1. Introduction
Many attempts have been made to identify a universal function whose extremum determines the development of a system far from equilibrium. Entropy production characterizes systems during nonequilibrium processes [
1,
2,
3,
4,
5], and its extrema may be used for determining the system behavior [
6,
7,
8,
9,
10,
11,
12,
13]. Recently, theoretical and experimental investigations have suggested the maximization of entropy production during nonequilibrium processes (the so-called maximum entropy-production principle, MEPP) [
10,
11,
12,
14,
15,
16]. According to MEPP, when a nonequilibrium system transitions from one state to another, it is characterized by the highest rate of entropy production. Analysis based on MEPP can thus be used to predict the transition point between two nonequilibrium states, such as those observed in the morphological changes of crystal growth, mode changes in droplet oscillation, and pattern changes in thermal convection [
12,
17,
18,
19]. For example, in the case of friction phenomena in the flow field, fluid velocity is treated as thermodynamic flux, and the transition point is predicted. For a pendant droplet that changes in the oscillation mode induced by the solutal Marangoni effect with viscous dissipation, the transition point of the oscillation mode is predicted from the intersection of the entropy-production curves determined from the velocity of the oscillating droplet, which is considered as thermodynamic flux [
18]. However, dissipative systems far from equilibrium frequently include solutions with several linearly stable branches, i.e., bistable behavior. In such cases, the selected solution depends on the initial conditions, and variational principles based on MEPP would not be required. The prediction of bistable behavior in various dissipative systems is considered an unsolved problem when using variational principles and overshadows the universality of MEPP. This problem is addressed in this study by examining these predictions for a situation involving bistable behavior, i.e., where hexagonal and roll flow patterns coexist during Rayleigh–Bénard convection.
  2. MEPP
The state of a nonequilibrium system is characterized by the thermodynamic flux 
 expressed as a function of the driving force of the entire system 
, which is proportional to differences in the temperature, concentration, pressure, etc. [
12,
17,
18,
19]:
      where 
 and 
 are constant coefficients.
Assume that there is a transition from state  to  as  increases. 
Each state is described by a local entropy-production 
 curve, which is characterized by the product of the local thermodynamic force 
 and thermodynamic flux 
 [
1]. Thus, the relationship between 
 and 
F can be expressed as [
12,
17,
18,
19]
      
According to MEPP, a nonequilibrium system develops in a manner that maximizes its entropy production under the given binding conditions [
10,
14]. The transition point between states 
 and 
 corresponds to the intersection of the two 
 curves A and not that of the two 
 lines B. Although the coefficients of 
 in 
 are not strictly equal to 
 in Equation (1), there is no change in the intersection. Thus, for simplicity, we use the same coefficients. Before intersection A, the entropy-production curve of state 
 lies above that of state 
, whereas after intersection A, the converse is true. It should be noted that when entropy production is expressed as a function of 
, 
 cannot distinguish between the nonequilibrium states because all 
 are present on a single quadratic curve 
 [
19], where 
 is the phenological coefficient [
1]. Furthermore, we cannot predict the transition point nor even understand whether another kind of nonequilibrium state exists in the system when the thermodynamic flux expressed as a function of the driving force can be described only on a single line. Even at equilibrium, if we know the thermodynamic properties of water only in the liquid phase at 1 atm, we cannot predict that water will undergo a phase transition to the gas phase at 100 °C. If we know the dependence of the chemical potential of water in both phases on temperature, we can predict the boiling point of water from the intersection of their two chemical-potential curves.
As shown in 
Figure 1, we find that intersection A represents the transition point between the nonequilibrium states. However, we wish to understand intersection B of the thermodynamic-flux lines in terms of the physical behavior of the nonequilibrium system. It transpires that this intersection is a starting point for the coexistence of two different nonequilibrium states. When 
, state 
 begins to manifest because the thermodynamic flux of state 
 is higher than that of state 
. However, the system mainly comprises state 
 because 
 of state 
 remains greater than that of state 
. Therefore, states 
 and 
 coexist until 
. The thermodynamic flux of state 
 increases continuously with 
F, so the proportion of state 
 will increase until it represents a major part of the system. Once 
, the system consists only of state 
.
  3. Thermodynamic Analysis of Rayleigh–Bénard Convection
To verify this MEPP prediction, precise experimental data are required to calculate the relationship between the thermodynamic flux and driving force of a dissipative system exhibiting bistability. One example is reported in the literature [
20], where hexagonal and roll patterns coexisted during Rayleigh–Bénard convection, subject to external temporal modulation of the reduced Rayleigh number 
, where 
 is the temperature difference between the bottom and top plates of the water-filled container and 
 is the critical temperature difference for the onset of convection without modulation. Here, 
 has the form
      
      where the time 
t and frequency 
 are scaled according to the vertical thermal-diffusion time and 
 is the amplitude of modulation. The 13-mode Lorenz model proposed by Hohenberg and Swift predicts a positive threshold-shift change in the convection onset from 
 to 
 [
21]. Above 
, the roll patterns that appear through supercritical bifurcation are unstable to the hexagonal patterns (reproduced in the inset of 
Figure 2a) [
21]. This continues as 
 increases until 
, beyond which the roll patterns are stable. Hexagonal patterns manifest through subcritical bifurcation, first becoming stable at 
, which continues until 
, where they are unstable to roll patterns. For 
, both hexagonal and roll patterns are stable. Meyer reported that the bistable region for the range of loop 
 was approximately two orders of magnitude larger than that for the loop between 
 and 
 [
22]. Therefore, it is easy to resolve the bistable region experimentally.
