Hilbert Spaces: Foundations & Structure

An interactive exploration of inner products, norms, metrics, and completeness

1. From Inner Products to Norms to Metrics

A Hilbert space carries three layers of structure, each derived from the one before:

The hierarchy: Inner Product $\langle \cdot, \cdot \rangle$ $\Rightarrow$ Norm $\|\cdot\|$ $\Rightarrow$ Metric $d(\cdot, \cdot)$

The inner product is the most fundamental — it lets us compute angles and orthogonality. From it, we derive the norm via $\|x\| = \sqrt{\langle x, x \rangle}$, which measures "length." From the norm, we get the metric via $d(x, y) = \|x - y\|$, which measures "distance."

Drag the teal and purple vector tips below. Watch how the inner product, norms, and distance all update — then switch inner products to see how the geometry changes.

u v
Inner Product $\langle u, v \rangle$: 5.00
Norm $\|u\| = \sqrt{\langle u, u \rangle}$: 3.16
Norm $\|v\| = \sqrt{\langle v, v \rangle}$: 2.24
Distance $d(u,v) = \|u - v\|$: 2.24
Angle $\theta = \arccos\left(\frac{\langle u,v\rangle}{\|u\|\|v\|}\right)$: 45.0°

Notice how switching to a weighted inner product changes all the derived quantities — the "unit circle" becomes an ellipse, distances stretch in certain directions, and angles between the same vectors change. The inner product is truly the foundation.

Check your understanding: If $u = (2, 0)$ and we use the weighted inner product $\langle x, y \rangle = 2x_1y_1 + x_2y_2$, what is $\|u\|$?

$\|u\| = $

2. Visualizing Beyond Three Dimensions

One of the biggest hurdles in understanding Hilbert spaces is that many important examples are infinite-dimensional. How can we build intuition for spaces we can't visualize directly?

Here's a key insight: distance accumulates across dimensions. In $\mathbb{R}^n$, the Euclidean distance between two points is:

$d(x, y) = \sqrt{\sum_{i=1}^{n} (x_i - y_i)^2}$

Each dimension contributes a term to this sum. More dimensions means more terms — and potentially much larger distances.

Drag the target ring below to set a desired distance. The system will show you which dimension is needed to make two standard basis vectors that far apart.

e₁ e₂ Dimension needed: 2 Target distance: 1.41
Target distance $d$: 1.41
Required dimension $n$: 2
Actual distance $\|e_1 - e_2\|$ in $\mathbb{R}^n$: 1.41
Formula: $\sqrt{1^2 + 1^2} = \sqrt{2}$

Per-coordinate squared contributions $(e_1 - e_2)_i^2$:

Notice how the dimension counter climbs as you expand the ring. To double the distance, you need to quadruple the dimension! This is because $\|e_1 - e_2\| = \sqrt{n}$ when both are standard basis vectors in $\mathbb{R}^n$.

Key insight: In high-dimensional spaces, "nearby" coordinate vectors can be astronomically far apart. The geometry of infinite-dimensional Hilbert spaces is nothing like our 3D intuition.
Check your understanding: In $\mathbb{R}^{100}$, what is $\|e_1 - e_2\|$ where $e_1$ and $e_2$ are standard basis vectors?

$\|e_1 - e_2\| = $

3. Completeness: Why Cauchy Sequences Matter

A Cauchy sequence is one where the terms get arbitrarily close to each other as $n \to \infty$. Formally: for every $\varepsilon > 0$, there exists $N$ such that $\|x_n - x_m\| < \varepsilon$ whenever $n, m > N$.

A space is complete if every Cauchy sequence converges to a limit within the space. This is THE defining property of Hilbert spaces (beyond having an inner product).

The problem with incompleteness: In the rationals $\mathbb{Q}$, the sequence $1, 1.4, 1.41, 1.414, 1.4142, \ldots$ is Cauchy (terms get closer together), but its limit $\sqrt{2}$ doesn't exist in $\mathbb{Q}$. The sequence "escapes" the space!

Below, place your landing pad where you think the Cauchy sequence will converge. Then press "Run Sequence" to watch it unfold. Toggle between ℝ (complete) and ℚ (incomplete) to see the difference.

