces, and which is bootstrapped from Cauchy–Schwarz.

A unifying principle

The reason Cauchy–Schwarz appears so often is that it is, in essence, the inequality that captures the geometric content of any inner-product structure. Any time you have a notion of “length” and “angle” that is compatible with addition and scaling, Cauchy–Schwarz is going to be lurking. The diversity of its applications — from statistics to quantum mechanics to optimisation to coding theory — is not a coincidence; it is the natural consequence of how often the inner-product structure shows up across mathematics and physics.

What makes the inequality particularly satisfying as a piece of mathematics is the gap between the modesty of its statement and the richness of its consequences. One line of algebra, three lines of proof — and from it follow the triangle inequality, the bounds on correlation in statistics, Heisenberg’s uncertainty principle, the structure of Hilbert spaces, and a great deal more. Cauchy proved the discrete version in 1821 because he needed it for a specific computation; Bunyakovsky and Schwarz extended it to integrals because they needed that for further specific computations. None of them could have anticipated that two centuries later, the inequality would be used so universally that working mathematicians often deploy it without comment, as one of the few facts that genuinely needs no introduction.

If you remember only one inequality from a first course in analysis, Cauchy–Schwarz is the right one to remember. It is the inequality that ties together the algebra and the geometry of inner-product spaces, and from it descends a remarkable share of the most useful results in modern mathematics.

Frequently asked

Why are both names — Cauchy and Schwarz — attached to it?

Augustin-Louis Cauchy proved the discrete (finite-sum) version in 1821 in his Cours d'analyse. Forty years later, the Russian Viktor Bunyakovsky published the integral form in 1859. Hermann Schwarz independently rediscovered the integral version in 1888 and proved it more cleanly; by the time the result was central to functional analysis, Schwarz's version was the more famous. Modern mathematicians call it Cauchy–Schwarz or Cauchy–Bunyakovsky–Schwarz depending on tradition; the Russian school strongly prefers the longer form. All three contributions are genuine.

What is the geometric meaning?

Recall that for vectors in the plane, ⟨u, v⟩ = |u| |v| cos θ, where θ is the angle between them. The Cauchy–Schwarz inequality is simply the statement that cos θ ≤ 1 in absolute value — a tautology in two dimensions, but a deep result when 'vectors' are functions in an infinite-dimensional space and 'angle' is defined via an inner product. The geometric content travels: in any inner-product space, vectors have lengths and angles, and the angle between two of them is at most 90° from being a right angle in the sense that the inequality remains tight.

When does the inequality become an equality?

Exactly when the two vectors are parallel — that is, when one is a scalar multiple of the other. In two dimensions this is the case where the angle between them is 0° or 180° and cos θ = ±1. In any inner-product space, equality in Cauchy–Schwarz characterises linear dependence: |⟨u, v⟩| = ‖u‖ ‖v‖ if and only if u = cv for some scalar c (or one of them is zero). This equality condition is useful in optimisation, where Cauchy–Schwarz often gives a bound and the equality case shows when the bound is attained.

Where is Cauchy–Schwarz used in practice?

Everywhere quadratic quantities meet linear ones. In statistics, Cauchy–Schwarz produces the bound Cov(X, Y)² ≤ Var(X) Var(Y), which says the correlation coefficient is between −1 and +1. In analysis, it provides the triangle inequality for L² norms, which underlies all of Hilbert space theory. In quantum mechanics, Heisenberg's uncertainty principle is a Cauchy–Schwarz inequality applied to the position and momentum operators. In information theory, the Fisher information lower bound (Cramér–Rao) is a Cauchy–Schwarz argument. The inequality is the single most-used inequality in modern mathematics.

Continue reading

Blog

The Weierstrass Function: A Curve That Is Continuous Everywhere and Smooth Nowhere

Until 1872, mathematicians assumed continuous functions had to be differentiable somewhere. Karl Weierstrass constructed one that is differentiable nowhere — and analysis had to be rebuilt.

Read →Blog

Jordan Normal Form: What Matrices Look Like When They Cannot Be Diagonalised

Most matrices can be diagonalised — but not all. For those that resist, Camille Jordan found the next best canonical form: a block-diagonal structure from which much of linear algebra follows.

Read →Blog

The Banach Fixed-Point Theorem: When Iteration Always Converges

If a function shrinks distances, repeated application from any starting point lands at the same destination. One short theorem from 1922 underlies numerical analysis and ODE existence proofs.

Read →