Carnot efficiency \(\eta_C=1-T_c/T_h\) is achievable only in the reversible, infinitely-slow limit — where the engine produces no power at all. Curzon & Ahlborn asked the practical question instead: for an engine that exchanges heat at finite rate with its reservoirs, what is the efficiency at maximum power?

Modelling an endoreversible engine — reversible in its working substance but leaking heat irreversibly through finite-conductance contacts — they found a strikingly simple answer:

It lies below Carnot, depends only on the reservoir temperatures, and matched real power-plant efficiencies remarkably well. (Novikov 1958 and Chambadal 1957 had precursor results.)

This is the paper that named the tradeoff at the heart of every finite-time process: speed costs efficiency, and there is an optimum. Everything downstream — endoreversible engine design, stochastic and quantum heat engines, the experiments — is a descendant of the move from "how efficient?" to "how well, in finite time?"

For seventy years, driving a system in finite time meant only that you dissipate: \(\langle W\rangle\ge\Delta F\). Jarzynski found that the fluctuating work \(W\) over many finite-time runs satisfies an exact equality, valid no matter how hard or fast you drive.

The first (Jarzynski) recovers the equilibrium free-energy difference from irreversible work; the second (Crooks) is a forward/reverse symmetry from which it follows. The old Second-Law inequality re-emerges as an average (by Jensen's inequality) of these exact relations.

These theorems made finite-time, irreversible processes exactly quantifiable — you can read equilibrium free energies off irreversible pulling experiments. They are the engine behind modern stochastic thermodynamics (Seifert), single-molecule experiments, and the quantum work-statistics measurements on the map's experimental frontier.

If dissipation is the price of haste, how do you pay the least? Salamon & Berry showed that thermodynamic state space carries a natural Riemannian geometry: a thermodynamic length \(\mathcal L\) measures the distance of a path through equilibrium states.

For a process of finite duration \(\tau\), the dissipated availability is bounded below by the squared length over the time:

So the minimum-dissipation protocol is the geodesic traversed at constant thermodynamic speed — turning optimal driving into a well-posed geometry problem.

This is the root of the modern theory of optimal finite-time protocols — Crooks (2007), Sivak & Crooks (2012, the friction tensor), and the counterdiabatic "shortcuts to adiabaticity" that beat the tradeoff on quantum hardware. The 1984 Andresen–Salamon–Berry review then named and codified the whole field.

Engines & efficiency (power vs efficiency) · Work & fluctuation relations (nonequilibrium driving) · Optimal protocols & geometry (minimal dissipation) · Information & erasure (Landauer, Maxwell's demon, thermodynamic-uncertainty and speed-limit bounds).

Classical / endoreversible FTT and thermodynamic geometry · Stochastic thermodynamics (heat and work along single fluctuating trajectories) · Quantum & control (quantum heat engines, shortcuts to adiabaticity, quantum speed limits).

theory analysis / proof only   classical exp. a table-top (colloidal or single-electron) demonstration   quantum exp. a real single-atom / ion / NMR / NV experiment. The pattern: FTT is overwhelmingly theory, with a growing — and increasingly quantum — experimental frontier.

Carnot efficiency \(\eta_C=1-T_c/T_h\) is an upper bound reached only in the quasistatic limit, where the cycle takes infinite time and therefore produces zero power. That is useless as a design target. The authors ask the operationally honest question: hold the working substance internally reversible, but let it exchange heat with the reservoirs at a finite rate, and maximize the power rather than the efficiency.

They model an endoreversible engine — reversible inside, with all irreversibility confined to the two heat-exchange links, each obeying a Newtonian (linear) heat law \(\dot Q = \kappa\,\Delta T\). The working fluid sits at internal temperatures \(T_{hw}\lt T_h\) and \(T_{cw}\gt T_c\); the temperature drops across the couplings are the knobs. Maximizing the power output over those drops gives, remarkably, a result independent of the conductances:

For a plant running between roughly 300 K and 850 K this predicts about 40%, close to observed values, whereas \(\eta_C\) would promise ~65%.

The square-root form proved unreasonably universal: it reappears in the small-dissipation and linear-irreversible limits of far more general models, and the search to explain why drove decades of work. This single result defined the central FTT premise — that time is a thermodynamic resource — and every optimal-protocol, thermodynamic-length, and stochastic-engine paper that follows is, in one sense, a descendant of it.

Chambadal was a practising power engineer, not a thermodynamics theorist looking for elegance. Faced with a reactor coupled to a turbine through a finite heat exchanger, he needed to know the efficiency an actual plant could sustain while delivering useful output — the same finite-rate constraint C&A would later formalize.

He modelled heat delivered from a fluid of finite thermal capacity cooling from \(T_h\) to an intermediate temperature \(T\), with the engine converting the extracted heat reversibly to a sink at \(T_c\). Maximizing the work per unit throughput over the exhaust temperature \(T\) yields the optimum at \(T=\sqrt{T_h T_c}\), and hence

the identical square-root law, obtained from a finite-reservoir rather than a finite-conductance argument.

Chambadal, together with Novikov, establishes that the maximum-power efficiency is not a quirk of the Newtonian-coupling model but a robust feature of finite-rate heat conversion. The FTT community now often writes "Chambadal–Novikov–Curzon–Ahlborn" to honour the buried lineage — a reminder that the field's founding result was discovered independently three times.

Like Chambadal, Novikov was concerned with the practical efficiency of nuclear stations whose reactor-to-turbine coupling limits the rate of heat delivery. The Carnot value was known to be unreachable at finite output; the useful question was the efficiency at the operating point that actually maximizes generated power.

Novikov's model places all the loss in a single thermal resistance between the hot source and an internally reversible engine, \(\dot Q_h=\kappa(T_h-T_{hw})\), with a perfect cold-side coupling. Optimizing the delivered power over the internal hot temperature \(T_{hw}\) again lands on

That the same expression emerges whether the bottleneck is one resistance (Novikov), a finite coolant (Chambadal), or two couplings (C&A) is the whole point: the square root is intrinsic to trading rate against reversibility.

Novikov's paper is the third independent arrival at the maximum-power efficiency and the one most cited in the Russian-language tradition. Collectively these precursors show FTT's cornerstone was "in the air" among mid-century power engineers, waiting for the Berry group to give the subject its name and its variational apparatus.

Before this paper, finite-time efficiency losses were treated case by case. The authors' ambition was structural: to build a systematic thermodynamics in which the duration of each process step is an explicit variable, and in which entropy production is computed from the actual paths rather than assumed away.

They analyse a step-Carnot cycle — a Carnot loop broken into discrete finite-time segments of heat exchange and adjustment — and account for the entropy generated in each segment through the finite-rate heat laws. The lost work over a cycle is the sum of the segment dissipations,

which exposes how the allocation of a fixed total cycle time among the steps controls the total dissipation — the seed of the optimal-control viewpoint that follows.

This is paper I of a long series; it supplies the vocabulary — "thermodynamics in finite time" — and the discretized scaffold on which the group's later variational and geometric results are built. It reframes Curzon–Ahlborn from an isolated formula into the first data point of a systematic theory.

