Completeness Relation and Resolution of Identity

The resolution of the identity is infrastructure at industrial scale. Tensor-network simulations of quantum materials — the DMRG algorithms that dominate one-dimensional many-body physics — work by truncating equation (1): insert a resolution on a bond, keep the few hundred projectors that carry the entanglement; the simulation's error is literally the norm of the discarded part of $\hat 1$. Feynman's path integral is the same move iterated — a position resolution at every time slice — and lattice-QCD codes execute it on supercomputers. In quantum computing, completeness runs the daily sanity checks: tomography certifies a measurement by verifying its outcome operators sum to the identity. Kronecker's bookkeeping symbol, resolved across Hilbert space, is load-bearing engineering.

Stand a vector in ordinary space and switch on a lamp along each coordinate axis. Each lamp casts a shadow — the component along that axis. Reverse the process: add the shadows tip-to-tail and the original vector reappears, exactly. That round trip — decompose into perpendicular shadows, reassemble, lose nothing — is what $\sum_n |n\rangle\langle n| = \hat 1$ says, promoted to however many complex dimensions the physics needs.

The right-hand panel is the failure mode that makes completeness a genuine condition. Drop one axis and the reassembled vector is systematically short: the surviving projectors sum to a projector onto a subspace, not the identity. In quantum mechanics the shortfall is missing probability — if an observable's eigenstates failed to span the space, the listed outcomes' probabilities would sum to less than one. Completeness says the inventory of outcomes is full. The wave-mechanics chapter used the relation implicitly whenever a state was expanded in energy eigenfunctions; Dirac Bra-Ket Notation used the finite-basis version and borrowed the general case from this article by name. Here we pay the debt.

Take an orthonormal basis $\{|n\rangle\}$ — orthonormal meaning $\langle m|n\rangle = \delta_{mn}$, complete meaning every state expands in it (independent properties; Complete Orthonormal Bases owns the distinction and the expansion theorem). Each basis vector contributes the rank-one projector $|n\rangle\langle n|$. The completeness relation says the projectors, summed over the basis, reconstitute the identity:

where: $\hat 1$ is the identity operator, mapping every state to itself. Each summand is an operator, so the equation asserts that the rank-one pieces tile the whole space with no gaps (excluded by completeness) and no overlaps (excluded by orthonormality). Sandwich (1) inside a norm and it becomes Parseval's identity; apply it to a ket and it becomes the basis expansion — hence the grand name resolution of the identity: the featureless $\hat 1$ resolved into labeled, physically meaningful pieces. Acting with (1) on an arbitrary state exhibits the expansion with coefficients attached:

where: $c_n$ is the amplitude of $|\psi\rangle$ along $|n\rangle$ — the shadow length in the hero picture. Read right to left, (2) is the expansion theorem of Complete Orthonormal Bases; read left to right, it is that theorem repackaged as an identity between operators, deployable before any state is in sight — which is the whole point.

Equation (1) is used far more often than it is stated, always by the same gesture: since $\sum_n|n\rangle\langle n|$ is $\hat 1$, write it into any expression, at any operator position, without changing the value — and the abstract expression falls apart into numbers. The simplest deployment splits an amplitude:

where: $b_n = \langle n|\phi\rangle$ and $c_n = \langle n|\psi\rangle$ are the states' components in the inserted basis, the conjugate on $b_n$ being the ket-to-bra antilinearity of Dirac Bra-Ket Notation. An amplitude can be computed in any basis — the answer is basis-independent, the effort is not. That article's insertion card showed the double deployment: two insertions turn $\langle m|\hat A\hat B|k\rangle$ into matrix multiplication, verified there in finite bases; equation (1) extends the license to the infinite-dimensional spaces where the physics lives. Insertion also converts between representations — insert the old basis's resolution on both sides of an operator:

where: $A_{nm}$ is the old-basis matrix and the overlaps $\langle a|n\rangle$ form the change-of-basis matrix — unitary, as developed in Unitary Operators and Transformations. Equation (4) is how one Hamiltonian appears as a differential operator in the position representation and a discrete matrix in an energy basis: same operator, two resolutions. The recurring patterns:

One warning: every insertion adds a summation index, carried until orthogonality or an eigenvalue relation collapses it. Free in validity, not in bookkeeping — insert where the collapse comes fastest.

