Arxiv:2001.10149V5 [Physics.Chem-Ph] 17 Dec 2020 a Prescribed Energy-Dissipation Law, in Which the Choice (CDF) [14, 15]

Field Theory of Reaction-Diﬀusion: Law of Mass Action with an Energetic Variational Approach

Yiwei Wang∗ and Chun Liu† Department of Applied Mathematics, Illinois Institute of Technology, Chicago, IL 60616, USA.

Pei Liu‡ School of Mathematics, University of Minnesota, Minneapolis, MN 55455, USA.

Bob Eisenberg§ Department of Applied Mathematics, Illinois Institute of Technology, Chicago, IL 60616, USA. and Department of Physiology and Biophysics, Rush University, 1750 W. Harrison, Chicago IL 60612.

We extend the energetic variational approach so it can be applied to a chemical reaction system with general mass action kinetics. Our approach starts with an energy-dissipation law. We show that the chemical equilibrium is determined by the choice of the free energy and the dynamics of the chemical reaction is determined by the choice of the dissipation. This approach enables us to couple chemical reactions with other effects, such as diffusion and drift in an electric field. As an illustration, we apply our approach to a non-equilibrium reaction-diffusion system in a specific but canonical setup. We show by numerical simulations that the input-output relation of such a system depends on the choice of the dissipation.

I. INTRODUCTION The macroscopic dynamics of chemical reactions are often described by the law of mass action, which states Many biological processes can be abstractly repre- that the rate of a reaction is proportional to the concen- sented as biochemical networks, in which chemical reac- trations of the reactants [17, 18]. The law of mass action tions are catalyzed by enzymes and combined to perform originally arises from the treatment of ideal gases (so- many of the functions of life. Examples include metabolic lutions) [19], where molecules/atoms only interact when pathways and the electron transport chain that power they collide. Although the mass-action type kinetics has life [1, 2]. In these systems, reactions occur in different been widely used for different chemical reaction systems, physical locations, so the products of one reaction move, it is a phenomenological theory, of which the underly- by diffusion (and perhaps migration and convection), to ing physical foundation is unclear, as molecules can in- become reactants for another reaction in a different lo- teract in many different ways. As aptly pointed out in cation. To describe a complex biological system, and Ref. [18], “the law of mass action is not a law in the consistently deal with the coupling between reaction and sense that it is inviolable, but rather is a useful model, diffusion, as well as other mechanical effects, one needs much like Ohm’s law or Newton’s law of cooling.” Since to turn to a variational theory. The variational princi- 1950’s, there has been a huge amount of work devoted ple guarantees a consistent mathematical formulation, in to studying the thermodynamics basis and mathemati- which all variables satisfy all equations, of all fields, and cal structures of chemically reacting systems [14, 16, 20– their boundary conditions, with one set of parameters in 54]. In particular, many papers extend the variational a certain region. principle for mechanical systems to the reaction kinetics For mechanical systems, inspired by the seminal work by building the analogies between Newtonian mechanics of Rayleigh [3] and Onsager [4, 5], various variational the- and chemical reactions [34, 35, 55, 56]. ories have been developed. Examples includes the Ener- The goal of this paper is to extend the framework of getic Variational Approach (EnVarA) [6, 7], the general EnVarA [6, 7, 57], which has dealt with flows in systems equation for the nonequilibrium reversible-irreversible with many components successfully for many years, to a coupling (GENERIC) [8–11], Doi’s Onsager principle chemical reaction system. We model a reaction system by [12, 13], and the Conservation-Dissipation Formalism