An engine with friction, finite heat conduction, and other losses has infinitely many ways to traverse a cycle in a given time. The design question is a functional optimization: over all admissible time-paths of the control variables, which one maximizes net work (or minimizes entropy production)?

They cast the total entropy produced as a functional of the control path \(x(t)\) and its rate \(\dot x(t)\),

and impose the Euler–Lagrange extremal condition to find the optimal schedule. The resulting extremals identify how fast to move in each phase of the cycle — slowing where dissipation is steep, hurrying where it is cheap — the concrete statement that timing is a design variable.

This paper imports the machinery of optimal control (later sharpened to Pontryagin's principle by Rubin) into thermodynamics. It defines the optimal-protocol lineage — Rubin, Salamon–Nitzan, Band–Kafri–Salamon — and, further downstream, the Schmiedl–Seifert optimal protocols of stochastic thermodynamics.

Curzon–Ahlborn fixed the cycle structure and optimized two temperature drops. Rubin asks a broader question: among all operating protocols of an engine whose only loss is finite-rate heat conduction, which is genuinely optimal? The freedom includes when to contact each reservoir and how to time the adiabats.

He formulates power as a control functional with the working-fluid temperature as the state and the reservoir contact as the control, then applies Pontryagin's maximum principle to extract the optimal switching structure. A central finding is that the optimal cycle is of bang-bang / constant-contact-temperature type,

and that under pure conductive loss the optimal engine reproduces the Curzon–Ahlborn efficiency at maximum power — now derived, not assumed.

Rubin's "Part I" (with sequels) established that FTT optimization is properly a problem in optimal control theory, and that the maximum-power efficiency is the output of a rigorous optimization when conduction dominates. It set the methodological standard for the endoreversible and optimal-control branches alike.

Different FTT criteria — maximum work, maximum power, minimum entropy production — could in principle disagree. The authors clarify the relationship and, crucially, characterize the shape of the optimal protocol: not just its value, but how the dissipation should be distributed in time.

For a process of fixed duration \(\tau\) the lost work is \(W_{\rm lost}=T_0\,\Delta S_{\rm gen}\), so minimizing one minimizes the other. Writing the entropy production as a time integral of a rate and applying the variational condition, the optimum satisfies an equipartition principle — the entropy-production rate is constant along the optimal path:

Dissipation should be spread evenly, never bunched — go slow through stiff regions, fast through soft ones, so that the instantaneous "friction" stays level.

Constant-rate (equipartition) optimality is the operational heart of thermodynamic geometry: it is exactly the statement that optimal protocols traverse thermodynamic length at uniform speed. This paper is the conceptual bridge from Berry-group optimal control to the Salamon–Berry length theorem three years later.

A finite hot body at initial temperature \(T_h\) delivering heat to a sink at \(T_c\) has a definite maximum extractable work — the exergy — but only if equilibrated reversibly. The realistic question is how to sequence engines so as to approach that ceiling, and what happens under finite-time constraints on the exchange.

Each infinitesimal Carnot cycle draws heat from the source at its current temperature \(T\), does reversible work, and rejects to \(T_c\); the source cools according to its heat capacity. Integrating the reversible work as the source relaxes from \(T_h\) to \(T_c\) gives the exergy of the finite reservoir,

with the well-known ideal result \(W_{\max}=C\big[(T_h-T_c)-T_c\ln(T_h/T_c)\big]\) for constant \(C\). The paper then studies how finite-time heat exchange degrades this ceiling.

This extends FTT from steady reservoirs to depletable sources — the relevant case for batteries, thermal storage, and any finite hot stream. The sequential-Carnot construction is a recurring FTT idiom, and the finite-reservoir exergy it computes is the reference against which real finite-time extraction is judged.

Move the piston too fast and the gas cannot keep up — irreversible pressure lags and dissipation grow. Move it too slowly and you waste the available time, doing less work than the schedule allows. Between these lies an optimal time-dependent volume \(V(t)\) that the authors seek subject to fixed endpoints and duration.

With the gas obeying an equation of state and a finite heat-transfer law, the net work is a functional of the piston path,

and the optimal expansion satisfies the corresponding Euler–Lagrange equation. The solution again displays the equipartition character — the optimal path keeps the dissipative penalty balanced through the stroke rather than loading it at one end.

This is FTT optimal control made tangible on the textbook piston-and-gas system, and it anticipates precisely the kind of driving-protocol optimization that stochastic thermodynamics would later perform for a colloid in an optical trap. It is the classical ancestor of the Schmiedl–Seifert minimum-work protocol.

The Gibbsian relations among extensive and intensive variables have an algebraic feel but no obvious geometry. Weinhold asked whether the stability conditions — the convexity of the energy surface — could be read as a genuine metric, turning thermodynamic reasoning into geometry.

He defines an inner product on the tangent space of thermodynamic states from the Hessian of the internal energy with respect to the extensive variables \(X^i\) (entropy, volume, mole numbers):

Thermodynamic stability makes this Hessian positive definite, so \(g^W\) is a legitimate Riemannian metric. Distances and angles on the state manifold now carry thermodynamic meaning, and the symmetry of the metric, \(g^W_{ij}=g^W_{ji}\) (equality of the mixed second derivatives of \(U\)), is exactly the statement of the Maxwell relations.

The Weinhold metric (energy representation) is one of the two founding thermodynamic metrics — the other being Ruppeiner's entropy-representation metric, conformally related to it. On its own it looked like elegant formalism; its physical payoff arrived when Salamon & Berry showed that Weinhold length bounds finite-time dissipation, wiring geometry directly into FTT.

Weinhold's metric assigns a length \(\mathcal L\) to any path between two equilibrium states. But length of what, physically? The authors sought a theorem connecting that geometric quantity to the actual thermodynamic cost of traversing the path at finite speed.

Writing the dissipated availability as a time integral of a metric quadratic form and applying the Cauchy–Schwarz inequality against a path of duration \(\tau\) yields

with equality when the path is traversed at constant thermodynamic speed — precisely the equipartition condition of Salamon–Nitzan. The optimal protocol is therefore a geodesic of the thermodynamic metric, covered at uniform rate, and its minimum cost scales as \(1/\tau\).

This inequality is the load-bearing result of thermodynamic geometry. It reappears — via Fisher information (Crooks 2007) and the friction tensor (Sivak–Crooks 2012) — throughout stochastic thermodynamics, and it turns "design the best protocol" into "find the shortest path" on a curved state space. It is the geometric half of FTT's foundation.

By 1984 the pieces — Curzon–Ahlborn efficiency, Berry-group extremals, Weinhold–Salamon geometry — existed but were spread across specialist journals. The authors' aim was synthesis: to articulate the guiding question (what are the real limits when time is finite?) and the shared toolkit that answers it.

The review organizes the field around endoreversibility (confine irreversibility to identifiable couplings), optimal control (time-schedule the process to minimize dissipation), and thermodynamic geometry (dissipation as squared length over time, \(\Delta A_{\rm diss}\ge \mathcal L^2/\tau\)). It stresses the central lesson that realistic bounds sit strictly between the reversible ideal and unconstrained operation, and that haste has a computable price.

This is the citation newcomers reach for and the document that fixed the field's identity and name in the broader literature. It set the agenda that the stochastic-thermodynamics generation (Jarzynski, Crooks, Seifert) would inherit and extend to the fluctuating, single-molecule scale.

