---
title: Non-Isothermal Maxwell-Stefan Equations
url: https://www.emergentmind.com/topics/non-isothermal-maxwell-stefan-equations
type: topic
---

# Non-Isothermal Maxwell-Stefan Equations

Non-isothermal Maxwell–Stefan equations are multicomponent diffusion systems in which the species balances are coupled to a temperature or energy balance, so that temperature is an unknown of the transport problem rather than a fixed parameter. In contrast with the classical isothermal theory, the diffusive driving forces involve temperature-weighted composition gradients such as \(\nabla(c_i T)\), \(\nabla(\rho_i\theta)\), or \(\nabla(n_i\theta)\), and the thermal equation may contain Fourier conduction, enthalpy transport by species diffusion, or purely advective transport, depending on the constitutive setting. The subject connects kinetic theory, continuum thermodynamics, and cross-diffusion PDE analysis: formal and rigorous derivations from Boltzmann and BGK models are available, as are local and global well-posedness results for several non-isothermal variants [1712.06194], [2303.17693], [2508.03311].

## 1. Defining structure and principal formulations

Across the literature, the unknowns are species concentrations, mass densities, or number densities—typically denoted by \(c_i\), \(\rho_i\), or \(n_i\)—together with diffusive fluxes \(J_i\) or species velocities \(u_i\), and a common temperature \(T\) or \(\theta\). The common structural core is a family of species continuity equations,
\[
\partial_t c_i+\nabla\cdot J_i=0,
\qquad
\partial_t \rho_i+\operatorname{div}J_i=0,
\qquad\text{or}\qquad
\partial_t n_i+\operatorname{div}(n_i v_i)=0,
\]
closed by Maxwell–Stefan friction relations expressing balance between thermodynamic driving forces and pairwise interspecies drag. In non-isothermal models, those driving forces are no longer purely compositional.

A representative algebraic closure is
\[
-\sum_{j=1}^n b_{ij}\rho_i\rho_j (u_i-u_j)=d_i,
\qquad
\sum_{i=1}^n \rho_i u_i=0,
\]
with \(d_i=\nabla(\rho_i\theta)\) after thermodynamic specialization and with a separate energy balance
\[
\partial_t(\rho e)+\operatorname{div}J_e=0,
\qquad
J_e=-K(\theta)\nabla\theta+\sum_{j=1}^n(\rho_j e_j+p_j)u_j,
\]
in the vanishing-barycentric-velocity formulation [2303.17693]. In other derivations, the macroscopic flux law is written directly in Maxwell–Stefan form,
\[
\nabla_x(c_i T)=-\sum_{j\neq i} k_{ij}(c_jJ_i-c_iJ_j),
\]
or, at the number-density level,
\[
\nabla_x(n_i\theta)=-\sum_{j=1}^N D_{ij}n_in_j(v_i-v_j),
\]
again coupled to an energy equation [1712.06196], [2509.14046].

| Formulation | Diffusive driving force | Thermal equation or flux |
|---|---|---|
| Monatomic Boltzmann limit | \(\nabla_x(c_iT)\) | \(\partial_t(c_{\mathrm{tot}}T)+\frac32\nabla_x\cdot\!\left(T\sum_i J_i\right)=0\) [1712.06194] |
| Zero-barycentric-velocity model | \(d_i=\nabla(\rho_i\theta)\) | \(J_e=-K(\theta)\nabla\theta+\sum_j(\rho_j e_j+p_j)u_j\) [2303.17693] |
| BGK diffusion limit | \(\nabla_x(n_i\theta)\) | convective plus diffusive heat flux with \(-\nabla_x(n_i\theta^2)/(m_i\nu_i)\) [2509.14046] |
| Maxwell–Stefan–Fourier simplification | \(d_i=\nabla(\rho_i\theta)\) | \(J_e=-\kappa(\theta)\nabla\theta\) after normalization [2606.13235] |

This variety is substantive rather than notational. The literature contains several structurally distinct non-isothermal Maxwell–Stefan systems, mainly differing in the closure imposed on the total flux, the pressure or barycentric constraint, and the constitution of the heat flux.

