Abstract Linear Algebra 12 | Cauchy-Schwarz Inequality

TL;DR

Cauchy–Schwarz inequality bounds the inner product by the product of norms: |⟨X, Y⟩| ≤ ‖X‖‖Y‖ in real and complex inner product spaces.

Briefing Cornell Notes

Briefing

Cauchy–Schwarz inequality becomes the bridge between inner products and geometry: it turns the inner product of two vectors into a quantity controlled by their lengths, making “angles” in abstract vector spaces precise. In an inner product space over real or complex numbers, the inequality says the absolute value of ⟨X, Y⟩ is at most the product of the norms, ‖X‖‖Y‖. Equality is limited to special alignment—when X and Y are linearly dependent—so the inequality doesn’t just bound values; it identifies when vectors behave like they lie on the same line.

The proof starts by handling the easy case. If X is the zero vector, linearity forces ⟨X, Y⟩ = 0, and the bound holds immediately because the right-hand side ‖X‖‖Y‖ also becomes zero. The argument then assumes X ≠ 0 and normalizes it: divide by ‖X‖ so the proof can focus on vectors with unit length. Concretely, set X̂ = X/‖X‖ so that ‖X̂‖ = 1, and consider an inner product of the form ⟨Y − ΛX̂, Y − ΛX̂⟩, where Λ is a real scalar (chosen real first to simplify the algebra).

Because inner products satisfy positivity, ⟨Y − ΛX̂, Y − ΛX̂⟩ must be nonnegative for every real Λ. Expanding using linearity in each slot (with conjugate-linearity in the first slot noted but simplified away by taking Λ real) produces a quadratic expression in Λ: it has a constant term ‖Y‖², a leading term Λ²‖X̂‖² (which equals Λ² since ‖X̂‖ = 1), and a middle term involving the real part of ⟨Y, X̂⟩. The nonnegativity for all real Λ forces the discriminant of this quadratic to be ≤ 0; otherwise the parabola would cross the x-axis twice, contradicting the “always nonnegative” condition.

That discriminant constraint yields an inequality for the real part: (Re⟨Y, X̂⟩)² ≤ ‖Y‖². Taking square roots gives |Re⟨Y, X̂⟩| ≤ ‖Y‖. From there, the proof extends to complex vector spaces by a rotation trick. Any complex number ⟨X, Y⟩ can be multiplied by a unit-modulus scalar (a complex number of modulus 1) so that the result lies on the real line. Since both sides of the inequality are real after this adjustment, the earlier real-case bound applies to the rotated inner product, and the same normalization step delivers the full complex version: |⟨X, Y⟩| ≤ ‖X‖‖Y‖.

The payoff is immediate. With Cauchy–Schwarz in hand, inner products can legitimately support geometric language—lengths come from norms, and angles can be defined via the ratio ⟨X, Y⟩/(‖X‖‖Y‖). The inequality also clarifies when equality occurs: only when X and Y are linearly dependent, meaning they point along the same line (corresponding to extreme angles like 0° or 180°).

Cornell Notes

Cauchy–Schwarz inequality is the key tool that links inner products to geometry in any real or complex inner product space. It guarantees that for vectors X and Y, the absolute value of the inner product is bounded by the product of their norms: |⟨X, Y⟩| ≤ ‖X‖‖Y‖. The proof uses positivity by forming ⟨Y − ΛX̂, Y − ΛX̂⟩ with X̂ = X/‖X‖, expanding to a quadratic in Λ, and requiring it to be nonnegative for all real Λ. That forces the discriminant to be ≤ 0, producing a bound on Re⟨Y, X̂⟩ and then on |⟨Y, X̂⟩|. For complex spaces, the argument finishes by rotating ⟨X, Y⟩ with a unit-modulus complex scalar so it becomes real, letting the real-case result apply.

Why does the proof start by checking the case X = 0?

If X is the zero vector, linearity in the inner product forces ⟨X, Y⟩ = ⟨0, Y⟩ = 0. The right-hand side ‖X‖‖Y‖ also equals 0 because ‖0‖ = 0. So the inequality |⟨X, Y⟩| ≤ ‖X‖‖Y‖ holds automatically.