Gaussian fluctuation theory assigns a probability to a deviation from equilibrium via the entropy cost of that deviation. Ruppeiner asks whether the covariance structure of these fluctuations defines a metric — and if so, what its intrinsic curvature encodes about the underlying physics.

He builds the metric from the second derivatives of the entropy (the entropy representation), which governs the fluctuation probability \(P\propto e^{\Delta S/k_B}\):

a metric conformally related to Weinhold's. Its scalar curvature \(R\) then has dimensions of volume and, empirically and by scaling arguments, behaves like a correlation volume: \(|R|\) grows with interaction strength, changes sign with the character of the interactions, and diverges at criticality as the correlation length does.

This review made thermodynamic curvature a diagnostic tool — used since for phase transitions, black-hole thermodynamics, and interacting gases. Within FTT it deepens the geometric picture beyond distance (which bounds dissipation) to curvature (which reports on the microphysics), and it is the standard reference for the entropy-metric side of thermodynamic geometry.

Salamon–Berry length is defined through equilibrium response functions that are not always easy to access. Crooks reformulates it in a way that is both computable from data and conceptually sharper, using the statistical geometry of the equilibrium probability distributions themselves.

Parametrizing a family of equilibrium (Gibbs) distributions by control variables \(\lambda\), the natural metric is the Fisher information \(g_{ij}(\lambda)=\big\langle \partial_i \ell\,\partial_j \ell\big\rangle\) of the log-likelihood \(\ell=\ln p\). The thermodynamic length is its geodesic distance,

and the divergence between endpoint distributions bounds the dissipation, refining \(\langle W_{\rm diss}\rangle\gtrsim \mathcal L^2/\tau\). Length becomes the number of statistically distinguishable intermediate states along the path — a quantity estimable from work measurements.

By welding thermodynamic geometry to Fisher information, Crooks unifies the classical FTT length with information geometry and free-energy estimation (Jarzynski/Crooks work relations). This is the pivot connecting the classical geometric pillar to the stochastic branch, and it directly enables the Sivak–Crooks friction-tensor construction.

Earlier length metrics were equilibrium constructs. To design an actual protocol \(\lambda(t)\) for driving a system (a colloid, a molecule) you need a metric that knows about the system's relaxation dynamics — how sluggishly it follows a moving control. Sivak & Crooks supply exactly that in the linear-response regime.

To leading order in the driving rate, the excess (dissipated) work is a quadratic form in the protocol velocity, with a metric equal to the integrated force–force correlation function:

This \(\zeta_{ij}\) is the friction tensor: it combines the thermodynamic (equilibrium) metric with the local relaxation time. Minimizing the dissipation is then a geodesic problem, and the optimal protocol traverses the friction-metric length at constant speed — moving slowly where the system is stiff or slow, quickly where it is soft or fast.

This is the practical culmination of the length program: a computable, dynamics-aware recipe for optimal driving that has been applied to protocol design in single-molecule pulling, bit erasure, and nonequilibrium free-energy estimation. It closes the loop from Weinhold's abstract metric to protocols one can actually run in a lab.

Written for a chemistry audience, the review answers what FTT has actually delivered — beyond the Curzon–Ahlborn headline — and where the live research fronts lie, from engine and separation-process optimization to the geometric bounds and their experimental tests.

Andresen threads the results through a single quantity — the excess dissipation incurred by finite speed — and the tools that bound it: endoreversible modelling, optimal-control scheduling, and the thermodynamic-length inequality

with its equipartition (constant-speed) prescription for the optimal protocol. The recurring message is that dissipation should be spread evenly over the process, and that geometry tells you how.

As an accessible, up-to-date entry point it functions as the field's second-generation review, explicitly gesturing toward the stochastic-thermodynamics extensions (fluctuation theorems, optimal protocols at the nanoscale) that carry the finite-time program into single-molecule and colloidal physics.

The second law says the average work to switch a control parameter obeys \(\langle W\rangle\ge\Delta F\), with equality only for infinitely slow (reversible) driving. It was assumed that fast driving therefore destroys access to \(\Delta F\). Jarzynski asked whether a cleverer average could still extract it.

Averaging \(e^{-\beta W}\) over the ensemble of realizations of a finite-time protocol — each starting from equilibrium at temperature \(T=1/(k_B\beta)\) — gives, exactly,

Jensen's inequality applied to this equality immediately returns the second law \(\langle W\rangle\ge\Delta F\), so the relation is a strict strengthening of it. The rare trajectories with \(W\lt\Delta F\) — apparent second-law "violations" — are exactly what the exponential average weights, which is why they must occur with nonzero probability.

The equality turned single-molecule pulling experiments into free-energy meters: repeat a fast pull many times, exponentially average the work, read off \(\Delta F\). It founded the fluctuation-theorem branch of stochastic thermodynamics and is the direct parent of Crooks 1999, which explains why it holds.

Jarzynski constrains one moment of the work distribution. Crooks asked for the complete relationship between the forward and reverse work distributions, which encodes the microscopic time-reversal asymmetry responsible for the arrow of time.

For a protocol run forward and its time-reverse, the work distributions satisfy

Two consequences are immediate. The distributions cross exactly at \(W=\Delta F\), giving a robust, average-free estimator of the free energy from experimental histograms. And multiplying through by \(e^{-\beta W}\) and integrating recovers Jarzynski's \(\langle e^{-\beta W}\rangle=e^{-\beta\Delta F}\) — so Crooks contains Jarzynski. The relation is the work-level face of the general entropy-production fluctuation theorem.

The crossing-point estimator made Crooks the workhorse for extracting free energies from noisy single-molecule data (famously RNA-hairpin unfolding). Together with Jarzynski it anchors the fluctuation-theorem pillar and generalizes to the trajectory entropy production formalized by Seifert.

Classical thermodynamics deals in ensemble averages. For a micron-scale colloid buffeted by thermal noise, work and heat fluctuate wildly from run to run. Sekimoto sought definitions of these quantities for an individual stochastic path, consistent with the Langevin dynamics.

For an overdamped particle in potential \(U(x,\lambda)\), \(\gamma\dot x=-\partial_x U+\xi\), he identifies the heat transferred from the bath as the work done by the friction and noise forces along the path, and the work as the energy injected by changing the control:

with the Stratonovich product \(\circ\) making the chain rule (and hence energy balance) hold. These combine into the trajectory-wise first law \(dU=d\!\!\;W+d\!\!\;Q\). Heat is now a functional of the individual path, not merely an ensemble average.

Stochastic energetics is the definitional bedrock beneath every fluctuation theorem: to state Jarzynski, Crooks, or Seifert one must first say what \(W\) and \(Q\) are along a trajectory, and this is where that is done. It makes the entire microscopic-engine and molecular-machine literature possible.

Sekimoto gave heat and work per path; the medium's entropy change follows from the heat. But what is the system's entropy on a single noisy trajectory, where there is no ensemble to average over at that instant? Without it, no trajectory-level second law can be written.

