PHYSICAL REVIEW LETTERS 120, 190602 (2018)

Universal Trade-Off between Power, Efficiency,

and Constancy in Steady-State Heat Engines
Patrick Pietzonka and Udo Seifert
II. Institut für Theoretische Physik, Universität Stuttgart, 70550 Stuttgart, Germany
(Received 8 May 2017; revised manuscript received 12 December 2017; published 11 May 2018)

Heat engines should ideally have large power output, operate close to Carnot efficiency and show
constancy, i.e., exhibit only small fluctuations in this output. For steady-state heat engines, driven by a
constant temperature difference between the two heat baths, we prove that out of these three requirements
only two are compatible. Constancy enters quantitatively the conventional trade-off between power and
efficiency. Thus, we rationalize and unify recent suggestions for overcoming this simple trade-off. Our
universal bound is illustrated for a paradigmatic model of a quantum dot solar cell and for a Brownian
gyrator delivering mechanical work against an external force.

DOI: 10.1103/PhysRevLett.120.190602

The efficiency η of heat engines operating cyclically and for an effective two-terminal device containing inelas-
between a hot heat reservoir at temperature T h and a cold tic electron-phonon processes [8]. However, as for cyclic
one at T c is universally bounded from above by the Carnot engines, it has recently been pointed out that in certain
value ηC ≡ 1 − T c =T h . Moreover, it was commonly limits Carnot efficiency at finite power can be reached in
believed that reaching Carnot efficiency inevitably comes models of steady-state engines. Specific proposals include
with a vanishing power P since such engines require quasi-
two-cycle engines with diverging affinities [9], a specially
static conditions leading to an infinite cycle-time. This
supposedly universal trade-off between power and effi- designed Feynman-Smoluchowski ratchet [10], and sys-
ciency has more recently been challenged by studies tems with a singular transport law [11,12].
hinting at the possibility to come at least arbitrarily close These observations taken together indicate that when
to Carnot efficiency with finite power by a particular searching for a universal trade-off it may not be enough to
coupling between subsystems [1] or by exploiting a work- focus only on the two characteristics power and efficiency,
ing substance at a critical point [2]. On the other hand, a and their relationship if one of them becomes maximal as it
general bound of the type P ≤ AðηC − ηÞ with a system- is typically done, see also [13–28]. As we will show in this
specific amplitude A has been proven for cyclic engines Letter, as a crucial third quantity, power fluctuations enter
both in linear response [3] and beyond [4]. These two quantitatively into this balance. Specifically, we will prove
strands of reasoning can be reconciled only if one allows for a huge class of steady-state heat engines, which includes
for a diverging amplitude A as the efficiency approaches the all thermodynamically consistent machines with a classical
Carnot value. discrete set of internal states and continuous ones modeled
For steady-state heat engines permanently coupled to with an overdamped Langevin dynamics, that there is a
two heat baths like in thermoelectric setups, the common universal trade-off between three desiderata: Finite (or even
argument from above invoking quasistatic conditions and large) power, an efficiency close to the Carnot value, and
hence an infinite cycle time is not directly applicable. It has constancy in the sense of small fluctuations in the power
usually been replaced with the idea that finite currents output are not compatible. Specifically, the bound
necessarily lead to dissipation which should spoil the
option of reaching the Carnot limit at finite power. This η Tc 1
P ≤ ð1Þ
view was challenged by a seminal paper from Benenti et al. ηC − η ΔP 2
[5], who pointed out that if time-reversal symmetry is
broken, like in the presence of a magnetic field, the usual constrains (mean) output power, efficiency, and the power
approach based on linear irreversible thermodynamics does fluctuations in finite time as measured by
not forbid Carnot efficiency at finite power. Subsequent
studies showed with somewhat more specific assumptions ΔP ≡ lim hðPðtÞ − PÞ2 it: ð2Þ
t→∞
that Carnot efficiency at finite power is not accessible. This
holds true, e.g., for any finite number of terminals in a Here, PðtÞ is the fluctuating power after time t evaluated in
multiterminal set-up within linear response theory [6,7], the steady state, for which h…i denotes averages. Since the

