betaFeedback

WorksPhysics and Astronomy

Probing quasiperiodic spectral accessibility with a finite-rate quantum Otto cycle

Abstract

Quasiperiodic systems combine sharply different spectral and localisation regimes with a dense hierarchy of dynamical time scales. We show that a finite-rate quantum Otto cycle can use this structure as a thermodynamic probe, revealing which parts of the spectrum become dynamically accessible and thermodynamically relevant during a driven transformation. For a generalised Aubry-André-Harper chain, the resulting response is not determined by the static quasiperiodic regime alone: extended, near-critical, and crossing protocols do not exhibit universal ordering of work fluctuations. Instead, the statistics reflect transition pathways that are simultaneously dynamically coupled, thermally occupied, and energetically significant. The same spectral accessibility separates thermodynamic recovery from microscopic adiabatic recovery. Positive mean work can reappear while the population transfer dynamics remain strongly nonadiabatic because finite-rate population transfer is redistributed toward energetically less costly channels rather than suppressed. The full work distribution reveals a further separation, with a finite rate window in which positive-work cycles form a majority although the mean work remains negative. This behaviour persists in finite-size analysis. A thermally weighted spectral susceptibility connects these effects to a distributed hierarchy of dynamically active spectral regions rather than to a single minimum-gap bottleneck. Finite thermal contacts preserve this picture while introducing a second time scale and intercycle memory. Quantum Otto cycles can therefore provide a fluctuation-resolved thermodynamic probe of quasiperiodic spectral accessibility, complementary to direct spectroscopic and eigenstate-resolved diagnostics.

Introduction

Quasiperiodic lattices occupy an intermediate position between periodic order and random disorder. They are deterministic and aperiodic, yet can support localisation, critical states, mobility edges, and fractal spectral structure without quenched randomness. Aubry-André-Harper (AAH) models provide a paradigmatic one-dimensional setting for this physics, with generalised forms hosting extended, localised and critical regimes [1, 2, 3, 4, 5]. Their non-uniform spectra and wave functions have motivated studies of localisation transitions, multifractality, mobility edges, and transport. The same structure also renders parameter driving non-trivial, as variations in the control parameters can reorganise both the spatial character of the eigenstates and the dynamical pathways connecting them.

This distinction becomes especially important away from equilibrium. Finite-rate ramps through localisation or critical regions can generate excitations when the evolving state cannot longer follow the instantaneous spectrum, whereas sudden quenches expose complementary information through return amplitudes and Loschmidt echoes [5, 6, 7, 8, 9, 10]. These limits connect quasiperiodic dynamics to Kibble-Zurek-type excitation production during slow critical ramps and to return-probability diagnostics after quenches. More generally, they show that an equilibrium phase label alone does not determine the response to a time-dependent transformation, with the parameter path and the time available to traverse it being also part of the dynamics.

Quasiperiodic localisation is experimentally accessible in platforms that permit increasingly detailed control and readout. Matter-wave localisation has been observed in one-dimensional incommensurate optical lattices [11, 12, 13], and generalised AAH models have been realised in programmable superconducting-qubit arrays, where extended, critical, and localised dynamics can be distinguished through spatial propagation-based observables and participation-based observables [14]. Such measurements provide microscopic information on the evolution of the state. They also motivate a complementary question: which signatures of this structured dynamics remain visible in coarser observables that do not require reconstructing spatial profiles, participation entropies, or individual eigenstates?

Quantum thermodynamic cycles provide one route to this question because unitary spectral dynamics is converted directly into energy exchange. Quasiperiodic working media have been proposed for steady-state heat engines that exploit mobility edges as energy-selective structures [15] and for cyclic machines whose

Hamiltonian is varied between different quasiperiodic regimes [16, 17, 18]. The proposals of related localisation-based engines have connected the spectral structure to reliability and power [19]. These studies establish that localisation properties can reshape thermodynamic operating regimes. However, in cyclic settings, isolated work strokes have been treated mainly through limiting transformations in which the state is effectively frozen during a rapid parameter change or populations remain attached to instantaneous energy levels during asymptotically slow driving [17, 18]. The finite-rate regime between these limits contains additional information since the operating boundary can shift with stroke duration, whereas different parts of a structured spectrum contribute unequally to the resulting energy exchange.

Finite-time Otto dynamics is well developed for other working media, including explicit unitary strokes and incomplete thermalization [20, 21, 22, 23], as well as critical driving and power optimization [24, 25, 26]. Fluctuation-resolved thermodynamics adds information that is absent from mean work alone. Analytically tractable two-level Otto engines make this structure explicit in closed work distributions [22], while nuclear magnetic resonance experiments have reconstructed joint work-heat statistics for a driven spin-1/2 Otto cycle [27]. Finite-time engines can exhibit broad and asymmetric work distributions [28, 29], threshold probabilities distinguish likely from average work extraction [30], and mean-to-fluctuation ratios provide separate measures of output precision [31, 32, 33, 34]. More broadly, power, efficiency, and precision obey fluctuation tradeoffs [35, 36], while full counting statistics extend the description to long-time energy currents and cumulants [37, 38]. Finite-rate work statistics have also been studied in quantum-critical ramps [39]. Particularly relevant here, the entropy of the work distribution in an AAH system can reveal localisation information that is not apparent in its lower moments [40]. However, to our knowledge, cycle-resolved work fluctuations have not previously been analysed in a quasiperiodic quantum thermal machine. This raises a distinct question whether extended, critical, and localised regimes leave a simple ordering in thermodynamic fluctuations once finite-rate dynamics and thermal occupation act together.

Here, we treat the driving rate as a physical control parameter of a single-particle quantum Otto cycle. During each isolated work stroke, the generalised AAH Hamiltonian is continuously varied and the state is propagated through the evolving spectrum. The conventional thermodynamic mode diagram thereby becomes rate dependent, and a path that produces positive work in the slow-driving reference can become work-consuming at faster driving, and the corresponding operating boundary can move or disappear as the available stroke time changes. We refer to this rate-dependent classification as a finite-time operating diagram. The finite-rate stroke is thus used as a probe of the transformation itself, connecting microscopic population transfer to mean thermodynamic output and to the complete work distribution.

We used spectral accessibility operationally to describe the portion of the evolving spectrum that contributes appreciably to this finite-rate thermodynamic response. Accessibility is determined jointly by the dynamical coupling between instantaneous branches, their thermal occupation, and the energetic consequences of the resulting transitions. The Hamiltonian path determines which spectral structures are encountered, while the bath temperatures select which of them carry appreciable thermodynamic weight. The resulting probe is coarser than direct spectroscopy or eigenstate-resolved localisation diagnostics, but it retains information about which parts of the structured spectrum actually participate in the driven energy exchange.

Contributions and scope. Our results first show that the static quasiperiodic regime does not impose a universal ordering on finite-rate work fluctuations. Extended, near-critical, and crossing protocols exchange their fluctuation ordering between the slow-driving, finite-rate, and state-frozen regimes, while matched controls preserve the enhanced finite-rate response of the crossing path. The relevant distinction is therefore dynamical, with the statistics depending on which parts of the spectrum become accessible during the stroke, rather than on the static phase label alone. Along the crossing path, this produces two further separations. Positive mean work recovers, while the population-transfer map remains strongly nonadiabatic, with a transition-resolved decomposition showing that the transferred population shifts toward energetically less costly channels. In addition, over a finite interval of driving rates, positive-work cycles form a majority even though the mean work remains negative. This stochastic separation persists across the Fibonacci sizes L=233,377,610,987L = 233, 377, 610, 987, while large branch displacements are increasingly associated with costly negative-work trajectories. At the microscopic level, a thermally weighted spectral susceptibility identifies a distributed hierarchy of dynamically active spectral regions that tracks excess-energy production. The narrowest isolated avoided crossing contributes only a small fraction of this response, so the operating boundary reflects the collective action of many thermally active transitions rather than a single minimum-gap bottleneck.

Finite thermal contacts extend this picture beyond the complete-reset cycle. Incomplete renewal introduces a second dynamical time scale, modifies work, power, efficiency, and fluctuations, and generates correlations between successive cycles. Repeated operation must therefore distinguish single-cycle work statistics from long-time current fluctuations [37, 41]. Temperature and Hamiltonian controls provide complementary limits, with cooling narrowing the thermally active spectral sector and exposing slow critical dynamics, while sufficiently rapid strokes approach the state-frozen regime. The analysis concerns a single-particle generalised AAH chain, positive-temperature reservoirs, and two-projective-measurement work statistics. Principal rate scans and finite-size analysis use linear retraced work strokes. The finite-size results establish persistence over the range studied and are not intended as a thermodynamic-limit claim.

Model, finite-time cycle, and thermodynamic metrics

A finite-time Otto cycle contains two physically distinct clocks. During the isolated work strokes the Hamiltonian changes, and the relevant time is the duration available for the state to reorganise through the evolving spectrum. During thermal contacts, the Hamiltonian is fixed while the reservoirs redistribute populations, and the relevant time is the duration available for relaxation. To separate these clocks, we first take both contacts to the complete-reset limit and isolate the finite duration of the work strokes. Then, the finite thermalization times are restored as an independent control in Sec. VII.

Quasiperiodic working medium and control paths

The working medium is a single particle on an open quasiperiodic chain of LL sites. Both the hopping amplitudes and the onsite energies carry the same incommensurate modulation. Changing the external controls does more than shift an overall energy scale, it reshapes the instantaneous spectrum and the spatial character of its eigenstates while the particle is being driven. We use the diagonal-plus-off-diagonal Aubry-André-Harper Hamiltonian

H(r,θ)=−∑n=1L−1Jn(r,θ)(∣n⟩⟨n+1∣+H.c.)+V(r,θ)∑n=1Lcos⁡(2παn+ϕ)∣n⟩⟨n∣,(1)H(r,\theta)=-\sum_{n=1}^{L-1}J_n(r,\theta)(|n\rangle\langle n+1|+\mathrm{H.c.})+V(r,\theta)\sum_{n=1}^{L}\cos(2\pi\alpha n+\phi)|n\rangle\langle n|, \tag*{(1)}

where the bond-dependent hopping is Jn(r,θ)=J+δJ(r,θ)cos⁡(2παn+ϕ)J_n(r,\theta)=J+\delta J(r,\theta)\cos(2\pi\alpha n+\phi). We parametrize the modulation amplitudes as δJ=r/1+tan⁡2θ/4\delta J=r/\sqrt{1+\tan^2\theta/4} and V=δJtan⁡θV=\delta J\tan\theta, so that r2=δJ2+(V/2)2r^2=\delta J^2+(V/2)^2 and V/δJ=tan⁡θV/\delta J=\tan\theta. The radial coordinate rr sets the overall modulation strength, while θ\theta determines how that modulation is shared between the hop and onsite terms. The phase ϕ\phi changes the quasiperiodic pattern relative to the open boundaries and α\alpha fixes its incommensurate spatial frequency. Unless otherwise stated, we use J=ℏ=kB=1J=\hbar=k_{\mathrm{B}}=1, α=(5−1)/2\alpha=(\sqrt{5}-1)/2, and ϕ=0\phi=0. The quantity LL counts the lattice sites and is also the dimension of the Hilbert space of one-particle.

The static model contains extended, critical, and localised regimes [3, 4, 17, 42], summarised schematically in Figure 1(a) for orientation. The finite-time calculation compares three paths within this same Hamiltonian family, where one remains on the extended side,

Working medium, static structure, and finite-time Otto protocol

Figure 1. Working medium, static structure, and finite-time Otto protocol. (a) Schematic summary of the static phase structure of the diagonal-plus-off-diagonal quasiperiodic model following Refs. [4, 17]. The blue, pale-yellow, and red shaded regions denote the extended, critical, and localised regimes, respectively. The arrows at θ=π/4\theta= \pi/4 mark the representative paths 0.4→0.50.4 \to0.5, 0.294→0.90.294 \to0.9, and 0.3593→1.10.3593 \to1.1. (b) The single-particle quasiperiodic chain is embedded in the four-corner Otto cycle. The central cartoon distinguishes onsite modulation from bond-dependent hopping. The work strokes A→BA \to B and C→DC \to D change rr, while B→CB \to C and D→AD \to A are the cold and hot thermal contacts. (c) Control profile r(t)r(t) over one cycle. The work strokes have durations τF\tau_F and τR\tau_R, while the cold and hot contacts last τc\tau_c and τh\tau_h, respectively, in the finite-contact extension. Times are measured in units ℏ/J\hbar/J.

Finite-time Otto cycle and thermodynamic metrics

ModeWoutW_{\mathrm{out}}QhQ_hQcQ_c
Engine++++−-
Refrigerator−-−-++
Heater−-−-−-
Accelerator−-++−-

Table 1. Thermodynamic operating modes. A plus sign denotes energy entering the indicated system for heat, and positive WoutW_{\mathrm{out}} denotes work delivered by the working medium.

another approaches the static boundary, and the third crosses it. These paths were chosen to probe progressively more structured spectral evolution. The actual hierarchy of narrow avoided crossings encountered by the crossing path is established in Sec. III and Appendix B, rather than assumed from the endpoint labels. Their precise endpoints are shown in Figure 1(a). Unless otherwise stated, we fix θ=π/4\theta=\pi/4 and Tc/J=0.1T_c/J=0.1. The bath temperature and the rate of work-stroke are treated as independent control parameters. Because these representative protocols have different excursion lengths, comparisons intended to isolate the role of the static spectral regime are performed separately using equal-excursion controls.

Figure 1 embeds the working medium in the four-stroke Otto protocol. During A→BA\to B and C→DC\to D the particle is isolated from the baths while rr is changed, while the hot and cold contacts repopulate the endpoint spectra at a fixed Hamiltonian. We first take the contacts to the complete-reset limit, so each work stroke starts from the appropriate Gibbs state. This removes incomplete thermalization and inter-stroke memory from the first part of the analysis and isolates the effect of work-stroke duration. If EnbE_n^b denotes the energy-ordered levels at the end point b∈{h,c}b\in\{h,c\} and βb=(kBTb)−1\beta_b=(k_{\mathrm{B}}T_b)^{-1}, the Gibbs populations are pbn=e−βbEnb/Zbp_{bn}=e^{-\beta_b E_n^b}/Z_b, with Zb=∑ne−βbEnbZ_b=\sum_n e^{-\beta_b E_n^b}. Corner AA is the hot Gibbs state of H(rh)H(r_h) and corner CC the cold Gibbs state of H(rc)H(r_c), while corners BB and DD are generated by unitary work strokes.

We take Qb>0Q_b>0 when heat enters the working medium and Wout>0W_{\mathrm{out}}>0 when the particle performs net work. If UXU_X is the mean energy in corner XX,

Qh=UA−UD,Qc=UC−UB,Q_h=U_A-U_D,\qquad Q_c=U_C-U_B,
Wout=(UA−UB)+(UC−UD)=Qh+Qc.(2)W_{\mathrm{out}}=(U_A-U_B)+(U_C-U_D)=Q_h+Q_c. \tag*{(2)}

The signs of these mean energy flows distinguish the conventional operating modes summarised in Table I. Within the engine sector, the mean efficiency is η=Wout/Qh\eta=W_{\mathrm{out}}/Q_h. If only the endpoints were specified, this classification would say nothing about the speed with which the spectrum traversed [16, 17, 18]. Here, BB and DD are generated by explicit time evolution, so shortening a work stroke can change not only the amount of work, but also whether the same endpoints operate as an engine at all. We use the term finite-time operating diagram for this rate-dependent classification.

