Simple Algorithm
Quantum Computing · Rydberg State · 30 min

The Spectral Theorem & Operator Functions

Why observables have real eigenvalues, and what e^{-iHt} really means.

Why is e−iHte^{-iHt} unitary? What do ρ\sqrt\rho and log⁡ρ\log\rho mean? One theorem answers all of these: normal operators are exactly those diagonalized by an orthonormal basis, and after that a function of an operator is just a function of numbers.

What you’ll learn

  • State and prove the finite-dimensional spectral theorem via the Schur decomposition.
  • Define f(A)f(A) for normal AA by functional calculus and use its algebraic properties.
  • Derive e−iθ n^⋅σ⃗/2=cos⁡θ2 I−isin⁡θ2 n^⋅σ⃗e^{-i\theta\,\hat n\cdot\vec\sigma/2}=\cos\tfrac\theta2\,\id-i\sin\tfrac\theta2\,\hat n\cdot\vec\sigma.
  • Prove ∥e−iHt−e−iH′t∥∞≤t∥H−H′∥∞\norm{e^{-iHt}-e^{-iH't}}_\infty\le t\norm{H-H'}_\infty and state the first-order Trotter bound.
  • Explain operator monotonicity and why A≥B≥0A\ge B\ge0 does not imply A2≥B2A^2\ge B^2.

Normal operators and the spectral theorem

DerivationProof via Schur triangularization
  1. A=UTU†,U unitary,T upper triangularA=UTU^\dagger,\qquad U\ \text{unitary},\quad T\ \text{upper triangular}

    Schur. Every A∈Cd×dA\in\C^{d\times d} has an eigenvector (its characteristic polynomial has a root in C\C). Normalize it, extend to an orthonormal basis, and induct on the orthogonal complement:

  2. (T†T)11=∣t11∣2,(TT†)11=∑j=1d∣t1j∣2(T^\dagger T)_{11}=\abs{t_{11}}^2,\qquad (TT^\dagger)_{11}=\sum_{j=1}^d\abs{t_{1j}}^2

    Unitary conjugation preserves normality, so TT is normal. Compare the (1,1)(1,1) entries of T†TT^\dagger T (column 1 of TT) and TT†TT^\dagger (row 1):

  3. T=diag⁡(λ1,…,λd) ⇒ A=∑iλi∣ui⟩ ⁣⟨ui∣T=\diag(\lambda_1,\dots,\lambda_d)\ \Rightarrow\ A=\sum_i\lambda_i\dyad{u_i}

    Equality forces t1j=0t_{1j}=0 for j>1j>1. Delete the first row and column (still normal and triangular) and repeat: TT is diagonal, and the columns uiu_i of UU are orthonormal eigenvectors.

  4. AA†=UDD∗U†=UD∗DU†=A†AAA^\dagger=UDD^*U^\dagger=UD^*DU^\dagger=A^\dagger A

    Conversely, if A=UDU†A=UDU^\dagger with DD diagonal, then both products equal U∣D∣2U†U\abs{D}^2U^\dagger. Grouping equal eigenvalues gives the projector form.

Check your understanding

Which matrix is normal but neither Hermitian nor unitary?

  1. YY
  2. (1101)\begin{pmatrix}1&1\\0&1\end{pmatrix}
  3. (0100)\begin{pmatrix}0&1\\0&0\end{pmatrix}
  4. diag⁡(1, 2i)\diag(1,\,2i)
Show the answer

D. diag⁡(1, 2i)\diag(1,\,2i)

Every diagonal matrix is normal. diag⁡(1,2i)\diag(1,2i) has a non-real eigenvalue (so it is not Hermitian) of modulus 22 (so it is not unitary). Normal means unitarily diagonalizable, with arbitrary complex eigenvalues.

