-
Efficiency at the maximum power output for simple two-level heat
engine
Sang Hoon Lee School of Physics, Korea Institute for Advanced
Study
http://newton.kias.re.kr/~lshlj82
in collaboration with Jaegon Um (CCSS, CTP and Department of
Physics and Astronomy, SNU) and Hyunggyu Park (School of Physics
& Quantum Universe Center, KIAS)
2016 Workshop on Special Topics in Statistical Physics &
Complex Systems (SPnCS) @ Chosun Univ., Gwangju, 22 December,
2016.
SHL, J. Um, and H. Park, e-print arXiv:1612.00518.
-
Carnot engine
source:
http://web2.uwindsor.ca/courses/physics/high_schools/2013/SteamEngine/images/PVgraph.jpg
S. Carnot, Rflexions Sur La Puissance Motrice Du Feu Et Sur Les
Machines Propres Dvelopper Cette Puissance (Bachelier Libraire,
Paris, 1824).
the
Sadi Carnot (1796-1832)
http://web2.uwindsor.ca/courses/physics/high_schools/2013/SteamEngine/images/PVgraph.jpg
-
Carnot engine
source:
http://web2.uwindsor.ca/courses/physics/high_schools/2013/SteamEngine/images/PVgraph.jpg
S. Carnot, Rflexions Sur La Puissance Motrice Du Feu Et Sur Les
Machines Propres Dvelopper Cette Puissance (Bachelier Libraire,
Paris, 1824).
the
Sadi Carnot (1796-1832)
the Carnot eciency C =Weng|Qh|
=
|Qh| |Qc||Qh|
= 1 TcTh
quasi-static(the 1st law of thermodynamics)
http://web2.uwindsor.ca/courses/physics/high_schools/2013/SteamEngine/images/PVgraph.jpg
-
the Carnot eciency C =Weng|Qh|
=
|Qh| |Qc||Qh|
= 1 TcTh
quasi-static(the 1st law of thermodynamics)
-
the Carnot eciency C =Weng|Qh|
=
|Qh| |Qc||Qh|
= 1 TcTh
quasi-static(the 1st law of thermodynamics)
0: cyclic process
the 2nd law of thermodynamics: Stot
= Seng
+Sres
= QhTh
+
QcTc 0
(per cycle)
! |Qc||Qh| Tc
Th! = 1 |Qc||Qh|
1 TcTh
= C
) C in general, and C is the theoretically maximum eciency.
quasi-static
-
the Carnot eciency C =Weng|Qh|
=
|Qh| |Qc||Qh|
= 1 TcTh
quasi-static(the 1st law of thermodynamics)
0: cyclic process
the 2nd law of thermodynamics: Stot
= Seng
+Sres
= QhTh
+
QcTc 0
(per cycle)
! |Qc||Qh| Tc
Th! = 1 |Qc||Qh|
1 TcTh
= C
) C in general, and C is the theoretically maximum eciency.
Weng reaches the maximum value for given |Qh| in the Carnot
engine,but the power P = Weng/ ! 0 where is the operating time
!1
quasi-static
-
Th
Tc
hot reservoir
cold reservoir
Thw
Tcw
Endoreversible engine P. Chambadal, Les Centrales Nuclaires
(Armand Colin, Paris, 1957). I. I. Novikov, Efficiency of an atomic
power generating installation, At. Energy 3, 1269 (1957);
The efficiency of atomic power stations, J. Nucl. Energy 7, 125
(1958). F. L. Curzon and B. Ahlborn, Efficiency of a Carnot engine
at maximum power output, Am. J. Phys. 43, 22 (1975).
-
endoreversibility
Th
Tc
hot reservoir
cold reservoir
Thw
Tcw
during t1
irreversible heat conduction
the input energy (linear heat conduction) Qh = t1(Th Thw)
the reversible engine
operated at Thw and Tcw!QhThw
=QcTcw
during t2
irreversible heat conduction
the heat rejected (linear heat conduction) Qc = t2(Tcw Tc)
Endoreversible engine P. Chambadal, Les Centrales Nuclaires
(Armand Colin, Paris, 1957). I. I. Novikov, Efficiency of an atomic
power generating installation, At. Energy 3, 1269 (1957);
The efficiency of atomic power stations, J. Nucl. Energy 7, 125
(1958). F. L. Curzon and B. Ahlborn, Efficiency of a Carnot engine
at maximum power output, Am. J. Phys. 43, 22 (1975).
-
endoreversibility
the (Chambadal-Novikov-)Curzon-Ahlborn eciency CA = 1r
TcTh
Th
Tc
hot reservoir
cold reservoir
Thw
Tcw
during t1
irreversible heat conduction
the input energy (linear heat conduction) Qh = t1(Th Thw)
the reversible engine
operated at Thw and Tcw!QhThw
=QcTcw
during t2
irreversible heat conduction
the heat rejected (linear heat conduction) Qc = t2(Tcw Tc)
maximizing power P =Qh Qct1 + t2
with respect to t1 and t2
Endoreversible engine P. Chambadal, Les Centrales Nuclaires
(Armand Colin, Paris, 1957). I. I. Novikov, Efficiency of an atomic
power generating installation, At. Energy 3, 1269 (1957);
The efficiency of atomic power stations, J. Nucl. Energy 7, 125
(1958). F. L. Curzon and B. Ahlborn, Efficiency of a Carnot engine
at maximum power output, Am. J. Phys. 43, 22 (1975).3/31/16,
12:03Endoreversible thermodynamics - Wikipedia, the free
encyclopedia
Page 2 of
3https://en.wikipedia.org/wiki/Endoreversible_thermodynamics
Power Plant (C) (C) (Carnot) (Endoreversible) (Observed)West
Thurrock (UK) coal-fired power
plant 25 565 0.64 0.40 0.36
CANDU (Canada) nuclear power plant 25 300 0.48 0.28
0.30Larderello (Italy) geothermal power
plant 80 250 0.33 0.178 0.16
As shown, the endoreversible efficiency much more closely models
the observed data. However, such anengine violates Carnot's
principle which states that work can be done any time there is a
difference intemperature. The fact that the hot and cold reservoirs
are not at the same temperature as the working fluidthey are in
contact with means that work can and is done at the hot and cold
reservoirs. The result istantamount to coupling the high and low
temperature parts of the cycle, so that the cycle collapses.[7] In
theCarnot cycle there is strict necessity that the working fluid be
at the same temperatures as the heat reservoirsthey are in contact
with and that they are separated by adiabatic transformations which
prevent thermalcontact. The efficiency was first derived by William
Thomson [8] in his study of an unevenly heated body inwhich the
adiabatic partitions between bodies at different temperatures are
removed and maximum work isperformed. It is well known that the
final temperature is the geometric mean temperature so that
the efficiency is the Carnot efficiency for an engine working
between and .
