Physica A 496 (2018) 137–146

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

Performance analysis for minimally nonlinear irreversible

refrigerators at finite cooling power
Rui Long *, Zhichun Liu, Wei Liu *
School of Energy and Power Engineering, Huazhong University of Science and Technology, 1037 Luoyu Road, Wuhan 430074, China

highlights

• The minimally nonlinear model is adopted to analyze the performance of refrigerators .

• COP and its bounds at given cooling power for different operating regions are deduced.
• COP bounds under the χ figure of merit at finite cooling power are calculated.
• Relative gain in COP in different operating regions is presented.

article info a b s t r a c t
Article history: The coefficient of performance (COP) for general refrigerators at finite cooling power have
Received 7 July 2017 been systematically researched through the minimally nonlinear irreversible model, and
Received in revised form 29 October 2017 its lower and upper bounds in different operating regions have been proposed. Under the
Available online 1 January 2018
tight coupling conditions, we have calculated the universal COP bounds under the χ figure
of merit in different operating regions. When the refrigerator operates
√ in the region with
Keywords:
lower external flux, we obtained the general bounds (0 < ε < ( 9 + 8εC − 3)/2) under
Minimally nonlinear irreversible
refrigerators the χ figure of merit. We have also calculated the universal bounds for maximum gain in
Coefficient of performance COP under different operating regions to give a further insight into the COP gain with the
Finite cooling power cooling power away from the maximum one. When the refrigerator operates in the region
located between maximum cooling power and maximum COP with lower external flux,
the upper bound for COP and the lower bound for relative gain in COP present large values,
compared to a relative small loss from the maximum cooling power. If the cooling power
is the main objective, it is desirable to operate the refrigerator at a slightly lower cooling
power than at the maximum one, where a small loss in the cooling power induces a much
larger COP enhancement.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction

Refrigerators are essential thermodynamic devices in our everyday life, which are widely used for cooling and refrig-
eration systems. Driven by the external power, a refrigerator absorbs heat from the cold reservoir with temperature Tc ,
and release it into the hot reservoir with temperature Th . Based on the second law of thermodynamics, the coefficient of
performance (COP) for a traditional refrigerator is bounded by the Carnot limit, εC = Tc /(Th − Tc ) [1], which is usually
achieved with infinite cycle duration, meanwhile the cooling power vanishes, which is unrealistic for actual applications

* Corresponding authors.
E-mail addresses: r_long@hust.edu.cn (R. Long), w_liu@hust.edu.cn (W. Liu).

https://doi.org/10.1016/j.physa.2017.12.112
0378-4371/© 2018 Elsevier B.V. All rights reserved.
138 R. Long et al. / Physica A 496 (2018) 137–146

although the attainability of Carnot efficiency at a non-zero power in some special cases has been reported recently [2–
7]. Pioneered by Curzon and Ahlborn [8], who considered finite cycle duration in studying practical heat engines, finite
time thermodynamics provides an alternative way to study and optimize heat devices to fulfill actual demands. And
many models have been developed to describe the operation characteristics of actual thermodynamic cycles, such as the
endoreversible model [8–10], low-dissipation model [11–13], irreversible models based on the Onsager relation [14–17],
quantum models [18–22], and Brownian models [23–25]. If the protocol driving the engine is optimized to reach maximum
amount of work during isothermal branches of the cycle, the stochastic heat engine can be tuned to fit the low-dissipation
model [26–28], which is only a special case of the minimally nonlinear irreversible model [17].
The universal optimization criterion for analyzing the performance of refrigerators is widely discussed. Many criteria have
been proposed to obtain the general bounds for the coefficient of performance to describe the actual operating conditions of
the refrigerators, such as the maximum cooling power criterion [29], the maximum COP criterion [30], the Ω criterion [31],
the χ criterion [12] and the ecological criterion [32]. And useful results have been provided. Under the maximum cooling
power condition, the COP lies in the range 0 < ε < εC /(2 + εC ) [33]. However it is not in accord with the operating
performance of actual refrigerators. Further study reveals that the real-life working conditions do not correspond to a
maximum cooling power but rather to a trade-off between maximum cooling power and maximum COP [29,34]. The Ω
criterion represents a compromise between the energy benefit and losses [10,17], and the χ criterion (the product of the
COP and the cooling power) indicates a trade-off between the maximum cooling power and maximum COP [35]. Through
√ the low dissipation refrigerator, Wang et al. [36] proposed that the COP at maximum χ was bounded between 0
studying
and ( 9 + 8εC − 3)/2, which is also√ obtained through the minimally nonlinear irreversible refrigerators [16]. And the CA
coefficient of performance εCA = εC + 1 − 1 is recovered under the conditions of symmetric dissipations. By investigating
general endoreversible refrigerators with non-isothermal processes, the CA coefficient of performance is also deduced under
maximum χ criterion [9]. In addition, Long and Liu [22] analyzed the performance of a quantum Otto refrigerator coupled
to a squeezed cold reservoir using the χ figure of merit. They claimed that the COP under the maximum χ figure of merit is
of the Curzon–Ahlborn (CA) style that cannot surpass the actual Carnot COP.
However, the above χ figure of merit is an ad-hoc criterion, the general universal optimization criterion should be
further investigated. There probably does not exist a general universal optimization criterion, but one always take a criterion
which fits his specific application. This means that the knowledge of maximum coefficient of performance at any finite
cooling power is even more important, which could offer special guidance to operate actual refrigerators under the desired
conditions. Motivated by the study of the efficiency at an arbitrary power output [28,37–41], and as a counterpart of our
previous paper on studying the performance of the minimally nonlinear heat engines at finite power [42]. In present paper,
we first introduce the model of minimally nonlinear refrigerators, and systemically discuss the coefficient of performance
and relative gain in coefficient of performance at any arbitrary cooling power for minimally nonlinear refrigerators,
respectively. Under the tight coupling conditions, the COP bounds at any finite cooling power are proposed and the COP
bounds in different operating regions under the χ figure of merit are deduced. Furthermore, under the non-tight coupling
conditions, general COP bounds and the bounds for the relative gain in COP at any arbitrary cooling power are also proposed,
and some extended discussions are presented to give a deep insight into the general COP for finite cooling power. Finally,
some important conclusions are drawn.

