The Martin-Siggia-Rose / Janssen-De Dominicis response-field formalism

Intuition Beginner

A noisy classical equation of motion is a rule that says how a field changes in time, plus a random kick at every instant. The response-field formalism is a systematic way to turn that noisy rule into a quantum-field-theory path integral, so that all the machinery built for path integrals — Feynman diagrams, the renormalisation group, generating functionals — becomes available for the noisy dynamics.

The trick is to introduce a second field, called the response field, that lives alongside the original one. The original field tracks where the system is. The new field acts like a bookkeeper: its job is to enforce the equation of motion, instant by instant, inside the integral. You can picture it as a Lagrange multiplier that hovers over each moment in time and insists that the field obey its noisy rule.

Once both fields are in the integral, the random noise can be averaged away cleanly. Averaging over the kicks leaves behind a single term that couples the response field to itself, and that term is the only place the noise survives. What remains is an ordinary-looking field theory with two fields and a fixed action.

The payoff is a dictionary. The cross-correlation between the field and its response partner turns out to be the response function: it measures how the system reacts to a small push. The self-correlation of the field is the correlation function: it measures how the system fluctuates on its own. In thermal equilibrium these two are tied together by the fluctuation-dissipation relation, which the formalism encodes as a hidden symmetry.

Visual Beginner

A schematic with two stacked timelines. The top timeline shows a wavy field value relaxing toward a valley while small random arrows kick it at each instant. The bottom timeline shows a second "response" track that lights up only where a small external push is applied, and an arrow connects a push on the response track to the reaction it produces on the field track a moment later.

The picture captures the dictionary: the field track carries the fluctuations (the correlation function), the response track carries the reaction to a push (the response function), and the arrow from push to reaction always points forward in time, which is the causality built into the construction.

Worked example Beginner

Take the simplest noisy relaxation: a single number $\varphi (t)$ that decays toward zero with rate $\gamma$ and gets random kicks. The rule is $\dot{\varphi }=-\gamma \varphi +\eta$, where $\eta$ is white noise with strength set by $D$. We want the response function and the correlation function, and we want to see the dictionary in action.

Step 1. The response function answers: if I give the system a small push of size $h$ at time $t^{\prime }$, how much does $\varphi$ at a later time $t$ change? Solve the rule with a push. The change relaxes away at rate $\gamma$, so the reaction is $R(t,t^{\prime })=e^{-\gamma (t-t^{\prime })}$ for $t>t^{\prime }$, and exactly zero for $t<t^{\prime }$. The reaction cannot precede the push: that is causality.

Step 2. The correlation function answers: how does $\varphi$ at one time relate to $\varphi$ at another, when no push is applied and only the noise is acting? Multiply two solutions and average over the noise. The answer is $C(t,t^{\prime })=\frac{D}{\gamma }{e}^{-\gamma |t-t^{\prime }|}$, an even function of the time gap, with equal-time value $C(t,t)=D/\gamma$.

Step 3. Compare the two. The response is one-sided in time and the correlation is symmetric. At equal-time separation the response jumps from zero to one as we cross from past to future, while the correlation has a smooth peak. The ratio of strengths, $D/\gamma$ for the correlation against $1$ for the response, is the fluctuation-dissipation relation for this system: the size of the spontaneous wiggles is fixed by the same $D$ that calibrates the noise.

Step 4. The response-field formalism reproduces both functions from one object. It writes a path integral over $\varphi$ and a partner field, and the partner's only role is to carry the push. The field-partner average gives the one-sided response $R$, the field-field average gives the symmetric correlation $C$, and the partner-partner average is zero — which is the statement that two pushes never correlate.

What this tells us: a noisy decay rule already contains two distinct two-point functions, a causal response and a symmetric correlation, and they are linked in equilibrium. The formalism is the tool that produces both from a single field theory, and it scales up unchanged from one number to a field on a lattice or a continuum.

Check your understanding Beginner

Formal definition Intermediate+

Let $\varphi :{\mathbb{R}}^d\times \mathbb{R}\to \mathbb{R}$ be a real scalar field obeying a first-order Langevin equation in physical time $t$, $$ \partial_t \phi(x, t) = F[\phi](x, t) + \eta(x, t), $$ where $F[\varphi ]$ is a local drift functional (for relaxational dynamics, $F[\varphi ]=-\Gamma \delta \mathcal{H}/\delta \varphi$ with $\mathcal{H}$ a free-energy functional and $\Gamma$ a kinetic coefficient) and $\eta$ is Gaussian white noise with mean zero and covariance $$ \langle \eta(x, t), \eta(x', t') \rangle = 2 D, \delta^d(x - x'), \delta(t - t'). $$ The dynamics is fixed by an initial condition; the noise average defines all observables. This is the same Langevin object whose Fokker-Planck partner 08.10.02 governs the probability density on configuration space.

