Equivalence of quantum heat machines, and quantum-thermodynamic signatures

Equivalence of quantum heat machines, and quantum-thermodynamic signatures

Review Speaker: Sang Hoon Lee Korea Institute for Advanced Study

http://newton.kias.re.kr/~lshlj82

The 7th Workshop on Non-Equilibrium Fluctuation Theorems, 18 November, 2016 @ Phoenix Island, Jeju Island

R. Uzdin, A. Levy, and R. Kosloff, Phys. Rev. X 5, 031044 (2015).

Discussion Leader: Jaegon Um (SNU)

quantum heat engines (QHE): thermal machines where the working substance is a quantum object


Q. What is quantum in quantum thermodynamics?


Q. What is quantum in quantum thermodynamics?e.g., four-stroke, two-stroke, and continuous engines



e.g., population evolving with the Langevin equation → density matrix evolving with the Lindblad equation



e.g., population evolving with the Langevin equation → density matrix evolving with the Lindblad equationinvolving Planck’s constant ~




mathematical formalism: the Liouville space formulation, the spectral norm action, the Strang decomposition, the symmetric rearrangement theorem, etc.

involving Planck’s constant ~





Appendix A Appendix B

Appendix C Appendix D






Appendix A Appendix B

Appendix C Appendix D


main results: (1) All engine types are equivalent in the quantum regime of small action with respect to . (2) QHE have quantum-thermodynamic signature, i.e., thermodynamic measurements can confirm the presence of quantum effects in the device.(3) Coherence enables power outputs that greatly exceed the power of stochastic (dephased) engines.

~

the success of thermodynamics: will it hold all the way to the atomic scale? It was anticipated that in the quantum regime new thermodynamic effects will surface.


However,

(1) When the baths are thermal, the Carnot efficiency limit is equally applicable to for a small quantum system.ref) R. Alicki, The Quantum Open System as a Model of the Heat Engine, J. Phys. A 12, L103 (1979); H. Spohn, Entropy Production for Quantum Dynamical Semigroups, J. Math. Phys. 19, 1227 (1978).

(2) Even classical fluctuation theorems hold without any alteration.ref) M. Campisi, P. Talkner, and P. Hänggi, Fluctuation Theorem for Arbitrary Open Quantum Systems, Phys. Rev. Lett. 102, 210401 (2009); M. Campisi, P. Hänggi, and P. Talkner, Colloquium: Quantum Fluctuation Relations: Foundations and Applications, Rev. Mod. Phys. 83, 771 (2011); H. T. Quan and H. Dong, Quantum Crooks Fluctuation Theorem and Quantum Jarzynski Equality in the Presence of a Reservoir, arXiv:0812.4955.


However,



efficiency at the maximum power (Chambadal-Novikov-Curzon-Ahlborn), finite-time thermodynamics, nonadiabatic couplings in finite-time quantum evolution: a new quantum frictionlike mechanism


However,



efficiency at the maximum power (Chambadal-Novikov-Curzon-Ahlborn), finite-time thermodynamics, nonadiabatic couplings in finite-time quantum evolution: a new quantum frictionlike mechanism

not a generic feature of quantum heat machines!

Is there really nothing generic, new, and profound in the thermodynamics of small quantum system?


This paper presents a generic thermodynamic behavior that is purely quantum in its essence and has no classical counterpart. Furthermore, it is shown that in the quantum regime, the generic coherent work extraction mechanism can significantly outperform the stochastic work extraction mechanism.


This paper presents a generic thermodynamic behavior that is purely quantum in its essence and has no classical counterpart. Furthermore, it is shown that in the quantum regime, the generic coherent work extraction mechanism can significantly outperform the stochastic work extraction mechanism.

most). Thus, the working substance cannot reach equilib-rium on its own. Furthermore, excluding a few nongenericcases, it is not possible to assign an equation of state thatestablishes a relation between thermodynamic quantities.Nevertheless, QHE’s satisfy the second law and thereforeare also bounded by the Carnot efficiency limit [1,72].Work strokes are characterized by zero contact with the

baths and an inherently time-dependent Hamiltonian. Theunitary evolution generated by this Hamiltonian can changethe energy of the system. On the other hand, the vonNeumann entropy and the purity remain fixed (unitaryevolution at this stage). Hence, the energy change of thesystem in this case constitutes pure work. The system’senergy change is actually an energy exchange with thework repository.When the system is coupled to a thermal bath and

the Hamiltonian is fixed in time, the bath can change thepopulations of the energy levels. In a steady state, thesystem reaches a Gibbs state where the density matrix hasno coherences in the energy basis and the population of thelevels is given by pn;b ¼ e−ðEn=TbÞ=

PNn¼1 e

−ðEn=TbÞ, whereN is the number of levels and “b” stands for “c” (cold) or“h” (hot). In physical models where the system thermalizesvia collision with bath particles, a full thermalization can beachieved in finite time [15,73–76]. However, it is notnecessary that the baths will bring the system close to aGibbs state for the proper operation of the engine. Inparticular, maximal efficiency (e.g., in Otto engines) can beachieved without full thermalization. Maximal power(work per cycle time) is also associated with partialthermalization [6,8]. The definitive property of a thermalbath is its aspiration to bring the system to a predefinedtemperature regardless of the initial state of the system. Theevolution in this stage does not conserve the eigenvalues ofthe density matrix of the system, and therefore, not onlyenergy but entropy as well is exchanged with the bath.Therefore, the energy exchange in this stage is consideredas heat.In contrast to definitions of heat and work that are based

on the derivative of the internal energy [1,77,78], ourdefinitions are obtained by energy balance when couplingonly one element (bath or external field) at a time. As wesee later, in some engine types, several agents change theinternal energy simultaneously. Even in this case, this pointof view of heat and work will still be useful for obtainingconsistent and physical definitions of heat and work.

B. Three types of engines

There are three core engine types that operate with twothermal baths: four-stroke engine, two-stroke engine, and acontinuous engine. A stroke is a time segment where acertain operation takes place, for example, thermalizationor work extraction. By definition, adjacent strokes in heatengines do not commute with each other. If they docommute (for example, see the “cold” and “hot” operations

in the two-stroke engine later), they can be combined into asingle stroke since the total effect of the two strokes can begenerated by applying the two operations simultaneously.Each stroke is a completely positive (CP) map [79], and

therefore, the one-cycle evolution operator of the engine isalso a CP map. For the extraction of work, it is imperativethat some of the stroke propagators do not commute [80].Otto engines and Carnot engines are examples of four-

stroke engines. The simplest quantum four-stroke engine isthe two-level Otto engine shown in Fig. 1(a). In the firststroke, only the cold bath is connected to the system. Thus,the internal energy changes are associated with heatexchange with the cold bath. The expansion and compres-sion of the levels are fully described by a time-dependentHamiltonian of the form HðtÞ ¼ fðtÞσz (the baths aredisconnected at this stage). In the second stroke, work isconsumed in order to expand the levels, and in the fourthstroke, work is produced when levels revert to their originalvalues. There is a net work extraction since the populationsin stages II and IV are different. In different engines, muchmore general unitary transformation can be used to extractwork. Nevertheless, this particular operation resembles theclassical expansion and compression of classical engines.The work is the energy exchanged with the system duringthe unitary stages: W ¼ WII þWIV ¼ ðhE3i − hE2iÞþðhE5i − hE4iÞ. We consider only energy expectation valuesfor two main reasons. First, investigations of work fluctua-tions revealed that quantum heat engines follow classicalfluctuation laws [72], and we search for quantum signaturesin heat engines. The second reason is that, in our view,the engine should not be measured during operation. Themeasurement protocol used in quantum fluctuationtheorems [3,4,72] eliminates the density-matrix coher-ences. These coherences have a critical component in

FIG. 1. (a) A two-level scheme of a four-stroke engine. (b) Atwo-particle scheme of a two-stroke engine. (c) A three-levelscheme of a continuous engine.

EQUIVALENCE OF QUANTUM HEAT MACHINES, AND … PHYS. REV. X 5, 031044 (2015)

031044-3

using the quantum heat engines:




PNn¼1 e










031044-3





PNn¼1 e










031044-3




PNn¼1 e










031044-3


example: Otto, Carnot enginesthe expansion and compression of the levels described bya net work extraction from the different populations in stages II and IV




PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3

hE1ihE2ihE3ihE4i

= hE5i

hE3ihE4i




PNn¼1 e










031044-3






PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3

hE1ihE2ihE3ihE4i

= hE5i

hE3ihE4i




PNn¼1 e










031044-3






PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3

the equivalence and quantum signature we study in thispaper. As shown in Sec. IV, measurements or dephasingdramatically change the engine. Thus, although we fre-quently calculate work per cycle, the measured quantityis the cumulative work, and it is measured only at theend of the process. The averaged quantities are obtained byrepeating the full experiment many times. Engines aredesigned to perform a task, and we assume that thiscompleted task is the subject of measurement. The engineinternal state is not measured.The heat per cycle taken from the cold bath is

Qc ¼ hE2i − hE1i, and the heat taken from the hot bathis Qh ¼ hE4i − hE3i. In a steady state, the average energyof the system returns to its initial value after one cycle [81]so that hE5i ¼ hE1i. From this result, it follows immedi-ately that Qc þQh þW ¼ 0; i.e., the first law of thermo-dynamics is obeyed. There is no instantaneous energyconservation of internal energy, as energy may be tempo-rarily stored in the interaction field or in the workrepository.In the two-stroke engine shown in Fig 1(b), the engine

consists of two parts (e.g., two qubits) [82]. One part maycouple only to the hot bath, and the other may couple onlyto the cold bath. In the first stroke, both parts interact withtheir bath (but do not necessarily reach equilibrium). In thesecond unitary stroke, the two engine parts are discon-nected from the baths and are coupled to each other. Theyundergo a mutual unitary evolution, and work is extractedin the process.In the continuous engine shown in Fig. 1(c), the two

baths and the external interaction field are connectedcontinuously. For example, in the three-level laser systemshown in Fig 1(c), the laser light represented by HwðtÞgenerates stimulated emission that extracts work from thesystem. This system was first studied in a thermodynamicscontext in Ref. [83], while a more comprehensive dynami-cal analysis of the system was given in Ref. [84]. It isimperative that the external field is time dependent. If it istime independent, the problem becomes a pure heat trans-port problem where Qh ¼ −Qc ≠ 0. In heat transport, theinteraction field merely “dresses” the level so that the bathssee a slightly modified system. The Lindblad generators aremodified accordingly, and heat flows without extracting orconsuming work [85]. Variations on these engine typesmay emerge because of realization constraints. For exam-ple, in the two-stroke engine, the baths may be continu-ously connected. This variation and others can still beanalyzed using the tools presented in this paper.

C. Efficiency vs work and heat

Since the early days of Carnot, efficiency receivedconsiderable attention for two main reasons. First, thisquantity is of great interest from both theoretical andpractical points of view. Second, unlike other thermody-namics quantities, the efficiency satisfies a universal bound

that is independent of the engine details. The Carnotefficiency bound is a manifestation of the second law ofthermodynamics. Indeed, for Markovian bath dynamics, itwas shown that quantum heat engines cannot exceed theCarnot efficiency [1]. Recently, a more general approachbased on a fluctuation theorem for QHE showed that theCarnot bound still holds for quantum engines [72]. Studiesin which higher-than-Carnot efficiency are reported [66] areinteresting, but they use nonthermal baths and therefore, notsurprisingly, deviate from results derived in the thermody-namic framework that deals with thermal baths. For exam-ple, an electric engine is not limited to Carnot efficiencysince its power source is not thermal. Although the presentwork has an impact on efficiency as well, we focus on workand heat separately in order to unravel quantum effects. Aswill be exemplified later, in some elementary cases, thesequantum effects do not influence the efficiency.

D. Bath description and Liouville space

The dynamics of the working fluid (system) interactingwith the heat baths is described by the Lindblad-Gorini-Kossakowski-Sudarshan (LGKS) master equation for thedensity matrix [79,86,87]:

ℏdtρ¼LðρÞ¼−i½Hs;ρ&þX

k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ

where the Ak operators depend on the temperature, relax-ation time of the bath, system bath coupling, and also on thesystem Hamiltonian Hs [79]. This form already encapsu-lates within the Markovian assumption of no bath memory.The justification for these equations arises from a “micro-scopic derivation” in the weak system-bath coupling limit[88]. In this derivation, a weak interaction field couplesthe system of interest to a large system (the bath) withtemperature T. This interaction brings the system into aGibbs state at temperature T. The Lindblad thermalizationoperators Ak used for the baths are described in the nextsection. The small Lamb shift is ignored.Equation (1) is a linear equation, so it can always be

rearranged into a vector equation. Given an index mappingρN×N → jρi1×N2 , the Lindblad equation now reads

iℏdtjρi ¼ ðHH þ LÞjρi ≐ Hjρi; ð2Þ

where HH is a Hermitian N2 × N2 matrix that originatesfrom Hs, and L is a non-Hermitian N2 × N2 matrix thatoriginates from the Lindblad evolution generators Ak. Thisextended space is called Liouville space [89]. In this paper,we use calligraphic letters to describe operators in Liouvillespace and ordinary letters for operators in Hilbert space.For states, however, jAi will denote a vector in Liouvillespace formed from AN×N by “vec-ing” A into a column inthe same procedure ρ is converted into jρi. A short review

RAAM UZDIN, AMIKAM LEVY, AND RONNIE KOSLOFF PHYS. REV. X 5, 031044 (2015)

031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

hE1ihE2ihE3ihE4i

= hE5i

hE3ihE4i




PNn¼1 e










031044-3






PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

hE1ihE2ihE3ihE4i

= hE5i

hE3ihE4i




PNn¼1 e










031044-3






PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

hE1ihE2ihE3ihE4i

= hE5i

hE3ihE4i




PNn¼1 e










031044-3






PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3




PNn¼1 e










031044-3











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

hE1ihE2ihE3ihE4i

= hE5i

hE3ihE4i











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

bath description & Liouville space

the Lindblad-Gorini-Kossakowski-Sudarshan (LGKS) master equation for the density matrix:











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

(in the weak system-bath coupling limit)











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

( )

[56] A. Mari and J. Eisert, Cooling by Heating: Very HotThermal Light Can Significantly Cool Quantum Systems,Phys. Rev. Lett. 108, 120602 (2012).

