The Onsager Reciprocal Relations

A metal bar can do two surprising things at once. Lay a temperature difference across it and a voltage appears (the Seebeck effect); push a current through it and it carries heat from one end to the other (the Peltier effect). These look like independent material properties, yet in 1931 Lars Onsager proved they are governed by one number, not two. Measure how strongly heat drives charge, and you have already measured how strongly charge drives heat. The same economy links thermal diffusion to its inverse, streaming potentials to electroosmosis, and cross-couplings throughout linear irreversible thermodynamics. The statement is compact: choose fluxes $J_i$ (currents of heat, charge, particles) and conjugate forces $X_j$ (gradients of $1/T$ , of $-\mu/T$ , of potential) so that the entropy production is the bilinear form $\dot S=\sum_i J_iX_i$ . Then the matrix of kinetic coefficients in the linear laws $J_i=\sum_j L_{ij}X_j$ is symmetric: $L_{ij}=L_{ji}$ . What makes this deep is that the symmetry is not a thermodynamic identity but a fossil of microscopic reversibility. Run the molecular movie backward and the equilibrium fluctuation dynamics looks the same; that time-reversal invariance descends, through Onsager's regression hypothesis, into the equality of cross-coefficients you can measure with a voltmeter and a thermometer. This article proves $L_{ij}=L_{ji}$ from the time-reversal symmetry of equilibrium correlations, derives the Kelvin thermoelectric relation $\Pi=TS$ as a direct corollary, and shows how a magnetic field — odd under time reversal — generalizes the result to the Onsager–Casimir relation $L_{ij}(\mathbf B)=\epsilon_i\epsilon_j L_{ji}(-\mathbf B)$ with its Hall-resistivity signature. The reciprocal relations sit at the head of a lineage that runs through the fluctuation–dissipation theorem to the Green–Kubo formulas.

Key takeaways

Onsager reciprocity: for fluxes and forces chosen so that entropy production is $\dot S=\sum_i J_iX_i$ , the kinetic matrix in $J_i=\sum_j L_{ij}X_j$ is symmetric, $L_{ij}=L_{ji}$ .
The symmetry is dynamical, not thermodynamic: it follows from the time-reversal invariance of the microscopic equations, encoded in the symmetry of equilibrium fluctuation correlations $\langle a_i(t)a_j(0)\rangle=\langle a_i(0)a_j(t)\rangle$ .
The proof combines the regression hypothesis ( $\langle\dot a_i\rangle=\sum_j L_{ij}X_j$ ) with the Gaussian fluctuation identity $\langle a_iX_j\rangle=-k_B\delta_{ij}$ and correlation symmetry.
Consequence: the Kelvin (Thomson) thermoelectric relation $\Pi=TS$ between Peltier and Seebeck coefficients, which Kelvin guessed but could not justify.
In a magnetic field or rotating frame (time-reversal-odd), reciprocity becomes the Onsager–Casimir relation $L_{ij}(\mathbf B)=\epsilon_i\epsilon_j L_{ji}(-\mathbf B)$ ; the Hall resistivity is odd in $\mathbf B$ .
Positivity of entropy production forces $L_{ii}\ge0$ and $L_{ii}L_{jj}\ge L_{ij}^2$ , and the Green–Kubo integrand's symmetry gives reciprocity a second, microscopic proof.
Key idea: choose entropy-conjugate fluxes and forces; then microscopic time-reversibility of the equilibrium fluctuation dynamics forces the kinetic matrix to be symmetric, $L_{ij}=L_{ji}$ — halving the independent transport coefficients and relating a priori unconnected effects.

The core insight

Choose fluxes and forces so that entropy production is the bilinear form $\dot S=\sum_i J_iX_i$ . Then the microscopic time-reversibility of equilibrium fluctuations, $\langle a_i(t)a_j(0)\rangle=\langle a_i(0)a_j(t)\rangle$ , together with the regression hypothesis that fluctuations decay by the macroscopic laws, forces the kinetic matrix to be symmetric: $L_{ij}=L_{ji}$ .