The noise-averaged generating functional is $$ Z[J, \hat{J}] = \Big\langle \int \mathcal{D}\phi; \delta\big[\partial_t\phi - F[\phi] - \eta\big]; \mathcal{J}[\phi]; e^{\int (J\phi)} \Big\rangle_\eta, $$ where the delta functional restricts the integral to the solution of the Langevin equation for each noise realisation and $\mathcal{J}[\varphi ]=\det (\delta ({\partial }_t\varphi -F)/\delta \varphi )$ is the Jacobian of the change of variables from $\eta$ to $\varphi$.

Introduce the response field $\hat{\varphi }(x,t)$ through the functional Fourier representation of the delta functional, $$ \delta\big[\partial_t\phi - F[\phi] - \eta\big] = \int \mathcal{D}\hat\phi; \exp!\Big(!-!\int dt, d^dx; \hat\phi,(\partial_t\phi - F[\phi] - \eta)\Big), $$ with $\hat{\varphi }$ integrated along the imaginary axis (the conventional rotation absorbs the factor of $i$). The exponent is now linear in $\eta$. Averaging over the Gaussian noise — a single Gaussian integral — gives $⟨e^{\int \hat{\varphi }\eta }{⟩}_{\eta }=e^{\int D{\hat{\varphi }}^2}$, leaving the MSR-Janssen-De Dominicis action $$ S[\phi, \hat\phi] = \int dt, d^dx; \Big[, \hat\phi,\big(\partial_t\phi - F[\phi]\big) - D,\hat\phi^2 ,\Big], $$ so that $Z=\int \mathcal{D}\varphi \mathcal{D}\hat{\varphi }\mathcal{J}[\varphi ]e^{-S[\varphi ,\hat{\varphi }]}$. The response field appears with the equation of motion in the first term and supplies the single noise vertex $-D{\hat{\varphi }}^2$ in the second.

The propagator dictionary is read from the quadratic part of $S$. The field-response correlation $⟨\varphi (x,t)\hat{\varphi }(x^{\prime },t^{\prime })⟩$ is the retarded response (Green) function $R(x-x^{\prime },t-t^{\prime })$, vanishing for $t<t^{\prime }$; the field-field correlation $⟨\varphi (x,t)\varphi (x^{\prime },t^{\prime })⟩$ is the correlation function $C$; and causality forces $$ \langle \hat\phi(x, t), \hat\phi(x', t') \rangle = 0 $$ identically, the response field having no self-contraction. The Jacobian $\mathcal{J}[\varphi ]=\det ({\partial }_t-\delta F/\delta \varphi )$ depends on the discretisation: in the Itô (prepoint) convention the equal-time step value $\theta (0)=0$ makes $\mathcal{J}$ a field-independent constant absorbed into the measure, while the Stratonovich (midpoint) convention $\theta (0)=1/2$ produces an extra $\frac{1}{2}\int \delta F/\delta \varphi$ term in the action.

Counterexamples to common slips

• The response-response propagator is not "small," it is exactly zero. A common slip is to treat $⟨\hat{\varphi }\hat{\varphi }⟩$ as a third independent two-point function; causality (the retardation of $R$) forces it to vanish to all orders, and any nonzero value signals a sign or contour error in the $\hat{\varphi }$ integration.
• The Jacobian is not automatically $1$. It equals $1$ in the Itô convention, but a Stratonovich-discretised Langevin equation carries the extra $\frac{1}{2}\delta F/\delta \varphi$ term; dropping it produces wrong loop corrections. The two conventions describe the same physics only when this term is tracked consistently.
• The noise vertex sign matters. The Gaussian average produces $+D{\hat{\varphi }}^2$ in the exponent, i.e. $-D{\hat{\varphi }}^2$ in the action $S=-\log (\text{weight})$; flipping it makes the $\hat{\varphi }$ integral divergent and the correlation function negative.

Key theorem with proof Intermediate+

Theorem (MSR response-field representation; Martin-Siggia-Rose 1973 ^{[Martin-Siggia-Rose]}, Janssen 1976 ^[Janssen], De Dominicis 1976 ^{[De-Dominicis]}). Let $\varphi$ obey the Langevin equation ${\partial }_t\varphi =F[\varphi ]+\eta$ with Gaussian white noise of covariance $2D{\delta }^d\delta$. Then the noise-averaged generating functional equals the doubled-field path integral $$ Z[J, \hat{J}] = \int \mathcal{D}\phi, \mathcal{D}\hat\phi; \exp!\Big(!-S[\phi, \hat\phi] + \int (J\phi + \hat{J}\hat\phi)\Big), $$ with $S[\varphi ,\hat{\varphi }]=\int dt{d}^{d}x[\hat{\varphi }({\partial }_t\varphi -F[\varphi ])-D{\hat{\varphi }}^2]$, the Itô Jacobian absorbed into the measure. The field-response two-point function $⟨\varphi \hat{\varphi }⟩={\delta }^2\log Z/\delta J\delta \hat{J}$ is the retarded response function $R$, the field-field function $⟨\varphi \varphi ⟩$ is the correlation function $C$, and $⟨\hat{\varphi }\hat{\varphi }⟩=0$.