Due to occasional confusion about the origins of the above
equation, it is sometimes named theChambadal-Novikov-Curzon-Ahlborn
efficiency.
See alsoHeat engine
An introduction to endoreversible thermodynamics is given in the
thesis by Katharina Wagner.[4] It is alsointroduced by Hoffman et
al.[9][10] A thorough discussion of the concept, together with many
applications inengineering, is given in the book by Hans Ulrich
Fuchs.[11]
References1. I. I. Novikov. The Efficiency of Atomic Power
Stations. Journal Nuclear Energy II, 7:125128, 1958. translated
from
Atomnaya Energiya, 3 (1957), 409.2. Chambadal P (1957) Les
centrales nuclaires. Armand Colin, Paris, France, 4 1-583. F.L.
Curzon and B. Ahlborn, American Journal of Physics, vol. 43, pp.
2224 (1975)4. M.Sc. Katharina Wagner, A graphic based interface to
Endoreversible Thermodynamics, TU Chemnitz, Fakultt fr
Naturwissenschaften, Masterarbeit (in English).
http://archiv.tu-chemnitz.de/pub/2008/0123/index.html5. A Bejan, J.
Appl. Phys., vol. 79, pp. 11911218, 1 Feb. 1996
http://dx.doi.org/10.1016/S0035-3159(96)80059-66. Callen, Herbert
B. (1985). Thermodynamics and an Introduction to Thermostatistics
(2nd ed. ed.). John Wiley &
Sons, Inc.. ISBN 0-471-86256-8.7. B. H. Lavenda, Am. J. Phys.,
vol. 75, pp. 169-175 (2007)8. W. Thomson, Phil. Mag. (Feb.
1853)
-
the (Chambadal-Novikov-)Curzon-Ahlborn eciency CA = 1r
TcTh
-
the (Chambadal-Novikov-)Curzon-Ahlborn eciency CA = 1r
TcTh
Q. Is this a universal formula for power-maximizing efficiency,
or does endoreversibility guarantee it?
-
the (Chambadal-Novikov-)Curzon-Ahlborn eciency CA = 1r
TcTh
Q. Is this a universal formula for power-maximizing efficiency,
or does endoreversibility guarantee it?
A. No. The linear heat conduction is essential.
Q = (Th Tc)
We introduce a different type of engine with non-(CN)CA optimal
efficiency.
L. Chen and Z. Yan, J. Chem. Phys. 90, 3740 (1988): F.
Angulo-Brown and R. Pez-Hernndez, J. Appl. Phys. 74, 2216
(1993):
(Dulong-Petit law of cooling)Q = (Th Tc)n
Q = (Tnh Tnc )
-
our simple two-level heat engine model
R1
R2
relaxation with
relaxation with
Q1
Q2
E1
E2
T1
T2
q
t1
t2
during t1
during t2
W = E1 E2W 0 = E1 E2
q/(1 q) = exp(E1/T1)
/(1 ) = exp(E2/T2)
0
E1
q(setting the Boltzmann constant
kB 1 for notational convenience)
-
our simple two-level heat engine model
R1
R2
relaxation with
relaxation with
Q1
Q2
E1
E2
T1
T2
q
t1
t2
during t1
during t2
W = E1 E2W 0 = E1 E2
q/(1 q) = exp(E1/T1)
/(1 ) = exp(E2/T2)
0
E2
(setting the Boltzmann constant
kB 1 for notational convenience)
-
our simple two-level heat engine model
R1
R2
relaxation with
relaxation with
Q1
Q2
E1
E2
T1
T2
q
t1
t2
during t1
during t2
W = E1 E2W 0 = E1 E2
q/(1 q) = exp(E1/T1)
/(1 ) = exp(E2/T2)
0
E2
(setting the Boltzmann constant
kB 1 for notational convenience)
-
our simple two-level heat engine model
R1
R2
relaxation with
relaxation with
Q1
Q2
E1
E2
T1
T2
q
t1
t2
during t1
during t2
W = E1 E2W 0 = E1 E2
q/(1 q) = exp(E1/T1)
/(1 ) = exp(E2/T2)
0
q
E1
(setting the Boltzmann constant
kB 1 for notational convenience)
-
our simple two-level heat engine model
R1
R2
relaxation with
relaxation with
Q1
Q2
E1
E2
T1
T2
q
t1
t2
during t1
during t2
W = E1 E2W 0 = E1 E2
q/(1 q) = exp(E1/T1)
/(1 ) = exp(E2/T2)
0
q
E1
(setting the Boltzmann constant
kB 1 for notational convenience)
-
our simple two-level heat engine model
R1
E1T1
q
R2
E2
T2
0
q
0
during duringstochastic Markov processes
1 2
|P1i(t1 = 0) |P1i(t1 = 1) = |P2i(t2 = 0) |P2i(t2 = 2)
|P2i(t2 = 2) = |P1i(t1 = 0)
W = E1 E2
W 0 = E1 E2
Q1 Q2
hWnetiT1
sh =T2T1
sc +hWnetiT1
sh =E1T1
scG1 (sc)
sc
sh
-
hW i = (E1 E2)P1 and hW 0i = (E1 E2)P2where P1 (P2) is the
population in E1 (E2) at R1 (R2), respectively
independent of t1
and t2
(no P1
and P2
dependency)
and ! Carnot
= 1 T2
/T1
when ' q
total entropy change S = Q1T1
+
Q2T2
hQ1i = (P1 P2)T1 ln[(1 q)/q], hQ2i = (P1 P2)T2 ln[(1 )/]
hWneti = hW i hW 0i = (P1 P2)(E1 E2)= (P1 P2){T1 ln[(1 q)/q] T2
ln[(1 )/]}
eciency =hWnetihQ1i
=hW i hW 0i
hQ1i= 1 T2
T1
ln[(1 )/]ln[(1 q)/q]
0
= (P1 P2)E1 = (P1 P2)E2(from the Schnakenberg entropy formula,
or equivalently, the 1st law:
hEi = hQi hW i for each half of the cycle)
-
hWneti = hW i hW 0i = (P1 P2)(E1 E2)= (P1 P2){T1 ln[(1 q)/q] T2
ln[(1 )/]}
eciency =hWnetihQ1i
=hW i hW 0i
hQ1i= 1 T2
T1
ln[(1 )/]ln[(1 q)/q]
P1(t1 !1, t2 !1) = q and P2(t1 !1, t2 !1) = , as expected
meaningful only for q > , or hWneti > 0P1 = (1 q)P1 + q(1
P1)P2 = (1 )P2 + (1 P2)
! P1 = q A1et01 , P2 = A2et
02
(0 t01 t1 and 0 t02 t2)P1(t
01 = 0, t2) = P2(t1, t2) and P2(t1, t
02 = 0) = P1(t1, t2)
! A1 = q P2, A2 = P1 and let t01 = t1, t02 = t2
! P1 =q(1 et1) + (1 et2)et1
1 e(t1+t2)
P2 =(1 et2) + q(1 et1)et2
1 e(t1+t2)
-
Let t1 = t2 = /2, then
in terms of , the maximum power is achieved for ! 0, asPower !