2. Minimally nonlinear irreversible refrigerators

For refrigerators, the cooling load (Q̇c ) is absorbed from the cold reservoir (Tc ), and a certain amount of heat (Q̇h ) is
evacuated to the hot reservoir (Th ) at the end of a cycle. After a cycle, the working fluid in the refrigerators returns to its
initial state; therefore, its entropy change per cycle is zero. The total entropy production rate σ̇ of the refrigerator can be
written as
( )
Q̇h Q̇c Ẇ 1 1
σ̇ = − = + Q̇c − (1)
Th Tc Th Th Tc
where the dot denotes the quantity per unit time for simultaneous refrigerators (working in a time-independent steady
state), or the quantity divided by the cycle time duration for sequential refrigerators (working in a time-periodic steady
state). Eq. (1) suggests considering a driving force X1 = F /Tc associated with the external force F performing work with
thermodynamically conjugate variable x and a flux J1 = ẋ, so that the input power can be rewritten as Ẇ = F ẋ = J1 X1 Th ; the
other thermodynamic force and its conjugate flux can also be defined as X2 = 1/Th − 1/Tc and J2 = Q̇c , respectively. In linear
irreversible refrigerators, the relations of the thermodynamic fluxes and forces are governed by linear relations [43]. By
adding a nonlinear term to the linear relations to consider the power dissipation into the heat reservoirs due to the fraction
loss, the minimally nonlinear irreversible model was proposed, and extended Onsager relations are adopted to illustrate the
refrigerators [16,44,45]:

J1 = L11 X1 + L12 X2 (2)

J2 = L21 X1 + L22 X2 − γc J12 (3)

 R. Long et al. / Physica A 496 (2018) 137–146 139

Fig. 1. Schematic illustration of the minimally nonlinear irreversible refrigerator.

Due to J12 = ẋ2 , quadratic nonlinear terms represent the reduce of the mechanical energy. It can be seen as the friction
losses, which are transformed into the heat, and are then transferred to the heat reservoirs (hot and cold reservoirs).
Therefore the irreversibility takes in form of J12 which has the dimension of energy. It is like the irreversible Joule heat
q = I 2 R, where I is the current and R is the internal resistance. Here we assume the dissipation is still weak, and does
not make the processes far from the equilibrium, which means the model is still in the near equilibrium state. Hence the
characteristics of the Onsager coefficients in Ref. [14,46] still hold in Eqs. (2) and (3). The Onsager coefficients (Lij ) with the
reciprocity L12 = L21 satisfy the relations: L11 ≥ 0, L22 ≥ 0, L11 L22 − L12 L21 ≥ 0. For most (or even all) physical systems,
Onsager coefficients depend on temperature and bias [47–51]. In this paper, for analytical investigation, the assumption
that Onsager coefficients is independent of the temperature and bias. The nonlinear term −γc J12 is interpreted as power
dissipation due to the fraction losses, which is transformed into the heat, and then transferred to the cold reservoir, where
γc denotes its strength (γc > 0). The heat evacuated to the hot reservoir can be calculated as

Q̇h = Q̇c + Ẇ = Q̇c + J1 X1 Th ≡ J3 (4)