## 2. Kinetic origins and asymptotic derivations

A central line of development derives non-isothermal Maxwell–Stefan equations from multispecies kinetic models under diffusive scaling. For monatomic gaseous mixtures, one starts from a system of Boltzmann equations
\[
\varepsilon \partial_t f_i + v\cdot \nabla_x f_i
=
\frac1\varepsilon \sum_{j=1}^n Q_{ij}(f_i,f_j),
\]
assumes that the solution remains close to a specieswise local Maxwellian with small velocities of order \(O(\varepsilon)\), and passes formally to the limit \(\varepsilon\to0^+\). In the derivation for Maxwellian molecules, pairwise elastic collisions conserve mass, total momentum, and kinetic energy, and the asymptotic analysis yields both species balances and a temperature equation. A key consequence is the synchronization of species temperatures:
\[
T_i(t,x)-T_j(t,x)=O(\varepsilon^2),
\]
so that all species converge to a common limit temperature \(T\) [1712.06194].

A related derivation replaces the local-Maxwellian ansatz by a maximum entropy principle. In that setting, the scaled Boltzmann system for a mixture of monatomic, inert, ideal gases is combined with constrained entropy maximization, yielding an exact scaled Maxwellian
\[
f_i^\alpha(t,x,v)
=
\left(\frac{5}{3m_0}\right)^{3/2}
\rho_i^\alpha
\left(\frac{m_i}{2\pi T}\right)^{3/2}
\exp\!\left(
-\frac{5}{3m_0}\frac{m_i|v-\alpha u_i^\alpha|^2}{2T}
\right).
\]
The resulting momentum balance contains the non-isothermal driving term \(\nabla(\rho_i^\alpha T)\), and the diffusion limit produces the system
\[
\partial_t \rho_i+\nabla\cdot(\rho_i u_i)=0,
\qquad
\frac{3m_0}{5m_i}\nabla(\rho_iT)
=
2\pi\sum_{j\ne i}\|b_{ij}\|_{L^1}\frac{m_j}{m_i+m_j}\rho_i\rho_j(u_j-u_i),
\]
together with
\[
\frac32\partial_t\sum_{i=1}^S (\rho_iT)
+\frac52\nabla\cdot\sum_{i=1}^S (\rho_i u_i T)=0.
\]
The same framework yields an explicit kinetic entropy production quadratic in the relative velocities \(|u_j^\alpha-u_i^\alpha|^2\) [2110.11170].

The first rigorous Boltzmann-to-non-isothermal-Maxwell–Stefan asymptotic result appears in the small-data setting of the multi-species Boltzmann equations under diffusive scaling,
\[
\partial_t F_i^\varepsilon+\frac1\varepsilon v\cdot\nabla_x F_i^\varepsilon
=
\frac1{\varepsilon^2}Q_i(\mathbf F^\varepsilon,\mathbf F^\varepsilon).
\]
There, the Maxwellian built from the non-isothermal Maxwell–Stefan solution is not an exact local equilibrium for the mixture collision operator, because cross-species interactions produce a nonzero defect \(Q(\mathbf M^\varepsilon,\mathbf M^\varepsilon)\neq0\). The analysis therefore combines perturbation around global equilibrium, a local coercivity estimate for the operator linearized around the non-equilibrium local Maxwellian, and uniform-in-\(\varepsilon\) hypocoercive estimates [2508.03311].

Formal BGK diffusion limits yield further variants. For the multispecies Gross–Krook model, the leading-order temperatures again synchronize,
\[
\theta_i^\varepsilon-\theta_j^\varepsilon = O(\varepsilon^2),
\]
and the limit system contains the Maxwell–Stefan-type balance
\[
\nabla_x(n_i\theta)=-\sum_{j=1}^N D_{ij}n_in_j(v_i-v_j)
\]
together with an energy equation whose flux contains both a convective term with \(\bar v_i\) and a diffusive term
\[
-\frac{\nabla_x(n_i\theta^2)}{m_i\nu_i}.
\]
This differs from non-isothermal derivations whose energy balance has only convective transport by the species velocities [2509.14046].

## 3. Closure relations, pressure constraints, and reduced temperature dynamics

Non-isothermal Maxwell–Stefan systems are typically underdetermined before an additional closure is imposed. In the monatomic formal Boltzmann limit, the limiting variables are \((c_i,J_i,T)\), giving \(2n+1\) unknowns but only \(2n\) independent equations. In the isothermal theory, one usually imposes \(\sum_i J_i=0\), but in the non-isothermal derivation based on \(\nabla(c_iT)\) the authors propose instead
\[
\sum_{i=1}^n J_i=-\alpha \nabla_x c_{\mathrm{tot}},
\qquad
\alpha>0,
\]
calling it the decoupling closure relation. When \(T\) is constant, this reduces to the classical isothermal constraint \(\sum_i J_i=0\) [1712.06194].