[57] D. Venturelli, R. Fazio, and V. Giovannetti, Minimal Self-Contained Quantum Refrigeration Machine Based onFour Quantum Dots, Phys. Rev. Lett. 110, 256801 (2013).

[58] M. Lostaglio, K. Korzekwa, D. Jennings, and T. Rudolph,Quantum Coherence, Time-Translation Symmetry, andThermodynamics, Phys. Rev. X 5, 021001 (2015).

[59] M. Lostaglio, D. Jennings, and T. Rudolph, Description ofQuantum Coherence in Thermodynamic ProcessesRequires Constraints Beyond Free Energy, Nat. Commun.6, 6383 (2015).

[60] P. Kammerlander and J. Anders, Quantum Measurementand Its Role in Thermodynamics, arXiv:1502.02673.

[61] M. T. Mitchison, M. P. Woods, J. Prior, and M. Huber,Coherence-Assisted Single-Shot Cooling by QuantumAbsorption Refrigerators, arXiv:1504.01593.

[62] J. Åberg, Catalytic Coherence, Phys. Rev. Lett. 113,150402 (2014).

[63] F. C. Binder, S. Vinjanampathy, K. Modi, and J. Goold,Quantacell: Powerful Charging of Quantum Batteries,New J. Phys. 17, 075015 (2015).

[64] K. Korzekwa, M. Lostaglio, J. Oppenheim, and D.Jennings, The Extraction of Work from Quantum Coher-ence, arXiv:1506.07875.

[65] S. Rahav, U. Harbola, and S. Mukamel, Heat Fluctuationsand Coherences in Quantum Heat Engines, Phys. Rev. A86, 043843 (2012).

[66] M. O. Scully, M. S. Zubairy, G. S. Agarwal, and H.Walther, Extracting Work from a Single Heat Bath viaVanishing Quantum Coherence, Science 299, 862 (2003).

[67] M. O. Scully, K. R. Chapin, K. E. Dorfman, M. BarnabasKim, and A. Svidzinsky, Quantum Heat Engine PowerCan Be Increased by Noise-Induced Coherence, Proc.Natl. Acad. Sci. U.S.A. 108, 15097 (2011).

[68] M. Campisi, J. Pekola, and R. Fazio, NonequilibriumFluctuations in Quantum Heat Engines: Theory, Example,and Possible Solid State Experiments, New J. Phys. 17,035012 (2015).

[69] Other types of engines consist of small variations and acombination of these types.

[70] R. Alicki and M. Fannes, Entanglement Boost forExtractable Work from Ensembles of Quantum Batteries,Phys. Rev. E 87, 042123 (2013).

[71] K. V. Hovhannisyan, M. Perarnau-Llobet, M. Huber, andA. Acín, Entanglement Generation Is Not Necessary forOptimal Work Extraction, Phys. Rev. Lett. 111, 240401(2013).

[72] M. Campisi, Fluctuation Relation for Quantum HeatEngines and Refrigerators, J. Phys. A 47, 245001 (2014).

[73] G. Gennaro, G. Benenti, and G. Massimo Palma, Entan-glement Dynamics and Relaxation in a Few-Qubit SystemInteracting with Random Collisions, Europhys. Lett. 82,20006 (2008).

[74] G. Gennaro, G. Benenti, and G. Massimo Palma, Relax-ation Due to Random Collisions with a Many-QuditEnvironment, Phys. Rev. A 79, 022105 (2009).

[75] T. Rybár, S. N. Filippov, M. Ziman, and V. Bužek,Simulation of Indivisible Qubit Channels in CollisionModels, J. Phys. B 45, 154006 (2012).

[76] M. Ziman, P. Štelmachovič, and V. Bužek, Description ofQuantum Dynamics of Open Systems Based on Collision-like Models, Open Syst. Inf. Dyn. 12, 81 (2005).

[77] R. Kosloff, Quantum Thermodynamics: A DynamicalViewpoint, Entropy 15, 2100 (2013).

[78] J. Anders and V. Giovannetti, Thermodynamics of DiscreteQuantum Processes, New J. Phys. 15, 033022 (2013).

[79] H.-P. Breuer and F. Petruccione, Open Quantum Systems(Oxford University Press, Oxford, 2002).

[80] R. Kosloff and T. Feldmann, Optimal Performance ofReciprocating Demagnetization Quantum Refrigerators,Phys. Rev. E 82, 011134 (2010).

[81] This is, of course, not true for the work repository.[82] A. E. Allahverdyan, K. Hovhannisyan, and G. Mahler,

Optimal Refrigerator, Phys. Rev. E 81, 051129 (2010).[83] H. E. D. Scovil and E. O. Schulz-DuBois, Three-Level

Masers as Heat Engines, Phys. Rev. Lett. 2, 262 (1959).[84] E. Geva and R. Kosloff, The Quantum Heat Engine and

Heat Pump: An Irreversible Thermodynamic Analysis ofthe Three-Level Amplifier, J. Chem. Phys. 104, 7681(1996).

[85] A. Levy and R. Kosloff, The Local Approach to QuantumTransport May Violate the Second Law of Thermodynam-ics, Europhys. Lett. 107, 20004 (2014).

[86] G. Lindblad, On the Generators of Quantum DynamicalSemigroups, Commun. Math. Phys. 48, 119 (1976).

[87] V. Gorini, A. Kossakowski, and E. C. G. Sudarshan,Completely Positive Dynamical Semigroup of n-LevelSystem, J. Math. Phys. (N.Y.) 17, 821 (1976).

[88] E. B. Davies, Markovian Master Equations, Commun.Math. Phys. 39, 91 (1974).

[89] S. Mukamel, Principles of Nonlinear Optical Spectroscopy(Oxford University Press, New York, 1995), Vol. 29.

[90] H. O. R. N. Roger and R. J. Charles, Topics in MatrixAnalysis (Cambridge University Press, Cambridge,England, 1994).

[91] T. F. Havel, Robust Procedures for Converting AmongLindblad, Kraus and Matrix Representations of QuantumDynamical Semigroups, J. Math. Phys. (N.Y.) 44, 534(2003).

[92] E. Andersson, J. D. Cresser, and M. J. W. Hall, Finding theKraus Decomposition from a Master Equation and ViceVersa, J. Mod. Opt. 54, 1695 (2007).

[93] A. Rivas, A. D. K. Plato, S. F. Huelga, and M. B. Plenio,Markovian Master Equations: A Critical Study, New J.Phys. 12, 113032 (2010).

[94] This can be seen by following the derivation in Ref. [79]and using the formalism introduced in Ref. [106].

[95] T. Jahnke and C. Lubich, Error Bounds for ExponentialOperator Splittings, BIT Numerical Math. 40, 735 (2000).

[96] M. D. Feit, J. A. Fleck, and A. Steiger, Solution of theSchrödinger Equation by a Spectral Method, J. Comput.Phys. 47, 412 (1982).

[97] H. De Raedt, Product Formula Algorithms for Solving theTime Dependent Schrödinger Equation, Comput. Phys.Rep. 7, 1 (1987).


031044-20













k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4












k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

the index mapping (to the Liouville space)











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

( )












































031044-20













k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4












k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

the index mapping (to the Liouville space)











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

( )












































031044-20











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

VI. CONCLUDING REMARKS

We identified coherent and stochastic work extractionmechanisms in quantum heat engines. While stroke engineshave both mechanisms, continuous engines only have thecoherent mechanism. We introduced the “norm action” ofthe engine using Liouville space and showed that when thisaction is small compared to ℏ, all three engine types areequivalent. This equivalence emerges because, for smallactions, only the coherent mechanism is important. Despitethe equivalence, before the engine cycle is completed, thestate of the different engine type differs by O½ðs=ℏÞ1$. Thisalso holds true for work and heat. Remarkably, at the end ofeach engine cycle, a much more accurate O½ðs=ℏÞ3$ equiv-alence emerges. Furthermore, the equivalence also holds fortransient dynamics, even when the initial state is very farfrom the steady state of the engine. It was shown that, forsmall actions, the coherent work extraction is considerablystronger than the stochastic work extraction mechanism.This enabled us to derive a power bound for stochasticengines that constitutes a quantum-thermodynamics signa-ture. Any power measurement that exceeds this boundindicates the presence of quantum coherence and theoperation of the coherent work extraction mechanism.Experimental schemes where the work is extracted by

changing the energy levels (e.g., Refs. [53–55]) correspondto a full swap in the multilevel embedding framework.Consequently, such setups have an inherently large action,and they are not suited for demonstrating the effectspresented here. In contrast, the scheme in Ref. [105] seemshighly suitable. There, the unitary operation that makes aswap between superconducting qubits is generated bycreating a magnetic flux through a superconducting ring.In the original paper, the authors use a flux that generates afull swap. However, by using weaker magnetic fields, theunitary operation will become a partial swap, and it shouldbe possible to attain the small action regime where theequivalence can be observed. In addition, NV centers indiamonds also have the potential for exploring heat engineequivalence in the quantum regime.The present derivation makes no assumption on the

direction of heat flows and the sign of work. Thus, ourresults are equally applicable to refrigerators and heaters.It is interesting to try and apply these concepts of

equivalence and quantum-thermodynamic signatures tomore general scenarios: non-Markovian baths, engineswith a nonsymmetric unit cell, and engines with quantumcorrelation between different particles (entanglement andquantum discord). We conjecture that in multiple particleengines, entanglement will play a similar role to that ofcoherence in single-particle engines.

ACKNOWLEDGMENTS

This work was support by the Israeli ScienceFoundation. Part of this work was supported by the

COST Action MP1209 “Thermodynamics in the quantumregime.”

APPENDIX A: LIOUVILLE SPACEFORMULATION OF QUANTUM DYNAMICS

Quantum dynamics is traditionally described in Hilbertspace. However, it is convenient, in particular, for openquantum systems, to introduce an extended space wheredensity operators are vectors and time evolution is gen-erated by a Schrödinger-like equation. This space is usuallyreferred to as Liouville space [89]. We denote the “densityvector” by jρi ∈ C1×N2

. It is obtained by reshaping thedensity matrix ρ into a larger single vector with indexα ∈ f1; 2;…N2g. The one-to-one mapping of the twomatrix indices into a single vector index fi; jg → α isarbitrary but has to be used consistently. The vector jρi isnot normalized to unity, in general. Its norm is equal to thepurity, P ¼ trðρ2Þ ¼ hρjρi, where hρj ¼ jρi† as usual. Theequation of motion of the density vector in Liouville spacefollows from dtρα ¼

Pβρβ∂ðdtραÞ=∂ρβ. Using this equa-

tion, one can verify that the dynamics of the density vectorjri is governed by a Schrödinger-like equation in the newspace,

i∂tjρi ¼ Hjρi; ðA1Þ

where the super-Hamiltonian H ∈ CN2×N2is given by

Hαβ ¼ i∂ðdtραÞ∂ρβ : ðA2Þ

A particularly useful index mapping is described inRef. [106] and in Ref. [90]. In this mapping, the Liouvilleindex of jρi is related to the original row and column indexof ρ via α ¼ colþ Nðrow − 1Þ. For this form, H can becompactly written in term of the original H and A:

H ¼ −iðH ⊗ I − I ⊗ HtÞ

þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ

where the superscript t stands for transposition and * forcomplex conjugation. H ¼ HH þ L is non-Hermitian foropen quantum systems. HH originates from the Hilbertspace Hamiltonian H, and L from the Lindblad terms. HH

is always Hermitian. The skew-Hermitian part ðL − L†Þ=2is responsible for purity changes. Yet, in Liouville space,the Lindblad operators Ak in Eq. (1) may also generate aHermitian term ðLþ L†Þ=2. Though Hermitian inLiouville space, this term cannot be associated with aHamiltonian in Hilbert space. If L ¼ 0, K is unitary. It isimportant to note that not all eigenvectors ofH in Liouville


031044-14






ACKNOWLEDGMENTS













þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ




031044-14

space can be populated exclusively. This is due to the factthat only positive ρ with unit trace are legitimate densitymatrices. The states that can be populated exclusivelydescribe steady states, while others correspond to transientchanges. We remind the reader that, in this paper, we usecalligraphic letters to describe operators in Liouville spaceand ordinary letters for operators in Hilbert space. Forstates, however, jAi will denote a vector in Liouville spaceformed from AN×N by “vec-ing” A into a column in thesame procedure ρ is converted into jρi.