According to Eq. (2), Eqs. (3) and (4) can be rewritten as

L21
J2 = J1 + L22 (1 − q2 )X2 − γc J12 (5)
L11

L21 Th
J3 = J1 + L22 (1 − q2 )X2 + γh J12 (6)
L11 Tc
√
where q = L12 / L11 L22 is the dimensionless coupling strength (|q| ≤ 1 [14]). q = 1 indicates that the heat leakage from the
high temperature heat reservoir to the low temperature one (the second terms in Eqs. (5) and (6)) vanishes. The nonlinear
term γh J12 characterizes power dissipation due to the fraction losses, which is transformed into the heat, and then transferred
to the hot reservoir, where γh = Th /L11 − γc denotes its strength (γh > 0). The existence of the two nonlinear terms results
in the decrease of the heat absorbed from the cold reservoir and the increase of the heat released to the heat reservoir.
The schematic illustration of the minimally nonlinear irreversible refrigerator is plotted in Fig. 1, where the heat fluxes and
dissipations mentioned above have been presented.
The input power can be rewritten as
L21 Th
Ẇ = J1 + J12 (7)
L11 εC L11
and the coefficient of performance is given by

Q̇c
L21
J
L11 1
+ L22 (1 − q2 )X2 − γc J12
ε= = L21 Th 2
(8)
Ẇ J
L11 εC 1
+ J
L11 1
140 R. Long et al. / Physica A 496 (2018) 137–146

To begin with, we first investigate the performance of the refrigerator operating at the maximum cooling power (R ≡ Q̇c ).
By taking the derivative of R with respect to J1 , we let ∂ R/∂ J1 = 0. Furthermore, the second derivative of R satisfies
∂ 2 R/∂ J12 < 0 at J1,max R , which means the cooling power R achieves its maximum value at J1,max R . Then, we have
L21
J1,max R = (9)
2γc L11
Substituting Eq. (9) into Eqs. (5) and (7) yields to the maximum cooling power and the corresponding power input
L221 L22 (1 − q2 )
Rmax = − εC (10)
4γ 2
c L11 Th

L221 2
Ẇmax R = ( + γh /γc + 1) (11)
4γ 2
c L11 εC
Thus we arrive at the coefficient at maximum cooling power
(1−q2 )
1− 4
γh /γc +1 q2 C ε
εmax R = (12)
2
εC
+ γh /γc + 1
According to Eqs. (5) and (10), after some algebra, the ratio of any finite cooling power and the maximum one can be
written as
(1−q2 )
R 2J
J1
− (J J1
)2 − 4
γh /γc +1 q2
ε C
1,max R 1,max R
= (13)
(1−q2 )
Rmax 1− 4
γh /γc +1 q2 C ε
Based on Eq. (13), we have
J1 √
=1± −Aδ R (14)
J1,max R
(1−q2 )
where A = 1 − γ /γ4 +1 q2 εC , and δ R = (R − Rmax )/Rmax is the relative deviation from the regime of maximum cooling
h c
power. Therefore, −1 ≤ δ R ≤ 0, which means any finite cooling power could not surpass the maximum one. The plus and
minus signs in Eq. (14) correspond to the regions with larger and lower external thermodynamic fluxes (J1 > J1,max R and
J1 < J1,max R ), respectively.
Similarly, the relative gain in the coefficient of performance δε can be defined as δε = (ε −εmax R )/εmax R . Based on Eqs. (8)
and (12), the ratio of the COP at any finite cooling power and that corresponding to the maximum one can be written as
ε R Ẇmax R 2 + (γh /γc + 1)εC
= = (1 + δ R) √ √ (15)
εmax R Rmax Ẇ 2(1 ± −Aδ R) + (1 ± −Aδ R)2 (γh /γc + 1)εC
According to Eqs. (13) and (15), the performance characteristics of the refrigerators are depicted in Fig. 2. Under the tight
coupling conditions (q2 = 1), when the external flux is larger than J1,max R (J1 > J1,max R ), decreasing the cooling power
from its maximal value reduces the COP of the refrigerator. When the external force is lower than J1,max R (J1 < J1,max R ),
the COP increases as the cooling power decreases. The minus sign in the above equations corresponds to the favorable case
when the external flux is decreased and the enhancement of COP occurs. The plus sign describes the opposite branch which
corresponds to the operating region with larger external flux. With the presence of non-tight coupling strength (|q| < 1),
second terms in Eqs. (5) and (6) indicate the heat leakage from the hot reservoir to the cold one. Thereby, the cooling power
is decreased, and so is the coefficient of the performance. In the operating region with lower external flux (J1 < J1,max R ),
decreasing the cooling power from its maximal value reduces the COP. While in the operating region with larger external
flux (J1 > J1,max R ), the COP first increases as the cooling power decreases, then decreases. There exists an optimal external
flux leading to the maximum coefficient of performance.
According to Eqs. (12) and (15), we have arrived at the coefficient of performance at any finite cooling power as the
function of the relative loss of cooling power (δ R):
(1−q2 ) 2
εC − 4
γh /γc +1 q2
εC
ε = (1 + δ R) √ √ (16)
2(1 ± −Aδ R) + (1 ± −Aδ R)2 (γh /γc + 1)εC