Seifert defines the trajectory system entropy from the instantaneous ensemble density \(p(x,t)\) evaluated along the path, \(s(t)=-k_B\ln p\big(x(t),t\big)\). Adding the medium entropy \(\Delta s_m=Q/T\) gives a total whose average increases (the second law) and which satisfies the exact integral fluctuation theorem

This holds for any initial state and any protocol, of any speed — the most general finite-time second law, with the inequality again following by Jensen's inequality from the exact equality.

This paper unifies Jarzynski, Crooks, and the earlier steady-state fluctuation theorems as special cases of one statement about total entropy production, and it establishes the definitions used throughout stochastic thermodynamics. It is the theoretical spine of Seifert's 2012 review and of the whole modern edifice.

Between 1997 and 2012 the ingredients — stochastic energetics, work relations, fluctuation theorems, optimal protocols — accumulated rapidly and somewhat piecemeal. The review's task was to present them as one structured subject with shared definitions, from the Langevin and master-equation levels up.

It builds everything on the trajectory-wise first law and the total entropy production, from which the fluctuation theorems follow as the master identities:

with Jarzynski and Crooks recovered as corollaries. The review then works through driven Brownian motion, molecular motors as biochemical networks, feedback and information, and the efficiency of small heat engines — the physics that motivates the Blickle and Martínez experiments in this corpus.

It is the most-cited synthesis of the field and the entry point most researchers use. By packaging trajectory thermodynamics as a mature theory it set the stage for the uncertainty-relation and speed-limit results (Barato–Seifert, Shiraishi) that sharpened the tradeoffs it catalogued.

Move an optically trapped particle (or stiffen its trap) from one state to another in fixed time \(\tau\). The mean work exceeds \(\Delta F\) by an excess that depends on the entire schedule \(\lambda(t)\). Which schedule makes that excess as small as possible?

For an overdamped particle the excess work is a functional of the protocol; minimizing it (Euler–Lagrange / optimal-control) yields a smooth interior motion — for a moving harmonic trap, the optimal center moves at constant velocity — bracketed by finite discontinuous jumps at \(t=0\) and \(t=\tau\). The minimal excess work scales as

recovering the classical \(1/\tau\) finite-time law, but with the counterintuitive feature that it is optimal to "waste" a jump at each end rather than start and stop smoothly. The lag between the particle and the trap is what these jumps exploit.

This is the microscopic descendant of Band–Kafri–Salamon's optimal piston expansion, now for a fluctuating particle, and it made optimal-protocol design a concrete experimental target. It set up the transport (Aurell) and geometric (Sivak–Crooks) reformulations, and its jumps have since been observed in colloidal experiments.

Schmiedl–Seifert solved specific traps analytically, but general anharmonic potentials and higher dimensions resisted case-by-case treatment. A structural principle was needed to characterize and compute the optimum in generality.

They show the mean dissipated work of driving the probability density from \(\rho_0\) to \(\rho_\tau\) in time \(\tau\) equals the minimal kinetic cost of transporting the mass between those densities — the quadratic Wasserstein distance:

with the optimal density current following the deterministic Burgers/optimal-transport flow. The problem of "design the best protocol" is thereby converted into "solve an optimal-transport problem," for which powerful numerical and analytical tools already exist.

This transport connection is one of the most fruitful in stochastic thermodynamics, exporting a mature branch of mathematics into protocol optimization. It underpins Aurell's 2012 refined second law and links directly to the thermodynamic-length picture, since Wasserstein distance is the transport analogue of \(\mathcal L\).

The bare second law is nearly vacuous for fast driving — it only says dissipation is non-negative. For rapid random processes one wants a quantitative floor: given the endpoint distributions and the time, how small can the dissipation possibly be?

Using the transport formulation, the dissipated work is bounded below by the squared quadratic-Wasserstein distance between initial and final densities, divided by the time:

saturated by the optimal-transport protocol of the 2011 paper. This is a genuine refinement: it is strictly stronger than \(\langle W_{\rm diss}\rangle\ge 0\), it is finite and explicit, and it makes plain the universal \(1/\tau\) cost of haste — the stochastic analogue of the classical \(\mathcal L^2/\tau\) length bound.

The refined second law is the transport counterpart of Salamon–Berry's thermodynamic-length inequality, and it gives optimal-protocol theory a hard performance guarantee. It is a cornerstone of the modern "geometry-and-transport" view of finite-time dissipation.

Landauer asked whether the manipulation of information is inherently dissipative. His key distinction is between logically reversible operations (which need not dissipate) and logically irreversible ones like erasure, which merge two distinct states into one and thus destroy information.

Resetting a bit from an unknown value to a definite "0" halves the accessible state space, reducing the memory's entropy by \(k_B\ln 2\). By the second law that entropy must be exported to the environment as heat, so the minimum heat dissipated and work required is

This is the Landauer bound: the irreducible thermodynamic cost of a logically irreversible bit operation, independent of implementation.

Landauer's principle is the founding link between information theory and thermodynamics, resolving Maxwell's demon (the demon must pay to erase its records) and setting the ultimate energy floor for computation. Within FTT it becomes a finite-time question — Proesmans 2020 asks what erasing in time \(\tau\) costs above \(k_B T\ln 2\) — and it was confirmed experimentally in colloidal-bit erasure setups.

Real memories must be reset quickly. The reversible \(k_B T\ln 2\) is reached only as \(\tau\to\infty\); for finite \(\tau\) there is an unavoidable excess. The question is the exact form of that excess and the optimal protocol that minimizes it.

Casting bit erasure as driving a double-well density to a single-well density, they use the optimal-transport bound to obtain the minimal work as the Landauer floor plus a term set by the (squared Wasserstein) transport length over the time:

i.e. the extra cost falls off as \(1/\tau\) and is governed by the same geometry-of-transport that controls all finite-time dissipation. They give the explicit optimal erasure protocol that saturates it.

This is Landauer's principle upgraded to a finite-time law, directly relevant to the energy budget of real, fast-switching computing hardware. It is the information-erasure instance of the Aurell refined second law, and it has guided finite-time-erasure experiments on colloidal and nanomagnetic bits.

Biomolecular processes — a motor stepping, a clock ticking — run in noisy steady states where output currents fluctuate. Is there a fundamental limit on how reproducible such a current can be, and what does it cost to achieve high reproducibility?

For an integrated current \(J\) accumulated over time in a steady state, the relative squared uncertainty is bounded below by the total entropy production \(\Sigma\) (in units of \(k_B\)):

Halving the relative fluctuations requires at least quadrupling the dissipation. Since \(\Sigma\) grows with the time of operation, sustained precision demands a steady entropy cost — an iron tradeoff between accuracy and energy that no design can evade.

The thermodynamic uncertainty relation (TUR) became one of the most active areas of stochastic thermodynamics, spawning a whole family of related bounds and serving as a model-free thermometer of dissipation: measure a current's fluctuations and you lower-bound the entropy production. It sharpened the qualitative tradeoffs catalogued in Seifert's 2012 review into a quantitative inequality.

Quantum mechanics has Mandelstam–Tamm / Margolus–Levitin speed limits bounding how fast a state can evolve. The authors asked for the classical, thermodynamic counterpart: given the entropy produced and the dynamics' overall activity, what is the minimum time to move between two probability distributions?