1. Useful relations in Liouville space

In Liouville space, the standard inner product of twooperators in Hilbert space trA†B reads

trA†B ¼ hAjBi:

In particular, the purity P ¼ hrjri is just the square of thedistance from the origin in Liouville space.A useful relation for HH is

HHjHi ¼ hHjHH ¼ 0: ðA4Þ

The proof is as follows:

HHij;mn ¼ Himδjn −Hnjδim: ðA5Þ

Therefore, using Eq. (A5) we get

HHjHi ¼X

β

HHαβHβ ¼

X

mn

HHijmnHmn ¼ ½H;H% ¼ 0.

ðA6Þ

This property is highly useful. We stress that Eq. (A4) is aproperty of Hermitian operators in Hilbert space, whereboth H and H are well defined. A general Hermitianoperator in Liouville space may not have a correspondingH in Hilbert space.Another property that immediately follows from

Eq. (A5) is

HHii;kk ¼ 0: ðA7Þ

This corresponds to a well-known property of unitaryoperation. If the system starts from a diagonal densitymatrix, then for short times, the evolution generated byHH,e−iHHdt ¼ I − iHHdtþOðdt2Þ does not change the pop-ulation in the leading order.

2. Expectation values and their time evolutionin Liouville space

The expectation value of an operator in Hilbert space ishAi ¼ trðρAÞ. Since ρ is Hermitian, the expectation value isequal to the inner product of A and ρ, and therefore,

hAi ¼ trðρAÞ ¼ hρjAi:

The dynamics of hAi under the Lindblad evolutionoperator is

ddt

hAi ¼ −ihAjHjρiþ!ρ

""""ddt

A#: ðA8Þ

Note that in Liouville space there is no commutator termsince H operates on jρi just from the left. If the totalHamiltonian is Hermitian and time independent, theconservation of energy follows immediately from apply-ing Eqs. (A8) and (A4) for A ¼ H.

APPENDIX B: GEOMETRIC MEANINGOF THE NORM ACTION

This appendix establishes the relation between the normaction and the path length in Liouville space. The actionconstitutes an upper bound on the length of the path overone cycle. The infinitesimal path dl between two statesjρðtþ dtÞi and jρðtÞi in Liouville space is given by

dl2 ¼ kjρðtþ dtÞi − jρðtÞik22¼ hρðtÞjH†HjρðtÞidt2=ℏ2 þOðdt3Þ; ðB1Þ

where kjρðtþ dtÞi − jρðtÞik22 ¼ trð½ρðtþ dtÞ − ρðtÞ%2Þ.Consequently, the path in Liouville space is given by

L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc

0kHkhρjρidt; ðB2Þ

where we have used the property of the spectral normhρðtÞjH†HjρðtÞi=hρjρi ≤ kHk2sp. Since the purity hρjρi isalways smaller than 1,

L ≤1

ℏ

Zτcyc

0kHkdt≡ s=ℏ: ðB3Þ

Thus, the path length per cycle in Liouville space isbounded by the action. For previous uses of the normaction to quantify quantum dynamics, see Refs. [107–110].This is also true for times shorter than the cycle time τcyc,

LðτÞ ¼Z

τ

0

$dldt

%dt ≤ s=ℏ: ðB4Þ

The triangle inequality implies kjρðτÞi − jρð0Þik2 ≤ LðτÞ;therefore,

maxðkjρðτÞi − jρð0Þik2Þ ≤ s=ℏ: ðB5Þ

Hence, the action limits the maximal state change duringthe cycle. For example, if the action is 10−3ℏ, the state willchange by 10−3 at the most.


031044-15











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4






ACKNOWLEDGMENTS













þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ




031044-14






ACKNOWLEDGMENTS













þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ




031044-14




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15

[ ] []Hilbert space

matrix

Liouville spacevector











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4






ACKNOWLEDGMENTS













þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ




031044-14






ACKNOWLEDGMENTS













þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ




031044-14




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15

[ ] []Hilbert space

matrix

Liouville spacevector

…

…











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4






ACKNOWLEDGMENTS













þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ




031044-14






ACKNOWLEDGMENTS













þ iX

k

!ðAk ⊗ A'

kÞ −1

2I ⊗ ðA†

kAkÞt − 1

2A†kAk ⊗ I

";

ðA3Þ




031044-14




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15

h⇢|⇢i

In unitary dynamics, the largest energy gap of the Hamiltonian sets a speed limit on the rate of change of a state.

of Liouville space and some of its properties is given inAppendix A.In unitary dynamics, the largest energy gap of the

Hamiltonian sets a speed limit on the rate of change ofa state (e.g., rotation speed in the Bloch sphere). SinceH isnot Hermitian, the energy scalar that sets a speed limit onthe evolution speed is the spectral norm (or operator norm)of H, kHk ¼ max

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffieigðH†HÞ

p(The spectral norm is the

largest singular value of H [90]). In particular, we show inAppendix B that the norm action, defined as

s ¼Zτ

0

kHðtÞkdt; ð3Þ

sets a limit on how much a state can change during a time τbecause of the operation ofH. For time-independent super-HamiltonianH, the evolution operator in Liouville space is

jρðtÞi ¼ Kjρðt0Þi ¼ e−iHðt−t0Þ=ℏjρðt0Þi: ð4Þ

Writing the evolution operator as an exponent of a matrixhas a significant advantage since commutator exponentia-tion is avoided. Furthermore, the action has a naturaldefinition in this formalism. In principal, it should bepossible to reformulate the derivations using densitymatrixes and the Kraus operators. However, it seems thatthe Hilbert space formalism is far more cumbersome andcomplicated (for example, see Refs. [91,92]).While the Lindblad description works very well for

sufficiently long times, it fails for very short times wheresome of the approximation breaks down. In scales wherethe bath still has a memory of the system’s past states, thesemigroup property of the Lindblad equation no longerholds: jρðtþ t0Þi ≠ e−iðHsþLÞðt−t0Þ=ℏjρðt0Þi. This will set acutoff limit for the validity of the engine-type equivalencein the Markovian approximation.Next we introduce the multilevel embedding scheme that

enables us to discuss various heat engines in the samephysical setup.

E. Multilevel embedding

Let the working substance of the quantum engine be anN-level system. These levels are fixed in time [i.e., they donot change as in Fig. 1(a)]. For simplicity, the levels areassumed to be nondegenerate. We divide the energy levelsinto a cold manifold and a hot manifold. During theoperation of the engine, the levels in the cold manifoldinteract only with the cold bath, and the levels in the hotmanifold interact only with the hot bath. Each thermalcoupling can be turned on and off as a function of time, butthe aliasing of a level to a manifold does not change in time.If the manifolds do not overlap, the hot and cold thermal

operations commute and they can be applied at the sametime or one after the other. The end result will be the same.

Nevertheless, our scheme also includes the possibility thatone level appears in both manifolds. This is the case for thethree-level continuous engine shown in Fig. 1(c). Forsimplicity, we exclude the possibility of more than onemutual level. If there are two or more overlapping levels,there is an inevitable heat transport in the steady state fromthe hot bath to the cold bath even in the absence of anexternal field that extracts work. In the context of heatengines, this can be interpreted as heat leak. This “no field–no transport” condition holds for many engines studied inthe literature. Nonetheless, this condition is not a necessarycondition for the validity of our results.This manifold division seems sensible for the continuous

engine and even for the two-stroke engine in Fig. 1(b), buthow can it be applied to the four-stroke engine shown inFig. 1(a)? The two levels interact with both baths andalso change their energy value in time, contrary to theassumption of fixed energy levels. Nevertheless, this engineis also incorporated into the multilevel embedding frame-work. Instead of two levels as in Fig. 1(a), consider thefour-level system shown in the dashed green lines in Fig. 2.Initially, only levels 2 and 3 are populated and coupled to

the cold bath (2 and 3 are in the cold manifold). In theunitary stage, an interaction Hamiltonian Hswap generates afull swap of populations and coherence according to therule 1 ↔ 2; 3 ↔ 4. Now, levels 1 and 4 are populated and 2and 3 are empty. Therefore, this system fully simulates theexpanding-level engine shown in Fig. 1(a). At the sametime, this system satisfies the separation into well-defined

FIG. 2. In the standard two-level Otto engine, there are two-level Eg;e (purple arrows) that change in time to E0

g;e. In themultilevel embedding framework, the levels (E1–4) are fixed intime (black dashed lines), but a time-dependent field (π pulse,swap operation) transfers the population (green arrows) to theother levels. For a swap operation, the two schemes lead to thesame final state and therefore are associate with the same work.Nonetheless, the multilevel scheme is more general since forweaker unitary transformation (instead of the π pulse), coher-ences are generated. We show that this type of coherence cansignificantly boost the power output of the engine.


031044-5






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5




trA†B ¼ hAjBi:






HHjHi ¼X

β

HHαβHβ ¼

X

mn


ðA6Þ


Eq. (A5) is







ddt


""""ddt

A#: ðA8Þ






L ¼Z

τcyc

0

$dldt

%dt ≤

1

ℏ

Zτcyc



L ≤1

ℏ

Zτcyc



LðτÞ ¼Z

τ

0

$dldt






031044-15






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5

H

multilevel embedding: the working substance of the quantum engine is a fixed nondegenerate N-level system composed of a cold manifold (interacting only with the cold bath) and a hot manifold (interacting only with the hot bath).






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5







s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5







s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5




PNn¼1 e










031044-3






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5

time-independent manifolds, as defined in the multilevelembedding scheme.The full swap used to embed the traditional four-stroke

Otto engine is not mandatory, and other unitary operationscan be applied. This extension of the four-stroke scheme iscritical for our work since the equivalence of enginesappears when the unitary operation is fairly close to theidentity transformation. A full swap turns one diagonalstate into another. Consequently, the steady state of anengine with full swap operation will not contain anycoherences in the energy basis. As will be shown later, apartial swap or a different “weaker than full swap” unitaryleads to steady-state coherences that dramatically enhancethe power output. Note that these coherences between thehot and cold manifold imply a superposition of the cold andhot states. In other words, in contrast to the full swap case,the particle is not localized exclusively on either the hot orcold manifold.Figures 3(a)–(c) show how the three types of engines are

represented in the multilevel embedding scheme. Theadvantage of the multilevel scheme now becomes clear.All three engine types can be described in the samephysical system with the same baths and the same couplingto external fields (work extraction). The engine types differonly in the order of the coupling to the baths and to thework repository. While the thermal operations commute ifthe manifolds do not overlap, the unitary operation nevercommutes with the thermal strokes.In the present paper, we use a direct sum structure for the

hot and cold manifolds. However, when there are two ormore particles in the engine [82], it is more natural toapply a tensor product structure for the manifolds of themultilevel embedding scheme.On the right of Fig. 3, we plotted a “brick” diagram for

the evolution operator. Black stands for unitary trans-formation generated by some external field, while blueand red stand for hot and cold thermal coupling,

respectively. When the bricks are on top of each other, itmeans that they operate simultaneously. Now we are in aposition to derive the first main results of this paper: thethermodynamic equivalence of the different engine types inthe quantum regime.

III. CONTINUOUS AND STROKE ENGINEEQUIVALENCE

We first discuss the equivalence of continuous and four-stroke engines. Nevertheless, all the arguments are valid forthe two-stroke engines as well, as explained later. Althoughour results are not limited to a specific engine model, itwill be useful to consider the simple engine shown in Fig. 4.We use this model to highlight a few points and alsofor numerical simulations. The Hamiltonian part of thesystem is

H0 þ cosðωtÞHw; ð5Þ

where H0 ¼ −ðΔEh=2Þj1ih1j − ðΔEc=2Þj2ih2jþ ðΔEc=2Þj3ih3jþ ðΔEh=2Þj4ih4j, Hw ¼ ϵðtÞj1ih2jþ ϵðtÞj3ih4jþH:c: and ω ¼ ðΔEh − ΔEcÞ=2ℏ.The driving frequency that couples the system to the

work repository is in resonance with the top and bottomenergy gaps. The specific partitioning into hot and coldmanifolds was chosen so that only one frequency (e.g., asingle laser) is needed for implementing the system insteadof two.We assume that the Rabi frequency of the drive ϵ is

smaller than the decay time scale of the baths, ϵ ≪ γc; γh.Under this assumption, the dressing effect of the drivingfield on the system-bath interaction can be ignored. It isjustified, then, to use “local” Lindblad operators obtained inthe absence of a driving field [85,93]. For plotting purposes(reasonable duty cycle), in the numerical examples, weoften use ϵ ¼ γc ¼ γh. While this poses no problemfor stroke-engine realizations, for experimental demonstra-tion of equivalence with continuous engines, one has toincrease the duty cycle so that ϵ ≪ γc; γh. In other words,the unitary stage should be made longer but with a weakerdriving field.

FIG. 3. Representation of the three types of engines (a)–(c) inthe multilevel embedding framework. In this scheme, the differ-ent engine types differ only in the order of coupling to the bathsand work repository. Since the interactions and energy levels arethe same for all engine types, a meaningful comparison ofperformance becomes possible.