hWneti/4 and the power is monotonically decreased as is
increased.
hWneti = (q )(1 e/2)2
1 e
{T1 ln[(1 q)/q] T2 ln[(1 )/]}
time: decoupled overall factor
hWneti( !1) = (q ) {T1 ln[(1 q)/q] T2 ln[(1 )/]}
Power hP i = hWneti
=
q
(1 e/2)2
1 e
{T1 ln[(1 q)/q] T2 ln[(1 )/]}
(still decoupled even when t1 6= t2)
-
Let t1 = t2 = /2, then
in terms of , the maximum power is achieved for ! 0, asPower !
hWneti/4 and the power is monotonically decreased as is
increased.
hWneti = (q )(1 e/2)2
1 e
{T1 ln[(1 q)/q] T2 ln[(1 )/]}
time: decoupled overall factor
hWneti( !1) = (q ) {T1 ln[(1 q)/q] T2 ln[(1 )/]}
Power hP i = hWneti
=
q
(1 e/2)2
1 e
{T1 ln[(1 q)/q] T2 ln[(1 )/]}
our goal: to find (q, ) = (q, ) maximizing hP i@hP i@q
q=q,=
=@hP i@
q=q,=
= 0
(still decoupled even when t1 6= t2)
hWneti = hW i hW 0i = (P1 P2)(E1 E2)= (P1 P2){T1 ln[(1 q)/q] T2
ln[(1 )/]}
eciency =hWnetihQ1i
=hW i hW 0i
hQ1i= 1 T2
T1
ln[(1 )/]ln[(1 q)/q]
substitute (q, ) = (q, ) here,then
op
(q, ) is the eciencyat the maximum power output
-
! T2q(1 q)
T1(1 )= 1
at (q, ) (global optimum)
3
equation system for given q and values,
dP1
dt
= (1 q)P1 + q(1 P1) ,dP2
dt
= (1 )P2 + (1 P2) .(7)
With the circular boundary condition P1(t01 = 0, t02 = t2) =
P2(t01 = t1, t02 = t2) and P2(t
01 = t1, t
02 = 0) = P1(t
01 = t1, t
02 =
t2) where the primed variables describe intermediate time,
thesolution at t01 = t1 and t
02 = t2 is
P1 =q(1 et1 ) + (1 et2 )et1
1 e(t1+t2) ,
P2 =(1 et2 ) + q(1 et1 )et2
1 e(t1+t2) ,(8)
and limt1,t2!1 P1 = q and limt1,t2!1 P2 = as expected.
Let t1 = t2 = /2 (each reservoir contacts with the systemduring
the equal time), then
hWneti = (q )(1 e/2)2
1 e"T1 ln
1 q
q
! T2 ln
1
!#,
(9)so the monotonically increasing factor (1 e/2)2/(1 e)for the
time scale is decoupled from the rest of the formulaand only plays
the role of overall factor. The amount of network is increased up
to
lim!1hWneti = (q )
"T1 ln
1 q
q
! T2 ln
1
!#. (10)
The average power output hPi, which is the amount of network per
unit time, is given by
hPi = (q )(1 e/2)2
(1 e)"T1 ln
1 q
q
! T2 ln
1
!#.
(11)In terms of , the maximum power is achieved forlim!0hPi =
hWneti/4 and the power is monotonically de-creased as is increased.
Therefore, from now on, we discard
the time dependence altogether and focus on other parame-ters,
i.e., denoting hWneti = hPi without considering the over-all factor
involving for notational convenience. Numerically,we obtain the net
work and eciency for (q, ) combination,as shown in Fig. 2. In Sec.
III B, we derive the condition forthe eciency at the maximum power
output.
B. Eciency at the maximum power output
1. The condition for the maximum power output
For a given T2/T1 value, the maximum power output con-dition for
the two-variable function is
@hPi@q
q=q,=
=@hPi@
q=q,=
= 0 , (12)
which leads to
1 T2T1
ln[(1 )/]ln[(1 q)/q] =
q
q
(1 q) ln[(1 q)/q] , (13a)
and
1 T2T1
ln[(1 )/]ln[(1 q)/q] =
(T2/T1)(q )(1 ) ln[(1 q)/q] , (13b)
from Eq. (11). By eliminating the left-hand side of Eqs.