Work-stroke dynamics: fast and slow limits

We measure energies in units of JJ and times in units of ℏ/J\hbar/J. Thus, a dimensionless stroke duration τ\tau corresponds to a physical duration τphys=(ℏ/J)τ\tau_{\mathrm{phys}} = (\hbar/J)\tau. Throughout the numerical calculations, we set J=ℏ=1J = \hbar= 1. For linear ramps,

rF(t)=rh+rc−rhτFt,0≤t≤τF,rR(t)=rc+rh−rcτRt,0≤t≤τR.(3)\begin{aligned} r_F(t) &= r_h + \frac{r_c-r_h}{\tau_F}t,\qquad0 \le t \le\tau_F,\\ r_R(t) &= r_c + \frac{r_h-r_c}{\tau_R}t,\qquad0 \le t \le\tau_R. \tag*{(3)} \end{aligned}

For the symmetric protocol, τF=τR=∣rc−rh∣/ν\tau_F = \tau_R = |r_c-r_h|/\nu, with ν=∣dr/dt∣\nu= |dr/dt|. In these units, the driving rate is expressed as ν/J\nu/J. The larger values correspond to shorter work strokes. Since the three paths have different excursion lengths, equal values of ν/J\nu/J do not correspond to equal stroke durations. Equivalently, the plotted rate ν/J\nu/J is the dimensionless derivative ∣d(rphys/J)/d(Jtphys/ℏ)∣|d(r_{\mathrm{phys}}/J)/d(Jt_{\mathrm{phys}}/\hbar)|. Platform-specific coherence and thermalization times enter after choosing the energy scale and reservoir rates of a concrete realisation.

Thermodynamic isolation and energy-ordered slow driving are distinct notions [20, 21]. In a very fast stroke the Hamiltonian changes before the wave function can reorganise appreciably in the site basis, defining the state-frozen reference. At the opposite extreme, an infinitely slow traversal follows the energy-ordered instantaneous eigenstates through the avoided-crossing structure and defines the strict slow-driving, energy-ordered reference. The quasiperiodic problem makes the distinction consequential because useful thermodynamic output can be recovered before this microscopic slow-driving limit is reached.

Let UFU_F be the forward propagator and ∣nh⟩|n_h\rangle and ∣mc⟩|m_c\rangle the energy eigenstates of the endpoint. The forward population map is PF(m∣n)=∣⟨mc∣UF∣nh⟩∣2P_F(m|n) = |\langle m_c|U_F|n_h\rangle|^2. For the reverse stroke, PR(ℓ∣k)=∣⟨ℓh∣UR∣kc⟩∣2P_R(\ell|k) = |\langle\ell_h|U_R|k_c\rangle|^2. Because the Hamiltonian is real and the reverse stroke retraces the same path at the same rate, PR=PFTP_R = P_F^{\mathsf{T}}. In energy-ordered labels, Pad(m∣n)=δmnP_{\mathrm{ad}}(m|n) = \delta_{mn} is the strict slow-driving map, whereas Psud(m∣n)=∣⟨mc∣nh⟩∣2P_{\mathrm{sud}}(m|n) = |\langle m_c|n_h\rangle|^2 is the state-frozen map. The open tridiagonal Hamiltonians have nondegenerate spectra over the paths considered, so instantaneous branches are ordered by eigenvalue. Near avoided crossings, their exchange of microscopic character is diagnosed by overlaps rather than by relabelling the energy order.

To locate the actual dynamics between these two limits, we use

Dad=∥PF−I∥FL,Dsud=∥PF−Psud∥FL.(4)D_{\mathrm{ad}} = \frac{\lVert P_F-I\rVert_F}{\sqrt{L}},\qquad D_{\mathrm{sud}} = \frac{\lVert P_F-P_{\mathrm{sud}}\rVert_F}{\sqrt{L}}. \tag*{(4)}

Thus, Dad=0D_{\rm ad}=0 means energy-ordered population following, while Dsud=0D_{\rm sud}=0 means state-frozen population dynamics. These are distances between population maps, not measures of phase coherence.

The Gibbs weights are branch dependent. To identify departures from energy-ordered following on the branches carrying thermal weight, we also use

ℓF=∑nphn[1−PF(n∣n)],\ell_{\rm F}=\sum_n p_{hn}[1-P_{\rm F}(n|n)],
ℓR=∑mpcm[1−PR(m∣m)],(5)\ell_{\rm R}=\sum_m p_{cm}[1-P_{\rm R}(m|m)], \tag*{(5)}

with ℓˉ=(ℓF+ℓR)/2\bar{\ell}=(\ell_{\rm F}+\ell_{\rm R})/2. The map distances characterise the complete transition matrix, whereas ℓˉ\bar{\ell} emphasises the thermally occupied sector. Together with the thermodynamic observables, these quantities separate operational recovery from the microscopic population as follows.

The work-stroke propagator is evaluated with unitary midpoint Cayley-Crank-Nicolson evolution. The convergence with respect to the time step and the independent adaptive propagation Schrödinger are documented in the Appendix A.

Single-cycle work and heat statistics

Even with fixed controls and bath temperatures, individual cycles exchange different amounts of energy because the thermal contacts sample different energy branches, and the finite-rate work strokes connect those branches probabilistically. Mean work and heat describe only the centre of a distribution of cycle outcomes. We resolve these outcomes with a TPM construction on each work stroke, combined into a four-corner cycle trajectory [43, 44]. The resulting distribution is an operational measurement-resolved quantity. Projective energy readout dephases the state, so monitored and unmonitored coherent engines can carry different fluctuation statistics [45]. For the complete-reset cycle used below, the stroke-start states are Gibbs diagonal endpoint energy readout preserves the mean energy, and the following reset removes residual coherence. The quoted complete-reset mean work and heat are invariant under retaining or discarding the TPM record. With coherence-retaining finite contacts, the monitoring prescription instead becomes part of the dynamical specification.

Upon starting a complete-reset trajectory at the corner CC, a reverse-stroke transition ∣kc⟩→∣ℓh⟩|k_c\rangle\to|\ell_h\rangle is followed by the hot reset ∣ℓh⟩→∣nh⟩|\ell_h\rangle\to|n_h\rangle and then the forward-stroke transition ∣nh⟩→∣mc⟩|n_h\rangle\to|m_c\rangle. The four-index record γ=(k,ℓ,n,m)\gamma=(k,\ell,n,m) is sufficient for work and hot heat. To identify the physical heat exchanged during the subsequent cold contact of the same realisation, append the reset outcome ∣mc⟩→∣kc′⟩|m_c\rangle\to|k'_c\rangle and write γ~=(k,ℓ,n,m,k′)\tilde{\gamma}=(k,\ell,n,m,k').

Then

Wγ=Ekc−Eℓh+Enh−Emc,Qh,γ=Enh−Eℓh,Qc,γ~phys=Ek′c−Emc,ΔEγ~=Ek′c−Ekc.(6)\begin{aligned} W_{\gamma}&=E_k^c-E_\ell^h+E_n^h-E_m^c,\\ Q_{h,\gamma}&=E_n^h-E_\ell^h,\\ Q_{c,\tilde{\gamma}}^{\mathrm{phys}}&=E_{k'}^c-E_m^c,\\ \Delta E_{\tilde{\gamma}}&=E_{k'}^c-E_k^c. \tag*{(6)} \end{aligned}

so the trajectory first law under our output-work convention is

Wγ=Qh,γ+Qc,γ~phys−ΔEγ~.(7)W_{\gamma}=Q_{h,\gamma}+Q_{c,\tilde{\gamma}}^{\mathrm{phys}}-\Delta E_{\tilde{\gamma}}. \tag*{(7)}

For a complete reset, kk and k′k' are independent draws from the same cold Gibbs distribution. Hence ⟨ΔE⟩=0\langle\Delta E\rangle=0. The mean balance therefore reduces to Wout=Qh+QcW_{\rm out}=Q_h+Q_c, while the work distribution and the joint work–hot-heat statistics require only the marginal four-index γ\gamma.

Complete resets make the hot and cold preparations statistically independent, so Πγ=pckPR(ℓ∣k)phnPF(m∣n)\Pi_\gamma=p_{ck}P_{\rm R}(\ell|k)p_{hn}P_{\rm F}(m|n). The resulting joint distribution is

Pcyc(W,Qh)=∑k,ℓ,n,mΠγ δ(W−Wγ)×δ(Qh−Qh,γ),(8)P_{\rm cyc}(W,Q_h)=\sum_{k,\ell,n,m}\Pi_\gamma\,\delta(W-W_\gamma)\times\delta(Q_h-Q_{h,\gamma}), \tag*{(8)}

and the work distribution is its marginal. Its first moments reproduce the thermodynamic averages ⟨W⟩=Wout\langle W\rangle=W_{\rm out} and ⟨Qh⟩=Qh\langle Q_h\rangle=Q_h, so the mean efficiency of the engine remains η=⟨W⟩/⟨Qh⟩\eta=\langle W\rangle/\langle Q_h\rangle. We retain the joint distribution when work-heat correlations are needed, but do not use the trajectory ratio Wγ/Qh,γW_\gamma/Q_{h,\gamma} as a primary observable because trajectories with Qh,γ=0Q_{h,\gamma}=0 make that ratio singular [28, 29].

For the finite-time engine, the most transparent work statistics are

P+=Pr⁡(W>0),σW=⟨(W−⟨W⟩)2⟩,P_+=\Pr(W>0),\qquad\sigma_W=\sqrt{\langle(W-\langle W\rangle)^2\rangle},
R=⟨W⟩σW.(9)\mathcal{R}=\frac{\langle W\rangle}{\sigma_W}. \tag*{(9)}

They answer different operational questions: P+P_+ is the probability that one cycle delivers positive work [30], σW\sigma_W sets the fluctuation scale, and R\mathcal{R} compares the mean output with that width. For ⟨W⟩>0\langle W\rangle>0, we refer to R\mathcal{R} as the reliability of the engine [31, 32, 33, 34]. Otherwise, it is retained only as a signed fluctuation diagnostic. These observables place mean thermodynamics, microscopic population transfer, and trajectory-level fluctuations on the same finite-rate axis.

Finite-rate accessibility and its spectral origin

How long must the isolated work stroke be before the engine operates again, and is that duration also sufficient for microscopic energy-ordered following? The crossing path makes these questions distinct because its thermally occupied states traverse a structured set of avoided crossings. Figure 2 compares the thermodynamic and microscopic responses on the same driving-rate axis.

Figure 2. Finite-rate response of the representative paths. (a) Mean output work versus driving rate. The extended, near-critical, and crossing curves use their representative hot-bath settings Th/J=1.0T_h/J = 1.0, 0.80.8, and 0.650.65, respectively. (b) Width ΔTh/J\Delta T_h/J of the hot-bath temperature interval supporting engine operation. Faint dashed lines mark the strict energy-ordered widths. The open marker identifies the numerically resolved onset of window closure, with no engine window reappearing at higher evaluated rates. (c) Crossing-path distances from the strict energy-ordered and state-frozen population maps, together with the thermally weighted leakage ℓˉ\bar{\ell}, shown on a logarithmic scale. The shaded band marks the displayed 0.03≤ν/J≤0.040.03 \leq\nu/J \leq0.04 interval bracketing recovery of positive mean work on the crossing path.

At an avoided crossing, an energy-ordered eigenstate can remain continuous, while its wave-function character is exchanged with a neighbouring branch. In the strict energy-ordered slow-driving limit, the state follows the evolving eigenstate through that exchange; a faster traversal can preserve more of the incoming microscopic character. For a single isolated crossing, this competition is captured by Landau–Zener physics [46]. However, along the quasiperiodic ramp, thermally occupied states encounter many crossings with widely different gaps and occupations.

The finite-temperature engine response samples a weighted collection of branches, each undergoing its own sequence of transfers. Appendix B resolves representative branch exchange, lower-tail spacing statistics, and a thermally weighted spectral susceptibility. Away from exact degeneracies, this susceptibility is proportional to a Gibbs-population-weighted quantum metric. At ν/J=0.04\nu/J = 0.04, the 20% of forward-stroke intervals with the highest susceptibility carry a positive excess-energy growth 62.8%. An exact additive decomposition of both work strokes gives a broader cycle-level response: the top 20% of the path ranked by the combined forward–reverse susceptibility carries 31.6% of the total work deficit, while the window ∣r−r∗∣/J≤0.005|r-r_*|/J \leq0.005 around the narrowest isolated avoided crossing contributes only 2.01%. The finite-rate response is therefore distributed across thermally occupied, dynamically coupled spectral regions spanning a broad hierarchy of gaps.

The susceptibility also guides a fixed-duration control. Keeping the total stroke time at τ=18.5167\tau= 18.5167 and reallocating 25% more time to the highest-susceptibility 20% of the path, compensated for by 25% less time in the lowest-susceptibility 20%, changes the crossing-path work from ⟨W⟩/J=−6.98×10−4\langle W\rangle/J = -6.98 \times10^{-4} for the uniform ramp to +3.12×10−3+3.12 \times10^{-3}. Reversing the allocation gives −4.79×10−3-4.79 \times10^{-3}. In all three cases Dad≃0.983D_{\mathrm{ad}} \simeq0.983, so the population map remains far from the energy-ordered following. The susceptibility therefore identifies where an increment of stroke time has the largest thermodynamic effect, while the detailed transition probabilities retain their dependence on dynamical phases and interference.

Figure 2(a) shows that shortening the work stroke has little effect on the extended-side path over most of the displayed range, while the near-critical and crossing paths lose output much earlier. The three curves use different representative hot-bath temperatures, so they compare finite-rate sensitivity along the chosen paths. In common Th/J=0.65T_h/J = 0.65, Tc/J=0.1T_c/J = 0.1 and stroke duration τ=18.5167\tau= 18.5167, the extended and near-critical paths give ⟨W⟩/J≃0.0130\langle W\rangle/J \simeq0.0130 and 0.04650.0465, while the crossing path is slightly work-consuming, ⟨W⟩/J≃−6.99×10−4\langle W\rangle/J \simeq-6.99 \times10^{-4}, and has the largest work width. The comparison of equal-excursions in Appendix B gives the same qualitative distinction. These controls identify a path-dependent enhancement of the finite-rate response.

The width of the engine-window in Figure 2(b) removes that single-temperature restriction. The extended path retains a broad hot-temperature interval, the near-critical window contracts, and the crossing-path window is the first to collapse. For the crossing path, increasing the duration of the stroke from τ≃18.5\tau\simeq18.5 in ν/J=0.04\nu/J = 0.04 to τ≃24.7\tau\simeq24.7 in 0.030.03 restores a finite engine window. These durations follow from ∣rc−rh∣/J≃0.7407|r_c-r_h|/J \simeq0.7407 and are quoted in the dimensionless time convention of Sec. II; the denser recovery brackets are listed in Appendix B, Table B1.

The operating-mode map is dominated by an engine-to-non-engine crossover. For the crossing protocol, the operation of the refrigerator is limited to a narrow low-temperature interval in the strict slow-driving reference, approximately 0.15≲Th/J≲0.1850.15 \lesssim T_h/J \lesssim0.185, and is absent in every finite driver evaluated. At ν/J=0.03\nu/J = 0.03 the finite-rate diagram still contains an engine sector, while the remaining points are accelerator or heater operation; from ν/J=0.04\nu/J = 0.04 onwards the evaluated crossing-path diagram contains only the latter non-engine modes. The finite-time boundary studied here is primarily an engine-to-non-engine crossover; refrigeration remains confined to the narrow low-temperature slow-driving sector.