FIG. 4. Illustration of the engine used in the numericalsimulation. By changing the time order of the coupling to Hwand to thermal baths, all three types of engines can be realized inthe model.


031044-6






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5
















031044-6
















031044-6
















031044-6

unitary transformation generated bysome external field
















031044-6

hot thermal coupling
















031044-6

cold thermal coupling






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5
















031044-6
















031044-6
















031044-6

unitary transformation generated bysome external field
















031044-6

hot thermal coupling
















031044-6

cold thermal coupling

simultaneously operating

the equivalence of continuous and four-stroke engine (and the two-stroke engine as well)
















031044-6

1234


the Hamiltonian part of the system:
















031044-6
















031044-6

1234


















031044-6
















031044-6

1234

the driving frequency that couples the system to the work repository


















031044-6
















031044-6

1234

the Rabi frequency of the drive


“the population switching” term, for energy eigenstates|1i, |2i, |3i, |4i


















031044-6
















031044-6

1234


















031044-6

e.g.,1p2(|1i+ |2i) , 1p

5(2|3i+ |4i) , · · ·

“the population switching” term, for energy eigenstates|1i, |2i, |3i, |4i

















031044-6


















031044-6
















031044-6

1234

















031044-6
















031044-6

123

4

















031044-6

The Lindblad equation is given by Eq. (1) with theHamiltonian (5) and with the following Lindblad operatorsin Hilbert space:

A1 ¼ffiffiffiffiffiγh

pe−ðΔEh=2ThÞj4ih1j; ð6Þ


p j1ih4j; ð7Þ

A3 ¼ffiffiffiffiffiγc

pe−ðΔEc=2TcÞj3ih2j; ð8Þ


p j2ih3j: ð9Þ

In all the numerical simulations, we use ΔEh ¼ 4,ΔEc ¼ 1, Th ¼ 5, Tc ¼ 1. The interaction with the bathsor with work repository can be turned on and off at will.Starting with the continuous engine, we choose a unit

cell that contains exactly 6m (m is an integer) completecycles of the drive (τd ¼ 2π=ω) so that τcyc ¼ 6mτd. Thedifference between the engine cycle time and the cycles ofthe external drive will become clear in stroke engines (also,the factor of 6 will be clarified).For the validity of the secular approximation used in the

Lindblad microscopic derivation [79], the evolution timescale must satisfy τ ≫ ð2πℏÞ=minðΔEh;ΔEcÞ. Therefore,mmust satisfym ≫ ðℏωÞ=minðΔEh;ΔEcÞ. Note that if theLindblad description is obtained from a different physicalmechanism (e.g., thermalizing collisions), then this con-dition is not required.Next, we transform to the interaction picture (denoted by

tilde) using the transformation U ¼ e−iH0t=ℏ, and performthe rotating wave approximation (RWA) by dropping termsoscillating at a frequency of 2ω. For the RWA to be valid,the amplitude of the field must satisfy ϵ ≪ ω. The resultingLiouville space super-Hamiltonian is

~H ¼ Lc þ Lh þ1

2Hw: ð10Þ

Note that Lh;c were not modified by the transformation tothe rotating system since ½Lh;c;H0& ¼ 0 in the microscopicderivation [94]. The oscillatory time dependence hasdisappeared because of the RWA and the interactionpicture. There is still an implicit time dependence thatdetermines which of the terms Lc;Lh;Hw is coupled to thesystem at a given time. We point out that when the RWA isnot valid, the dynamics becomes considerably more com-plicated. First, even the basic unitary evolution has nosimple analytical solution. Second, the Lindblad descrip-tion of the continuous engine becomes more complicated.Thus, our analysis is restricted to the validity regime ofthe RWA.The Lindblad Markovian dynamics and the RWA set a

validity regime for our theory. This regime is the defaultregime used in quantum open systems (see Refs. [77,79]).It is intriguing to study how the results presented here are

modified by the breakdown of the RWA or by bath memoryeffects. However, this analysis is beyond the scope of thepresent paper.Now that we have established a regime of validity and

the super-Hamiltonian that governs the system, we canturn to the task of transforming from one engine type toother types and study what properties change in thistransformation. The engine-type transformation is basedon the Strang decomposition [95–97] for two noncom-muting operators A and B (the operators may not beHermitian):

eðAþBÞdt ¼ e12AdteBdte

12Adt þO½ðs=ℏÞ3& ≅ e

12AdteBdte

12Adt;

ð11Þ

where the norm action (3), s ¼ ðkAkþ kBkÞdt, must besmall for the expansion to be valid. kAk is the spectralnorm of A. In Appendix C, we derive the condition s ≪12ℏ for the validity of Eq. (11). We use the symbol ≅ todenote equality with correction O½ðs=ℏÞ3&.Let the evolution operator of the continuous engine over

the chosen cycle time τcyc ¼ 6mτd be

~Kcont ¼ e−i ~Hτcyc=ℏ: ð12Þ

By first splitting Lc and then splitting Lh, we get

~Kfour stroke ¼ e−ið3LcÞðτcyc=6ℏÞe−ið32HwÞðτcyc=6ℏÞe−ið3LhÞðτcyc=3ℏÞ

× e−ið32HwÞðτcyc=6ℏÞe−ið3LcÞðτcyc=6ℏÞ: ð13Þ

Note that the system is periodic so the first and laststages are two parts of the same thermal stroke.Consequently, Eq. (13) describes an evolution operatorof a four-stroke engine, where the unit cell is symmetric.This splitting is illustrated in Figs. 5(a) and 5(b). Thereare two thermal strokes and two work strokes thattogether constitute an evolution operator that describesa four-stroke engine. The cumulative evolution time aswritten above is ðmþmþ2mþmþmÞτd¼6mτd¼ τcyc.Yet, to maintain the same cycle time as chosen for thecontinuous engine, the coupling to the baths and fieldwere multiplied by 3. In this four-stroke engine, eachthermal or work stroke operates, in total, only a third ofthe cycle time compared to the continuous engine. Hence,the coupling must be 3 times larger in order to generatethe same evolution.By virtue of the Strang decomposition, ~Kfour stroke ≅

~Kcont if s ≪ ℏ. The action parameter s of the engine

is defined as s ¼R τcyc=2−τcyc=2 k

~Hkdt ¼ ð12 kHwkþ kLhkþkLckÞτcyc. Note that the relation ~Kfour stroke ≅ ~Kcont

holds only when the engine action is small comparedto ℏ. This first appearance of a quantum scale will bediscussed later.


031044-7











k

AkρA†k−

1

2A†kAkρ−1

2ρA†

kAk;

ð1Þ






031044-4

( )

















031044-6

123

4





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

the interaction picture (denoted by tilde) using the transformation





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

+ the rotating wave approximation (RWA) by dropping terms oscillating at a frequency of ( )





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

06/11/2016, 23)06Interaction picture - Wikipedia

Page 1 of 4https://en.wikipedia.org/wiki/Interaction_picture

Interaction pictureFrom Wikipedia, the free encyclopedia

In quantum mechanics, the interaction picture (also known as the Dirac picture) is an intermediate representationbetween the Schrödinger picture and the Heisenberg picture. Whereas in the other two pictures either the state vector or theoperators carry time dependence, in the interaction picture both carry part of the time dependence of observables.[1] Theinteraction picture is useful in dealing with changes to the wave functions and observable due to interactions. Most fieldtheoretical calculations[2] use the interaction representation because they construct the solution to the many bodySchrödinger equation as the solution to the free particle problem plus some unknown interaction parts.

Equations that include operators acting at different times, which hold in the interaction picture, don't necessarily hold in theSchrödinger or the Heisenberg picture. This is because time-dependent unitary transformations relate operators in onepicture to the analogous operators in the others.

Contents1 Definition

1.1 State vectors1.2 Operators

1.2.1 Hamiltonian operator1.2.2 Density matrix

1.3 Time-evolution equations in the interaction picture2 Expectation values

2.1 Time-evolution of states2.2 Time-evolution of operators2.3 Time-evolution of the density matrix

3 Use of interaction picture4 References5 See also

DefinitionOperators and state vectors in the interaction picture are related by a change of basis (unitary transformation) to those sameoperators and state vectors in the Schrödinger picture.

To switch into the interaction picture, we divide the Schrödinger picture Hamiltonian into two parts,

Any possible choice of parts will yield a valid interaction picture; but in order for the interaction picture to be useful insimplifying the analysis of a problem, the parts will typically be chosen so that H0,S is well understood and exactlysolvable, while H1,S contains some harder-to-analyze perturbation to this system.

If the Hamiltonian has explicit time-dependence (for example, if the quantum system interacts with an applied externalelectric field that varies in time), it will usually be advantageous to include the explicitly time-dependent terms with H1,S,leaving H0,S time-independent. We proceed assuming that this is the case. If there is a context in which it makes sense to



Interaction pictureFrom Wikipedia, the free encyclopedia

In quantum mechanics, the interaction picture (also known as the Dirac picture) is an intermediate representationbetween the Schrödinger picture and the Heisenberg picture. Whereas in the other two pictures either the state vector or theoperators carry time dependence, in the interaction picture both carry part of the time dependence of observables.[1] Theinteraction picture is useful in dealing with changes to the wave functions and observable due to interactions. Most fieldtheoretical calculations[2] use the interaction representation because they construct the solution to the many bodySchrödinger equation as the solution to the free particle problem plus some unknown interaction parts.

Equations that include operators acting at different times, which hold in the interaction picture, don't necessarily hold in theSchrödinger or the Heisenberg picture. This is because time-dependent unitary transformations relate operators in onepicture to the analogous operators in the others.

Contents1 Definition

1.1 State vectors1.2 Operators

1.2.1 Hamiltonian operator1.2.2 Density matrix

1.3 Time-evolution equations in the interaction picture2 Expectation values

2.1 Time-evolution of states2.2 Time-evolution of operators2.3 Time-evolution of the density matrix

3 Use of interaction picture4 References5 See also

DefinitionOperators and state vectors in the interaction picture are related by a change of basis (unitary transformation) to those sameoperators and state vectors in the Schrödinger picture.

To switch into the interaction picture, we divide the Schrödinger picture Hamiltonian into two parts,

Any possible choice of parts will yield a valid interaction picture; but in order for the interaction picture to be useful insimplifying the analysis of a problem, the parts will typically be chosen so that H0,S is well understood and exactlysolvable, while H1,S contains some harder-to-analyze perturbation to this system.

If the Hamiltonian has explicit time-dependence (for example, if the quantum system interacts with an applied externalelectric field that varies in time), it will usually be advantageous to include the explicitly time-dependent terms with H1,S,leaving H0,S time-independent. We proceed assuming that this is the case. If there is a context in which it makes sense to



have H0,S be time-dependent, then one can proceed by replacing by the corresponding time-evolution operator inthe definitions below.

State vectors

A state vector in the interaction picture is defined as[3]

where |ψS(t)〉is the state vector in the Schrödinger picture.

Operators

An operator in the interaction picture is defined as

Note that AS(t) will typically not depend on t, and can be rewritten as just AS. It only depends on t if the operator has"explicit time dependence", for example due to its dependence on an applied, external, time-varying electric field.

Hamiltonian operator

For the operator H0 itself, the interaction picture and Schrödinger picture coincide,

This is easily seen through the fact that operators commute with differentiable functions of themselves. This particularoperator then can be called H0 without ambiguity.

For the perturbation Hamiltonian H1,I, however,

where the interaction picture perturbation Hamiltonian becomes a time-dependent Hamiltonian—unless [H1,S, H0,S] = 0 .

It is possible to obtain the interaction picture for a time-dependent Hamiltonian H0,S(t) as well, but the exponentials need tobe replaced by the unitary propagator for the evolution generated by H0,S(t), or more explicitly with a time-orderedexponential integral.

Density matrix

The density matrix can be shown to transform to the interaction picture in the same way as any other operator. In particular,let ρI and ρS be the density matrix in the interaction picture and the Schrödinger picture, respectively. If there is probabilitypn to be in the physical state |ψn〉, then




State vectors



Operators









Density matrix





State vectors



Operators









Density matrix





State vectors



Operators









Density matrix

The density matrix can be shown to transform to the interaction picture in the same way as any other operator. In particular,let ρI and ρS be the density matrix in the interaction picture and the Schrödinger picture, respectively. If there is probabilitypn to be in the physical state |ψn〉, then 06/11/2016, 23)06Interaction picture - Wikipedia


Evolution Picture

of: Heisenberg Interaction Schrödinger

Ket state constant

Observable constant

Density matrix constant

Time-evolution equations in the interaction picture

For a general operator , the expectation value in the interaction picture is given by

Using the density matrix expression for expectation value, we will get

Expectation values

Time-evolution of states

Transforming the Schrödinger equation into the interaction picture gives:

This equation is referred to as the Schwinger–Tomonaga equation.

Time-evolution of operators

If the operator AS is time independent (i.e., does not have "explicit time dependence"; see above), then the correspondingtime evolution for AI(t) is given by

In the interaction picture the operators evolve in time like the operators in the Heisenberg picture with the HamiltonianH' =H0.

