Hilbert Spaces 5 | Proof of Jordan-von Neumann Theorem

TL;DR

Jordan–von Neumann’s theorem characterizes inner-product norms: a normed space is an inner product space iff its norm satisfies the parallelogram law for all vectors.

Briefing Cornell Notes

Briefing

A normed space becomes a genuine inner product space exactly when its norm satisfies the parallelogram law. That criterion—Jordan–von Neumann’s theorem—matters because it turns a geometric identity about lengths into the full algebraic structure of an inner product, which then unlocks the standard Hilbert-space machinery (orthogonality, projections, Fourier-type tools, and more).

The proof strategy starts by assuming the parallelogram law holds for every pair of vectors X and Y in a real normed vector space: the norm of X+Y and X−Y control the norm of X and Y through a fixed identity. From there, an inner product is defined indirectly using the norm. In the real case, the construction is designed so that plugging the same vector into both slots recovers the original norm: the square root of the defined inner product equals ||X||. Positive definiteness follows quickly from this recovery property: if the defined inner product is zero, the norm must be zero, forcing the vector to be the zero vector. Symmetry is also immediate in the real setting because swapping the arguments does not change the norm-based formula.

The real work is proving linearity in the second argument. The proof uses the parallelogram law as a computational engine. By rewriting differences of norm-squares into sums (adding and subtracting the same norm-squared term), the parallelogram law can be applied even though it originally speaks only about sums like ||U+Z|| and ||U−Z||. Solving small linear systems for the vectors that appear inside those norms lets the argument isolate expressions of the form “half of an inner product,” yielding a key scaling rule: the inner product with (1/2)Z in the second slot equals (1/2) times the inner product with Z.

Once the factor 1/2 case is established, additivity is tackled next. The proof again engineers the right norm-squared expressions so the parallelogram law can be applied, choosing decompositions that cause terms to cancel. This produces the identity that the inner product is additive in the second argument when the first argument is fixed: ⟨W, Z+Z’⟩ becomes ⟨W, Z⟩+⟨W, Z’⟩. With additivity in hand, homogeneity for integers follows by repeating the additivity argument, first for natural numbers and then extending to negative scalars using the inner product’s behavior under sign changes.

To reach arbitrary real scalars, the proof uses approximation: any positive real number can be approximated by dyadic rationals of the form k/2^n. Continuity of the norm lets the scaling rule pass from these approximations to all real λ. That completes linearity in the second argument, and with symmetry and positive definiteness already secured, the constructed form satisfies all inner product axioms.

Finally, the theorem is stated cleanly: a normed space is an inner product space if and only if the parallelogram law holds. The complex case follows the same blueprint but requires extra terms to handle conjugation, leading to the same conclusion.

Cornell Notes

Jordan–von Neumann’s theorem links geometry to algebra: a norm comes from an inner product exactly when it satisfies the parallelogram law. Starting from that identity (assumed for all vectors X and Y), a norm-based formula defines a candidate inner product in the real case. The construction automatically recovers the original norm via ⟨X,X⟩ = ||X||^2, giving positive definiteness, and it is symmetric because the real norm-based expression is unchanged by swapping arguments. The main effort proves linearity in the second argument by repeatedly rewriting norm-squared differences into sums that fit the parallelogram law, first deriving scaling by 1/2, then additivity, then homogeneity for all real scalars using dyadic rational approximation and continuity.

Why does recovering ||X|| from ⟨X,X⟩ immediately establish positive definiteness?

The construction is set up so that √⟨X,X⟩ = ||X||, meaning ⟨X,X⟩ = ||X||^2. If ⟨X,X⟩ = 0, then ||X||^2 = 0, so ||X|| = 0. In a normed space, ||X|| = 0 happens only when X is the zero vector, so the inner product is positive definite.

How can the parallelogram law be used when linearity computations naturally produce differences like ||U||^2 − ||V||^2?

The proof “adds and subtracts” the same norm-squared term to convert a difference into a sum of two terms that match the parallelogram-law pattern. After this adjustment, each part can be expressed using vectors of the form (something)+Z and (something)−Z, allowing the parallelogram law to be applied. The algebra then isolates the desired inner-product expression.

What is the significance of proving the factor-1/2 rule first?

Establishing ⟨W, (1/2)Z⟩ = (1/2)⟨W, Z⟩ is a cornerstone for full homogeneity. Once the 1/2 scaling is available, additivity lets the argument extend to integer multiples (via repeated sums), and then dyadic rationals k/2^n follow by combining these steps. That prepares the ground for approximating arbitrary real scalars.

How does additivity in the second argument emerge from the parallelogram law?

Additivity is proved by choosing a decomposition of the vectors inside the norm-squared expressions so that applying the parallelogram law yields four terms, with two canceling due to the specific sign pattern. What remains can be rewritten in terms of the original inner-product definition, producing ⟨W, Z+Z’⟩ = ⟨W, Z⟩ + ⟨W, Z’⟩.

Why does the proof need approximation by k/2^n and continuity of the norm?

After proving homogeneity for integers and dyadic rationals, the argument still lacks scaling for arbitrary real λ. Any positive real can be approximated by dyadic rationals k/2^n. Continuity of the norm ensures the inner-product scaling identity survives the limit as these approximations converge to λ, extending homogeneity to all real scalars (and then to negatives as well).

Review Questions

What exact role does the parallelogram law play in proving linearity, and why isn’t symmetry or positive definiteness enough?
How does the proof convert norm-squared differences into a form where the parallelogram law can be applied?
Which steps establish homogeneity first (1/2, then integers, then dyadic rationals), and how does continuity finish the extension to all real scalars?

Key Points

1
Jordan–von Neumann’s theorem characterizes inner-product norms: a normed space is an inner product space iff its norm satisfies the parallelogram law for all vectors.
2
A norm-based formula can define a candidate inner product so that ⟨X,X⟩ = ||X||^2, ensuring positive definiteness.
3
In the real case, symmetry follows directly from the norm-based definition because swapping arguments leaves the expression unchanged.
4
Linearity in the second argument is proved by rewriting expressions into sums that match the parallelogram law, then applying it repeatedly with carefully chosen vector substitutions.
5
The proof first derives scaling by 1/2 in the second argument, then uses additivity to extend scaling to integers and dyadic rationals k/2^n.
6
Homogeneity for all real scalars comes from approximating arbitrary λ by dyadic rationals and using continuity of the norm.
7
The complex case follows the same overall blueprint but requires additional terms to handle conjugation.

Highlights

The parallelogram law is the single geometric condition that forces a norm to come from an inner product.

The constructed inner product is engineered so that taking ⟨X,X⟩ recovers the original norm exactly.

Linearity is obtained by converting norm-squared differences into parallelogram-law-ready sums and then using cancellation from a tailored decomposition.

After proving homogeneity for dyadic rationals, continuity extends the result to every real scalar.

In the real setting, symmetry is essentially automatic once the norm-based formula is fixed. 

Topics

Jordan–von Neumann Theorem
Parallelogram Law
Polarization Identity
Inner Product Linearity
Normed Spaces