arXiv:2001.10149v5 [physics.chem-ph] 17 Dec 2020 a prescribed energy-dissipation law, in which the choice (CDF) [14, 15]. However, it is not straightforward to of the free energy determines the chemical equilibrium (if apply these variational principles to a chemical reaction it exists), and the choice of the dissipation determines the system, which cannot be understood from Newtonian me- dynamics of the chemical reaction. The classical law of chanics [16]. mass action can be derived from a particular choice of the energy-dissipation law. Our approach is non-equilibrium and provides a basis to couple the chemical reaction with ∗ [email protected] the effect of other fields, such as diffusion, drift in an † [email protected] electric field, as well as the thermal effects. As an il- ‡ [email protected] lustration, we apply our approach to a non-equilibrium § [email protected] reaction-diffusion system, which can be viewed as an ab- 2 stract building block of biological networks with inputs A. Energetic Variational Approach and outputs. Our analysis shows that the input-output relation of such a system depends on the choice of the dis- We start with a brief introduction to the classical en- sipation, which indicates the dissipation can be inferred ergetic variational approach (EnVarA), which was devel- from experimental measurements. oped from the variational principle, proposed by Rayleigh [3] for purely frictional systems, that Onsager tried to extend to physical systems in general [4, 5]. II. FIELD THEORY OF REACTION-DIFFUSION The starting point of an energetic variational approach is a prescribed energy-dissipation law for an isothermal To clarify our ideas, we first consider a single reversible and closed system, which comes from the first and second reaction law of thermodynamics [7]. Indeed, for a thermodynamic process without transfer of matter, the first law of ther- αA + βB )−−−−* γC, (1) modynamics is often formulated as where α, β and γ are stoichiometric coefficients. Let ci d (K + U) = W˙ + Q,˙ (6) (i = A, B, C) denote the concentration of each species. d Due to conservation of elements, one must have that is the rate of change of the kinetic energy K and the d d (γc + αc ) = 0, (γc + βc ) = 0, (2) internal energy U can be attributed to either the work dt A C dt B C W˙ done by the external environment or the heat Q˙ . To analyze heat, one needs to introduce the entropy S, which which is known as the stoichiometric constraint [27]. As satisfies a time dependent version of the second law of a consequence, each component of c = (c , c , c )T sat- A B C thermodynamics: isfies the ordinary differential equation dS d T = Q˙ + 4, 4 ≥ 0, (7) c = σ r(c), (3) dt dt i i where T is the temperature and 4 is the entropy produc- T where σ = (−α, −β, γ) is the stoichiometric vector, and tion. Subtracting the two laws, one arrives at an energy- r(c) is known as the reaction rate [17]. dissipation law The law of mass action gives a particular form of the reaction rate r(c), d Etotal(t) = −D(t), (8) dt α β γ r(c) = kf cAcB − krcC , (4) for isothermal and closed system (W˙ = 0). Here Etotal is where kf and kr are rate constants for forward and the total energy, which is the sum of the Helmholtz free reverse directions. This form originally arose from energy F = K − T S and the kinetic energy K. D is the the treatment of ideal gases (solutions) [19], where rate of energy dissipation that is related to the entropy molecules/atoms only interact when they collide. At an production. equilibrium, in which the concentrations are not chang- For a given energy-dissipation law, the energetic vari- ing, we have ational approach provides a paradigm to determine the dynamics of system through two distinct variational pro- ∞ α ∞ β (cA ) (cB ) kr cesses: the Least Action Principle (LAP) and the Max- ∞ γ = , Keq, (5) (cC ) kf imum Dissipation Principle (MDP) [6, 7]. Specifically, the Least Action Principle states that the dynamics of a kr where Keq = is called the equilibrium constant. The kf Hamiltonian system is determined by a critical point of law of mass action is an empirical law without a clear R T the action functional A(x) = 0 K − Fdt with respect physical interpretation, so it can not be immediately ap- to x (the trajectory in Lagrangian coordinates, if appli- plied to biological and electrochemical systems in which cable) [7, 58], i.e., chemical reactions are coupled with other effects. In the remainder of this section, we show that the sim- Z T Z ple reaction kinetics (3) can be modeled by an extended δA = (finertial − fconv) · δxdxdt, (9) 0 Ω energetic variational approach,like a mechanical system, which provides a basis of coupling chemical reaction with where finertial is the inertial force and fconv is the conser- other mechanisms, including mechanical effects such as vative force. Formally, the LAP represents the fact that diffusion, drift in an electric field, as well as the ther- force multiplies by distance is equal to the work, i.e., mal effects. As an application, we also provide an ener- δE = force × δx, where x is the location, δ represents the getic variational formulation to a reaction-diffusion sys- variation/differential. In the meantime, for a dissipative tem, which is a typical example of mechano-chemical or system (D ≥ 0), we follow Onsager [4, 5] and determine chemomechanical systems. the dissipative force fdiss by minimizing the dissipation 3 functional D with respect to the “rate” xt, known as the B. EnVarA with chemical reaction Maximum Dissipation Principle (MDP), i.e., As well as any other variational principles, classical en- 1 Z δ D = fdiss · δxt dx. (10) ergetic variational approaches deal with mechanical sys- 2 Ω tems, which are indeed based on the Newton’s second law The dissipation D is often assumed to be quadratic in F = ma. In general, chemical reactions cannot be understood from Newtonian mechanics, as there is no clear terms of the “rate” xt [12], that is mechanical interpretation for the chemical potential [16]. Z Many papers try to build an Onsager type variational D[x, x ] = G(x)x · x dx, (11) t t t theory for chemical reaction systems [20, 35, 37, 56, 61– 64].For example, Mielke established the gradient flow where G(x) is a positive semi-definite matrix for given x. structure for reaction-diffusion systems with reversible The assumption (11) corresponds to the linear response mass-action kinetics by using the dual dissipation poten- theory in non-equilibrium thermodynamics [4, 5, 59]. tial [37]. As an extension of the GENERIC framework, According to force balance (Newton’s second law, in Grmela showed the geometry associated with the law of which the inertial force plays the role of ma), we have mass action is the the contact geometry. He extended the δA 1 δD mass-action kinetics to account for the influence of inertia = , (12) and fluctuations, which can be adopted to complicated δx 2 δxt reaction systems involving many intermediate reactions which defines the dynamics of the system. [38]. The EnVarA framework shifts the main task of model- For the reaction (1) with the law of mass action, it ing to the construction of the energy-dissipation law. As has been discovered for a long time that there exists a an illustration, we consider a simple example originally Lyapunov functional [16, 21, 37, 38, 46, 50, 65, 66], which proposed by Lord Rayleigh [3], a spring-mass system, in is the free energy of the system. The free energy can be which a Hookean spring of which one end is attached to written down in various equivalent form; here we adopt a wall and another end to a mass m. Then, a thermodynamics based form m k F(c , c , c ; U ,U ,U ) K = x2, F = x2, D = γx2, A B C A B C 2 t 2 t Z = RT cA(ln cA − 1) + cB(ln cB − 1) + cC (ln cC − 1) where k is the spring constant, and γ is damping coef- Ω ficient. The corresponding action functional is defined + cAUA + cBUB + cC UC dx, as (16) for the chemical reaction (1). The first three terms in Z T m k 2 2 (16) form the free energy of a mixture of ideal gases with- A = xt − x dt. 0 2 2 out chemical reactions, which corresponds to the entropy. Then the LAP, i.e. variation of A with respect to the Indeed, for a mixture of ideal gases with N species, the trajectory x(t) gives rise to chemical potential of a substance j is expressed by [67] 0 δA µj = µ + RT ln xj, (17) = −mxtt − kx. (13) δx 0 where µ is the reference chemical potential, and xj is Meanwhile, the MDP, taking the variation of D with re- the concentration of the substance j. Since the chemical spect to xt gives potential is defined relative to its value at an arbitrary reference state, we can take µ0 = 0. The free energy of 1 δD the mixture of ideal gases, corresponding to the chemical = γxt. (14) 2 δxt potential (17) with µ0 = 0, is given by