Time-evolution of the density matrix

Transforming the Schwinger–Tomonaga equation into the language of the density matrix (or equivalently, transforming thevon Neumann equation into the interaction picture) gives:



Evolution Picture

of: Heisenberg Interaction Schrödinger

Ket state constant

Observable constant

Density matrix constant

Time-evolution equations in the interaction picture

For a general operator , the expectation value in the interaction picture is given by

Using the density matrix expression for expectation value, we will get

Expectation values

Time-evolution of states

Transforming the Schrödinger equation into the interaction picture gives:

This equation is referred to as the Schwinger–Tomonaga equation.

Time-evolution of operators

If the operator AS is time independent (i.e., does not have "explicit time dependence"; see above), then the correspondingtime evolution for AI(t) is given by

In the interaction picture the operators evolve in time like the operators in the Heisenberg picture with the HamiltonianH' =H0.

Time-evolution of the density matrix

Transforming the Schwinger–Tomonaga equation into the language of the density matrix (or equivalently, transforming thevon Neumann equation into the interaction picture) gives:






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

06/11/2016, 23)08Rotating wave approximation - Wikipedia

Page 1 of 4https://en.wikipedia.org/wiki/Rotating_wave_approximation

Rotating wave approximationFrom Wikipedia, the free encyclopedia

The rotating wave approximation is an approximation used in atom optics and magnetic resonance. In thisapproximation, terms in a Hamiltonian which oscillate rapidly are neglected. This is a valid approximationwhen the applied electromagnetic radiation is near resonance with an atomic transition, and the intensity islow.[1] Explicitly, terms in the Hamiltonians which oscillate with frequencies are neglected, whileterms which oscillate with frequencies are kept, where is the light frequency and is atransition frequency.

The name of the approximation stems from the form of the Hamiltonian in the interaction picture, as shownbelow. By switching to this picture the evolution of an atom due to the corresponding atomic Hamiltonian isabsorbed into the system ket, leaving only the evolution due to the interaction of the atom with the light fieldto consider. It is in this picture that the rapidly oscillating terms mentioned previously can be neglected.Since in some sense the interaction picture can be thought of as rotating with the system ket only that part ofthe electromagnetic wave that approximately co-rotates is kept; the counter-rotating component is discarded.

Contents1 Mathematical formulation

1.1 Making the approximation2 Derivation3 References

Mathematical formulationFor simplicity consider a two-level atomic system with ground and excited states and , respectively(using the Dirac bracket notation). Let the energy difference between the states be so that is thetransition frequency of the system. Then the unperturbed Hamiltonian of the atom can be written as

.

Suppose the atom experiences an external classical electric field of frequency , given by , e.g. a plane wave propagating in space. Then under the dipole approximation

the interaction Hamiltonian between the atom and the electric field can be expressed as

,

where is the dipole moment operator of the atom. The total Hamiltonian for the atom-light system istherefore The atom does not have a dipole moment when it is in an energy eigenstate, so

This means that defining allows the dipole operator to be written as06/11/2016, 23)08Rotating wave approximation - Wikipedia


(with denoting the complex conjugate). The interaction Hamiltonian can then be shown to be (see theDerivation section below)

where is the Rabi frequency and is the counter-rotating frequency. Tosee why the terms are called `counter-rotating' consider a unitary transformation to the interaction or Diracpicture where the transformed Hamiltonian is given by

where is the detuning between the light field and the atom.

Making the approximation

This is the point at which the rotating wave approximation is made. The dipole approximation has beenassumed, and for this to remain valid the electric field must be near resonance with the atomic transition.This means that and the complex exponentials multiplying and can be considered tobe rapidly oscillating. Hence on any appreciable time scale the oscillations will quickly average to 0. Therotating wave approximation is thus the claim that these terms may be neglected and thus the Hamiltoniancan be written in the interaction picture as

Finally, transforming back into the Schrödinger picture, the Hamiltonian is given by

Another criterion for rotating wave approximation is the weak coupling condition, that is, the Rabifrequency should be much less than the transition frequency.[1]

At this point the rotating wave approximation is complete. A common first step beyond this is to remove theremaining time dependence in the Hamiltonian via another unitary transformation.

DerivationGiven the above definitions the interaction Hamiltonian is




















.



,


This means that defining allows the dipole operator to be written as





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

the resulting Liouville space super-Hamiltonian






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7









.



,


This means that defining allows the dipole operator to be written as06/11/2016, 23)08Rotating wave approximation - Wikipedia






























.



,


This means that defining allows the dipole operator to be written as





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7


the Strang decomposition (details in Appendix C) for two noncommuting operators:





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7







p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

~ ~ ~ ~ ~ ~ ~





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7







p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5

the spectral norm

~ ~ ~ ~ ~ ~ ~





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7







p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






s ¼Zτ

0

kHðtÞkdt; ð3Þ















031044-5

the spectral norm





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

~ ~ ~ ~ ~ ~ ~






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

~ ~ ~ ~ ~ ~ ~






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

APPENDIX C: STRANG DECOMPOSITIONVALIDITY

Let K be an operator generated by two noncommutingoperators A and B:

K ¼ eðAþBÞdλ; ðC1Þ

where we use dλ ¼ dt=ℏ for brevity. The splitted operatoris

Ks ¼ e12AdλeBdλe

12Adλ: ðC2Þ

Our goal is to quantify the difference between K and Ks,kKs −Kk, where k · k stands for the spectral norm. Inprinciple, other submultiplicative matrix norms can be used(such as the Hilbert-Schmidt norm). However, the spectralnorm more accurately captures aspects of quantum dynam-ics [108–111]. K can be expanded as

K ¼X ðAþ BÞndλn

n!: ðC3Þ

Ks, on the other hand, is

Ks ¼X∞

k;l;m¼0

ðA=2Þkdλk

k!Bldλl

l!ðA=2Þmdλm

m!

¼X∞

n¼0

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k

ðn − l − kÞ!dλn: ðC4Þ

Because of the symmetric splitting, the terms up to n ¼ 2(including n ¼ 2) are identical for both operators.Therefore, the difference can be written as

kKs −Kk ¼!!!!X∞

n¼3

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k

ðn − l − kÞ!dλn

−X

n¼3

ðAþ BÞndλn

n!

!!!!: ðC5Þ

Next, we apply the triangle inequality and the submulti-plicativity property to get

kKs −Kk ≤!!!!X∞

n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


þX∞

n¼3

ðkAkþ kBkÞndλn

n!

!!!!: ðC6Þ

Using the binomial formula two times, one finds

X∞

n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


¼X∞

n¼3

ðkAkþ kBkÞndλn

n!; ðC7Þ

and therefore,

kKs −Kk ≤ 2X∞

n¼3

ðkAkþ kBkÞndλn

n!

¼ 2R2½ðkAkþ kBkÞdλ&: ðC8Þ

The right-hand side is the Taylor remainder of a powerseries of an exponential with ðkAkþ kBkÞdλ as an argu-ment. The Taylor remainder formula for the exponentfunction is RkðxÞ ¼ eξðjxjkþ1Þ=ðkþ 1Þ!, where 0 ≤ ξ ≤ 1(for now, we assume x < 1). Setting k ¼ 2 and ξ ¼ 1(worst case), we finally obtain

kKs −Kk ≤e3½ðkAkþ kBkÞdλ&3 ≤

"sℏ

#3

; ðC9Þ

s ¼ðkAkþ kBkÞdt; ðC10Þ

where we call s the norm action of the evolution operator.To get an estimation where the leading non-neglected termof K, ðAþ BÞ2dλ2=2, is larger than the remainder, werequire that

ðkAþ BkÞ2dλ2=2 ≥"sℏ

#3

: ðC11Þ

Using the triangle inequality, we get the estimated con-dition for the Strang decomposition:

s ≤ ℏ=2: ðC12Þ

This condition explains why it was legitimate to limit therange of x to 1 in the remainder formula.

APPENDIX D: SYMMETRIC REARRANGEMENTTHEOREM

The goal of this appendix is to explain why theequivalence of evolution operators leads to equivalenceof work and equivalence of heat. In addition, we show whythis is also valid for transients. For the equivalence of theevolution operator, we require that the super-Hamiltonian issymmetric and that the action is small:

HðtÞ ¼ Hð−tÞ; ðD1Þ

s ¼Z

þτ=2

−τ=2kHkdt ≪ ℏ: ðD2Þ


031044-16





Ks ¼ e12AdλeBdλe

12Adλ: ðC2Þ



n!: ðC3Þ


Ks ¼X∞

k;l;m¼0

ðA=2Þkdλk

k!Bldλl

l!ðA=2Þmdλm

m!

¼X∞

n¼0

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k




n¼3

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k


−X

n¼3

ðAþ BÞndλn

n!

!!!!: ðC5Þ



n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


þX∞

n¼3

ðkAkþ kBkÞndλn

n!

!!!!: ðC6Þ


X∞

n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


¼X∞

n¼3

ðkAkþ kBkÞndλn

n!; ðC7Þ

and therefore,

kKs −Kk ≤ 2X∞

n¼3

ðkAkþ kBkÞndλn

n!




"sℏ

#3

; ðC9Þ




#3

: ðC11Þ


s ≤ ℏ=2: ðC12Þ





s ¼Z

þτ=2



031044-16





Ks ¼ e12AdλeBdλe

12Adλ: ðC2Þ



n!: ðC3Þ


Ks ¼X∞

k;l;m¼0

ðA=2Þkdλk

k!Bldλl

l!ðA=2Þmdλm

m!

¼X∞

n¼0

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k




n¼3

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k


−X

n¼3

ðAþ BÞndλn

n!

!!!!: ðC5Þ



n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


þX∞

n¼3

ðkAkþ kBkÞndλn

n!

!!!!: ðC6Þ


X∞

n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


¼X∞

n¼3

ðkAkþ kBkÞndλn

n!; ðC7Þ

and therefore,

kKs −Kk ≤ 2X∞

n¼3

ðkAkþ kBkÞndλn

n!




"sℏ

#3

; ðC9Þ




#3

: ðC11Þ


s ≤ ℏ=2: ðC12Þ





s ¼Z

þτ=2



031044-16





Ks ¼ e12AdλeBdλe

12Adλ: ðC2Þ



n!: ðC3Þ


Ks ¼X∞

k;l;m¼0

ðA=2Þkdλk

k!Bldλl

l!ðA=2Þmdλm

m!

¼X∞

n¼0

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k




n¼3

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k


−X

n¼3

ðAþ BÞndλn

n!

!!!!: ðC5Þ



n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


þX∞

n¼3

ðkAkþ kBkÞndλn

n!

!!!!: ðC6Þ


X∞

n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


¼X∞

n¼3

ðkAkþ kBkÞndλn

n!; ðC7Þ

and therefore,

kKs −Kk ≤ 2X∞

n¼3

ðkAkþ kBkÞndλn

n!




"sℏ

#3

; ðC9Þ




#3

: ðC11Þ


s ≤ ℏ=2: ðC12Þ





s ¼Z

þτ=2



031044-16

~ ~ ~ ~ ~ ~ ~





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

~ ~ ~ ~ ~ ~ ~





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7






p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

~ ~ ~ ~ ~ ~ ~





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

A. Dynamical aspect of the equivalence

The equivalence of the evolution operators ~Kfour stroke ≅~Kcont has two immediate important consequences. First,both engines have the same steady-state solution over onecycle j~ρsi:

~Kfour strokeðτcycÞj~ρsi ≅ ~KcontðτcycÞj~ρsi ¼ j~ρsi; ð14Þ

!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ

At time instances that are not integer multiples of τcyc, thestates of the engines will differ significantly (O½ðs=ℏÞ1&)since ~Kfour strokeðt < τcycÞ ≠ ~Kcontðt < τcycÞ. In otherwords, the engines are still significantly different fromeach other. The second consequence is that the two engineshave the same transient modes as well. When monitored atmultiples of τcyc, both engines will have the same relaxationdynamics to the steady state if they started from the sameinitial condition. In the remainder of the paper, when theevolution operator is written without a time tag, this meansthat we are considering the evolution operator of acomplete cycle.We point out that there are higher-order decompositions

where the correction terms are smaller than O½ðs=ℏÞ3&.However, it turns out that these decompositions inherentlyinvolve negative coefficients [98]. A negative coefficientimplies a thermal stroke of the form eþiLdt=ℏ (instead ofe−iLdt=ℏ). This type of evolution cannot be generated by aMarkovian bath. Therefore, among the symmetric decom-positions, the Strang decomposition seems to be the onlyone that can be used for decomposing Markovian thermalengine evolution operators.