How does normalization (setting X̂ = X/‖X‖) simplify the argument?

Normalization reduces the problem to unit vectors. With X̂ = X/‖X‖, one has ‖X̂‖ = 1. When expanding ⟨Y − ΛX̂, Y − ΛX̂⟩, the Λ² term becomes Λ²‖X̂‖² = Λ², making the quadratic expression cleaner and letting the final bound scale back to the original ‖X‖.

What role does the nonnegativity of ⟨Y − ΛX̂, Y − ΛX̂⟩ play?

Inner products satisfy positivity: ⟨v, v⟩ ≥ 0 for any vector v. Here v = Y − ΛX̂, so ⟨Y − ΛX̂, Y − ΛX̂⟩ must be ≥ 0 for every real Λ. After expansion, this becomes a quadratic polynomial in Λ; requiring it to stay nonnegative for all real Λ forces its discriminant to be ≤ 0.

How does the discriminant condition translate into the Cauchy–Schwarz bound?

The quadratic in Λ has a constant term ‖Y‖², a leading term Λ², and a middle term involving Re⟨Y, X̂⟩. If the discriminant were positive, the quadratic would have two real roots and would dip below zero between them, contradicting nonnegativity for all Λ. Therefore the discriminant must be ≤ 0, which yields (Re⟨Y, X̂⟩)² ≤ ‖Y‖² and hence |Re⟨Y, X̂⟩| ≤ ‖Y‖. The proof then upgrades this to the absolute value bound.

How is the complex case handled without redoing the whole quadratic argument?

For complex inner products, the proof uses a rotation by a unit-modulus complex scalar. Any complex number z = ⟨X, Y⟩ can be written as z = |z|·c where c has modulus 1, chosen so that c·z lies on the real axis. Because the inequality being derived concerns real quantities after this rotation, the real-case bound applies to the rotated inner product. Then the normalization step gives the full complex inequality |⟨X, Y⟩| ≤ ‖X‖‖Y‖.

Review Questions

In the proof, what vector is substituted into the positivity condition, and why does that substitution produce a quadratic in Λ?
Where exactly does the discriminant argument enter, and what contradiction would occur if the discriminant were positive?
Why does multiplying by a unit-modulus complex scalar help extend the real-variable result to complex inner product spaces?

Key Points

1
Cauchy–Schwarz inequality bounds the inner product by the product of norms: |⟨X, Y⟩| ≤ ‖X‖‖Y‖ in real and complex inner product spaces.
2
Equality in Cauchy–Schwarz occurs only when X and Y are linearly dependent, meaning they lie on the same line.
3
The proof uses positivity by considering ⟨Y − ΛX̂, Y − ΛX̂⟩, which must be nonnegative for every real Λ.
4
Normalizing X to X̂ = X/‖X‖ reduces the expansion because ‖X̂‖ = 1 makes the Λ² coefficient simple.
5
Expanding the inner product yields a quadratic polynomial in Λ whose discriminant must be ≤ 0 to avoid violating nonnegativity.
6
For complex spaces, a unit-modulus scalar “rotates” ⟨X, Y⟩ onto the real line so the real-case bound applies.
7
Cauchy–Schwarz underpins the geometric meaning of inner products by controlling how “angles” relate to lengths.

Highlights

The inequality |⟨X, Y⟩| ≤ ‖X‖‖Y‖ is derived from the fact that ⟨Y − ΛX̂, Y − ΛX̂⟩ ≥ 0 for all real Λ.

Normalizing to a unit vector X̂ turns the leading term of the quadratic into Λ², simplifying the discriminant step.

The complex case is finished by rotating the inner product with a complex number of modulus 1 so it becomes real.

Equality happens precisely when X and Y are linearly dependent—vectors align along one line.

Topics

Inner Products
Cauchy–Schwarz Inequality
Norms
Quadratic Discriminant
Complex Rotation