(13a)and (13b), we obtain the following simple relation
T2q(1 q)
T1(1 ) = 1 , (14a)
or
=12
2666641
r1 4T2
T1q
(1 q)377775 , (14b)
because 0 < < q < 1/2. By substituting as a function
ofq
in Eq. (14b) to Eq. (13a) or Eq. (13b), we get the
optimumcondition f (T2/T1, q) = 0, which is explicitly
f (T2/T1, q) = ln
1 qq
! T2
T1ln
26666666666666664
1 +r
1 4T2T1
q
(1 q)
1 r
1 4T2T1
q
(1 q)
37777777777777775
q
12+
12
r1 4T2
T1q
(1 q)q
(1 q)
= ln
1 qq
! (1
C
) ln
266666666664
1 +q
(1 2q)2 + 4C
q
(1 q)1
q(1 2q)2 + 4
C
q
(1 q)
377777777775
q
12+
12
q(1 2q)2 + 4
C
q
(1 q)q
(1 q) = 0 .
(15)
Furthermore, the condition in Eq. (15) leads to the
followingform of op from Eq. (6),
op =q
12+
12
r1 4T2
T1q
(1 q)q
(1 q) ln[(1 q)/q] (16a)
or in terms of C
= 1 T2/T1,
op =q
12+
12
q(1 2q)2 + 4
C
q
(1 q)q
(1 q) ln[(1 q)/q] . (16b)
-
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
quadratic form
FIG. 3. Numerically found q and values satisfying Eq. (15), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The quadratic form indicates Eq.
(19).
We verify that the (q, ) is indeed the maximum point byusing the
relations of second derivatives
0BBBBB@@2hPi@q2
q=q,=
1CCCCCA
0BBBBB@@2hPi@2
q=q,=
1CCCCCA
0BBBBB@@2hPi@q@
q=q,=
1CCCCCA
2
=
T
21(1 )q
(1 q) (2q q )2 4T 21 2(1 )2
2(1 )2q2(1 q)2
= T1(q )2
(1 )q3(1 q)3 > 0 ,(17)
where we use the relation in Eq. (14a), and
@2hPi@q2
q=q,=
= T1[q + (1 2q)]q
2(1 q)2 < 0 ,
@2hPi@2
q=q,=
= T2[q + (1 2q)]2(1 )2 < 0 .
(18)
Therefore, the procedure to calculate the eciency forgiven T2/T1
at the maximum power output seems straight-forward now. First, find
the q value satisfying Eq. (15) andsubstitute the q value to Eq.
(16a). The only (but very crucial)problem is that f (T2/T1, q) = 0
is a transcendental equationwhose closed-form solution is
unattainable.
2. Asymptotic behaviors for extreme cases
The upper bound for q is given by the condition T2/T1 =
0,satisfying ln[(1 q)/q] = 1/(1 q) and q(T2/T1 = 0) '0.217 811 705
719 800 found numerically and (T2/T1 =0) = 0 exactly from Eq.
(14b). T2/T1 = 1 always satisfiesf (T2/T1, q) = 0 regardless of q
values, but the T2/T1 ' 1asymptotic case yields the lower bound for
q(T2/T1 ' 1) =(T2/T1 ' 1) ' 0.083 221 720 199 517 7 found
numerically.
schematically . . .
q
= q
no net work
q(T2/T1 1) 0.0832217201995177= (T2/T1 1)
q(T2/T1 = 0) 0.217811705719800(T2/T1 = 0) = 0
as T2/T1 (C) is decreased (increased),respectively
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
op
c = 1 T2 / T1
at (q*, *)CA = 11c
c/(2c)c/2
FIG. 5. The eciency at the maximum power op as the functionof
the Carnot eciency
C
in Eq. (16b) using numerically found op-timal q values, along
with various asymptotic cases: the Curzon-Ahlborn eciency CA in Eq.
(21), the upper bound C/(2 C) andthe lower bound
C
/2 in Ref. [19].
Figure 3 shows the numerical solution (q, ) as a functionof
C
= 1 T2/T1, where the asymptotic behaviors derivedabove hold when
T2/T1 ' 0 and T2/T1 ' 1. It seems thatq
is monotonically decreased and is monotonically in-creased, as
T2/T1 is increased, i.e., qmin = q
(T2/T1 ' 1),q
max = q
(T2/T1 = 0), min = 0, and max =
(T2/T1 ' 1).Figure 4 illustrates the situation on the (q, )
plane. We numer-ically obtain the quadratic function q(
C
) = qmin + aC + b2C
by fitting it to the numerical solution near C
= 0, in the range
C
2 [0, 0.05] in this case, which isq
(C
) ' qmin + 0.0457524C + 0.02767322C . (19)To proceed further, we
have to take into account that q
is also a function of C
. Unfortunately, it is not possible toobtain the closed form
solution for q(
C
), so we rely on the
! T2q(1 q)
T1(1 )= 1
at (q, ) (global optimum)
! T2q(1 q)
T1(1 )= 1
-
! T2q(1 q)
T1(1 )= 1
at (q, ) (global optimum)
) = 12
1
r1 4T2
T1q(1 q)
!
substituting (q) to
1 T2T1
ln[(1 )/]ln[(1 q)/q]
=
T2T1
q
(1 ) ln[(1 q)/q]
Finding the global maximum theoretically
1 T2T1
ln[(1 )/]ln[(1 q)/q]
=
q
q(1 q) ln[(1 q)/q] or
! f(T2
/T1
, q) = 0 ! q(T1
, T2
) ! (T1
, T2
) ! (hWnet
imax
, op
)
-
! f(T2
/T1
, q) = 0 ! q(T1
, T2
) ! (T1
, T2
) ! (hWnet
imax
, op
)
f(T2/T1, q) = ln
1 q
q
T2
T1ln
0
BB@1 +
r1 4T2
T1q(1 q)
1r
1 4T2T1
q(1 q)
1
CCAq 1
2+
1
2
r1 4T2
T1q(1 q)
q(1 q)
numerically found q(C) and (C)
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q*(c0) = *(c0)
q*(c=1)
optim
al tr
ansi
tion
rate
s
c
q**
c0 and 1 asymptotes
-
schematically . . .
q
= q
net power < 0
-
schematically . . .