B. Thermodynamic aspect of the equivalence

The equivalence of the one-cycle evolution operators ofthe two engines does not immediately imply that theengines are thermodynamically equivalent. Generally, instroke engines, the heat and work depend on the dynamicsof the state inside the cycle, which is very different(O½ðs=ℏÞ1&) from the constant state of the continuousengine. However, in this section, we show that all thermo-dynamics properties are equivalent in both engines up toO½ðs=ℏÞ3& corrections, similarly to the evolution operator.We start by evaluating the work and heat in the continuousengine. By considering infinitesimal time elements whereLc;Lh, andHw operate separately, one obtains that the heatand work currents are jcðhÞ ¼ hH0jð1=ℏÞLcðhÞj~ρsðtÞi andjw ¼ hH0jð1=2ℏÞHwj~ρsðtÞi, where hH0j ¼ jH0i† is thevectorized form of the field-free Hamiltonian H0 of thesystem [see Eq. (5)]. See Appendix A for the use of bracketnotation to describe expectation values hAi ¼ trðAρÞ ¼hAjρi. In principle, to calculate hAi in the rotatingframe using j~ρsðtÞi, hAj must be rotated as well.However, because of the property hH0jH0 ¼ 0 shown inAppendix A, hH0j is not affected by this rotation.In the continuous engine, the steady state satisfies

j~ρsðtÞi ¼ j~ρsi, so the total heat and work in the steadystate in one cycle are

FIG. 5. Graphical illustrations of the super-Hamiltonians ofvarious engines (a)–(d). The horizontal axis corresponds to time.The brick size corresponds to the strength of the coupling to thework repository or to the baths. The Hamiltonians are related toeach other by applying the Strang decomposition to the evolutionoperators (12), (13), and (20). The symmetric rearrangementtheorem ensures that in the limit of small action, any rearrange-ment that is symmetric with respect to the center and conservesthe area of each color does not change the total power and heatover one cycle.


031044-8




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

Wcont ¼!H0

""""1

2ℏHw

""""~ρs#τcyc; ð16Þ

QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ

""""~ρs#τcyc: ð17Þ

These quantities should be compared to the work andheat in the four-stroke engine. Instead of carrying outthe explicit calculation for this specific four-strokesplitting, we use the symmetric rearrangement theorem(SRT) derived in Appendix D. Symmetric rearrange-ment of a Hamiltonian is a change in the order ofcouplings ϵðtÞ; γcðtÞ; γhðtÞ that satisfies

RϵðtÞdt ¼ const,R

γcðtÞdt ¼ const,RγhðtÞdt ¼ const, and with the sym-

metry ϵðtÞ ¼ ϵð−tÞ; γcðtÞ ¼ γcð−tÞ; γcðtÞ ¼ γcð−tÞ. Anysuper-Hamiltonian obtained using the Strang splittingof the continuous engine [for example, Htwo strokeðtÞ,Hfour strokeðtÞ] constitutes a symmetric rearrangement ofthe continuous engine. The SRT exploits the symmetryof the Hamiltonian to show that symmetric rearrangementchanges heat and work only in O½ðs=ℏÞ3%. In Appendix D,we show that

Wfour stroke ≅ Wcont; ð18Þ

Qfour strokecðhÞ ≅ Qcont

cðhÞ: ð19Þ

Thus, we conclude that up to ðs=ℏÞ3 corrections, theengines are thermodynamically equivalent. When s ≪ 1,work, power, heat, and efficiency converge to the samevalue for all engine types. Clearly, inside the cycle, thework and heat in the two engines are significantly different(O½ðs=ℏÞ1%), but after a complete cycle, they becomeequivalent. The symmetry makes this equivalence moreaccurate as it holds up to ðs=ℏÞ3 [rather than ðs=ℏÞ2].Interestingly, the work done in the first half of the cycle is12W

cont þO½ðs=ℏÞ2%. However, when the contribution ofthe other half is added, the O½ðs=ℏÞ2% correction cancelsout and Eq. (18) is obtained (see Appendix D).We emphasize that the SRT and its implications (18) and

(19) are valid for transients and for any initial state—notjust for steady-state operation. In Fig. 6(a), we show thecumulative work as a function of time for a four-strokeengine and a continuous engine. The vertical lines indicatea complete cycle of the four-stroke engine. In additionto the parameter common to all examples specified before,we used ϵ ¼ γc ¼ γh ¼ 10−4, and the equivalence of workat the vertical lines is apparent. In Fig. 6(b), the field andthermal coupling were increased to ϵ ¼ γc ¼ γh ¼5 × 10−3. Now the engines perform differently, even atthe end of each cycle. This example is a somewhat extremesituation where the system changes quite rapidly (conse-quence of the initial state we chose). In other cases, such assteady-state operation, the equivalence can be observed formuch larger action values.

The splitting used in Eq. (13) was based on first splittingLc and thenHw. Other engines can be obtained by differentsplitting of ~Kcont. For example, consider the two-strokeengine obtained by splitting Lc þ Lh:

~Ktwo stroke ¼ e−i32ðLcþLhÞðτcyc=3Þe−ið32HwÞðτcyc=3Þ

× e−i32ðLcþLhÞðτcyc=3Þ: ð20Þ

Note that in the two-stroke engine, the thermal coupling hasto be 3

2 stronger compared to the continuous case in order toprovide the same action. Using the SRT, we obtain thecomplete equivalence relations of the three main enginetypes:

Wtwo stroke ≅ Wfour stroke ≅ Wcont; ð21Þ

Qtwo strokecðhÞ ≅ Qfour stroke

cðhÞ ≅ QcontcðhÞ; ð22Þ

~Ktwo stroke ≅ ~Kfour stroke ≅ ~Kcont: ð23Þ

Note that since K ¼ e−iH0τcyc ~K, the equivalence of theevolution operators holds also in the original frame, notjust in the interaction frame. Another type of engine existswhen the interaction with the work repository is carriedout by two physically distinct couplings. This happens

FIG. 6. (a) The equivalence of heat engine types in transientevolution when the engine action is small compared to ℏ. (a) Thecumulative power transferred to the work repository is plotted asa function of time. All engines start in the excited state j4i, whichis very far from the steady state of the system. At complete enginecycles (vertical lines), the power in all engines is the same.(b) Once the action is increased (here, the field ϵ was increased),the equivalence no longer holds.


031044-9




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7





p j1ih4j; ð7Þ




p j2ih3j: ð9Þ





~H ¼ Lc þ Lh þ1

2Hw: ð10Þ







12AdteBdte

12Adt;

ð11Þ













031044-7

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

naturally if E4 − E3 ≠ E2 − E1 so that two differentdriving lasers have to be used and the Hamiltonian isH0 þ cos½ðE2 − E1Þt%Hw1 þ cos½ðE4 − E3Þt%Hw2. In suchcases, one can make the splitting shown in Fig. 5(d). Inthis numerical example, we used Hw1 ¼ ϵðtÞj1ih2jþ H:c:and Hw2 ¼ ϵðtÞj3ih4jþ H:c: Since there are two differentwork strokes in addition to the thermal stroke, this engineconstitutes a four-stroke engine.

C. Power and energy flow balance

The average power and heat flow in the equivalenceregime are independent of the cycle time:

PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ

Using the steady-state definition (15), one obtains thesteady-state energy balance equation:

Pw þ Jc þ Jh ¼ 0: ð26Þ

Equation (26) does not necessarily hold if the system is notin a steady state, as energy may be temporarily stored in thebaths or in the work repository.Figure 7 shows the power in a steady state as a function

of the action. The action is increased by increasing the timeduration of each stroke (see top illustration in Fig. 7). Thefield and the thermal coupling are ϵ ¼ γh ¼ γc ¼ 5 × 10−4.The coupling strengths to the bath and work repository arenot changed. When the engine action is large compared toℏ, the engines behave very differently [Fig. 7(a)]. On theother hand, in the equivalence regime, where s is small withrespect to ℏ, the power of all engines types converges to thesame value. In the equivalence regime, the power risesquadratically with the action since the correction to thepower is s3=τcyc ∝ τ2cyc. This power plateau in the equiv-alence regime is a manifestation of quantum interferenceeffects (coherence in the density matrix), as will be furtherdiscussed in the next section.The behavior of different engines for large action with

respect to ℏ is very rich and strongly depends on the ratiobetween the field and the bath coupling strength.Finally, we comment that the same formalism and results

can be extended for the case in which the drive is slightlydetuned from the gap.

D. Lasing condition via the equivalenceto a two-stroke engine

Laser medium can be thought of as a continuous enginewhere the power output is light amplification. It iswell knownthat lasing requires population inversion. Scovil et al. [83]were the first to show the relation between the populationinversion lasing condition and the Carnot efficiency.

Using the equivalence principle presented here, the mostgeneral form of the lasing condition can be obtainedwithout any reference to light-matter interaction.Let us start by decomposing the continuous engine into

an equivalent two-stroke engine. For simplicity, it isassumed that the hot and cold manifolds have some overlapso that, in the absence of the driving field, this bath leadsthe system to a unique steady state ρ0. If the driving field istiny with respect to the thermalization rates, then the systemwill be very close to ρ0 in the steady state.To see when ρ0 can be used for work extraction, we need

to discuss passive states. A passive state is a state that isdiagonal in the energy basis, and with populations thatdecrease monotonically with energy [99]. The energy of apassive state cannot be decreased (or work cannot beextracted from the system) by applying some unitarytransformation (the Hamiltonian after the transformationis the same as it was before the transformation) [70,99].Thus, if ρ0 is passive, work cannot be extracted from thedevice, regardless of the details of the driving field (as longas it is weak and the equivalence holds).A combination of thermal baths will lead to an energy

diagonal ρ0. Consequently, to enable work extraction,

FIG. 7. Power as a function of action for various engine types in asteady state. The four-stroke variant (green line) is described inFig. 5(d). The action is increased by increasing the stroke duration(top illustration). (a) For large action with respect to ℏ, the enginessignificantly differ in performance. In this example, all engineshave the same efficiency, but they extract different amounts of heatfrom the hot bath. (b) In the equivalence regimewhere the action issmall, all engine types exhibit the same power and also the sameheat flows. The condition s < ℏ=2 that follows from the Strangdecomposition agrees with the observed regime of equivalence.The time-symmetric structure of the engines causes the deviationfrom equivalence to be quadratic in the action.


031044-10

not just the evolution operator, but all thermodynamics properties are equivalent in both engines up to




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8

not just the evolution operator, but all thermodynamics properties are equivalent in both engines up to




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

the symmetric rearrangement theorem (SRT): details in Appendix D

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

the symmetric rearrangement theorem (SRT): details in Appendix D

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9

Wcont ¼!H0

""""1

2ℏHw


QcontcðhÞ ¼

!H0

""""1

ℏLcðhÞ








cðhÞ: ð19Þ
















031044-9




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10

/ s2?





Ks ¼ e12AdλeBdλe

12Adλ: ðC2Þ



n!: ðC3Þ


Ks ¼X∞

k;l;m¼0

ðA=2Þkdλk

k!Bldλl

l!ðA=2Þmdλm

m!

¼X∞

n¼0

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k




n¼3

Xn

l¼0

Xn−l

k¼0

ðA=2Þk

k!Bl

l!ðA=2Þn−l−k


−X

n¼3

ðAþ BÞndλn

n!

!!!!: ðC5Þ



n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


þX∞

n¼3

ðkAkþ kBkÞndλn

n!

!!!!: ðC6Þ


X∞

n¼3

Xn

l¼0

Xn−l

k¼0

kA=2kk

k!kBkl

l!kA=2kn−l−k


¼X∞

n¼3

ðkAkþ kBkÞndλn

n!; ðC7Þ

and therefore,

kKs −Kk ≤ 2X∞

n¼3

ðkAkþ kBkÞndλn

n!




"sℏ

#3

; ðC9Þ




#3

: ðC11Þ


s ≤ ℏ=2: ðC12Þ





s ¼Z

þτ=2



031044-16

Let the initial state at time t ¼ −τ=2 be

j~ρii ¼ j~ρð−τ=2i: ðD3Þ

This state leads to a final state at τ=2,

j~ρfi ¼ j~ρðτ=2i: ðD4Þ

Our goal is to evaluate a symmetric expectation valuedifference of the form

dAtot ¼ ½hAðt2Þi − hAðt1Þi% þ ½hAð−t1Þi − hAð−t2Þi%¼ ½hAj~ρðt2Þi − hAj~ρðt1Þi%þ ½hAj~ρð−t1Þi − hAj~ρð−t2Þi%;

t2; t1 ≥ 0; ðD5Þ

that is, the change in the expectation value of A in thesegment ½t1; t2% and its symmetric counterpart in negativetime [e.g., the green areas in Fig. 11(a)]. When A is equal toH0, this difference will translate into work or heat. We startwith the expansion

½hAðt2Þi − hAðt1Þi% ¼ hAjKt1→t2 − Ij~ρðt1Þi

¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ

For the negative side, we get

½hAð−t1Þi − hAð−t2Þi% ¼ hAjI −K−t1→−t2 jrð−t1Þi

¼!A""""−iHð−t1Þ δtℏ þ 1

2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ

Next, we use the fact that

j~ρðt1Þi ¼ j~ρð0Þi − iZ

t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ

j~ρð−t1Þi ¼ j~ρð0Þiþ iZ

t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ

When adding the two segments, the second order termscancel out and we get

δAtot ¼ −2ihAjHðt1Þj~ρð0ÞiδtþO$%

sℏ

&3': ðD10Þ

Note that the result is expressed using j~ρð0Þi, which is notgiven explicitly. To correctly relate it to j~ρð−τ=2Þi, we haveto use the symmetric rearrangement properties of theevolution operator.

1. Symmetric rearrangement

In Fig. 11(a), there is an illustration of some time-dependent Hamiltonian with reflection symmetryHðtÞ ¼ Hð−tÞ. We use H to denote a Liouville spaceoperator which may be any unitary operation or MarkovianLindblad operation. Assume that in addition to the sym-metric bins of interest (green bins), the remainder of thetime is also divided into bins in a symmetric way so thatthere is still a reflection symmetry in the bin partitioningalso. Now, we permute the bins in the positive side asdesired and then make the opposite order in the negativeside so that the reflection symmetry is kept. An example ofsuch an operation is shown in Fig. 11(b). Because of theStrang decomposition, we know that the total evolutionoperator will stay the same under this rearrangement up tothird order:

K−τ2→

τ2¼ T sym½K%−τ

2→τ2þO

$%sℏ

&3'; ðD11Þ

where T sym½x% stands for evaluation of x after a symmetricreordering.