The microscopic response evolves on a much slower scale. Figure 2(c) shows that DsudD_{\mathrm{sud}} changes by orders of magnitude as the stroke is lengthened, so the dynamics is progressively leaving the state-frozen limit, while DadD_{\mathrm{ad}} remains of order unity throughout the displayed range. At ν/J=0.03\nu/J = 0.03, where the engine operation has returned, Dsud≃0.079D_{\mathrm{sud}} \simeq0.079 and Dad≃0.982D_{\mathrm{ad}} \simeq0.982. The Gibbs-weighted leakage is simultaneously ℓˉ≃0.966\bar{\ell} \simeq0.966, confirming that the thermally occupied branches themselves remain strongly displaced from the energy-ordered following.

The distances from the population-map and leakage quantify redistribution, whereas the work deficit also weights the energies carried by those transfers. Resolving the transition Wad−WνW_{\mathrm{ad}} - W_{\nu} by transition shows that, when the crossing protocol is slowed from ν/J=0.04\nu/J = 0.04 to 0.030.03, the work deficit decreases by approximately 24%, while ℓˉ\bar{\ell} changes by less than 0.2% and DadD_{\mathrm{ad}} remains essentially unchanged. Most of the reduction comes from the reverse stroke and from changes among large branch-displacement transitions. Positive mean work recovers because the still strongly off-diagonal transferred population shifts toward channels with smaller work penalties. The exact decomposition and the rate-resolved values are given in the Appendix A 1.

Engine operation is restored at work-stroke durations of order tens of ℏ/J\hbar/J, while the complete transition map and the Gibbs-weighted branch dynamics remain far from energy-ordered following. The work distribution provides an additional level of information, distinguishing positive mean output from typical-cycle behavior and fluctuation-based reliability.

Work fluctuations and distinct finite-rate criteria

From mean operation to cycle-by-cycle reliability

Positive mean work and typical-cycle output recover at distinct rates. The contours in the two heat maps of Figure 3 are separated. In Figure 3(b), the contour P(W>0)=1/2P(W > 0) = 1/2 extends beyond the contour ⟨W⟩=0\langle W\rangle= 0, so a majority of cycles can already produce positive work while the mean is still negative. Figure 3(a) adds a second separation: throughout the displayed positive-mean region, the signed mean-to-width ratio remains below about 0.05. Majority-positive behavior appears before positive mean work, while the recovered engine region remains strongly fluctuation dominated.

Figure 3. Stochastic operation and branch redistribution for the crossing path. (a) Signed mean-to-width ratio ⟨W⟩/σW\langle W\rangle/\sigma_W over the (Th/J,ν/J)(T_h/J,\nu/J) plane. (b) Positive-work probability P(W>0)P(W>0) over the same parameter range. In panels (a,b), the solid contour marks the mean-work boundary ⟨W⟩=0\langle W\rangle=0, while the dashed contour marks the majority-positive boundary P(W>0)=1/2P(W>0)=1/2; contours interpolate between the computed driving rates as visual guides. (c) TPM cycle-work distribution at Th/J=0.65T_h/J=0.65 and ν/J=0.04\nu/J=0.04, with negative- and positive-work sectors shaded separately. The dotted line marks W=0W=0 and the dashed line the mean work. (d) Distribution of the total branch displacement KK, conditioned on the sign of the cycle work, at the same operating point.

The difference between the first two criteria follows from

⟨W⟩=P+⟨W∣W>0⟩+(1−P+)⟨W∣W≤0⟩.(10)\langle W\rangle= P_{+}\langle W \mid W > 0\rangle+ (1-P_{+})\langle W \mid W \leq0\rangle. \tag*{(10)}

A small majority of positive-work cycles can coexist with a negative mean when the negative-work histories carry larger losses. The representative point in Figure 3(c) makes this imbalance visible and locates it within the finite-rate quasiperiodic dynamics. At Th/J=0.65T_h/J = 0.65 and ν/J=0.04\nu/J = 0.04, just over half of the cycles have W>0W > 0, yet the mean remains slightly negative. Slowing the same path to ν/J=0.03\nu/J = 0.03 restores a positive mean work, with engine reliability R≃0.049R \simeq0.049; therefore, fluctuations remain large compared to the mean. The unbinned TPM construction used for the quoted moments, sign probabilities, and conditional means is given in Appendix C.

The fluctuation scale depends on the driving protocol as well as the static spectral regime. Equal-excursion controls at common bath temperatures separate this question from the different path lengths of the representative protocols. For Δr/J=0.1\Delta r/J = 0.1, the localised control has the largest σW\sigma_W in the strict slow-driving reference, whereas the critical control is the widest in the state-frozen reference and at the representative finite rate ν/J=0.04\nu/J = 0.04. A sliding equal-excursion scan places the strongest finite-rate response near the critical-localised crossover. The work-width ordering is protocol dependent: it changes between the slow-driving, finite-rate, and state-frozen protocols. The numerical controls are summarised in the Appendix Table B4.

Branch-redistribution structure of the negative-work tail

To resolve the transition histories behind the negative-work tail, we define for a complete TPM cycle with branch changes n→mn \to m and k→ℓk \to\ell in the two work strokes the total displacement in the energy-ordered branch index as K=∣m−n∣+∣ℓ−k∣K = |m-n| + |\ell-k|. This is a spectral-branch displacement, not a spatial localisation measure. Large KK means that the two work strokes carry the trajectory far from the branches it would follow in the energy-ordered limit.

Figure 3(d) shows that the negative-work sector has a significantly heavier large tail KK. At the representative point, 13.28% of negative-work trajectories have K≥50K \geq50, compared to 4.24% of positive-work trajectories, and this same tail carries 27.98% of the total negative-work loss. The energetic severity also increases with displacement: the mean loss is about 0.11J0.11J for K<10K < 10, 0.31J0.31J for 50≤K<6050 \leq K < 60, 0.43J0.43J for 60≤K<8060 \leq K < 80, and 0.83J0.83J for K≥100K\ge100. The large enrichment KK persists throughout the threshold range 20≤K0≤10020\le K_0\le100, while the conditional energetic loss grows with displacement. Appendix Table C1 gives the threshold-resolved values. The same redistribution signature also appears on smoother paths, but the crossing brings it into relevance at much slower driving. At ν/J=0.04\nu/J=0.04, the mean negative-work displacement is almost branch preserving on the extended path, ⟨K∣W<0⟩≃0.2\langle K\mid W<0\rangle\simeq0.2, but is already about 29 on the crossing path. The near-critical control reaches a comparable redistribution only near ν/J=10\nu/J=10. For the crossing path, the negative tail of the large-displacement becomes prominent at a driving rate roughly 250 times slower than for the near-critical control. The negative-work sector is concentrated in the thermally occupied low-energy manifold. At ν/J=0.04\nu/J=0.04, the lowest 32 hot-side branches contain 50.8% of the Gibbs population, but generate 71.1% of the negative cycles and 61.4% of the total loss of work. negative. The effect involves many levels, but is strongly weighted toward the occupied low-energy manifold. Together with the displacement-resolved loss data, this links strong branch redistribution in the thermally populated sector directly to the negative-work tail.

Finite-size persistence and quasiperiodic arithmetic

A finite quasiperiodic chain carries two kinds of size dependence at once. Increasing LL enlarges the one-particle spectrum and resolves progressively finer pieces of the incommensurate spatial structure. We follow the Fibonacci sequence L=233,377,610,987L=233,377,610,987 at fixed inverse-golden modulation and crossing protocol to determine how the fluctuation-dominated, strongly nonadiabatic regime evolves as that structure is resolved more finely. This finite-size question is distinct from previously studied reliability scaling in many-body-localised and finite-size Otto engines [19, 47].

Persistence with increasing system size

Table II compares the complete work statistics available at the two endpoint sizes with transition diagnostics at all four Fibonacci sizes. The endpoint stochastic comparison changes very little between L=233L = 233 and 987987: the work-distribution width is close to 0.244J0.244J, the probability of positive work stays near 0.530.53, and the signed mean-to-width ratio is about 0.050.05. The broad, fluctuation-dominated cycle statistics established in § 1.3 are essentially unchanged between the two endpoint sizes.

233377610987
⟨W⟩/J\langle W\rangle/J0.0118––0.0132
σW/J\sigma_W/J0.2428––0.2438
R\mathcal{R}0.049––0.054
P(W>0)P(W > 0)0.5260––0.5279
skewness-0.554––-0.546
excess kurtosis3.275––3.263
DadD_{\mathrm{ad}}0.98250.98760.99200.9952
DsudD_{\mathrm{sud}}0.07870.06190.04880.0384
ℓˉ\bar{\ell}0.96590.97620.98540.9910

Table II. Finite-size persistence at ν/J=0.03\nu/J = 0.03. Work statistics are shown at the endpoint sizes L=233L = 233 and 987987, while transition diagnostics span the full Fibonacci sequence considered.

The microscopic response is equally persistent. Across the four sizes, the distance from the state-frozen map decreases while the distance from the energy-ordered map and the thermally weighted leakage remain close to unity. The larger systems retain strongly nonadiabatic population transfer on the reservoir-populated branches at fixed rate, with increasing distance from energy-ordered following and decreasing distance from the state-frozen map.

Boundary shifts and the role of quasiperiodic arithmetic

The precise boundaries shift with size, but the interval between them persists. At the common rate ν/J=0.041\nu/J = 0.041, all four sizes satisfy ⟨W⟩<0\langle W\rangle< 0 and P(W>0)>1/2P(W > 0) > 1/2: the mean work changes from −1.96×10−3J-1.96 \times10^{-3}J at L=233L = 233 to −4.91×10−4J-4.91 \times10^{-4}J at L=987L = 987, and P+P_+ changes from 0.50430.5043 to 0.50640.5064. Bracketing the two zero crossings separately shows that the estimated width of the sign-separated interval remains approximately 4.24.2–4.4×10−34.4 \times10^{-3} in ν/J\nu/J across L=233,377,610,987L = 233, 377, 610, 987, even though both boundary locations move. The majority-positive and positive-mean criteria retain the same ordering and a finite separation across the four sizes despite the shifts in boundary location. Appendix Table B2 gives the four-size brackets and the large-LL convergence result; Figure B1(b) shows that the microscopic nonadiabaticity persists across the same sequence.

Figure B1. Finite-size robustness of the spectral and transition structure. (a) Low-spacing tail within the lowest energy-ordered sector carrying 99% of the cold-endpoint Gibbs probability. Adjacent spacings are collected from 101 uniformly spaced Hamiltonians along each work-stroke path, and the 0.1% quantile is plotted. The crossing path remains separated from both controls through L=987L = 987. (b) Microscopic transition diagnostics at ν/J=0.04\nu/J = 0.04 for L=233,377,610,987L = 233, 377, 610, 987. The finite-rate population map remains far from energy-ordered following while staying close to the state-frozen population map, with near-unit Gibbs-weighted leakage. Together with the ν/J=0.03\nu/J = 0.03 values in Table II, these data show persistent microscopic nonadiabaticity across the thermodynamic recovery region.

For the inverse-golden modulation, the Fibonacci size sequence also has an arithmetic meaning. The chain length is increased while the irrational frequency α=(5−1)/2\alpha= (\sqrt{5} - 1)/2 itself is held fixed, and the chosen LL values follow the natural denominators associated with its continued-fraction structure. Independent rational-approximant calculations show that the TPM observables approach the inverse-golden result smoothly even as the smallest instantaneous spacing varies nonmonotonically.

Changing α\alpha is a different physical operation from increasing LL. The silver-ratio control preserves the qualitative separation between majority-positive and positive-mean operations, whereas other irrational choices can modify the endpoint thermodynamics of the same nominal ramp. The inverse-golden Fibonacci sequence establishes finite-size persistence of the stochastic regime while retaining sensitivity to quasiperiodic frequency, Hamiltonian path, and thermal occupation as separate controls.

Control parameters, temperature, and critical-dynamics limits

The finite-rate mechanism relies on two distinct forms of spectral selection. Changing the quasiperiodic Hamiltonian changes the spectral structures traversed during the work strokes, whereas changing the bath temperatures changes the structures that carry appreciable thermodynamic weight. Figure 4 separates these two effects and connects them to the low-temperature critical-ramp and sudden-quench limits.

Figure 4. Hamiltonian-path control and the thermal crossover. (a) Mean work at Th/J=0.65T_{\mathrm{h}}/J = 0.65, Tc/J=0.1T_{\mathrm{c}}/J = 0.1, and ν/J=0.04\nu/J = 0.04 as the modulation angle θ\theta changes the balance between diagonal and off-diagonal quasiperiodic modulation. The vertical dotted line marks the baseline path θ=π/4\theta= \pi/4. (b) Signed mean-to-width ratio R=⟨W⟩/σW\mathcal{R} = \langle W\rangle/\sigma_W and positive-work probability P+=P(W>0)P_{+} = P(W > 0) for the same scan. The horizontal reference lines mark R=0\mathcal{R} = 0 and P+=1/2P_{+} = 1/2. (c) Effective finite-window work-deficit slope peffp_{\mathrm{eff}} and hot-state participation Neff(h)=1/∑npnhN_{\mathrm{eff}}^{(h)} = 1/\sum_n p_n^h when the two bath temperatures are lowered together at fixed ratio Th=6.5TcT_{\mathrm{h}} = 6.5T_{\mathrm{c}}. The colored markers identify the temperatures used in panel (d). (d) Raw work deficit ΔW=Wad−Wν\Delta W = W_{\mathrm{ad}} - W_{\nu} versus rate at three representative temperatures, labeled directly by Tc/JT_{\mathrm{c}}/J. The curves show the slope change summarised in panel (c).

Quasiperiodic and Hamiltonian controls

The finite-rate stochastic separation persists across the phase offsets examined. Generalised quasiperiodic models already show that transport and localisation diagnostics can vary strongly with modulation parameters [42]. Varying the quasiperiodic phase ϕ\phi over the selected values produces only weak changes in the crossing-path response, and a silver-ratio control retains the qualitative separation between majority-positive and positive-mean operation. Changing the irrational frequency more broadly can alter the endpoint thermodynamics of the same nominal ν\nu ramp and can even remove the strict energy-ordered engine. The response remains sensitive to the quasiperiodic frequency while varying only weakly across the phase realisations examined; the full scans are collected in Appendix § 1.17.

The modulation angle θ\theta is a more direct Hamiltonian control because it changes how the same radial amplitude is divided between diagonal and off-diagonal quasiperiodic modulation. Figure 4(a) shows that this change alone can move the finite-rate crossing cycle into and out of positive mean work at Th/J=0.65T_h/J = 0.65, Tc/J=0.1T_c/J = 0.1, and ν/J=0.04\nu/J = 0.04. The operating interval is bounded on both sides: paths with too little or too much diagonal modulation become work-consuming even though the radial endpoints are unchanged.