Hence, the force balance condition (12) yields N Z X F[xi] = RT xi(ln xi − 1)dx. (18) mxtt + kx + γxt = 0. (15) Ω i=1 In an overdamped case (m γ), the mxtt term can be The last three terms in (16) can be viewed as internal neglected [60], and the system becomes a gradient flow energies stored inside the molecular A, B and C. In the with the dynamics given by case without chemical reaction, since cA, cB and cC do 1 δF not change with respect to time, these terms are con- x = − , t γ δx stants that can be ignored. From a modeling perspective, as also pointed out in [38], Ui are parameters that In the following, we always working on the overdamped determine the equilibrium of the system. For the given region, and neglect the kinetic energy in (8). free energy F(cA, cC , cC ; UA,UB,UC ) defined in (16), the 4 corresponding chemical potential of each species is given to model the reaction kinetics of the chemical reaction by (1). Here D[R,Rt] is a dissipation of the system. Differ- δF ent choices of D[R,Rt] determine different reaction ki- µi = = RT ln ci + Ui, i = A, B, C (19) netics. δci Unlike mechanical systems, chemical reactions are of- At a chemical equilibrium, the chemical potential of both ten far from thermodynamic equilibrium, so the dissi- sides of the reaction are equal, i.e., the affinity pation D[R,Rt] may not be quadratic in terms of Rt [35, 59]. In order to deal with the general form of the γµ − αµ − βµ = 0, (20) C A c dissipation, we need to extend the classical EnVarA. As- which indicates that sume D(R,Rt) takes the form ∞ α ∞ β (cA ) (cB ) 1 ∆U D[R,Rt] = (Γ(R,Rt),Rt) ≥ 0, (25) ln ∞ γ = (γUC −αUA −βUB) := . (21) (cC ) RT RT where (., .) is an inner product, since Here ∆U = γUC −αUA −βUB is the difference of internal energy between the state {αA, βB} and the state {γC}. d δF Then the equilibrium constant K is defined as [18] F[R] = ,R , (26) eq dt δR t ∞ α ∞ β (c ) (c ) ∆U A B RT Keq , ∞ γ = e , (22) the energy-dissipation law (24) implies (cC ) which is an exponential representation of the difference δF Γ(R,Rt) = − , (27) in internal (‘chemical’) energies. δR In our approach, we always assume the existence of the which is the equation for the chemical kinetics. Interest- free energy F, which is different from most of previous ingly, notice that approaches. Those approaches start with the mass-action kinetics and show the existence of the free energy under δF X = σ µ , (28) the detailed balance condition [37, 65]. For a general sys- δR i i tem, the free energy F might contains various different i=1 mechanism and cannot be derived by mathematics alone is exactly the affinity of chemical reaction, as defined by until a physical model is specified. Here we assume that De Donder [68, 69]. The affinity plays a role of the “force” Ui are constants to illustrate our approach. Confronta- that drives chemical reactions, and Rt can be identified as tion with real experimental data will undoubtedly moti- the reaction velocity (or rate of conversion [56]). Just as vate more complex models. It should be emphasized that in a mechanical system, the dissipation of this chemical the choice of the free energy F determines the chemical reaction system gives the relation between the reaction equilibrium (if it exists) of the system. velocity Rt and the chemical force. Next we discuss two As pointed out in [27], one of the difficulties in applying typical choices of the dissipations. variational principles to a chemical reaction arose from General Law of mass action: The law of mass action the stoichiometric constraint (2). To overcome this diffi- can be derived from the energy-dissipation law (24) by culty, Oster and Perelson treated the reaction kinetics in choosing a differential geometric context and introduced the “reaction trajectory”. The idea of using reaction trajectory, Rt D[R,Rt] = RT Rt ln γ + 1 . (29) also known as extent of reaction or degree of advance- krcC ment, as a new stable variable can be traced back to De Donder [56], and has been used for both determinis- Indeed, the energetic variational procedure gives tic and stochastic descriptions of chemical reactions for Rt δ a long time [34, 39, 46]. Roughly speaking, a reaction RT ln γ + 1 = − F[R]. (30) trajectory accounts for the “number” of forward chemical krcC δR reactions that has occurred by time t. By introducing Notice that the reaction trajectory R(t), the concentrations of A, B and C for the single chemical reaction (1) are given by γ ! δ cc F[R] = RT ln − αUA − βUB + γUC , δR α β ci(t) = ci(0) + σiR(t), (23) cAcB which can be viewed as the kinematics of the chemical which indicates that reaction that embodies the constraint (2). ! By using the reaction trajectory, we can reformulate R cα cβ ∆U ln t + 1 = ln A B − , (31) the free energy F(cA, cB, cC ; UA,UB,UC ) defined in (16) γ γ krc c RT in terms of R(t), and use the energy-dissipation law C C d where the right-hand side is determined by the difference F[R; U ,U ,U ] = −D[R,R ], (24) dt A B C t of internal energy ∆U between the state {αA, βB} and 5 the state {γC}. Although (31) looks complicated, direct the chemical reaction, we have Rt ≈ 0 near the equilib- computation shows that rium, then the Taylor expansion gives us α β ! Rt η1(R) 2 γ 1 cAcB α β γ η1(R) ln + 1 ≈ |Rt| . (36) Rt = krcC γ − 1 = kf cAcB − krcC , (32) η2(R) η2(R) Keq cC Thus, one can view the dissipation (34) as a linear ap- which is the classical law of mass action. Here the rela- ∆U proximation near equilibrium to (33). However, it is be- kr tion Keq = e RT = is used to get the last equality. It kf lieved that, except for the special case that is close to is worth mentioning that the dissipation (29) is identical equilibrium, the driving force for chemical reaction is a to a widely used form of the entropy production [16, 49] nonlinear functional of the system variables [35, 59]. It is straightforward to extend the above EnVarA de- rf scription to a general reversible chemical reaction system 4 = (rf − rr) ln , rr contains N species {X1,X2,...XN } and M reactions, given by where rf and rr are forward the reverse reaction rates. l l l l l l As a generalization of (29), we can consider a more α1X1 + α2X2 + . . . αN XN )−−−−* β1X1 + β2X2 + . . . βN XN , general form of the dissipation T N for l = 1,...,M. Let c = (c1, c2, . . . , cN ) ∈ R be Rt the concentrations of all species. The kinematics of the D[R,Rt] = η1(R)Rt ln( + 1), (33) system are then given by η2(R)

