nLab Bloch's theorem

Redirected from "Bloch waves".

Contents

Idea

In solid state physics, Bloch’s theorem (Bloch 1929) says that the energy eigenstates of a single electron in a crystal — hence of a Schrödinger operator with periodic potential — may be chosen to be Bloch waves (Bloch states): plane waves e ik⋅xe^{i k \cdot x} modulated by a function u ku_k which is periodic with respect to the crystal lattice Λ\Lambda:

ψ k(x)=e ik⋅xu k(x),u k(x+a)=u k(x)for alla∈Λ. \psi_k(x) \;=\; e^{i k \cdot x}\, u_k(x) \,, \;\;\;\;\;\; u_k(x + a) \;=\; u_k(x) \;\;\; \text{for all}\; a \in \Lambda \,.

The wave vector kk here is called the crystal momentum or quasi-momentum. Since e iG⋅xe^{i G \cdot x} is itself lattice-periodic when GG is in the reciprocal lattice Λ *\Lambda^\ast (Def. ), the crystal momentum is well-defined only modulo Λ *\Lambda^\ast, and hence ranges over a torus: the Brillouin torus.

Conceptually, Bloch’s theorem is an instance of harmonic analysis on the abelian group Λ≃ℤ d\Lambda \simeq \mathbb{Z}^d of lattice translations: Since these commute with the Hamiltonian, the Hilbert space of states decomposes as a direct integral over the irreducible unitary representations of Λ\Lambda, hence over its characters, which form the Pontrjagin dual group of Λ\Lambda — this is the Brillouin torus. Accordingly, the Hamiltonian decomposes into a family of Bloch Hamiltonians H(k)H(k) parametrized by points kk of the Brillouin torus, whose eigenvalues, as functions of kk, are the energy bands of electron band theory.

For tight-binding models with NN orbitals per unit cell (Def. , Rem. ) the Bloch Hamiltonian is simply a map from the Brillouin torus to the space of Hermitian N×NN \times N matrices (Prop. ), which is the more regular the more local the hopping is (Prop. ). The condition that the crystal is a band insulator means that these maps take values in the subspace of Hermitian matrices with a spectral gap at the Fermi energy, which is homotopy equivalent to a disjoint union of Grassmannians (Prop. ). Therefore deformation classes of gapped Bloch Hamiltonians are classified by the topological K-theory of the Brillouin torus, which is the starting point of the K-theory classification of topological phases of matter (Rem. ).

The tight-binding case captures the general situation of a Schrödinger operator with periodic potential (Prop. ) restricted to an isolated family of bands, at least whenever the corresponding Bloch bundle over the Brillouin torus is trivializable, as is automatic in the presence of time-reversal symmetry in dimension d≤3d \leq 3 (Prop. , Prop. , Rem. ).

The analogous statement with spatial t0ranslations replaced by time translations, hence for Hamiltonians periodic in time, is the theorem of Floquet theory (Floquet 1883), which historically predates Bloch’s theorem; hence one also speaks of Bloch-Floquet theory.

Under Bloch's theorem, the Hamiltonian of a single electron in a crystal lattice is decomposed into one Bloch Hamiltonian of internal orbital motion per plane wave vector k ranging over the Brillouin torus of electron quasimomenta in the crystal. The result is a map from the Brillouin torus to the space of accessible internal Bloch Hamiltonians.

Statement

Setup

Definition

(crystal lattice and Brillouin torus)

Let d∈ℕd \in \mathbb{N} (the spatial dimension, typically d∈{1,2,3}d \in \{1,2,3\}).

  1. A crystal lattice is a lattice Λ⊂ℝ d\Lambda \subset \mathbb{R}^d in dd-dimensional Euclidean space, hence a discrete subgroup of the form Λ=ℤa 1+⋯+ℤa d≃ℤ d\Lambda = \mathbb{Z} a_1 + \cdots + \mathbb{Z} a_d \simeq \mathbb{Z}^d for a linear basis (a 1,⋯,a d)(a_1, \cdots, a_d) of ℝ d\mathbb{R}^d. A fundamental domain for the action of Λ\Lambda on ℝ d\mathbb{R}^d is called a unit cell.

  2. Its reciprocal lattice is

    Λ *≔{G∈ℝ d|∀a∈ΛG⋅a∈2πℤ}. \Lambda^\ast \;\coloneqq\; \Big\{ G \in \mathbb{R}^d \;\Big\vert\; \underset{a \in \Lambda}{\forall} \; G \cdot a \,\in\, 2\pi \mathbb{Z} \Big\} \,.
  3. Its Brillouin torus is the Pontrjagin dual group of Λ\Lambda, hence the group of its characters, which is identified with the quotient group of ℝ d\mathbb{R}^d by the reciprocal lattice:

    T^ d≔Hom Grp(Λ,U(1))≃ℝ d/Λ *, \widehat{T}{}^d \;\coloneqq\; \mathrm{Hom}_{Grp}\big(\Lambda,\, U(1)\big) \;\simeq\; \mathbb{R}^d / \Lambda^\ast \,,

    via [k]↦(a↦e ik⋅a)[k] \mapsto \big(a \mapsto e^{i k \cdot a}\big). A fundamental domain B⊂ℝ dB \subset \mathbb{R}^d for this quotient (traditionally the Voronoi cell of 0∈Λ *0 \in \Lambda^\ast) is called a Brillouin zone.

    We write dkd k for the Haar measure on T^ d\widehat{T}{}^d normalized to total volume 1, hence for the Lebesgue measure on BB divided by vol(B)\mathrm{vol}(B).