3. Performance under the tight coupling conditions

According to Ref. [15,17], the low-dissipation model can be described by the minimally nonlinear irreversible model
with the tight coupling condition. In this section, we focus on the performance under the tight coupling condition,
which contributes to analyzing the low-dissipation refrigerators at an arbitrary cooling power. And at this situation, the
refrigerators present best efficiency because there is no heat leakage.
 R. Long et al. / Physica A 496 (2018) 137–146 141

Fig. 2. The relation of relative cooling power and coefficient of performance with the relative thermodynamic flux under different coupling strengths,
where the Carnot coefficient of performance is 0.5 and the dissipation ratio (γh /γc ) equals to 1.

3.1. General COP bounds at finite cooling power

Under the tight coupling conditions (q2 = 1), Eq. (16) can be written as
√
−δ R)εC
(1 ∓
ε= √ (17)
2 + (1 ± −δ R)(γh /γc + 1)εC
where the minus sign corresponds to the favorite region of enhanced COP J1 < J1,max R ; the plus sign corresponds to the
region of lowered COP J1 > J1,max R .
According to Eq. (17), under the asymmetric dissipation limits γh /γc → ∞ and γh /γc → 0, we obtain the lower and
upper bounds on the COP at any finite cooling power.

ε (R)− = 0 (18)
√
(1 ∓ −δ R)εC
ε (R)+ = √ (19)
2 + (1 ± −δ R)εC
To be more specific, in the region J1 > J1,max R , the COP at any finite cooling power are located as
√
−δ R)εC
(1 −
0 < ε (R)J1 >J1,max R < √ (20)
2 + (1 + −δ R)εC
Similarly, in the region J1 < J1,max R , we have
√
(1 + −δ R)εC
0 < ε (R)J1 <J1,max R < √ (21)
2 + (1 − −δ R)εC
Furthermore, under the symmetric dissipation condition (γh /γc = 1), the COP is given by
√
(1 ∓ −δ R)εC
ε (R)sym = √ (22)
2 + 2(1 ± −δ R)εC
The above COP bounds at any finite cooling power are plotted in Fig. 3. The lower bound (ε (R)− = 0) coincide with the
vertical coordinate. In the region J1 > J1,max R , as the cooling power increases (δ R increases from −1 to 0), the upper bound
for the COP increases. Furthermore, the COP vanishes when the cooling power turns to be zero. While in the favorite region
J1 < J1,max R , when the cooling power decreases away from the maximum one, the coefficient of performance increases. And
the COP turns to be the Carnot coefficient of performance at vanishing cooling power.
In addition, for δ R → 0, Eqs. (18) and (19) become the general COP bounds under the maximum cooling power,
which are also deduced through the minimally nonlinear irreversible refrigerators [33] and an autonomous exoreversible
thermoelectric refrigerator [29] under the asymmetric dissipation limits:
1
0 < εmax R < εC (23)
2 + εC
142 R. Long et al. / Physica A 496 (2018) 137–146

Fig. 3. The COP bounds at any finite cooling power. The lower bound (ε (R)− = 0) coincide with the vertical coordinate. The COP under the symmetric
dissipation condition is also plotted based on Eq. (22). The Carnot coefficient of performance is 1.

Meanwhile, Eq. (22) becomes

sym 1
εmax R = εC (24)
2 + 2 εC

3.2. COP bounds under the χ figure of merit

As a compromise between the maximum COP and cooling power, the χ figure of merit is widely adopted to optimize
general refrigerators [35]. Here we defined a normalized χ criterion χ̃ = R εεR , which can be expressed as
max max R

2 + (γh /γc + 1)εC

χ̃ = (1 + δ R)2 √ √ (25)
2(1 + −δ R) + (1 + −δ R)2 (γh /γc + 1)εC
According to Eq. (25), under the asymmetric dissipation limits γh /γc → ∞ and γh /γc → 0, we obtain the lower and
upper bounds on the normalized χ at any finite cooling power.
(1 + δ R)2
χ̃ (R)− = √ (26)
(1 ± −δ R)2