c = c0 + σR, (37) where η1(R) > 0 and η2(R) > 0, then D[R,Rt] ≥ 0 for the admissible R. By choosing η (R) and η (R) prop- M 1 2 where c0 is the initial concentrations, R ∈ R repre- erly, we can have a concentration dependent reaction N×M sents M reaction trajectories of M reactions, σ ∈ R rate, which is often used to provide a thermodynamic l l with σil = βi − αi is the stoichiometric matrix. The re- description of an autocatalytic chemical reaction [18]. action kinetics of this chemical reaction network can be Linear Response Theory: In nonequilibrium thermo- described by the energy-dissipation dynamics, it is often assumed that the dissipation of the d total energy is a quadratic function of “rate” of change F[c(R)] = −D[R, ∂ R], (38) of state variables, which is known as the linear response dt t theory. [4, 5, 59]. In our case, the linear response theory where gives a form of the dissipation term N 2 X D[R,Rt] = η(R)|Rt| . (34) F[c] = ci(ln ci − 1) + ciUi, (39) i=1 Then the variational procedure gives with Ui be the internal energy, and the dissipation can γ ! be taken as ∂ cC η(R)Rt = − F[R] = RT ln − ∆U. ∂R α β M ∂ R cAcB X t l D[R, ∂tR] = − ∂tRl ln + 1 (40) ηl(c(R)) By choosing η(R) = RT, the reaction rate is given by l=1 to be consistent with mass action kinetics. Then the α β ! 1 cAcB variational procedure gives the dynamics of the chemical r = Rt = ln γ , (35) reaction Keq cC ∂ R δF ln t l + 1 = − , (41) a form of which is more complicated than the law of mass η (c(R)) δR action. l l where δF is the affinity of the l-th chemical reaction. Remark II.1. The law of mass action gives a simple δRl form of the reaction rate r in terms of concentrations, however, the dissipation in terms of R and R becomes t C. Reaction-Diffusion System complicated (See eq. (29)). On the other hand, if the dissipation is taken to be simple that described by linear response theory, the the reaction rate r becomes compli- The above EnVarA description of a chemical reaction cated (See eq. (35) ). provides a way to couple chemical reactions with other mechanical mechanisms, such as diffusion and electro- Some early variational treatments of chemical reactions diffusion, in a unified variational framework. As an illus- are based on the linear response assumption [55], which tration, we apply the EnVarA to a reaction-diffusion sys- arose from the near equilibrium assumption. Indeed, for tem, which is a simple example of a mechano-chemical or 6 chemo-mechanical system. Reaction-diffusion type par- share the same free energy but have different dissipa- tial differential equations are used widely to model bio- tion mechanisms [32, 40]. In [43], the authors develop a logical processes [66], such as molecular motors [70], prion novel approach that couples chemical kinetics with non- diseases [71], and tumor growth [72]. dissipative time reversible mechanics, such as elastic de- Consider a reaction-diffusion system in a fixed domain formations, which has potential applications in biology. Ω with the reaction given by (1), then the kinematics for We refer interested readers to [11, 43] for the mathemat- the concentrations cA, cB and cC are given by ical formula of such a type of coupling.