Where this lives today

The reciprocal relations are the theoretical backbone of thermoelectric materials science. The figure of merit $ZT=S^2\sigma T/\kappa$ that governs the efficiency of solid-state coolers and waste-heat generators is built entirely from the Onsager coefficients, and the Kelvin relation $\Pi=TS$ eliminates one free parameter in every device model, from Peltier chip design to radioisotope thermoelectric generators on deep-space probes. The positivity bound $L_{eq}\le\sqrt{L_{ee}L_{qq}}$ is exactly why raising $ZT$ requires decoupling the electronic and thermal conductivities — the search that drives phonon-glass electron-crystal materials. In condensed-matter transport, the Onsager–Casimir relations underpin the whole phenomenology of magnetotransport. The evenness of longitudinal magnetoresistance and the oddness of the Hall resistivity in the applied field are used as consistency checks on measurements, and reciprocity relations for nonlocal resistances in mesoscopic conductors (the Büttiker relations for multi-terminal devices) are a direct descendant. In molecular simulation, the reciprocal relations validate Green–Kubo calculations of coupled transport coefficients, and in the emerging field of spin caloritronics — the spin Seebeck and spin Peltier effects — Onsager reciprocity again links each new cross-effect to its inverse.

The physical picture

Imagine a bar of metal carrying both a temperature gradient and a voltage. A temperature difference drives an electric current (Seebeck); a current carries heat (Peltier). Onsager's claim is that these two cross-couplings share a single coefficient — measure one and you know the other. The reason lives underneath the macroscopic laws: the same carriers, obeying the same collision dynamics, mediate "heat-driven charge" and "charge-driven heat," and one symmetric matrix ties them together. The deepest picture is statistical. In equilibrium a quantity $a_i$ that fluctuates spontaneously to a nonzero value decays back, on average, exactly as a macroscopic disturbance of the same size would — this is Onsager's regression hypothesis, the subject of its own article. Now run the molecular movie backward: for variables even under time reversal, the equilibrium dynamics is unchanged. So the correlation between "how variable $i$ now relates to $j$ a moment later" must equal "how $j$ now relates to $i$ a moment later." That time-symmetry of the correlations, $\langle a_i(t)a_j(0)\rangle=\langle a_j(t)a_i(0)\rangle$ , is literally the source of $L_{ij}=L_{ji}$ . A mechanical analogy helps: a symmetric linear-response matrix is like the symmetry of the elastic compliance tensor, or of a network's mutual capacitances and inductances (Maxwell's reciprocity). Onsager's theorem is the dissipative, dynamical generalization of those static reciprocity theorems — the first exact result of non-equilibrium statistical mechanics, and the one Onsager's 1968 Nobel citation singled out. There is a crucial catch that a magnetic field introduces. A charged particle's helical trajectory under $\mathbf B$ is not its own time-reverse unless you also flip $\mathbf B\to-\mathbf B$ , because the Lorentz force is odd in both velocity and field. So in a field, reciprocity relates a system to its field-reversed twin — the Onsager–Casimir refinement — and the antisymmetric (Hall) part of the transport tensor survives precisely because of it.

Figure

Punchline: microscopic reversibility locks the two cross-correlations exactly on top of each other

The core equations

Onsager's framework starts from the second law written locally: the entropy produced by irreversible processes is a rate, and that rate is non-negative. Choosing the right pairing of currents and driving gradients makes the entropy-production rate a simple bilinear form.

\dot{S}=\sum_i J_i\,X_i\ \ge\ 0 \qquad (1)

where: $\dot S$ — entropy-production rate (per unit volume, or per system); $J_i$ — thermodynamic flux (heat, charge, particle current); $X_i$ — the conjugate thermodynamic force, e.g. $\nabla(1/T)$ or $-\nabla(\mu/T)$ . Only for this conjugate pairing does reciprocity hold — the requirement that $J_iX_i$ reproduce $\dot S$ is what fixes which quantity counts as a "force." Near equilibrium the forces are small, so the fluxes are analytic in them and may be linearized. The coefficients of that linearization are the kinetic (Onsager) coefficients that carry all the transport information.