Under that closure, the total concentration solves
\[
\partial_t c_{\mathrm{tot}}-\alpha \Delta c_{\mathrm{tot}}=0,
\]
while the temperature obeys an advection-type equation. In the PDE analysis of the associated cross-diffusion system, the temperature equation is written as
\[
\partial_t T
-\left(\frac23\partial_t \log c_{\mathrm{tot}}\right)T
-\left(\frac{5\alpha}{3}\nabla\log c_{\mathrm{tot}}\right)\cdot \nabla T
=0,
\]
and the temperature is represented along characteristics of the velocity field
\[
\mathcal V(t,x)=-\frac{5\alpha}{3}\nabla\log c_{\mathrm{tot}}.
\]
This decoupling is the technical basis for the reduction of the original Maxwell–Stefan system to a quasi-linear parabolic system for \(n-1\) concentrations [1712.06196].

A different closure appears in thermodynamically consistent zero-barycentric-velocity models. There, the mixture velocity is fixed to \(v=0\), the compatibility constraint is
\[
\sum_{i=1}^n \rho_i u_i=0,
\]
and thermodynamic consistency requires the pressure constraint
\[
\nabla p=0.
\]
Because the barycentric velocity vanishes, the total density is then time-independent:
\[
\partial_t p=0.
\]
The algebraic Maxwell–Stefan system is singular, and inversion is performed on the subspace orthogonal to the density vector rather than by an additional Fick-type law for the total density [2303.17693].

In the rigorous Boltzmann asymptotics for small perturbations, the closure
\[
\sum_{i=1}^N c_i u_i=-\alpha \nabla_x c_{\mathrm{tot}}
\]
again produces a heat equation for \(c_{\mathrm{tot}}\),
\[
\partial_t c_{\mathrm{tot}}=\alpha \Delta_x c_{\mathrm{tot}},
\]
while the relation \(\nabla_x(c_{\mathrm{tot}}T)=0\) allows the temperature equation to be rewritten as
\[
\partial_t T+\frac{\alpha}{3}\frac{|\nabla_x T|^2}{T}+\frac{2\alpha}{3}\Delta_x T=0.
\]
Different closures therefore lead to materially different thermal equations: transport-type in some settings, nonlinear parabolic in others [2508.03311].

## 4. Thermodynamic structure, entropy variables, and dissipation

The mathematically most developed non-isothermal Maxwell–Stefan theories are built from a free-energy closure. In the vanishing-barycentric-velocity model, the Helmholtz free energy density of species \(i\) is
\[
\psi_i(\rho_i,\theta)
=
\frac{\rho_i}{m_i}\left(\log\frac{\rho_i}{m_i}-1\right)
-
C_w\rho_i\theta(\log\theta-1),
\]
from which one obtains
\[
\rho_i e_i=C_w\rho_i\theta,\qquad
\mu_i=\frac1{m_i}\log\frac{\rho_i}{m_i}-C_w\theta\log\theta,\qquad
p_i=\frac{\rho_i\theta}{m_i}.
\]
With these constitutive laws the driving force simplifies to \(d_i=\nabla(\rho_i\theta)\), and the heat flux becomes
\[
J_e=-K(\theta)\nabla\theta+\theta\sum_{i=1}^n \frac{C_wm_i+1}{m_i}\rho_i u_i
\]
[2303.17693].

The singular Maxwell–Stefan algebraic system is then rewritten as \(M(\rho)v=d\), where the matrix \(M(\rho)\) has kernel \(\operatorname{span}\{\rho\}\). The Bott–Duffin inverse \(M^{\mathrm{BD}}\) exists and is symmetric positive definite on the constraint subspace
\[
L=\{y\in\mathbb R^n:\rho\cdot y=0\},
\]
which yields \(v=M^{\mathrm{BD}}d\). This inversion underlies the entropy-variable formulation.