∂tci(x, t) + ∇ · (ciui) = σi∂tR(x, t), i = A, B, C (42) III. INPUT-OUTPUT RELATION AND THE where ui is the macroscopic velocity of different species DISSIPATION induced by the diffusion process, R is the reaction trajectory for the chemical reaction (1). As pointed out previously, in the EnVarA framework, The energy-dissipation law of the reaction-diffusion the dynamics of chemical reactions are determined by system can be formulated as the choice of the dissipation. Notice that the equilibrium constant K is determined by the choice of the free en- d eq F(cA, cB, cC ) = −(Dchem + Dmech) (43) ergy, measurements of just Keq cannot distinguish differ- dt ent dissipation mechanisms. In the meantime, although chemical reactions are believed to operate far away from where the free energy F(cA, cB, cC ) is given by Eq. (16), equilibrium [35, 59], directly simulating the ODE system which is same as for a pure reaction system. Dchem is the dissipation arises from the chemical reaction, which for the two dissipations (32) and (35) produce almost identical results since both systems move to the equilib- is given by Dchem = (Γ(R,Rt),Rt) ≥ 0 as in the last rium so quickly. To distinguish different reaction kinetics, subsection. Dmech is the dissipation due to the diffusion process, which is often taken as [73] it is necessary to study a non-equilibrium system, which can predict different dependence of rate on concentra- Z 2 2 2 tions and different time courses of the chemical reaction. Dmech = ηA(c)|uA| + ηB(c)|uB| + ηC (c)|uC | dx In this section, we study a particular setup, shown in Fig. 1, that can be realized in experiments. to model the friction of the fluid fluxes. It is important to notice that in this case, the dynamics of both the mechanical and chemical parts are derived from the same free energy. Notice that 3 d X δF !!### " F(c , c , c ) = (∇µ , u) + ,R , (44) dt A B C i δR t i=1 by using the generalized energetic variational approach, the equations for R and ui can be derived as  δF  Γ(R,Rt) = − FIG. 1: Setup of an open nonequilibrium system  δR (45)  δF Such a setup is chosen to give reproducible input-  ηi(c)ui = −ci∇ , i = A, B, C. δci output functions for different dissipations. In this system, a narrow channel connects two bath, as shown in Here the first equation is the same as (27), and the sec- Fig. 1. We assume the chemical reaction ond equation is actually Fick’s Law of diffusion [7]. By A + B −−* C (46) choosing ηi(c) = ci(i = A, B, C) and combining (45) with )−− (42), we can obtain a reaction-diffusion system happens inside the channel, and the average concentra- ∂ c = ∇ · (∇c ) + σ r(x, t) tions of A and B in the left bath can be maintained by t i i i the boundary condition. The species in the left bath where r(x, t) is the reaction rate determined by the choice are sources, and the species in the right bath are outputs. The chemical reaction is the “transfer function”. of Dchem(R, ∂tR) as discussed in the last subsection. The sources provided by the “left bath” can keep the sys- Remark II.2. It is worth mentioning that here we only tem away from the equilibrium. couple the chemical reaction with dissipative mechanics This system can be viewed as an abstract representa- (e.g. diffusion). The chemical and mechanical parts tion of one component of complex biological networks, in 7

10 which is the output of our system. The flux of C or the rate of change of amount of C in the right bath is d 8 defined as C . The initial concentrations of A, B dt out 0 0 and C in the channel are constants cA(x) = cB(x) = c0 0 6 and cC (x) = 0.1. We fix Keq = 0.1 and assume the free energy is given by 4 Z F = cA ln (0.1cA) − 1)

