4.6 Self-adjoint operators and the spectral theorem

This is the closing lesson of the chapter. It is short on new mechanics — most of the algebra was set up in the previous five lessons — but heavy on consequence. The spectral theorem is the deep statement that ties together everything we have built: the eigenvalues of 4.4, the inner-product geometry of 4.5, and the mode-and-modal-sum picture of Foundations 6.5. It is the algebraic reason separation of variables works.

Self-adjoint matrices

A real matrix $A$ is symmetric if it equals its own transpose: $A^T = A$. In components, $a_{ij} = a_{ji}$ for all $i, j$.

Symmetric matrices have a remarkable property hinted at in the 4.4 worked example: their eigenvalues are always real, and their eigenvectors can always be chosen orthogonal. Stated as a theorem:

Spectral theorem (real symmetric form). Every real symmetric $n \times n$ matrix $A$ has $n$ real eigenvalues $\lambda_1, \ldots, \lambda_n$ (counted with multiplicity) and an orthonormal basis of $\mathbb{R}^n$ consisting of corresponding eigenvectors $\mathbf{v}_1, \ldots, \mathbf{v}_n$.

In symbols: $A$ can be written as

where $D$ is the diagonal matrix of eigenvalues and $Q$ is the orthogonal matrix whose columns are the eigenvectors (so $Q^T Q = I$). This is the eigendecomposition of $A$. In the eigenbasis, the matrix is simply diagonal — every linear operation involving $A$ (matrix powers, exponentials, inverses, function evaluation) becomes componentwise multiplication in that basis.

There is a complex-vector-space generalisation. A complex matrix $A$ is Hermitian if $A^\dagger = A$, where $A^\dagger$ is the conjugate transpose (transpose and complex-conjugate every entry). Hermitian matrices have exactly the same spectral structure as real symmetric matrices: real eigenvalues, orthonormal eigenvectors (with orthogonality measured under the Hermitian inner product $\langle \mathbf{u}, \mathbf{v} \rangle = \sum \overline{u_k}\, v_k$). The Hamiltonian operator in Foundations 6.8 is Hermitian; that is why its eigenvalues — the allowed energies — are real numbers rather than complex ones, and why its eigenstates can be chosen orthogonal.

Why symmetric matrices have these properties

A short proof outline, because the result is so central:

Eigenvalues are real. Suppose $A \mathbf{v} = \lambda \mathbf{v}$ with $\mathbf{v}$ possibly complex. Take the Hermitian inner product with $\mathbf{v}$:

The right side is $\lambda$ times a real positive number $\|\mathbf{v}\|^2$. The left side is a complex number; by the Hermitian property of $A$,

so it equals its own complex conjugate, which means it’s real. The quotient of a real number by a positive real number is real, so $\lambda$ is real. ✓

Eigenvectors of distinct eigenvalues are orthogonal. Suppose $A \mathbf{v}_1 = \lambda_1 \mathbf{v}_1$ and $A \mathbf{v}_2 = \lambda_2 \mathbf{v}_2$ with $\lambda_1 \neq \lambda_2$. Compute $\langle \mathbf{v}_1, A \mathbf{v}_2 \rangle$ two ways:

• Direct: $\langle \mathbf{v}_1, A \mathbf{v}_2 \rangle = \lambda_2\, \langle \mathbf{v}_1, \mathbf{v}_2 \rangle$.
• Via adjoint: $\langle \mathbf{v}_1, A \mathbf{v}_2 \rangle = \langle A^\dagger \mathbf{v}_1, \mathbf{v}_2 \rangle = \langle A \mathbf{v}_1, \mathbf{v}_2 \rangle = \lambda_1\, \langle \mathbf{v}_1, \mathbf{v}_2 \rangle$.

(Using that $\lambda_1$ is real, so it pops out of the inner product unchanged.)

Setting the two equal: $\lambda_1 \langle \mathbf{v}_1, \mathbf{v}_2 \rangle = \lambda_2 \langle \mathbf{v}_1, \mathbf{v}_2 \rangle$, hence $(\lambda_1 - \lambda_2) \langle \mathbf{v}_1, \mathbf{v}_2 \rangle = 0$. Since $\lambda_1 \neq \lambda_2$, the inner product is zero. ✓

Within a repeated eigenvalue’s eigenspace you can always pick orthogonal eigenvectors by Gram–Schmidt, completing the orthonormal basis.

This proof is short and elementary, yet it does the entire heavy lifting of mode-decomposition physics. Notice what it depends on: just the symmetry / Hermitian property of $A$ and the basic inner-product algebra of 4.5. No deeper machinery is required.

Self-adjoint operators on function spaces

The spectral theorem generalises to self-adjoint operators on infinite-dimensional inner-product spaces (Hilbert spaces). The setting most relevant for the bookshelf is the space of square-integrable functions $L^2[a, b]$ with the inner product $\langle f, g \rangle = \int_a^b f(x)\, g(x)\, dx$, and the operator is a linear differential operator like $-\partial_x^2$ acting on functions satisfying chosen boundary conditions.