2. Symmetric rearrangement theorem

From Eq. (D11), we see that if the initial state is the samefor a system described by K, and for a system described by

FIG. 11. The Hamiltonians in (a) and (b) are related bysymmetric rearrangement of the time segments. Up to a smallcorrection Oðs3Þ, the change in expectation values of anobservable A that takes place during the green segments is thesame in both cases. This effect explains why work and heat arethe same in various types of engines when s is small compared toℏ (equivalence regime).


031044-17







t2; t1 ≥ 0; ðD5Þ



¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ




2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ



t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ


t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ



sℏ

&3': ðD10Þ




K−τ2→


2→τ2þO

$%sℏ

&3'; ðD11Þ






031044-17







t2; t1 ≥ 0; ðD5Þ



¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ




2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ



t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ


t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ



sℏ

&3': ðD10Þ




K−τ2→


2→τ2þO

$%sℏ

&3'; ðD11Þ






031044-17

)

)







t2; t1 ≥ 0; ðD5Þ



¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ




2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ



t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ


t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ



sℏ

&3': ðD10Þ




K−τ2→


2→τ2þO

$%sℏ

&3'; ðD11Þ






031044-17







t2; t1 ≥ 0; ðD5Þ



¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ




2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ



t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ


t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ



sℏ

&3': ðD10Þ




K−τ2→


2→τ2þO

$%sℏ

&3'; ðD11Þ






031044-17







t2; t1 ≥ 0; ðD5Þ



¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ




2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ



t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ


t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ



sℏ

&3': ðD10Þ




K−τ2→


2→τ2þO

$%sℏ

&3'; ðD11Þ






031044-17

T sym½K", the final state at t ¼ τ=2 is the same for bothsystems up to a third-order correction:

!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ

Using Eqs. (D8) and (D9), we see that

j~ρð0Þi ¼j~ρðτ2Þiþ j~ρð− τ

2Þi2

þO%"

sℏ

#2&; ðD13Þ

and because of Eq. (D12), it also holds that

T sym½j~ρð0Þi" ¼ j~ρð0ÞiþO%"

sℏ

#2&

¼j~ρðτ2Þiþ j~ρð− τ

2Þi2

þO%"

sℏ

#2&; ðD14Þ

using this in Eq. (D10), we get

δAtot ¼ −2ihAjHðt1Þj~ρðτ2Þiþ j~ρð− τ

2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ

Expression (D15) no longer depends on the position of thetime segment but only on its duration and on the value ofH.Thus, the SRT states that the expression above also holdsfor any symmetric rearrangement,

dAtot ¼ T sym½dAtot" þO%"

sℏ

#3&: ðD16Þ

If we replace A by H0 and Hðt1Þ by Lc;Lh, or Hw, weimmediately get the invariance of heat and work tosymmetric rearrangement (up to s3). If j~ρ½−ðτ=2Þ"i is thesame for all engines, then j~ρðτ=2Þi is also the same for allengine types up to Oðs3Þ. Consequently, for all strokeengines, the expressions for work and heat are

W ¼ −2ihH0jZ

t∈tw

HwðtÞdtℏj~ρðτ2Þiþ j~ρð− τ

2Þi2

þO%"

sℏ

#3&;

ðD17Þ

QcðhÞ ¼ −2ihH0jZ

t∈tcðhÞ

LcðhÞðtÞdtℏj~ρðτ2Þiþ j~ρð− τ

2Þi2

þO%"

sℏ

#3&: ðD18Þ

Using the identity j~ρðτ=2Þiþ j~ρ½−ðτ=2Þ"i ¼ j~ρðtÞiþj~ρð−tÞiþO½ðs=ℏÞ2" that follows from Eq. (D13), theintegration over time of the energy flows jw ¼hH0jð1=2ℏÞHwj~ρðtÞi and jcðhÞ ¼ hH0jð1=ℏÞLcðhÞj~ρðtÞi

for continuous engines yields expressions (D17) and(D18) once more. This implies that the SRT (D17) and(D18) holds even if the different operations Lc;Lh, andHwoverlap with each other.We emphasize that all the above relations hold for any

initial state and not only in the steady state wherej~ρðτ=2Þi ¼ j~ρ½−ðτ=2Þ"i. The physical implication is thatin the equivalence regime, different engines are thermo-dynamically indistinguishable when monitored at the endof each cycle, even when the system is not in itssteady state.

[1] R. Alicki, The Quantum Open System as a Model of theHeat Engine, J. Phys. A 12, L103 (1979).

[2] H. Spohn, Entropy Production for Quantum DynamicalSemigroups, J. Math. Phys. (N.Y.) 19, 1227 (1978).

[3] M. Campisi, P. Talkner, and P. Hänggi, FluctuationTheorem for Arbitrary Open Quantum Systems, Phys.Rev. Lett. 102, 210401 (2009).

[4] M. Campisi, P. Hänggi, and P. Talkner, Colloquium:Quantum Fluctuation Relations: Foundations and Appli-cations, Rev. Mod. Phys. 83, 771 (2011).

[5] H. T. Quan and H. Dong, Quantum Crooks FluctuationTheorem and Quantum Jarzynski Equality in the Presenceof a Reservoir, arXiv:0812.4955.

[6] F. L. Curzon and B. Ahlborn, Efficiency of a CarnotEngine at Maximum Power Output, Am. J. Phys. 43, 22(1975).

[7] I. Novikov, The Efficiency of Atomic Power Stations(A Review), J. Nucl. Energy 7, 125 (1958).

[8] M. Esposito, K. Lindenberg, and C. Van den Broeck,Universality of Efficiency at Maximum Power, Phys. Rev.Lett. 102, 130602 (2009).

[9] R. Uzdin and R. Kosloff, Universal Features in theEfficiency at Maximal Work of Hot Quantum Otto Engines,Europhys. Lett. 108, 40001 (2014).

[10] P. Salamon, J. D. Nulton, G. Siragusa, T. R. Andersen, andA. Limon, Principles of Control Thermodynamics, Energy26, 307 (2001).

[11] B. Andresen, Current Trends in Finite-Time Thermody-namics, Angew. Chem., Int. Ed. Engl. 50, 2690 (2011).

[12] R. Kosloff and T. Feldmann, A Discrete Four StrokeQuantum Heat Engine Exploring the Origin of Friction,Phys. Rev. E 65, 055102 (2002).

[13] F. Plastina, A. Alecce, T. J. G. Apollaro, G. Falcone,G. Francica, F. Galve, N. L. Gullo, and R. Zambrini,Irreversible Work and Inner Friction in QuantumThermodynamic Processes, Phys. Rev. Lett. 113,260601 (2014).

[14] A. del Campo, J. Goold, and M. Paternostro, More Bangfor Your Buck: Super-Adiabatic Quantum Engines, Sci.Rep. 4, 6208 (2014).

[15] R. Uzdin and R. Kosloff, The Multilevel Four-Stroke SwapEngine and Its Environment, New J. Phys. 16, 095003(2014).

[16] R. Kosloff, A Quantum Mechanical Open System as aModel of a Heat Engine, J. Chem. Phys. 80, 1625 (1984).


031044-18


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18







t2; t1 ≥ 0; ðD5Þ



¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ




2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ



t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ


t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ



sℏ

&3': ðD10Þ




K−τ2→


2→τ2þO

$%sℏ

&3'; ðD11Þ






031044-17


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18

/2

2







t2; t1 ≥ 0; ðD5Þ



¼!A""""−iHðt1Þ

δtℏ− 1

2Hðt1Þ2

δt2

ℏ2

""""~ρðt1Þ#

þO$%

sℏ

&3': ðD6Þ




2Hð−t1Þ2 δt

2

ℏ2

""""~ρð−t1Þ#

þO$%

sℏ

&3': ðD7Þ



t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2';

ðD8Þ


t1

0HðtÞ dt

ℏj~ρð0ÞiþO

$%sℏ

&2':

ðD9Þ



sℏ

&3': ðD10Þ




K−τ2→


2→τ2þO

$%sℏ

&3'; ðD11Þ






031044-17


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18


!!!!~ρ"τ2

#$¼ T sym

%!!!!~ρ"τ2

#$&þO

%"sℏ

#3&: ðD12Þ



2Þi2

þO%"

sℏ

#2&; ðD13Þ



sℏ

#2&


2Þi2

þO%"

sℏ

#2&; ðD14Þ



2Þi2

δtþO%"

sℏ

#3&:

ðD15Þ



sℏ

#3&: ðD16Þ


W ¼ −2ihH0jZ

t∈tw


2Þi2

þO%"

sℏ

#3&;

ðD17Þ


t∈tcðhÞ


2Þi2

þO%"

sℏ

#3&: ðD18Þ





















031044-18

/2

2

?!

quantum-thermodynamic signature: a signal that is impossible to produce by the corresponding classical engine

passivity must be broken by population inversion.Therefore, we obtain the standard population inversioncondition. Note that the derivation does not require anEinstein rate equation or any information on the processesof emission and absorption of photons.Furthermore, it now becomes clear that if “coherent

baths” are used [66] so that ρ0 is no longer diagonal in theenergy basis (and therefore no longer passive), it is possibleto extract work even without population inversion.In conclusion, using the equivalence principle, it is

possible to import known results from work extractionin stroke schemes to continuous machines.

IV. QUANTUM-THERMODYNAMIC SIGNATURE

Can the measurements of thermodynamics quantitiesreveal quantum effects in heat engines? To answer this, wefirst need to define the corresponding classical engine.The term “classical engine” is rather ambiguous. There

are different protocols of modifying the system so that itbehaves classically. To make a fair comparison to the fullyquantum engine, we look for the minimal modification thatsatisfies the following conditions:(1) The dynamics of the device should be fully de-

scribed using population dynamics (no coherences,no entanglement).

(2) The modification should not alter the energy levelsof the system, the couplings to the baths, and thecoupling to the work repository.

(3) The modification should not introduce a new sourceof heat or work.

To satisfy the first requirement, we introduce a dephasingoperator that eliminates the coherences [100] and leads to astochastic description of the engine. Clearly, a dephasingoperator satisfies the second requirement. To satisfy thethird requirement, we require “pure dephasing,” a dephas-ing in the energy basis. The populations in the energy basisare invariant to this dephasing operation. Such a naturalsource of energy-basis dephasing emerges if there is somescheduling noise [101]. In other words, if there is someerror in the switching time of the strokes.Let us define a “quantum-thermodynamic signature” as a

signal that is impossible to produce by the correspondingclassical engine as defined above.Our goal is to derive a threshold for power output that a

stochastic engine cannot exceed but a coherent quantumengine can.Before analyzing the effect of decoherence, it is instruc-

tive to distinguish between two different work extractionmechanisms in stroke engines.

A. Coherent and stochastic work extractionmechanisms

Let us consider the work done in the work stroke of atwo-stroke engine [as in Fig. 5(c)]:

W ¼ hH0je−ið1=2ℏÞHwτw j~ρi − hH0j~ρi; ð27Þ

where τw is the duration of the work stroke. Writingthe state as a sum of population and coherencesj~ρi ¼ j~ρpopiþ j~ρcohi, we get

W ¼!H0

""""X

n¼1

ð−i 12ℏHwτwÞ2n−1

ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1

ð−i 12ℏHwτwÞ2n

ð2nÞ!

""""~ρpop#: ð28Þ

This result follows from the generic structure ofHamiltonians in Liouville space. Any H that originatesfrom a Hermitian Hamiltonian in Hilbert space (in contrastto Lindblad operators as a source) has the structure shownin Fig. 8(b) (see Appendix A for Liouville space derivationof this property). In other words, it connects only pop-ulations to coherences and vice versa, but it cannotconnect populations to populations directly [102]. Inaddition, since hH0j acts as a projection on populationspace, one gets that odd powers of Hw can only operate oncoherences and even powers can only operate on popula-tions. Thus, the power can be extracted using two differentmechanisms: a coherent mechanism that operates oncoherences and a stochastic mechanism that operates onpopulations.The effects of the “stochastic” termsPn¼1ð−i 1

2ℏHwτwÞ2n=ð2nÞ! on the populations are equiv-alently described by a single doubly stochastic operator. Ifthere are no coherences (next section), this leads to a simpleinterpretation in terms of full swap events that take placewith some probability.

FIG. 8. Panel (a), left side: Dephasing operations (slanted line,operator D) commute with thermal baths so the dephased enginein the left side of (a) is equivalent to the one on the right. In thenew engine, the unitary evolution is replace by DUD. If Deliminates all coherences, the effect of DUD on the populationscan always be written as a doubly stochastic operator. (b) AnyHermitian Hamiltonian in Liouville space has the structure shownin (b). Thus, first-order changes in populations critically dependon the existence of coherence.


031044-11

consider the work done in the work stroke of a two-stroke engine




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11

consider the work done in the work stroke of a two-stroke engine




!Lc þ Lh þ

1

2Hw

"j~ρsi ¼ 0: ð15Þ








031044-8


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11

possible? uniquely?