0031-9007=18=120(19)=190602(6) 190602-1 © 2018 American Physical Society

 PHYSICAL REVIEW LETTERS 120, 190602 (2018)

mean output work grows linearly in time as does its if this transition involves the release of an electron to a bath,
variance, converting work fluctuations into power fluctua- baij ¼ −1. In both cases, the chemical potentials enter the
tions requires the additional factor of t to reach a finite limit expression in the exponent providing a contribution to the
for ΔP, with which we characterize the constancy, or total free energy involved in such a transition. The last term
stability [29], of the engine. In particular for nanoscopic in the exponent is nonzero if this transition additionally
heat engines, power fluctuations due to thermal noise are involves a step of length dij against an external force f.
not negligible compared to the mean work output on Generalizations to several species of particles, further
relevant time scales and should therefore be taken into baths, or the case of rotary motion against an applied
account for a thermodynamic description. The crucial role torque should be obvious.
power fluctuations play in the above bound is comple- In the steady state, where state i is realized with
mentary to their effect in the statistics of efficiency in a probability pi, this engine can be characterized by three
finite time [30–33]. important mean currents, the heat current
As a main first consequence of this new bound for
X
steady-state heat engines, it is obvious that as long as the jh ¼ ðpi khij − pj khji ÞðEj − Ei − bhij μh Þ ð5Þ
power fluctuations remain finite (and T c as well), approach- i<j
ing Carnot efficiency implies that the power has to vanish at
least linearly. The explicit occurrence of ΔP as an ampli- from the hot bath to the engine, the heat current
tude in a putative linear relationship between power and
X
efficiency gap to Carnot, however, offers a second option. If jc ¼ − ðpi kcij − pj kcji ÞðEj − Ei − bcij μc Þ ð6Þ
the fluctuations blow up, then a finite (or even diverging) i<j
power is not ruled out as η → ηC , which unifies quantita-
tively the various observations recalled above. from the engine to the cold bath, and the “work current”
The bound (1) can be rearranged as a bound on efficiency
jw ¼ jh − jc ¼ P; ð7Þ
ηC
η≤ ; ð3Þ
1 þ 2PT c =ΔP which, due to the first law, is the power delivered by the
engine. Running the engine for a finite time t, each of the
determined by mean and fluctuations of the output power.
currents will fluctuate around these mean values with a
Thus, the efficiency of any steady-state heat engine is
dispersion [38]
bounded from above by this simple expression independent
of the specific design of the engine. The formal similarity of
Dα ≡ lim hðjα ðtÞ − jα Þ2 it=2; ð8Þ
(3) with a bound derived for the efficiency of molecular t→∞
motors [34] indicates as common origin of these bounds the
thermodynamic uncertainty relation [35,36], which where α ¼ h, c, w. The mean entropy production rate
describes a universal inequality between entropy produc- becomes
tion and mean and variance of an arbitrary current.
For a proof of the universal trade-off (1), we consider an σ ¼ jc =T c − jh =T h ¼ jw ðηC =η − 1Þ=T c : ð9Þ
engine characterized by a set of internal states fig with
internal energies fEi g. A transition between state i and j Since this Markovian network is thermodynamically
takes place with a rate kcij if it is mediated by the contact to consistent, one can directly apply the thermodynamic
the cold bath and with a rate khij if it is mediated by the hot uncertainty relation, which reads for any of the three
currents [35,36]
bath. One of these rates can be zero, which means that this
particular transition always involves the other bath. For any
σDα ≥ j2α : ð10Þ
nonzero rate, the corresponding backward rate does not
vanish either, and it obeys the local detailed balance
Evaluating this relation for the work current, α ¼ w, leads,
condition [37]
with ΔP ¼ 2Dw , (7) and (9), to the bound (1).
Two related, but not identical, forms of this bound can be
kc;h
ij
¼ exp½ðEi − Ej þ bc;h
ij μc;h − fdij Þ=T c;h ; ð4Þ derived similarly by applying the thermodynamic uncer-
kc;h
ji tainty relation to either the heat current from the hot bath or
the one entering the cold bath. Expressed as bound on
where we set Boltzmann’s constant to 1 throughout the power, they read explicitly
Letter. Here, Ei;j are the internal energies of the two states.
If the transition from i to j requires the transport of an P ≤ ðηC − ηÞηDh =T c ð11Þ
electron from the bath with temperature T a ða ¼ c; hÞ and
chemical potential μa to the system, then baij ¼ 1. Likewise, and

