Functional Analysis: Operators, Spaces, Spectra

Build a functional analyst’s mental model for infinite-dimensional spaces and operators by tracking which notion of convergence you need, what extra structure (Banach or Hilbert) buys you, and how spectra generalize eigenvalues. Learn where finite-dimensional intuition fails and when it returns.

Finite-dimensional linear algebra trains you to trust three things. Norms feel interchangeable, bounded sets feel tame, and operators behave like matrices with eigenvalues that explain everything. Functional analysis breaks each habit in a specific way. Convergence splits into inequivalent notions, completeness becomes a theorem-making machine, and spectra keep information even when eigenvectors disappear. The payoff is not more abstraction for its own sake. It is a reliable way to choose the right space, the right topology, and the right operator class so your argument actually closes.

Why infinite dimensions feel different

The first shock is that sequences can converge in one sense and fail badly in another, and your theorems silently depend on picking the sense that makes your limiting step legitimate. In practice, the difference between pointwise, uniform, and norm convergence is the difference between being able to pass a limit through a supremum, an integral, or an operator.

See how common function sequences behave under different convergence notions.

Completeness is where this becomes structural. A normed space that is not complete forces you to step outside it to take limits of Cauchy sequences, and those missing limits are exactly where existence proofs go to die. Completing a space is not aesthetic, it is making sure your analytic closure properties match your algebraic ambitions.

Completion first
When an argument uses Cauchy sequences, completeness is the hidden hypothesis that makes the final object live where you need it.

“Nice” operators are the ones that respect the convergence you have. If your problem gives you only weak or pointwise control, expecting strong or uniform continuity from an operator is a category error. Most functional analysis technique is choosing hypotheses that upgrade convergence just enough to make the operator behave.

Normed, Banach, and Hilbert structure

A normed space gives you size and continuity. A Banach space gives you the right to take limits without leaving home. A Hilbert space gives you angles, orthogonality, and projections that act like honest geometry. These are not incremental upgrades, they are different toolkits.

Compare what each structure unlocks and what it costs.

What Banach structure buys

Completeness makes fixed-point arguments, inverse mapping principles, and closed graph reasoning behave as you expect. Duality becomes usable because bounded linear functionals separate points and support extension theorems. You can build objects via limits without constantly checking you stayed inside the space.

What Hilbert structure buys

An inner product turns duality into geometry via the Riesz representation theorem and makes minimization problems look like orthogonal projection. When you can say x is the closest point in a closed subspace, you are really invoking projection theorems that have no Banach analogue in general.

A good litmus test is whether your argument wants orthogonality or just completeness. If you need to decompose errors into perpendicular components, you are asking for Hilbert. If you need to run a convergence machine, Banach is often enough.

Bounded operators are continuity

In a normed space setting, a linear map is continuous exactly when it is bounded. That is why boundedness is the right notion. It is not a growth condition added on top of continuity. For linear maps, it is continuity captured in a single inequality.

Explore how the operator norm packages continuity across common spaces.

Here is the core equivalence you use constantly. For a linear operator $T:X\to Y$ between normed spaces, the following are equivalent.

• $T$ is continuous at $0$
• $T$ is continuous everywhere
• there exists $C$ such that $\|Tx\|\le C\|x\|$ for all $x$
• the operator norm $\|T\|=\sup_{\|x\|=1}\|Tx\|$ is finite

Once you internalize this, most estimates become operator-norm estimates. You stop proving continuity by epsilon-delta and start proving it by bounding $\|Tx\|$ in terms of $\|x\|$.

Unbounded operators are not a minor technical variation. They usually live on a proper dense domain, they break naive algebra like everywhere-defined composition, and their spectral behavior belongs to a different world. Treat them like partially-defined objects, not like bounded operators with a large constant.

Domain matters
An unbounded operator is often best thought of as a closed graph on a dense domain, not as a map on the whole space.

Dual spaces and weak topologies

The dual space is where functional analysis turns linear algebra into analysis. The Hahn–Banach theorem makes bounded linear functionals abundant, and abundance of functionals means you can define weaker notions of convergence that are still strong enough to control many problems.

Try switching topologies and watch convergence and compactness change.

Norm convergence is too strict for many limiting arguments. Weak convergence replaces control of $\|x_n-x\|$ with control of $\varphi(x_n)$ for all bounded linear functionals $\varphi$. It feels like losing information, but it often restores compactness and existence.