2 + εC
χ̃ (R)+ = (1 + δ R)2 √ √ (27)
2(1 ± −δ R) + (1 ± −δ R)2 εC
To step further, we could calculate the COP under the maximum χ figure of merit by deriving χ̃ with respect to R at Rmax χ̃ .
By substituting Rmax χ̃ into Eq. (17), we have
√
8(γh /γc + 1)εC + 9 − 3
εmax χ̃ = , at J1 < J1,max R (28)
2(γh /γc + 1)
√
1 8(γh /γc + 1)εC + 9 − 3
εmax χ̃ = √ , at J1 > J1,max R (29)
γh /γc + 1 4(γh /γc + 1)εC + 7 − 8(γh /γc + 1)εC + 9
Eq. (28) is also obtained through the minimum nonlinear irreversible model [16]. According to Eqs. (28) and (29), under the
asymmetric limits: γh /γc → ∞ and γh /γc → 0, we have
√
8εC + 9 − 3
0 < εmax χ̃ < , at J1 < J1,max R (30)
√ 2
8εC + 9 − 3
0 < εmax χ̃ < √ , at J1 > J1,max R (31)
4εC + 7 − 8εC + 9
Eq. (30) are the lower and upper bounds for the COP which have been obtained through the minimally nonlinear
irreversible model [16] and the low dissipation model [13,36]. Here we reveal that those bounds only hold in the favorite
operating region. In the operating region with larger external flux, Eq. (31) illustrate the lower and upper COP bounds. Accord
 R. Long et al. / Physica A 496 (2018) 137–146 143

Fig. 4. The COP bounds under χ figure of merit in different operating regions. The lower bound coincide with the vertical coordinate.

to Eqs. (30) and (31), the above two bounds are plotted in Fig. 4. In the operating region with larger external flux, the COP
under the maximum χ figure of merit is much less that in the favorite region.

3.3. Relative gain in the COP

Base on Eq. (15), the relative gain in COP, δε = (ε − εmax R )/εmax R , is

√
∓2 −δ R(1 + (γh /γc + 1)εC )
δε = √ (32)
2 + (1 ± −δ R)(γh /γc + 1)εC
According to Eq. (32), the relative gain in COP achieves its minimum and maximum values at asymmetric dissipation
limits γh /γc → ∞ and γh /γc → 0, respectively. Therefore, we have
√
∓2 −δ R
δε = √ , at γh /γc → ∞ (33)
1 ± −δ R
√
∓2 −δ R(1 + εC )
δε = √ , at γh /γc → 0 (34)
2 + (1 ± −δ R)εC
To step further, we can arrive at the bounds for the relative COP gain at any arbitrary cooling power.
√ √
−2 −δ R −2 −δ R(1 + εC )
√ < δε < √ , if J1 > J1,max R (35)
1 + −δ R 2 + (1 + −δ R)εC
√ √
2 −δ R(1 + εC ) 2 −δ R
√ < δε < √ , if J1 < J1,max R (36)
2 + (1 − −δ R)εC 1 − −δ R
Based on Eqs. (35) and (36), the bounds for relative gain in COP at any arbitrary cooling power are plotted in Fig. 5. In the
region J1 > J1,max R , as the cooling power increases (δ R increases from −1 to 0), the upper and lower bounds for the relative
gain in COP increase, meanwhile, whose difference first increases then decreases. In the region J1 < J1,max R , when the cooling
power decreases away from the maximum cooling power, and the relative gain in COP increases, however, the difference
of upper and lower bounds increases. All the bounds for the relative gain in COP coincide at the maximum cooling power
condition, i.e. δ R → 0, where the derivative δε with δ R does not exist, indicating that when working near the maximum
cooling power region, the gain in COP is much larger than the loss of the cooling power, and that if the cooling power is the
main objective, it is desirable to operate the refrigerator at a slightly lower cooling power than at the maximum one in the
region J1 < J1,max R , where the refrigerator attains significantly larger COP enhancement.

4. Performance under the non-tight coupling conditions

According to Eq. (16), the COP achieves its minimum and maximum values at asymmetric dissipation limits γh /γc → ∞
and γh /γc → 0, respectively. The lower bounds are given by Eq. (18), i.e ε (R, q)− = ε (R)− = 0. The upper bounds are
144 R. Long et al. / Physica A 496 (2018) 137–146

Fig. 5. The bounds (Eqs. (35) and (36)) of relative gain in COP at any arbitrary cooling power. The Carnot coefficient of performance is 1.