Proof. Start from $Z=⟨\int \mathcal{D}\varphi \delta [{\partial }_t\varphi -F[\varphi ]-\eta ]\mathcal{J}[\varphi ]e^{\int J\varphi }{⟩}_{\eta }$, where the delta functional selects the unique Langevin trajectory and $\mathcal{J}$ is the Jacobian of $\eta \mapsto \varphi$. Represent the delta functional by its functional Fourier transform over an auxiliary field $\hat{\varphi }$: $$ \delta[\partial_t\phi - F[\phi] - \eta] = \int \mathcal{D}\hat\phi; e^{-\int \hat\phi,(\partial_t\phi - F[\phi] - \eta)}, $$ the $\hat{\varphi }$ contour running along the imaginary axis. Insert this and exchange the order of integration so the noise average acts on the only $\eta$-dependent factor: $$ \big\langle e^{\int \hat\phi,\eta}\big\rangle_\eta = \exp!\Big(\tfrac12 \int \hat\phi(x,t), \langle\eta,\eta\rangle, \hat\phi(x',t')\Big) = \exp!\Big(\int D,\hat\phi^2\Big), $$ using the Gaussian moment-generating identity with covariance $2D{\delta }^d\delta$. Collecting terms, the exponent becomes $-\int [\hat{\varphi }({\partial }_t\varphi -F[\varphi ])-D{\hat{\varphi }}^2]=-S[\varphi ,\hat{\varphi }]$. In the Itô convention the Jacobian $\mathcal{J}[\varphi ]=\det ({\partial }_t-\delta F/\delta \varphi )$ equals $\det ({\partial }_t)$, a field-independent constant, since the equal-time response $\theta (0)=0$ removes the $\delta F/\delta \varphi$ contribution; it is absorbed into $\mathcal{D}\hat{\varphi }$. Adding the sources gives the stated $Z$.

For the propagators, take the free (Gaussian-$F$) limit $F[\varphi ]=-\Gamma (-{\nabla }^2+r)\varphi =:-A\varphi$. The action is $S_0=\int [\hat{\varphi }({\partial }_t+A)\varphi -D{\hat{\varphi }}^2]$, a quadratic form in $(\varphi ,\hat{\varphi })$ with kernel $$ M = \begin{pmatrix} 0 & (-\partial_t + A)^T \ (\partial_t + A) & -2D \end{pmatrix}, $$ where the upper-left block is the $\varphi \varphi$ entry. Inverting $M$ gives the propagator matrix. The $\varphi \hat{\varphi }$ entry is $({\partial }_t+A{)}^{-1}$, the retarded Green function $R$ with $R(t-t^{\prime })=0$ for $t<t^{\prime }$ by the causal pole structure; the $\varphi \varphi$ entry is $({\partial }_t+A{)}^{-1}(2D)(-{\partial }_t+A{)}^{-1}$, the symmetric correlation $C$; and the $\hat{\varphi }\hat{\varphi }$ entry is $0$ because $M$ has a vanishing upper-left block, so $⟨\hat{\varphi }\hat{\varphi }⟩=0$. $\square$

Bridge. The MSR representation builds toward the entire perturbative and renormalisation-group apparatus for stochastic dynamics: the foundational reason a noisy classical equation can be diagrammed is exactly that the response-field doubling converts the noise average into a single quadratic vertex $-D{\hat{\varphi }}^2$, and this is exactly the same delta-functional-plus-Gaussian-average move that appears again in 08.10.08 when stochastic quantisation builds the Euclidean measure as a Langevin equilibrium. The construction generalises the Onsager-Machlup path probability — integrating $\hat{\varphi }$ back out of $S$ recovers the Onsager-Machlup functional $\frac{1}{4D}\int ({\partial }_t\varphi -F{)}^2$, so the response field is dual to the squared-residual weighting of trajectories. The central insight is that the two propagators of the doubled theory, $R=⟨\varphi \hat{\varphi }⟩$ and $C=⟨\varphi \varphi ⟩$, separate cause from fluctuation: $R$ is one-sided in time and $C$ is symmetric, and putting these together with the equilibrium condition yields the fluctuation-dissipation theorem of 08.12.01 as a Ward identity, developed in the Master tier below.

