nLab quantum adiabatic theorem

Redirected from "quantum adiabatic transport".

Context

Quantum systems

quantum logic


quantum physics


quantum probability theory – observables and states


quantum information


quantum technology


quantum computing

Contents

Idea

In quantum physics, the quantum adiabatic theorem for a parameterized quantum system says the following. Under a sufficiently slow motion of the external parameters along a path γ:[0,1]→X\gamma \,\colon\, [0,1] \to X in parameter space XX, eigenstates of the Hamiltonian at γ(0)\gamma(0) approximately evolve into eigenstates at γ(1)\gamma(1), even if the corresponding eigenvalues change significantly. More generally, this holds for states in any part of the spectrum which stays separated from the rest of the spectrum by a gap.

For adiabatic transport along a loop, γ(0)=γ(1)\gamma(0) = \gamma(1), this implies that the eigenspaces of the system are acted on by unitary operators. These are the holonomies of the Berry connection, known as Berry phases in the case of eigen-dimension 1, and as non-abelian Berry phases in the case of degenerate eigenspaces (see below).

The adiabatic parameter-action on ground states of (topologically ordered) quantum materials is one model for quantum computation: see at holonomic quantum computation and adiabatic quantum computation. If these adiabatic quantum gates depend only on the isotopy class of the parameter path, then one speaks of topological quantum computation. This is made explicit in Arovas, Schrieffer & Wilczek 1984, Freedman, Kitaev, Larsen & Wang 2003, p. 7 and Nayak, Simon, Stern, Freedman & Das Sarma 2008, §II.A.2 (p. 6); see also at braiding of anyonic defects and below.

The following graphic is meant to illustrate this general idea for the case of adiabatic transformation along the braiding of nodal points in the Brillouin torus of semi-metal quantum materials (see the discussion at braid group statistics – anyonic band nodes):

(graphics from SS22)

Statement

Preliminaries

Throughout, ℋ\mathcal{H} denotes a complex Hilbert space, ℬ(ℋ)\mathcal{B}(\mathcal{H}) its algebra of bounded operators equipped with the operator norm ‖−‖\Vert - \Vert, and we use units with ℏ=1\hbar = 1.

Adiabatic time evolution

Definition

(path of Hamiltonians)

A path of Hamiltonians is a function

[0,1] ⟶ ℬ(ℋ) s ↦ H(s) \begin{array}{ccc} [0,1] &\longrightarrow& \mathcal{B}(\mathcal{H}) \\ s &\mapsto& H(s) \end{array}

to self-adjoint operators, which is twice continuously differentiable in the operator norm.

Remark

(on paths of Hamiltonians)

  1. Typically such a path arises from a parameterized quantum system, namely from a family of Hamiltonians

    X ⟶ ℬ(ℋ) x ↦ H^(x) \begin{array}{ccc} X &\longrightarrow& \mathcal{B}(\mathcal{H}) \\ x &\mapsto& \hat H(x) \end{array}

    over a smooth manifold XX of external parameters, by restriction along a smooth path γ:[0,1]→X\gamma \colon [0,1] \to X in parameter space:

    H(s)≔H^(γ(s)). H(s) \;\coloneqq\; \hat H\big(\gamma(s)\big) \,.
  2. The restriction to bounded operators is for simplicity of exposition. Everything below generalizes to unbounded Hamiltonians which are bounded below and have a common domain DD, such that s↦H(s)s \mapsto H(s) is C 2C^2 as a function to bounded operators from DD, equipped with the graph norm? of H(0)H(0), to ℋ\mathcal{H} (Avron, Seiler & Yaffe 1987, §2, Teufel 2003).

  3. The C 2C^2-assumption is convenient but stronger than necessary: Kato 1950 only assumes that the first and second derivatives of the spectral projections (Def. below) are piecewise continuous.

Running through such a path in physical time t∈[0,τ]t \in [0,\tau] means that the system evolves under the time-dependent Hamiltonian t↦H(t/τ)t \mapsto H(t/\tau). In terms of the rescaled time parameter s≔t/τ∈[0,1]s \coloneqq t/\tau \in [0,1], the Schrödinger equation then reads as follows:

Definition

(actual time evolution)

For τ∈(0,∞)\tau \in (0,\infty) write

ε≔1/τ \varepsilon \;\coloneqq\; 1/\tau

(the adiabatic parameter), and let

[0,1] ⟶ ℬ(ℋ) s ↦ U ε(s) \begin{array}{ccc} [0,1] &\longrightarrow& \mathcal{B}(\mathcal{H}) \\ s &\mapsto& U_\varepsilon(s) \end{array}

be the solution of

iε∂ sU ε(s)=H(s)U ε(s),U ε(0)=id ℋ. \mathrm{i}\, \varepsilon \, \partial_s U_\varepsilon(s) \;=\; H(s)\, U_\varepsilon(s) \,, \;\;\;\;\; U_\varepsilon(0) \;=\; \mathrm{id}_{\mathcal{H}} \,.

This exists, is unique, and is unitary for all ss, being given by the time-ordered exponential of −iε∫H-\tfrac{\mathrm{i}}{\varepsilon}\int H. (It is called the dynamical transformation in Kato 1950, §2.)