An operator $\hat L$ on a function space is self-adjoint if $\langle f, \hat L g \rangle = \langle \hat L f, g \rangle$ for all $f, g$ in its domain. For differential operators this condition mixes the operator’s coefficients with the boundary conditions; the Sturm–Liouville form is the standard rewriting that makes the self-adjointness manifest.

The infinite-dimensional spectral theorem says:

Spectral theorem (Sturm–Liouville form). A regular Sturm–Liouville operator on a bounded interval, with self-adjoint boundary conditions, has a countably infinite sequence of real eigenvalues $\lambda_1 \leq \lambda_2 \leq \cdots \to \infty$ and a corresponding sequence of orthonormal eigenfunctions $\{\phi_n\}$ that forms a complete basis of $L^2$ — any sufficiently well-behaved function can be expanded as $f(x) = \sum_n c_n \phi_n(x)$ with $c_n = \langle f, \phi_n \rangle$.

This is the theorem that makes Foundations 6.5 — Modes and mode sums work. The clamped-string spatial operator $-\partial_x^2$ with boundary conditions $X(0) = X(L) = 0$ is self-adjoint; its eigenfunctions $\sin(n \pi x / L)$ are therefore guaranteed to be orthogonal and complete. The Fourier sine series of any reasonable function on $[0, L]$ converges to that function in the $L^2$ sense.

The Hamiltonian operator $\hat H = -\hbar^2 \nabla^2 / (2m) + V(\mathbf{r})$ in Foundations 6.8 is self-adjoint (Hermitian); its energy eigenvalues are therefore real, and its energy eigenstates form a complete orthonormal basis for the Hilbert space of wavefunctions. That is why every quantum mechanics textbook starts with this theorem.

The unified picture

Putting everything in the chapter together:

• Linear algebra is the language of linear transformations. Vectors are the things being transformed; matrices and operators are the transformations.
• Eigenvalues and eigenvectors capture the “principal directions” of any linear transformation — the directions in which the action reduces to simple scaling.
• Inner products endow the space with geometry — length, angle, perpendicularity, projection.
• The spectral theorem unifies the previous three for self-adjoint operators: the principal directions are guaranteed to exist, they are guaranteed to be orthogonal, and they are guaranteed to be complete. The operator is therefore fully described by listing its eigenvalues and eigenfunctions.

Almost every linear problem in this bookshelf reduces, under the hood, to “find the eigenvalues and eigenfunctions of a particular self-adjoint operator, then expand the answer in that basis.” The wave equation has the Laplacian; the heat equation has the Laplacian; the Helmholtz equation has the Laplacian; the Schrödinger equation has the Hamiltonian; the linearised Navier–Stokes equations have a (complicated, sometimes non-self-adjoint) flow operator. In every well-behaved case, the spectral theorem promises an orthonormal eigenbasis, and the physics decomposes mode by mode.

What we use this for

The spectral theorem is the theoretical underpinning rather than a tool you wield directly, but it is invoked tacitly wherever modes appear:

• Mode expansions for PDEs (Foundations 6.5) — the existence of a complete orthonormal mode basis is the spectral theorem applied to the spatial operator.
• Fourier series and Fourier transform (Foundations 7) — the eigenfunctions of $-\partial_x^2$ on $[0, L]$ with periodic boundary conditions are $\{e^{i 2\pi n x / L}\}$, the complex Fourier basis. Completeness of that basis is the spectral theorem.
• Helmholtz cavity modes (Foundations 6.7) — the room modes are the eigenfunctions of $-\nabla^2$ with Dirichlet or Neumann boundary conditions; their orthogonality and completeness underpin all of room acoustics.
• Energy eigenstates of a quantum system (Foundations 6.8) — the Hamiltonian is self-adjoint; its eigenstates form a complete orthonormal basis, into which any state can be decomposed.
• Principal component analysis (PCA), modal decomposition of vibrating structures, proper orthogonal decomposition (POD) for fluid flows — all are instances of the spectral theorem for an empirically constructed symmetric matrix.

Closing the chapter

The chapter started with the question “what is a vector?” and ended with the spectral theorem. The arc, in one paragraph: vectors and matrices are the elementary objects; matrix–vector multiplication is the elementary operation; eigenvalues capture invariant directions; inner products supply geometry; the spectral theorem guarantees that for self-adjoint operators, the invariant directions are orthogonal and complete. Every linear problem on this bookshelf is, at some level, an application of that final guarantee.

The next chapter, Foundations 5 — Linear ODEs, already used this material — the eigenvalues of a $2 \times 2$ matrix decided whether the damped oscillator was overdamped, critically damped, or underdamped. Now you can return to it knowing exactly what the algebra was doing.