q
= q
net power < 0
C from 0 to 1(T2/T1 from 1 to 0)
( ), T1 = 1, T2 = 1/100
0.1 0.2 0.3 0.4 0.5q
0.1
0.2
0.3
0.4
0.5
0
0.05
0.1
0.15
0.2
0.25
0.3
( ), T1 = 1, T2 = 1/10
0.1 0.2 0.3 0.4 0.5q
0.1
0.2
0.3
0.4
0.5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
( ), T1 = 1, T2 = 1/2
0.1 0.2 0.3 0.4 0.5q
0.1
0.2
0.3
0.4
0.5
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
( ), T1 = 1, T2 = 9/10
0.1 0.2 0.3 0.4 0.5q
0.1
0.2
0.3
0.4
0.5
0
0.0002
0.0004
0.0006
0.0008
0.001
0.0012
P (T2/T1 = 9/10)
P (T2/T1 = 1/2)
P (T2/T1 = 1/10)
P (T2/T1 = 1/100)
q
q
q
q
hWneti( !1) = (q ) {T1 ln[(1 q)/q] T2 ln[(1 )/]}
q(C ! 0) = (C ! 0) ' 0.083 221 720 199 517 7 q0
the root of
2
1 2q = ln1 q
q
q(C = 1) ' 0.217 811 705 719 800(C = 1) = 0
the root of
1
1 q = ln1 q
q
-
) = 12
1
r1 4T2
T1q(1 q)
!
) op
= 1 T2T1
ln[(1 )/]ln[(1 q)/q]
=
q 12+
1
2
r1 4T2
T1
q(1 q)
q(1 q) ln[(1 q)/q]
=q 1
2+
1
2
p(1 2q)2 + 4Cq(1 q)
q(1 q) ln[(1 q)/q]
= ln
1 q
q
(1 C) ln
1 +
p(1 2q)2 + 4Cq(1 q)
1p
(1 2q)2 + 4Cq(1 q)
!q 1
2+
1
2
p(1 2q)2 + 4Cq(1 q)
q(1 q)
f(T2/T1, q) = ln
1 q
q
T2
T1ln
0
BB@1 +
r1 4T2
T1q(1 q)
1r
1 4T2T1
q(1 q)
1
CCAq 1
2+
1
2
r1 4T2
T1q(1 q)
q(1 q)
! f(T2
/T1
, q) = 0 ! q(T1
, T2
) ! (T1
, T2
) ! (hWnet
imax
, op
)
the series expansion at C ! 0
-
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q*(c0) = *(c0)
q*(c=1)
optim
al tr
ansi
tion
rate
s
c
q**
c0 and 1 asymptotes
the numerically found functional form of
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd *
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
quadratic form near c=0
FIG. 3. Numerically found q and values satisfying Eq. (16), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The quadratic form indicates Eq.
(21).
We verify that the (q, ) is indeed the maximum point byusing the
relations of second derivatives
0BBBBB@@2hPi@q2
q=q,=
1CCCCCA
0BBBBB@@2hPi@2
q=q,=
1CCCCCA
0BBBBB@@2hPi@q@
q=q,=
1CCCCCA
2
=
T
21(1 )q
(1 q) (2q q )2 4T 21 2(1 )2
2(1 )2q2(1 q)2
= T1(q )2
(1 )q3(1 q)3 > 0 ,(18)
where we use the relation in Eq. (15a), and
@2hPi@q2
q=q,=
= T1[q + (1 2q)]q
2(1 q)2 < 0 ,
@2hPi@2
q=q,=
= T2[q + (1 2q)]2(1 )2 < 0 .
(19)
Therefore, the procedure to calculate the eciency forgiven T2/T1
at the maximum power output seems straight-forward now. First, find
the q value satisfying Eq. (16) andsubstitute the q value to Eq.
(17a). The only (but very crucial)problem is that f (T2/T1, q) = 0
is a transcendental equationwhose closed-form solution is
unattainable.
2. Asymptotic behaviors based on series expansion
If we treat q as constant for a moment (which is clearly notthe
case, of course, as q is also a function of T2/T1 to satisfy
schematically . . .
q
= q
no net work
q(T2/T1 1) 0.0832217201995177= (T2/T1 1)
q(T2/T1 = 0) 0.217811705719800(T2/T1 = 0) = 0
as T2/T1 (C) is decreased (increased),respectively
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
f = 0), the function f has the expansion form
f (T2/T1, q) ="ln
1 q
q
! 1
1 q#+
(1 ln
T2
T1
!+ ln
q
(1 q))
T2
T1
!
+ O2666664
T2
T1
!23777775
(20a)
when T2/T1 ' 0, and
f (T2/T1, q) ="ln
1 q
q
! 2
1 2q#
1 T2T1
!+
12(1 2q)
1 T2
T1
!2
+ O2666664
1 T2
T1
!33777775
(20b)
when T2/T1 ' 1. The upper bound for q is given by the con-dition
T2/T1 = 0, satisfying ln[(1 q)/q] = 1/(1 q) andq
(T2/T1 = 0) ' 0.217 811 705 719 800 found numericallyand (T2/T1
= 0) = 0 exactly from Eq. (15b). T2/T1 = 1always satisfies f
(T2/T1, q) = 0 regardless of q values, butthe T2/T1 ' 1 asymptotic
case yields the lower bound forq
(T2/T1 ' 1) = (T2/T1 ' 1) ' 0.083 221 720 199 517 7found
numerically. Figure 3 shows the numerical solution(q, ) as a
function of
C
= 1 T2/T1, where the asymp-totic behaviors derived above hold
when T2/T1 ' 0 andT2/T1 ' 1. It seems that q is monotonically
decreasedand is monotonically increased, as T2/T1 is increased,
i.e.,q
min = q
(T2/T1 ' 1), qmax = q(T2/T1 = 0), min = 0, andmax = (T2/T1 ' 1).
For later, we numerically obtain thefunctional form of q(
C
) up to the quadratic term by fitting aquadratic function
near
C
= 0, in the range C
2 [0, 0.05] inthis case, which is
q
(C
) ' qmin + 0.0457524C + 0.02767322C . (21)
-
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q*(c0) = *(c0)
q*(c=1)
optim
al tr
ansi
tion
rate
s
c
q**
c0 and 1 asymptotes
the numerically found functional form of
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd *
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
quadratic form near c=0
FIG. 3. Numerically found q and values satisfying Eq. (16), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The quadratic form indicates Eq.