The mathematical entropy density is
\[
h(\rho,\theta)
=
\sum_{i=1}^n \frac{\rho_i}{m_i}\left(\log\frac{\rho_i}{m_i}-1\right)
-
C_w\rho\log\theta,
\]
and the entropy variables are
\[
w_i
=
\frac1{m_i}\log\frac{\rho_i}{m_i}
-
\frac1{m_n}\log\frac{\rho_n}{m_n},
\qquad
w=\log\theta.
\]
In these variables the fluxes can be written as
\[
J_i=-\sum_{j=1}^{n-1}A_{ij}(\rho,\theta)\nabla w_j-B_i(\rho,\theta)\nabla w,
\qquad
J_e=-\sum_{j=1}^{n-1}B_j(\rho,\theta)\nabla w_j-a(\rho,\theta)\nabla w,
\]
with full Onsager matrix
\[
Q(\rho,\theta)=
\begin{pmatrix}
A & B\\
B^\top & a
\end{pmatrix}
\]
positive semidefinite. This positive semidefiniteness is the structural reason the PDE behaves as a nonlinear parabolic system in entropy variables [2303.17693].

Entropy production is also explicit in kinetic derivations. For the maximum-entropy Boltzmann closure, the total entropy production has the form
\[
D^\alpha
=
\alpha^2 \frac{5\pi}{3Tm_0}
\sum_{i=1}^S\sum_{j\ne i}
\|b_{ij}\|_{L^1}
\frac{m_i m_j}{m_i+m_j}
\rho_i^\alpha\rho_j^\alpha
|u_j^\alpha-u_i^\alpha|^2,
\]
which is nonnegative and identifies Maxwell–Stefan friction as a kinetic dissipation mechanism [2110.11170]. In the BGK diffusion limit, under the normalization \(m_i\nu_i=1\), the macroscopic entropy production splits into a diffusion/friction term involving \(D_{ij}n_in_j|v_i-v_j|^2\) and a thermal term involving \((n_i/\theta)|\nabla_x\theta|^2\) [2509.14046].

## 5. Analytical results and well-posedness theory

The first PDE well-posedness results for a non-isothermal Maxwell–Stefan system were local in time. Starting from the system
\[
\partial_t c_i+\nabla\cdot J_i=0,
\qquad
\nabla(c_iT)=-\sum_{j\neq i}k_{ij}(c_jJ_i-c_iJ_j),
\qquad
\sum_{i=1}^n J_i=-\alpha \nabla c_{\rm tot},
\]
one eliminates one species, rewrites the flux law as \(\mathbf D=\mathbf F\mathbf J\), analyzes the singular Maxwell–Stefan matrix by Perron–Frobenius arguments, and reduces the problem to a quasi-linear parabolic system
\[
\partial_t \mathbf c' - \operatorname{div}\!\left(T\,\mathbf B \nabla \mathbf c'\right) = \mathbf r(\mathbf c').
\]
Because the reduced diffusion matrix has spectrum bounded away from zero and \(T\) remains positive, the system is normally elliptic, and \(L^p\)-maximal regularity yields local-in-time existence and uniqueness of strong solutions [1712.06196].

Global weak existence and weak-strong uniqueness have been proved for a different non-isothermal Maxwell–Stefan system with heat conduction and vanishing barycentric velocity. Under \(b_{ij}=b_{ji}>0\), positive initial data, fixed total density bounded away from vacuum, and conductivity growth
\[
c_K(1+\theta^2)\le K(\theta)\le C_K(1+\theta^2),
\]
there exists a global weak solution with \(\rho_i>0\) and \(\theta>0\) almost everywhere. The proof uses the boundedness-by-entropy method: time discretization by implicit Euler, higher-order regularization, a discrete entropy inequality, and compactness arguments based on Aubin–Lions. If a strong solution exists with bounded positive densities and bounded temperature, then any weak solution with the same initial data coincides with it; the proof proceeds through a relative entropy inequality and Grönwall’s lemma [2303.17693].

For the small-perturbation closure-based system arising in the rigorous Boltzmann asymptotics, global classical solutions are available in Sobolev spaces for \(s>3\), with positivity preserved for all time. The proof combines coercivity of the Maxwell–Stefan matrix on \(\operatorname{Span}(\mathbf 1)^\perp\), the heat equation for \(c_{\mathrm{tot}}\), the relation \(\nabla_x(c_{\mathrm{tot}}T)=0\), Sobolev product estimates, and a nonlinear iteration scheme. On the kinetic side, the same work establishes a unique global classical solution to the perturbed Boltzmann equation, uniform in Knudsen number \(\varepsilon\), thereby rigorously justifying the non-isothermal Maxwell–Stefan asymptotic limit [2508.03311].