190602-2
 PHYSICAL REVIEW LETTERS 120, 190602 (2018)
(a) (b)
two states can occur either nonradiatively through inter-
actions with the surrounding phonon bath at temperature T c
with rates kclr and kcrl or are mediated by the black body
radiation of the sun with temperature T h at rates khlr and khrl .
The transition rates to and from the two electron reservoirs
(i ¼ l, r) are chosen according to the Fermi-Dirac distri-
(c) bution as

kc0i ¼ Γi =½1 þ expðxi Þ; kci0 ¼ Γi =½1 þ expð−xi Þ ð13Þ

with frequencies Γi and affinities xi ¼ ðEi − μi Þ=T c . The

transition rates between the two states in contact with the
cold phonon bath or hot photon bath (a ¼ c, h, respec-
tively) are determined by the Bose-Einstein distribution

kalr ¼ Γa =½expðxa Þ − 1; karl ¼ Γa =½1 − expð−xa Þ; ð14Þ

with frequencies Γj and affinities xa ¼ ðEr − El Þ=T a . All

four pairs of rates satisfy the local detailed balance
condition (4).
FIG. 1. (a) Model for a photoelectric device, transporting In a network representation of this system [Fig. 1(b)], we
electrons between two reservoirs through a quantum dot with can identify two independent cycle currents. The variables
two energy levels. Transitions to and from the baths (blue solid X1 and X2 count the net number of electrons transported
arrows) and nonradiative transitions between the energy levels from state l to state r via radiative and nonradiative
(blue dashed arrows) occur in contact with the cold bath. transitions, respectively. The total number of transported
Radiative transitions (red curved arrows) occur in contact with electrons is thus Xe ¼ X1 þ X2 (up to a single electron
the hot bath. (b) The network representation of the three possible
that might still be in the two-level system, which does not
states of the quantum dot allows for an identification of the two
cycle currents associated with radiative (j1 ) and nonradiative affect the statistics of fluctuations in the long-time limit).
transitions (j2 ). (c) Output power and efficiency (inset) of the The joint fluctuations of these counting variables are
photoelectric device as a function of the scaling parameter Ω characterized by the average currents hX1;2 ðtÞi ¼ j1;2 t
entering the rates and affinities according to xc ¼ 10, xh ¼ 0.2, and their covariance hðXi ðtÞ − ji tÞðXj ðtÞ − jj tÞi ¼ 2Dij t
xl ¼ 1, xr ¼ 1.2 þ 7=Ω, Γl ¼ Γr ¼ Γh ¼ ΩΓ0 , Γc ¼ Ω−1.5 Γ0 , for large t, which can be calculated directly from the rates
where Γ0 is apfrequency of reference. The power fluctuations,
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (13) and (14) [40].
quantified by ΔP =ð10Γ0 Þ, are shown as a grey shaded region. For the mean power, mean heat current from the hot
The three bounds on the output power (1), (11), (12) are shown as reservoir, and mean heat current into the cold reservoir, one
dashed, dash-dotted and dotted curves, respectively. obtains

