Functional analysis and operator domains
Treat whole functions as vectors and learn why infinite-dimensional operators need more than matrix intuition.
On this page
The question for today
What you’ll understand
- Recognize Hilbert-space geometry and bounded linear operators.
- Demonstrate why differentiation is unbounded in the L2 norm.
- Explain how domains and boundary conditions affect adjoints, spectra, and quantum evolution.
Linear algebra does not end when a vector has infinitely many components. A wavefunction or a signal can be regarded as a vector in a complete inner-product space, and an evolution law can be expressed through linear operators. This extension is powerful but not automatic. Every linear map between finite-dimensional normed spaces is bounded, and a matrix acts on its declared input space. In infinite dimensions differentiation and many quantum observables are unbounded, so their domains are essential parts of their definitions rather than technical footnotes.
Geometry of function spaces
A Hilbert space is a complete inner-product space. In L2, the inner product integrates one function's complex conjugate times the other, with functions identified almost everywhere. Orthogonality, projection, and least-squares approximation still make sense. Completeness ensures that norm-convergent approximations remain inside the space. An orthonormal basis allows reconstruction through square-summable coefficients, but the resulting convergence is initially in norm; it does not automatically give pointwise values at every coordinate.
A bounded linear operator satisfies a uniform inequality relating output norm to input norm. Boundedness is equivalent to continuity for linear operators between normed spaces. It means one constant controls all inputs, not that the output itself is bounded independently of input size. On L2, multiplication by a bounded measurable function is bounded. An ideal point-evaluation functional is not generally well-defined there because changing one point does not change the L2 element.
Differentiation magnifies arbitrarily fine structure
On the periodic interval [0,2π], normalized waves e^(inx) have identical L2 norm, but their derivatives have norm proportional to |n|. No finite constant can bound all derivative outputs by the original L2 inputs. Differentiation is therefore unbounded in that norm. Moreover, an arbitrary L2 function need not possess an L2 weak derivative, so the derivative operator is not defined on the entire Hilbert space. A natural domain uses the Sobolev space H1 with the required boundary behavior.
An unbounded operator can still be well-behaved when densely defined and closed. Closedness means that if inputs and their operator outputs both converge in Hilbert norm, the limits remain a valid input-output pair. The graph norm combines the input norm and output norm to control both. These ideas replace the false expectation that every useful infinite-dimensional map behaves like a finite matrix. They also explain why truncating to finitely many frequencies hides unboundedness: each finite cutoff has a finite maximum frequency.
The domain decides which operator you actually mean
The formal momentum expression −i d/dx is not a complete operator specification. Integration by parts produces boundary terms, and the chosen domain determines whether those terms vanish and whether the adjoint has the same domain. Symmetry means the inner-product identity holds on the operator's domain. Self-adjointness additionally requires equality with the adjoint including its domain. For unbounded operators, symmetry alone does not guarantee the spectral and unitary-evolution conclusions used in quantum mechanics.
On a finite interval, periodic boundary conditions yield one self-adjoint momentum realization. More generally a phase-twisted periodic condition produces shifted momentum eigenvalues. Setting the wavefunction to zero at both endpoints makes the first-derivative expression symmetric on an appropriate dense domain but not self-adjoint. A self-adjoint Hamiltonian generates a strongly continuous unitary group; states outside its domain can still evolve unitarily even when a strong time derivative at zero is unavailable. Domain statements distinguish these claims precisely.
Work it through
Boundary phase changes the spectrum
For P=−i d/dx on [0,L], impose f(L)=exp(iθ)f(0). Find its plane-wave eigenvalues.
The eigenvalue equation −if'=pf has solutions . For nonzero C, the boundary condition becomes .
Equality of these phases requires pL=θ+2πn for an integer n. Thus the allowed values are .
Normalize with |C|=1/√L. Distinct integer modes are orthogonal because their phase differences are integer multiples of 2π across the interval.
Changing θ changes the operator domain and shifts its spectrum. The same differential expression therefore describes different operators when its global boundary data change.
The eigenvalues are . A domain choice can change measurable spectral data without changing the local derivative formula.
The living notebook
Explore the model
Interactive lab · spectral
A cutoff makes an unbounded operator look finite
Change mode count N and coefficient decay p in Σ sin(nx−t)/nᵖ. Each added mode has frequency n, so its spatial derivative would acquire an additional factor n, although derivatives are not plotted. This finite family motivates high-frequency sensitivity; it does not compute adjoint domains, prove convergence, or certify self-adjointness.
- Sum at x = π/2
- 0.867
- Smallest wavelength
- 2π / 5
- Highest-mode coefficient
- 0.2
A finite sum of sin(nx − t)/nᵖ. Playback changes t; increasing the cutoff includes finer spatial scales. This is not a proof about a continuum limit or renormalization. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forA finite matrix approximation can conceal the domain and unboundedness of the continuum operator.
Pause & reason
Check your understanding
Your turn
Try it yourself
Let (Mf)(x)=x f(x) on L2([0,1]). Prove boundedness and determine its operator norm.
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
The upper bound is immediate, but proving the exact norm requires inputs concentrated near x=1.
Compare with the solution
Since x²≤1, integrating x²|f|² gives ||Mf||₂≤||f||₂, so ||M||≤1. Choose normalized indicators supported on [1−ε,1]. Their output norm is at least 1−ε. Letting ε→0 shows ||M||≥1, hence ||M||=1. The supremum need not be attained by a nonzero L2 function concentrated at the single endpoint.
Keep the idea
Take it with you
- Hilbert-space geometry preserves projection and orthogonality while adding completeness.
- Unbounded operators require explicit domains and cannot be understood from finite matrices alone.
- Self-adjointness includes domain equality and supports the spectral structure of quantum observables.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.