They prove that the time to evolve between distributions is bounded below by a statistical distance divided by the geometric mean of the entropy production rate and the dynamical activity \(\mathcal A\) (the mean number of jumps / total rate of transitions):

where \(\mathcal L\) is the total variation (or similar) distance between the endpoints. The crucial new ingredient beyond the TUR is dynamical activity: even a dissipationless but fast-shuffling process is time-limited by how much microscopic traffic it can muster.

This classical speed limit complements the thermodynamic uncertainty relation — one bounds precision by dissipation, the other bounds speed by dissipation and activity. Together they define the modern "cost of going fast" toolkit, the stochastic-era sharpening of the classical \(1/\tau\) finite-time laws.

Textbook engines are macroscopic and deterministic. When the working substance is a single Brownian particle, work and heat per cycle become random variables comparable to \(k_B T\). The experiment tests whether the concepts of work, heat, and efficiency still make operational sense there — the physical embodiment of stochastic energetics.

A single ~micron melamine sphere in water is held in an optical trap whose stiffness is modulated in time to supply the two isothermal "compression/expansion" strokes; the surrounding water's temperature is switched (via laser heating) to provide the two isochoric temperature changes — a Stirling cycle for one particle. Tracking the bead's position gives the work per cycle directly from stochastic energetics,

They recorded the distribution of work over many cycles (not just its mean), the mean power output, and the efficiency, confirming that averaged quantities obey the expected thermodynamics while individual cycles fluctuate — and can even transiently run "backwards."

This is the experimental proof that finite-time, fluctuating heat engines are real and measurable, validating the stochastic-energetics framework of Sekimoto and Seifert on hardware. It opened the microscopic-engine field and set the stage for the Brownian Carnot engine of Martínez et al.

Realizing a true Carnot cycle — two isotherms and two adiabats — at the colloidal scale is hard because "adiabatic" (no heat exchange) is difficult when the particle is permanently coupled to a bath. The experiment asks whether the Carnot construction, and its efficiency, survive in the fluctuating regime.

A single optically trapped microsphere serves as the working substance. The two bath temperatures are engineered by adding an external random (electric/noisy) force that inflates the particle's kinetic temperature — letting a "hot" reservoir far above room temperature be synthesized on demand — while the trap stiffness is varied to form the strokes. Measuring position and applied forces yields per-cycle work, heat, and the stochastic efficiency \(\eta=-W/Q_h\). The ensemble-average efficiency obeys the Carnot bound,

yet the single-cycle efficiency has a broad distribution whose tail momentarily surpasses \(\eta_C\) — a fluctuation effect, immediately reconciled with thermodynamics because it is only the ensemble average that is bounded.

This experiment demonstrated finite-time Carnot operation and the fluctuating nature of efficiency at the nanoscale, directly probing the interface between the classical Carnot/Curzon–Ahlborn limits and stochastic thermodynamics. Together with Blickle–Bechinger it anchors the colloidal-experiment pillar and closes the loop from 1975 endoreversible theory to 2016 single-particle measurement.

Classically a thermodynamic cycle is a closed loop in a \((p,V)\) or \((T,S)\) diagram, but a finite quantum system has no pressure or volume in the usual sense. The authors resolve this by working with the system's energy eigenvalues \(E_n(\lambda)\) — tuned by an external control parameter \(\lambda\) (a trap frequency, a field, a level spacing) — and its occupation probabilities \(p_n\). The first law is split unambiguously: changing \(E_n\) at fixed \(p_n\) is work, changing \(p_n\) at fixed \(E_n\) (through contact with a bath) is heat.

For internal energy \(U=\sum_n p_n E_n\), differentiating gives the quantum first law

A quantum isochoric stroke holds \(\lambda\) fixed and lets the populations relax toward a bath, so only heat flows; a quantum adiabatic stroke varies \(\lambda\) slowly enough that each \(p_n\) is frozen (no level crossings), so only work is done. Assembling two isochores and two adiabats yields the quantum Otto cycle; replacing the adiabats with quantum isothermals yields the quantum Carnot cycle, whose efficiency still caps at \(\eta_C = 1 - T_c/T_h\).

The framework is platform-agnostic but is developed for the simplest working substances — a single spin-1/2 in a tunable field and a particle in a box or harmonic trap of variable frequency. These are exactly the substrates later realized in trapped-ion, NMR, and NV experiments, so the paper's abstractions map directly onto hardware.

By pinning down heat versus work at the level of populations and eigenvalues, Quan et al. made finite-time questions well posed: once you accelerate a stroke, populations no longer follow the instantaneous eigenstates, coherence and "quantum friction" appear, and efficiency drops below the quasi-static value. That gap is precisely what shortcut-to-adiabaticity and quantum-speed-limit work later attack. It is the conceptual root of the quantum-engine program. [Phys. Rev. E 76, 031105, 2007]

The working substance is a quantum harmonic oscillator with time-dependent frequency \(\omega(t)\). The four Otto strokes are: adiabatic compression \(\omega_c \to \omega_h\) (work in), hot isochore (heat in at \(\omega_h\)), adiabatic expansion \(\omega_h \to \omega_c\) (work out), and cold isochore (heat out). In the quasi-static limit the efficiency is fixed purely by the compression ratio,

independent of the bath temperatures — the quantum Otto signature.

What makes the harmonic case tractable is that the dynamics close on a small set of operators (the Hamiltonian, the squeezing generator, and position-momentum correlations) forming an \(SU(1,1)\) Lie algebra. Sweeping \(\omega(t)\) at finite speed generates parametric squeezing: the state is no longer diagonal in the instantaneous energy basis, and the extra "parasitic" internal energy is quantum friction. This friction raises the entropy produced per cycle and forces power and efficiency apart — you cannot maximize both by simply running faster.

Optimizing cycle time, the authors show the maximum-power operating point and how it degrades as strokes shorten. Crucially, the coherence generated by fast driving can be undone: a suitably shaped \(\omega(t)\) (a shortcut/frictionless trajectory that returns the state to a diagonal form at stroke's end) restores the quasi-static efficiency at finite power — the harmonic engine's version of a shortcut to adiabaticity. The review lays out the phase diagram of engine, refrigerator, heater, and accelerator modes as a function of the control parameters.

The harmonic Otto oscillator is the theoretical workhorse against which single-atom and spin engines are benchmarked, and its exact solvability makes it the natural testbed for counterdiabatic driving and speed-limit bounds. This review is the canonical reference tying those threads together. [Entropy 19, 136, 2017]

The adiabatic theorem guarantees a system stays in its instantaneous eigenstate \(|n(t)\rangle\) only if the drive is slow compared with the inverse gap. For a fast sweep the state leaks into other levels (nonadiabatic transitions), spoiling both quantum control protocols and thermodynamic strokes. The question: can one keep perfect eigenstate following without the slowness penalty?

The answer is to augment the reference Hamiltonian \(H_0(t)\) with a control term \(H_{\rm CD}(t)\) constructed from the eigenstates themselves,

so that under \(H_0 + H_{\rm CD}\) the exact solution is precisely the instantaneous eigenstate at any drive speed. The added field supplies the "extra push" that would otherwise be missing at finite rate, actively steering amplitude to stay on track rather than waiting for slowness to suppress transitions.