achieved at γh /γc → 0:
εC − 4(1/q2 − 1)εC2
ε (R, q)+ = (1 + δ R) √ √ (37)
2(1 ± −Bδ R) + (1 ± −Bδ R)2 εC
where B = 1 − 4(1/q2 − 1)εC .
Similarly, based on Eq. (15), when γh /γc → ∞, δε (R, q) = δε (R), and the bounds for relative gain in COP (δε ) read
√
−2 −δ R (1 + δ R)(2 + εC )
√ < δε(R, q) < √ √ − 1, if J1 > J1,max R (38)
1 + −δ R 2(1 + −Bδ R) + (1 + −Bδ R)2 εC
√
(1 + δ R)(2 + εC ) 2 −δ R
√ √ − 1 < δε(R, q) < √ , if J1 < J1,max R (39)
2(1 − −Bδ R) + (1 − −Bδ R)2 εC 1 − −δ R
When |q| < 1, the existence of second terms in Eqs. (5) and (6) indicate heat leakage from the hot reservoir to the cold
one, which does not impact the power input, however, decreases the cooling power and the coefficient of performance.
Based on Eqs. (37)–(39), the upper bound for coefficient of performance and the bounds for relative gain in coefficient of
performance at any arbitrary cooling power with different coupling strengths are presented in Fig. 6. As shown in Fig. 6(a),
under the non-tight coupling conditions (|q| < 1), the curve of the upper COP bound (ε (R, q)+ ) with δ R is loop-like for any q2 .
There exists an optimal cooling power leading to the maximum value of the ε (R, q)+ at finite cooling power (Rmax ε(R,q)+ ) in
the region J1 < J1,max R . While under the tight-coupling condition, ε (R, q)+ achieves its maximum value at vanishing cooling
power. Meanwhile, in the region J1 > J1,max R , ε (R, q)+ turns to be zero.
The bounds for relative gain in coefficient of performance at any arbitrary cooling power are presented in Fig. 6(b). In the
region J1 > J1,max R , the upper bound increases with increasing δ R and the lower bound is independent of q2 . In the region
J1 < J1,max R , the upper bound is independent of q2 . Under the non-tight coupling conditions, as the cooling power decreases
away from the maximum one, the lower bound for relative gain in coefficient of performance first increases, reaches its
maximum value, then decreases. There exists an optimal cooling power leading to the maximum value of δε (R, q)− at finite
cooling power (Rmax δε(R,q)− ). While under the tight-coupling condition, when the cooling power decreases away from the
maximum one, δε (R, q)− increases, and achieves its maximum value at vanishing cooling power. Furthermore, it can be
seen that if the coefficient and cooling power should both be considered, it is more appealing to run the refrigerator in the
region with lower thermodynamic flux J1 < J1,max R , where the upper bound for COP and the lower bound for relative gain
in COP both present large values, compared to a relative small derivation from the maximum cooling power.

5. Conclusions

For refrigerators, the real-life working conditions do not correspond to a maximum cooling power or maximum COP,
but rather to a trade-off between maximum cooling power and maximum COP, however so far there is no general trade-off
criterion that exactly reflects the actual demand of real-life refrigerators, which induces an urgency to study the COP bounds
at any given cooling power, thus to give a deep insight into the operating characteristics of real-life refrigerators. In this
paper, the coefficient of performance for minimally nonlinear irreversible refrigerators at finite cooling power has been
 R. Long et al. / Physica A 496 (2018) 137–146 145

Fig. 6. The upper bound for coefficient of performance (a) and the bounds for relative gain in coefficient of performance (b) at any arbitrary cooling power
with different coupling strengths. The Carnot coefficient of performance is 0.5.

systematically studied. The lower and upper COP bounds under the tight coupling conditions for different operating regions
have been deduced, which can represent the generalization of the bounds (0 < ε < εC /(2 + εC )) on COP at maximum
cooling power. As the χ figure of merit indicates a kind of compromise between the maximum COP and cooling power. We
deduced the value of χ at finite cooling power, and calculated the COP corresponding
√ to the maximum χ figure of merit. In
the favorite operating region, we obtained the general bounds (0 < ε < ( 9 + 8εC − 3)/2) under the χ figure of merit. In
the operating region with larger external flux, new upper bound on COP has been deduced, which is much less than that in
the favorite region. For quantum systems, a detailed study on thermoelectric quantum refrigerators
√ under finite power is
presented in Refs. [37,38], and the COP bound has been deduced, which is ε + = εC [1 − 1.09 (εC + 1)(1 + δ R)]. The upper
bound on COP equals Carnot COP at zero power input, but decays with increasing power input, which corresponds to the
favorite operating region in this study.
In order to further illustrate the COP gain with the cooling power away from the maximum one, we calculated the bounds
for the maximum gain in COP in different operating regions. More generally, under different coupling strengths, the universal
upper and lower bounds for the COP and the relative gain in COP at any arbitrary cooling power have also been discussed. If
the coefficient and cooling power should both be considered, it is more appealing to run the refrigerator in the region located
between maximum cooling power and maximum COP with lower external thermodynamic flux, where the upper bound for
COP and the lower bound for relative gain in COP achieve large augment, compared to a relative small loss in the cooling
power. If the cooling power is the main objective, it is desirable to operate the refrigerator at a slightly lower cooling power
than at the maximum one, where the refrigerator attains significantly larger COP enhancement.