The stochastic criteria sharpen this path dependence. In Figure 4(b), the signed mean-to-width ratio R\mathcal{R} follows the same broad optimum as the mean work, whereas P+P_{+} continues to improve to a slightly different path before turning over. The displacement is small but systematic. The path that maximizes average output is slightly displaced from the path that maximizes the probability of positive work. At the same time, R\mathcal{R} stays small throughout the scan, so the entire comparison is fluctuation dominated.

Temperature and the crossover in active spectral weight

Temperature changes the thermodynamically active part of the spectrum without changing the Hamiltonian path. At the baseline temperatures Tc/J=0.1T_{\mathrm{c}}/J = 0.1 and Th/J=0.65T_{\mathrm{h}}/J = 0.65, the hot Gibbs state occupies a broad manifold of branches, consistent with the many-level negative-work mechanism identified in § 1.3. Lowering both reservoirs together at fixed ratio Th=6.5TcT_{\mathrm{h}} = 6.5T_{\mathrm{c}} suppresses the high-energy contribution and progressively restricts the cycle to the low-energy sector. Figure 4(c) quantifies this contraction with the hot-state participation Neff(h)N_{\mathrm{eff}}^{(h)}, which falls from more than eighty effective branches at Tc/J=0.1T_{\mathrm{c}}/J = 0.1 to fewer than twenty near Tc/J=0.006T_{\mathrm{c}}/J = 0.006.

The finite-rate scaling changes at the same time. Over the fixed rate window used in Figure 4(d), the effective work-deficit slope evolves from peff≃0.88p_{\mathrm{eff}} \simeq0.88 at Tc/J=0.1T_{\mathrm{c}}/J = 0.1 toward peff≃0.48p_{\mathrm{eff}} \simeq0.48 at Tc/J=0.006T_{\mathrm{c}}/J = 0.006. The raw curves in panel (d) confirm that the change occurs in the slope of the work-deficit curve rather than only in isolated fit values. The concurrent decrease of Neff(h)N_{\mathrm{eff}}^{(h)} and peffp_{\mathrm{eff}} identifies a thermal crossover in which the thermally active spectral weight and the finite-rate response evolve together as the occupied energy window contracts. The participation ratio quantifies that contraction, while the effective exponent also depends on the detailed transition structure within the active window.

In the still lower-temperature limit, the thermodynamic interpretation becomes especially simple. The strict energy-ordered cycle returns the occupied ground branch to itself, whereas a finite-rate stroke creates excitation energies δEF\delta E_{\mathrm{F}} and δER\delta E_{\mathrm{R}} on the forward and reverse ramps. The complete-reset cycle then obeys

Wad−Wν=δEF+δER.(11)W_{\mathrm{ad}} - W_{\nu} = \delta E_{\mathrm{F}} + \delta E_{\mathrm{R}}. \tag*{(11)}

In this limit the work deficit equals the excitation energy produced by the two ramps, providing a direct bridge from a thermodynamic observable to critical-ramp excitation physics.

Critical-ramp and sudden-quench limits

The Kibble–Zurek mechanism provides a natural limiting language for slow driving across a continuous critical region: adiabatic following fails near the part of the ramp with the longest intrinsic response time, producing excitations with rate-dependent scaling [5, 6, 7, 8, 24]. For the quasiperiodic crossing path considered here, this picture is complicated by a hierarchy of narrow avoided crossings rather than a single isolated scale. A single-scale asymptotic interpretation would require an accessible slow-ramp window that is insensitive to this hierarchy; the present finite interval instead supports effective rate-dependent slopes.

Extending the joint cooling scan into the low-temperature regime yields a reproducible finite-window law. At L=233L = 233 and Tc/J=0.006T_{\mathrm{c}}/J = 0.006, a fit over the full evaluated interval 0.006≤ν/J≤0.0300.006 \leq\nu/J \leq0.030 gives peff=0.480p_{\mathrm{eff}} = 0.480. Nested four-point windows vary from peff=0.407p_{\mathrm{eff}} = 0.407 on 0.006≤ν/J≤0.0140.006 \leq\nu/J \leq0.014 to 0.5290.529 on 0.010≤ν/J≤0.0300.010 \leq\nu/J \leq0.030, with intermediate windows lying between these values. Equation (11) confirms that the same rate dependence appears in the work deficit and in the total excitation energy. The systematic window drift, together with the size- and approximant-dependence of the quasiperiodic gap hierarchy, identifies these slopes as effective finite-window descriptors. All quoted low-temperature exponents refer to the L=233L = 233 scan. The resulting peffp_{\mathrm{eff}} values provide a controlled finite-window connection to quasiperiodic Kibble–Zurek dynamics.

This distinction also separates the low-temperature critical-ramp limit from the finite-temperature engine boundary. The latter is reached while many thermally weighted branches remain active and before microscopic adiabatic following is restored. The former progressively isolates the lowest-energy excitation physics. Temperature connects these two regimes of the same driven Hamiltonian while changing the number and structure of the branches that carry thermodynamic weight.

The opposite limiting comparison is the sudden quench. In the state-frozen reference, the branch-resolved characteristic function develops the expected Loschmidt-type structure, and the crossing path shows a pronounced return minimum that is absent in smoother controls [5, 10, 48, 49]. This provides a useful connection between the fast reference transformation and standard quench diagnostics.

Across the four sizes, the continuously driven characteristic-function minima saturate and the physical return probability remains finite. The finite-rate ramps occupy a nonsingular crossover regime, while the pronounced Loschmidt-type structure belongs to the sudden reference. The corresponding temperature, fit-window, and return diagnostics are given in Appendix § 1.18.

Finite contacts, efficiency, power, and entropy production

The complete-reset cycle isolates the work-stroke dynamics by allowing each bath to restore its Gibbs populations fully. Finite thermal contacts restore the second clock introduced in Sec. II: the time required to renew those endpoint populations. Incomplete renewal leaves population memory in the stationary cycle, making thermalization itself a second accessibility constraint while also adding the contact duration to the performance problem [22, 34, 47].

Incomplete thermalization as a second accessibility constraint

We model contact with bath b∈{h,c}b \in\{h,c\} by the population semigroup

p↦(1−λb)p+λbpβb,λb=1−e−Γbτb,(12)\mathbf{p} \mapsto(1-\lambda_b)\mathbf{p}+\lambda_b\mathbf{p}_{\beta_b}, \qquad\lambda_b = 1-e^{-\Gamma_b\tau_b}, \tag*{(12)}

where Γb\Gamma_b is the relaxation rate and τb\tau_b the contact duration. The complete-reset reference is recovered only asymptotically as λb→1\lambda_b \to1. Equation (12) is a single-rate Gibbs-renewal model; the numerical contact thresholds below therefore refer to this channel. A coherence-retaining Gibbs channel provides a complementary comparison of weak-contact sensitivity. The stationary population map and fixed-point construction are given in Appendix D.

Finite contacts introduce a second boundary in addition to the work-stroke boundary of Sec. III. The unitary strokes may already be slow enough to support positive mean work, yet the stationary cycle can still fail if the baths do not renew the endpoint populations strongly enough. Table III shows that the required symmetric reset strength depends strongly on the chosen finite-rate protocol. The near-critical control requires almost complete population renewal, whereas the extended and crossing protocols tolerate appreciably incomplete contacts. If either bath is fully reset, the required strength of the other is lower, showing that the two contacts jointly set the population memory retained by the stationary cycle.

Pathν/J\nu/JTh/JT_h/Jλcrit\lambda_{\mathrm{crit}}
Extended0.251.00.46678
Near-critical0.250.80.99995
Crossing0.030.650.83861

Table III. Finite-contact accessibility under the population semigroup. The symmetric reset strength λcrit\lambda_{\mathrm{crit}} marks the onset of positive stationary mean work for each representative finite-rate protocol.

Residual coherence can modify this contact boundary. To estimate its influence, we also propagate the full density matrix through the affine Gibbs channel ρ↦(1−λ)ρ+λρβ\rho\mapsto(1-\lambda)\rho+\lambda\rho_\beta, which retains a fraction 1−λ1-\lambda of the pre-contact coherence. On the extended path at symmetric λ=0.40\lambda=0.40, the population model gives W/J=−7.02×10−4W/J=-7.02\times10^{-4} whereas the coherence-retaining channel gives W/J=1.89×10−3W/J=1.89\times10^{-3}, reversing the operating sign. On the crossing path both models change sign within the same numerical bracket 0.83<λcrit<0.840.83<\lambda_{\mathrm{crit}}<0.84; the population value in Table III is 0.83861. Thus the crossing-path contact boundary is insensitive to this particular coherence control at the quoted resolution, while the smoother path is qualitatively coherence sensitive in the weak-contact regime. Equation (12) remains the population-level model used for the quantitative operating analysis. Appendix § 1.1.4 gives the coherence-retaining channel and its convergence details.

Power, stochastic accessibility, and irreversibility

A finite contact must be paid for in cycle time. Defining the contact exposure xb=Γbτbx_b=\Gamma_b\tau_b, the stationary cycle duration is

τcyc=2∣rc−rh∣ν+xhΓh+xcΓc,(13)\tau_{\mathrm{cyc}}=\frac{2\lvert r_c-r_h\rvert}{\nu}+\frac{x_h}{\Gamma_h}+\frac{x_c}{\Gamma_c}, \tag*{(13)}

and the mean power is P=Wout/τcycP=W_{\mathrm{out}}/\tau_{\mathrm{cyc}} in the engine sector. With time measured in units of ℏ/J\hbar/J, the cycle duration in Eq. (13) converts directly to physical time once JJ and the reservoir relaxation rates are specified. The plotted bath rate Γb/J\Gamma_b/J corresponds to the dimensionless physical rate ℏΓb,phys/J\hbar\Gamma_{b,\mathrm{phys}}/J. For example, the crossing

path has τ≃24.7\tau\simeq24.7 at ν/J=0.03\nu/J=0.03 and τ≃43.6\tau\simeq43.6 at the L=233L=233 grid maximum-power rate ν/J=0.017\nu/J=0.017. For ℏΓb,phys/J=1\hbar\Gamma_{b,\mathrm{phys}}/J=1, a contact exposure xb=5x_b=5 corresponds to a contact duration 5ℏ/J5\hbar/J.

The rapid-driving limit determines how the stroke-time cost competes with output. The smooth path ending at rc=0.5r_c=0.5 remains engine-like in the state-frozen limit, whereas the crossing path becomes non-engine-like; the latter combines a fast-drive loss of engine operation with the time cost of very slow ramps. Figure 5(a) shows the resulting broad maximum-power region. On the evaluated rate grid with Γh=Γc=J\Gamma_h=\Gamma_c=J, the grid maximum lies around $\nu/J\approx 0.018 across the four sizes (the tabulated L=233L=233 point is ν/J=0.017\nu/J=0.017), while the peak power on the grid changes by only a few percent from L=233L=233 to 987. Faster baths shift the optimum toward larger ν\nu, but yield only modest additional power once the work-stroke duration dominates the cycle.

Figure 5. Finite contacts reshape mean performance and one-cycle stochastic accessibility on the crossing path. (a) Stationary mean power versus driving rate for symmetric contacts with Γh=Γc=J\Gamma_h=\Gamma_c=J and xh=xc=5x_h=x_c=5 at L=233,377,610,987L=233,377,610,987. Markers are direct primary calculations and connecting lines are visual guides. (b) Signed one-cycle mean-to-width ratio R=⟨W⟩/σW\mathcal{R}=\langle W\rangle/\sigma_W over the (ν/J,x)(\nu/J,x) control plane at L=233L=233, with x=Γτbx=\Gamma\tau_b. The solid contour marks ⟨W⟩=0\langle W\rangle=0 and the dashed contour P+=P(W>0)=1/2P_+=P(W>0)=1/2. The star marks the grid maximum-power point for Γ/J=1\Gamma/J=1 and the diamond a <1%<1\% power-loss compromise point.

Mean power is only one operational criterion. Figure 5(b) shows how the one-cycle signed mean-to-width ratio R=⟨W⟩/σW\mathcal{R}=\langle W\rangle/\sigma_W varies with driving rate and contact exposure. The mean-work and majority-positive contours remain distinct, and incomplete thermalization can reverse their ordering relative to the complete-reset case. In particular, a narrow region appears in which the stationary mean work is already positive while P+<1/2P_+<1/2: mean engine operation can be recovered before a majority of individual cycles deliver positive work.

The star and diamond in Figure 5(b) show that the grid maximum-power point and the best stochastic-quality point occur at different locations. Table IV compares that point with a point chosen inside the 1% maximum-power band. A small reduction in power buys larger work per cycle, efficiency, and one-cycle reliability, while P+P_+ also increases. The finite-time engine has a broad operating compromise, with different optima on the evaluated grid for power, work per cycle, efficiency, and fluctuation measures, consistent with the broader power–precision tradeoffs of stochastic heat engines [35, 36].

Max. powerCompromiseWeak contact
ν/J\nu/J0.0170.0160.017
xx5.05.40.8
W/JW/J0.02860.03020.0012
η\eta0.1260.1310.014
104P/J210^4\mathcal{P}/J^22.9482.9210.133
R\mathcal{R}0.1300.1380.006
P+P_+0.5580.5610.489
Σ˙TPM\dot{\Sigma}_{\mathrm{TPM}}0.01690.01590.0077

Table IV. Representative L=233L=233, Γ/J=1\Gamma/J=1 crossing-path operating points. The compromise point has the largest evaluated R\mathcal{R} within the 1% maximum-power band. Here Σ˙TPM=ΣTPM/τcyc\dot{\Sigma}_{\mathrm{TPM}}=\Sigma_{\mathrm{TPM}}/\tau_{\mathrm{cyc}}.

For the TPM-resolved stationary trajectory, the mean entropy production is ΣTPM=−βhQh−βcQc\Sigma_{\mathrm{TPM}}=-\beta_hQ_h-\beta_cQ_c. It contains contributions from population relaxation during the contacts and from the measurement-induced dephasing associated with the TPM work protocol. Their decomposition is given in Appendix § 1.16; an unmeasured coherent cycle carries a different entropy accounting because the TPM dephasing step is absent.

The entropy-production rate decreases toward weak contacts, while useful output requires sufficient renewal. Table IV shows that reducing Σ˙TPM=ΣTPM/τcyc\dot{\Sigma}_{\mathrm{TPM}}=\Sigma_{\mathrm{TPM}}/\tau_{\mathrm{cyc}} can coincide with a collapse of work, efficiency, power, and reliability. Irreversibility forms one coordinate of the operational trade-off alongside output and fluctuation measures. Finite contacts also retain information from one cycle to the next, introducing a temporal correlation scale beyond stationary one-cycle fluctuations.

Repeated cycles and long-time stochastic thermodynamics

Finite contacts correlate successive cycles, so statistics accumulated over many repetitions contain information beyond a single stationary-cycle distribution [41, 50]. The population left at the end of one cycle becomes the input to the next, so successive work outcomes are no longer independent. The relevant question is whether the one-cycle reliability remains representative of repeated operation or whether intercycle memory changes the accumulated variance appreciably.

Full counting statistics captures these intercycle correlations within the same stationary cycle map [37, 38, 41]. For a work counting field χ\chi, let Tχ\mathcal{T}_{\chi} denote the one-cycle population propagator with each trajectory weighted by its work increment.