The adiabatic limit is the limit ε→0\varepsilon \to 0, i.e. τ→∞\tau \to \infty, in which the path of Hamiltonians is traversed infinitely slowly. In the rescaled time ss the dynamics becomes fast in this limit. For instance, if H(s)=HH(s) = H is constant and Hψ=EψH \psi = E \psi, then U ε(s)ψ=e −iEs/εψU_\varepsilon(s)\psi = e^{-\mathrm{i} E s/\varepsilon} \psi.

Gap condition and spectral projections

Definition

(gap condition)

A path of Hamiltonians (Def. ) satisfies the gap condition if, for all s∈[0,1]s \in [0,1], the spectrum of H(s)H(s) decomposes into two disjoint non-empty closed subsets

σ(H(s))=σ *(s)⊔σ ⊥(s) \sigma\big(H(s)\big) \;=\; \sigma_\ast(s) \,\sqcup\, \sigma_\perp(s)

such that

  1. there is a contour Γ(s)⊂ℂ\Gamma(s) \subset \mathbb{C} encircling σ *(s)\sigma_\ast(s) but not σ ⊥(s)\sigma_\perp(s), which may be chosen to depend continuously on ss,

  2. the gap between the two parts is bounded below uniformly in ss:

    g≔inf s∈[0,1]dist(σ *(s),σ ⊥(s))>0. g \;\coloneqq\; \inf_{s \in [0,1]} \mathrm{dist}\big( \sigma_\ast(s),\, \sigma_\perp(s) \big) \;\gt\; 0 \,.

We speak of gapped ground states if σ *(s)\sigma_\ast(s) is the bottom part of the spectrum: supσ *(s)<infσ ⊥(s)\sup \sigma_\ast(s) \lt \inf \sigma_\perp(s).

In the special case that σ *(s)={λ(s)}\sigma_\ast(s) = \{\lambda(s)\} is a single eigenvalue, we say that the selected eigenstates are exactly degenerate, where the degeneracy may be 1. This is the case considered in Kato 1950.

Definition

(spectral projection)

Given a path satisfying the gap condition (Def. ), the spectral projections onto σ *(s)\sigma_\ast(s) are given by the Riesz formula

P(s)≔12πi∮ Γ(s)(z−H(s)) −1dz. P(s) \;\coloneqq\; \frac{1}{2\pi\mathrm{i}} \oint_{\Gamma(s)} \big( z - H(s) \big)^{-1} \, \mathrm{d}z \,.

We write

ℋ *(s)≔im(P(s))⊂ℋ \mathcal{H}_\ast(s) \;\coloneqq\; \mathrm{im}\big(P(s)\big) \;\subset\; \mathcal{H}

for their image spaces, which in the case of gapped ground states are the spaces of (quasi-)ground states.

Proposition

(properties of the spectral projections)

The spectral projections P(s)P(s) (Def. ) satisfy:

  1. Each P(s)P(s) is an orthogonal projection which commutes with H(s)H(s).

  2. The function s↦P(s)s \mapsto P(s) is C 2C^2 in the operator norm.

  3. The dimension of ℋ *(s)\mathcal{H}_\ast(s) is independent of ss (possibly infinite).

  4. The derivative P′≔∂ sPP' \coloneqq \partial_s P is off-diagonal with respect to the decomposition ℋ=ℋ *(s)⊕ℋ *(s) ⊥\mathcal{H} = \mathcal{H}_\ast(s) \oplus \mathcal{H}_\ast(s)^\perp, in that

    PP′P=0,P′=P′P+PP′, P \, P' \, P \;=\; 0 \,, \;\;\;\;\; P' \;=\; P' P + P P' \,,

    and hence

    P′=[[P′,P],P]. P' \;=\; \big[ [P', P],\, P \big] \,.
Proof

The first statement is standard functional calculus (see at Riesz formula). The second follows since the resolvent (z−H(s)) −1(z - H(s))^{-1} is C 2C^2 in ss, locally uniformly for zz on the contour. The third follows from the general fact that projections at operator-norm distance ‖P(s)−P(s′)‖<1\Vert P(s) - P(s') \Vert \lt 1 are unitarily equivalent, together with continuity of PP and connectedness of [0,1][0,1]. For the last statement, differentiate P 2=PP^2 = P to get P′P+PP′=P′P' P + P P' = P', and then multiply this from both sides by PP to get 2PP′P=PP′P2 P P' P = P P' P. With this, [[P′,P],P]=P′P−2PP′P+PP′=P′\big[[P',P],P\big] = P'P - 2 P P' P + P P' = P'. (These are equations (17) and (19) in Kato 1950, and (2.5) in Avron, Seiler & Yaffe 1987.)