For tight-binding models

Definition

(tight-binding Hamiltonians)

For N∈ℕ ≥1N \in \mathbb{N}_{\geq 1}, a tight-binding Hamiltonian with NN orbitals per unit cell is a bounded self-adjoint operator HH on the Hilbert space

ℋ≔ℓ 2(Λ;ℂ N)≃ℓ 2(Λ)⊗ℂ N \mathcal{H} \;\coloneqq\; \ell^2\big(\Lambda;\, \mathbb{C}^N\big) \;\simeq\; \ell^2(\Lambda) \otimes \mathbb{C}^N

which commutes with all lattice translations (T bψ)(a)≔ψ(a−b)(T_b \psi)(a) \coloneqq \psi(a - b), b∈Λb \in \Lambda.

Equivalently, HH is a convolution operator

(Hψ)(a)=∑b∈Λh(a−b)⋅ψ(b) (H \psi)(a) \;=\; \underset{b \in \Lambda}{\sum} h(a - b) \cdot \psi(b)

with hopping matrices h:Λ→Mat N×N(ℂ)h \colon \Lambda \to \mathrm{Mat}_{N \times N}(\mathbb{C}) given by

h ij(a)≔⟨δ a⊗e i,H(δ 0⊗e j)⟩ h_{i j}(a) \;\coloneqq\; \big\langle \delta_a \otimes e_i ,\, H (\delta_0 \otimes e_j) \big\rangle

(the amplitude for an electron to hop from orbital jj in cell 00 to orbital ii in cell aa), which satisfy h(−a)=h(a) †h(-a) = h(a)^\dagger.

We say that the hopping is:

  • of finite range if h(a)=0h(a) = 0 for all but finitely many a∈Λa \in \Lambda;

  • exponentially decaying if ‖h(a)‖≤Ce −β|a|\Vert h(a) \Vert \leq C e^{- \beta \vert a \vert} for some C,β>0C, \beta \gt 0;

  • rapidly decaying if ‖h(a)‖⋅(1+|a|) m\Vert h(a) \Vert \cdot (1 + \vert a \vert)^m is bounded, for all m∈ℕm \in \mathbb{N};

  • summable if ∑ a∈Λ‖h(a)‖<∞\sum_{a \in \Lambda} \Vert h(a) \Vert \lt \infty.

Remark

(terminology: “tight binding” and “Bloch Hamiltonian”)

  • The term tight binding refers to the original motivation for such models (Bloch 1929, Slater & Koster 1954, see Ashcroft & Mermin 1976, Ch. 10): If the electrons of interest are tightly bound to the atoms (ions) of the crystal, then their wave functions are well approximated by linear combinations of finitely many localized atomic orbitals? (the LCAO method: “linear combination of atomic orbitals”). This means that the single-electron Hilbert space L 2(ℝ d)L^2(\mathbb{R}^d) may be truncated to the span of these orbitals, which (after orthonormalization) is ≃ℓ 2(Λ;ℂ N)\simeq \ell^2(\Lambda; \mathbb{C}^N) if there are NN relevant orbitals per unit cell. The matrix elements of the Hamiltonian between orbitals in different cells involve the overlap of their wave functions, which decays rapidly with distance; hence the hopping matrices h(a)h(a) are dominated by small |a|\vert a \vert, and are typically truncated to finite range (often to “nearest neighbours”).

  • What this reflects in modern understanding is not so much tight binding as the isolation of a family of NN bands (typically those near the Fermi energy) from the rest of the spectrum by spectral gaps: In this situation, the band-projected continuum Hamiltonian is exactly (not just approximately) unitarily equivalent to a tight-binding Hamiltonian with exponentially decaying hopping, provided that the corresponding Bloch bundle is trivializable, in which case the role of the “orbitals” is played by exponentially localized Wannier functions (Prop. below). In this sense, “tight-binding model” is used today essentially synonymously with “lattice model with finitely many degrees of freedom per unit cell”, irrespective of how tightly the electrons are actually bound.

  • Accordingly, the term Bloch Hamiltonian for the operators H(k)H(k) (Def. below, and similarly for the fiber operators in Prop. ) reflects that these are what remains of the Hamiltonian after diagonalizing the lattice translation symmetry, following Bloch: one Hamiltonian for each value of the crystal momentum, acting only on the internal (orbital, or intra-cell) degrees of freedom.

Proposition

(Bloch’s theorem for tight-binding models)

Given a tight-binding Hamiltonian HH (Def. ), the Fourier transform

ℓ 2(Λ;ℂ N) ⟶ℱ L 2(T^ d;ℂ N) ψ ↦ (k↦∑a∈Λe −ik⋅aψ(a)) \array{ \ell^2\big(\Lambda;\, \mathbb{C}^N\big) &\overset{\mathcal{F}}{\longrightarrow}& L^2\big(\widehat{T}{}^d;\, \mathbb{C}^N\big) \\ \psi &\mapsto& \Big( k \mapsto \underset{a \in \Lambda}{\sum} e^{- i k \cdot a}\, \psi(a) \Big) }

is a unitary operator which transforms HH into the operator of multiplication by the matrix-valued function

H(−):T^ d⟶Herm(N),H(k)≔∑a∈Λe −ik⋅ah(a), H(-) \;\colon\; \widehat{T}{}^d \longrightarrow \mathrm{Herm}(N) \,, \;\;\;\;\;\; H(k) \;\coloneqq\; \underset{a \in \Lambda}{\sum} e^{- i k \cdot a}\, h(a) \,,

in that

(ℱHℱ −1ϕ)(k)=H(k)⋅ϕ(k). \big(\mathcal{F} H \mathcal{F}^{-1} \phi\big)(k) \;=\; H(k) \cdot \phi(k) \,.

(For non-summable hopping the sum defining H(k)H(k) is understood as an L 2L^2-limit, so that H(−)H(-) is an essentially bounded measurable function.)