The scaled cumulant-generating function (SCGF) is

ψ(χ)=lim⁡N→∞1Nln⁡⟨eχXN⟩=ln⁡ρmax⁡(Tχ)(14)\psi(\chi)=\lim_{N\to\infty}\frac{1}{N}\ln\left\langle e^{\chi X_N}\right\rangle=\ln\rho_{\max}(\mathcal{T}_{\chi}) \tag*{(14)}

where XNX_N is the accumulated work after NN cycles. Derivatives of ψ\psi at χ=0\chi=0 give the long-time cumulants per cycle. In particular, for the stationary process the asymptotic work variance can be written as

σ∞2=σ12+2∑ℓ≥1Cov⁡(Wq,Wq+ℓ)(15)\sigma_{\infty}^{2}=\sigma_{1}^{2}+2\sum_{\ell\geq1}\operatorname{Cov}(W_q,W_{q+\ell}) \tag*{(15)}

so the difference between σ∞2\sigma_{\infty}^{2} and the one-cycle variance σ12\sigma_{1}^{2} directly measures the accumulated effect of intercycle correlations. The quantity R∞=⟨W⟩/σ∞\mathcal{R}_{\infty}=\langle W\rangle/\sigma_{\infty} is an asymptotic reliability prefactor: for accumulated work XNX_N, ⟨XN⟩/Var⁡(XN)∼NR∞\langle X_N\rangle/\sqrt{\operatorname{Var}(X_N)}\sim\sqrt{N}\mathcal{R}_{\infty}, rather than tending to R∞\mathcal{R}_{\infty} itself.

For the L=233L=233 crossing protocol, Table V shows that this correction depends strongly on the operating regime. At the refined maximum-power point, the long-time variance per cycle differs from the one-cycle value by less than 0.3%0.3\%. The corresponding reliability prefactor changes only from R1≃0.1300\mathcal{R}_1\simeq0.1300 to R∞≃0.1302\mathcal{R}_{\infty}\simeq0.1302. Thus, at the maximum-power strong-contact point, single-cycle stationary statistics reproduce the long-time variance and reliability prefactor at the sub-percent level.

Max. powerNear threshold
ν/J\nu/J0.0170.030
xx5.001.83
R1R_10.13003.70×10−43.70 \times10^{-4}
R∞R_\infty0.13023.86×10−43.86 \times10^{-4}
Δσ2/σ12\Delta\sigma^2/\sigma_1^2−0.28%-0.28\%−7.78%-7.78\%

Table V. One-cycle and long-time work reliability. Comparison of the maximum-power point (ν/J,x)=(0.017,5.00)(\nu/J,x)=(0.017,5.00) with a near-threshold point (0.030,1.83)(0.030,1.83). Here R1=⟨W⟩/σ1R_1=\langle W\rangle/\sigma_1, R∞=⟨W⟩/σ∞R_\infty=\langle W\rangle/\sigma_\infty, and the final row gives the relative intercycle correction to the work variance per cycle.

Close to the finite-contact threshold, intercycle memory becomes quantitatively important. The variance correction grows to about −7.8%-7.8\%, while the reliability prefactor changes from approximately 3.70×10−43.70\times10^{-4} to 3.86×10−43.86\times10^{-4}. The negative correction means that the net sum of intercycle work covariances is negative. In this sense the repeated-cycle memory acts as an anticorrelation that suppresses the accumulated variance. This anticorrelation slightly improves the long-time reliability prefactor relative to the one-cycle estimate, while the near-threshold mean output and both reliability measures remain very small.

The same tilted-cycle construction can count hot and cold heat together with work. For the TPM-resolved population process defined by Eq. (12), the Gibbs-renewal contacts obey local detailed balance and the retraced work strokes satisfy PR=PTP_R=P_T; over the evaluated counting-field range, the entropy generating function displays the Gallavotti–Cohen symmetry and the joint SCGF satisfies the cycle first-law identity to numerical precision. The coherence-retaining contact channel is treated separately and is outside this monitored population construction. In the useful strong-contact regime, the one-cycle variance already approximates the long-time variance closely; near the contact threshold, the long-time statistics resolve the remaining cycle-to-cycle memory.

Discussion and conclusion

The finite-rate Otto cycle resolves a hierarchy of dynamical information that is hidden in the usual slow-driving and state-frozen limits. Along the quasiperiodic crossing path, thermodynamic recovery does not coincide with microscopic energy-ordered following. Positive mean work returns while the transition matrix remains strongly redistributed, because population transferred away from the energy-ordered channels is progressively redirected toward energetically less costly processes. Across the recovery region, the total amount of population redistribution changes only weakly while its energetic penalty decreases substantially. The relevant distinction is therefore not simply how much population is transferred, but how much energetic weight those transitions carry.

The spectral analysis identifies where this redistribution is generated. Thermally occupied regions with large spectral susceptibility carry a disproportionate share of the excess-energy growth, while the complete-cycle deficit remains distributed over a broad set of spectral regions. The narrowest isolated avoided crossing accounts for only a small fraction of the deficit in the representative recovery regime. The equal-duration schedule control provides an independent dynamical test: reallocating time toward high-susceptibility regions improves the complete-cycle work, whereas reallocating the same time toward low-susceptibility regions worsens it, even though the transition map remains far from energy-ordered following. Spectral accessibility is therefore determined jointly by dynamical coupling, thermal occupation, and energetic weight rather than by a single spectral scale.

The work distribution resolves a further separation of dynamical scales. The majority-positive and positive-mean boundaries remain distinct over a finite rate interval for all four Fibonacci sizes studied. Within the negative-work sector, the conditional energetic loss increases strongly with branch displacement and is concentrated in the thermally occupied low-energy manifold. Mean operation, majority-cycle success, fluctuation-based reliability, and microscopic energy-ordered following consequently probe different coarse grainings of the same finite-rate unitary transformation. Table VI summarizes this hierarchy together with the additional information introduced by finite thermal contacts and repeated operation.

Diagnostic layerPhysical content
Finite-rate accessibilityThe response is distributed across thermally occupied, dynamically coupled spectral regions. Equal-duration time reallocation guided by the spectral susceptibility changes the work systematically, while the narrowest isolated crossing contributes only a small fraction of the complete-cycle deficit.
Mean work and branch followingPositive mean work can recover while the transition map remains strongly redistributed because the surviving off-diagonal transfer becomes energetically less costly.
Fluctuation resolutionThe majority-positive and positive-mean boundaries remain separated across the four sizes, while large-displacement negative-work trajectories carry systematically larger losses.
Finite thermal contactsWithin the specified Gibbs-renewal model, incomplete population renewal creates a second accessibility constraint and separates the optima found on the evaluated grid for work, power, efficiency, reliability, and irreversibility.
Repeated operationStrong-contact operation is well represented by one-cycle variance, while contact-threshold operation develops appreciable intercycle memory.

Table VI. Hierarchy of finite-time thermodynamic information. Each diagnostic resolves a different layer of the same driven quasiperiodic cycle.

The quasiperiodic structure determines the rate and path dependence of this hierarchy. Matched-temperature, matched-duration, and equal-excursion controls retain the enhanced finite-rate sensitivity of the crossing route, while changes in modulation, phase, and diagonal–off-diagonal balance shift the accessibility boundary by reshaping the spectral path and endpoint thermodynamics. Cooling restricts the active spectrum toward the low-energy sector, where the finite-rate response develops reproducible Kibble–Zurek-like scaling over finite windows. At the opposite end of the rate axis, the state-frozen reference develops pronounced Loschmidt-type structure, whereas the continuous ramps interpolate smoothly toward that limit. The fluctuation ordering is likewise protocol dependent: the localised control is broadest in the strict slow-driving reference, while the critical control dominates in the state-frozen limit and the finite-rate maximum shifts toward the critical-localised crossover.

Finite thermal contacts add an independent dynamical timescale. Within the Gibbs-renewal model, incomplete population renewal can suppress work extraction even when the isolated strokes already lie in the positive-work regime. The coherence-retaining control further shows that weak-contact operation can depend on the dynamical channel. Because work-stroke and contact durations both enter the cycle time, finite-contact operation produces distinct compromises among work per cycle, power, efficiency, reliability, and irreversibility. The same incomplete renewal generates intercycle correlations. These correlations remain weak near the useful maximum-power region considered here but become appreciable near the contact threshold, where long-time counting statistics are needed to characterize the fluctuations consistently.

Taken together, these results establish a finite-rate thermodynamic picture of quasiperiodic spectral accessibility. The cycle responds to a thermally weighted network of dynamical pathways whose energetic relevance depends on the driving protocol, rather than to a single avoided crossing or static localisation label. Mean work, trajectory statistics, microscopic transition maps, and long-time fluctuations then resolve successive levels of the same structured dynamics. Thermodynamic recovery, stochastic success, and microscopic following are consequently distinct physical notions, and their separation provides an operational route for resolving finite-rate spectral structure through quantum thermodynamic observables.

Outlook. The present work establishes a controlled single-particle setting in which finite-rate spectral accessibility can be followed from microscopic population transfer to mean thermodynamics and full work statistics. A first research program is to determine which aspects of this hierarchy are specifically enhanced by quasiperiodic order and which belong more generally to structured multilevel dynamics. This requires matched periodic, disordered, and engineered controls in which the thermally active bandwidth, gap distribution, path length, and driving duration are placed on comparable footing. Such a comparison would turn the present AAH result into a broader test of spectral-accessibility mechanisms. Introducing finite filling and interactions provides the next stage of the same program. Occupation constraints, level repulsion, scattering, and many-body level structure would then allow the separation between microscopic following, positive mean operation, and stochastic reliability to be tested when the accessible spectrum is generated collectively rather than by a single-particle band problem. These questions require new Hamiltonian families and many-body dynamical calculations beyond the controlled single-particle setting considered here.

A second research program concerns the reservoir and coherence sectors of the finite-rate mechanism. The Gibbs-renewal contacts used here isolate the work-stroke dynamics and provide a transparent two-clock description. Microscopic reservoirs would instead introduce energy-dependent transition rates, eigenstate-selective couplings, and bath-induced coherence or dephasing, allowing the spectral structure of the working medium and that of the environment to be treated on the same footing. The coherence-retaining control already shows that weak-contact operation can depend on the dynamical channel, motivating a direct comparison of monitored and coherence-preserving fluctuation statistics [45]. Within the same program, the gauge-invariant formulation of finite-time Otto thermodynamics developed in Ref. [51] is directly relevant to the central separation found here: positive mean work recovers while energy-ordered microscopic following remains strongly broken. Applying this decomposition to the driven AAH strokes could determine whether that recovery is carried predominantly by the sector associated with the instantaneous spectrum and populations, by coherence generated during finite-rate driving, or by a cancellation between the two contributions. This would distinguish thermodynamic recovery arising from the energetic reweighting of strongly redistributed populations from recovery that depends essentially on quantum coherence. Addressing this question requires access to the full density matrix throughout the driven strokes, rather than only the population transition maps sufficient for the present analysis, and therefore defines a dedicated coherence-resolved study.

A third research program is to use the finite-rate cycle as an inference tool for spectral structure. The present results show that temperature, driving rate, and path select different dynamically active regions of the spectrum and leave distinct signatures in work and heat statistics. Scanning these controls systematically could provide an overconstrained thermodynamic data set from which properties of the underlying spectral dynamics may be inferred without reconstructing the complete microscopic state. Establishing such an inverse protocol requires combining many operating points and determining which spectral information is uniquely encoded in the thermodynamic response. The low-temperature regime offers a complementary route toward asymptotic critical dynamics. Larger systems, slower ramps, and scaling windows stable against the quasiperiodic gap hierarchy would allow the finite-window Kibble–Zurek-like behavior found here to be connected to a controlled asymptotic scaling theory. These directions extend the same central question—how structured quantum dynamics becomes visible through thermodynamic observables—toward universality, many-body physics, microscopic reservoirs, coherence-resolved energetics, and spectral inference.

Author contributions

S.C. contributed to the methodology, numerical investigation, and software implementation. L.C.C., N.G.A., and G.X. contributed to the conceptualization and validation of the work. G.D.M.N. contributed to the conceptualization, methodology, theoretical analysis, numerical investigation, visualization, and preparation of the original draft. All authors discussed the results, reviewed the manuscript, and approved the submitted version. OpenAI ChatGPT (GPT-5.6 Sol) was used for language editing, bibliographic organization, and code-assisted analytical and numerical cross-checks. The authors specified the physical model and validation criteria, checked the resulting calculations against analytic identities and numerical convergence tests, and assume full responsibility for the manuscript.

Acknowledgments

This research is supported by Zhejiang Provincial Natural Science Foundation of China under Grant No. LQN25A040012, the start-up fund from Xingzhi College, Zhejiang Normal University, and the National Natural Science Foundation of China under Grant No. 12604416. LCC acknowledges support from the Brazilian National Council for Scientific and Technological Development (CNPq) through grant 305636/2026-0, the National Institute of Science and Technology for Applied Quantum Computing through CNPq grant 408884/2024-0, Goiás Research Foundation (FAPEG) through grant 202510267001843, and São Paulo Research Foundation (FAPESP) through grant 2025/23726-4.

Data and code availability

Data and numerical code supporting the findings of this study are available from the corresponding author upon reasonable request. ... erence with that of a finite-rate cycle and serves as the cycle-level measure of the finite-rate work penalty.

Finite-rate propagation and transition diagnostics

For a linear ramp of duration τ=∣rc−rh∣/ν\tau= |r_c-r_h|/\nu, the interval is divided into N=⌈τ/Δτmax⁡⌉N = \lceil\tau/\Delta\tau_{\max}\rceil equal steps of size h=τ/Nh = \tau/N. The Hamiltonian is evaluated at each step’s midpoint. For ℏ=1\hbar= 1, the midpoint Cayley update is

(I+ihHj+1/22)∣ψj+1⟩=(I−ihHj+1/22)∣ψj⟩.(16)\left(I+\frac{ihH_{j+1/2}}{2}\right)|\psi_{j+1}\rangle=\left(I-\frac{ihH_{j+1/2}}{2}\right)|\psi_j\rangle. \tag*{(16)}

This second-order update is unitary up to the linear-solver tolerance. Choosing NN from the ceiling function keeps the final control parameter exactly at the requested endpoint even when τ/Δτmax⁡\tau/\Delta\tau_{\max} is not an integer.

The crossing-path transition and work anchors use the dimensionless timestep Δτmax⁡=0.0025\Delta\tau_{\max}=0.0025 and agree with a full-basis adaptive DOP853 reference with relative and absolute tolerances 2×10−112\times10^{-11} and 2×10−132\times10^{-13}, respectively. The stochastic maps in Figure 3 use Δτmax⁡=0.005\Delta\tau_{\max}=0.005. Relative to the finer 0.0025 grid, this changes the mean work by at most 6.6×10−6J6.6\times10^{-6}J over the common evaluated plane and P(W>0)P(W>0) by at most 1.3×10−61.3\times10^{-6} at the three representative rates. The finite-contact power scans also use Δτmax⁡=0.005\Delta\tau_{\max}=0.005, which sets their quoted numerical resolution.