Kato’s parallel transport

Definition

(Kato’s parallel transport)

Given a C 1C^1 path s↦P(s)s \mapsto P(s) of orthogonal projections, its Kato transport, called the adiabatic transformation in Kato 1950, §3, is the solution

[0,1] ⟶ ℬ(ℋ) s ↦ K(s) \begin{array}{ccc} [0,1] &\longrightarrow& \mathcal{B}(\mathcal{H}) \\ s &\mapsto& K(s) \end{array}

of

∂ sK(s)=[P′(s),P(s)]K(s),K(0)=id ℋ. \partial_s K(s) \;=\; \big[ P'(s),\, P(s) \big] \, K(s) \,, \;\;\;\;\; K(0) \;=\; \mathrm{id}_{\mathcal{H}} \,.
Proposition

(Kato transport is unitary and intertwining)

The Kato transport K(s)K(s) (Def. ) is unitary for all ss, and it intertwines the projections:

K(s)P(0)=P(s)K(s). K(s)\, P(0) \;=\; P(s)\, K(s) \,.

Hence it restricts to unitary isomorphisms

K(s)| ℋ *(0):ℋ *(0)→∼ℋ *(s), K(s)\vert_{\mathcal{H}_\ast(0)} \;\colon\; \mathcal{H}_\ast(0) \xrightarrow{\;\;\sim\;\;} \mathcal{H}_\ast(s) \,,

and similarly between the orthogonal complements ℋ *(0) ⊥→∼ℋ *(s) ⊥\mathcal{H}_\ast(0)^\perp \xrightarrow{\sim} \mathcal{H}_\ast(s)^\perp.

Proof

The generator [P′,P][P', P] is anti-self-adjoint, since [P′,P] †=[P,P′]=−[P′,P][P',P]^\dagger = [P, P'] = -[P', P], so that K(s)K(s) is unitary. For the intertwining property, consider Q(s)≔K(s) †P(s)K(s)Q(s) \coloneqq K(s)^\dagger P(s) K(s). Then

Q′=K †(P′−[[P′,P],P])K=0 Q' \;=\; K^\dagger \Big( P' - \big[ [P',P], P \big] \Big) K \;=\; 0

by Prop. , so that Q(s)=Q(0)=P(0)Q(s) = Q(0) = P(0). Kato’s original argument instead observes that P(s)K(s)P(s) K(s) satisfies the same differential equation as K(s)K(s), and hence coincides with K(s)P(0)K(s)P(0) (Kato 1950, (20-22)). The statement about the orthogonal complements follows since [P′,P]=[(1−P)′,1−P][P',P] = \big[(1-P)', 1-P\big], so that KK is also the Kato transport of 1−P1 - P.

Remark

(Kato transport as parallel transport of the Berry connection)

For ψ 0∈ℋ *(0)\psi_0 \in \mathcal{H}_\ast(0), the transported vectors ψ(s)≔K(s)ψ 0∈ℋ *(s)\psi(s) \coloneqq K(s)\psi_0 \in \mathcal{H}_\ast(s) satisfy (using Prop. )

P(s)ψ′(s)=0. P(s)\, \psi'(s) \;=\; 0 \,.

This says that they are covariantly constant with respect to the connection ∇≔P∘d\nabla \coloneqq P \circ \mathrm{d} on the vector bundle whose fibers are the spaces ℋ *\mathcal{H}_\ast. This is the sub-bundle of the trivial bundle with fiber ℋ\mathcal{H} cut out by the projections PP. Therefore the Kato transport is the parallel transport of this connection, which is known as the Berry connection:

  • If H^\hat H is a family of Hamiltonians over parameter space XX (Rem. ) satisfying the gap condition at every point, then the spaces ℋ *(H^(x))\mathcal{H}_\ast\big(\hat H(x)\big) form a vector bundle ℰ→X\mathcal{E} \to X with connection ∇=P∘d\nabla = P \circ \mathrm{d}. Kato transport along a loop γ\gamma in XX is the holonomy

    hol ∇(γ)∈U(ℰ γ(0)) \mathrm{hol}_\nabla(\gamma) \;\in\; \mathrm{U}\big(\mathcal{E}_{\gamma(0)}\big)

    of this connection. It depends on the loop, not just on its base point. In general it also depends on more than the homotopy class of the loop.

  • In terms of a local orthonormal frame (ψ a) a(\psi_a)_a of ℰ\mathcal{E}, the connection form has components

    A ab=i⟨ψ a|dψ b⟩. A_{a b} \;=\; \mathrm{i}\, {\langle \psi_a \vert \mathrm{d} \psi_b \rangle} \,.

    This is the non-abelian Berry connection (Wilczek & Zee 1984). In the case of rank 1, it is the original Berry connection A=i⟨ψ|dψ⟩A = \mathrm{i}\langle \psi \vert \mathrm{d}\psi \rangle (Simon 1983, Berry 1984), whose holonomy along a loop is multiplication by the Berry phase exp(i∮ γA)∈U(1)\exp\big(\mathrm{i}\oint_\gamma A\big) \in \mathrm{U}(1).

Remark

(transport of several spectral subspaces)

More generally, for a family of mutually orthogonal projections P n(s)P_n(s) with ∑ nP n(s)=1\sum_n P_n(s) = 1, the solution of

∂ sK(s)=12∑ n[P n′(s),P n(s)]K(s),K(0)=id ℋ \partial_s K(s) \;=\; \tfrac{1}{2} \textstyle{\sum_n} \big[ P_n'(s),\, P_n(s) \big] \, K(s) \,, \;\;\;\;\; K(0) = \mathrm{id}_{\mathcal{H}}

is unitary and intertwines all of these projections simultaneously, K(s)P n(0)=P n(s)K(s)K(s) P_n(0) = P_n(s) K(s) (Kato 1950, §5). For two projections P 1=PP_1 = P and P 2=1−PP_2 = 1-P this reduces to Def. .