Exercises Intermediate+

Advanced results Master

Theorem (fluctuation-dissipation as a Ward identity; Janssen 1976 ^[Janssen], Zinn-Justin Ch. 17 ^{[Zinn-Justin]}). For relaxational dynamics $F=-\Gamma \delta \mathcal{H}/\delta \varphi$ with $D=\Gamma T$, the MSR action is invariant in equilibrium under the field transformation $\varphi (t)\mapsto \varphi (-t)$, $\hat{\varphi }(t)\mapsto \hat{\varphi }(-t)-\frac{1}{T}{\partial }_t\varphi (-t)$ (a time reversal combined with a response-field shift). The Ward identity associated with this invariance is exactly the fluctuation-dissipation theorem $R(t)=\frac{1}{T}\theta (t){\partial }_{t}C(t)$.

The transformation is a symmetry only when the noise strength and kinetic coefficient satisfy $D=\Gamma T$, which is the detailed-balance / Einstein condition that the Langevin dynamics relaxes to $e^{-\mathcal{H}/T}$. The shift mixes $\hat{\varphi }$ with ${\partial }_t\varphi$, so the response field is genuinely the carrier of the dissipative half of the relation. The associated conserved Ward identity, obtained by demanding that $⟨\cdots ⟩$ be invariant under the transformation, links the field-response correlator to the time-derivative of the field-field correlator at every order in perturbation theory — the FDT survives renormalisation. Out of equilibrium, $D\ne \Gamma T$, the symmetry is broken and the FDT fails, which is the field-theoretic statement of the difference between equilibrium and driven steady states.

Theorem (supersymmetric formulation; Zinn-Justin Ch. 17 ^{[Zinn-Justin]}). The equilibrium MSR action with its Jacobian can be written in a superspace with two Grassmann coordinates $\theta ,\bar{\theta }$ and a superfield $\Phi (x,t,\theta ,\bar{\theta })=\varphi +\bar{\theta }\psi +\bar{\psi }\theta +\bar{\theta }\theta \hat{\varphi }$, where $\psi ,\bar{\psi }$ are anticommuting ghosts that exponentiate the functional Jacobian $\det ({\partial }_t-\delta F/\delta \varphi )$. The action becomes a single superspace integral, and the time-reversal/FDT symmetry plus the BRST symmetry generated by the ghosts organise into an $\mathrm{OSp}$ supersymmetry.

The ghost fields $\psi ,\bar{\psi }$ are not optional bookkeeping: in the Stratonovich convention the Jacobian is field-dependent, and exponentiating it as $\det =\int \mathcal{D}\bar{\psi }\mathcal{D}\psi e^{\int \bar{\psi }({\partial }_t-\delta F/\delta \varphi )\psi }$ supplies exactly the fermionic partners that complete the superfield. The supersymmetry is the algebraic skeleton from which the FDT and the Parisi-Sourlas dimensional-reduction phenomena follow; it makes the equilibrium structure of stochastic dynamics manifest in a way the component formulation hides.

Theorem (KPZ via MSR; Kardar-Parisi-Zhang 1986 ^[KPZ]). The Kardar-Parisi-Zhang growth equation ${\partial }_{t}h=\nu {\nabla }^{2}h+\frac{\lambda }{2}(\nabla h{)}^2+\eta$ for an interface height $h(x,t)$ has the MSR action $S=\int [\hat{h}({\partial }_{t}h-\nu {\nabla }^{2}h-\frac{\lambda }{2}(\nabla h{)}^2)-D{\hat{h}}^2]$. The nonlinear $(\nabla h{)}^2$ term produces a three-point vertex $\hat{h}(\nabla h{)}^2$; in $1+1$ dimensions the dynamic exponent is $z=3/2$, the roughness exponent $\alpha =1/2$, and the growth exponent $\beta =\alpha /z=1/3$, the exponents of the KPZ universality class.

The KPZ exponents in $1+1$ dimensions are protected by a fluctuation-dissipation symmetry special to that dimension, and the Cole-Hopf map $w=e^{(\lambda /2\nu )h}$ linearises the equation into a diffusion equation with multiplicative noise, equivalently the partition function of a directed polymer in a random medium. The MSR field theory is the systematic route to the exponents in higher dimensions, where no exact solution exists; the perturbative RG flow of the coupling $\lambda$ has a strong-coupling fixed point inaccessible to the epsilon-expansion, which is what makes KPZ a genuinely hard and still-active problem.

Synthesis. The response-field formalism puts together the foundational reason a noisy classical equation of motion becomes a field theory: the delta functional enforcing the Langevin equation is Fourier-represented over a response field $\hat{\varphi }$, and the Gaussian noise average collapses to the single vertex $-D{\hat{\varphi }}^2$, so the doubled action $S=\int [\hat{\varphi }({\partial }_t\varphi -F)-D{\hat{\varphi }}^2]$ carries the whole stochastic content. This is exactly the analytic dual of the operator Fokker-Planck description of 08.10.02: the central insight is that the same Langevin dynamics admits two equivalent renderings, an operator one on probability densities and a path-integral one on trajectories, and the response field is what makes the trajectory rendering a bona-fide field theory with Feynman rules. Putting these together with the propagator dictionary — $R=⟨\varphi \hat{\varphi }⟩$ retarded, $C=⟨\varphi \varphi ⟩$ symmetric, $⟨\hat{\varphi }\hat{\varphi }⟩=0$ — gives the fluctuation-dissipation theorem of 08.12.01 as a time-reversal Ward identity, and the bridge is the recognition that this Ward identity is a fragment of a larger supersymmetry whose breaking distinguishes equilibrium from driven steady states.