Functional calculus

The map f↦f(A)f\mapsto f(A) is an algebra homomorphism: (fg)(A)=f(A)g(A)(fg)(A)=f(A)g(A), (f+g)(A)=f(A)+g(A)(f+g)(A)=f(A)+g(A) and fˉ(A)=f(A)†\bar f(A)=f(A)^\dagger. All the f(A)f(A) are diagonal in one basis, so they commute; spec⁡f(A)=f(spec⁡A)\operatorname{spec}f(A)=f(\operatorname{spec}A); and ∥f(A)∥∞=max⁡k∣f(λk)∣\norm{f(A)}_\infty=\max_k\abs{f(\lambda_k)}. This one definition supplies every operator function in quantum information:

  • ρ\sqrt\rho and ρ−1/2\rho^{-1/2} (fidelity, the pretty-good measurement), and log⁡ρ\log\rho (von Neumann entropy −Tr⁡ρlog⁡ρ-\Tr\rho\log\rho, with 0log⁡0=00\log0=0);
  • ∣A∣=A†A\abs{A}=\sqrt{A^\dagger A} for any AA (trace norm, polar decomposition A=V∣A∣A=V\abs{A});
  • e−iHte^{-iHt} (Hamiltonian evolution) and e−βH/Tr⁡e−βHe^{-\beta H}/\Tr e^{-\beta H} (thermal states).
DerivationPauli exponentials from the spectral theorem
  1. σn^2=∣n^∣2 I=I\sigma_{\hat n}^2=\abs{\hat n}^2\,\id=\id

    Let n^∈R3\hat n\in\R^3 be a unit vector and σn^=nxX+nyY+nzZ\sigma_{\hat n}=n_xX+n_yY+n_zZ. The anticommutation relations {σi,σj}=2δijI\acomm{\sigma_i}{\sigma_j}=2\delta_{ij}\id give

  2. P±=12(I±σn^),σn^=P+−P−,P++P−=IP_\pm=\tfrac12\big(\id\pm\sigma_{\hat n}\big),\qquad \sigma_{\hat n}=P_+-P_-,\qquad P_++P_-=\id

    So σn^\sigma_{\hat n} is Hermitian with eigenvalues ±1\pm1 (traceless, so one of each), with spectral projectors

  3. e−iθσn^/2=e−iθ/2P++eiθ/2P−e^{-i\theta\sigma_{\hat n}/2}=e^{-i\theta/2}P_++e^{i\theta/2}P_-

    Apply f(λ)=e−iθλ/2f(\lambda)=e^{-i\theta\lambda/2}:

  4. e−iθσn^/2=cos⁡θ2 I−isin⁡θ2 σn^e^{-i\theta\sigma_{\hat n}/2}=\cos\tfrac\theta2\,\id-i\sin\tfrac\theta2\,\sigma_{\hat n}

    Collect terms. This is Rn^(θ)R_{\hat n}(\theta), the rotation by θ\theta about n^\hat n on the Bloch sphere. At θ=2π\theta=2\pi it equals −I-\id.

What e^{-iHt} really means

For time-independent H=∑kEkPkH=\sum_kE_kP_k the Schrödinger equation iddt∣ψ⟩=H∣ψ⟩i\tfrac{d}{dt}\ket\psi=H\ket\psi (with ℏ=1\hbar=1) is solved in one line:

∣ψ(t)⟩=U(t)∣ψ(0)⟩,U(t)=e−iHt=∑ke−iEktPk.\ket{\psi(t)}=U(t)\ket{\psi(0)},\qquad U(t)=e^{-iHt}=\sum_ke^{-iE_kt}P_k.

Each energy component only rotates its phase. Populations ∥Pkψ∥2\norm{P_k\psi}^2 never change; relative phases advance at the Bohr frequencies Ek−ElE_k-E_l. Since ∣e−iEkt∣=1\abs{e^{-iE_kt}}=1, U(t)U(t) is unitary, with U(s)U(t)=U(s+t)U(s)U(t)=U(s+t) and U(t)†=U(−t)U(t)^\dagger=U(-t).

Conversely, every continuous one-parameter unitary group on Cd\C^d is e−iHte^{-iHt} for a unique Hermitian HH (the finite-dimensional case of Stone's theorem, treated in Beyond Finite Dimensions). Hermiticity of the generator is equivalent to conservation of probability.