General

The adiabatic theorem says that the actual dynamics follows the spectral subspaces ℋ *(s)\mathcal{H}_\ast(s) bordered by gaps. To make this precise, one compares the actual time evolution U εU_\varepsilon (Def. ) with a modified evolution which preserves these spectral subspaces exactly:

Definition

(adiabatic evolution)

Given a path of Hamiltonians satisfying the gap condition (Def. ), its adiabatic evolution (Avron, Seiler & Yaffe 1987, (1.0), following Kato 1950) is the solution

[0,1] ⟶ ℬ(ℋ) s ↦ U ε A(s) \begin{array}{ccc} [0,1] &\longrightarrow& \mathcal{B}(\mathcal{H}) \\ s &\mapsto& U^A_\varepsilon(s) \end{array}

of

iε∂ sU ε A(s)=(H(s)+iε[P′(s),P(s)])U ε A(s),U ε A(0)=id ℋ. \mathrm{i}\, \varepsilon \, \partial_s U^A_\varepsilon(s) \;=\; \Big( H(s) + \mathrm{i}\,\varepsilon \big[ P'(s),\, P(s) \big] \Big) \, U^A_\varepsilon(s) \,, \;\;\;\;\; U^A_\varepsilon(0) \;=\; \mathrm{id}_{\mathcal{H}} \,.

The correction term iε[P′,P]\mathrm{i}\varepsilon[P',P] is the generator of Kato’s parallel transport (Def. ), rescaled to the time parameter of the Schrödinger equation. It is the unique correction to HH which makes the evolution decouple ℋ *\mathcal{H}_\ast from its complement while approximating the actual evolution as well as possible (Avron, Seiler & Yaffe 1987, §1). Accordingly, the adiabatic evolution splits into a geometric and a dynamical part:

Proposition

(structure of the adiabatic evolution)

The adiabatic evolution (Def. ) factors as

U ε A(s)=K(s)D ε(s). U^A_\varepsilon(s) \;=\; K(s) \, D_\varepsilon(s) \,.

Here K(s)K(s) is the Kato transport of P(s)P(s) (Def. ), and D ε(s)D_\varepsilon(s) is the solution of

iε∂ sD ε(s)=(K(s) †H(s)K(s))D ε(s),D ε(0)=id ℋ, \mathrm{i}\,\varepsilon\, \partial_s D_\varepsilon(s) \;=\; \big( K(s)^\dagger H(s) K(s) \big) \, D_\varepsilon(s) \,, \;\;\;\;\; D_\varepsilon(0) \;=\; \mathrm{id}_{\mathcal{H}} \,,

which is unitary and commutes with P(0)P(0). Consequently, U ε A(s)U^A_\varepsilon(s) is unitary and exactly intertwines the spectral projections:

U ε A(s)P(0)=P(s)U ε A(s). U^A_\varepsilon(s) \, P(0) \;=\; P(s) \, U^A_\varepsilon(s) \,.
Proof

Using ∂ sK=[P′,P]K\partial_s K = [P',P] K, we have

iε∂ s(KD ε)=iε[P′,P]KD ε+K(K †HK)D ε=(H+iε[P′,P])KD ε, \mathrm{i}\varepsilon\, \partial_s \big( K D_\varepsilon \big) \;=\; \mathrm{i}\varepsilon [P',P] K D_\varepsilon + K \big( K^\dagger H K \big) D_\varepsilon \;=\; \big( H + \mathrm{i}\varepsilon[P',P] \big) K D_\varepsilon \,,

which gives the factorization, by uniqueness of solutions. The generator K †HKK^\dagger H K of D εD_\varepsilon is self-adjoint, so that D εD_\varepsilon is unitary. It commutes with P(0)P(0), since by Prop. and Prop. we have

K †HKP(0)=K †HPK=K †PHK=P(0)K †HK. K^\dagger H K \, P(0) \;=\; K^\dagger H P K \;=\; K^\dagger P H K \;=\; P(0)\, K^\dagger H K \,.

Hence D ε(s)D_\varepsilon(s) commutes with P(0)P(0), and the intertwining property of U ε AU^A_\varepsilon follows from that of KK. (For a direct proof of the intertwining property see Avron, Seiler & Yaffe 1987, Lem. 2.3.)

In words: The adiabatic evolution moves the spectral subspace ℋ *(0)\mathcal{H}_\ast(0) onto ℋ *(s)\mathcal{H}_\ast(s) by Kato’s unitary parallel transport K(s)K(s), while inside the spectral subspace it runs the actual dynamics D εD_\varepsilon. The adiabatic theorem says that this correctly describes the actual dynamics up to errors of order ε=1/τ\varepsilon = 1/\tau:

Theorem

(quantum adiabatic theorem)

Given a path of Hamiltonians (Def. ) satisfying the gap condition (Def. ), there is a constant C∈(0,∞)C \in (0,\infty), depending on the path but not on ε\varepsilon, such that for all ε>0\varepsilon \gt 0:

sup s∈[0,1]‖U ε(s)−U ε A(s)‖≤Cε. \sup_{s \in [0,1]} \big\Vert U_\varepsilon(s) - U^A_\varepsilon(s) \big\Vert \;\leq\; C \, \varepsilon \,.