If the hopping is summable, then furthermore:

  1. (Bloch waves) For every k∈ℝ dk \in \mathbb{R}^d and every eigenvector v∈ℂ Nv \in \mathbb{C}^N of H(k)H(k) with eigenvalue EE, the bounded function

    ψ k(a)≔e ik⋅av \psi_k(a) \;\coloneqq\; e^{i k \cdot a}\, v

    is a generalized eigenvector of HH, Hψ k=Eψ kH \psi_k = E\, \psi_k, and is quasi-periodic: ψ k(a+b)=e ik⋅bψ k(a)\psi_k(a + b) = e^{i k \cdot b}\, \psi_k(a).

  2. (band structure) H(−)H(-) is continuous, and so are its ordered eigenvalues E 1(k)≤E 2(k)≤⋯≤E N(k)E_1(k) \leq E_2(k) \leq \cdots \leq E_N(k), the band functions. The spectrum of HH is the union of the energy bands:

    σ(H)=⋃k∈T^ dσ(H(k))=⋃n=1NE n(T^ d), \sigma(H) \;=\; \underset{k \in \widehat{T}{}^d}{\bigcup} \sigma\big(H(k)\big) \;=\; \overset{N}{\underset{n = 1}{\bigcup}}\, E_n\big(\widehat{T}{}^d\big) \,,

    which is a union of at most NN closed intervals.

Proof

Unitarity of ℱ\mathcal{F} is the Plancherel theorem? for the Pontrjagin duality between Λ\Lambda and T^ d\widehat{T}{}^d, and the transformation of HH is the convolution theorem. Self-adjointness H(k) †=H(k)H(k)^\dagger = H(k) follows from h(−a)=h(a) †h(-a) = h(a)^\dagger.

For the Bloch waves, compute:

(Hψ k)(a)=∑bh(a−b)e ik⋅bv=e ik⋅a∑ce −ik⋅ch(c)v=e ik⋅aH(k)v=Eψ k(a). (H \psi_k)(a) \;=\; \underset{b}{\sum} h(a - b)\, e^{i k \cdot b}\, v \;=\; e^{i k \cdot a}\, \underset{c}{\sum} e^{- i k \cdot c}\, h(c)\, v \;=\; e^{i k \cdot a}\, H(k)\, v \;=\; E\, \psi_k(a) \,.

For summable hopping, H(−)H(-) is a uniform limit of continuous functions, hence continuous, and continuity of the ordered eigenvalues follows by Weyl’s inequality |E n(k)−E n(k′)|≤‖H(k)−H(k′)‖\big\vert E_n(k) - E_n(k') \big\vert \leq \big\Vert H(k) - H(k') \big\Vert. Finally, the spectrum of a multiplication operator by a continuous function on a compact space is the union of the pointwise spectra.

Definition

(Bloch Hamiltonian)

The function H(−):T^ d→Herm(N)H(-) \colon \widehat{T}{}^d \to \mathrm{Herm}(N) in Prop. is called the Bloch Hamiltonian of the tight-binding Hamiltonian HH.

Conversely, the hopping matrices are recovered from the Bloch Hamiltonian as its Fourier coefficients

h(a)=∫ T^ de ik⋅aH(k)dk, h(a) \;=\; \int_{\widehat{T}{}^d} e^{i k \cdot a}\, H(k)\, d k \,,

so that tight-binding Hamiltonians are in bijection with (essentially bounded, measurable) Bloch Hamiltonians.

Proposition

(locality of hopping is regularity of the Bloch Hamiltonian)

Under the bijection of Def. , the hopping (Def. ) is:

  • of finite range ⇔\;\Leftrightarrow\; the Bloch Hamiltonian is a trigonometric polynomial (in particular real analytic);

  • exponentially decaying ⇔\;\Leftrightarrow\; the Bloch Hamiltonian is real analytic (and then it extends holomorphically to a strip |Im(k)|<β\vert \mathrm{Im}(k) \vert \lt \beta);

  • rapidly decaying ⇔\;\Leftrightarrow\; the Bloch Hamiltonian is smooth;

  • summable ⇒\;\Rightarrow\; the Bloch Hamiltonian is continuous.

Proof

These are standard facts about Fourier series on the torus T^ d\widehat{T}{}^d: smoothness corresponds to rapid decay of Fourier coefficients via integration by parts; and real analyticity corresponds to exponential decay via the holomorphic extension of the Fourier series to a complex strip and shifting the integration contour in the formula for h(a)h(a) (Def. ) into that strip.

In practice, physically admissible Bloch Hamiltonians are thus real-analytic, if not trigonometric polynomials. For deformation-theoretic questions, however, this regularity is immaterial (Prop. below).

Remark

(periodic vs. non-periodic convention)

If the orbital ii sits at position r ir_i within the unit cell, part of the physics literature uses instead of H(k)H(k) (Prop. ) the matrices

H˜(k) ij≔∑a∈Λe −ik⋅(a+r i−r j)h ij(a),henceH˜(k)=D(k)H(k)D(k) †,D(k)≔diag(e −ik⋅r 1,⋯,e −ik⋅r N), \widetilde{H}(k)_{i j} \;\coloneqq\; \underset{a \in \Lambda}{\sum} e^{- i k \cdot (a + r_i - r_j)}\, h_{i j}(a) \,, \;\;\;\;\;\; \text{hence} \;\;\;\;\;\; \widetilde{H}(k) \;=\; D(k)\, H(k)\, D(k)^\dagger \,, \;\;\; D(k) \;\coloneqq\; \mathrm{diag}\big( e^{- i k \cdot r_1}, \cdots, e^{-i k \cdot r_N} \big) \,,

which have the same spectrum but are not periodic in kk (see Bena & Montambaux 2009 for discussion in the example of graphene): instead H˜(k+G)=D(G)H˜(k)D(G) †\widetilde{H}(k + G) = D(G)\, \widetilde{H}(k)\, D(G)^\dagger for G∈Λ *G \in \Lambda^\ast.