The construction generalises in three load-bearing directions: to dynamic critical phenomena, where the Hohenberg-Halperin models A through J classify the relaxational and conserved Langevin equations and the MSR field theory delivers the dynamic exponent $z$ by epsilon-expansion; to the KPZ equation and directed polymers 08.14.07, where the nonlinear growth vertex pushes the theory to a strong-coupling fixed point; and to reaction-diffusion systems via the Doi-Peliti second-quantised formalism, where a different but parallel field doubling turns a classical master equation into a field theory. The foundational reason all three work is the same single move; the central insight that unifies them is that classical stochastic dynamics, equilibrium or not, is a quantum field theory in disguise once the response field is admitted.

Full proof set Master

Proposition (Gaussian noise average produces the ${\hat{\varphi }}^2$ vertex). Let $\eta$ be Gaussian with mean zero and covariance $⟨\eta (z)\eta (z^{\prime })⟩=2D\delta (z-z^{\prime })$, where $z=(x,t)$. Then for any field $\hat{\varphi }$, $⟨e^{\int \hat{\varphi }\eta }{⟩}_{\eta }=e^{\int D{\hat{\varphi }}^2}$.

Proof. The moment-generating functional of a centred Gaussian field with covariance $G(z,z^{\prime })$ is $⟨e^{\int j\eta }⟩=\exp (\frac{1}{2}\int j(z)G(z,z^{\prime })j(z^{\prime }))$, a standard completion-of-the-square identity for Gaussian measures (the field analogue of 08.06.01). Take $j=\hat{\varphi }$ and $G(z,z^{\prime })=2D\delta (z-z^{\prime })$. Then $$ \big\langle e^{\int \hat\phi\eta}\big\rangle_\eta = \exp!\Big(\tfrac12 \int \hat\phi(z), 2D,\delta(z - z'), \hat\phi(z'), dz, dz'\Big) = \exp!\Big(\int D,\hat\phi(z)^2, dz\Big). $$ The delta function collapses the double integral to a single one, giving the local vertex $+D{\hat{\varphi }}^2$ in the exponent, hence $-D{\hat{\varphi }}^2$ in the action. $\square$

Proposition (causal response propagator). For the free relaxational action $S_0=\int [\hat{\varphi }({\partial }_t+\gamma )\varphi -D{\hat{\varphi }}^2]$ on the real line, the field-response propagator is $R(t)=e^{-\gamma t}\theta (t)$, vanishing for $t<0$.

Proof. The $\varphi$-$\hat{\varphi }$ propagator is the inverse of the operator ${\partial }_t+\gamma$ acting between the two fields, i.e. the retarded Green function $R$ of $({\partial }_t+\gamma )R(t)=\delta (t)$. In frequency space, $(-i\omega +\gamma )\tilde{R}(\omega )=1$, so $\tilde{R}(\omega )=(\gamma -i\omega {)}^{-1}$. The single pole sits at $\omega =-i\gamma$ in the lower half-plane. Closing the inverse-transform contour in the upper half-plane for $t<0$ encloses no pole, giving $R(t)=0$; closing in the lower half-plane for $t>0$ picks up the residue, giving $R(t)=e^{-\gamma t}$. So $R(t)=e^{-\gamma t}\theta (t)$, retarded. The lower-half-plane pole — equivalently the choice of the response field's integration contour — is what encodes causality. $\square$

Proposition (vanishing response-response correlator). In the MSR theory the two-point function $⟨\hat{\varphi }(z)\hat{\varphi }(z^{\prime })⟩$ vanishes identically, to all orders in perturbation theory.