J_i=\sum_j L_{ij}\,X_j \qquad (2)

where: $L_{ij}$ — matrix of Onsager coefficients. The diagonal entries are the familiar transport laws (Fourier's $\kappa$ , Fick's $D$ , Ohm's $\sigma$ ); the off-diagonal entries are the cross-effects (thermoelectric, thermodiffusive, electrokinetic). The reciprocal relations are the central claim: this matrix is symmetric.

L_{ij}=L_{ji} \qquad (3)

where: $L_{ij}$ — coefficient coupling flux $i$ to force $j$ . Symmetry holds for conjugate flux/force pairs of equal time-reversal parity; it halves the number of independent transport coefficients. The engine of the proof is a Gaussian fluctuation identity. Writing the entropy of a fluctuation as $\Delta S=-\tfrac12\sum_{ij}s_{ij}a_ia_j$ with conjugate forces $X_i=\partial\Delta S/\partial a_i=-\sum_j s_{ij}a_j$ , integration by parts in the fluctuation measure $P\propto e^{\Delta S/k_B}$ gives a clean orthogonality.

\langle a_i\,X_j\rangle=-k_B\,\delta_{ij} \qquad (4)

where: $k_B$ — Boltzmann constant; the identity says a fluctuating variable is "orthogonal" to every conjugate force but its own, with strength $k_B$ . This is the load-bearing algebraic fact of the whole derivation. Microscopic reversibility enters as the time-symmetry of equilibrium correlations. For variables even under time reversal, evolving $a_i$ forward for a time $t$ and correlating with $a_j$ gives the same answer as evolving $a_j$ .

\langle a_i(t)\,a_j(0)\rangle=\langle a_i(0)\,a_j(t)\rangle=\langle a_j(t)\,a_i(0)\rangle \qquad (5)

where: $\langle\cdot\rangle$ — equilibrium average; the middle equality is stationarity, the outer one time reversal. Together they make the two-time correlation symmetric under $i\leftrightarrow j$ . When some variables are odd under time reversal or a magnetic field is present, reversal must flip both, giving the Onsager–Casimir generalization and its Hall signature.

L_{ij}(\mathbf{B})=\epsilon_i\epsilon_j\,L_{ji}(-\mathbf{B}),\qquad \rho_{xy}(\mathbf B)=-\rho_{xy}(-\mathbf B) \qquad (6)

where: $\mathbf B$ — magnetic field (time-reversal odd); $\epsilon_i=\pm1$ — the time-reversal parity of variable $a_i$ . The diagonal coefficients stay even in $\mathbf B$ ; the Hall resistivity is odd — a directly measurable fingerprint of the Casimir refinement.

Derivation AOnsager reciprocity from the regression of equilibrium fluctuations

The goal is to prove $L_{ij}=L_{ji}$ from two ingredients only: the time-reversal symmetry of equilibrium fluctuation correlations and Onsager's regression hypothesis that a spontaneous fluctuation decays, on average, by the same linear laws as a macroscopic deviation. The strategy is to write the entropy of small fluctuations as a quadratic form, extract the Gaussian orthogonality identity $\langle a_iX_j\rangle=-k_B\delta_{ij}$ , differentiate the two-time correlation, and let the correlation symmetry collapse the two sides. Watch where the arrow of time enters — not in the linear laws, which are symmetric-looking, but in the single input that equilibrium correlations are even under $t\to-t$ .

Step 1. Write the entropy of a fluctuation $\mathbf a=(a_1,\dots,a_n)$ from equilibrium as $\Delta S=-\tfrac12\sum_{ij}s_{ij}a_ia_j$ with the symmetric, positive-definite Hessian $s_{ij}$ , so the fluctuation probability is $P\propto e^{\Delta S/k_B}$ (Einstein's formula). Why: near a maximum, entropy is quadratic; the Gaussian weight is what lets every moment be computed exactly.

Step 2. Define the conjugate forces $X_i=\partial\Delta S/\partial a_i=-\sum_j s_{ij}a_j$ , and integrate by parts in the Gaussian measure to get $\langle a_iX_j\rangle=-k_B\delta_{ij}$ . Why: this orthogonality is the algebraic heart of the proof — it turns the correlation of a variable with a force into a Kronecker delta, so the coefficient matrix falls straight out at the end.