So H˜\widetilde{H} is not a function on the Brillouin torus but a section of a (here: trivialized) bundle of matrix algebras over it. Only the periodic convention of Prop. exhibits the Bloch Hamiltonian as a map out of the Brillouin torus. The same issue reappears for continuum models in Rem. below.

Gapped Bloch Hamiltonians

Definition

(gapped Bloch Hamiltonians)

For μ∈ℝ\mu \in \mathbb{R} (the Fermi energy) and n∈{0,⋯,N}n \in \{0, \cdots, N\} write

Herm μ(N)≔{A∈Herm(N)|μ∉σ(A)}=∐n=0NHerm μ (n)(N) \mathrm{Herm}_\mu(N) \;\coloneqq\; \big\{ A \in \mathrm{Herm}(N) \,\big\vert\, \mu \notin \sigma(A) \big\} \;=\; \overset{N}{\underset{n = 0}{\coprod}}\, \mathrm{Herm}^{(n)}_\mu(N)

for the open subspace of Hermitian matrices with a spectral gap at μ\mu, decomposed according to the number nn of eigenvalues below μ\mu (counted with multiplicity).

A continuous Bloch Hamiltonian H(−)H(-) (Def. ) is gapped at μ\mu if it takes values in Herm μ(N)\mathrm{Herm}_\mu(N), hence, by connectedness of T^ d\widehat{T}{}^d, in Herm μ (n)(N)\mathrm{Herm}^{(n)}_\mu(N) for a fixed nn. In this case nn is the number of valence bands (filled bands) and N−nN - n the number of conduction bands.

By compactness of T^ d\widehat{T}{}^d and Prop. , H(−)H(-) is gapped at μ\mu iff μ∉σ(H)\mu \notin \sigma(H), i.e. iff the crystal is a band insulator at Fermi energy μ\mu.

Proposition

(space of gapped Hamiltonians is a union of Grassmannians)

The map

Herm μ (n)(N) ⟶P <μ Gr n(ℂ N) A ↦ im(P <μ(A)) \array{ \mathrm{Herm}^{(n)}_\mu(N) &\overset{P_{\lt \mu}}{\longrightarrow}& \mathrm{Gr}_n\big(\mathbb{C}^N\big) \\ A &\mapsto& \mathrm{im}\big(P_{\lt \mu}(A)\big) }

which sends a gapped Hermitian matrix to the image of its spectral projection? onto the eigenvalues below μ\mu, regarded as a point in the Grassmannian Gr n(ℂ N)≃U(N)/(U(n)×U(N−n))\mathrm{Gr}_n(\mathbb{C}^N) \simeq U(N)/\big(U(n) \times U(N-n)\big), is a homotopy equivalence. In fact it exhibits a deformation retraction onto the subspace of flattened Hamiltonians {μ⋅1+1−2P|P∈Gr n(ℂ N)}\big\{\mu \cdot 1 + 1 - 2 P \,\big\vert\, P \in \mathrm{Gr}_n(\mathbb{C}^N)\big\}, via the homotopy

A t≔(1−t)⋅A+t⋅(μ⋅1+1−2P <μ(A)),t∈[0,1]. A_t \;\coloneqq\; (1 - t) \cdot A \,+\, t \cdot \big( \mu \cdot 1 + 1 - 2 P_{\lt \mu}(A) \big) \,, \;\;\;\;\; t \in [0,1] \,.

Accordingly, a gapped Bloch Hamiltonian T^ d→Herm μ (n)(N)\widehat{T}{}^d \to \mathrm{Herm}^{(n)}_\mu(N) is, up to homotopy, equivalently a map T^ d→Gr n(ℂ N)\widehat{T}{}^d \to \mathrm{Gr}_n(\mathbb{C}^N), hence a rank nn subbundle of the trivial vector bundle T^ d×ℂ N\widehat{T}{}^d \times \mathbb{C}^N: the valence bundle 𝒱→T^ d\mathcal{V} \to \widehat{T}{}^d with fibers 𝒱 k=im(P <μ(H(k)))\mathcal{V}_k = \mathrm{im}\big(P_{\lt \mu}(H(k))\big).

Proof

Since the spectrum of AA stays away from μ\mu under small perturbations, the spectral projection is given locally by the Riesz projection formula

P <μ(A)=12πi∮ C(z−A) −1dz P_{\lt \mu}(A) \;=\; \frac{1}{2\pi i} \oint_C (z - A)^{-1} \, d z

for a fixed contour CC enclosing the eigenvalues below μ\mu, and is hence continuous (even real analytic) in AA. The matrix A tA_t commutes with AA and has eigenvalues (1−t)λ+t(μ∓1)(1-t)\lambda + t(\mu \mp 1) for each eigenvalue λ<μ\lambda \lt \mu (resp. λ>μ\lambda \gt \mu) of AA, which remain on the same side of μ\mu for all t∈[0,1]t \in [0,1]. Hence A t∈Herm μ (n)(N)A_t \in \mathrm{Herm}^{(n)}_\mu(N), with A 0=AA_0 = A and with A 1A_1 the flattened Hamiltonian corresponding to P <μ(A)P_{\lt \mu}(A), which is fixed throughout if AA is already flat.

Proposition

(regularity is immaterial for deformation classes)

  1. Every continuous gapped Bloch Hamiltonian is homotopic, through gapped Bloch Hamiltonians, to one that is a trigonometric polynomial (hence comes from finite-range hopping, Prop. ).