(21).
We verify that the (q, ) is indeed the maximum point byusing the
relations of second derivatives
0BBBBB@@2hPi@q2
q=q,=
1CCCCCA
0BBBBB@@2hPi@2
q=q,=
1CCCCCA
0BBBBB@@2hPi@q@
q=q,=
1CCCCCA
2
=
T
21(1 )q
(1 q) (2q q )2 4T 21 2(1 )2
2(1 )2q2(1 q)2
= T1(q )2
(1 )q3(1 q)3 > 0 ,(18)
where we use the relation in Eq. (15a), and
@2hPi@q2
q=q,=
= T1[q + (1 2q)]q
2(1 q)2 < 0 ,
@2hPi@2
q=q,=
= T2[q + (1 2q)]2(1 )2 < 0 .
(19)
Therefore, the procedure to calculate the eciency forgiven T2/T1
at the maximum power output seems straight-forward now. First, find
the q value satisfying Eq. (16) andsubstitute the q value to Eq.
(17a). The only (but very crucial)problem is that f (T2/T1, q) = 0
is a transcendental equationwhose closed-form solution is
unattainable.
2. Asymptotic behaviors based on series expansion
If we treat q as constant for a moment (which is clearly notthe
case, of course, as q is also a function of T2/T1 to satisfy
schematically . . .
q
= q
no net work
q(T2/T1 1) 0.0832217201995177= (T2/T1 1)
q(T2/T1 = 0) 0.217811705719800(T2/T1 = 0) = 0
as T2/T1 (C) is decreased (increased),respectively
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
f = 0), the function f has the expansion form
f (T2/T1, q) ="ln
1 q
q
! 1
1 q#+
(1 ln
T2
T1
!+ ln
q
(1 q))
T2
T1
!
+ O2666664
T2
T1
!23777775
(20a)
when T2/T1 ' 0, and
f (T2/T1, q) ="ln
1 q
q
! 2
1 2q#
1 T2T1
!+
12(1 2q)
1 T2
T1
!2
+ O2666664
1 T2
T1
!33777775
(20b)
when T2/T1 ' 1. The upper bound for q is given by the con-dition
T2/T1 = 0, satisfying ln[(1 q)/q] = 1/(1 q) andq
(T2/T1 = 0) ' 0.217 811 705 719 800 found numericallyand (T2/T1
= 0) = 0 exactly from Eq. (15b). T2/T1 = 1always satisfies f
(T2/T1, q) = 0 regardless of q values, butthe T2/T1 ' 1 asymptotic
case yields the lower bound forq
(T2/T1 ' 1) = (T2/T1 ' 1) ' 0.083 221 720 199 517 7found
numerically. Figure 3 shows the numerical solution(q, ) as a
function of
C
= 1 T2/T1, where the asymp-totic behaviors derived above hold
when T2/T1 ' 0 andT2/T1 ' 1. It seems that q is monotonically
decreasedand is monotonically increased, as T2/T1 is increased,
i.e.,q
min = q
(T2/T1 ' 1), qmax = q(T2/T1 = 0), min = 0, andmax = (T2/T1 ' 1).
For later, we numerically obtain thefunctional form of q(
C
) up to the quadratic term by fitting aquadratic function
near
C
= 0, in the range C
2 [0, 0.05] inthis case, which is
q
(C
) ' qmin + 0.0457524C + 0.02767322C . (21)f(T2/T1 = 0, q
) = ln
1 q
q
1
1 q = 0
! q ' 0.217 811 705 719 800
-
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q*(c0) = *(c0)
q*(c=1)
optim
al tr
ansi
tion
rate
s
c
q**
c0 and 1 asymptotes
the numerically found functional form of
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd *
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
quadratic form near c=0
FIG. 3. Numerically found q and values satisfying Eq. (16), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The quadratic form indicates Eq.
(21).
We verify that the (q, ) is indeed the maximum point byusing the
relations of second derivatives
0BBBBB@@2hPi@q2
q=q,=
1CCCCCA
0BBBBB@@2hPi@2
q=q,=
1CCCCCA
0BBBBB@@2hPi@q@
q=q,=
1CCCCCA
2
=
T
21(1 )q
(1 q) (2q q )2 4T 21 2(1 )2
2(1 )2q2(1 q)2
= T1(q )2
(1 )q3(1 q)3 > 0 ,(18)
where we use the relation in Eq. (15a), and
@2hPi@q2
q=q,=
= T1[q + (1 2q)]q
2(1 q)2 < 0 ,
@2hPi@2
q=q,=
= T2[q + (1 2q)]2(1 )2 < 0 .
(19)
Therefore, the procedure to calculate the eciency forgiven T2/T1
at the maximum power output seems straight-forward now. First, find
the q value satisfying Eq. (16) andsubstitute the q value to Eq.
(17a). The only (but very crucial)problem is that f (T2/T1, q) = 0
is a transcendental equationwhose closed-form solution is
unattainable.
2. Asymptotic behaviors based on series expansion
If we treat q as constant for a moment (which is clearly notthe
case, of course, as q is also a function of T2/T1 to satisfy
schematically . . .
q
= q
no net work
q(T2/T1 1) 0.0832217201995177= (T2/T1 1)
q(T2/T1 = 0) 0.217811705719800(T2/T1 = 0) = 0
as T2/T1 (C) is decreased (increased),respectively
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
f = 0), the function f has the expansion form
f (T2/T1, q) ="ln
1 q
q
! 1
1 q#+
(1 ln
T2
T1
!+ ln
q
(1 q))
T2
T1
!