For real H(r)H(r) and a reversed protocol, the forward and reverse transition matrices obey PR=PFTP_R=P_F^{T}. The propagated matrices satisfy row and column normalization and the transpose relation within numerical precision. The normalized distances from the energy-ordered and state-frozen reference maps are

Dad=∥PF−I∥FL,D_{\mathrm{ad}}=\frac{\lVert P_F-I\rVert_F}{\sqrt{L}},
Dsud=∥PF−Psud∥FL,D_{\mathrm{sud}}=\frac{\lVert P_F-P_{\mathrm{sud}}\rVert_F}{\sqrt{L}},
Psud(m∣n)=∣⟨mc∣nh⟩∣2.(17)P_{\mathrm{sud}}(m|n)=|\langle m_c|n_h\rangle|^2. \tag*{(17)}

These diagnostics compare population transition matrices; phase-sensitive information resides in the full propagator. For orientation, a uniformly mixing matrix gives Dad=1−1/LD_{\mathrm{ad}}=\sqrt{1-1/L}, so values near unity diagnose diffuse population transfer. The associated energy scale is supplied separately by the Gibbs-weighted leakage and the energy-weighted work deficit. For an initial Gibbs distribution pnp_n, the probability weight transferred away from its energy-ordered channel is ℓ=∑npn[1−PF(n∣n)]\ell=\sum_n p_n[1-P_F(n|n)]; the reverse leakage is defined with PRP_R and the cold Gibbs distribution.

The cycle-level work deficit Wfric=Wad−WνW_{\mathrm{fric}}=W_{\mathrm{ad}}-W_\nu compares the work output of the strict energy-ordered reference with that of a finite-rate cycle and serves as the cycle-level measure of the finite-rate work penalty.

Transition-resolved work deficit

The work deficit can be resolved directly into the transitions generated during the two isolated strokes. For the forward stroke,

ΔWF=∑n,mphnPF(m∣n)(Emc−Enc).(18)\Delta W_F=\sum_{n,m}p_{h n}P_F(m|n)\left(E_m^c-E_n^c\right). \tag*{(18)}

while the reverse stroke contributes

ΔWR=∑k,ℓpckPR(ℓ∣k)(Eℓh−Ekh).(19)\Delta W_R=\sum_{k,\ell}p_{c k}P_R(\ell|k)\left(E_\ell^h-E_k^h\right). \tag*{(19)}

Their sum gives

Wad−Wν=ΔWF+ΔWR.(20)W_{\mathrm{ad}}-W_\nu=\Delta W_F+\Delta W_R. \tag*{(20)}

The deficit is nonnegative under the assumptions of the complete-reset protocol. For either stroke, order the positive-temperature Gibbs populations as p1≥p2≥⋯≥pLp_1\ge p_2\ge\cdots\ge p_L and the endpoint energies as E1≤E2≤⋯≤ELE_1\le E_2\le\cdots\le E_L. A unitary transition matrix is unistochastic and hence doubly stochastic, so by the Birkhoff–von Neumann theorem it is a convex combination of permutation matrices. The rearrangement inequality then gives

ETPp≥ETp,(21)\mathbf{E}^{T}P\mathbf{p}\ge\mathbf{E}^{T}\mathbf{p}, \tag*{(21)}

because the energy-ordered assignment pairs the largest populations with the lowest endpoint energies. Applying this result to the two strokes separately yields

ΔWF≥0,ΔWR≥0,Wad−Wν≥0.(22)\Delta W_F\ge0,\qquad\Delta W_R\ge0,\qquad W_{\mathrm{ad}}-W_\nu\ge0. \tag*{(22)}

The proof assumes positive-temperature Gibbs populations; degeneracies make the inequalities non-strict. The diagonal terms in Eqs. (A3) and (A4) vanish identically. At fixed endpoints and bath temperatures, the Gibbs weights and endpoint energy differences are independent of the driving rate, so the rate dependence of the work deficit comes entirely from the redistribution of transition probabilities.

Table A1 shows the decomposition for the crossing protocol at L=233L=233, Th/J=0.65T_h/J=0.65, and Tc/J=0.1T_c/J=0.1. Across the mean-work recovery boundary, from ν/J=0.04\nu/J=0.04 to 0.03, the total deficit decreases by 23.69%, whereas the mean thermally weighted leakage decreases by only 0.185%. Defining

ν/J\nu/J⟨W⟩/J\langle W\rangle/J(Wad−Wν)/J(W_{\mathrm{ad}}-W_{\nu})/JΔWF/J\Delta W_F/JΔWR/J\Delta W_R/Jℓˉ\bar{\ell}DadD_{\mathrm{ad}}
0.040−6.98×10−4-6.98\times10^{-4}0.052890.009830.043070.967650.98285
0.0300.011830.040360.007500.032860.965850.98249
0.0170.029280.022920.004320.018600.961710.98170
Reduction of total work deficit23.69%
Reduction of mean leakage0.185%
Reduction of ϵleak\epsilon_{\mathrm{leak}}23.55%
Reverse-stroke share of deficit reduction81.45%
Net recovery from d≥20d\geq2098.16%
Largest 20 transitions / positive recovery weight4.30%

Table A1. Transition-resolved origin of mean-work recovery on the crossing path. Rate-resolved work deficit and microscopic population redistribution for L=233L=233, Th/J=0.65T_h/J=0.65, and Tc/J=0.1T_c/J=0.1. The lower block resolves the change from ν/J=0.04\nu/J=0.04 to 0.030.03. Here ϵleak=(Wad−Wν)/(ℓF+ℓR)\epsilon_{\mathrm{leak}}=(W_{\mathrm{ad}}-W_{\nu})/(\ell_F+\ell_R) is the signed net work-deficit-to-leakage ratio and includes cancellations among transition contributions.

ϵleak=Wad−WνℓF+ℓR,(23)\epsilon_{\mathrm{leak}}=\frac{W_{\mathrm{ad}}-W_\nu}{\ell_F+\ell_R}, \tag*{(23)}

this net work-deficit-to-leakage ratio falls by 23.55%. The probability of leaving the energy-ordered channels remains nearly unchanged while the net work deficit associated with that redistributed population decreases substantially.

The reverse stroke accounts for 81.45% of the reduction in the total deficit between the two rates. Grouping transition changes by branch displacement, with d=∣m−n∣d = \lvert m - n\rvert on the forward stroke and d=∣ℓ−k∣d = \lvert\ell- k\rvert on the reverse stroke, the sector d≥20d \ge20 accounts for 98.16% of the net recovery; smaller-displacement changes partially oppose it. The recovery is distributed over many transitions: the twenty largest positive transition-level changes contain only 4.30% of the total positive recovery weight. The mean work recovers through a collective shift of the off-diagonal transition map toward channels carrying smaller net work penalties.

Finite-rate and finite-size robustness

This appendix collects the rate- and size-resolved diagnostics supporting the accessibility analysis. It locates the mean-work and engine-window boundaries, follows the thermally relevant low-spacing hierarchy and transition dynamics through L=987L = 987, and records the numerical convergence scale. The propagation errors remain well below the finite-rate variations reported here.

Figure B1(a) follows the large-LL spectral hierarchy with a low-spacing quantile that remains above the numerical conditioning floor. For each size and path, adjacent instantaneous spacings are pooled from 101 uniformly spaced Hamiltonians along the work stroke. The plotted 0.1% quantile is restricted to the lowest energy-ordered levels carrying 99% of the cold-endpoint Gibbs probability at Tc/J=0.1T_c/J = 0.1. The crossing path retains a substantially lower spacing tail than both controls through L=987L = 987. At that size the quantile is smaller by factors of approximately 6.1 and 4.9 relative to the rc=0.5r_c = 0.5 and rc=0.9r_c = 0.9 controls, respectively. The same ordering is obtained from the full spectrum, confirming the low-spacing hierarchy in both the thermally restricted and full spectral data.

The complementary ν/J=0.04\nu/J = 0.04 size dependence is shown in Figure B1(b), where the precise mean-work boundary is already finite-size dependent in Table B1. Nevertheless, DadD_{\rm ad} is close to unity, DsudD_{\rm sud} is small, and the mean Gibbs-weighted leakage ℓˉ\bar{\ell} is near unity across L=233,377,610,987L = 233, 377, 610, 987. In particular, Dad:0.983→0.995D_{\rm ad}: 0.983 \to0.995, Dsud:0.0679→0.0331D_{\rm sud}: 0.0679 \to0.0331, and ℓˉ:0.968→0.992\bar{\ell}: 0.968 \to0.992 between the smallest and largest displayed sizes. The corresponding ν/J=0.03\nu/J = 0.03 four-size values are listed in main Table II; the Appendix uses the complementary rate to provide an independent view of the same size sequence.

Numerical questionConditionNumerical result
Strict-reference engine intervalcrossing path, L=233L=233, Tc/J=0.1T_c/J=0.10.212<Th/J<1.4030.212<T_h/J<1.403
Finite-rate engine intervalsame path, ν/J=0.03\nu/J=0.030.454<Th/J<0.9740.454<T_h/J<0.974
Mean-work sign changeL=233L=233, Th/J=0.65T_h/J=0.650.0394<ν/J<0.03970.0394<\nu/J<0.0397
Disappearance of any engine windowL=233L=233, scan over ThT_h0.0397<ν/J<0.03980.0397<\nu/J<0.0398
Finite-size shift of the work boundaryν/J=0.04\nu/J=0.04⟨W⟩/J=−6.99×10−4\langle W\rangle/J=-6.99\times10^{-4} (L=233L=233), +1.9×10−5+1.9\times10^{-5} (377), +7.66×10−4+7.66\times10^{-4} (987)
Positive-work probability near that boundaryν/J=0.04\nu/J=0.04P(W>0)=0.506P(W>0)=0.506 (L=233L=233) and 0.508 (987)
Resolved branch-character exchangeL=89L=89, r∗≃1.044r_\ast\simeq1.044same/exchanged overlap weights change from 0.844/0.156 at δr=10−4\delta r=10^{-4} to 0.179/0.821 at δr=5×10−4\delta r=5\times10^{-4}

Table B1. Numerical landmarks for the crossing protocol. Representative values locating the finite-rate operating boundaries, their finite-size shift, and the associated branch exchange.

The propagation and reciprocity convergence data for these calculations are collected in Appendix A. At L=987L = 987, reducing the dimensionless timestep Δτ\Delta\tau from 0.005 to 0.0025 changes the work only at the 7×10−77 \times10^{-7}–8×10−7J8 \times10^{-7}J scale at the two rates used in the size calculations, well below the finite-rate changes reported in the main text.

At L=987L = 987 and ν/J=0.041\nu/J = 0.041, reducing the dimensionless timestep from 0.005 to 0.0025 changes ⟨W⟩/J\langle W\rangle/J by 8.3×10−78.3 \times10^{-7} and P+P_+ by 1.2×10−61.2 \times10^{-6}, well below the distances from the two sign boundaries.

Matched controls and thermally weighted spectral response

To separate spectral-path sensitivity from the different representative temperatures and stroke durations, we compare the three original paths at common Th/J=0.65T_h/J = 0.65, Tc/J=0.1T_c/J = 0.1, and common work-stroke duration τ=18.5167\tau= 18.5167. The results are summarised in Table B3. The crossing path remains the only work-consuming protocol and retains the largest work width, while its transition map and thermally occupied sector are much farther from energy-ordered following than the two controls.

TABLE B3. Matched-duration path control and spectral susceptibility. All three paths use L=233L = 233, Th/J=0.65T_h/J = 0.65, Tc/J=0.1T_c/J = 0.1, and τ=18.5167\tau= 18.5167. The rate is adjusted for each excursion to keep the stroke duration fixed. χFmax⁡\chi_F^{\max} and χRmax⁡\chi_R^{\max} are the maxima of Eq. (24) on the forward and reverse strokes.

Pathν/J\nu/J⟨W⟩/J\langle W\rangle/JσW/J\sigma_W/JDadD_{\mathrm{ad}}ℓˉ\bar{\ell}χFmax⁡\chi_F^{\max}χRmax⁡\chi_R^{\max}
Extended0.005400.013030.027560.01730.004531.030.43
Near-critical0.032730.046490.190130.35890.2405011.6411.51
Crossing0.04000−6.99×10−4-6.99 \times10^{-4}0.258140.98280.967656.42×1036.42 \times10^38.16×1038.16 \times10^3

Table B3.

A direct spectral-response diagnostic is obtained from

grr(n)(r)=∑m≠n∣⟨m(r)∣∂rH∣n(r)⟩∣2[Em(r)−En(r)]2,g_{rr}^{(n)}(r) = \sum_{m \ne n} \frac{\left\lvert\langle m(r) \lvert\partial_r H \rvert n(r) \rangle\right\rvert^2}{\left[E_m(r) - E_n(r)\right]^2},
χb(r)=J2∑npbngrr(n)(r),(24)\chi_b(r) = J^2 \sum_n p_{bn} g_{rr}^{(n)}(r), \tag*{(24)}

where pbnp_{bn} are the Gibbs weights at the beginning of the corresponding stroke, carried along the energy-ordered labels. Away from exact degeneracies, grr(n)g_{rr}^{(n)} is the band quantum metric along the driven coordinate rr [52]; hence χb/J2\chi_b/J^2 is a population-weighted sum of band quantum metrics. We refer to χb\chi_b as the thermally weighted spectral susceptibility. A full mixed-state thermal fidelity metric also contains population-dependent terms. Equation (B1) weights instantaneous interlevel couplings by Gibbs occupation and inverse squared spacing; finite-rate transition probabilities additionally depend on dynamical phases and interference.

For the crossing path, log⁡χb\log\chi_b is strongly rank-correlated with the local excess-energy increments. At ν/J=0.04\nu/J=0.04, the Spearman rank coefficients between log⁡χb\log\chi_b and the absolute excess-energy increments are 0.842 on the forward stroke and 0.650 on the reverse stroke. The 20% of forward-stroke intervals with the largest χb\chi_b carry 62.8% of the positive excess-energy growth, and the largest 30% carry 76.6%. At the same stroke duration, the maximum forward susceptibility increases from approximately 1.0 on the extended path and 11.6 on the near-critical path to 6.4×1036.4\times10^3 on the crossing path; the reverse-stroke maximum reaches 8.2×1038.2\times10^3.

To resolve the complete-cycle energetic contribution spatially, define the instantaneous forward excess energy

δEF(r)=Tr⁡[ρF(r)H(r)]−∑nphnEn(r),(25)\delta E_F(r)=\operatorname{Tr}[\rho_F(r)H(r)]-\sum_n p_{hn}E_n(r), \tag*{(25)}

and analogously δER(r)\delta E_R(r) with the cold Gibbs weights on the reverse stroke. For spatial bins [rj,rj+1][r_j,r_{j+1}], the signed increments CjFC_j^F and CjRC_j^R are the corresponding changes in δEF\delta E_F and δER\delta E_R, with the reverse bins mapped onto the same ascending-rr grid. They satisfy exactly

∑j(CjF+CjR)=ΔWF+ΔWR.(26)\sum_j(C_j^F+C_j^R)=\Delta W_F+\Delta W_R. \tag*{(26)}

At ν/J=0.04\nu/J=0.04, ranking bins by χh+χc\chi_h+\chi_c shows that the top 10%, 20%, and 30% of the path carry 15.88%, 31.64%, and 49.96% of the signed cycle deficit. Reaching 50%, 80%, and 90% of the positive cycle deficit requires 29, 59, and 78 of 100 spatial bins, respectively. The reverse-stroke contribution is broader than the forward one and dominates the total deficit.