  2. If two trigonometric-polynomial gapped Bloch Hamiltonians are continuously homotopic through gapped Bloch Hamiltonians, then they are so through trigonometric-polynomial gapped Bloch Hamiltonians.

Hence (with Prop. ) the deformation classes of gapped tight-binding Hamiltonians — whether of finite-range, exponentially decaying, rapidly decaying or just summable hopping — are the homotopy classes of maps

π 0Map(T^ d,Herm μ (n)(N))≃[T^ d,Gr n(ℂ N)]. \pi_0 \, \mathrm{Map}\big( \widehat{T}{}^d ,\, \mathrm{Herm}^{(n)}_\mu(N) \big) \;\simeq\; \Big[ \widehat{T}{}^d ,\, \mathrm{Gr}_n\big(\mathbb{C}^N\big) \Big] \,.
Proof

Since Herm μ(N)⊂Herm(N)\mathrm{Herm}_\mu(N) \subset \mathrm{Herm}(N) is open, every compact subset K⊂Herm μ(N)K \subset \mathrm{Herm}_\mu(N) has positive distance δ K\delta_K from its complement; so any two maps into KK at uniform distance <δ K\lt \delta_K are connected by the straight-line homotopy within Herm μ(N)\mathrm{Herm}_\mu(N). Now the Fejér means (convolution with the dd-fold product of Fejér kernels) of a continuous function on T^ d\widehat{T}{}^d are trigonometric polynomials, preserve Hermiticity, and converge uniformly; and for a homotopy T^ d×[0,1]→Herm μ(N)\widehat{T}{}^d \times [0,1] \to \mathrm{Herm}_\mu(N) they do so uniformly also in the homotopy parameter, by equicontinuity. Applying this to a given map or homotopy, and then joining endpoints by straight-line homotopies (which preserve the property of being a trigonometric polynomial) yields the claim.

Remark

(K-theory classification and symmetries)

For Schrödinger operators with periodic potential

Proposition

(Bloch-Floquet decomposition)

Let V:ℝ d→ℝV \colon \mathbb{R}^d \to \mathbb{R} be a Λ\Lambda-periodic potential, V(x+a)=V(x)V(x + a) = V(x) for a∈Λa \in \Lambda, which is bounded (or more generally suitably locally integrable, see Reed & Simon 1978, §XIII.16), and consider the Schrödinger operator

H=−ℏ 22m eΔ+V H \;=\; - \frac{\hbar^2}{2 m_e}\, \Delta \,+\, V

on L 2(ℝ d)L^2(\mathbb{R}^d). Then:

  1. The Bloch-Floquet transform (also: Zak transform), given on Schwartz functions by

    (Uψ) k(x)≔∑a∈Λe −ik⋅(x+a)ψ(x+a),k,x∈ℝ d, (U \psi)_k(x) \;\coloneqq\; \underset{a \in \Lambda}{\sum} e^{- i k \cdot (x + a)}\, \psi(x + a) \,, \;\;\;\;\; k, x \in \mathbb{R}^d \,,

    produces functions (Uψ) k(U\psi)_k which are Λ\Lambda-periodic in xx and satisfy (Uψ) k+G=e −iG⋅x(Uψ) k(U \psi)_{k + G} = e^{- i G \cdot x} (U \psi)_k for G∈Λ *G \in \Lambda^\ast. It extends to a unitary operator onto a direct integral of Hilbert spaces

    U:L 2(ℝ d)→∼∫ B ⊕L 2(ℝ d/Λ)dk U \;\colon\; L^2(\mathbb{R}^d) \xrightarrow{\;\sim\;} \int^\oplus_B L^2\big(\mathbb{R}^d/\Lambda\big)\, d k

    with inverse ψ(x)=∫ Be ik⋅x(Uψ) k(x)dk\psi(x) = \int_B e^{i k \cdot x}\, (U \psi)_k(x)\, d k.

  2. This transform decomposes HH into fiber operators:

    UHU −1=∫ B ⊕H(k)dk,H(k)≔ℏ 22m e(−i∇+k) 2+V, U H U^{-1} \;=\; \int^\oplus_B H(k)\, d k \,, \;\;\;\;\;\; H(k) \;\coloneqq\; \frac{\hbar^2}{2 m_e} \big( - i \nabla + k \big)^2 \,+\, V \,,

    where H(k)H(k) acts on L 2(ℝ d/Λ)L^2\big(\mathbb{R}^d/\Lambda\big) (i.e. with periodic boundary conditions on the unit cell), being self-adjoint on the Sobolev space H 2(ℝ d/Λ)H^2\big(\mathbb{R}^d/\Lambda\big).

  3. Each H(k)H(k) is bounded below and has compact resolvent, hence purely discrete spectrum E 1(k)≤E 2(k)≤⋯→∞E_1(k) \leq E_2(k) \leq \cdots \to \infty (eigenvalues ordered and counted with multiplicity).

  4. The family k↦H(k)k \mapsto H(k) extends to an entire analytic family of type (A) (in the sense of Kato) on ℂ d\mathbb{C}^d, and satisfies

    H(k+G)=τ GH(k)τ G −1,τ G≔e −iG⋅x,G∈Λ *. H(k + G) \;=\; \tau_G \, H(k) \, \tau_G^{-1} \,, \;\;\;\;\; \tau_G \;\coloneqq\; e^{- i G \cdot x} \,, \;\;\; G \in \Lambda^\ast \,.

    Hence the band functions E nE_n are continuous and Λ *\Lambda^\ast-periodic, thus descend to continuous functions on the Brillouin torus T^ d\widehat{T}{}^d (and they are real analytic wherever E n(k)E_n(k) is a simple eigenvalue).