Step 3. Invoke the regression hypothesis: the conditional average of a fluctuation decays by the macroscopic law $\tfrac{\mathrm d}{\mathrm dt}\langle a_i(t)\rangle=\sum_j L_{ij}\langle X_j(t)\rangle$ . Multiply by $a_k(0)$ and average over equilibrium, giving $\tfrac{\mathrm d}{\mathrm dt}\langle a_i(t)a_k(0)\rangle=\sum_j L_{ij}\langle X_j(t)a_k(0)\rangle$ . Why: the regression hypothesis is the physical bridge — it inserts the very coefficients $L_{ij}$ we want into the dynamics of an equilibrium correlation, an object that has its own symmetry.

Step 4. Evaluate at $t\to0^+$ : $\langle X_j(t)a_k(0)\rangle\to\langle X_ja_k\rangle=-k_B\delta_{jk}$ , so the right side is $-k_B L_{ik}$ . Now use the correlation symmetry $\langle a_i(t)a_k(0)\rangle=\langle a_k(t)a_i(0)\rangle$ (stationarity + time reversal): its short-time derivative is symmetric under $i\leftrightarrow k$ , forcing $-k_BL_{ik}=-k_BL_{ki}$ , hence $L_{ik}=L_{ki}$ . Why: the whole theorem is here — the time derivative of a symmetric object is symmetric, and the orthogonality identity converts that into the equality of cross-coefficients. *Verified: verify_onsager-reciprocal-relations.py builds a random symmetric positive-definite Hessian $s$ , forms the covariance $k_B s^{-1}$ , and confirms $\langle a_iX_j\rangle=-k_B\delta_{ij}$ to machine precision; it also checks that $L=L^\top$ is equivalent to the symmetry of the fluctuation-relaxation matrix and that an asymmetric $L$ breaks the correlation symmetry (reciprocity is nontrivial).*

Go deeper · Derivation B · 2 minThe Kelvin thermoelectric relation Pi equals T S from reciprocity

Derivation BThe Kelvin thermoelectric relation Pi equals T S from reciprocity

The goal is to show that the historically empirical Kelvin (Thomson) relation between the Peltier and Seebeck coefficients, $\Pi=TS$ , is forced by the single equality $L_{eq}=L_{qe}$ . The strategy is to write the coupled charge-and-heat linear laws with entropy-conjugate forces, extract the Seebeck coefficient as the open-circuit voltage per temperature gradient, extract the Peltier coefficient as the heat carried per unit charge at uniform temperature, and then impose reciprocity. The subtlety to respect is the choice of forces: using $\nabla(1/T)$ rather than $\nabla T$ is what makes the coefficient matrix the symmetric one, so the factors of $T$ land correctly.

Step 1. Write the coupled laws $J_e=L_{ee}X_e+L_{eq}X_q$ , $J_q=L_{qe}X_e+L_{qq}X_q$ , with $X_e=-\tfrac1T\nabla(\bar\mu/e)$ (electrochemical) and $X_q=\nabla(1/T)$ chosen so that $\dot S=J_eX_e+J_qX_q$ . Why: only these entropy-conjugate forces make $L$ symmetric; a different pairing would rescale the coefficients and hide the relation.

Step 2. The Seebeck coefficient is the voltage built up per unit temperature gradient at open circuit, $S=(\nabla(\bar\mu/e)/\nabla T)|_{J_e=0}$ ; solving $J_e=0$ expresses it through the ratio $L_{eq}/L_{ee}$ . Why: an open-circuit thermocouple develops exactly the voltage that stops the thermally driven current, so $S$ is set by how strongly heat couples to charge, $L_{eq}$ , relative to the pure electrical response $L_{ee}$ .

Step 3. The Peltier coefficient is the heat carried per unit charge at uniform temperature, $\Pi=(J_q/J_e)|_{\nabla T=0}=L_{qe}/L_{ee}$ . Why: at uniform $T$ the only driving force is electrical, so every unit of charge current drags along a heat current in proportion to the reverse cross-coefficient $L_{qe}$ .