The compactness surprise is Banach–Alaoglu. The closed unit ball of $X^*$ is compact in the weak* topology, even though it is typically wildly noncompact in norm. That one fact underwrites many existence proofs in PDE, optimization, and probability because it supplies convergent subsequences when norm methods cannot.

Weak and weak* topologies are not chosen for elegance. They are chosen because they are the coarsest topologies that keep the functionals you care about continuous. Once you decide which evaluations must be continuous, the topology is forced.

Adjoints, self-adjointness, and Riesz in Hilbert spaces

Hilbert spaces turn duality into an identity. The Riesz representation theorem says every continuous linear functional on a Hilbert space $H$ has the form $x\mapsto\langle x,y\rangle$ for a unique $y\in H$. So $H^*$ is not just isomorphic to $H$, it is geometrically the same object once you fix the inner product convention.

Adjoints are the operator-level version. For a bounded operator $T:H\to H$, there exists a unique bounded operator $T^*$ such that $\langle Tx,y\rangle=\langle x,T^*y\rangle$ for all $x,y$. This is not an optional extra structure. It lets you transfer statements about ranges into statements about kernels, makes normal equations precise, and gives you positivity through $\langle Tx,x\rangle$.

Self-adjointness is where spectral intuition starts to look like diagonalization again. A self-adjoint operator behaves like a real symmetric matrix in the ways that matter. Its spectrum is real, it supports functional calculus, and its geometry is stable under perturbations in ways that fail outside the self-adjoint world.

Compact operators and Fredholm behavior

Compactness is the functional analysis trick for recovering finite-dimensional behavior without being finite-dimensional. A compact operator maps the unit ball to a relatively compact set, so bounded sequences have images with convergent subsequences. That single property forces a surprisingly rigid spectral picture.

Explore how compactness creates a singular value style structure.

Compact operators often behave like matrices with decaying singular values. The spectrum of a compact operator on an infinite-dimensional Banach space is discrete away from $0$. Nonzero spectral values are eigenvalues with finite multiplicity, and $0$ is the only possible accumulation point. That is the sense in which finite-dimensional behavior reappears.

The Fredholm alternative packages the main consequence. For many operators that are identity plus compact, solvability is governed by a finite-dimensional obstruction. Either the equation $(I-K)x=y$ has a unique solution for every $y$, or the homogeneous equation has nontrivial solutions and the inhomogeneous equation is solvable exactly when $y$ satisfies finitely many orthogonality constraints.

Compactness heuristic
If you can reduce your operator to identity plus compact, expect a clean existence theory with only finite-dimensional defects.

Spectra beyond eigenvalues

The spectrum is what remains when eigenvalues are too small a notion. For a bounded operator $T$ on a Banach space, the spectrum $\sigma(T)$ is the set of $\lambda$ for which $T-\lambda I$ fails to be invertible as a bounded operator. The resolvent set is its complement, where $(T-\lambda I)^{-1}$ exists and is bounded. The point spectrum is the eigenvalues, but it can be empty even when the spectrum is large.

Visualize how spectrum, point spectrum, and resolvent differ across operators.

Two anchors keep the mental model straight. First, $\sigma(T)$ is always nonempty and compact, and it lies in the disk of radius $\|T\|$. Second, the spectral radius satisfies

$$r(T)=\sup\{|\lambda|:\lambda\in\sigma(T)\}=\lim_{n\to\infty}\|T^n\|^{1/n}.$$

So even when you cannot compute eigenvalues, powers of the operator still reveal asymptotic behavior.

The spectral theorem viewpoint is not just a statement about diagonalizing. It is a statement about replacing operators with measures or multiplication operators, so that complicated dynamics become pointwise multiplication in a transformed representation. In Hilbert spaces, normal and self-adjoint operators are the settings where this replacement is sharpest.

Thinking like a functional analyst

Most of the craft is upstream of the proof. You decide what you want to converge, what must be continuous, and which compactness principle can actually fire. Then you pick the space and topology that make those statements true with the weakest hypotheses.

Three reliable questions guide the choice.

• What is the natural a priori bound? Pick the norm that your estimates naturally control.
• What convergence do you truly have? If only evaluations or integrals converge, weak or weak* is probably the right topology.
• Which operator class matches the phenomenon? Compact for discrete spectral effects, self-adjoint for real spectral geometry, bounded for stability under perturbation and composition.

Your next step is to take a single theorem you use often, maybe a projection argument, a compactness-existence step, or an invertibility claim, and rewrite it with the exact space, topology, and operator class stated explicitly. That exercise makes the hidden functional analysis in your intuition visible and reusable.