One insertion-built quantity deserves its own card, because it carries serious physics later: the trace,

where: the sum runs over any complete orthonormal basis — and the theorem below is that the choice does not matter. Basis-independence makes the trace physical: a single number belonging to the operator itself; for a Hermitian operator, the sum of its eigenvalues. The trace is the engine of the density-operator formalism (Density Operator and Mixed States computes every average as a trace), and its cyclic property is daily currency in quantum information.

The deepest use of the resolution is not decomposing states but assembling operators. Let a Hermitian observable $\hat A$ have a complete orthonormal eigenbasis $\{|a_n\rangle\}$ — reality and orthogonality are the theorems of Hermitian Operators and Observables; the eigenvalue equation is Eigenvalues and Eigenstates' business. Insert the eigenbasis resolution on both sides and let the eigenvalue equation collapse the double sum:

where: the middle element $\langle a_n|\hat A|a_m\rangle = a_m\delta_{nm}$ is diagonal precisely because the basis is the operator's own. Equation (6) — the spectral form — says a Hermitian operator is its menu of outcomes: each eigenvalue, weighted by the projector asking "are you in state $|a_n\rangle$?". When eigenvalues repeat, the rank-one pieces of one eigenvalue group into a single eigenspace projector — a regrouping owned by the next article, Projection Operators. The spectral form makes functions of operators effortless. Because the pieces are orthogonal and idempotent, powers act eigenvalue-wise, and so does any decent function:

where: $f$ may be a polynomial, exponential, square root — anything defined on the spectrum. "Eigenvalues go through the function while eigenvectors stand still" was derived one eigenstate at a time in Eigenvalues and Eigenstates; equation (7) assembles it into the whole operator. The most important customer is $e^{-i\hat Ht/\hbar}$: the dynamics chapter builds time evolution out of exactly this equation. For continuous spectra the sum becomes an integral: the position basis satisfies

where: the Kronecker delta has become a Dirac delta because the $|x\rangle$ are not normalizable states but delta-normalized ideal elements — the caveat Position Operator attached to its position spikes. Sandwiching (8) gives $\psi(x) = \langle x|\psi\rangle$; inserting the momentum version into $\langle x|\psi\rangle$ produces the position–momentum Fourier transform, the bracket-language home of Momentum-Space Wavefunction. We state (8) and use it in that spirit; the honest theory — including spectra part discrete, part continuous — is owned by Continuous Spectra and Dirac Delta Normalization, with the mathematical rescue in Rigged Hilbert Space.

Example 1 — inventory check in a tilted basis (confidence builder). In $\mathbb{C}^3$ take the discrete-Fourier basis $|f_n\rangle = \tfrac{1}{\sqrt3}(1,\omega^{n-1},\omega^{2(n-1)})$, $\omega = e^{2\pi i/3}$ — orthonormal, three vectors in three dimensions, hence complete. Audit the state $|\psi\rangle = \tfrac13(2,\,-1,\,2i)$.

Step 1. The state is normalized: $\|\psi\|^2 = \tfrac19(4 + 1 + 4) = 1$ ✓.

Step 2. Components via equation (2), $c_n = \langle f_n|\psi\rangle$ (conjugate the basis entries): $|c_1|^2 = 5/27 = 0.185$, $|c_2|^2 = 0.022$, $|c_3|^2 = 0.792$.

Step 3. Parseval audit: $0.185 + 0.022 + 0.792 = 1.000$ ✓ — the inventory is full, and reassembling $\sum_n c_n|f_n\rangle$ returns $(2, -1, 2i)/3$ exactly (machine-verified). Components are basis-relative; their square-sum is not.