Step 4. Impose reciprocity $L_{eq}=L_{qe}$ . Then the ratio of the two cross-effects collapses and $\Pi=TS$ — the relation Kelvin conjectured in 1854 on thermodynamic grounds he could not justify, now a corollary of microscopic reversibility. Why: reciprocity is precisely the statement that "heat drives charge" and "charge drives heat" share one coefficient, so the two coefficients built from them differ only by the factor of $T$ that the entropy-conjugate forces carry. *Verified: verify_onsager-reciprocal-relations.py takes a 2×2 thermoelectric $L$ obeying $L_{eq}=L_{qe}$ and positivity $L_{ee}L_{qq}\ge L_{eq}^2$ , computes $S=L_{eq}/(TL_{ee})$ and $\Pi=L_{qe}/L_{ee}$ , and confirms $\Pi=TS$ to machine precision for the worked numbers below.*

Go deeper · Derivation C · 1 minThe Onsager-Casimir relation in a magnetic field

Derivation CThe Onsager-Casimir relation in a magnetic field

The goal is to generalize reciprocity when some variables are odd under time reversal or a magnetic field is present, arriving at $L_{ij}(\mathbf B)=\epsilon_i\epsilon_j L_{ji}(-\mathbf B)$ and its Hall-resistivity corollary. The strategy is to track how time reversal acts on the correlation function once a field is switched on: reversing time flips odd-parity variables and reverses the field, so the correlation symmetry now relates a system to its field-reversed twin. The point to appreciate is that the antisymmetric (Hall) part of the transport tensor survives precisely because the field breaks ordinary reciprocity.

Step 1. Under time reversal $a_i\to\epsilon_ia_i$ and $\mathbf B\to-\mathbf B$ (the field couples to velocities and currents, which flip sign). So the correlation obeys $\langle a_i(t)a_j(0)\rangle_{\mathbf B}=\epsilon_i\epsilon_j\langle a_i(0)a_j(t)\rangle_{-\mathbf B}$ . Why: a charged particle's helical path is not its own time-reverse unless the field is also reversed — the field must ride along in the reversal.

Step 2. Repeat the regression-and-differentiation argument keeping the field as a parameter, so the short-time slope of the correlation gives $-k_BL_{ij}(\mathbf B)$ on one side and $-k_B\epsilon_i\epsilon_jL_{ji}(-\mathbf B)$ on the other. Equating yields $L_{ij}(\mathbf B)=\epsilon_i\epsilon_jL_{ji}(-\mathbf B)$ . Why: the machinery of the zero-field proof carries over verbatim; only the bookkeeping of parities and the field reversal is new.

Step 3. Read off the transport consequences. Diagonal coefficients are even in $\mathbf B$ , $L_{ii}(\mathbf B)=L_{ii}(-\mathbf B)$ ; the off-diagonal (Hall) part is antisymmetric in indices and odd in $\mathbf B$ , giving $\rho_{xy}(\mathbf B)=-\rho_{xy}(-\mathbf B)$ and $\sigma_{xy}=-\sigma_{yx}$ . Why: these parities are experimentally standard — the longitudinal magnetoresistance is even in the field, the Hall resistivity reverses sign with the field, and both follow from the Casimir relation. *Verified: verify_onsager-reciprocal-relations.py evaluates a Drude magnetoconductivity tensor at $\pm\mathbf B$ and confirms $\sigma_{xx}$ is even, $\sigma_{xy}$ is odd, and $\sigma_{xy}=-\sigma_{yx}$ — the Onsager–Casimir parity bookkeeping.*

Limiting cases

How reciprocity specializes and where it must be modified:

Case	Result	Meaning
Time-reversal-even variables	$L_{ij}=L_{ji}$	Standard Onsager symmetry when all $a_i$ are even under $t\to-t$ (densities of mass, charge, energy).
Magnetic field / rotation	$L_{ij}(\mathbf B,\boldsymbol\omega)=\epsilon_i\epsilon_j L_{ji}(-\mathbf B,-\boldsymbol\omega)$	Onsager–Casimir relation; the Hall resistivity is odd in $\mathbf B$ , the longitudinal part even.
Isotropic medium (Curie principle)	many $L_{ij}=0$	Fluxes and forces of different tensorial rank (scalar reaction vs vector current) do not couple, so those off-diagonal coefficients vanish identically.
Far from equilibrium	$J_i=\sum_j L_{ij}X_j+\tfrac12\sum_{jk}L_{ijk}X_jX_k+\cdots$	Reciprocity is a strictly linear-regime theorem; the nonlinear coefficients $L_{ijk}$ are not constrained by it.