2 + cB(ln (0.1cB) − 1) + cC ln (cC ) − 1)dx. We focus on two types of dissipations, a generalized law 0 of mass action 0 0.5 1 1.5 2 D1(R, ∂tR) = Rt ln(Rt + 1), (48) and a dissipation based on the linear response assumption 2 FIG. 2: The output Cout(t) as a function of input c0 when t = 1 D2(R, ∂tR) = |Rt| . (49) for dissipation (48) [circle] and dissipation (49) [square]. These two dissipations (48) and (49) are almost same near a equilibrium (see (36)). By numerical simulations, which a enzyme localizes a particular chemical reaction we show that the input-output relation depends on the and moves the reactants into products. This representa- choice of the dissipation in this nonequilibrium setup. tion links chemical reactions to the two terminal devices Fig. 2 shows the output Cout(t) as a function of the of electrical and electronic engineering [1, 28, 47, 52]. input c0 at t = 1 for the two choices of dissipation. For Each reaction is a separately defined device (loosely small c0, the outputs are nearly same for the two dissipa- speaking) with an input and output and its own input- tion functionals. However, the output for dissipation (48) output relations. The enzymes can be thought as two ter- is much larger than that for the dissipation (49) when c0 minal devices, as diodes, that move reactants into prod- is large. Formally, from the computations in Sec. II, we ucts, from one chemical state to another, much as chan- 1 cAcB know Rt = ln , for the dissipation (48), while nels are diodes that move ions from one physical location Keq cC 1 cAcB Rt = − 1 for the dissipation (49). For c0 = 0.1, to another through a reaction path [28, 74]. Although Keq cC treating chemical reaction systems by electric circuit the- the system is at the equilibrium, so Cout = 0. When ory has existed for a long time, the spatial effect seems c0 is large, the dissipation (49) will determine a larger to be overlooked. Reactions in biology occur in different reaction rate. d physical locations, so the products of one enzyme’s reac- We also consider dt Cout as a function of t for two tion move, by diffusion (and perhaps migration and con- choices of dissipation for various of c0. The results are vection), to become reactants for the reaction catalyzed shown in Fig. 3. The time courses in Fig. 3 show that for d by an enzyme in a different location. different dissipations and different inputs, dt Cout tends Mathematically, since the channel is very narrow, we to a constant, which is a function of input for a given can treat this problem as one one-dimensional, with the dissipation. domain given by [−, ]. We fix = 0.1 through this Although the dissipation (48) and (49) are almost the section. As mentioned previously, the concentrations of same when near equilibrium, the above simulations indi- A and B in the left bath are maintained, which gives cate that in a non-equilibrium setting, the input-output us the Dirichlet boundary conditions of A and B in the relationship might be very different for different choices of left-end of the channel. We can impose the boundary dissipations since the system is maintained far from equi- conditions librium due to inputs of reactants through the boundary c (−, t) = c , ∂ c (, t) = 0, condition. This suggests that one might be able to deter- A 0 x A mine the dissipation through experimental measurements cB(−, t) = c0, ∂xcB(, t) = 0, and solving the inverse problem [75]. ∂xcC (−, t) = 0, cC (, t) = 0.1, and treat c0 as the single input of our system. IV. SUMMARY Since cC satisfies the Dirichlet boundary condition on the right-end of the channel, we can define the amount of C diffuse into right bath by time T as In this paper, we apply a generalized energetic variational approach (EnVarA) to a reversible chemical reac- Z Z tion system, which enables us to couple chemical reac- Cout(T ) = R(x, T )dx− (c(x, T )−c0(x))dx, (47) − − tions with other mechanical effects, such as diffusion, as 8

(a) (b) 1 4 0.25 0.8 0.50 3 1.00 0.6

2 0.4

1 0.2 0.25 0.50 1.00 0 0 0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1

d FIG. 3: dt Cout as a function of t for two dissipations for various of c0 (c0 = 1, 0.5 and 0.25 from top to bottom in each figure). (a) Dissipation (48), (b) Dissipation (49). well as the thermal effect. In our approach, the chemical Appendix A: Numerical Method equilibrium (if it exists) is determined by the choice of the free energy, and the dynamics of a chemical reaction is In the appendix, we give a detailed description of the determined by the choice of the dissipation. The classical numerical method that we used to study the reaction- law of mass action can be derived through a particular diffusion system in Sec. III, which is based on the ener- form of the dissipation. getic variational formulation proposed in this paper. To distinguish different dissipations, we study a non- From a numerical perspective, it is often a challenge equilibrium reaction-diffusion system with boundary ef- to construct a numerical scheme that preserves the posi- fects. This system can be viewed as an abstract represen- tivity and conservation of elements for reaction-diffusion tation of a building block of complex biological networks, systems [79, 80]. The energetic variational formulation in which a enzyme that localizes a particular chemical presented in this paper opens a new door to design a reaction and moves reactants into products. Our simula- positive, energy-stable numerical schemes to reaction- tion results show that the input-output of such a system diffusion type equations. depends on the choice of the dissipation. If the experimental system is reasonably reproducible, the dissipation -19.2 mechanism can be obtained by experimental measure- Dissipation D1 ments and studying an inverse problem. Dissipation D2 The energetic variational form proposed here also opens a new door to design a positiveness preserving and -19.4 energy stable numerical schemes for reaction-diffusion type equations. For instance, such an energetic variational form will enable us to design Lagrangian-Eulerian schemes for reaction-diffusion systems by applying some -19.6 recently developed methods for general diffusions [73, 76– 78]. -19.8

0 0.02 0.04 0.06 0.08 0.1 ACKNOWLEDGE

FIG. 4: Change of the discrete energy for two dissipations with −4 The authors acknowledge the partial support of NSF respect to time in numerical simulations (τ = 10 ). (Grant DMS-1759536). We thank Prof. Hong Qian, Prof. Huaxiong Huang and Dr. Shixin Xu for suggestions and Here we only proposed a numerical scheme for the one- helpful discussions. dimensional reaction-diﬀusion system considered in Sec. 9