The original context is coherent chemical dynamics — accelerating adiabatic passage in a few-level molecular system driven by shaped laser pulses. The construction is fully general, however, and applies to any driven quantum system with a discrete spectrum, which is why it migrated straight into quantum thermodynamics.

This is the birth of counterdiabatic driving. Berry (2009) recast it as "transitionless driving," and del Campo (2014) turned it into "super-adiabatic" engines that run frictionlessly at finite power. Every shortcut-to-adiabaticity heat engine traces its \(H_{\rm CD}\) to this paper. [J. Phys. Chem. A 107, 9937, 2003]

Berry asks the inverse-engineering question directly: given a reference \(H_0(t)\) with eigenstates \(|n(t)\rangle\), find the Hamiltonian \(H(t)\) for which the adiabatic approximation to \(H_0\) becomes the exact evolution — no leakage, no slowness requirement.

He shows \(H = H_0 + H_1\), where the additional piece

is exactly the transitionless generator (equivalent to Demirplak & Rice's counterdiabatic field). For a driven two-level system this reduces to a transparent result: \(H_1\) is an extra field component orthogonal to the original one, with magnitude set by the rate of change of the mixing angle, \(\propto \hbar\,\dot\theta\). Run the drive twice as fast and the required counterterm simply doubles.

The formulation is abstract and universal — spins, oscillators, cold atoms, molecules — with the two-level and harmonic cases worked explicitly. The catch it makes plain: \(H_1\) depends on the full instantaneous spectrum and may be non-local or hard to implement, which motivates all subsequent work on approximate or locally-realizable shortcuts.

Berry's paper is the reference that turned counterdiabatic driving into a standard tool. Its two-level formula is what experimentalists actually implement, and it underpins the super-adiabatic engines of del Campo (2014) and the unifying review of Guery-Odelin (2019). [J. Phys. A: Math. Theor. 42, 365303, 2009]

In a real finite-time Otto cycle, fast compression/expansion excites the working substance out of the instantaneous eigenbasis. This quantum friction forces extra heat dumping on the isochores, lowering efficiency, while running slowly to avoid it kills the power (energy per cycle divided by cycle time). The generic result is a strict trade-off. Can it be broken?

They add the transitionless driving term \(H_{\rm CD}(t)\) to each adiabatic (frequency-sweep) stroke of a harmonic Otto engine, so the state returns to a thermal, diagonal form at stroke's end regardless of how fast \(\omega(t)\) is swept. Friction is eliminated, and the finite-time efficiency is restored to the ideal \(\eta = 1 - \omega_c/\omega_h\). The finite power is then set only by the isochoric relaxation times, not by the adiabatic slowness constraint,

The energetic cost of the counterdiabatic field itself is accounted for and can be kept modest for suitable protocols.

The concrete model is again the harmonic (and hard-wall box) quantum Otto engine — analytically tractable and directly relevant to ion-trap and cold-atom realizations, where \(\omega(t)\) is a controllable trap frequency.

This is the proof of principle that quantum control can buy thermodynamic advantage: the shortcut toolkit of Demirplak-Rice and Berry becomes an engine design principle. It set off a wave of work on the cost and optimality of counterdiabatic engines, culminating in the review by Guery-Odelin et al. (2019). [Sci. Rep. 4, 6208, 2014]

A decade of STA work had produced several apparently distinct techniques for accelerating adiabatic processes. The review's job is to show they are facets of one idea — engineer the fast dynamics so the system lands in the target adiabatic state — and to catalogue where each is best applied.

Three main routes are laid out. Counterdiabatic (transitionless) driving adds \(H_{\rm CD}\) to suppress transitions exactly. Invariant-based inverse engineering uses a Lewis-Riesenfeld dynamical invariant \(I(t)\) with \(\big[i\hbar\,\partial_t - H(t),\,I(t)\big]=0\): one designs \(I(t)\) with the desired boundary behavior and reads off the required \(H(t)\). Fast-forward scaling drives a wavefunction along an accelerated version of its adiabatic trajectory. The review makes precise how these are related by unitary/gauge transformations, so a hard-to-implement counterterm in one picture becomes a tractable field in another.

A dedicated thread covers STA-enhanced heat engines and refrigerators: using shortcuts to remove quantum friction from finite-time strokes, the energetic cost of the shortcut, and quantum-speed-limit bounds on how much time a shortcut can actually save. This is the connective tissue between the STA and FTT literatures.

As the field's canonical review it fixes definitions and notation, surveys experimental realizations in cold atoms, ions, and NV centers, and frames the open questions (cost, robustness, many-body shortcuts) that guide subsequent work. For the finite-time quantum-engine program it is the standard entry point. [Rev. Mod. Phys. 91, 045001, 2019]

For closed systems the Mandelstam-Tamm and Margolus-Levitin bounds set a minimum time to reach an orthogonal (distinguishable) state, controlled respectively by the energy spread \(\Delta E\) and the mean energy above the ground state,

Real engines and erasure protocols are open, dissipative, and driven, so the question is what bounds their speed. A universal answer is prerequisite to asking how quickly a heat engine can complete a cycle.

Deffner & Lutz derive a bound from the geometry of density matrices, measuring the "distance" traveled by the Bloch angle (or Bures angle) between initial and final states and dividing by the time-averaged speed set by the generator of the open-system dynamics. For a driven system with an arbitrary (non-Markovian) master equation, the minimal time is

where \(\mathcal L\) is the geometric distance and the denominator the time-averaged rate. The key physical finding: non-Markovian memory (information flowing back from the environment) increases the instantaneous speed and therefore shortens the minimal time relative to the Markovian case.

The bound is worked out for a driven two-level system coupled to a structured reservoir, the canonical model of an open qubit undergoing damped Rabi dynamics — precisely the situation of a qubit working substance in contact with a finite-time bath.

By giving a speed limit for exactly the driven-dissipative processes that occur in quantum engines, erasure, and control, this paper connects the abstract Heisenberg-uncertainty speed limits to concrete finite-time thermodynamics — bounding, for instance, the shortest time a shortcut-to-adiabaticity stroke can achieve. [Phys. Rev. Lett. 111, 010402, 2013]

How short can a quantum evolution be? The review collects the answers — the fundamental limits on the rate at which a state can change, how they generalize from pure closed states to mixed open ones, and what they imply for how fast one can drive, control, and run thermodynamic protocols.

It presents the two historical bounds — Mandelstam-Tamm, set by the energy fluctuation \(\Delta E\), and Margolus-Levitin, set by the mean energy \(\langle E\rangle\) — and their unified geometric form,

then generalizes to mixed states and open dynamics via Bures/relative-purity metrics. It clarifies when each bound is tight and how driving reshapes them.

A central theme is the QSL as a resource limit: it lower-bounds the time of any thermodynamic stroke, caps the achievable power of a finite-time engine, sets the ultimate cost of shortcuts to adiabaticity, and defines what "optimal" quantum control can hope to achieve. The review ties the speed limit directly to the power-efficiency and speed-dissipation trade-offs of the FTT literature.