Common trapReciprocity holds only for the entropy-conjugate fluxes and forces

The symmetry $L_{ij}=L_{ji}$ is not a property of any convenient pairing of currents and gradients — it holds only for the conjugate fluxes and forces whose product reproduces the entropy production $\dot S=\sum J_iX_i$ . Repackaging the fluxes (measuring energy flux instead of heat flux, or using $\nabla T$ rather than $\nabla(1/T)$ as the force) generally destroys the apparent symmetry of the coefficient matrix. Reciprocity is also a near-equilibrium, linear-regime statement; and in magnetic fields or rotating frames one must use the Onsager–Casimir form with reversed $\mathbf B$ . Confusing global stationarity with detailed balance is a related trap: a driven steady state can be stationary while violating reciprocity's premises.

Worked examples

Example 1 — Thermoelectric cross-coefficients and the Kelvin relation. This computes the Seebeck and Peltier coefficients from a symmetric Onsager matrix and confirms $\Pi=TS$ , the payoff of reciprocity.

Step 1. Take a thermoelectric material at $T=300\ \mathrm K$ with (schematic SI-ish) coefficients $L_{ee}=6.0\times10^{6}$ , $L_{qq}=2.0\times10^{9}$ , and cross-coefficient $L_{eq}=L_{qe}=9.0\times10^{7}$ (reciprocity imposed).

Step 2. The Seebeck coefficient (open-circuit voltage per temperature gradient) is $S=L_{eq}/(TL_{ee})=9.0\times10^{7}/(300\times6.0\times10^{6})=9.0\times10^{7}/1.8\times10^{9}=\mathbf{0.05}$ (V/K in these units).

Step 3. The Peltier coefficient (heat carried per unit charge at uniform $T$ ) is $\Pi=L_{qe}/L_{ee}=9.0\times10^{7}/6.0\times10^{6}=\mathbf{15}$ .

Step 4. Check Kelvin: $TS=300\times0.05=\mathbf{15}=\Pi$ . The Peltier coefficient equals $T$ times the Seebeck coefficient exactly — one measurement fixes both cross-effects. This is the single-equation content of reciprocity, and it eliminates one free parameter from every thermoelectric device model.

Example 2 — Positivity of entropy production bounds the cross-coefficient. The second law does more than fix the sign of $\dot S$ ; it constrains the whole matrix.

Step 1. Positivity of $\dot S=\sum_{ij}L_{ij}X_iX_j\ge0$ requires the symmetric part of $L$ to be positive semi-definite: $L_{ii}\ge0$ for every $i$ , and $L_{ii}L_{jj}\ge L_{ij}^2$ (with reciprocity $L_{ij}=L_{ji}$ ).

Step 2. For the numbers above, $L_{ee}L_{qq}=6.0\times10^{6}\times2.0\times10^{9}=\mathbf{1.2\times10^{16}}$ , while $L_{eq}^2=(9.0\times10^{7})^2=\mathbf{8.1\times10^{15}}$ .

Step 3. Since $1.2\times10^{16}\ge8.1\times10^{15}$ , the matrix is positive-definite and the determinant $\det L=L_{ee}L_{qq}-L_{eq}^2=3.9\times10^{15}>0$ . A thermoelectric material cannot have an arbitrarily large cross-coupling: the second law caps $L_{eq}\le\sqrt{L_{ee}L_{qq}}$ , which is exactly why the thermoelectric figure of merit $ZT$ has a ceiling set by the diagonal transport.