Acknowledgments

We acknowledge the support received from the National Natural Science Foundation of China (51706076, 51736004). In
addition, Rui Long wants to thank, in particular, the patience, care and support from Panpan Mao over the passed years. Will
you marry me?

References

[1] S. Carnot, Reflexions sur la Puissance Motorice Du Feu Et sur Les Machines, Ecole Polytechnique, Paris, 1824.
[2] V. Holubec, A. Ryabov, Work and power fluctuations in a critical heat engine, 2017, arXiv preprint arXiv:170502371.
[3] M. Polettini, M. Esposito, Carnot efficiency at divergent power output, 2016, arXiv preprint arXiv:161108192.
[4] J.S. Lee, H. Park, Carnot efficiency is attainable in an irreversible process, 2016, arXiv preprint arXiv:161107665.
[5] M. Campisi, R. Fazio, The power of a critical heat engine, Nat. Commun. 7 (2016) 11895.
[6] C.V. Johnson, Approaching the Carnot limit at finite power: An exact solution, 2017, arXiv preprint arXiv:170306119.
[7] V. Holubec, A. Ryabov, Diverging, but negligible power at Carnot efficiency: theory and experiment, 2017, arXiv preprint arXiv:170806261.
[8] F.L. Curzon, B. Ahlborn, Efficiency of a Carnot engine at maximum power output, Amer. J. Phys. 43 (1) (1975) 22.
[9] R. Long, W. Liu, Coefficient of performance and its bounds for general refrigerators with nonisothermal processes, J. Phys. A 47 (32) (2014) 325002.
[10] R. Long, W. Liu, Unified trade-off optimization for general heat devices with nonisothermal processes, Phys. Rev. E 91 (4) (2015) 042127.
146 R. Long et al. / Physica A 496 (2018) 137–146