  5. The spectrum of HH is the union of the energy bands:

    σ(H)=⋃n∈ℕE n(T^ d), \sigma(H) \;=\; \underset{n \in \mathbb{N}}{\bigcup}\, E_n\big(\widehat{T}{}^d\big) \,,

    a union of closed intervals, possibly separated by band gaps.

See Reed & Simon 1978, §XIII.16 and Kuchment 2016. Moreover, under mild conditions on VV, the spectrum of HH is purely absolutely continuous, i.e. there are no flat bands (Thomas 1973, see Kuchment 2016 for the general situation).

Corollary

(Bloch’s theorem)

In the situation of Prop. , for every k∈ℝ dk \in \mathbb{R}^d and n∈ℕn \in \mathbb{N}, let u n,ku_{n,k} be an eigenfunction of H(k)H(k) for the eigenvalue E n(k)E_n(k). Then the Bloch wave

ψ n,k(x)≔e ik⋅xu n,k(x) \psi_{n,k}(x) \;\coloneqq\; e^{i k \cdot x}\, u_{n,k}(x)

is a generalized eigenfunction of HH,

Hψ n,k=E n(k)ψ n,k,ψ n,k(x+a)=e ik⋅aψ n,k(x)fora∈Λ, H \psi_{n,k} \;=\; E_n(k)\, \psi_{n,k} \,, \;\;\;\;\;\; \psi_{n,k}(x + a) \;=\; e^{i k \cdot a}\, \psi_{n,k}(x) \;\;\; \text{for}\; a \in \Lambda \,,

and every ψ∈L 2(ℝ d)\psi \in L^2(\mathbb{R}^d) is a superposition of Bloch waves:

ψ(x)=∑n∫ Bc n(k)ψ n,k(x)dk,c n(k)=⟨u n,k,(Uψ) k⟩. \psi(x) \;=\; \underset{n}{\sum} \int_B c_n(k)\, \psi_{n,k}(x) \, d k \,, \;\;\;\;\; c_n(k) \;=\; \big\langle u_{n,k} ,\, (U\psi)_k \big\rangle \,.
Remark

(twisted periodicity)

In contrast to the tight-binding case (Prop. ), the fiber operators H(k)H(k) in Prop. are not periodic in kk, but only up to unitary conjugation by τ G\tau_G. One may instead use the functions e ik⋅x(Uψ) k(x)e^{i k \cdot x}\, (U\psi)_k(x), which are Λ *\Lambda^\ast-periodic in kk, but these are quasi-periodic in xx, so that the fiber Hilbert spaces {ϕ|ϕ(x+a)=e ik⋅aϕ(x)}\big\{ \phi \,\big\vert\, \phi(x + a) = e^{i k \cdot a} \phi(x) \big\} now depend on kk. Either way, HH decomposes not into a function on the Brillouin torus but into a section of an operator bundle over the Brillouin torus. (This is the continuum analog of Rem. .)

A function on the Brillouin torus is obtained only after choosing a trivialization, which for isolated families of bands is the content of the following.

Definition

(isolated bands and Bloch bundle)

In the situation of Prop. , a family of NN consecutive bands E m,⋯,E m+N−1E_m, \cdots, E_{m + N - 1} is called isolated if for all kk we have E m−1(k)<E m(k)E_{m-1}(k) \lt E_m(k) (if m>1m \gt 1) and E m+N−1(k)<E m+N(k)E_{m+N-1}(k) \lt E_{m+N}(k); by compactness of T^ d\widehat{T}{}^d the corresponding gaps are then bounded away from zero. In this case the spectral projections

P(k)≔12πi∮ C(z−H(k)) −1dz P(k) \;\coloneqq\; \frac{1}{2\pi i} \oint_{C} \big(z - H(k)\big)^{-1}\, d z

(for CC a contour enclosing exactly E m(k),⋯,E m+N−1(k)E_m(k), \cdots, E_{m + N - 1}(k), which may be held fixed locally in kk) form a real analytic family of rank NN projections with P(k+G)=τ GP(k)τ G −1P(k + G) = \tau_G\, P(k)\, \tau_G^{-1}.

The Bloch bundle of the isolated family is the resulting rank NN real analytic Hermitian vector bundle over the Brillouin torus:

ℰ≔{(k,u)∈ℝ d×L 2(ℝ d/Λ)|u∈imP(k)}/((k,u)∼(k+G,τ Gu)) G∈Λ *⟶T^ d. \mathcal{E} \;\coloneqq\; \Big\{ (k, u) \in \mathbb{R}^d \times L^2\big(\mathbb{R}^d/\Lambda\big) \;\Big\vert\; u \in \mathrm{im}\, P(k) \Big\} \Big/ \big( (k, u) \sim (k + G, \tau_G u) \big)_{G \in \Lambda^\ast} \;\; \longrightarrow \;\; \widehat{T}{}^d \,.
Proposition

(triviality of Bloch bundles)

If d≤3d \leq 3 and HH is time-reversal symmetric (as is the case for HH as in Prop. , with real potential and no magnetic field), then the Bloch bundle (Def. ) of every isolated family of bands is trivializable, even real-analytically.

This is due to Panati 2007, building on Nenciu 1983, see also Brouder et al. 2007. The idea is that time-reversal symmetry implies c 1(ℰ)=−c 1(ℰ)c_1(\mathcal{E}) = - c_1(\mathcal{E}), hence c 1(ℰ)=0c_1(\mathcal{E}) = 0, while for d≤3d \leq 3 the first Chern class is the only obstruction to triviality (cf. Rem. ).