Origins · 1 minonsager reciprocal relations and how this was discovered

onsager reciprocal relations

proved by Lars Onsager, 1931 (Phys. Rev. 37, 405; 38, 2265) · The Onsager reciprocal relations were established by the Norwegian-American physical chemist Lars Onsager in two 1931 Physical Review papers, "Reciprocal Relations in Irreversible Processes I and II." Onsager took the empirical cross-effects of nineteenth-century transport physics — thermoelectricity, thermal diffusion, electrokinetics — and proved that their coupling coefficients are equal, tracing the equality back to the time-reversal invariance of the microscopic equations of motion via his regression hypothesis. The result was so far ahead of its time that it went largely unappreciated for a decade; it is now recognized as the founding theorem of linear irreversible thermodynamics. Onsager received the 1968 Nobel Prize in Chemistry specifically "for the discovery of the reciprocal relations bearing his name, which are fundamental for the thermodynamics of irreversible processes." Hendrik Casimir extended the relations to time-reversal-odd variables and magnetic fields in 1945.

History

1854

— William Thomson (Kelvin) conjectures the thermoelectric relation Pi = T S on thermodynamic grounds he cannot fully justify

1905

— Einstein's fluctuation theory supplies the statistical foundation later used in the reciprocity proof

1931

— Lars Onsager publishes "Reciprocal Relations in Irreversible Processes I and II," proving L_ij = L_ji from microscopic reversibility

1945

— Hendrik Casimir generalizes the relations to time-reversal-odd variables and magnetic fields

1962

— de Groot and Mazur's monograph systematizes linear irreversible thermodynamics around the reciprocal relations

1968

— Onsager receives the Nobel Prize in Chemistry for the reciprocal relations

Going further

References. L. Onsager, "Reciprocal Relations in Irreversible Processes I & II," Phys. Rev. 37, 405 (1931); 38, 2265 (1931) · H. B. G. Casimir, "On Onsager's Principle of Microscopic Reversibility," Rev. Mod. Phys. 17, 343 (1945) · S. R. de Groot & P. Mazur, Non-Equilibrium Thermodynamics (North-Holland, 1962), Ch. IV–VI · L. D. Landau & E. M. Lifshitz, Statistical Physics Part 1, 3rd ed., §120–122 · F. Reif, Fundamentals of Statistical and Thermal Physics (McGraw-Hill, 1965), §15.8 · M. Kardar, Statistical Physics of Particles (Cambridge, 2007), Ch. 3 · R. K. Pathria & P. D. Beale, Statistical Mechanics, 3rd ed. (Elsevier, 2011), §15.3–15.5.

Provenance. Equations and structure follow Onsager (1931), Casimir (1945), Landau–Lifshitz §120–122, Reif §15.8, and de Groot–Mazur. A verifier recomputes: (1) the Gaussian moment identity $\langle a_iX_j\rangle=-k_B\delta_{ij}$ by integration by parts on a random positive-definite Hessian; (2) the equivalence of $L=L^\top$ with the symmetry of the fluctuation-relaxation matrix, and the failure for asymmetric $L$ ; (3) the Kelvin relation $\Pi=TS$ and the worked numbers $S=0.05$ , $\Pi=15$ ; (4) the positivity bound $L_{ee}L_{qq}\ge L_{eq}^2$ ; (5) the Onsager–Casimir Hall parity from a Drude magnetoconductivity tensor; (6) the symmetry of the Green–Kubo flux-flux time-integral. All cited relations ( $\Pi=TS$ , $L_{ij}=L_{ji}$ , Hall oddness) are standard textbook results.

Builds on: Entropy and the Second Law · Equilibrium Fluctuation Theory · The Fluctuation-Dissipation Theorem · Time-Reversal Symmetry of Microscopic Dynamics Leads to: The Fluctuation-Dissipation Theorem · The Einstein Relation · The Langevin Equation · The Ornstein-Uhlenbeck Process

Why do I have to use

\nabla(1/T)

as the force instead of the more intuitive

\nabla T

Because reciprocity only holds for the entropy-conjugate forces — those whose product with the flux gives the entropy production $\dot S=\sum J_iX_i$ . The local entropy balance produces exactly $\mathbf J_q\cdot\nabla(1/T)$ , so $\nabla(1/T)$ is the correct force for the heat flux. Switching to $\nabla T=-T^2\nabla(1/T)$ rescales the coefficient by a factor of $T^2$ and, crucially, misaligns the cross-coefficients so that $L_{eq}\ne L_{qe}$ in the new variables. The symmetry is a statement about the specific matrix defined from the entropy representation, not about transport coefficients in general.

