Bounded Linear Operators: A Complete Student Guide

A bounded linear operator T between normed spaces is a linear map for which there is a constant M with ||T(x)|| <= M ||x|| for every x. The smallest such M is the operator norm ||T||. Boundedness is the right notion of "well-behaved" for linear maps between infinite-dimensional spaces: it is equivalent to continuity, whereas mere linearity is not enough.

They matter because almost every object of applied analysis is a bounded operator: matrices acting on vectors, integral operators, differentiation on the right spaces, the Fourier transform on L^2. The operator norm measures the maximum amplification the operator can produce, and the collection of all bounded operators B(X, Y) is itself a Banach space when Y is complete.

Key Concepts

• Bounded equals continuous: a linear map is bounded exactly when it is continuous, exactly when it is continuous at 0; discontinuity at one point means discontinuity everywhere.
• Operator norm: ||T|| = sup{||T(x)|| : ||x|| <= 1}, the largest stretch factor; also equals sup_{x != 0} ||T(x)|| / ||x||.
• Composition: ||ST|| <= ||S|| ||T||, so operator norms are submultiplicative.
• Examples: every matrix as an operator on R^n or C^n; the integral operator (Tf)(x) = integral_0^x f(t) dt on C[0,1]; the right-shift on l^2.
• B(X): bounded operators from a Banach space to itself form a Banach algebra under composition.
• Invertibility: if ||I - T|| < 1 then T is invertible, via the Neumann series.

Formulas

• Operator norm: ||T|| = sup_{||x|| <= 1} ||T(x)||; the units are (units of output) per (units of input).
• Fundamental estimate: ||T(x)|| <= ||T|| ||x|| for all x; this is how the norm is used in every error bound.
• Submultiplicativity: ||ST|| <= ||S|| ||T||, proved by applying the fundamental estimate twice.
• Matrix norms: the operator norm induced by the Euclidean vector norm is the largest singular value; it is bounded by the Frobenius norm sqrt(sum |a_ij|^2).
• Integral operator bound: for (Tf)(x) = integral_0^x f(t) dt on C[0,1] with the sup norm, |(Tf)(x)| <= x ||f||_infinity <= ||f||_infinity, so ||T|| <= 1.

Worked Example

Show that differentiation is not bounded on polynomials with the sup norm, but integration is. Let T(p) = p' on polynomials over [0,1] with ||p|| = max |p|. Take p_n(x) = x^n: ||p_n|| = 1 for every n, but T(p_n)(x) = n x^{n-1} has ||T(p_n)|| = n, which grows without bound, so no finite M works and T is unbounded (hence discontinuous). Now take (Sf)(x) = integral_0^x f(t) dt on C[0,1]. For any f, |(Sf)(x)| <= integral_0^x |f(t)| dt <= x ||f||_infinity <= ||f||_infinity, so ||S|| <= 1; testing f = 1 gives (S1)(x) = x with norm 1, so ||S|| = 1 exactly. Smoothing (integration) is bounded while roughening (differentiation) is not.

Common Misconceptions

• "Linear implies continuous." In infinite dimensions there exist everywhere-discontinuous linear maps; boundedness is the extra condition that restores continuity.
• "The operator norm is hard to compute, so it is useless." Upper bounds are usually what proofs need, and the fundamental estimate ||T(x)|| <= ||T|| ||x|| only needs an upper bound.
• "Unbounded operators cannot be studied." They can, via domains and the closed graph theorem, but they need more care; quantum mechanics is built on unbounded operators.

Real-World Applications

• Numerical linear algebra: the operator (spectral) norm of a matrix is its largest singular value, governing error amplification in Ax = b via the condition number.
• Signal processing: filters are convolution operators; boundedness on l^2 means the filter cannot blow up finite-energy signals.
• Control theory: transfer functions define bounded operators on signal spaces exactly when the system is stable.
• Quantum mechanics: observables are (often unbounded) operators, while unitary evolution operators are bounded with norm 1.

A bounded linear operator is a linear map with a finite maximum stretch factor, and boundedness is the same as continuity. The operator norm drives every estimate through ||T(x)|| <= ||T|| ||x||, matrices and integral operators are the model examples, and differentiation shows why boundedness must be checked rather than assumed.