1 1.5 2 √2 ≈ 1.414 your prediction
Your predicted limit: 1.41
Actual limit of sequence: √2 ≈ 1.4142...
Distance to prediction: 0.00
Status: Place your landing pad, then run the sequence

In the incomplete space ℚ, the sequence desperately wants to converge to $\sqrt{2}$, but that point simply doesn't exist! The terms pile up around an empty hole. This is catastrophic for analysis — we can't take limits, can't define derivatives or integrals reliably.

Check your understanding: Consider the sequence $a_n = (1 + 1/n)^n$ in ℝ. Is this sequence Cauchy? Does it converge?

4. Hilbert vs Banach: The Parallelogram Law

Both Hilbert spaces and Banach spaces are complete normed vector spaces. What makes Hilbert spaces special? They have an inner product underlying their norm.

The parallelogram law is the geometric test: a norm comes from an inner product if and only if:

$\|u + v\|^2 + \|u - v\|^2 = 2\|u\|^2 + 2\|v\|^2$

This says: in a parallelogram, the sum of squared diagonals equals the sum of squared sides (times 2).

Click to place vectors u and v on the grid below. Try different norms (L¹, L², L∞) and see which ones satisfy the parallelogram law!

Click to place vector u

The L² norm (Euclidean) always satisfies the parallelogram law — it comes from the standard inner product. But L¹ and L∞ fail! This is why $\ell^1$ and $\ell^\infty$ are Banach spaces but not Hilbert spaces.

Check your understanding: The parallelogram law guarantees that we can recover an inner product from a norm via the polarization identity. In a real Hilbert space, what is the formula?

5. L² as the Canonical Infinite-Dimensional Example

The space $L^2[0, 2\pi]$ of square-integrable functions is the most important infinite-dimensional Hilbert space. Its inner product is:

$\langle f, g \rangle = \int_0^{2\pi} f(x) g(x) \, dx$

This lets us measure "angles" and "distances" between functions! Two functions are orthogonal if their inner product is zero — their "positive overlap" exactly cancels their "negative overlap."

Scrub through the integration below to watch the inner product $\langle f, g \rangle$ accumulate from left to right. See where positive and negative contributions come from, and discover why $\sin(x)$ and $\cos(x)$ are orthogonal!

2π
f(x) and g(x) f g f(x)·g(x) — pointwise product Running integral ⟨f,g⟩|₀ᵗ
Integration range: [0, 2π]
Running inner product $\langle f, g \rangle|_0^t$: 0.00
Final $\langle f, g \rangle$ (at $t = 2\pi$): 0.00
Orthogonal? —

Notice how $\langle \sin, \cos \rangle$ oscillates up and down but returns exactly to zero at $t = 2\pi$. The positive and negative regions perfectly cancel — that's orthogonality! This is the foundation of Fourier analysis.

The power of L²: The functions $\{1, \sin(x), \cos(x), \sin(2x), \cos(2x), \ldots\}$ form an orthogonal basis for $L^2[0, 2\pi]$. Any square-integrable function can be written as a (possibly infinite) sum of these basis elements — that's the Fourier series!
Check your understanding: What is $\langle \sin(x), \sin(x) \rangle = \int_0^{2\pi} \sin^2(x) \, dx$?

$\langle \sin, \sin \rangle = $
(Hint: use the identity $\sin^2(x) = \frac{1 - \cos(2x)}{2}$)

Summary

You've now explored the foundational structure of Hilbert spaces:

  1. Inner products induce norms induce metrics — changing the inner product changes all of geometry.
  2. High dimensions behave strangely — distances accumulate across dimensions in ways that defy 3D intuition.
  3. Completeness is essential — Cauchy sequences must converge within the space, or analysis breaks down.
  4. The parallelogram law distinguishes Hilbert from Banach — only norms from inner products satisfy it.
  5. L² is the canonical example — functions become points, inner products become integrals, and Fourier analysis emerges naturally.

These concepts form the foundation for quantum mechanics, signal processing, PDEs, and much of modern analysis. The geometry of infinite dimensions is strange but learnable — and now you've taken the first steps.