Proof. At quadratic order, the kernel of $S_0$ in the basis $(\varphi ,\hat{\varphi })$ has a zero in the $\varphi \varphi$ slot: $$ S_0 = \tfrac12 \int (\phi;; \hat\phi) \begin{pmatrix} 0 & -\partial_t + \gamma \ \partial_t + \gamma & -2D \end{pmatrix} \begin{pmatrix} \phi \ \hat\phi \end{pmatrix}. $$ The propagator is the inverse of this kernel. A block matrix $\left(\begin{array}{cc}0& B^T\\ B& C\end{array}\right)$ with invertible $B$ has inverse $\left(\begin{array}{cc}-B^{-1}C(B^T{)}^{-1}& B^{-1}\\ (B^T{)}^{-1}& 0\end{array}\right)$, whose lower-right block — the $\hat{\varphi }\hat{\varphi }$ entry — is zero. So $⟨\hat{\varphi }\hat{\varphi }⟩=0$ at tree level. Interaction vertices in the MSR action always carry at least one response field ($\hat{\varphi }({\partial }_t\varphi -F)$ has exactly one $\hat{\varphi }$ per vertex, and the noise vertex ${\hat{\varphi }}^2$ carries two), but every internal line out of a vertex that could contribute to $⟨\hat{\varphi }\hat{\varphi }⟩$ requires a $⟨\hat{\varphi }\hat{\varphi }⟩$ sub-contraction, which vanishes by induction. Hence $⟨\hat{\varphi }\hat{\varphi }⟩=0$ to all orders, the field-theoretic statement of normalisation $Z[0,0]=1$ and probability conservation. $\square$

Proposition (Onsager-Machlup as the $\hat{\varphi }$-integrated action). Integrating $\hat{\varphi }$ out of the MSR action $S=\int [\hat{\varphi }({\partial }_t\varphi -F[\varphi ])-D{\hat{\varphi }}^2]$ yields the Onsager-Machlup functional $S_{\mathrm{OM}}[\varphi ]=\frac{1}{4D}\int ({\partial }_t\varphi -F[\varphi ]{)}^2$ (up to the Jacobian).

Proof. The action is quadratic in $\hat{\varphi }$. Complete the square: $$ -D\hat\phi^2 + \hat\phi(\partial_t\phi - F) = -D\Big(\hat\phi - \frac{\partial_t\phi - F}{2D}\Big)^2 + \frac{(\partial_t\phi - F)^2}{4D}. $$ The Gaussian integral over the shifted field ${\hat{\varphi }}^{\prime }=\hat{\varphi }-({\partial }_t\varphi -F)/2D$ contributes a field-independent normalisation. The remaining $\varphi$-dependent exponent is $-S_{\mathrm{OM}}[\varphi ]=-\frac{1}{4D}\int ({\partial }_t\varphi -F[\varphi ]{)}^2$, the Onsager-Machlup path weight ^{[Onsager-Machlup]}. The deterministic trajectory ${\partial }_t\varphi =F[\varphi ]$ is the maximum-probability path, the residual measuring departures from it. The Jacobian $\det ({\partial }_t-\delta F/\delta \varphi )$ rides along as the prefactor; in the Itô convention it is constant. $\square$

Proposition (FDT from the equilibrium Ward identity). For relaxational dynamics with $D=\Gamma T$, the time-reversal-plus-shift transformation $\varphi (t)\mapsto \varphi (-t)$, $\hat{\varphi }(t)\mapsto \hat{\varphi }(-t)-\frac{1}{T}{\partial }_t\varphi (-t)$ leaves $S$ invariant, and the resulting Ward identity is $R(t)=\frac{1}{T}\theta (t){\partial }_{t}C(t)$.

Proof. Substitute the transformation into $S=\int [\hat{\varphi }({\partial }_t\varphi +\Gamma \frac{\delta \mathcal{H}}{\delta \varphi })-D{\hat{\varphi }}^2]$ and use $D=\Gamma T$. The shift in $\hat{\varphi }$ generates cross terms; the $-\frac{1}{T}{\partial }_t\varphi \cdot {\partial }_t\varphi$ piece is a total derivative and the $\frac{1}{T}{\partial }_t\varphi \cdot \Gamma \frac{\delta \mathcal{H}}{\delta \varphi }=\frac{\Gamma }{T}\frac{d\mathcal{H}}{dt}$ piece is also a total time-derivative, both integrating to boundary terms that vanish in equilibrium. The $D{\hat{\varphi }}^2$ term reproduces itself under the combined reflection and shift precisely when $D=\Gamma T$. Invariance of $S$ implies invariance of all correlators; applying the transformation to $⟨\varphi (t)\hat{\varphi }(0)⟩=R(t)$ and matching to $⟨\varphi (t)\varphi (0)⟩=C(t)$ under the shift gives $R(t)=\frac{1}{T}\theta (t){\partial }_{t}C(t)$, the fluctuation-dissipation theorem of 08.12.01. $\square$

Proposition (KPZ $1+1$ exponents from Galilean and FDT symmetries). The KPZ equation ${\partial }_{t}h=\nu {\nabla }^{2}h+\frac{\lambda }{2}(\nabla h{)}^2+\eta$ in one spatial dimension has dynamic exponent $z=3/2$, roughness exponent $\alpha =1/2$, growth exponent $\beta =1/3$.