DerivationDuhamel's formula and a perturbation bound
  1. F(s)=e−iH(t−s)e−iH′s,F(0)=e−iHt,F(t)=e−iH′tF(s)=e^{-iH(t-s)}e^{-iH's},\qquad F(0)=e^{-iHt},\quad F(t)=e^{-iH't}

    How much does a Hamiltonian error V=H′−HV=H'-H matter? Interpolate between the two evolutions:

  2. F′(s)=e−iH(t−s)(iH−iH′)e−iH′s=−i e−iH(t−s) V e−iH′sF'(s)=e^{-iH(t-s)}\big(iH-iH'\big)e^{-iH's}=-i\,e^{-iH(t-s)}\,V\,e^{-iH's}

    Differentiate; each exponential commutes with its own generator:

  3. e−iH′t−e−iHt=−i∫0te−iH(t−s) V e−iH′s dse^{-iH't}-e^{-iHt}=-i\int_0^te^{-iH(t-s)}\,V\,e^{-iH's}\,ds

    Integrate from 00 to tt (Duhamel's formula):

  4. ∥e−iHt−e−iH′t∥∞≤t ∥H−H′∥∞\norm{e^{-iHt}-e^{-iH't}}_\infty\le t\,\norm{H-H'}_\infty

    Take operator norms; unitaries have norm 11. Errors in the generator grow at most linearly in time.

Predict before you look

A Hamiltonian is miscalibrated by VV with ∥V∥∞=10−3\norm{V}_\infty=10^{-3} (ℏ=1\hbar=1). After evolving for t=200t=200, what is the largest operator-norm error ∥e−iHt−e−iH′t∥∞\norm{e^{-iHt}-e^{-iH't}}_\infty the bound allows?

Show the answer

0.2

t∥V∥∞=200×10−3=0.2t\norm{V}_\infty=200\times10^{-3}=0.2. The bound is nearly tight when VV commutes with HH: the error is then max⁡k∣1−e−ivkt∣=2∣sin⁡(vkt/2)∣≈0.1997\max_k\abs{1-e^{-iv_kt}}=2\abs{\sin(v_kt/2)}\approx0.1997 for an eigenvalue vk=10−3v_k=10^{-3}.

Order, monotonicity and trace inequalities

Hermitian matrices are partially ordered: A≥BA\ge B means A−BA-B is positive semidefinite. Functional calculus interacts with this order in surprising ways. A function ff is operator monotone if A≥BA\ge B implies f(A)≥f(B)f(A)\ge f(B) for Hermitian matrices of every size with spectra in its domain.

Think first Find the catch: does A≥B≥0A\ge B\ge0 imply A2≥B2A^2\ge B^2? Test A=(2111)A=\begin{pmatrix}2&1\\1&1\end{pmatrix}, B=(1000)B=\begin{pmatrix}1&0\\0&0\end{pmatrix}.

A−B=(1111)≥0A-B=\begin{pmatrix}1&1\\1&1\end{pmatrix}\ge0 (eigenvalues 0,20,2) and B≥0B\ge0. But A2−B2=(4332)A^2-B^2=\begin{pmatrix}4&3\\3&2\end{pmatrix} has determinant 8−9=−1<08-9=-1<0, hence a negative eigenvalue. Squaring is monotone on numbers, not on matrices: non-commutativity breaks it.

Trace functions behave better. The Golden–Thompson inequality Tr⁡eA+B≤Tr⁡(eAeB)\Tr e^{A+B}\le\Tr\big(e^Ae^B\big) holds for all Hermitian A,BA,B, while its naive three-matrix extension fails. Lieb's concavity theorem (1973), joint concavity of (A,B)↦Tr⁡(K†ApKB1−p)(A,B)\mapsto\Tr\big(K^\dagger A^pKB^{1-p}\big) on positive matrices for 0≤p≤10\le p\le1, led to Lieb and Ruskai's proof of strong subadditivity of von Neumann entropy, the backbone of quantum Shannon theory.