In particular, transitions out of the spectral subspace are suppressed:

sup s∈[0,1]‖(1−P(s))U ε(s)P(0)‖≤Cε. \sup_{s \in [0,1]} \big\Vert \big( 1 - P(s) \big) \, U_\varepsilon(s) \, P(0) \big\Vert \;\leq\; C \, \varepsilon \,.

This is due to Kato 1950, (29) for the case of an exactly degenerate eigenvalue of finite multiplicity (Def. ). In the above generality it is due to Nenciu 1980, 1981 for bounded Hamiltonians, and Avron, Seiler & Yaffe 1987, Thm. 2.4 for unbounded Hamiltonians. A textbook account is in Teufel 2003.

Proof

(sketch)

The key observation is that the off-diagonal operator [P′,P][P',P] is a commutator with HH. Namely, with the resolvent R(z)≔(z−H) −1R(z) \coloneqq (z - H)^{-1}, the operator

Y(s)≔−12πi∮ Γ(s)R(z)P′(s)R(z)dz Y(s) \;\coloneqq\; -\frac{1}{2\pi \mathrm{i}} \oint_{\Gamma(s)} R(z) \, P'(s) \, R(z) \, \mathrm{d}z

satisfies

[H(s),Y(s)]=[P′(s),P(s)]. \big[ H(s),\, Y(s) \big] \;=\; \big[ P'(s),\, P(s) \big] \,.

This follows from the identity HR(z)=R(z)H=zR(z)−1H R(z) = R(z) H = z R(z) - 1, which gives [H,RXR]=RX−XR[H, R X R] = R X - X R for any XX, together with the Riesz formula for PP. Moreover, YY is C 1C^1 in ss, since HH is C 2C^2.

Now consider the wave operator Ω(s)≔U ε A(s) †U ε(s)\Omega(s) \coloneqq U^A_\varepsilon(s)^\dagger \, U_\varepsilon(s), so that ‖U ε(s)−U ε A(s)‖=‖Ω(s)−1‖\Vert U_\varepsilon(s) - U^A_\varepsilon(s) \Vert = \Vert \Omega(s) - 1 \Vert. The two differential equations give

Ω′=−U ε A†[P′,P]U ε=−U ε A†[H,Y]U ε. \Omega' \;=\; - U^{A \dagger}_\varepsilon \, [P', P] \, U_\varepsilon \;=\; - U^{A \dagger}_\varepsilon \, [H, Y] \, U_\varepsilon \,.

This is of order 1, but it oscillates fast and may be integrated by parts:

U ε A†[H,Y]U ε=−iε∂ s(U ε A†YU ε)+iεU ε A†(Y′−[P′,P]Y)U ε. U^{A \dagger}_\varepsilon \, [H, Y] \, U_\varepsilon \;=\; - \mathrm{i}\varepsilon \, \partial_s \big( U^{A \dagger}_\varepsilon \, Y \, U_\varepsilon \big) + \mathrm{i}\varepsilon \, U^{A \dagger}_\varepsilon \big( Y' - [P',P]\,Y \big) U_\varepsilon \,.

Integrating, and using unitarity of U εU_\varepsilon and U ε AU^A_\varepsilon, yields

‖Ω(s)−1‖≤ε(‖Y(0)‖+‖Y(s)‖+∫ 0 s(‖Y′‖+‖P′‖‖Y‖)). \big\Vert \Omega(s) - 1 \big\Vert \;\leq\; \varepsilon \Big( \Vert Y(0) \Vert + \Vert Y(s) \Vert + \textstyle{\int_0^s} \big( \Vert Y' \Vert + \Vert P' \Vert \, \Vert Y \Vert \big) \Big) \,.

For the second statement, observe that (1−P(s))U ε A(s)P(0)=0(1-P(s))\, U^A_\varepsilon(s)\, P(0) = 0, by Prop. .

This is the argument of Avron, Seiler & Yaffe 1987, §2 (wave operator (2.6), integration by parts Lemma 2.5), which in turn generalizes that of Kato 1950, §4.

Remark

(on the adiabatic theorem)

  1. (size of the constant) By the above proof, the constant CC in Theorem is controlled by ‖Y‖\Vert Y \Vert and ‖Y′‖\Vert Y' \Vert. Roughly, it scales like

    ‖H′‖g 2,‖H″‖g 2,‖H′‖ 2g 3, \frac{\Vert H' \Vert}{g^2} \,, \;\; \frac{\Vert H'' \Vert}{g^2} \,, \;\; \frac{\Vert H' \Vert^2}{g^3} \,,

    so that the adiabatic regime is τ≫‖H′‖/g 2\tau \gg \Vert H' \Vert / g^2, etc. Explicit bounds of this form are given in Jansen, Ruskai & Seiler 2007.