P ≤ ðηC − ηÞηDc =½ð1 − ηÞ2 T c ; ð12Þ P ¼ Δμðj1 þ j2 Þ; ð15Þ

respectively. Obviously, in order to reach a finite value for jh ¼ ΔEj1 ; ð16Þ

the power as η → ηC , the fluctuations in all three currents
have to diverge at least ∼1=ðηC − ηÞ since each of the three jc ¼ ðΔE − ΔμÞj1 − Δμj2 ; ð17Þ
bounds (1), (11), (12) has to be respected.
For an illustration of these bounds, we consider a simple respectively, with ΔE ≡ Er − El and Δμ ≡ μr − μl . For the
but instructive model for a solar cell introduced in Ref. [39] corresponding dispersion coefficients, one gets
as shown in Fig. 1(a). It consists of a two level quantum dot
Dw ¼ Δμ2 ðD11 þ D22 þ 2D12 Þ; ð18Þ
that can either be empty (state 0) or contain an electron in
one of the levels El < Er (states l and r, respectively).
Electrons are transported through the dot from a left Dh ¼ ΔE2 D11 ; ð19Þ
reservoir with chemical potential μl and temperature T c
to a right reservoir with higher chemical potential μr and Dc ¼ ðΔE − ΔμÞ2 D11 þ Δμ2 D22 − 2ðΔE − ΔμÞΔμD12 :
the same temperature T c . They enter the level El from the ð20Þ
left reservoir at a rate kc0l , and they jump back at a rate kcl0 .
Analogously, the level Er is connected to the right reservoir In the “strong coupling” limit of negligible nonradiative
via the transition rates kcr0 and kc0r . Transitions between the transitions (Γc → 0 and hence X2 , j2 , D22 , and D12 → 0),

190602-3
 PHYSICAL REVIEW LETTERS 120, 190602 (2018)

the efficiency of the photoelectric device becomes (a) (b)

η ¼ Δμ=ΔE. By gradually increasing the chemical poten-

tial difference Δμ with fixed ΔE, this ratio can approach ηC
from below. In Fig. 1(c), this limit is realized by increasing
the chemical potential μr via a scaling parameter Ω while
keeping μl , the energies, and the temperatures fixed, as
detailed in the caption. We show that finite power can be
achieved in this limit by increasing the rates Γl , Γr , and Γh
while reducing Γc [44]. Crucially, as predicted by relation
(1), the fluctuations of the output power diverge in such a
scenario.
The uncertainty relation (10) becomes tight in the linear
response limit for unicyclic networks [35]. Hence, for
strong coupling and vanishing cycle affinity FIG. 2. (a) Schematic trajectory of a Brownian gyrator with the
two-dimensional potential Vðx1 ; x2 Þ shown as contour lines.
kc0l khlr kcr0 The anisotropic noise leads to a gyration in counter-clockwise
ln ¼ xr − xl − xh ¼ ðηC ΔE − ΔμÞ=T c ; ð21Þ direction against the external force with parameter k. (b) Scaled
kc0r khrl kcl0 power (top) and efficiency (bottom) as a function of the force
parameter k. The respective bounds (11) and (3) become tighter
as realized for a large scaling parameter Ω in Fig. 1(c), all with decreasing c. Parameters: T c ¼ 1, T h ¼ 7, μ ¼ 1, u1 ¼ 1,
three bounds on the power output (1), (11), (12) saturate. u2 ¼ 1.2.
Beyond the linear response and strong-coupling limit,
the bounds can become weaker. As explored in the
Supplemental Material [40], depending on the values of thermodynamics [37], along an individual trajectory we
the transition frequencies, each of the three bounds can identify the work
become the strongest one, demonstrating that these are
independent bounds. đw ¼ −dx · Fext ðxÞ; ð23Þ
Since the thermodynamic uncertainty relation (10) has
also been proven for overdamped Langevin dynamics performed against the external force, the heat
[45,46], the resulting trade-offs apply as well to steady-
state heat engines modeled on such a continuous state đqc ¼ dx1 ½−∂VðxÞ=∂x1 þ Fext
1 ðxÞ ð24Þ
space. For an analytically solvable illustration, we consider
the so-called Brownian gyrator [47], which has recently dissipated into the cold heat bath, and the heat
been realized experimentally in an electronic [48] and in a
colloidal system [49]. The setup, shown in Fig. 2(a), đqh ¼ −dx2 ½−∂VðxÞ=∂x2 þ Fext
2 ðxÞ ð25Þ
consists of a point particle with mobility μ in two
dimensions and an anisotropic harmonic potential VðxÞ ¼ extracted from the hot heat bath. The corresponding
pffiffiffiffiffiffiffiffiffiffi
ðu1 x21 þ u2 x22 Þ=2 þ cx1 x2 with u1;2 > 0 and 0 < c < u1 u2 . integrated currents follow from these differentials through
Its motion obeys the Langevin equation the Stratonovich integration scheme.
The average mechanical power can be calculated as [40]
dx=dt ¼ μ½−∂V=∂x þ Fext ðxÞ þ ζðtÞ; ð22Þ  
k k
P ¼ jw ¼ hđw=dti ¼ 4Pmax 1− ; ð26Þ
where the noise term has mean hζðtÞi ¼ 0 and correlations ks ks
hζ i ðtÞζ j ðt0 Þi ¼ 2μT i δij δðt − t0 Þ; i.e., the two components of
the fluctuating force are associated with heat reservoirs at with the stall parameter ks ≡ cηC =ð2 − ηC Þ and the maxi-
two different temperatures T i . We choose T 1 ≡ T c < mal power Pmax ≡ μk2s ðT c þ T h Þ=ð2ðu1 þ u2 ÞÞ. The sys-
T 2 ≡ T h . In the established case with Fext ðxÞ ¼ 0, the tem operates as a heat engine, delivering mechanical work
coupling c ≠ 0 in the potential mediates a steady transfer of for 0 < k < ks. Its efficiency is then given by
heat from the hot to the cold reservoir, that is accompanied
by a persistent directed gyration of the particle [47]. In η ¼ jw =jh ¼ 2k=ðc þ kÞ ≤ ηC : ð27Þ
order to extract mechanical work from this gyration, we
load the engine with a nonconservative external counter The diffusion coefficients Dw;c;h of the three currents can be
force Fext ðxÞ. As the simplest, linear realization of such a calculated analytically as well. It turns out that the
force, we choose Fext ðxÞ ¼ ðFext 1 ; F2 Þ ¼ kðx2 ; −x1 Þ
ext T T
uncertainty relation (10) becomes the same for each
with a parameter k. Following the rules of stochastic current,

190602-4
 PHYSICAL REVIEW LETTERS 120, 190602 (2018)

Dw σ Dc σ Dh σ ½cηC ð1 − k=ks Þ2 Carnot (or less than 1 for isothermal machines), can be
¼ ¼ ¼ 1 þ ≥ 1; ð28Þ
j2w j2c j2h ðu1 þ u2 Þ2 ð1 − ηC Þ bounded or shown to vanish. Exploring this issue, using the
techniques introduced here, will be left to future work.
with the entropy production rate σ as defined in Eq. (9). The Finally, it will be interesting to investigate whether and how
resulting bounds on the power and efficiency are shown in constancy enters bounds for genuine quantum heat engines
Fig. 2(b). These bounds become strong when the ratio that exploit coherences—see [54] and references therein.
Dw σ=j2w in Eq. (28) is close to one, as it is the case for
moderate temperature differences or small couplings c with
respect to the parameters u1;2 . For k close to ks , the bounds
[1] A. E. Allahverdyan, K. V. Hovhannisyan, A. V. Melkikh,
on both power and efficiency become tight in linear order in
and S. G. Gevorkian, Carnot Cycle at Finite Power: Attain-
ðks − kÞ, independently of all other parameters. Because of ability of Maximal Efficiency, Phys. Rev. Lett. 111, 050601
a tight coupling between heat and work currents, this (2013).
regime corresponds to a weak driving of the system within [2] M. Campisi and R. Fazio, The power of a critical heat
linear response, for which the uncertainty relation can engine, Nat. Commun. 7, 11895 (2016).
generally be saturated. For k → ks, as the efficiency [3] K. Brandner, K. Saito, and U. Seifert, Thermodynamics of
approaches the Carnot limit, the power vanishes. Since Micro- and Nano-Systems Driven by Periodic Temperature
the constancy of the power remains finite, this typical Variations, Phys. Rev. X 5, 031019 (2015).
behavior of a heat engine can be interpreted as a conse- [4] N. Shiraishi, K. Saito, and H. Tasaki, Universal Trade-Off
quence of the universal relation (1). The unusual case of Relation between Power and Efficiency for Heat Engines,
Phys. Rev. Lett. 117, 190601 (2016).
finite power close to Carnot efficiency can yet be obtained
[5] G. Benenti, K. Saito, and G. Casati, Thermodynamic
by scaling the mobility μ, which comes at the cost of Bounds on Efficiency for Systems with Broken Time-
diverging power fluctuations [40]. Reversal Symmetry, Phys. Rev. Lett. 106, 230602 (2011).
In summary, we have derived a bound providing a [6] K. Brandner and U. Seifert, Multi-terminal thermoelectric
universal trade-off between power, efficiency, and con- transport in a magnetic field: Bounds on Onsager coef-
stancy, i.e., fluctuations in the power output. A finite (or ficients and efficiency, New J. Phys. 15, 105003 (2013).
even diverging) power as Carnot efficiency is approached, [7] K. Brandner and U. Seifert, Bound on thermoelectric power
necessarily requires the latter to diverge. The three versions in a magnetic field within linear response, Phys. Rev. E 91,
of the bound hold beyond the linear response regime of a 012121 (2015).
small temperature difference between the two heat reser- [8] K. Yamamoto, O. Entin-Wohlman, A. Aharony, and N.
Hatano, Efficiency bounds on thermoelectric transport in
voirs. We have derived these results for steady-state engines
magnetic fields: The role of inelastic processes, Phys. Rev.
described by thermodynamically consistent Markovian B 94, 121402 (2016).
dynamics on a discrete state space and for overdamped [9] M. Polettini and M. Esposito, Carnot efficiency at divergent
Langevin dynamics. They also apply to apparently non- power output, Europhys. Lett. 118, 40003 (2017).
Markovian heat engines, provided that there is a suffi- [10] J. S. Lee and H. Park, Carnot efficiency is reachable in an
ciently fine-grained level of description on which the irreversible process, Sci. Rep. 7, 10725 (2017).
system obeys Markovian stochastic dynamics, however [11] N. Shiraishi, Attainability of Carnot efficiency with autono-
complex this description might be. The generalization to mous engines, Phys. Rev. E 92, 050101 (2015).
underdamped Langevin dynamics might be somewhat [12] J. Koning and J. O. Indekeu, Engines with ideal efficiency
more subtle. and nonzero power for sublinear transport laws, Eur. Phys. J.
B 89, 248 (2016).
One might expect that similar results could be derived for
[13] F. L. Curzon and B. Ahlborn, Efficiency of a Carnot engine
cyclic, i.e., periodically driven, heat engines [15] that can at maximum power output, Am. J. Phys. 43, 22 (1975).
be experimentally realized for colloidal systems [50,51] [14] C. Van den Broeck, Thermodynamic Efficiency at Maxi-
and with active baths [52]. In particular, a certain formal mum Power, Phys. Rev. Lett. 95, 190602 (2005).
analogy of (1) with the bound derived in [4] for cyclic [15] T. Schmiedl and U. Seifert, Efficiency at maximum power:
engines is striking. However, the analysis of a periodically An analytically solvable model for stochastic heat engines,
driven Brownian clock in [53] shows that the steady-state Europhys. Lett. 81, 20003 (2008).
bound (10), relating entropy production, mean current, and [16] Z. C. Tu, Efficiency at maximum power of Feynman’s
dispersion, cannot naively be extended to periodically ratchet as a heat engine, J. Phys. A 41, 312003 (2008).
driven systems. Therefore, it remains an exciting open [17] M. Esposito, K. Lindenberg, and C. Van den Broeck,
Universality of Efficiency at Maximum Power, Phys.
question whether there are periodically driven heat engines
Rev. Lett. 102, 130602 (2009).
that beat the, for steady-state engines universal, bound (1). [18] N. Nakpathomkun, H. Q. Xu, and H. Linke, Thermoelectric
We have focused on the question whether Carnot efficiency at maximum power in low-dimensional systems,
efficiency can be reached with finite power. More generally, Phys. Rev. B 82, 235428 (2010).
one can ask for any given class of machines whether the [19] U. Seifert, Efficiency of Autonomous Soft Nano-Machines
power at the maximal efficiency, which may be less than at Maximum Power, Phys. Rev. Lett. 106, 020601 (2011).

190602-5
 PHYSICAL REVIEW LETTERS 120, 190602 (2018)

[20] Y. Izumida and K. Okuda, Efficiency at maximal power of [38] The fluctuating
P currents jα ðtÞ are formallyPdefined as
minimal nonlinear irreversible heat engines, Europhys. Lett. jh ðtÞ ¼ i<j nhij ðtÞðEj −Ei −bhij μh Þ=t, jc ðtÞ ¼ − i<j ncij ðtÞ×
97, 10004 (2012). P
ðEj −Ei −bcij μc Þ=t, and jw ðtÞ ¼ i<j ½−bhij nhij ðtÞðμh − μc Þ þ
[21] N. Golubeva and A. Imparato, Efficiency at Maximum
ðnhij ðtÞ þ ncij ðtÞÞfdij =t, with naij ðtÞ being the net number of
Power of Interacting Molecular Machines, Phys. Rev. Lett.
109, 190602 (2012). transitions between i and j, mediated by the hot (a ¼ h) or
[22] C. de Tomas, J. M. M. Roco, A. C. Hernández, Y. Wang, cold (a ¼ c) bath up to the time t.
and Z. C. Tu, Low-dissipation heat devices: Unified trade- [39] B. Rutten, M. Esposito, and B. Cleuren, Reaching optimal
off optimization and bounds, Phys. Rev. E 87, 012105 efficiencies using nanosized photoelectric devices, Phys.
(2013). Rev. B 80, 235122 (2009).
[23] J. Stark, K. Brandner, K. Saito, and U. Seifert, Classical [40] See Supplemental Material (including Refs. [41–43]) at
Nernst Engine, Phys. Rev. Lett. 112, 140601 (2014). http://link.aps.org/supplemental/10.1103/PhysRevLett.120
[24] R. S. Whitney, Most efficient Quantum Thermoelectric .190602 for detailed calculations for the quantum dot solar
at Finite Power Output, Phys. Rev. Lett. 112, 130601 cell and the Brownian gyrator.
(2014). [41] Z. Koza, General technique of calculating the drift velocity
[25] J.-H. Jiang, Thermodynamic bounds and general properties and diffusion coefficient in arbitrary periodic systems,
of optimal efficiency and power in linear responses, Phys. J. Phys. A 32, 7637 (1999).
Rev. E 90, 042126 (2014). [42] M. Baiesi, C. Maes, and K. Netočný, Computation of
[26] O. Raz, Y. Subaşı, and R. Pugatch, Geometric Heat Engines current cumulants for small nonequilibrium systems,
Featuring Power that Grows with Efficiency, Phys. Rev. J. Stat. Phys. 135, 57 (2009).
Lett. 116, 160601 (2016). [43] H. Touchette, Introduction to dynamical large deviations of
[27] M. Bauer, K. Brandner, and U. Seifert, Optimal perfor- Markov processes, Physica (Amsterdam) A, DOI: 10.1016/
mance of periodically driven, stochastic heat engines under j.physa.2017.10.046, 2018.
limited control, Phys. Rev. E 93, 042112 (2016). [44] A similar scaling has been used to achieve Carnot efficiency
[28] K. Proesmans, B. Cleuren, and C. Van den Broeck, Power- at divergent power output in Ref. [9].
Efficiency-Dissipation Relations in Linear Thermodynam- [45] M. Polettini, A. Lazarescu, and M. Esposito, Tightening the
ics, Phys. Rev. Lett. 116, 220601 (2016). uncertainty principle for stochastic currents, Phys. Rev. E