Key takeaways

  • Normal   ⟺  \iff unitarily diagonalizable: A=∑kλkPkA=\sum_k\lambda_kP_k with orthogonal spectral projectors. Hermitian, unitary and positive operators are the normal ones with real, unimodular and non-negative spectra.
  • Functional calculus f(A)=∑kf(λk)Pkf(A)=\sum_kf(\lambda_k)P_k defines e−iHte^{-iHt}, ρ\sqrt\rho, log⁡ρ\log\rho and ∣A∣\abs A; it is an algebra homomorphism with spec⁡f(A)=f(spec⁡A)\operatorname{spec}f(A)=f(\operatorname{spec}A).
  • e−iθn^⋅σ⃗/2=cos⁡θ2 I−isin⁡θ2 n^⋅σ⃗e^{-i\theta\hat n\cdot\vec\sigma/2}=\cos\tfrac\theta2\,\id-i\sin\tfrac\theta2\,\hat n\cdot\vec\sigma, and a 2π2\pi rotation is −I-\id.
  • Duhamel: ∥e−iHt−e−iH′t∥≤t∥H−H′∥\norm{e^{-iHt}-e^{-iH't}}\le t\norm{H-H'}; first-order Trotter error is at most t22r∥[A,B]∥\tfrac{t^2}{2r}\norm{\comm{A}{B}}.
  • Matrix order is subtle: tpt^p is operator monotone exactly for 0≤p≤10\le p\le1, while trace inequalities (Golden–Thompson, Lieb) power entropy theory.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

What is e−iπX/2e^{-i\pi X/2} as a matrix?

  1. XX
  2. −iX-iX
  3. iXiX
  4. −I-\id
Show the answer

B. −iX-iX

cos⁡π2 I−isin⁡π2 X=−iX\cos\tfrac\pi2\,\id-i\sin\tfrac\pi2\,X=-iX. As a gate it equals XX up to the global phase −i-i, but as a matrix it is −iX-iX, and the difference matters once the operation is controlled.

Question 2

Which operator is RX(2π)=e−iπXR_X(2\pi)=e^{-i\pi X}?

  1. I\id
  2. XX
  3. −iX-iX
  4. −I-\id
Show the answer

D. −I-\id

cos⁡π I−isin⁡π X=−I\cos\pi\,\id-i\sin\pi\,X=-\id. A full turn returns every Bloch vector to itself but multiplies the state by −1-1, the hallmark of spin-½ (SU(2)SU(2) double-covers SO(3)SO(3)).

Question 3

Let A≥B≥0A\ge B\ge0. Which conclusion is always valid?

  1. A2≥B2A^2\ge B^2
  2. eA≥eBe^A\ge e^B
  3. A≥B\sqrt A\ge\sqrt B
  4. A−1≥B−1A^{-1}\ge B^{-1} (when both are invertible)
Show the answer

C. A≥B\sqrt A\ge\sqrt B

t↦tt\mapsto\sqrt t is operator monotone (Löwner–Heinz with p=12p=\tfrac12). Squaring and exponentiating are increasing on numbers but not monotone on matrices.

Question 4

Which condition guarantees eA+B=eAeBe^{A+B}=e^Ae^B for square matrices A,BA,B?

  1. AA and BB are both Hermitian
  2. [A,B]=0\comm{A}{B}=0
  3. Tr⁡(AB)=Tr⁡(BA)\Tr(AB)=\Tr(BA)
  4. AA and BB are both unitary
Show the answer

B. [A,B]=0\comm{A}{B}=0

If AA and BB commute, the binomial theorem applies to (A+B)n(A+B)^n and the series factor. Otherwise the BCH series adds 12[A,B]+…\tfrac12\comm{A}{B}+\dots to the exponent.

Further reading

  • R. Bhatia, Matrix Analysis, Graduate Texts in Mathematics 169 (Springer, 1997).
  • R. A. Horn & C. R. Johnson, Matrix Analysis (Cambridge, 2nd ed., 2013).
  • E. H. Lieb & M. B. Ruskai, "Proof of the strong subadditivity of quantum-mechanical entropy," J. Math. Phys. (1973).
  • A. M. Childs, Y. Su, M. C. Tran, N. Wiebe & S. Zhu, "Theory of Trotter error with commutator scaling," Phys. Rev. X (2021).
  • A. Gilyén, Y. Su, G. H. Low & N. Wiebe, "Quantum singular value transformation and beyond: exponential improvements for quantum matrix arithmetics," STOC (2019).