  2. (higher order and exponential accuracy) The error of order ε\varepsilon comes from the boundary terms in the integration by parts. If s↦H(s)s \mapsto H(s) is smooth and constant near the endpoints of the path, then repeated integration by parts shows that the transitions (1−P(1))U ε(1)P(0)(1-P(1))\, U_\varepsilon(1)\, P(0) out of the spectral subspace at the end of the path are of order ε n\varepsilon^n for all nn (Avron, Seiler & Yaffe 1987, Cor. 2.6). Under suitable analyticity assumptions they are even exponentially small in 1/ε1/\varepsilon (Nenciu 1993, Hagedorn & Joye 2002, Teufel 2003).

  3. (level crossings and gapless cases) If σ *(s)\sigma_\ast(s) touches the rest of the spectrum at finitely many points ss, then the conclusion of Theorem still holds, but with o(1)o(1) in place of O(ε)O(\varepsilon) (Kato 1950, §4). There are versions of the adiabatic theorem without any gap condition, again without an error rate in general (Avron & Elgart 1999).

  4. (on the hypotheses) The theorem concerns a fixed path, traversed ever more slowly. Heuristic “adiabaticity conditions” formulated only in terms of instantaneous matrix elements may fail (Marzlin & Sanders 2004, Amin 2009).

  5. (many-body systems) For quantum lattice systems, ‖H′‖\Vert H' \Vert grows with the volume, so that the above bounds become useless in the thermodynamic limit. There are versions of the adiabatic theorem with error bounds uniform in the system size, for local observables and assuming a uniform gap (Bachmann, De Roeck & Fraas 2017, 2018, Monaco & Teufel 2019, Henheik & Teufel 2022). A related approach is the quasi-adiabatic continuation of Hastings & Wen 2005 (see also Bachmann, Michalakis, Nachtergaele & Sims 2012). It provides quasi-local unitaries which map ground states along gapped paths onto each other, whereas the generator [P′,P][P',P] of Kato’s transport is in general not quasi-local.

For gapped ground states

We now specialize to the case of gapped ground states (Def. ). Here the dynamical part D εD_\varepsilon of the adiabatic evolution (Prop. ) becomes trivial up to a global phase. Hence the adiabatic theorem says that slow motion of parameters acts on ground states by the unitary parallel transport of the Berry connection.

Exactly degenerate ground states

Assume first that the ground states are exactly degenerate, so that σ *(s)={E 0(s)}\sigma_\ast(s) = \{E_0(s)\} is a single eigenvalue and

ℋ 0(s)≔ℋ *(s)=ker(H(s)−E 0(s)) \mathcal{H}_0(s) \;\coloneqq\; \mathcal{H}_\ast(s) \;=\; \mathrm{ker}\big( H(s) - E_0(s) \big)

is the ground state space.

Proposition

(adiabatic evolution of degenerate ground states)

In this case the adiabatic evolution (Def. ) on ground states is

U ε A(s)P(0)=e −iε∫ 0 sE 0(u)duK(s)P(0). U^A_\varepsilon(s) \, P(0) \;=\; e^{- \tfrac{\mathrm{i}}{\varepsilon} \int_0^s E_0(u) \, \mathrm{d}u} \, K(s) \, P(0) \,.
Proof

On ℋ 0(0)\mathcal{H}_0(0), the generator of D εD_\varepsilon is K †HKP(0)=K †HPK=E 0P(0)K^\dagger H K P(0) = K^\dagger H P K = E_0 \, P(0), by Prop. . So D ε(s)D_\varepsilon(s) acts on ℋ 0(0)\mathcal{H}_0(0) as multiplication by the phase exp(−iε∫ 0 sE 0)\exp\big(-\tfrac{\mathrm{i}}{\varepsilon}\int_0^s E_0\big), and the claim follows by Prop. . (Cf. Avron, Seiler & Yaffe 1987, p. 34.)

Combined with Theorem , this gives the main statement of Kato 1950, (29):

Corollary

(adiabatic transport of ground states)

Given a path of Hamiltonians with exactly degenerate gapped ground states, the actual time evolution on ground states satisfies, for all s∈[0,1]s \in [0,1],

‖U ε(s)P(0)−e −iθ ε(s)K(s)P(0)‖≤Cε,whereθ ε(s)≔1ε∫ 0 sE 0(u)du. \Big\Vert U_\varepsilon(s)\, P(0) \;-\; e^{- \mathrm{i}\, \theta_\varepsilon(s)} \, K(s)\, P(0) \Big\Vert \;\leq\; C\, \varepsilon \,, \;\;\;\;\;\; \text{where} \;\; \theta_\varepsilon(s) \;\coloneqq\; \frac{1}{\varepsilon}\int_0^s E_0(u)\, \mathrm{d}u \,.

Hence in the adiabatic limit ε→0\varepsilon \to 0, the time evolution acts on ground states, up to the dynamical phase θ ε\theta_\varepsilon, by the unitary operator

ℋ 0(0) →∼ ℋ 0(s) ψ ↦ K(s)ψ, \begin{array}{ccc} \mathcal{H}_0(0) &\xrightarrow{\;\;\sim\;\;}& \mathcal{H}_0(s) \\ \psi &\mapsto& K(s)\, \psi \,, \end{array}

which is the parallel transport of the Berry connection (Rem. ).

Remark

(geometric nature of adiabatic ground state transport)

The unitary K(s)| ℋ 0(0)K(s)\vert_{\mathcal{H}_0(0)} in Cor. has the following properties.

  1. It is independent of the running time τ=1/ε\tau = 1/\varepsilon.

  2. It is independent of the parameterization of the path, since the defining equation ∂ sK=[P′,P]K\partial_s K = [P',P]\, K is invariant under reparameterization s↦f(s)s \mapsto f(s). So KK depends only on the curve of ground state spaces s↦ℋ 0(s)s \mapsto \mathcal{H}_0(s). For a parameterized system (Rem. ) it therefore depends only on the unparameterized path in parameter space XX.

  3. The dynamical phase θ ε(s)\theta_\varepsilon(s) does depend on τ\tau and on the parameterization, and it diverges as ε→0\varepsilon \to 0. But it is a global phase on all of ℋ 0\mathcal{H}_0, the same for all ground states. So it does not affect the relative phases and amplitudes between ground states.

Therefore the physically meaningful part of the adiabatic evolution of ground states is the purely geometric operation of parallel transport in the vector bundle ℰ→X\mathcal{E} \to X of ground state spaces, with respect to the Berry connection ∇=P∘d\nabla = P \circ \mathrm{d} (Rem. ).

In particular, adiabatic transport along a loop γ\gamma in parameter space, based at x 0∈Xx_0 \in X, maps the ground state space back to itself, acting on it by the holonomy

hol ∇(γ)∈U(ℋ 0(x 0)): \mathrm{hol}_\nabla(\gamma) \;\in\; \mathrm{U}\big( \mathcal{H}_0(x_0) \big) \,:
  • For non-degenerate ground states, dimℋ 0=1\mathrm{dim}\, \mathcal{H}_0 = 1, this is multiplication by the Berry phase exp(i∮ γA)∈U(1)\exp\big( \mathrm{i}\oint_\gamma A \big) \in \mathrm{U}(1) (Simon 1983, Berry 1984).

  • For degenerate ground states, dimℋ 0=n≥2\mathrm{dim}\, \mathcal{H}_0 = n \geq 2, this is a unitary n×nn \times n matrix, the non-abelian Berry phase (Wilczek & Zee 1984). It is given by the path-ordered exponential of the non-abelian Berry connection form A ab=i⟨ψ a|dψ b⟩A_{a b} = \mathrm{i}\langle \psi_a \vert \mathrm{d}\psi_b \rangle.

Quasi-degenerate ground states

In practice, ground state degeneracy is often only approximate. The bottom part σ *(s)\sigma_\ast(s) of the spectrum is then a narrow band of width

δ(s)≔supσ *(s)−infσ *(s) \delta(s) \;\coloneqq\; \sup \sigma_\ast(s) - \inf \sigma_\ast(s)

much smaller than the gap gg to the rest of the spectrum. Write E¯ 0(s)\bar E_0(s) for the midpoint of this band. The above conclusions still hold, provided the running time τ\tau is long compared to the inverse gap but short compared to the inverse splitting:

Proposition

(adiabatic evolution of quasi-degenerate ground states)

With CC as in Theorem , we have for all s∈[0,1]s \in [0,1]

‖U ε(s)P(0)−e −iε∫ 0 sE¯ 0(u)duK(s)P(0)‖≤Cε+12ε∫ 0 sδ(u)du. \Big\Vert U_\varepsilon(s)\, P(0) \;-\; e^{- \tfrac{\mathrm{i}}{\varepsilon} \int_0^s \bar E_0(u) \, \mathrm{d}u} \, K(s)\, P(0) \Big\Vert \;\leq\; C\, \varepsilon \;+\; \frac{1}{2\varepsilon} \int_0^s \delta(u)\, \mathrm{d}u \,.

Hence, up to a global phase, the ground states are transported by K(s)K(s) as in Cor. , within the time window

C≪τ≪1/sup sδ(s). C \;\ll\; \tau \;\ll\; 1/\sup_s \delta(s) \,.
Proof

On ℋ *(0)\mathcal{H}_\ast(0), the generator of D εD_\varepsilon (Prop. ) is K †HKP(0)=E¯ 0P(0)+BK^\dagger H K P(0) = \bar E_0 P(0) + B, where B≔K †(H−E¯ 0)PKB \coloneqq K^\dagger (H - \bar E_0) P K is self-adjoint with ‖B‖≤δ/2\Vert B \Vert \leq \delta/2. Since the scalar part commutes with everything, we have D ε(s)P(0)=e −iε∫ 0 sE¯ 0D˜ ε(s)P(0)D_\varepsilon(s) P(0) = e^{-\tfrac{\mathrm{i}}{\varepsilon}\int_0^s \bar E_0} \widetilde D_\varepsilon(s) P(0), where iε∂ sD˜ ε=BD˜ ε\mathrm{i}\varepsilon\, \partial_s \widetilde D_\varepsilon = B \widetilde D_\varepsilon. This implies ‖(D˜ ε(s)−1)P(0)‖≤1ε∫ 0 s‖B‖\Vert (\widetilde D_\varepsilon(s) - 1) P(0) \Vert \leq \tfrac{1}{\varepsilon}\int_0^s \Vert B \Vert. Now combine with Theorem .