A further analytical direction treats degenerate heat conduction in a Maxwell–Stefan–Fourier system with
\[
\kappa(\theta)=\theta^\alpha,
\qquad
0<\alpha\le 2.
\]
In that case the entropy inequality controls only \(\nabla\theta^{\alpha/2}\), not \(\nabla\theta\) or \(\nabla\log\theta\), so the identification of the heat flux cannot rely on standard entropy methods. The main new tool is a renormalized energy estimate that yields
\[
\theta\in L^\infty(0,T;L^{\alpha+2}(\Omega)),
\qquad
\nabla\theta^{\alpha+1}\in L^2(\Omega\times(0,T)),
\]
which provides compactness of the temperature and permits passage to the limit in
\[
J_e=-\theta^\alpha \nabla\theta.
\]
This gives global weak solutions and, in addition, an entropy production defect measure accounting for possible loss of dissipation in the degenerate limit [2606.13235].

## 6. Extensions, variants, and relation to the isothermal theory

Non-isothermal Maxwell–Stefan equations admit several physically distinct extensions. For reactive mixtures of polyatomic gases with continuous internal energy, the diffusive limit of a reactive Boltzmann system yields
\[
\partial_t n_i+\nabla_x\cdot J_i=-\nu_i A,
\]
together with the non-isothermal Maxwell–Stefan law
\[
\nabla_x n_i-\frac{n_i}{T}\nabla_x T
=
\sum_{j\neq i} D_{ij}(n_jJ_i-n_iJ_j),
\]
and a temperature equation containing both transport and the reaction source \(-EA\). Here temperature enters the flux law explicitly through the term \(-\frac{n_i}{T}\nabla_x T\), and the coefficients \(D_{ij}\) as well as the chemical production term \(A\) are derived from the collision kernels [1906.11766].

A broader thermodynamic generalization arises from a GENERIC/Hamiltonian framework for binary mixtures with multiple partial velocities and partial temperatures. After reduction to a single-temperature, single-momentum description, the generalized diffusion-velocity law contains chemical-potential gradients, temperature gradients, and electric forcing; after reduction to mechanical equilibrium it becomes a non-isothermal variant of the dusty gas model. In this framework, temperature differences between constituents are identified as a possible new source of the Soret coefficient [2206.14930].

The relation to the isothermal Maxwell–Stefan theory is precise in many models. Constant temperature reduces \(\nabla(c_iT)\), \(\nabla(\rho_i\theta)\), or \(\nabla(n_i\theta)\) to the classical composition-driven forces, and the closure \(\sum_i J_i=-\alpha\nabla c_{\mathrm{tot}}\) collapses to \(\sum_i J_i=0\) when \(T\) is constant [1712.06194]. The isothermal theory supplies important operator-theoretic and entropy-based baselines: normal ellipticity from Perron–Frobenius analysis for the reduced diffusion operator [1007.1775], and entropy-variable symmetrization for singular Maxwell–Stefan matrices coupled to incompressible flow [1310.3376].

Two recurrent simplifications from the isothermal literature do not transfer automatically. First, the equimolar or zero-total-flux condition is not universal in non-isothermal settings: some models retain \(\sum_i J_i=0\) through the zero-barycentric-velocity constraint, whereas others replace it by a Fick-type law for the total concentration [2303.17693], [1712.06194]. Second, the specieswise local Maxwellian used in asymptotic derivations need not be a true local equilibrium of the full mixture collision operator; the rigorous Boltzmann analysis shows that cross-interactions obstruct that property and generate a nontrivial source term in the perturbation equation [2508.03311].

The modern theory therefore presents non-isothermal Maxwell–Stefan equations not as a single canonical PDE, but as a family of thermodynamically and kinetically motivated cross-diffusion systems. Their unifying feature is the replacement of purely isothermal diffusion by a coupled mass–heat dynamics in which temperature directly enters the driving forces, the entropy production, and the analytic structure of the equations.

Source: https://www.emergentmind.com/topics/non-isothermal-maxwell-stefan-equations