Proof sketch. The KPZ action has two exact symmetries. First, Galilean invariance: the shift $h\to h+\mathbf{v}\cdot x$, $x\to x-\lambda \mathbf{v}t$ leaves the equation form-invariant and forces the scaling relation $\alpha +z=2$, protecting the coupling $\lambda$ from renormalisation. Second, in $d=1$ a fluctuation-dissipation symmetry makes the stationary interface a random walk, fixing the roughness exponent $\alpha =1/2$ exactly (the equal-time height-difference correlation grows linearly in separation). Combining $\alpha =1/2$ with $\alpha +z=2$ gives $z=3/2$, and the growth exponent follows from the scaling relation $\beta =\alpha /z=(1/2)/(3/2)=1/3$. The MSR field theory supplies these symmetries as Ward identities and confirms the exponents at the strong-coupling fixed point; the exact $\alpha =1/2$ in $d=1$ does not extend to higher dimensions, where the strong-coupling exponents are known only numerically and the perturbative epsilon-expansion misses the relevant fixed point. $\square$

Connections Master

• Fokker-Planck equation and equilibrium distribution 08.10.02. The MSR response-field path integral is the trajectory-space dual of the operator Fokker-Planck description of the same Langevin equation. Where 08.10.02 evolves a probability density on configuration space and identifies $e^{-S}/Z$ as the stationary solution, the present unit weights whole trajectories and reads off response and correlation functions directly; the two descriptions agree on every equal-time observable, and the detailed-balance condition $D=\Gamma T$ of 08.10.02 is exactly the FDT symmetry of the MSR action.

• Langevin updates and lattice numerics 08.10.08. Stochastic quantisation builds the Euclidean measure as the equilibrium of a Langevin process in fictitious time; the MSR formalism is the analytic engine that turns that Langevin process into a perturbatively-calculable field theory. The delta-functional-plus-noise-average construction here is the same move that underlies the diagrammatic expansion of stochastic quantisation, and the response field supplies the propagators that the lattice numerics of 08.10.08 estimate empirically.

• Fluctuation-dissipation theorem 08.12.01. The FDT enters the MSR formalism as a Ward identity: the response function $⟨\varphi \hat{\varphi }⟩$ and the correlation function $⟨\varphi \varphi ⟩$ are tied together by a time-reversal symmetry of the action that holds exactly when the noise strength matches the temperature. This is the dynamical, field-theoretic face of the static FDT of 08.12.01, and its breaking is the field-theoretic signature of a non-equilibrium driven steady state.

• Path integral formulation of statistical mechanics 08.07.01. The MSR construction is a path integral over trajectories, doubled by the response field. Where 08.07.01 sets up the path-integral measure for equilibrium statistical mechanics, the present unit extends the same machinery to time-dependent stochastic dynamics, so that generating functionals, Feynman diagrams, and the renormalisation group apply to relaxation and growth, not only to equilibrium partition functions.

• Gaussian field theory and free boson 08.06.01. The single noise vertex $-D{\hat{\varphi }}^2$ comes from a Gaussian moment-generating identity, and the free MSR theory is a Gaussian field theory in the doubled $(\varphi ,\hat{\varphi })$ variables. The propagator dictionary — retarded response, symmetric correlation, vanishing response-response — is the dynamic generalisation of the static Gaussian propagator of 08.06.01, with causality selecting the retarded branch.

• Brownian motion, the Wiener measure, and the path integral 08.14.01. The white noise $\eta$ driving the Langevin equation is the time-derivative of a Wiener process, and the response-field path integral is a field-theoretic packaging of the Wiener measure on noise realisations. The Itô-versus-Stratonovich Jacobian ambiguity that fixes the MSR functional determinant is the stochastic-calculus subtlety that 08.14.01 introduces at the level of single-particle Brownian motion.

Historical & philosophical context Master

The response-field formalism originates in Paul Martin, Eric Siggia, and Harvey Rose's 1973 paper Statistical dynamics of classical systems (Phys. Rev. A 8, 423) ^{[Martin-Siggia-Rose]}, written in the language of operators rather than path integrals. Their construction introduced, alongside the physical field operator, a conjugate "response operator" whose commutators with the field reproduce the response functions of the classical stochastic dynamics, building a closed operator algebra modelled on the canonical formalism of quantum field theory. The motivation was the statistical theory of turbulence and classical fluids, where a systematic perturbation theory for correlation and response functions had been lacking; the MSR operator algebra gave one. The construction was technically intricate, and its full power became clear only when it was recast as a functional integral.