As the standard pedagogical entry point to QSLs, it consolidates a scattered literature (including Deffner-Lutz 2013) into one framework and frames the open problems — many-body speed limits, classical analogues, and speed limits with a fixed energy budget — that drive current work. [J. Phys. A: Math. Theor. 50, 453001, 2017]

Landauer's principle sets the minimum heat to erase one bit, \(Q \ge k_B T\ln 2\), reached only in the quasi-static limit. Any finite-time reset costs more, with the excess scaling as \(\Sigma \sim 1/\tau\) for protocol time \(\tau\). The question: does quantum mechanics change the finite-time correction to Landauer, and if so in which direction?

Driving a quantum bit quickly creates coherences between energy eigenstates that a classical bit cannot have. Analyzing the full distribution of dissipated work (not just its mean), the authors show these coherences fatten the tail: rare trajectories dissipate anomalously much. Because the average excess dissipation is bounded below by the fluctuations, the finite-time erasure cost acquires an extra positive quantum term,

so quantum coherence hinders rather than helps the approach to the Landauer limit. Suppressing coherence-generating driving (staying closer to the adiabatic basis) mitigates the penalty.

The analysis is for a driven qubit (or few-level system) weakly coupled to a thermal bath and reset in finite time — the canonical model of a quantum memory element or logical bit undergoing Landauer erasure.

It sharpens the quantum version of Landauer's principle at the level of fluctuations, connecting quantum work statistics to the thermodynamic cost of computation and warning that naive fast erasure carries a distinctly quantum overhead. It sits at the intersection of the quantum-Landauer and work-statistics strands of FTT. [Phys. Rev. Lett. 125, 160602, 2020]

For classical Markov machines the TUR states that reducing the relative fluctuations of an output current \(J\) requires paying more entropy production \(\Sigma\),

This is a hard trade-off: a steadier engine is a more dissipative one. The question is whether quantum systems obey the same bound, or whether coherence lets a machine be both precise and thermodynamically cheap.

Working with quantum master equations for open steady-state machines, the authors show that when the dynamics sustain quantum coherence between energy eigenstates (i.e. the generator is not effectively classical), the classical TUR can be violated: the product on the left can drop below \(2k_B\), so a given precision is achieved at lower dissipation than any classical machine allows. They identify coherence — not merely non-Markovianity — as the resource responsible, and derive modified, looser quantum bounds that the coherent current does respect.

The analysis targets quantum thermal machines in nonequilibrium steady state — e.g. multi-level systems or coupled qubits bridging a hot and cold reservoir and carrying a steady heat or charge current, the continuous (autonomous) counterpart of a cyclic engine.

Precision (low output noise) is as important as efficiency for a real device, and this paper shows quantum coherence is a genuine thermodynamic resource for it, relaxing a bound thought universal. It links the thermodynamic-uncertainty program to quantum-engine design and to the broader question of what coherence buys in finite-time thermodynamics. [Phys. Rev. Research 1, 033021, 2019]

Can the four-stroke logic of a heat engine survive when the working substance is one quantum particle, with thermal fluctuations comparable to the extracted work? Realizing it requires a controllable trap, two baths at different temperatures, and a way to store and read out mechanical work — all for a single ion.

The working substance is a single \(^{40}\)Ca\(^+\) ion held in a funnel-shaped (tapered) linear Paul trap, so its radial confinement depends on axial position. The hot bath is synthesized by applying electric-field noise to a trap electrode (driving the ion's radial motion to a high effective temperature); the cold bath is laser Doppler cooling. As the radial energy is periodically heated and cooled in step with the position-dependent trap frequency, radial thermal energy is converted into coherent axial oscillation — the engine's flywheel — realizing an Otto cycle whose work output is stored in that axial motion,

Work and power were inferred from the amplified axial oscillation amplitude.

A single \(^{40}\)Ca\(^+\) ion in a linear Paul trap with a tapered geometry, driven at radial frequencies of order MHz. The reported output power was on the scale of \(\sim 10^{-22}\,\text{W}\), with an extracted efficiency of order a fraction of a percent — small in absolute terms but a large fraction of the Otto ideal for the operating parameters. Note: the device operated in a thermal (not manifestly quantum-coherent) regime; the achievement is the single-atom scale, not coherence per se.

The defining demonstration that thermodynamic cycles run on one atom, this experiment anchored the finite-time quantum-engine field to real hardware and set the stage for probing genuinely quantum effects (spin coherence, squeezed baths) in later single-particle engines. [Science 352, 325, 2016]

Beyond showing an engine runs, one wants to reconstruct its complete thermodynamics at the single-spin level: the distributions of work and heat over the finite-time cycle, and the resulting efficiency including fluctuation corrections invisible to a classical treatment.

The working substance is a spin-1/2 \(^{13}\)C nucleus whose Zeeman-like energy gap is set by an effective magnetic field controlled with radio-frequency pulses. The two adiabatic strokes vary the gap (work), and the two isochores thermalize the spin to effective hot/cold states prepared by control. The Otto efficiency again reads

with \(\Delta_{h,c}\) the energy gaps at the two ends of the sweep. Work statistics were obtained by an interferometric (characteristic-function) measurement using an ancilla spin.

Liquid-state NMR at room temperature: the \(^{13}\)C nuclear spin of a \(^{13}\)C-labeled chloroform molecule serves as the engine, while the \(^1\)H spin acts as the ancilla used to read out the work distribution via a Ramsey-type interferometer. The setup gives near-perfect single-qubit control and readout, enabling full tomography of the cycle. The measured efficiency was consistent with the finite-time Otto prediction. Device detail: single-molecule spin engine; the ensemble NMR signal reports the single-spin thermodynamics.

It is the most completely characterized microscopic quantum engine of its generation, turning the abstract quantum Otto cycle of Quan/Kosloff into measured work and heat distributions and providing a clean testbed for finite-time and fluctuation predictions. [Phys. Rev. Lett. 123, 240601, 2019]

Theory predicted that in the limit of small engine action (short strokes, small energy exchange per cycle) a coherent quantum engine and its stochastic classical analogue diverge: coherence can boost the output power beyond the classical bound. The experiment set out to observe that crossover directly, by comparing an "Otto" and a coherent "quantum" operating mode of the same engine.

Uzdin's theory frames stroke-based (Otto) and continuous (coherent) engines as thermodynamically equivalent only when coherence is negligible; when coherence survives, the coherent engine's power exceeds the classical-equivalent value. The experiment prepares the working levels so that quantum coherence persists across strokes and measures the extracted power, finding

the signature of a coherence-enhanced engine. Outside that regime the two modes coincide, as predicted.

The engine is realized in an ensemble of nitrogen-vacancy (NV) centers in diamond, whose electronic spin sublevels provide the working energy levels; the strokes are driven optically and with microwaves, and the baths are engineered from the NV level structure. The NV ensemble gives a large, well-controlled signal while each center behaves as the microscopic engine. Device detail: solid-state NV-center ensemble; the advantage is observed at small per-cycle action.

This is among the first experiments to isolate a concrete, measurable regime where a quantum engine beats its classical equivalent, giving empirical teeth to the claim that coherence is a thermodynamic resource — the practical payoff of the quantum-engine theory line. [Phys. Rev. Lett. 122, 110601, 2019]

In quantum mechanics work done over a driven process is not a standard observable — it depends on two energy measurements whose back-action disturbs the very coherence one wants to study. The naive two-projective- measurement scheme destroys quantum features, so a non-invasive route to the work statistics is needed to test fluctuation theorems in a truly quantum setting.