Heat flux 
 is a dimensionless quantity given by the ratio of the convective heat flux to the heat flux conducted through the fluid. When 
, convective flow occurs and 
 becomes positive. Meyer, Cannell, and Ahlers performed experimental observations of Rayleigh–Bénard convection and the heat flux, as shown in 
Figure 2a [
20]. They detected the bistable region where both hexagonal and roll patterns were stable. The model quantitatively and qualitatively agrees with the experimental results in the pure-hexagonal 
, bistable 
, and pure-roll 
 regions for 
, where 
, 
, and 
.
Let us begin by predicting the transition point between heat conduction and heat convection using MEPP with static measurements without modulation. This is easy to predict because of the linear relationship between  and  in each state.
Here,  corresponds to the driving force of the entire system . The values of  may be linearly fitted as a function of  in each state. The heat flux for heat conduction is obviously zero. The best-fit line for  of heat convection with  is . The entropy production of the heat-conduction and heat-convection regions is easily calculated from Equation (2), where  and , respectively. Thus, we obtain the intersection . It is clear that the transition occurs at , because of the definition of heat flux wherein the positive value of  signifies the occurrence of heat flux produced by only convective flow; however, it is important that this is predicted using MEPP.
Next, we analyze Rayleigh–Bénard convection subject to external temporal modulation on the MEPP basis. The transition points cannot be distinguished easily because the values of 
 show rounded bifurcations due to modulation from heat conduction to convection; for the heat-convection region (
), they align approximately along a single line that changes only slightly in slope. The derivative of 
 with respect to 
 enables recognizing the point of change in the slope (
Figure 3a).
For 
, 
 increases monotonically with 
, whereas for 
, it remains approximately constant. Thus, we can divide the heat-convection region in two based on the change in 
. The values of 
 in the two regions can be linearly fitted as functions of 
, yielding 
 and 
 (
Figure 3b). The former corresponds to the hexagonal convection and the latter to roll convection. Using the method described above yields two curves for entropy production of the hexagonal and roll flows, respectively: 
 and 
. These curves intersect at one point, 
, even though the two curves appear to overlap.
This intersection corresponds to a transition point between the hexagons and rolls, and it is in good agreement with 
 where the rolls become stable to the hexagonal patterns in both the experimental and theoretical results. Furthermore, the intersection of the two 
 lines occurs at 
. The value of 
 is close to a starting point for the bistable region, 
, where both hexagonal and roll patterns are stable in the experimental and theoretical results. The actual patterns obtained by Meyer et al. are shown in 
Figure 4.
For 
, the roll patterns gradually overlap with the stable hexagonal patterns, and coexisting patterns persist until 
. As shown in 
Figure 3b, the 
 of the rolls lies above that of the hexagons, whereas 
 of the rolls lies below that of the hexagons. For 
, 
 and 
 for the rolls lie above those for the hexagons, and the actual patterns show that only the rolls are stable.
The experimental values of 
 represent the total heat flux produced by the three directional components of heat flux, i.e., 
x, 
y, and 
z. The vertical component, which is aligned with the temperature difference 
 between the bottom and top plates, accounts for a large proportion of the total heat flux; hence, it obscures the slight change produced by the horizontal components [
19]. Thus, the experimental results do not show the jump in 
, as indicated by the schematic in 
Figure 1. If the effect of the vertical component is removed from the total flux and the heat flux is analyzed only in the direction perpendicular to 
, the jump in 
 can be observed, and more accurate transition points may be obtained.
  4. Discussion
The state with the highest entropy is the state where intensive variables are uniform in the entire system. However, the dissipative system applied by external forces, such as difference in temperature, concentration, and pressure fields, never develops into a uniform state. Under the given binding conditions, in order for the dissipative system to approach more rapidly the state with a uniform field, heat, molecules, and the fluid momentum must transfer faster. In Rayleigh–Bénard convection, the heat transfer of the system changes from heat conduction to heat convection in a hexagonal pattern, and then to heat convection in a roll pattern, and finally to turbulence as the difference in temperature increases. The state of heat transfer changes such that the system more rapidly approaches the uniform temperature field. Therefore, the system should realize the state with higher damping and higher energy consumption by changing the flow state to approach the uniform field more rapidly.
The accuracy of the transition point predicted from our thermodynamic approach depends on the precision of the data on the relationship between the thermodynamic flux and driving force. Thus, the predicted transition point necessarily involves some uncertainty. However, as shown in 
Figure 2 and 
Figure 3, in Rayleigh–Bénard convection, the state with the highest entropy production is more stable than a state with lower entropy production. Bistability occurs when a new state has higher thermodynamic flux and the existing state has higher entropy production. It is a significant achievement that Rayleigh–Bénard convection cannot develop against the above rules.