Proposition

(isolated bands are tight-binding models)

Given an isolated family of NN bands (Def. ) with real-analytically trivializable Bloch bundle, choose a real-analytic orthonormal frame, i.e. real-analytic maps u 1,⋯,u N:ℝ d→L 2(ℝ d/Λ)u_1, \cdots, u_N \colon \mathbb{R}^d \to L^2\big(\mathbb{R}^d/\Lambda\big) such that (u j,k) j=1 N\big(u_{j,k}\big)_{j = 1}^N is an orthonormal basis of imP(k)\mathrm{im}\, P(k) and u j,k+G=τ Gu j,ku_{j, k + G} = \tau_G\, u_{j,k}. Then:

  1. The matrices

    H ℰ(k) ij≔⟨u i,k,H(k)u j,k⟩ H_{\mathcal{E}}(k)_{i j} \;\coloneqq\; \big\langle u_{i,k} ,\, H(k)\, u_{j,k} \big\rangle

    are Λ *\Lambda^\ast-periodic in kk and hence constitute a real-analytic Bloch Hamiltonian

    H ℰ:T^ d⟶Herm(N) H_{\mathcal{E}} \;\colon\; \widehat{T}{}^d \longrightarrow \mathrm{Herm}(N)

    (Def. ), whose eigenvalues at kk are E m(k),⋯,E m+N−1(k)E_m(k), \cdots, E_{m+N-1}(k).

  2. The Wannier functions

    w j,a≔U −1(k↦e −ik⋅au j,k),i.e.w j,a(x)=∫ Be ik⋅(x−a)u j,k(x)dk, w_{j,a} \;\coloneqq\; U^{-1}\big( k \mapsto e^{- i k \cdot a}\, u_{j,k} \big) \,, \;\;\;\;\;\; \text{i.e.} \;\;\; w_{j,a}(x) \;=\; \int_B e^{i k \cdot (x - a)}\, u_{j,k}(x)\, d k \,,

    for j∈{1,⋯,N}j \in \{1, \cdots, N\} and a∈Λa \in \Lambda, form an orthonormal basis of the band subspace PL 2(ℝ d)P L^2(\mathbb{R}^d), where P≔U −1(∫ B ⊕P(k)dk)UP \coloneqq U^{-1} \big(\int^\oplus_B P(k)\, d k\big) U. They are lattice translates of each other, w j,a(x)=w j,0(x−a)w_{j,a}(x) = w_{j,0}(x - a), and are exponentially localized.

  3. The unitary isomorphism

    ℓ 2(Λ;ℂ N) →∼ PL 2(ℝ d) δ a⊗e j ↦ w j,a \array{ \ell^2\big(\Lambda;\, \mathbb{C}^N\big) &\xrightarrow{\;\sim\;}& P L^2\big(\mathbb{R}^d\big) \\ \delta_a \otimes e_j &\mapsto& w_{j,a} }

    identifies the band-projected Hamiltonian PHPP H P with the tight-binding Hamiltonian (Def. ) whose Bloch Hamiltonian is H ℰH_{\mathcal{E}}, hence with exponentially decaying hopping matrices

    h(a) ij=⟨w i,a,Hw j,0⟩=∫ T^ de ik⋅aH ℰ(k) ijdk. h(a)_{i j} \;=\; \big\langle w_{i,a} ,\, H\, w_{j,0} \big\rangle \;=\; \int_{\widehat{T}{}^d} e^{i k \cdot a}\, H_{\mathcal{E}}(k)_{i j}\, d k \,.
Proof

Periodicity of H ℰH_{\mathcal{E}} follows since τ G\tau_G is unitary and intertwines H(k)H(k) with H(k+G)H(k+G) (Prop. ). From U(T aψ) k=e −ik⋅a(Uψ) kU(T_a \psi)_k = e^{- i k \cdot a}\, (U\psi)_k (with (T aψ)(x)=ψ(x−a)(T_a \psi)(x) = \psi(x - a)) one gets w j,a=T aw j,0w_{j,a} = T_a w_{j,0} and

⟨w i,a,w j,b⟩=∫ Be ik⋅(a−b)⟨u i,k,u j,k⟩dk=δ ijδ ab, \big\langle w_{i,a} ,\, w_{j,b} \big\rangle \;=\; \int_B e^{i k \cdot (a - b)} \big\langle u_{i,k} ,\, u_{j,k} \big\rangle\, d k \;=\; \delta_{i j}\, \delta_{a b} \,,

and similarly for the matrix elements of HH. Exponential localization of the w j,aw_{j,a} follows from real analyticity of the frame as in Prop. .

Remark

(Bloch Hamiltonians of electrons in crystals)

In summary, the Hamiltonian of a single electron in a crystal, restricted to any isolated family of NN bands with trivializable Bloch bundle — such as any isolated family of bands in a time-reversal symmetric crystal in d≤3d \leq 3 (Prop. ) — is equivalently a real-analytic Bloch Hamiltonian

H ℰ:T^ d⟶Herm(N) H_{\mathcal{E}} \;\colon\; \widehat{T}{}^d \longrightarrow \mathrm{Herm}(N)

(Prop. ), unique up to “gauge transformations” H ℰ(k)↦g(k)H ℰ(k)g(k) †H_{\mathcal{E}}(k) \mapsto g(k)\, H_{\mathcal{E}}(k)\, g(k)^\dagger by real-analytic g:T^ d→U(N)g \colon \widehat{T}{}^d \to U(N), corresponding to the choice of frame (which do not change the isomorphism class of the valence bundle). If the Fermi energy μ\mu lies in a gap within this family of bands, then H ℰH_{\mathcal{E}} takes values in the space Herm μ (n)(N)≃Gr n(ℂ N)\mathrm{Herm}^{(n)}_\mu(N) \simeq \mathrm{Gr}_n(\mathbb{C}^N) of gapped Hamiltonians (Prop. ), and its deformation class, which may be represented by finite-range hopping (Prop. ), is classified by topological K-theory (Rem. ).