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11

Continuous engines, on the other hand, have only acoherent work extraction mechanism. This can be seenfrom the expression for their work output,

Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ

where again we used the population projection property ofhH0j and the structure of Hw [Fig. 8(b)]. We conclude thatin contrast to stroke engines, continuous engines have nostochastic work extraction mechanism. This differencestems from the fact that in continuous engines, the steadystate is stationary. Consequently, there are no higher-orderterms that can give rise to a population-population stochas-tic work extraction mechanism. This is a fundamentaldifference between stroke engines and continuous engines.This effect is pronounced outside the equivalence regimewhere the stochastic terms become important (see Sec. V).

B. Engines subjected to pure dephasing

Consider the engine shown in Fig. 8(a). The slanted lineson the baths indicate that there is an additional dephasingmechanism that takes place in parallel to the thermalization[103]. Let us denote the evolution operator of the puredephasing by D. In principle, to analyze the deviation fromthe coherent quantum engine, first the steady state has to besolved and then work and heat can be compared. Even forsimple systems, this is a difficult task. Hence, we shall take adifferent approach and derive an upper bound for the powerof stochastic engines. It is important that the bound containsonlyquantities that are unaffected by the level of coherence inthe system. For example, the dipole expectation value, doescontain information on the coherence. We construct a boundin terms of the parameters of the system (e.g., the energylevels, coupling strengths, etc.), which is independent of thestate of the system. In the pure dephasing stage, the energydoes not change. Hence, the total energy change in theDUDstage is associated with work.Let Dcomp ¼ jpopihpopj be a projection operator on the

population space. This operator generates a completedephasing that eliminates all coherences. In such a case,the leading order in the work expression becomes

W ¼ hH0jDcompe−ið1=2ℏÞHwτwDcompj~ρi

¼ τ2w8ℏ2

hH0jH2wj~ρpopiþO½ðs=ℏÞ4&; ð30Þ

where we used hH0jD ¼ hH0j andDcompj~ρi ¼ j~ρpopi. SinceDcomp eliminates coherences, W does not contain a linearterm in time. Next, by using the following relation,hH0jBjρi ≤

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffihH0jH0ihρjρi

pkBk,

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffihH0jH0i

p¼

ffiffiffiffiffiffiffiffiffiffiffiffiffitrðH2

0Þp

,we find that for s ≪ ℏ the power of a stochastic enginesatisfies

Pstoch ≤z

8ℏ2

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffitrðH2

0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;

z ¼ 1=2 four-stroke; ð31Þ

where Δw is the gap of the interaction Hamiltonian(maximal eigenvalue minus minimal eigenvalue of Hw),and d is the duty cycle—the fraction of time dedicated towork extraction (d ¼ τw=τcyc, e.g., d ¼ 1=3 in all theexamples in this paper). We also used the fact thathρpopjρpopi is always smaller than the purity hρjρi andtherefore smaller than 1. Note that, as we required, thisbound is state independent, and the right-hand side ofEq. (31) contains no information on the coherences in thesystem. Thus, we conclude that for power measurements,

P > Pstoch ⇒ quantum-thermodynamic signature: ð32Þ

As shown earlier, in coherent quantum engines (in theequivalence regime), the work scales linearly with τcyc [seeEqs. (16) and (18)], and therefore, the power is constant as afunction of τcyc. When there are no coherences, the powerscales linearly with τcyc.Numerical results of power as a function of cycle time

are shown in Fig. 9. The power is not plotted as a functionof action as before because, at the same cycle time, thecoherent engine and the dephased engine have differentactions. The coupling parameters are as in Fig. 7. Theaction of the dephased engine is

FIG. 9. The power output of the three types of engines (two-stroke blue, four-stroke red, continuous black) with and withoutdephasing [top horizontal solid lines are without dephasing—same as in Fig. 7(b)]. The power of the continuous dephasedengine is zero. The dashed lines show the stochastic upper boundson the power of two-stroke (dashed blue line) and four-stroke(dashed red line) engines. Any power measurement in the shadedarea of each engine indicates the presence of quantum interfer-ence in the engine. This plot also demonstrates that for short cycletimes (low action), coherent engines produce much more powercompared to stochastic dephased engines.


031044-12


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11

possible? uniquely?


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


















W ¼!H0

""""X

n¼1


ð2n − 1Þ!

""""~ρcoh#

þ!H0

""""X

n¼1


ð2nÞ!

""""~ρpop#: ð28Þ





031044-11


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12

cannot connect populations to populations directly: well known in the context of the Zeno effect

NON-SELECTIVE, CONTINUOUS MEASUREMENTS 167

Let us assume that the system is initially in an eigenstate 10n ) belonging to the eigenvalue an of A,

IV)(0)) = (3.347)

For sufficiently small t we then have according to the Schrödinger equation,

110) = [I — iHt — —1 H2 t2 ± ...]111)n ). 2 (3.348)

The first ideal measurement of A is carried out at time t = 0. The probability of obtaining the eigenvalue a, in this measurement is given by

w(0) = Pni/i)(0))12 _ 1 — (AE)202 +...,

where the dots indicate terms of higher order and

(AE) , = (OniI/2 10n) — (Orilli

is the energy uncertainty in the state W). The quantity w(0) is just the probability that the system is still in the initial state PO after time O. After time T = 109, that is after k measurements this probability is thus

tOnn(T) Pe., [1 _ (AE)2no2]k . (3.351)

For large k and fixed T, that is in the limit 0 Ik = r 0, this leads to

-k [i — (AE) 2 122-

n k z,- exp [—(AE) 2n 7-0] 1. (3.352)

This equation tells us that the system remains with probability 1 in the initial state 10n ) if a continuous, ideal measurement of the observable A is carried out on the system. As a result of the continuous state reduction induced by the measurements the system cannot leave its initial state. Formally, the reason for this fact is that in the limit of small 0 the probability of leaving the state IOn ) is proportional to the square of 0, that is we have 1 — w(8) oc 0 2 , while the number k of measurements increases as 8 -1 . The state reduction induced by the succession of measurements is thus faster than possible transitions into other states. In analogy to Zeno's paradox this phenomenon is known as the quantum Zeno effect.

3.5.2 Density matrix equation The quantum Zeno effect described above results, or course, from an idealization which presupposes an ideal, continuous measurement. In order to investigate what happens for non-ideal measurements we consider the following model for an indirect continuous monitoring of the system.

on )2

(3.349)

(3.350)












































031044-20


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10

the action of the dephased enginesdeph ¼

!kLckþ kLhkþ

""""1

2Hw

""""þ kLdephasingk#τcyc:

ð33Þ

If the dephasing is significant, the action is large andequivalence cannot be observed. In other words, a fullystochastic engine in a quantum system has a large actionand cannot satisfy s ≪ ℏ.The stochastic power bounds for a two-stroke engine

(dashed blue line) and for a four-stroke engine (dashed redline) define a power regime (shaded areas) that is inacces-sible to fully stochastic engines. Thus, any power meas-urement in this regime unequivocally indicates the presenceof quantum coherences in the engine.In practice, the dephasing time may be very small but

different from zero. When the cycle time is large comparedto the dephasing time, the system behaves as if there iscomplete dephasing. If, however, the cycle time is smallwith respect to the decoherence time (close to the origin ofFig. 9), the power will form a plateau of finite power insteadof reducing to zero.Note that to measure power, the measurement is carried

out on the work repository and not on the engine.Furthermore, the engine must operate for many cycles toreduce fluctuations in the accumulated work. To calculatethe average power, the accumulated work is divided by thetotal operation time and compared to the stochastic powerthreshold (31).Also, note that a complete dephasing would have

resulted in zero power output for the continuousengine (29).In summary, the quantum-thermodynamics signature in

stroke engines can be observed in the weak action limit.

V. OVER-THERMALIZATION EFFECT INCOHERENT QUANTUM HEAT ENGINE

In all the numerical examples studied so far, the unitaryaction and the thermal action were roughly comparable forreasons that will soon become clear. In this section, westudy some generic features that take place when thethermal action takes over.Let us now consider the case where the unitary con-

tribution to the action kHωkτ is small with respect to ℏ. Allthe time intervals are fixed, but we can control thethermalization rate γ (for simplicity, we assume it is thesame value for both baths). Common sense suggests thatincreasing γ should increase the power output. At somestage, this increase will stop since the system will alreadyreach thermal equilibrium with the bath (or baths in two-stroke engines). Yet, Fig. 10 shows that there is a verydistinctive peak where an optimal coupling takes place. Inother words, in some cases, less thermalization leads tomore power. We call this effect over-thermalization. Thiseffect is generic and not unique to the specific model used

in the numerical simulations. The parameters used for theplot are ϵ ¼ γc ¼ γh ¼ 2 × 10−4, and the number of drivescycles per engine cycle is m ¼ 600.The peak and the saturation are a consequence of the

interplay between the two different work extraction mech-anisms (see Sec. IVA). For low γ, the coherences in thesystem are significant, and the leading term in the power ishH0j−ið1=2ℏÞHwj~ρcohid (where d is the duty cycle). Inprinciple, all Lindblad thermalization processes are asso-ciated with some level of decoherence. This decoherencegenerates an exponential decay of j~ρcohi that explains thedecay on the right-hand side of the peak. At a certain stage,the linear term becomes so small that the stochastic second-order term −ð1=8ℏ2ÞhH0jH2

wj~ρpopid2τcyc dominates thepower. j~ρpopi eventually saturates for large γ, and therefore,the stochastic second-order term leads to a powersaturation. Interestingly, in the example shown in Fig. 10,we observe that the peak is obtained when γ and ϵ areroughly equal. Of course, what really matters is the thermalactionwith respect to unitary action and not just the values ofthe parameters γ and ϵ. We point out that this effect for acontinuous engine can be seen in Fig. 3 of Ref. [20] and inFig. 11 of Ref. [104]. In the present work, the mechanismthat generates this general effect has been clarified.If thermalization occurs faster, the thermal stroke can be

shortened, and this increases the power. However, thiseffect is small with respect to the exponential decay of thecoherences. We conclude that even without additionaldephasing as in the previous section, excessive thermalcoupling turns the engine into a stochastic machine. Forsmall unitary action, this effect severely degrades the poweroutput. The arguments presented here are valid for anysmall-action coherent quantum engine.

FIG. 10. The over-thermalization effect is the decrease of powerwhen the thermalization rate is increased. Over-thermalizationdegrades the coherent work extraction mechanism withoutaffecting the stochastic work extraction mechanism. When thecoherent mechanism gets weak enough, the power is dominatedby the stochastic power extraction mechanisms and powersaturation is observed (dashed lines). The continuous enginehas no stochastic work extraction mechanism, and therefore, itdecays to zero without reaching saturation.


031044-13


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12


Pcont ¼!H0

""""1

2ℏHw

""""~ρ#

¼!H0

""""1

2ℏHw

""""~ρcoh#; ð29Þ






¼ τ2w8ℏ2




pkBk,


p¼


0Þp


Pstoch ≤z

8ℏ2


0Þ − trðH0Þ2q

Δ2wd2τcyc;

z ¼ 1 two-stroke;








031044-12




PW ¼ Wτcyc

¼!H0

""""1

2ℏHw

""""~ρs#; ð24Þ

JcðhÞ ¼QcðhÞ

τcyc¼

!H0

""""1

ℏLcðhÞ

""""~ρs#: ð25Þ















031044-10

the action of the dephased enginesdeph ¼

!kLckþ kLhkþ

""""1

2Hw


ð33Þ
















031044-13

sdeph ¼!kLckþ kLhkþ

""""1

2Hw


ð33Þ
















031044-13

over-thermalization effect in coherent quantum heat engine


""""1

2Hw


ð33Þ
















031044-13



""""1

2Hw


ð33Þ
















031044-13

?!



""""1

2Hw


ð33Þ
















031044-13

?!


""""1

2Hw


ð33Þ
















031044-13



""""1

2Hw


ð33Þ
















031044-13

?!


""""1

2Hw


ð33Þ
















031044-13

“over-thermalization”: less thermalization leads to more power!


""""1

2Hw


ð33Þ
















031044-13

“over-thermalization”: less thermalization leads to more power!


""""1

2Hw


ð33Þ
















031044-13

summary and conclusions• coherent and stochastic work extraction mechanisms in quantum heat engines

• stroke engines (having both) vs continuous engines (only the coherent one)• norm action all three engine types are equivalent, because only the

coherent mechanism is important• before the engine cycle is completed, the difference: • at the end of each engine cycle, the difference:

• experimental schemes• usually involving large action• however, e.g., A. O. Niskanen, Y. Nakamura, and J. P. Pekola, Information Entropic

Superconducting Microcooler, Phys. Rev. B 76, 174523 (2007): the unitary operation that makes a swap between superconducting qubits is generated by creating a magnetic flux through a superconducting ring

• no assumption on the direction of heat flows and the sign of work• equally applicable to refrigerators and heaters

• open questions: more general scenarios• non-Markovian baths, engines with a nonsymmetric unit cell, and engines

with quantum correlation between different particles (entanglement and quantum discord)

• conjecture: in multiple particle engines, entanglement will play a similar role to that of coherence in single-particle engines

O[(s/~)]

s ⌧ ~ !

O[(s/~)3]