Is Onsager reciprocity the same thing as the fluctuation-dissipation theorem?

They are close cousins, both expressing dissipative (transport) coefficients through equilibrium correlation functions, but they are not identical. Reciprocity is the symmetry statement $L_{ij}=L_{ji}$ , a consequence of time-reversal invariance applied to cross-correlations. The fluctuation-dissipation theorem is the stronger, frequency-resolved statement relating the full response function to the equilibrium fluctuation spectrum. Onsager reciprocity is recovered from the FDT in the static (zero-frequency) limit, where the Green–Kubo integrand's manifest symmetry gives $L_{ij}=L_{ji}$ directly.

Does reciprocity constrain nonlinear transport far from equilibrium?

No. The proof is built for the linear regime where $J_i=\sum_j L_{ij}X_j$ . Once the forces are large enough that $J_i=\sum_j L_{ij}X_j+\tfrac12\sum_{jk}L_{ijk}X_jX_k+\cdots$ , the higher-order coefficients $L_{ijk}$ are not fixed by the reciprocity argument. Far-from-equilibrium physics is instead constrained by the fluctuation theorems (Jarzynski, Crooks), which are exact symmetry relations for the distribution of entropy production valid arbitrarily far from equilibrium — a different and more general class of result.

Why does a magnetic field break ordinary reciprocity?

Time-reversal symmetry is the engine of the proof, and a magnetic field is odd under time reversal. A charged particle spiraling in a field, run backward in time, spirals the wrong way — unless you also reverse the field. So the correlation symmetry no longer relates a system to itself but to its field-reversed twin: $\langle a_i(t)a_j(0)\rangle_{\mathbf B}=\langle a_j(t)a_i(0)\rangle_{-\mathbf B}$ . This is the Onsager–Casimir relation $L_{ij}(\mathbf B)=\epsilon_i\epsilon_j L_{ji}(-\mathbf B)$ , and it is exactly why the Hall effect (an antisymmetric, field-odd transport coefficient) can exist at all — ordinary reciprocity would forbid an antisymmetric part.

What does "regression of fluctuations" actually mean in the proof?

It is Onsager's hypothesis (proved within linear response, and the subject of its own article) that the average decay of a spontaneous equilibrium fluctuation follows the same linear laws as a macroscopically prepared deviation. In the reciprocity proof it enters as $\tfrac{\mathrm d}{\mathrm dt}\langle a_i(t)\rangle=\sum_j L_{ij}\langle X_j(t)\rangle$ : this inserts the very transport coefficients $L_{ij}$ into the dynamics of an equilibrium correlation function, whose independent time-reversal symmetry then forces those coefficients to be symmetric. Without regression there would be no link between the measurable $L_{ij}$ and the microscopic correlation symmetry.

If the linear laws look time-symmetric, where does irreversibility come from?

The linear laws $J_i=\sum_j L_{ij}X_j$ describe dissipation — they are not themselves time-reversal invariant, since a flux down a gradient produces entropy. What is time-reversal invariant is the underlying microscopic dynamics, and it is that reversibility (the symmetry of equilibrium fluctuation correlations) which forces the symmetry of the dissipative coefficient matrix. So reciprocity is a genuinely surprising bridge: a symmetry of the reversible microscopic world surviving as a constraint on the irreversible, dissipative macroscopic transport. The entropy production $\dot S=\sum_{ij}L_{ij}X_iX_j\ge0$ carries the arrow of time; reciprocity constrains the coefficients in that expression.

How does reciprocity halve the number of measurements for a transport system?

An $n$ -flux system has an $n\times n$ coefficient matrix with $n^2$ entries in general. Symmetry $L_{ij}=L_{ji}$ reduces the independent entries to $n(n+1)/2$ — for a 2-flux thermoelectric system, from 4 to 3, so measuring the Seebeck coefficient hands you the Peltier coefficient for free. In multicomponent diffusion (an $n$ -species mixture) the economy is larger: the cross-diffusion matrix has $n(n-1)/2$ fewer independent coefficients than it would without reciprocity, which is why experimentalists routinely use Onsager symmetry as a self-consistency check on measured transport matrices.