That recasting was carried out independently by Hans-Karl Janssen in On a Lagrangean for classical field dynamics and renormalization group calculations of dynamical critical properties (Z. Phys. B 23, 377 (1976)) ^[Janssen] and by Cyrano De Dominicis in Techniques de renormalisation de la théorie des champs et dynamique des phénomènes critiques (J. Phys. Paris Colloq. 37, C1-247 (1976)) ^{[De-Dominicis]}. Both wrote the noise-averaged Langevin generating functional as a path integral over the field and a conjugate response field with the action $S=\int [\hat{\varphi }({\partial }_t\varphi -F)-D{\hat{\varphi }}^2]$, making the renormalisation group directly applicable to stochastic dynamics. The combined construction is universally called the Martin-Siggia-Rose-Janssen-De Dominicis (MSRJD) formalism. Its first major payoff was the systematic theory of dynamic critical phenomena: Pierre Hohenberg and Bertrand Halperin's 1977 review Theory of dynamic critical phenomena (Rev. Mod. Phys. 49, 435) ^{[Hohenberg-Halperin]} classified relaxational and conserved stochastic dynamics into the now-standard models A through J and used the response-field field theory to compute the dynamic critical exponent $z$ in the epsilon-expansion.

The philosophical interest of the formalism is its assertion that classical statistical dynamics is a quantum field theory in disguise. The doubling of fields — a physical field and its response partner — is the classical-stochastic analogue of the doubling of contours in the Schwinger-Keldysh formalism for non-equilibrium quantum systems, and the two formalisms converge in the classical limit. The equilibrium FDT appears not as an external input but as a Ward identity of a hidden supersymmetry, a structural fact that Jean Zinn-Justin's textbook ^{[Zinn-Justin]} makes central. The same supersymmetric structure underlies Parisi and Sourlas's dimensional reduction for random-field systems, tying the response-field formalism to the deepest results in the theory of disordered systems. Beyond equilibrium, the 1986 Kardar-Parisi-Zhang equation ^[KPZ] — a nonlinear Langevin equation for surface growth — became the canonical hard application of the formalism and the defining example of a strong-coupling stochastic fixed point that the perturbative renormalisation group cannot reach, a problem that remains active four decades later.

Bibliography Master

@article{MartinSiggiaRose1973,
  author  = {Martin, P. C. and Siggia, E. D. and Rose, H. A.},
  title   = {Statistical dynamics of classical systems},
  journal = {Physical Review A},
  volume  = {8},
  year    = {1973},
  pages   = {423--437}
}

@article{Janssen1976,
  author  = {Janssen, H.-K.},
  title   = {On a {L}agrangean for classical field dynamics and renormalization group calculations of dynamical critical properties},
  journal = {Zeitschrift f{\"u}r Physik B},
  volume  = {23},
  year    = {1976},
  pages   = {377--380}
}

@article{DeDominicis1976,
  author  = {De Dominicis, C.},
  title   = {Techniques de renormalisation de la th{\'e}orie des champs et dynamique des ph{\'e}nom{\`e}nes critiques},
  journal = {Journal de Physique (Paris) Colloque},
  volume  = {37},
  year    = {1976},
  pages   = {C1-247--C1-253}
}

@article{HohenbergHalperin1977,
  author  = {Hohenberg, P. C. and Halperin, B. I.},
  title   = {Theory of dynamic critical phenomena},
  journal = {Reviews of Modern Physics},
  volume  = {49},
  year    = {1977},
  pages   = {435--479}
}

@article{KardarParisiZhang1986,
  author  = {Kardar, Mehran and Parisi, Giorgio and Zhang, Yi-Cheng},
  title   = {Dynamic scaling of growing interfaces},
  journal = {Physical Review Letters},
  volume  = {56},
  year    = {1986},
  pages   = {889--892}
}

@article{OnsagerMachlup1953,
  author  = {Onsager, Lars and Machlup, Stefan},
  title   = {Fluctuations and irreversible processes},
  journal = {Physical Review},
  volume  = {91},
  year    = {1953},
  pages   = {1505--1512}
}

@book{ZinnJustin2002,
  author    = {Zinn-Justin, Jean},
  title     = {Quantum Field Theory and Critical Phenomena},
  publisher = {Oxford University Press},
  edition   = {4},
  year      = {2002}
}

@book{Tauber2014,
  author    = {T{\"a}uber, Uwe C.},
  title     = {Critical Dynamics: A Field Theory Approach to Equilibrium and Non-Equilibrium Scaling Behavior},
  publisher = {Cambridge University Press},
  year      = {2014}
}

@book{Kamenev2011,
  author    = {Kamenev, Alex},
  title     = {Field Theory of Non-Equilibrium Systems},
  publisher = {Cambridge University Press},
  year      = {2011}
}

@book{Kardar2007Fields,
  author    = {Kardar, Mehran},
  title     = {Statistical Physics of Fields},
  publisher = {Cambridge University Press},
  year      = {2007}
}

@article{Ito1951SDE,
  author  = {It{\^o}, Kiyosi},
  title   = {On stochastic differential equations},
  journal = {Memoirs of the American Mathematical Society},
  volume  = {4},
  year    = {1951},
  pages   = {1--51}
}