On the other hand, if the Bloch bundle of the family of bands under consideration is non-trivial (which requires broken time-reversal symmetry if d≤3d \leq 3, e.g. by magnetic fields), then this family admits no basis of exponentially localized Wannier functions (Brouder et al. 2007). But a larger isolated family containing it (e.g. valence and conduction bands together) may have trivial Bloch bundle, in which case the original family appears as a non-trivial valence bundle of a tight-binding model as above. This is how Chern insulators are realized by tight-binding models.

Wannier functions

The functions w j,aw_{j,a} in Prop. are called Wannier functions (Wannier 1937). Under the Bloch-Floquet transform, they are dual to frames of the Bloch bundle, and their spatial localization is dual to the regularity of the frame in kk: Wannier functions are exponentially localized iff the frame is real analytic (Prop. ). The existence of exponentially localized Wannier functions was established by Kohn 1959 for d=1d = 1, by des Cloizeaux 1964 and Nenciu 1983 for single isolated bands, and by Brouder et al. 2007 and Panati 2007 for composite families of bands (in d≤3d \leq 3, under time-reversal symmetry).

Since the frame is unique only up to gauge transformations g:T^ d→U(N)g \colon \widehat{T}{}^d \to U(N), so are the Wannier functions, and choosing them “maximally localized” is a standard problem in computational solid state physics (Marzari & Vanderbilt 1997, Marzari et al. 2012).

Variants and generalizations

  • Time-periodic Hamiltonians. The analog of Bloch theory for Hamiltonians periodic in time is Floquet theory (Floquet 1883).

  • Magnetic fields. In the presence of a uniform magnetic field the lattice translations commute with the Hamiltonian only up to phases, i.e. they form a projective representation of Λ\Lambda, the magnetic translations (Zak 1964). For rational magnetic flux per unit cell, Bloch theory applies to a correspondingly enlarged magnetic unit cell. For effective Hamiltonians of Bloch electrons in electric and magnetic fields see Kohn 1959, Wannier 1960, Panati, Spohn & Teufel 2003.

  • Hyperbolic lattices. For lattices in hyperbolic space the translation group is a non-abelian Fuchsian group. Its 1-dimensional unitary representations form a torus of dimension 2g2g (with gg the genus of the compactified unit cell), on which an abelian Bloch theory applies, while the higher-dimensional representations lead to “non-abelian Bloch states” (Maciejko & Rayan 2021, 2022, Attar & Boettcher 2022).

  • Interacting electrons. For multi-electron systems the simultaneous translation of all electrons by lattice vectors still commutes with the Hamiltonian, so that many-body eigenstates may be labeled by their total crystal momentum (see the references below).

References

Original articles

The analogous theorem for ordinary differential equations with time-periodic coefficients:

  • Gaston Floquet: Sur les équations différentielles linéaires à coefficients périodiques, Annales scientifiques de l’École Normale Supérieure, Série 2, 12 (1883) 47-88 [doi:10.24033/asens.220]

The original article on Bloch’s theorem, including the tight-binding approximation:

Introducing Brillouin zones:

Textbook accounts

In solid state physics:

  • Neil W. Ashcroft, N. David Mermin: Ch. 8 (and Ch. 10 for the tight-binding method) of: Solid State Physics, Holt, Rinehart and Winston (1976)

  • Charles Kittel: Ch. 7 of: Introduction to Solid State Physics, Wiley (8th ed. 2005)

  • Sheng San Li: Section 4.3 of: Semiconductor Physical Electronics, Springer (2006) [doi:10.1007/0-387-37766-2_4]

  • David Vanderbilt: Ch. 2 of: Berry Phases in Electronic Structure Theory – Electric Polarization, Orbital Magnetization and Topological Insulators, Cambridge University Press (2018) [doi:10.1017/9781316662205]

In mathematical physics:

On absolute continuity of the spectrum:

  • Lawrence E. Thomas: Time dependent approach to scattering from impurities in a crystal, Commun. Math. Phys. 33 (1973) 335-343 [doi:10.1007/BF01646745]

See also:

Tight-binding models

Besides Bloch 1929 and Ashcroft & Mermin 1976, Ch. 10:

  • John C. Slater, George F. Koster: Simplified LCAO Method for the Periodic Potential Problem, Phys. Rev. 94 (1954) 1498 [doi:10.1103/PhysRev.94.1498]

On the periodic vs. non-periodic convention for Bloch Hamiltonians (Rem. ):

Analyticity and the Bloch bundle

On analytic dependence of Bloch waves on the crystal momentum, triviality of the Bloch bundle and existence of exponentially localized Wannier functions:

Effective dynamics of Bloch electrons

Bloch Hamiltonians and topological phases

On Bloch Hamiltonians in the context of topological phases of matter (see also at K-theory classification of topological phases of matter):

Generalizations

Further discussion

For hyperbolic lattices

Variant for lattices in hyperbolic spaces:

For interacting electrons

  • Yuejin Guo, Jean-Marc Langlois, William A. Goddard: Electronic Structure and Valence-Bond Band Structure of Cuprate Superconducting Materials, Science 239 4842 (1988) 896-899 [doi:10.1126/science.239.4842.896, jstor:1700316]

  • Jingsan Hu, Jianfei Gu, Weiyi Zhang: Bloch’s band structures of a pair of interacting electrons in simple one- and two-dimensional lattices, Physics Letters A 414 (2021) 127634 [doi:10.1016/j.physleta.2021.127634]

Wannier functions

The original article:

On their exponential localization see the references above.

On maximally localized Wannier functions:

In the context of fractional Chern insulators exhibiting a fractional quantum anomalous Hall effect:

Last revised on November 15, 2026 at 15:30:54. See the history of this page for a list of all contributions to it.