[11] M. Esposito, R. Kawai, K. Lindenberg, C. Van den Broeck, Efficiency at maximum power of low-dissipation carnot engines, Phys. Rev. Lett. 105 (15)
(2010) 150603.
[12] C. de Tomás, A.C. Hernández, J.M.M. Roco, Optimal low symmetric dissipation Carnot engines and refrigerators, Phys. Rev. E 85 (1) (2012) 010104.
[13] Y. Hu, F. Wu, Y. Ma, J. He, J. Wang, A. Hernández, et al., Coefficient of performance for a low-dissipation carnot-like refrigerator with nonadiabatic
dissipation, Phys. Rev. E 88 (6) (2013) 062115.
[14] C. Van den Broeck, Thermodynamic efficiency at maximum power, Phys. Rev. Lett. 95 (19) (2005) 190602.
[15] Y. Izumida, K. Okuda, Efficiency at maximum power of minimally nonlinear irreversible heat engines, Europhys. Lett. 97 (1) (2012) 10004.
[16] Y. Izumida, K. Okuda, A. Calvo Hernández, J.M.M. Roco, Coefficient of performance under optimized figure of merit in minimally nonlinear irreversible
refrigerator, Europhys. Lett. 101 (1) (2013) 10005.
[17] R. Long, Z. Liu, W. Liu, Performance optimization of minimally nonlinear irreversible heat engines and refrigerators under a trade-off figure of merit,
Phys. Rev. E 89 (6) (2014) 062119.
[18] L.A. Correa, J.P. Palao, G. Adesso, D. Alonso, Optimal performance of endoreversible quantum refrigerators, Phys. Rev. E 90 (6) (2014) 062124.
[19] Y. Zheng, D. Poletti, Work and efficiency of quantum Otto cycles in power-law trapping potentials, Phys. Rev. E 90 (1) (2014) 012145.
[20] G. Thomas, R.S. Johal, Coupled quantum Otto cycle, Phys. Rev. E 83 (3) (2011) 031135.
[21] R. Uzdin, R. Kosloff, Universal features in the efficiency at maximal work of hot quantum otto engines, Europhys. Lett. 108 (4) (2014) 40001.
[22] R. Long, W. Liu, Performance of quantum Otto refrigerators with squeezing, Phys. Rev. E 91 (6) (2015) 062137.
[23] C. Jarzynski, O. Mazonka, Feynman’s ratchet and pawl: an exactly solvable model, Phys. Rev. E 59 (6) (1999) 6448–6459.
[24] Y. Lee, A. Allison, D. Abbott, H. Stanley, Minimal Brownian ratchet: An exactly solvable model, Phys. Rev. Lett. 91 (22) (2003) 220601.
[25] D. Segal, Stochastic pumping of heat: Approaching the Carnot efficiency, Phys. Rev. Lett. 101 (26) (2008) 260601.
[26] T. Schmiedl, U. Seifert, Efficiency at maximum power: An analytically solvable model for stochastic heat engines, Europhys. Lett. 81 (2) (2007) 20003.
[27] V. Holubec, An exactly solvable model of a stochastic heat engine: optimization of power, power fluctuations and efficiency, J. Stat. Mech. Theory Exp.
2014 (5) (2014) P05022.
[28] V. Holubec, A. Ryabov, Efficiency at and near maximum power of low-dissipation heat engines, Phys. Rev. E 92 (5) (2015) 052125.
[29] Y. Apertet, H. Ouerdane, A. Michot, C. Goupil, P. Lecoeur, On the efficiency at maximum cooling power, Europhys. Lett. 103 (4) (2013) 581–586.
[30] B. Jiménez de Cisneros, L. Arias-Hernández, A. Hernández, A Linear irreversible thermodynamics and coefficient of performance, Phys. Rev. E 73 (5)
(2006) 057103.
[31] A.C. Hernández, A. Medina, J. Roco, J. White, S. Velasco, Unified optimization criterion for energy converters, Phys. Rev. E 63 (3) (2001) 037102.
[32] R. Long, W. Liu, Ecological optimization and coefficient of performance bounds of general refrigerators, Physica A 443 (2016) 14–21.
[33] Y. Izumida, K. Okuda, J.M.M. Roco, A.C. Hernández, Heat devices in nonlinear irreversible thermodynamics, Phys. Rev. E 91 (5) (2015) 052140.
[34] Z. Yan, J. Chen, A class of irreversible Carnot refrigeration cycles with a general heat transfer law, J. Phys. D: Appl. Phys. 23 (2) (1990) 136.
[35] R. Long, W. Liu, Coefficient of performance and its bounds with the figure of merit for a general refrigerator, Phys. Scr. 90 (2) (2015) 025207.
[36] Y. Wang, M. Li, Z.C. Tu, A.C. Hernández, J.M.M. Roco, Coefficient of performance at maximum figure of merit and its bounds for low-dissipation Carnot-
like refrigerators, Phys. Rev. E 86 (1) (2012) 011127.
[37] R.S. Whitney, Most efficient quantum thermoelectric at finite power output, Phys. Rev. Lett. 112 (13) (2014) 130601.
[38] R.S. Whitney, Finding the quantum thermoelectric with maximal efficiency and minimal entropy production at given power output, Phys. Rev. B
91 (11) (2015) 115425.
[39] V. Holubec, A. Ryabov, Erratum: Efficiency at and near maximum power of low-dissipation heat engines [Phys. Rev. E 92, 052125 (2015)] 93(5) (2016)
059904.
[40] A. Dechant, N. Kiesel, E. Lutz, Underdamped stochastic heat engine at maximum efficiency, 2016, arXiv preprint arXiv:160200392.
[41] V. Holubec, A. Ryabov, Maximum efficiency of low-dissipation heat engines at arbitrary power, J. Stat. Mech. Theory Exp. 2016 (7) (2016) 073204.
[42] R. Long, W. Liu, Efficiency and its bounds of minimally nonlinear irreversible heat engines at arbitrary power, Phys. Rev. E 94 (5) (2016) 052114.
[43] B.J. de Cisneros, L.A. Arias-Hernández, A.C. Hernández, Linear irreversible thermodynamics and coefficient of performance, Phys. Rev. E 73 (5) (2006)
645–666.
[44] H. Goldsmid, Thermoelectric Refrigeration, Pion Ltd., London, 1964.
[45] G. Mahan, Introduction to thermoelectrics, APL Mat. 4 (10) (2016) 104806.
[46] L. Onsager, Reciprocal relations in irreversible processes. I, Phys. Rev. 37 (4) (1931) 405–426.
[47] M. Zebarjadi, K. Esfarjani, A. Shakouri, Nonlinear peltier effect in semiconductors, Appl. Phys. Lett. 91 (12) (2007) 122104.
[48] B. Muralidharan, M. Grifoni, Performance analysis of an interacting quantum dot thermoelectric setup, Phys. Rev. B 85 (15) (2012) 155423.
[49] J. Meair, P. Jacquod, Scattering theory of nonlinear thermoelectricity in quantum coherent conductors, J. Phys.: Condens. Matter 25 (8) (2013) 082201.
[50] J. Azema, P. Lombardo, A.-M. Daré, Conditions for requiring nonlinear thermoelectric transport theory in nanodevices, Phys. Rev. B 90 (20) (2014)
205437.
[51] R.S. Whitney, Nonlinear thermoelectricity in point contacts at pinch off: A catastrophe aids cooling, Phys. Rev. B 88 (6) (2013) 064302.