III. Our numerical discretization is based a discrete en- By employing a discrete energetic variational approach, ergetic variational approach [73, 81], which follows the we get strategy of “discretize-then-variation”. More specifically,  0 we can adopt a finite difference scheme on a staggered  Γ(Rj(t)) = (−(µA)j − (µB)j + (µC )j) grid for the spatial discretization of R and the accumu- 0 (µi)k+1 − (µi)k (A2) t R  (Ji)k+1/2 = , lated fluxes Ji = 0 ciuidt (i = A, B, C). Assume [0, l] is h the compuatational domain, let Xj = jh (j = 0,...,N) where be the equidistant grid point and Xj+1/2 = (j + 1/2)h (j = 0,...,N − 1) be the corresponding half-integer grid n+1 n+1 ∞ point, where h = k/N. (µi)j = ln(ci)j − ln(ci )j, (A3) Let EN and CN be the spaces of functions defined on {X | j = 0,...,N} and {X | j = 0,...,N − 1}, j = 0,...N and k = 0,...N − 1. The fully discrete j j+1/2 scheme can be obtained by applying the implicit Euler respectively, We can approximate R and ci in EN and 0 discretization to (A2), that is approximate Ji in CN . Then the kinematic ci = ci + σ R + ∂ J becomes i x  Rn+1 − Rn  j j P3 n+1  Γ( ) = − i=1 σi(µi)j , (Ji)j+1/2 − (Ji)j−1/2  τ (c ) (t) = (c0) + σ R (t) + , (A1)  i j i j i j h (A4)  (J )n+1 − (J )n n+1 n+1  i k+1/2 i k+1/2 (µi)k+1 − (µi)k where i = A, B, C. Inserting (A1) into (43), we get the  = / τ h discrete energy in terms of Rj and Jj+1/2. On the meantime, for the dissipation (48) and (49), the discrete dis- As a numerical test, we compute our system with c0 = sipation functional can be written as 0.25 for dissipations (48) and (49). The computed dis-

N crete free energy as a function of time is showed in Fig. X D = Γ(R0 (t))R0 (t)) 4. The simulation result indicates that our numerical h j j scheme is energy stable, although a careful numerical j=1 analysis is certainly needed. N−1 X 0 2 0 2 0 2 + |(JA)k+1/2| + |(JB)k+1/2| + |(JC )k+1/2| . k=1

[1] U. Alon, Science 301, 1866 (2003). ucation, 2015). [2] T. Okada and A. Mochizuki, Physical review letters 117, [18] J. P. Keener and J. Sneyd, Mathematical physiology, 048101 (2016). vol. 1 (Springer, 1998). [3] J. Strutt, Proceedings of the London Mathematical So- [19] P. Waage and C. M. Gulberg, Journal of chemical edu- ciety 1, 357 (1871). cation 63, 1044 (1986). [4] L. Onsager, Physical review 37, 405 (1931). [20] J. Wei, The Journal of Chemical Physics 36, 1578 (1962). [5] L. Onsager, Physical review 38, 2265 (1931). [21] D. Shear, Journal of theoretical biology 16, 212 (1967). [6] C. Liu, in Multi-Scale Phenomena in Complex Fluids: [22] N. Z. Shapiro and L. S. Shapley, Journal of the Society Modeling, Analysis and Numerical Simulation (World for Industrial and Applied Mathematics 13, 353 (1965). Scientiﬁc, 2009), pp. 286–337. [23] R. Aris, Archive for rational mechanics and analysis 19, [7] M.-H. Giga, A. Kirshtein, and C. Liu, Handbook of math- 81 (1965). ematical analysis in mechanics of viscous ﬂuids pp. 1–41 [24] P. H. Sellers, SIAM Journal on Applied Mathematics 15, (2017). 637 (1967). [8] M. Grmela and H. C. Öttinger, Physical Review E 56, [25] R. Aris, Archive for Rational Mechanics and Analysis 27, 6620 (1997). 356 (1968). [9] H. C. Öttinger and M. Grmela, Physical Review E 56, [26] F. Horn and R. Jackson, Archive for rational mechanics 6633 (1997). and analysis 47, 81 (1972). [10] M. Grmela, Journal of Physics Communications 2, [27] G. F. Oster and A. S. Perelson, Archive for rational me- 032001 (2018). chanics and analysis 55, 230 (1974). [11] M. Pavelka, V. Klika, and M. Grmela, Multiscale [28] A. S. Perelson and G. F. Oster, Archive for Rational thermo-dynamics: introduction to GENERIC (Walter de Mechanics and Analysis 57, 31 (1974). Gruyter GmbH & Co KG, 2018). [29] T. G. Kurtz, The Journal of Chemical Physics 57, 2976 [12] M. Doi, Journal of Physics: Condensed Matter 23, (1972). 284118 (2011). [30] H. Othmer, Chemical Engineering Science 31, 993 [13] M. Doi, Chin. Phys. B 24, 1674 (2015). (1976). [14] W.-A. Yong, Physical Review E 86, 067101 (2012). [31] B. L. Clarke, Stability of Complex Reaction Networks [15] L. Peng, Y. Hu, and L. Hong, The European Physical (John Wiley & Sons, Ltd, 2007), pp. 1–215. Journal E 42, 73 (2019). [32] M. A. Biot, Quarterly of Applied Mathematics 39, 517 [16] H. Ge and H. Qian, Physical Review E 94, 052150 (2016). (1982). [17] R. Chang and K. Goldsby, Chemistry (McGraw-Hill Ed- [33] C. Truesdell, Rational thermodynamics (Springer, 1984). 10