A refined two-level fit near r/J≃1.041003r/J\simeq1.041003 gives a minimum gap Δ/J≃4.81×10−8\Delta/J\simeq4.81\times10^{-8} and Landau–Zener exponents of order 10−1110^{-11} at both ν/J=0.04\nu/J=0.04 and 0.03, so this pair is effectively diabatic at both rates. Under the additive regional definition above, the window ∣r−r∗∣/J≤0.005|r-r_\ast|/J\leq0.005 contributes 4.83% and 5.19% of the forward deficits at the two rates, but only 2.01% and 2.27% of the complete-cycle deficits because the reverse contribution is both larger and more broadly distributed. The cycle-level work penalty is therefore distributed well beyond this minimum-gap pair.

LL⟨W⟩/J\langle W\rangle/J at 0.041P+P_{+} at 0.041⟨W⟩=0\langle W\rangle= 0 bracketP+=1/2P_{+} = 1/2 bracketΔνsep/J\Delta\nu_{\mathrm{sep}}/J
233−1.962×10−3-1.962 \times10^{-3}0.504270.039375–0.0396880.0425–0.04500.00417
377−1.241×10−3-1.241 \times10^{-3}0.505330.0400–0.04100.0425–0.04500.00429
610−7.705×10−4-7.705 \times10^{-4}0.506020.0400–0.04100.0425–0.04500.00437
987−4.908×10−4-4.908 \times10^{-4}0.506400.0400–0.04100.0450–0.04750.00440

Table B2. Finite-size separation of the mean-work and majority-positive boundaries. All values use the crossing path at Th/J=0.65T_h/J = 0.65 and Tc/J=0.1T_c/J = 0.1. The two middle columns give the numerical brackets containing the zeros of ⟨W⟩\langle W\rangle and P(W>0)−1/2P(W > 0) - 1/2. The final column is the interval width obtained by linear interpolation within those brackets. At the common rate ν/J=0.041\nu/J = 0.041, all four sizes lie inside the sign-separated interval.

The geometric diagnostic can also be tested operationally without changing the total stroke time. Divide the path into 100 equal spatial intervals and rank them by χΣ(r)=χh(r)+χc(r)\chi_{\Sigma}(r) = \chi_h(r) + \chi_c(r). In the high-susceptibility schedule, the top 20% receive 1.25 times the uniform interval duration and the bottom 20% receive 0.75 times that duration; the middle 60% are unchanged. The low-susceptibility control reverses these allocations, and the reverse stroke retraces the same spatial schedule. Because the top and bottom sets have equal width, all three schedules have exactly the same total duration.

At ν/J=0.04\nu/J = 0.04 (τ=18.5167\tau= 18.5167), the high-susceptibility schedule changes ⟨W⟩/J\langle W\rangle/J from −6.98×10−4-6.98 \times10^{-4} to +3.12×10−3+3.12 \times10^{-3} and lowers the work deficit by 7.22%, whereas the opposite allocation gives −4.79×10−3-4.79 \times10^{-3} and raises the deficit by 7.74%. The ordering occurs on both work strokes. At ν/J=0.03\nu/J = 0.03 the same control gives 0.01634, 0.01184, and 0.00670 for the high-susceptibility, uniform, and low-susceptibility schedules, respectively. Throughout, Dad≃0.982D_{\mathrm{ad}} \simeq0.982–0.983 and the mean Gibbs-weighted leakage changes by less than 5×10−45 \times10^{-4}; the energetic improvement occurs while energy-ordered following remains essentially unchanged.

Two-projective-measurement work statistics

For the forward stroke, the output work wF=Enh−Emcw_F = E_n^h - E_m^c occurs with probability pnhPF(m∣n)p_n^h P_F(m|n). For the reverse stroke, wR=Ekc−Eℓhw_R = E_k^c - E_\ell^h occurs with probability pkcPR(ℓ∣k)p_k^c P_R(\ell|k). Complete Gibbs resets make these two stroke outcomes independent, so Pcyc(W)P_{\mathrm{cyc}}(W) is their convolution. The first two moments and sign-conditioned quantities can be written directly from the discrete outcomes:

⟨W⟩=∑aπaWa,σW2=∑aπa(Wa−⟨W⟩)2,\langle W\rangle= \sum_a \pi_a W_a, \qquad\sigma_W^2 = \sum_a \pi_a \left(W_a - \langle W\rangle\right)^2,
P+=∑a:Wa>0πa,R=⟨W⟩σW,(27)P_+ = \sum_{a:W_a>0} \pi_a, \qquad\mathcal{R} = \frac{\langle W\rangle}{\sigma_W}, \tag*{(27)}

where aa labels an outcome pair with probability πa\pi_a and work WaW_a. The signed ratio is retained when ⟨W⟩<0\langle W\rangle< 0; it is called engine reliability only in the positive-mean regime.

Exact sign probabilities and conditional means are evaluated from the unbinned discrete work distribution. Each stroke has L2L^2 level-pair outcomes. We sort the reverse-stroke work values and form cumulative probability and probability-weighted-work sums. For each forward outcome, a binary search identifies the reverse outcomes for which wR>−wFw_R > -w_F. Prefix sums then give P(W>0)P(W > 0), the positive-work sum, and the complementary nonpositive-work sum. O(L2log⁡L)O(L^2 \log L) sorting and search work after the transition matrices have been computed. The conditional means are the exact sign-restricted work sums divided by their probabilities. Any binned density is only a visualization of the discrete convolution; all quoted moments and sign probabilities use the unbinned outcomes.

The sign-separated regime remains visible after excluding near-zero outcomes. At L=233L = 233, Th/J=0.65T_h/J = 0.65, and ν/J=0.041\nu/J = 0.041, a positive-work event of 10−3J10^{-3}J gives P(W>10−3J)=0.5015>1/2P(W > 10^{-3}J) = 0.5015 > 1/2.

To test whether the branch-displacement diagnostic has energetic content, we resolve the negative-work probability and loss above several KK thresholds. Table C1 shows that the negative/positive tail enrichment increases steadily with threshold, while the same large-KK sectors carry a substantial fraction of the negative-work loss. The mean conditional loss also grows across broad displacement bins, from −0.112J-0.112J for K<10K < 10 to −0.314J-0.314J for 50≤K<6050 \le K < 60, −0.430J-0.430J for 60≤K<8060 \le K < 80, and −0.831J-0.831J for K≥100K \ge100. The lowest 32 hot-side branches contain 50.8% of the Gibbs weight but account for 71.1% of negative cycles and 61.4% of the negative-work loss.

ThresholdP(K≥K0∣W<0)P(K \ge K_0 \mid W < 0)P(K≥K0∣W>0)P(K \ge K_0 \mid W > 0)Loss fraction
K0=20K_0 = 200.66100.50970.7854
K0=40K_0 = 400.24550.10770.4278
K0=50K_0 = 500.13280.04240.2798
K0=60K_0 = 600.06730.01600.1706
K0=80K_0 = 800.01540.002280.0519
K0=100K_0 = 1000.002340.0002620.0103

Table C1. Threshold dependence of the negative-work displacement tail. Values refer to L=233L = 233, Th/J=0.65T_h/J = 0.65, and ν/J=0.04\nu/J = 0.04. The probability columns are conditional on work sign, and the last column is the fraction of the total negative-work loss carried by the indicated tail.

Population contacts and stationary operation

For a contact with bath b∈{h,c}b \in\{h,c\}, the population map can be written

Cb=(1−λb)I+λbpβb1T,λb=1−e−Γbτb.(28)C_b = (1-\lambda_b)I + \lambda_b \mathbf{p}_{\beta_b}\mathbf{1}^{\mathsf{T}}, \qquad\lambda_b = 1-e^{-\Gamma_b\tau_b}. \tag*{(28)}

TABLE B4. Protocol dependence of work fluctuations and operating modes. All fluctuation controls use L=233L = 233, Th/J=0.65T_h/J = 0.65, Tc/J=0.1T_c/J = 0.1, and equal excursions Δr/J=0.1\Delta r/J = 0.1 unless stated otherwise. The regional scan moves an equal-width interval across the rr axis. The operating-mode classification for the crossing path uses the temperature grid 0.15≤Th/J≤1.600.15 \le T_h/J \le1.60.

(a) Equal-excursion work widths

ControlPathSlow referenceν/J=0.04\nu/J = 0.04State-frozen
Extended0.02700.04330.1131
Critical0.02280.05980.1243
localised0.03660.04640.1056

(b) Largest width in the sliding equal-excursion scan

Protocolrh/J→rc/Jr_h/J \rightarrow r_c/Jmidpoint rm/Jr_m/JσW/J\sigma_W/J
Slow reference1.65→1.751.65 \rightarrow1.751.700.0373
ν/J=0.04\nu/J = 0.041.00→1.101.00 \rightarrow1.101.050.0608
ν/J=0.25\nu/J = 0.251.00→1.101.00 \rightarrow1.101.050.1179
State-frozen0.95→1.050.95 \rightarrow1.051.000.1248

(c) Crossing-path operating modes

ProtocolRefrigerator sectorEngine sectorOther modes on the grid
Slow reference0.15≲Th/J≲0.1850.15 \lesssim T_h/J \lesssim0.185presentaccelerator, heater
ν/J=0.03\nu/J = 0.03absentpresentaccelerator, heater
ν/J=0.04,0.05,0.10,0.25,0.50,1,10\nu/J = 0.04, 0.05, 0.10, 0.25, 0.50, 1, 10absentabsentaccelerator, heater
State-frozenabsentabsentaccelerator, heater

Table B4.

Here τb\tau_b is the contact duration and Γb\Gamma_b the contact rate. The dimensionless contact coordinate is xb=Γbτbx_b = \Gamma_b\tau_b, so the cycle time contains xb/Γbx_b/\Gamma_b. For nonzero λb\lambda_b, the Gibbs distribution is the unique fixed point of this semigroup; λb=1\lambda_b = 1 is a complete reset. Exactly zero contact on both baths is excluded because it does not select a unique stationary cycle.

With column probability vectors, one full cycle maps

p⟼Mp,M=ChPRCcPF.(29)p \longmapsto Mp,\qquad M = C_hP_RC_cP_F. \tag*{(29)}

The stationary population p∗p_* satisfies Mp∗=p∗Mp_* = p_* and 1Tp∗=1\mathbf{1}^{\mathsf{T}}p_* = 1. The reported critical contact strengths are the values at which the stationary mean work changes sign for the specified paths and this population semigroup.

For the coherence-sensitivity control, each contact instead acts on the density matrix in the corresponding endpoint energy basis as

Tb[ρ]=(1−λb)ρ+λbρβb.(30)\mathcal{T}_b[\rho] = (1-\lambda_b)\rho+ \lambda_b\rho_{\beta_b}. \tag*{(30)}

The unitary strokes are propagated with the same Cayley–Crank–Nicolson amplitudes as the population calculation, and the affine cycle is iterated to its stationary density matrix. Equation (D3) retains the fraction 1−λb1-\lambda_b of pre-contact off-diagonal coherence and serves as a controlled coherence-retaining reference channel. At symmetric λ=0.40\lambda= 0.40 on the extended path it gives W/J=1.894×10−3W/J = 1.894 \times10^{-3} compared with −7.018×10−4-7.018 \times10^{-4} from the population semigroup. On the crossing path the two models give negative work at λ=0.83\lambda= 0.83 and positive work at 0.840.84, bracketing their sign changes within the same interval. Propagator unitarity errors are below 10−1410^{-14} in this control, and the stationary density matrices are iterated until the Frobenius-norm change is below 10−1310^{-13}.

Tilted cycle, long-time cumulants, and entropy

Label a measured population trajectory through one cycle as ih→jc→kc→ℓh→mhi_h \to j_c \to k_c \to\ell_h \to m_h. For a work counting field sws_w, the tilted work-stroke matrices are

[Fsw]ji=PF(j∣i)exp⁡[sw(Eih−Ejc)],(31)[F_{s_w}]_{ji} = P_F(j|i)\exp[s_w(E_i^h-E_j^c)], \tag*{(31)}
[Rsw]ℓk=PR(ℓ∣k)exp⁡[sw(Ekc−Eℓh)].(32)[R_{s_w}]_{\ell k} = P_R(\ell|k)\exp[s_w(E_k^c-E_\ell^h)]. \tag*{(32)}

For heat counting fields scs_c and shs_h, the tilted cold- and hot-contact matrices are

[Csc]kj=[Cc]kjexp⁡[sc(Ekc−Ejc)],(33)[C_{s_c}]_{kj} = [C_c]_{kj}\exp[s_c(E_k^c-E_j^c)], \tag*{(33)}
[Hsh]mℓ=[Ch]mℓexp⁡[sh(Emh−Eℓh)].(34)[H_{s_h}]_{m\ell} = [C_h]_{m\ell}\exp[s_h(E_m^h-E_\ell^h)]. \tag*{(34)}

These fields count the energy changes of the realised contact transitions; in particular, scs_c counts the physical cold-contact heat for jc→kcj_c \to k_c. Using the column-vector convention of Eq. (28), the joint tilted one-cycle operator is

Ts=HshRswCscFsw,s=(sw,sh,sc).(35)T_{\mathbf{s}} = H_{s_h}R_{s_w}C_{s_c}F_{s_w}, \qquad\mathbf{s}=(s_w,s_h,s_c). \tag*{(35)}

The spectral radius of TsT_{\mathbf{s}} gives the long-time scaled cumulant-generating function,

ψ(s)=lim⁡N→∞1Nln⁡⟨eswXN(w)+shXN(h)+scXN(c)⟩=ln⁡ρmax⁡(Ts).(36)\psi(\mathbf{s}) = \lim_{N\to\infty}\frac{1}{N}\ln\left\langle e^{s_w X_N^{(w)}+s_h X_N^{(h)}+s_c X_N^{(c)} }\right\rangle = \ln\rho_{\max}(T_{\mathbf{s}}). \tag*{(36)}

where XN(w)X_N^{(w)}, XN(h)X_N^{(h)}, and XN(c)X_N^{(c)} are the accumulated work and heats over NN cycles and ρmax⁡\rho_{\max} is the largest-modulus eigenvalue. Derivatives at zero fields give the long-time cumulants per cycle. In particular, intercycle covariance terms shift the long-time variance per cycle away from the one-cycle variance whenever memory is appreciable. For the TPM-resolved stationary population trajectory, the mean entropy production is

ΣTPM=−βhQh−βcQc=Σcontact,pop+ΔSwork,deph.(37)\Sigma_{\mathrm{TPM}}=-\beta_h Q_h-\beta_c Q_c=\Sigma_{\mathrm{contact,pop}}+\Delta S_{\mathrm{work,deph}}. \tag*{(37)}

Here Σcontact,pop\Sigma_{\mathrm{contact,pop}} is the contribution generated by population relaxation during the two contacts, and ΔSwork,deph\Delta S_{\mathrm{work,deph}} records the additional entropy change associated with the energy-basis measurements and dephasing used to define the TPM work trajectory. An unmeasured coherent cycle has different entropy accounting because this projective dephasing step is absent. At the L=233L=233 crossing-path grid maximum-power point, the contact and measurement/dephasing contributions are approximately 1.55 and 0.097, giving ΣTPM≃1.65\Sigma_{\mathrm{TPM}}\simeq1.65. For the monitored population process, the Gibbs-renewal contacts obey local detailed balance and the retraced work strokes obey PR=PFP_R=P_F; the resulting entropy generating function satisfies the Gallavotti–Cohen symmetry over the evaluated counting-field range. This fluctuation symmetry refers specifically to the monitored population model; the coherence-retaining channel is treated separately. The joint work and heat statistics also satisfy the cycle first-law identity, with the difference between accumulated work and heat accounted for by the change in system energy at the initial and final cycle boundaries.