Remark

(topological ground state degeneracy)

This quasi-degenerate situation is that of topologically ordered quantum materials. Here the splitting δ\delta of the ground state degeneracy is exponentially small in the system size LL, δ∼e −L/ξ\delta \sim e^{- L/\xi}, while the gap gg remains finite as L→∞L \to \infty (Hastings & Wen 2005). So the time window in Prop. , in which adiabatic transport of ground states is the unitary parallel transport KK, is wide for large systems.

Outlook: Adiabatic quantum gates

The above statements make adiabatic transport of gapped ground states a means for implementing quantum gates. The (quasi-)degenerate ground state space ℋ 0\mathcal{H}_0 serves as the code space, and slowly cycling the external parameters through a loop γ\gamma executes the gate hol ∇(γ)\mathrm{hol}_\nabla(\gamma), up to a global phase:

References

General

The original formulation:

  • Max Born, Vladimir A. Fock: Beweis des Adiabatensatzes, Zeitschrift für Physik 51 (1928) 165–180 [doi:10.1007/BF01343193]

  • Tosio Kato: On the Adiabatic Theorem of Quantum Mechanics, J. Phys. Soc. Jpn. 5 (1950) 435-439 [doi:10.1143/JPSJ.5.435]

    (introducing the “adiabatic transformation” U′(s)=[E′(s),E(s)]U(s)U'(s) = [E'(s), E(s)]\, U(s), now known as Kato’s parallel transport, and proving that the actual time evolution of an isolated eigenspace approaches it, up to a dynamical phase, with error O(1/τ)O(1/\tau); or o(1)o(1) in the presence of finitely many level crossings)

Textbook accounts:

Rigorous generalizations and refinements, including higher-order and exponentially small error estimates:

  • Gheorghe Nenciu: On the adiabatic theorem of quantum mechanics, J. Phys. A: Math. Gen. 13 (1980) L15 [doi:10.1088/0305-4470/13/2/002]

  • Gheorghe Nenciu: Adiabatic theorem and spectral concentration, Commun. Math. Phys. 82 (1981) 121-135 [doi:10.1007/BF01206948]

  • Joseph E. Avron, Ruedi Seiler, Laurence G. Yaffe: Adiabatic theorems and applications to the quantum Hall effect, Commun. Math. Phys. 110 (1987) 33-49 [doi:10.1007/BF01209015], Erratum: Commun. Math. Phys. 156 (1993) 649-650 [doi:10.1007/BF02096867]

    (introducing the “adiabatic evolution” generated by H(s)+iτ[P′(s),P(s)]H(s) + \tfrac{\mathrm{i}}{\tau}[P'(s), P(s)] for spectral subspaces bordered by gaps, and proving that it approximates the actual dynamics to order 1/τ1/\tau, with tunneling out of the spectral subspace smaller than any power of 1/τ1/\tau for smooth switching)

  • Gheorghe Nenciu: Linear adiabatic theory. Exponential estimates, Commun. Math. Phys. 152 (1993) 479-496 [doi:10.1007/BF02096616]

  • George A. Hagedorn, Alain Joye: Elementary exponential error estimates for the adiabatic approximation, J. Math. Anal. Appl. 267 (2002) 235-246 [doi:10.1006/jmaa.2001.7765]

With explicit bounds (motivated by adiabatic quantum computation):

On the failure of naive “adiabaticity conditions” when the rigorous hypotheses are dropped:

Without a spectral gap condition:

See also:

Holonomy and Berry phases

Interpretation of adiabatic transport as parallel transport in a vector bundle of eigenspaces over parameter space, with holonomy given by Berry phases:

On the non-abelian generalization to degenerate eigenspaces:

Textbook account:

  • Arno Bohm, Ali Mostafazadeh, Hiroyasu Koizumi, Qian Niu, Joseph Zwanziger: The Geometric Phase in Quantum Systems, Texts and Monographs in Physics, Springer (2003) [doi:10.1007/978-3-662-10333-3]

On adiabatic transport subject to repeated quantum measurement:

In condensed matter theory

Discussion in solid state physics and in view of gapped topological phases of matter (quantum Hall effect, topological insulators), see also Avron, Seiler & Yaffe 1987:

With error bounds uniform in the system size, for gapped quantum many-body systems:

Review:

For adiabatic quantum computation

On adiabatic quantum computation via non-degenerate ground states:

Review:

For holonomic quantum computation

On holonomic quantum computation by non-abelian Berry phases of degenerate ground states:

For topological quantum computation

References which make explicit that the standard model of topological quantum computation by braiding of anyons is a form of adiabatic quantum computation:

Last revised on November 15, 2026 at 17:46:48. See the history of this page for a list of all contributions to it.