Step 4. Sabotage it: keep only $|f_1\rangle, |f_2\rangle$. The sum $\hat P = |f_1\rangle\langle f_1| + |f_2\rangle\langle f_2|$ is still idempotent, but $\langle\psi|\hat P|\psi\rangle = \boxed{0.21}$: 79% of the probability leaked into the missing direction — an incomplete inventory is wrong by exactly the norm of what it forgot.

Example 2 — rebuilding a Hamiltonian from its spectrum (exam level). The two-level Hamiltonian of Eigenvalues and Eigenstates, $H = \bigl(\begin{smallmatrix}1.0 & 0.3\\ 0.3 & 2.0\end{smallmatrix}\bigr)$ eV, $E_\mp = 1.5 \mp \sqrt{0.34}$: assemble the spectral form, audit the trace, build $e^{-iHt/\hbar}$.

Step 1. Eigendata (recalled, recomputed in this article's verifier): $E_- = \boxed{0.917\ \text{eV}}$, $E_+ = \boxed{2.083\ \text{eV}}$, orthonormal eigenvectors $|E_\mp\rangle$.

Step 2. Spectral assembly, equation (6): $E_-|E_-\rangle\langle E_-| + E_+|E_+\rangle\langle E_+|$ rebuilds the matrix entry for entry (verified to $10^{-12}$) — the observable reconstructed from outcomes and the states that yield them.

Step 3. Trace audit, equation (5): diagonal route $1.0 + 2.0 = 3.0$ eV; eigenvalue route $0.917 + 2.083 = \boxed{3.0\ \text{eV}}$ — the same number, as basis-independence demands. A one-line sanity check for every diagonalization.

Step 4. A function of the operator, equation (7): $e^{-iHt/\hbar} = e^{-iE_-t/\hbar}|E_-\rangle\langle E_-| + e^{-iE_+t/\hbar}|E_+\rangle\langle E_+|$ — at $t = 1\,\mathrm{fs}$ it matches the summed exponential series to $10^{-10}$ and is unitary to $10^{-12}$. The eigenphases drift apart with beat time $2\pi\hbar/(E_+ - E_-) = \boxed{3.55\ \text{fs}}$ — the number The Time-Evolution Operator will turn into physical oscillation.

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §1.2–1.3 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 1 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I, Ch. II · Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., Appendix A · Townsend, A Modern Approach to Quantum Mechanics, Ch. 1.

Provenance. All equations are standard textbook results. A verifier should confirm: the completeness relation and its insertion uses (Sakurai §1.2–1.3, Cohen-Tannoudji Vol. I Ch. II); the three-face equivalence (Shankar Ch. 1); trace basis-independence and cyclicity; the spectral decomposition and functions of operators (Sakurai §1.3); the continuous and mixed-spectrum forms. For this edition every numeric claim was machine-verified (numpy/scipy, 27 checks): the ℂ³ DFT-basis resolution and worked-example components (0.185/0.022/0.792, Parseval to 10⁻¹², 79% leak on dropping one vector), trace invariance and Tr(AB)=Tr(BA), the two-level spectral reconstruction (0.917/2.083 eV, trace 3.0 eV, beat 3.55 fs), $e^{-iHt/\hbar}$ via the spectral form against the matrix exponential, and the double-insertion change of basis. Per the chapter ownership map: the expansion theorem and Gram–Schmidt live in Complete Orthonormal Bases; the finite-basis insertion exhibit lives in Dirac Bra-Ket Notation (extended here to the general license); projector theory and the degenerate regrouping live in Projection Operators; delta normalization proper lives in Continuous Spectra and Dirac Delta Normalization.

Builds on: Hilbert Space and State Vectors · Dirac Bra-Ket Notation · Inner Product and Norm of States · Complete Orthonormal Bases · Projection Operators

Leads to: Born Rule and Measurement Probabilities · Hermitian Operators and Observables · Expectation Values and Variance · The Postulates of Quantum Mechanics