Quasiperiodic parameter robustness

The finite-rate sign and fluctuation patterns were examined under changes in the quasiperiodic phase, irrational frequency, rational approximants, and modulation balance. Across the eight phase values examined at the reference size, the principal stochastic pattern varies weakly. Replacing the inverse golden-mean frequency with one alternate irrational frequency preserves the qualitative mean/majority-sign separation at the selected path. Across these controls, the principal finite-rate sign separation varies weakly with phase offset and remains present for the alternate irrational frequency.

Rational Fibonacci approximants to the inverse golden-mean frequency change the finite minimum-gap sequence nonmonotonically, while the TPM observables approach their irrational-frequency values smoothly. The smooth convergence of the cycle statistics follows the Gibbs-weighted transition structure even as the extremal spacing varies nonmonotonically. Varying the irrational frequency can also change endpoint thermodynamics along the fixed path, and changing θ\theta changes the relative diagonal and off-diagonal modulation. Finite-rate accessibility consequently reflects the complete Hamiltonian path through the spectrum.

Temperature, finite-window critical scaling, and return diagnostics

The low-temperature analysis is performed by lowering the hot and cold reservoir temperatures together so that the complete-reset Gibbs states progressively concentrate on the lowest endpoint levels. In the joint ground-state limit the strict energy-ordered cycle returns the occupied branch to itself. For the two finite-rate ramps, the excess final energies above the corresponding energy-ordered branches are denoted δEF\delta E_F and δER\delta E_R. Direct evaluation verifies

Wad−Wν=δEF+δERW_{\mathrm{ad}} - W_\nu= \delta E_F + \delta E_R

to relative error approximately 1.1×10−61.1 \times10^{-6} for the evaluated low-temperature data.

Power-law fits in the accessible low-temperature rate window are treated as effective finite-window diagnostics. At L=233L = 233 and Tc/J=0.006T_c/J = 0.006, the full interval 0.006≤ν/J≤0.0300.006 \le\nu/J \le0.030 gives an exponent 0.48038. Nested four-point windows give 0.40695 on 0.006–0.014, 0.46783 on 0.008–0.020, and 0.52877 on 0.010–0.030; the corresponding five-point fits are 0.44460 and 0.50415. Independent pure-state rate-window scans show the same systematic drift, while finite-size gap fits depend on whether the fixed irrational frequency or Fibonacci rational approximants are used. These trends identify a reproducible preasymptotic scaling regime and provide the quantitative basis for the finite-window Kibble–Zurek connection discussed in the main text.

Thermal broadening changes the effective scaling because more energy branches receive appreciable Gibbs weight. In the stored joint-temperature scan, the fitted work-deficit exponent is 0.480 at Tc/J=0.006T_c/J = 0.006, where relatively few low-energy branches dominate, and rises to 0.875 at the baseline temperatures Tc/J=0.1T_c/J = 0.1 and Th/J=0.65T_h/J = 0.65. At the higher baseline temperatures, the finite-rate response involves the full Gibbs-weighted transition and work statistics across many occupied branches.

For the rapid-driving comparison, the state-frozen reference is used as the sudden-quench endpoint map. The branch-resolved characteristic function develops a pronounced minimum on the crossing path. Across the four-size range, the continuous-ramp characteristic-function minimum saturates while the physical return remains finite and nearly size independent. The sudden reference carries the pronounced Loschmidt/characteristic-function structure, while the finite-rate ramps remain in a nonsingular crossover regime over the evaluated window.

References

  1. [1]S. Aubry and G. André, Analyticity breaking and anderson localization in incommensurate lattices, in Annals of the Israel Physical Society, Vol. 3 (1980) p. 133.
  2. [2]J. Biddle, J. Priour, D. J., B. Wang, and S. Das Sarma, Localization in one-dimensional lattices with non-nearest-neighbor hopping: Generalized anderson and aubry-andré models, Phys. Rev. B 83, 075105 (2011).DOI
  3. [3]S. Ganeshan, J. H. Pixley, and S. Das Sarma, Nearest neighbor tight binding models with an exact mobility edge in one dimension, Phys. Rev. Lett. 114, 146601 (2015).pubmed.ncbi.nlm.nih.gov/25910146
  4. [4]F. Liu, S. Ghosh, and Y. D. Chong, Localization and adiabatic pumping in a generalized aubry-andré-harper model, Phys. Rev. B 91, 014108 (2015).DOI
  5. [5]X. Tong, Y.-M. Meng, X. Jiang, C. Lee, G. D. de Moraes Neto, and X. Gao, Dynamics of a quantum phase transition in the aubry-andré-harper model with p-wave superconductivity, Phys. Rev. B 103, 104202 (2021).DOI
  6. [6]J. Dziarmaga, Dynamics of a quantum phase transition and relaxation to a steady state, Adv. Phys. 59, 1063 (2010).DOI
  7. [7]A. Polkovnikov, K. Sengupta, A. Silva, and M. Vengalattore, Colloquium: Nonequilibrium dynamics of closed interacting quantum systems, Rev. Mod. Phys. 83, 863 (2011).DOI
  8. [8]A. Sinha, M. M. Rams, and J. Dziarmaga, Kibble-zurek mechanism with a single particle: Dynamics of the localization-delocalization transition in the aubry-andré model, Phys. Rev. B 99, 094203 (2019).DOI
  9. [9]C. Yang, Y. Wang, P. Wang, X. Gao, and S. Chen, Dynamical signature of localization-delocalization transition in a one-dimensional incommensurate lattice, Phys. Rev. B 95, 184201 (2017).DOI
  10. [10]M. Heyl, A. Polkovnikov, and S. Kehrein, Dynamical quantum phase transitions in the transverse-field ising model, Phys. Rev. Lett. 110, 135704 (2013).pubmed.ncbi.nlm.nih.gov/23581343
  11. [11]G. Roati, C. D’Errico, L. Fallani, M. Fattori, C. Fort, M. Zaccanti, G. Modugno, M. Modugno, and M. In-guscio, Anderson localization of a non-interacting bose-einstein condensate, Nature 453, 895 (2008).
  12. [12]H. P. Lüschen, S. Scherg, T. Kohlert, M. Schreiber, P. Bordia, X. Li, S. Das Sarma, and I. Bloch, Single-particle mobility edge in a one-dimensional quasiperiodic optical lattice, Phys. Rev. Lett. 120, 160404 (2018).DOI
  13. [13]F. A. An, K. Padavić, E. J. Meier, S. Hegde, S. Ganeshan, J. H. Pixley, S. Vishveshwara, and B. Gadway, Interactions and mobility edges: Observing the generalized aubry-andré model, Phys. Rev. Lett. 126, 040603 (2021).DOI
  14. [14]H. Li, Y.-Y. Wang, Y.-H. Shi, K. Huang, X. Song, G.-H. Liang, Z.-Y. Mei, B. Zhou, H. Zhang, J.-C. Zhang, S. Chen, S. P. Zhao, Y. Tian, Z.-Y. Yang, Z. Xiang, K. Xu, D. Zheng, and H. Fan, Observation of criti- cal phase transition in a generalized aubry-andré-harper model with superconducting circuits, npj Quantum Information 9, 40 (2023).
  15. [15]C. Chiaracane, M. T. Mitchison, A. Purkayastha, G. Haack, and J. Goold, Quasiperiodic quantum heat engines with a mobility edge, Phys. Rev. Research 2, 013093 (2020).
  16. [16]H.-G. Xu and S. Cheng, Quantum transport property and thermodynamic applications of one-dimensional off-diagonal quasiperiodic system, Phys. Rev. B 112, 024204 (2025).DOI
  17. [17]S. Suo, A. Zhou, Y. Chen, S. Cheng, and X. Gao, Wigner distribution, wigner entropy, and quantum refrigerator of a one-dimensional system with off-diagonal and diagonal quasiperiodic modulations, Front. Phys. 21, 125206 (2026).DOI
  18. [18]A. Zhou, F. Lu, S. Cheng, and X. Gao, Thermodynamic modes of a quasiperiodic mobility-edge system in a quantum otto cycle, Phys. Rev. B 114, 034206 (2026).
  19. [19]N. Yunger Halpern, C. D. White, S. Gopalakrishnan, and G. Rafael, Quantum engine based on many-body localization, Phys. Rev. B 99, 024203 (2019).DOI
  20. [20]Y. Rezek and R. Kosloff, Irreversible performance of a quantum harmonic heat engine, New J. Phys. 8, 83 (2006).DOI
  21. [21]S. Lee, M. Ha, J.-M. Park, and H. Jeong, Finite-time quantum otto engine: Surpassing the quasistatic efficiency due to friction, Phys. Rev. E 101, 022127 (2020).DOI
  22. [22]G. Jiao, S. Zhu, J. He, Y. Ma, and J. Wang, Fluctuations in irreversible quantum otto engines, Phys. Rev. E 103, 032130 (2021).DOI
  23. [23]T. McKeever, O. Diba, and A. Nazir, Finite-time quantum otto engine subject to control noise and enhancement techniques, Phys. Rev. Research 7, 033172 (2025).
  24. [24]Revathy B. S., V. Mukherjee, U. Divakaran, and A. del Campo, Universal finite-time thermodynamics of many-body quantum machines from kibble-zurek scaling, Phys. Rev. Research 2, 043247 (2020).DOI
  25. [25]V. R. Arezzo, D. Rossini, and G. Piccitto, Many-body quantum heat engines based on free fermion systems, Phys. Rev. B 109, 224309 (2024).DOI
  26. [26]H.-G. Xu, J. Jin, N. G. de Almeida, and G. D. de Moraes Neto, Exploring the role of criticality in the quantum otto cycle fueled by the anisotropic quantum rabi-star model, Phys. Rev. B 110, 134318 (2024).arxiv.org/abs/2407.09027
  27. [27]T. Denzler, J. F. G. Santos, E. Lutz, and R. M. Serra, Nonequilibrium fluctuations of a quantum heat engine, Quantum Sci. Technol. 9, 045017 (2024).arxiv.org/abs/2104.13427
  28. [28]T. Denzler and E. Lutz, Efficiency fluctuations of a quantum heat engine, Physical Review Research 2, 032062 (2020).arxiv.org/abs/2104.13427
  29. [29]M. F. Anka, T. R. de Oliveira, and D. Jonathan, Work and efficiency fluctuations in a quantum otto cycle with idle levels, Phys. Rev. E 109, 064129 (2024).DOI
  30. [30]V. Cavina, A. Mari, and V. Giovannetti, Optimal processes for probabilistic work extraction beyond the second law, Sci. Rep. 6, 29282 (2016).DOI
  31. [31]H. J. D. Miller and M. Mehboudi, Geometry of work fluctuations versus efficiency in microscopic thermal machines, Phys. Rev. Lett. 125, 260602 (2020).DOI
  32. [32]R. Shastri and B. P. Venkatesh, Optimization of asymmetric quantum otto engine cycles, Phys. Rev. E 106, 024123 (2022).DOI
  33. [33]N. Jaseem, S. Vinjanampathy, and V. Mukherjee, Quadratic enhancement in the reliability of collective quantum engines, Phys. Rev. A 107, L040202 (2023).
  34. [34]G. G. Damas, N. G. de Almeida, X. Gao, and G. D. de Moraes Neto, Coherence-preserving fluctuation diagnostics for an engineered population-inverted qubit otto engine, Adv. Quantum Technol. 9, e70412 (2026).
  35. [35]P. Pietzonka and U. Seifert, Universal trade-off between power, efficiency, and constancy in steady-state heat engines, Phys. Rev. Lett. 120, 190602 (2018).DOI
  36. [36]H. J. D. Miller, M. H. Mohammady, M. Perarnau-Llobet, and G. Guarnieri, Thermodynamic uncertainty relation in slowly driven quantum heat engines, Phys. Rev. Lett. 126, 210603 (2021).DOI
  37. [37]M. Esposito, U. Harbola, and S. Mukamel, Nonequilibrium fluctuations, fluctuation theorems, and counting statistics in quantum systems, Rev. Mod. Phys. 81, 1665 (2009).DOI
  38. [38]S. Mohanta and B. K. Agarwalla, Full statistics of nonequilibrium heat and work for many-body quantum otto engines and universal bounds: A nonequilibrium green’s function approach, Phys. Rev. E 108, 064127 (2023).DOI
  39. [39]Z. Ma, A. K. Mitchell, and E. Sela, Quantum work statistics across a critical point: Full crossover from sudden quench to the adiabatic limit, Phys. Rev. Lett. 135, 130402 (2025).
  40. [40]A. Kiely, E. O’Connor, T. Fogarty, G. T. Landi, and S. Campbell, Entropy of the quantum work distribution, Phys. Rev. Research 5, L022010 (2023).
  41. [41]G. Watanabe, B. P. Venkatesh, P. Talkner, and A. del Campo, Quantum performance of thermal machines over many cycles, Phys. Rev. Lett. 118, 050601 (2017).DOI
  42. [42]F. Lu, A. Zhou, S. Cheng, and X. Gao, Wigner distribution, wigner entropy, and anomalous transport of a generalized aubry-andré model, Phys. Rev. B 112, 174206 (2025).
  43. [43]P. Talkner, E. Lutz, and P. Hänggi, Fluctuation theorems: Work is not an observable, Phys. Rev. E 75, 050102(R) (2007).DOI
  44. [44]M. Campisi, P. Hänggi, and P. Talkner, Colloquium: Quantum fluctuation relations: Foundations and applications, Rev. Mod. Phys. 83, 771 (2011).DOI
  45. [45]J. Son, P. Talkner, and J. Thingna, Monitoring quantum otto engines, PRX Quantum 2, 040328 (2021).DOI
  46. [46]C. Zener, Non-adiabatic crossing of energy levels, Proceedings of the Royal Society of London. Series A 137, 696 (1932).DOI
  47. [47]G. G. Damas, C. Cruz, N. G. de Almeida, X. Gao, and G. D. de Moraes Neto, Finite-size reliability of homothetic quantum otto engines (2026), arXiv:2607.29050 [quant-ph].arxiv.org/abs/2607.29050
  48. [48]M. Heyl, Dynamical quantum phase transitions: a review, Rep. Prog. Phys. 81, 054001 (2018).DOI
  49. [49]N. A. Khan, S. Ye, Z. Zhou, S. Cheng, and X. Gao, Chebyshev polynomial approach to loschmidt echo: Application to quench dynamics in two-dimensional quasicrystals, Phys. Rev. E 109, 065311 (2024).DOI
  50. [50]C. T. Garcez, G. G. Damas, N. G. de Almeida, and G. D. de Moraes Neto, Collective thermalization, work reliability, and resource bounds in a population-inverted dicke otto engine (2026), arXiv:2609.30132 [quant-ph].arxiv.org/abs/2609.30132
  51. [51]M. B. Sambú, T. R. de Oliveira, and L. C. Céléri, Gauge-invariant thermodynamics of a finite-time quantum otto engine (2026) arXiv:2607.09857 [quant-ph].
  52. [52]M. Kolodrubetz, D. Sels, P. Mehta, and A. Polkovnikov, Geometry and non-adiabatic response in quantum and classical systems, Phys. Rep. 697, 1 (2017).DOI

Paper details

vlnc:W_yc4pbksx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents