Carnot efficiency is attainable in an irreversible process

Jae Sung Lee and Hyunggyu Park

School of Physics and Quantum Universe Center,
Korea Institute for Advanced Study, Seoul 02455, Korea
(Dated: November 24, 2016)
In thermodynamics, there exists a conventional belief that the Carnot efficiency is reachable
arXiv:1611.07665v1 [cond-mat.stat-mech] 23 Nov 2016

only when a process is reversible. However, there is no theorem proving that the Carnot efficiency
is impossible in an irreversible process. Here, we show that the Carnot efficiency is attainable in
an irreversible process through investigation of the Feynman-Smoluchowski ratchet (FSR). Thus,
this finding gives us a new possibility to develop a novel design of thermodynamic engines with
high efficiency regardless of the reversibility. Our result also answers the long-standing question of
whether the Carnot efficiency is possible in the FSR.

PACS numbers: 05.70.-a, 05.40.-a, 05.70.Ln

Thermodynamics is a field of science dealing with the T1 T2

relationship between energy, work, and heat [1]. It was
practically initiated to develop a heat engine with high
efficiency. Here, the heat engine is a device transform-
ing heat energy into useful mechanical work. Therefore,
one of the most interesting subjects in thermodynamics
is the study of the maximum possible efficiency attain-
able by a heat engine. The maximum efficiency of a heat
engine operating in two thermal baths of different tem-
peratures T1 and T2 (T1 > T2 ) is fairly well understood;
the efficiency cannot be greater than the Carnot efficiency
C = 1 T2 /T1 [2].
The efficiency reaches C when the process of the heat
engine is perfectly reversible [2]. Formally, defining Q1
and Q2 as the average heat transferred from thermal
baths at temperatures T1 and T2 during one engine cy-
cle over a time duration cyc , respectively, then, the ef-
ficiency and the entropy production per cycle S are
defined as
|Q2 | |Q1 | |Q2 |
1 , S + . (1)
|Q1 | T1 T2
The 2nd law of thermodynamics guarantees that S 0,
with the equality satisfied only for a reversible process.
It is easy to see that = C for a reversible process, and
it has been widely accepted that the Carnot efficiency is
impossible for an irreversible process, since C > when
S > 0.
However, such exact reversible dynamics do not exist in
the real world. Therefore, the attainability of the Carnot
efficiency should be decided through a limiting process.
It is convenient to rewrite Eq. (1) as
T2 S
C = . (2)
|Q1 |
Approaching a reversible process means the limit S
0, which can be realized in a quasi-static process [3]. In
this limit, the Carnot efficiency is reachable with zero en-
tropy production. We note that there is another limiting
process S/|Q1 | 0 with positive entropy production,
pawl
x0
F
h y
forward
vanes
backward x
FIG. 1. Schematic of the FSR model. One-dimensional vanes
(the pawl) move only horizontally (vertically), and are in con-
tact with thermal baths T1 (T2 ), respectively. x is the position
of one vane, y is the height from the bottom of the vanes to
the tip of the pawl, F is a constant external force, x0 is the
distance between neighboring vanes, h is height of a vane, and
is angle of the pawl. The pawl is pulled down by a spring.
i.e., S & O(1). In this case, the efficiency will also
approach C , which means that the Carnot efficiency is
reachable in an irreversible process. However, no such
example has yet been discovered, so it has been com-
monly misunderstood that C is only attainable in the
reversible limit.
In this work, we study the efficiency of the Feynman-
Smoluchowski ratchet (FSR) [4, 5] and find that there
indeed exists a limiting process satisfying S/|Q1 | 0
with non-vanishing entropy production. There are vari-
ous configurations of the FSR setup and here we use the
FSR model introduced by Sekimoto [6], which is illus-
trated in Supplemental Material (SM) Fig. 1. This FSR
can be modeled as shown in Fig. 1. The one-dimensional
vanes move only horizontally and the pawl moves only
vertically. They are in contact with thermal baths T1
and T2 , respectively, and the ratcheting interaction oc-
curs outside of the baths. x is the position of one vane
and y is the height from the bottom of vanes to the tip
of the pawl. x and y are stochastic variables due to ther-
mal noise. Since the pawl cannot penetrate the bottom,
y 0. The vanes and the pawl are pulled by the con-
stant external force F and the harmonic force ky, re-

(a) pawl-open (b) pawl-closed
pawl
h y
vanes
(c) (d) no-collision collision
forward
potential
potential
backward
Fx0
nx0 (n+1)x0 nx0
x x
FIG. 2. Schematics of the pawl-open and pawl-closed states.
(a) Pawl-open state (y > h). (b) Pawl-closed state (y h).
(c) Potential in the pawl-open state. Only a linear potential
with slope F is felt by the vanes. (d) Potential in the pawl-
closed state when nx0 x < (n + 1)x0 (n is an integer).
An infinite potential wall at nx0 prevents a backward hop.
There are two regions: the no-collision (nx0 < x < nx0 + a)
and collision regions (nx0 + a x < (n + 1)x0 ), depending
on whether a collision between the vanes and the pawl takes
place. U0 is the potential energy of the pawl at y = h.
spectively. Here, the direction against F is defined as
forward. x0 is the distance between neighboring vanes,
h is the height of a vane, and is the angle of the pawl.
Then, the corresponding Langevin equation can be writ-
ten as
vane:v = x,
mv = F gv (x, y) 1 v + 1 , (3)
pawl:u = y,
mp u = gp (x, y) ky 2 u + 2 (y 0),(4)
where m and mp are the masses of the vanes and the pawl
respectively, i is the damping coefficient of heat bath i,
and i (t) is the Gaussian noise of heat bath i at time t
satisfying hi (t)j (t )i = 2i Ti ij (t t ) (the Boltzmann
constant is set to kB = 1). gv (x, y) and gp (x, y) denote
the forces applied to the vanes and the pawl, respectively,
which are generated when a vane and the pawl collide or
interact. The detailed forms of the forces are given in
SM Fig. 2.
We define two states in this model: the pawl-open and
pawl-closed states as shown in Fig. 2(a) and (b), respec-
tively. In the pawl-open state (y > h), both forward and
backward hopping movements of the vanes are possible.
Here, one hop denotes movement of x from nx0 < x <
(n + 1)x0 (n is an integer) to n x0 < x < (n + 1)x0
(n = n 1). Since gv (x, y) = 0 in this state, only a lin-
ear potential with slope F is felt by the vanes, as shown
in Fig. 2(c). Furthermore, no energy is transferred from
the vanes to the pawl in this state.
2
In the pawl-closed state (y h), the pawl completely
forbids a backward hop of the vanes at x = nx0 . This
blockage is felt by the vanes as an infinite potential bar-
pawl
rier located at x = nx0 , as illustrated in Fig. 2(d). Note
that no energy is transferred to the pawl by this block-
ing collision because a horizontal force does not induce
h y any (vertical) y displacement. For nx0 < x < nx0 + a
(a = x0 h), there is no collision between the vanes and
vanes the pawl, i.e., gv (x, y) = gp (x, y) = 0. Therefore, the
vanes feels only a linear potential of slope F in this re-
gion. For nx0 + a x < (n + 1)x0 , a collision can occur,
forward so an extra interaction potential will be felt by the vanes.
U0 In this collision region, some energy is transferred from
backward
the vanes to the pawl by the collision, which is eventually
Fx0 dissipated as heat Qcol into the heat bath 2. For vanes
going over (n+1)x0 , the pawl will be lifted up to y = h by
nx0+a (n+1)x0 a vane. Then, the vanes will overcome an energy barrier
of height U0 +F x0 with U0 = kh2 /2. In this one-step for-
ward hopping process, the energy delivered to the pawl
from the vanes is U0 , which is dissipated as heat Qhop in
the heat bath 2, i.e., Qhop = U0 [7]. For later discussion
we define the average time for one forward hop as hop .
We now consider the following limit:
T2 < T1 U0 , F x0 . (5)
In the large U0 limit (large k or h), the FSR will almost
always be in the pawl-closed state due to huge energy bar-
riers. Along with very large F x0 , the vanes will rarely
reach the collision region against a very steep energy hill.
Therefore, in the above limit, the vanes will spend most of
their time in the no-collision region (nx0 < x < nx0 + a).
Then, Eq. (3), the dynamics of vanes, can be approxi-
mately written as
vane: v = x,
mv = F 1 v + 1 (x nx0 ), (6)
with an infinite energy barrier at x = nx0 . Similarly,
Eq. (4) can be practically written as
pawl: u = y,
mp u = ky 2 u + 2 (y 0), (7)
with an infinite energy barrier at y = 0. This implies that
the steady-state probability distributions of the vanes
and the pawl are almost the same as the equilibrium
distributions of the Langevin equations 6 and 7, respec-
tively, in the limit of Eq. (5). Hence, the probabilities for
the pawl-open and pawl-closed states, po and pc , become
Z r r
2k ky 2
T2 UT0
po dy e 2T2
e 2 ,
h T 2 U 0
pc = 1 po 1, (8)
respectively. This will be demonstrated numerically later
for the case of small p0 . Note that all higher-order cor-
rections are exponentially small in U0 .
In this limit, we can estimate the power hW is and heat
dissipation rate into the heat bath 2 hQ2 is , where h is
 3

denotes the steady-state average. These can be written Using Eqs. (9), (10), and (12), we calculate the effi-
as ciency and entropy production in the limit of Eq. (5)
and mp /m 1. First, the efficiency is given by
hW is = (rf rb )F x0 ,
hW is (1 e )F x0
hQ 2 is = hQ col is + hQ hop is , (9) (, ) = ,(13)
is + hQ 2 is
hW (1 e )F x0 + U0
where rf and rb are the rates of forward and backward
hopping, respectively. The power hW is is the work rate where rb /rf e with
of lifting the load hanging from the axle of the vanes. The r
U0 U0 + F x0 No T2
heat dissipation can be separated into two terms based = and = . (14)
on collisions and hopping, as discussed before. T2 T1 Nc U0
First, consider the rates in the pawl-closed state. Since To be a useful heat engine (positive work extraction
the dynamics are almost identical to the equilibrium against the load), 1 e should be larger than zero.
state, the rate of forward hopping that overcomes an en- Moreover, since F x0 > 0, we have the condition for as
ergy barrier of height U0 + F x0 can be estimated from
the Arrhenius rate equation as rf,c Nc pc e(F x0 +U0 )/T1 ln < < m , (15)
where Nc is the hopping-attempt frequency [8, 9]. The
backward hopping rate rb,c is simply zero in this case. where
In the pawl-open state, both forward and backward hops  
1 1
are possible, but the forward hopping rate rf,o is exponen- m = U0 . (16)
T2 T1
tially smaller ( eF x0 /kB T1 ) than the backward hopping
rate rb,o for large F . So, the backward hopping rate will For fixed U0 , T1 , and T2 , we find the maximum effi-
be almost identical to the hopping-attempt frequency, ciency by varying F x0 (or ) in the range of Eq. (15):
i.e., rb,o No po with exponentially small corrections. d()/d|= = 0. The result is
Note that No and Nc can differ, but this difference will    
not be very large. The vanes spend most of time in the 1
1 ln m

ln m + ln 1 ln m + O ,
pawl-closed state and become fully relaxed. As the pawl 2 m 2 m
opens for a very short period ( hop po 1/U0 ), we (17)
expect that the vane statistics does not deviate signifi- which is well inside of the range of Eq. (15). Plugging
cantly from the full relaxed one. Therefore, No /Nc can Eq. (17) into Eq. (13), it is easy to see
be reasonably assumed to be a constant of O(1). Then,
we have T22 ln U0
(
, ) C , (18)
U0 +F x0
2T1 U0

rf = rf,o + rf,c rf,c Nc e T1
,
r where we used rb /rf |= = e 1/U0 . This clearly
T2 UT0 shows that the Carnot efficiency C can be reached in the
rb = rb,o + rb,c = rb,o No e 2. (10)
U0 limit of Eq. (5) with F x0 U0 (T1 /T2 1) (T1 /2) ln U0
U0 1

Now consider hQ 2 is . Since Qhop = U0 , hQ hop is = 21 ln m ) and rf Nc e T2 + 2 ln U0 .

( =
rf,c U0 . Qint originates from the energy transfer due to At = ,
the entropy production during one hopping
the collision between a vane and the pawl in the collision period S becomes
region of the pawl-closed state. In our elastic collision !
model (SM Fig. 2), one can easily show that the trans- 1 hQ 2 is hQ 1 is 1
S = hop hSis ln U0 , (19)
ferred energy per collision is proportional to mp /m for rf T2 T1 2
mp /m 1. The collision frequency cannot be calcu-
lated simply in our setup, nevertheless we may expect where hop ( 1/rf ) is the average time for one forward
s is the steady state entropy production rate, and
hop, hSi
hQ col is = F (mp /m), (11) hQ1 is = hQ 2 is + hW
is is the average heat transfer rate
from heat bath 1. Since Eq. (19) can be very large, the
where F (z) is a vanishing function as z goes to zero. FSR operates definitely in an irreversible process, while
Equation (11) will be checked numerically later. Then, retaining the Carnot efficiency. In terms of Eq. (2), both
we can make hQ col is arbitrarily small compared to S and |Q1 | during one hopping period diverge, but in
hQ hop is , i.e. hQ col is hQ hop is , by taking an appro- a different manner to S ln U0 and |Q1 | U0 , thus
priately small value of mp /m. Therefore, in this limit, its ratio approaches zero in the U0 limit (see also
we get Eq. (18)), which is in contrast to the conventional re-
versible Carnot engine. We note that hW is rf U0 and
hQ 2 is hQ hop is rf U0 , (12)
hSis rf ln U0 are vanishingly small due to exponentially
 4

101 h = 10. Since U0 = kh2 /2 = 100 is much larger than

100
the thermal energy, no forward or backward hops can
occur within our simulation time ( = 7.5 107 ), so x
P(x)

10-1
10-2 remains between 0 and x0 when n = 0. Figures 3(a)
10-3
(a) F=2 (b) F=20 and (c) show the probability distributions of x and y for
0 1 2 3 0 0.2 0.4 0.6 0.8 1 F = 2, respectively. They show clear deviations from
x x the equilibrium distributions (solid lines), as expected
100 for small F . However, for F = 20, we can see perfect
agreement in Fig. 3(b) and (d). We also checked the va-
10-1
lidity of Eq. (11). To facilitate more collisions, a lighter
P(y)

10-2 load (small F ) was used, by setting F = 1, with x0 = 2,

10-3 (c) F=2 (d) F=20 k = 100 and h = 1. Since U0 = 50 is still too large,
0 1 2 3 0 1 2 3
no hopping occurs ( = 2 109 ) and the heat dissipa-
y y tion into heat reservoir 2 is solely from energy transfer
10-1 through collisions: hQ 2 is = hQ col is . Figure 3(e) shows
(e) the log-log plot for hQ col is versus mp /m, which shows
a power-law scaling of hQ col is with an exponent 0.27,
<Qcol>s

which confirms Eq. (11). It is infeasible to measure the

efficiency numerically for large U0 because hop grows ex-
.

10-2 ponentially with U0 . Instead, we obtained efficiency data

10-3 10-2
mp/m
FIG. 3. Numerical results. (A) and (B ) show the probability
distributions of x for F = 2 and F = 20, respectively. (C )
and (D) show the probability distributions of y for F = 2
and F = 20, respectively. Solid curves denote equilibrium
distributions of Eqs. 6 and 7, respectively. (E ) Log-log plot
of hQ col is versus mp /m. The dashed line is a guide line with
slope 0.27.
diverging hopping period hop , which is similar to an or-
dinary Carnot engine operating in a quasi-static way.
We performed a numerical simulation to check the
validity of our theory. In this simulation, we numeri-
cally integrated the Langevin Eqs. (3) and (4), using a
second-order integrator [10]. To implement the interac-
tion forces gv (x, y) and gp (x, y), we assumed that colli-
sions between a vane and the pawl were elastic and in-
stantaneous (see SM Fig. 2). For convenience, we used
dimensionless variables byprescaling time, length, and en-
ergy in units of 0 /k0 , T2 /k0 , and T2 , respectively.
Here 0 and k0 are constants with dimensions of the
damping coefficient and the spring R constant, respectively.
Heat can be calculated as Q1 = 0 dt v (1 v + 1 ) and
R
Q2 = 0 dt u(2u+2 ) during , where denotes the
Stratonovich integral [6, 1113]. Then, hQ i is Qi / in
a steady state. For convenience, we take T1 = 2, T2 = 1,
m = 10, 1 = 2 = 1, and = 45 .
We first check whether the vanes and the pawl are al-
most always in equilibrium without the collision forces
described by Eqs. (6) and (7), respectively, in the limit
of Eq. (5). We set mp = 0.1, x0 = 11, k = 2, and
at U0 = 5, which are presented in SM Fig. 3. This result
also supports our theory.
10-1 100 In conclusion, we have described a heat engine that
can operate with the Carnot efficiency in an irreversible
process. The key observation is that the entropy pro-
duction divergence is weaker than the divergence of the
heat currents to attain the Carnot efficiency. Therefore,
this finding gives us a new possibility to develop a novel
design of thermodynamic engines, especially for micro-
scopic ones actively studied recently [1416], with high
efficiency regardless of the reversibility. Note that our
conclusion is different from the result of Ref. [17] which
shows that the efficiency can be close to Carnot efficiency
at finite entropy production rate for long but finite time.
However, our results are consistent with the recent rig-
orous bound claiming that power should go to zero when
the efficiency approaches the Carnot efficiency [18].
Our result also answers the long-standing debate on
the attainability of the Carnot efficiency in the FSR.
Feynman first claimed that the efficiency of the FSR
can reach C in a reversible process in the 1960s [5].
However, in 1996, Parrodo and Espa nol [19] criticized
Feynmans argument, stating that the reversible pro-
cess was in fact not reversible, and that irreversible heat
flows must exist in the process [20]. They also claimed
that, when a system is connected to two different ther-
mal baths simultaneously, the dynamics should be al-
ways irreversible, thus the Carnot efficiency is unattain-
able. Ever since, the FSR efficiency has been discussed
in a great number of studies [2123], and the claim made
by Parrodo and Espa nol is generally accepted [2426].
However, our conclusion clearly disproves this claim and
shows that the FSR can attain the Carnot efficiency even
though the dynamics are irreversible.
This research was supported by the NRF grant No.
2011-35B-C00014 (JSL).

[1] E. Fermi, Thermodynamics, Dover edition (Courier Cor-
poration, 1959).
[2] C. Kittel and H. Kroemer, Thermal physics, 2nd Ed. (W.
H. Freeman and Company, 1980), Ch. 8.
[3] F. L. Curzon and B. Ahlborn, Am. J. Phys. 43, 22(1975).
[4] M. von Smoluchowski, Phys. Zeitschr. 13, 1069 (1912) .
[5] R. P. Feynman, The Feynman Lectures on Physics, Vol.
1. (Massachusetts, USA: Addison-Wesley) (1963), Ch. 46.
[6] K. Sekimoto, Prog. Theor. Phys. 130, 17 (1998) .
[7] In principle, Qhop can be larger than U0 . However, in
the U0 limit, the probability for Qhop > U0 is
exponentially small. Thus, it is sufficient to set Qhop = U0
in the zeroth-order calculation.
[8] P. Hanggi, P. Talkner, and M. Borkovec, Rev. Mod. Phys.
62, 251 (1990).
[9] S. A. Arrhenius, Z. Phys. Chem. 4,226 (1889).
[10] E. Vanden-Eijnden and G. Ciccotti, Chem. Phys. Lett.
429, 310 (2006).
[11] T. Hatano and S. -I. Sasa, Phys. Rev. Lett. 86, 3463
(2001).
[12] J. D. Noh and J. -M. Park, Phys. Rev. Lett. 108, 240603
(2012).
[13] J. S. Lee, C. Kwon, and H. Park, Phys. Rev. E 87,
020104(R) (2013).
5
Rold
[14] I. A. Martnez, E. an, L. Dinis, D. Petrov, J. M. R.
Parrondo, and R.A. Rica, Nature Phys. 12, 67 (2016).
[15] V. Blickle and C. Bechinger, Nature. Phys. 8, 143 (2012).
[16] J. V. Koski, V. F. Maisi, J. P. Pekola, and D. V. Averin,
Proc. Natl. Acad. Sci. 111, 13786 (2014).
[17] M. Polettini, G. Verley, and M. Esposito, Phys. Rev.
Lett. 114, 050601 (2015).
[18] N. Shiraishi, K. Saito, and H. Tasaki, preprint
arXiv:1605.00356v1 (2016).
[19] J. M. R. Parrondo and P. Espa nol, Am. J. Phys. 64, 1125
(1996) .
[20] This was also argued independently and demonstrated
numerically in K. Sekimoto, Phys. Soc. Jpn. 66, 1234
(1997).
[21] M. O. Magnasco and G. Stolovitzky, J. Stat. Phys. 93,
615 (1998).
[22] P. Reimann, Phys. Rep. 361, 57 (2002) and references
therein.
[23] P. Hanggi P and F. Marchesoni, Rev. Mod. Phys. 81, 387
(2009) and references therein.
[24] C. Van den Broeck, Carnot efficiency revisited. Adv.
Chem. Phys. 135, 189 (2007).
[25] I. Derenyi and R. D. Astumian, Phys. Rev. E 59, R6219
(1999).
[26] T. Hondou, K. Sekimoto, Phys. Rev. E 62, 6021 (2000).
 6

T1 vanes T2
y
x

pawl
F

-sU(y) y

x pawl

vanes
cross-sectional image

FIG. 4. (Supplemental Material Fig.1) Schematic of the FSR. The FSR consists of two components: vanes and a pawl.
Both are in contact with different thermal baths of temperatures T1 and T2 , and their ratcheting interaction occurs outside of
the baths as shown in the boxed area. Ratcheting is achieved by interaction between the symmetric vanes and an angled pawl
in our representation, while it takes place between a ratchet wheel with angled teeth and a simple pawl in the original FSR [5].
However, both provide essentially the same rectifying function. Since only rotational motion is allowed, the dynamics of the
vanes and the pawl can be described by their angles x and y, respectively, which are stochastic variables due to thermal noise.
A restoring force U pulls down the pawl and a constant load F hangs on the axle of the vanes.
 7

(a)
gp G

-gv
gv

-G -gp

(b)

-G=-gv G=gv

FIG. 5. (Supplemental Material Fig.2) Schematic of the elastic collision model. (a) Collision occurred when vanes move
in the forward direction. When a vane and the pawl collide together, action and reaction forces denoted by G and G acting
on the pawl and a vane respectively, which are orthogonal to the inclined plane of the pawl, are produced. gv (= G cos ) and
gp (= G sin ) are its horizontal and vertical components, respectively. Since only vertical (horizontal) motion is allowed for
the pawl (vanes) by fixed boundary conditions, the horizontal (vertical) component of G should be canceled out by boundary
forces and their motions are affected only by gv (gp ) as described in Eq. (4) (Eq. (5)) in the main text. Collision is almost
instantaneous, so the velocity change during collision are governed by action-reaction forces only: mv gv and mp u gp .
With = 45 , we get mv + mp u 0, which looks like a momentum-conservation law in one dimensional motion of two particles
with mass m and mp . Assuming the elastic collision, we have the energy conservation equation as dt d
[mv 2 /2 + mp u2 /2] = 0.
Using these two equations, we can calculate velocities after collision without any detailed knowledge of action-reaction forces.
(b) Collision occurred when vanes move in the backward direction. In this case, only horizontal forces are produced, which
does not affect the motion of the pawl, i.e., gp = 0. To the vane, gv produces an infinite potential barrier.

0.3

0.2
0.01
0.1
(F)

0.1
mp/m=1
0

-0.1

0 0.2 0.4 0.6 0.8 1

-1 -1
Fx0U0 (T1/T2-1)

FIG. 6. (Supplemental Material Fig.3) Plot of the efficiency as a function of F . The horizontal axis is rescaled as
F x0 U01 (T1 /T2 1)1 = 1 /m . As U0 = 5 is rather small, many forward hops occur as well as a lot of collision events.
At finite mp /m, hQ col is is not negligible and affects the efficiency negatively. The solid line is drawn by Eq. (14) in the main
text at the value of U0 = 5, which should be correct only in the U0 and mp /m 0 limits. Thus, this line is not the
asymptotic line as mp /m approaches zero, but can be used as an upper-bound guide line. As expected, the efficiency increases
and seems to approach the vicinity of the guide line as mp /m decreases. This strongly supports that our theory works well.
One data point is obtained from averaging efficiencies in 10 steady states and we used = 2 1011 .