+ O2666664
T2
T1
!23777775
(20a)
when T2/T1 ' 0, and
f (T2/T1, q) ="ln
1 q
q
! 2
1 2q#
1 T2T1
!+
12(1 2q)
1 T2
T1
!2
+ O2666664
1 T2
T1
!33777775
(20b)
when T2/T1 ' 1. The upper bound for q is given by the con-dition
T2/T1 = 0, satisfying ln[(1 q)/q] = 1/(1 q) andq
(T2/T1 = 0) ' 0.217 811 705 719 800 found numericallyand (T2/T1
= 0) = 0 exactly from Eq. (15b). T2/T1 = 1always satisfies f
(T2/T1, q) = 0 regardless of q values, butthe T2/T1 ' 1 asymptotic
case yields the lower bound forq
(T2/T1 ' 1) = (T2/T1 ' 1) ' 0.083 221 720 199 517 7found
numerically. Figure 3 shows the numerical solution(q, ) as a
function of
C
= 1 T2/T1, where the asymp-totic behaviors derived above hold
when T2/T1 ' 0 andT2/T1 ' 1. It seems that q is monotonically
decreasedand is monotonically increased, as T2/T1 is increased,
i.e.,q
min = q
(T2/T1 ' 1), qmax = q(T2/T1 = 0), min = 0, andmax = (T2/T1 ' 1).
For later, we numerically obtain thefunctional form of q(
C
) up to the quadratic term by fitting aquadratic function
near
C
= 0, in the range C
2 [0, 0.05] inthis case, which is
q
(C
) ' qmin + 0.0457524C + 0.02767322C . (21)f(T2/T1 = 0, q
) = ln
1 q
q
1
1 q = 0
! q ' 0.217 811 705 719 800
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(12q0)4{2a21+[(12q0)a212(1q0)q0a2]
ln[(1q0)/q0]}}/[(1q
20)
2q
20]. Using Eqs. (24), (25) and (26), Eq. (27) becomes sim-
ply
op =12
C
+182
C
+7 24q0 + 24q20
96(1 2q0)2 3C
+ O 4C
. (28)
-
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(12q0)4{2a21+[(12q0)a212(1q0)q0a2]
ln[(1q0)/q0]}}/[(1q
20)
2q
20]. Using Eqs. (24), (25) and (26), Eq. (27) becomes sim-
ply
op =12
C
+182
C
+7 24q0 + 24q20
96(1 2q0)2 3C
+ O 4C
. (28)
-
= ln
1 q
q
(1 C) ln
1 +
p(1 2q)2 + 4Cq(1 q)
1p
(1 2q)2 + 4Cq(1 q)
!q 1
2+
1
2
p(1 2q)2 + 4Cq(1 q)
q(1 q)0 = f(C , q)
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(12q0)4{2a21+[(12q0)a212(1q0)q0a2]
ln[(1q0)/q0]}}/[(1q
20)
2q
20]. Using Eqs. (24), (25) and (26), Eq. (27) becomes sim-
ply
op =12
C
+182
C
+7 24q0 + 24q20
96(1 2q0)2 3C
+ O 4C
. (28)
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd *
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
asymptotic form
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0(the horizontal axis). The asymptotic form on the right
side indicatesEq. (31).
schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ O 3C
. (22)
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
op
c = 1 T2 / T1
at (q*, *)CA = 11c
c/(2c)c/2
1+a[ln(1c)b](1c)
FIG. 5. The eciency at the maximum power op as the function
ofthe Carnot eciency
C
in Eq. (19) using numerically found optimalq
values, along with various asymptotic cases: the
Curzon-Ahlborneciency CA in Eq. (28), the upper bound C/(2C) and
the lowerbound
C
/2 in Ref. [19], and the function in Eq. (32) for C
0.65.
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 a0)a0(1 2a0)32
C
+O 3C
= 0 .
(23)
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =a0(1 a0)2(1 2a0) . (25)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ O 3C
.
(26)
Using Eqs. (24) and (25), Eq. (26) becomes simply
op =12
C
+182
C
+ O 3C
, (27)
which has exactly the same coecients up to the quadraticterm to
those of the Curzon-Ahlborn eciency [35] definedas
CA = 1 p
T2/T1 = 1 p
1 C
, (28)
with the expansion form
CA =12
C
+182
C
+1
163
C
+5
1284
C
+ O(5C
) , (29)
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd *
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(1
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(1
-
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q*(c0) = *(c0)
q*(c=1)
optim
al tr
ansi
tion
rate
s
c
q**
c0 and 1 asymptotes
= ln
1 q
q
(1 C) ln
1 +
p(1 2q)2 + 4Cq(1 q)
1p
(1 2q)2 + 4Cq(1 q)
!q 1
2+
1
2
p(1 2q)2 + 4Cq(1 q)
q(1 q)0 = f(C , q)
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
asymptotic form
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0(the horizontal axis). The asymptotic form on the right
side indicatesEq. (31).
schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ O 3C
. (22)
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
op
c = 1 T2 / T1
at (q*, *)CA = 11c
c/(2c)c/2
1+a[ln(1c)b](1c)
FIG. 5. The eciency at the maximum power op as the function
ofthe Carnot eciency
C
in Eq. (19) using numerically found optimalq
values, along with various asymptotic cases: the
Curzon-Ahlborneciency CA in Eq. (28), the upper bound C/(2C) and
the lowerbound
C
/2 in Ref. [19], and the function in Eq. (32) for C
0.65.
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 a0)a0(1 2a0)32
C
+O 3C
= 0 .
(23)
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =a0(1 a0)2(1 2a0) . (25)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
0
q(C ! 0) = q0 = (C ! 0) ' 0.083 221 720 199 517 7
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(12q0)4{2a21+[(12q0)a212(1q0)q0a2]
ln[(1q0)/q0]}}/[(1q
20)
2q
20]. Using Eqs. (24), (25) and (26), Eq. (27) becomes sim-
ply
op =12
C
+182
C
+7 24q0 + 24q20
96(1 2q0)2 3C
+ O 4C
. (28)
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd *
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
asymptotic form
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0(the horizontal axis). The asymptotic form on the right
side indicatesEq. (31).
schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ O 3C
. (22)
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
op
c = 1 T2 / T1
at (q*, *)CA = 11c
c/(2c)c/2
1+a[ln(1c)b](1c)
FIG. 5. The eciency at the maximum power op as the function
ofthe Carnot eciency
C
in Eq. (19) using numerically found optimalq
values, along with various asymptotic cases: the
Curzon-Ahlborneciency CA in Eq. (28), the upper bound C/(2C) and
the lowerbound
C
/2 in Ref. [19], and the function in Eq. (32) for C
0.65.
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 a0)a0(1 2a0)32
C
+O 3C
= 0 .