[29] V. Holubec, An exactly solvable model of a stochastic heat 94, 052104 (2016).
engine: optimization of power, power fluctuations and [46] T. R. Gingrich, G. M. Rotskoff, and J. M. Horowitz, Infer-
efficiency, J. Stat. Mech. (2014) P05022; V. Holubec and ring dissipation from current fluctuations, J. Phys. A 50,
A. Ryabov, Work and power fluctuations in a critical heat 184004 (2017).
engine, Phys. Rev. E 96, 030102(R) (2017). [47] R. Filliger and P. Reimann, Brownian gyrator: A Minimal
[30] G. Verley, M. Esposito, T. Willaert, and C. Van den Broeck, Heat Engine on the Nanoscale, Phys. Rev. Lett. 99, 230602
The unlikely Carnot efficiency, Nat. Commun. 5, 4721 (2007).
(2014). [48] K.-H. Chiang, C.-L. Lee, P.-Y. Lai, and Y.-F. Chen,
[31] T. R. Gingrich, G. M. Rotskoff, S. Vaikuntanathan, and P. L. Electrical autonomous Brownian gyrator, Phys. Rev. E
Geissler, Efficiency and large deviations in time-asymmetric 96, 032123 (2017).
stochastic heat engines, New J. Phys. 16, 102003 (2014). [49] A. Argun, J. Soni, L. Dabelow, S. Bo, G. Pesce, R.
[32] M. Polettini, G. Verley, and M. Esposito, Efficiency Sta- Eichhorn, and G. Volpe, Experimental realization of a
tistics at All Times: Carnot Limit at Finite Power, Phys. Rev. minimal microscopic heat engine, Phys. Rev. E 96,
Lett. 114, 050601 (2015). 052106 (2017).
[33] K. Proesmans and C. Van den Broeck, Stochastic efficiency: [50] V. Blickle and C. Bechinger, Realization of a micrometre-
five case studies, New J. Phys. 17, 065004 (2015). sized stochastic heat engine, Nat. Phys. 8, 143 (2012).
[34] P. Pietzonka, A. C. Barato, and U. Seifert, Universal bound [51] I. A. Martínez, E. Roldán, L. Dinis, D. Petrov, J. M. R.
on the efficiency of molecular motors, J. Stat. Mech. (2016) Parrondo, and R. A. Rica, Brownian Carnot engine, Nat.
124004. Phys. 12, 67 (2016).
[35] A. C. Barato and U. Seifert, Thermodynamic Uncertainty [52] S. Krishnamurthy, S. Ghosh, D. Chatterji, R. Ganapathy,
Relation for Biomolecular Processes, Phys. Rev. Lett. 114, and A. K. Sood, A micrometre-sized heat engine operating
158101 (2015). between bacterial reservoirs, Nat. Phys. 12, 1134 (2016).
[36] T. R. Gingrich, J. M. Horowitz, N. Perunov, and J. L. [53] A. C. Barato and U. Seifert, Cost and Precision of Brownian
England, Dissipation Bounds All Steady-State Current Clocks, Phys. Rev. X 6, 041053 (2016).
Fluctuations, Phys. Rev. Lett. 116, 120601 (2016). [54] K. Brandner, M. Bauer, and U. Seifert, Universal
[37] U. Seifert, Stochastic thermodynamics, fluctuation theo- Coherence-Induced Power Losses of Quantum Heat
rems, and molecular machines, Rep. Prog. Phys. 75, 126001 Engines in Linear Response, Phys. Rev. Lett. 119,
(2012). 170602 (2017).

190602-6