The work distribution \(P(W)\) is accessed through its characteristic function \(\chi(u)=\langle e^{iuH_\tau}e^{-iuH_0}\rangle\), which is the Fourier transform of \(P(W)\). Coupling the driven spin to an ancilla qubit and reading out the ancilla's phase and contrast (a Ramsey interferometer) yields \(\chi(u)\) directly, from which \(P(W)\) is reconstructed. The measured distribution satisfies the Jarzynski equality and, with the reverse protocol, the Crooks relation,

Liquid-state NMR: the driven system is a \(^{13}\)C nuclear spin in chloroform, rapidly quenched by a shaped RF field, with the \(^1\)H spin as the ancilla/interferometer. Room-temperature NMR provides the high-fidelity single- and two-qubit control needed to realize the characteristic-function measurement cleanly. Device detail: two-spin \(^{13}\)C/\(^1\)H molecule; the closed-system (unitary) quench is the process under test.

By making quantum work statistics measurable, this experiment turned the quantum fluctuation theorems from formal identities into laboratory facts and provided the technique later reused to measure entropy production (Batalhao 2015) and to characterize spin engines (Peterson 2019). [Phys. Rev. Lett. 113, 140601, 2014]

A finite-time drive on an isolated quantum system produces entropy even without a bath, because the state lags the instantaneous equilibrium. The goal is to measure that irreversible entropy production directly and test the microscopic origin of time's arrow in a controlled quantum experiment.

Reconstructing the work statistics for both the forward quench and its time-reversed partner, the mean irreversible entropy production is the relative entropy (Kullback-Leibler distance) between them,

The measured \(\langle\Sigma\rangle\) grows with the driving rate — faster quenches are more irreversible — and its positivity encodes the second law and the arrow of time. The experiment shows entropy production is set by how far the forward and reversed trajectories can be told apart.

The same liquid-state NMR platform: a \(^{13}\)C nuclear spin driven through a finite-time quench with a \(^1\)H ancilla for interferometric readout of the work characteristic function. Room-temperature, high-control NMR makes the delicate forward/reverse comparison feasible. Device detail: single driven spin in an effectively closed system.

It provides the experimental counterpart to finite-time entropy-production theory, linking the speed of a protocol to its irreversibility and grounding the second law in measurable quantum statistics — directly relevant to the dissipation costs that limit finite-time engines and erasure. [Phys. Rev. Lett. 115, 190601, 2015]

The Jarzynski equality relates nonequilibrium work to an equilibrium free-energy difference, \(\langle e^{-\beta W}\rangle = e^{-\beta\Delta F}\), and must hold no matter how violently the system is driven. Testing it quantum-mechanically requires reconstructing the full work distribution \(P(W)\), including the rare low-work events that dominate the exponential average.

A single ion's motional mode is a quantum harmonic oscillator; a sudden displacement/quench does work on it. By preparing a thermal motional state and using the ion's internal levels to projectively measure the motional energy before and after the drive, the team samples the two-point work distribution \(P(W)\) and forms the exponential average,

confirming the equality across driving strengths from near-equilibrium to strongly non-equilibrium. The rare "negative-dissipation" trajectories required for the identity are resolved directly.

A single trapped ion (\(^{171}\)Yb\(^+\)) in a linear Paul trap; the motional (phonon) mode is the driven harmonic oscillator, and the ion's internal electronic states are used to map out the phonon-number distribution via state-dependent readout. Laser cooling and coherent motional control set the initial thermal state and the quench. Device detail: single \(^{171}\)Yb\(^+\) ion; the working system is its quantized motion, not a spin.

A clean single-particle confirmation of a cornerstone nonequilibrium identity in a fully quantum setting, it complements the NMR work-statistics experiments and establishes trapped ions as a precision platform for quantum stochastic thermodynamics. [Nat. Phys. 11, 193, 2015]

Szilard's 1929 thought experiment argues that knowing which side of a box a single particle occupies lets a demon extract \(k_B T\ln 2\) of work per bit — provided the measurement's information is properly accounted for. Realizing it demands a two-state system, a fast single-shot measurement, and feedback that exploits the outcome before it is erased.

A single-electron box has two charge states (electron on one island or the other) whose energies are tuned by a gate. The experiment (i) measures which state is occupied, (ii) applies a gate protocol conditioned on that outcome so the electron's random thermal transition performs work against the drive, and (iii) extracts on average up to the Szilard/Landauer limit per cycle,

with the equality approached in the reversible, slow-feedback limit. Measured heat and work agreed with the information-thermodynamic prediction, closing the loop between Maxwell's demon and Landauer's principle.

A single-electron box built from metallic tunnel junctions (a superconductor-normal-metal single- electron transistor as the charge detector), operated at cryogenic (sub-kelvin) temperature. Charge states are read out in single shots and manipulated by a gate voltage; feedback is applied electronically in real time. Device detail: the engine is a discrete two-charge-state system — thermodynamically quantum by its discreteness, though the dynamics are classical stochastic jumps rather than coherent.

A defining demonstration of information thermodynamics — that a measured bit is a genuine thermodynamic resource worth \(k_B T\ln 2\) — it anchors the Landauer/Maxwell-demon strand of finite-time thermodynamics in experiment and pairs naturally with the quantum-Landauer erasure theory (Miller 2020) on the cost side. [Proc. Natl. Acad. Sci. USA 111, 13786, 2014]

Of the 47 works, 38 are theory — the proofs, bounds and optimal protocols that define the field; 3 are classical experiments (colloidal Brownian and single-electron engines); and 6 are quantum experiments — trapped-ion, NMR and NV realisations. The experimental column is thin, but it is growing fastest exactly where the physics is hardest to reach: at the scale of single atoms and qubits, where thermal and quantum fluctuations meet.

① Speed vs efficiency — Curzon–Ahlborn → quantum engines. The efficiency-at-maximum-power tradeoff runs through endoreversible design into the quantum Otto cycle, and out onto hardware in the single-ion Roßnagel engine and its NMR cousins.

② Exact irreversible laws — Jarzynski–Crooks → stochastic thermodynamics. The fluctuation theorems become trajectory thermodynamics (Seifert's framework), and are read out directly in colloidal and NMR work-statistics measurements.

③ Geometry of dissipation — Salamon–Berry → optimal control. Thermodynamic length grows into the friction tensor of optimal protocols and the counterdiabatic shortcuts to adiabaticity that beat the tradeoff on quantum hardware.

Quantum advantage in thermodynamic tasks — can a quantum engine or eraser genuinely outperform its classical bound, and by how much? Finite-time bounds for many-body and strongly-coupled systems — the thermodynamic-uncertainty relation and quantum speed limits are still being sharpened. Information engines at the quantum scale — Landauer erasure and Maxwell-demon feedback pushed into the coherent regime. And closing the theory–experiment gap: turning optimal-protocol theory into control pulses real devices can run.

In the real world nothing is quasi-static. Finite-time thermodynamics is how physics prices the clock — and the arc of this survey is that pricing being carried, theorem by theorem and experiment by experiment, all the way down to single atoms and qubits.