(23)
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =a0(1 a0)2(1 2a0) . (25)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ O 3C
.
(26)
Using Eqs. (24) and (25), Eq. (26) becomes simply
op =12
C
+182
C
+ O 3C
, (27)
which has exactly the same coecients up to the quadraticterm to
those of the Curzon-Ahlborn eciency [35] definedas
CA = 1 p
T2/T1 = 1 p
1 C
, (28)
with the expansion form
CA =12
C
+182
C
+1
163
C
+5
1284
C
+ O(5C
) , (29)
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd *
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(1
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
c1 asymptote
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0 (thehorizontal axis). The
C
! 1 asymptote indicates Eq. (34).schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ a33C
+ O 4C
. (22)
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 q0)q0(1 2q0)32
C
+ c3(q0, a1, a2)3C
+ O 4C
= 0 ,
(23)
where c3(q0, a1, a2) = [10q60 + 3a21 6q0(a21 + a2) 6q50(5 +
6a1+8a2)12q30(1+6a1+16a21+9a2)+q20(1+18a1+132a21+42a2)+q40(31+90a1+96a
21+120a2)]/[6(12q0)5(1q0)2q20].
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =q0(1 q0)2(1 2q0) . (25)
Similarly, the coecient a2 in Eq. (22) can also be written
interms of q0 alone, by letting c3(q0, a1, a2) = 0 in Eq. (23)
andusing the relations in Eqs. (24) and (25), as
a2 =7q0(1 q0)24(1 2q0) . (26)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
C
in Eq. (22). Then,
op =1
(1 2q0) ln[(1 q0)/q0]C
+
a1q03q20+2q30
+[q20+2a1q0(1+4a1)] ln[(1q0)/q0]
(12q0)3
ln2[(1 q0)/q0]2
C
+ d3(q0, a1, a2)3C
+ O 4C
,
(27)
where d3(q0, a1, a2) = {2(1 2q0)2a1[q20 + 2a1 q0(1 +4a1)]
ln[(1q0)/q0]+2[2q40+a14a212a2+4q0(4a21+3a2)+4q30(1+a1+4a2)2q20(1+3a1+8a21+12a2)]
ln2[(1q0)/q0]+(1
-
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q*(c0) = *(c0)
q*(c=1)
optim
al tr
ansi
tion
rate
s
c
q**
c0 and 1 asymptotes
= ln
1 q
q
(1 C) ln
1 +
p(1 2q)2 + 4Cq(1 q)
1p
(1 2q)2 + 4Cq(1 q)
!q 1
2+
1
2
p(1 2q)2 + 4Cq(1 q)
q(1 q)0 = f(C , q)
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
asymptotic form
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0(the horizontal axis). The asymptotic form on the right
side indicatesEq. (31).
schematically . . .
q
= q
no net work
as C is increased
q(C 0) = (C 0) 0.083 221 720 199 517 7
q(C = 1) 0.217 811 705 719 800(C = 1) = 0
FIG. 4. Illustration of the optimal transition rates (q, ) for
the max-imum power output as the T2/T1 value varies.
2. Asymptotic behaviors obtained from series expansion
The upper bound for q is given by the condition C
= 1,satisfying ln[(1 q)/q] = 1/(1 q) and q(
C
= 1) '0.217 811 705 719 800 found numerically and (
C
= 1) = 0exactly from Eq. (16b).
C
= 0 always satisfies Eq. (18) re-gardless of q values, so
finding the optimal q is meaningless(in fact, when
C
= 0, the operating regime for the engineis shrunk to the line q
= and there cannot be any positivework). Therefore, let us examine
the case
C
' 0 using theseries expansion of q with respect to
C
, as
q
= q0 + a1C + a22C
+ O 3C
. (22)
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
op
c = 1 T2 / T1
at (q*, *)CA = 11c
c/(2c)c/2
1+a[ln(1c)b](1c)
FIG. 5. The eciency at the maximum power op as the function
ofthe Carnot eciency
C
in Eq. (19) using numerically found optimalq
values, along with various asymptotic cases: the
Curzon-Ahlborneciency CA in Eq. (28), the upper bound C/(2C) and
the lowerbound
C
/2 in Ref. [19], and the function in Eq. (32) for C
0.65.
Substituting Eq. (22) into Eq. (18) and expanding the
left-handside with respect to
C
again, we obtain
2 (1 2q0) ln[(1 q0)/q0]2q0 1 C
+q0(1 q0) 2a1(1 2q0)
2(1 a0)a0(1 2a0)32
C
+O 3C
= 0 .
(23)
Letting the linear coecient to be zero yields
21 2q0 = ln
1 q0
q0
!, (24)
from which the lower bound for q(C
! 0) = q0 =(
C
! 0) ' 0.083 221 720 199 517 7 found numerically[lim
C
!0 U(C , q) = 1 2q, thus (C ! 0) = q(C ! 0)by Eq. (16b)]. Figure
3 shows the numerical solution (q, )as a function of
C
, where the asymptotic behaviors derivedabove hold when
C
' 0 and C
' 1. It seems that q ismonotonically increased and is
monotonically decreased,as
C
is increased, i.e., qmin = q(
C
! 0), qmax = q(C = 1),min = 0, and
max =
(C
! 0). Figure 4 illustrates the situ-ation on the (q, ) plane.
The linear coecient a1 in Eq. (22)can be written in terms of q0
when we let the coecient of thequadratic term in Eq. (23) to be
zero, as
a1 =a0(1 a0)2(1 2a0) . (25)
With the relations of coecients in hand, we find theasymptotic
behavior of op in Eq. (19) by expanding it withrespect to
C
after substituting q as the series expansion of
0 0
4
0
0.05
0.1
0.15
0.2
0 0.2 0.4 0.6 0.8 1
q* a
nd
*
c = 1 T2 / T1
q**
q*(c0) = *(c0)q*(c=1)
asymptotic form
FIG. 3. Numerically found q and values satisfying Eq. (18), as
afunction of
C
= 1T2/T1, along with the q(C ! 0) = (C ! 0)and q(
C
= 1) values presented in Sec. III B 2. (C
= 1) = 0(the horizon