[34] P. Réti and L. Ropolyi, Reaction Kinetics and Catalysis ical Physics 133, 104104 (2010). Letters 25, 109 (1984). [58] V. I. Arnol’d, Mathematical methods of classical mechan- [35] A. N. Beris, B. J. Edwards, B. J. Edwards, et al., ics, vol. 60 (Springer Science & Business Media, 2013). Thermodynamics of flowing systems: with internal mi- [59] S. R. de Groot and P. Mazur, Non-equilibrium Thermo- crostructure, 36 (Oxford University Press on Demand, dynamics (Courier Corporation, 1984). 1994). [60] Z. Schuss, Siam Review 22, 119 (1980). [36] A. N. Gorban, I. V. Karlin, and A. Y. Zinovyev, Physics [61] P. Van Rysselberghe, The Journal of Chemical Physics Reports 396, 197 (2004). 36, 1329 (1962). [37] A. Mielke, Nonlinearity 24, 1329 (2011). [62] M. Feinberg, Archive for rational mechanics and analysis [38] M. Grmela, Physica D: Nonlinear Phenomena 241, 976 49, 187 (1972). (2012). [63] J. Bataille, D. Edelen, and J. Kestin, Journal of Non- [39] J. Keizer, Statistical thermodynamics of nonequilibrium Equilibrium Thermodynamics 3, 153 (1978). processes (Springer Science & Business Media, 2012). [64] A. Mielke, D. M. Renger, and M. A. Peletier, Journal of [40] M. Liero and A. Mielke, Philosophical Transactions of Non-Equilibrium Thermodynamics 41, 141 (2016). the Royal Society A: Mathematical, Physical and Engi- [65] L. Desvillettes and K. Fellner, Journal of mathematical neering Sciences 371, 20120346 (2013). analysis and applications 319, 157 (2006). [41] A. Mielke, Discr. Cont. Dynam. Systems Ser. S 6, 479 [66] B. Perthame, in Parabolic Equations in Biology (2013). (Springer, 2015), pp. 1–21. [42] A. van der Schaft, S. Rao, and B. Jayawardhana, SIAM [67] G. Lebon, D. Jou, and J. Casas-Vázquez, Understand- Journal on Applied Mathematics 73, 953 (2013). ing non-equilibrium thermodynamics, vol. 295 (Springer, [43] V. Klika and M. Grmela, Physical Review E 87, 012141 2008). (2013). [68] T. De Donder, Mémoires de la Classe des sciences. [44] A. Gorban and V. Kolokoltsov, Mathematical Modelling Académie royale de Belgique. Collection in 8 9, 1 (1927). of Natural Phenomena 10, 16 (2015). [69] T. De Donder, Thermodynamic theory of affinity, vol. 1 [45] D. F. Anderson and T. G. Kurtz, Stochastic analysis of (Stanford university press, 1936). biochemical systems, vol. 1 (Springer, 2015). [70] F. Jülicher, A. Ajdari, and J. Prost, Reviews of Modern [46] D. F. Anderson, G. Craciun, M. Gopalkrishnan, and Physics 69, 1269 (1997). C. Wiuf, Bulletin of mathematical biology 77, 1744 [71] S. Fornari, A. Schäfer, M. Jucker, A. Goriely, and (2015). E. Kuhl, Journal of the Royal Society Interface 16, [47] R. Rao and M. Esposito, Physical Review X 6, 041064 20190356 (2019). (2016). [72] A. Hawkins-Daarud, K. G. van der Zee, and J. Tins- [48] H. Qian, S. Kjelstrup, A. B. Kolomeisky, and D. Bedeaux, ley Oden, International journal for numerical methods in Journal of Physics: Condensed Matter 28, 153004 (2016). biomedical engineering 28, 3 (2012). [49] H. Ge and H. Qian, Journal of Statistical Physics 166, [73] C. Liu and Y. Wang, Journal of Computational Physics 190 (2017). p. 109566 (2020). [50] A. Mielke, R. I. Patterson, M. A. Peletier, and [74] R. Eisenberg, Journal of Membrane Biology 115, 1 D. Michiel Renger, SIAM Journal on Applied Mathemat- (1990). ics 77, 1562 (2017). [75] M. Burger, R. S. Eisenberg, and H. W. Engl, SIAM Jour- [51] J. Haskovec, S. Hittmeir, P. Markowich, and A. Mielke, nal on Applied Mathematics 67, 960 (2007). SIAM Journal on Mathematical Analysis 50, 1037 [76] O. Junge, D. Matthes, and H. Osberger, SIAM Journal (2018). on Numerical Analysis 55, 419 (2017). [52] M. Feinberg, Foundations of Chemical Reaction Network [77] J. A. Carrillo, B. Düring, D. Matthes, and D. S. Theory, vol. 202 (Springer, 2019). McCormick, Journal of Scientific Computing 75, 1463 [53] Z. Fang and C. Gao, SIAM Journal on Applied Dynam- (2018). ical Systems 18, 1163 (2019). [78] J. A. Carrillo, K. Craig, and F. S. Patacchini, Calculus [54] A. N. Gorban, arXiv preprint arXiv:1902.05351 (2019). of Variations and Partial Differential Equations 58, 53 [55] M. A. Biot, Journal of the Mechanics and Physics of (2019). Solids 25, 289 (1977). [79] A. Sandu, Journal of Computational Physics 170, 589 [56] D. Kondepudi and I. Prigogine, Modern thermodynamics: (2001). from heat engines to dissipative structures (John Wiley [80] L. Formaggia and A. Scotti, SIAM Journal on Numerical & Sons, 2014). Analysis 49, 1267 (2011). [57] B. Eisenberg, Y. Hyon, and C. Liu, The Journal of Chem- [81] C. Liu and Y. Wang, arXiv preprint arXiv:2003.10413 (2020).