---
title: Complex Langevin Dynamics
url: https://www.emergentmind.com/topics/complex-langevin-dynamics
type: topic
---

# Complex Langevin Dynamics

Complex Langevin dynamics (CLD) is a stochastic-quantisation method for theories with a complex action, in which the original real degrees of freedom are complexified and evolved in a fictitious Langevin time instead of being sampled by importance sampling. In the standard formulation one replaces \(x\in\mathbb{R}^n\) by \(z=x+iy\in\mathbb{C}^n\), evolves \(dz=K(z)\,dt+d\eta\) with complex drift \(K(z)=-\partial_z S(z)\), and computes holomorphic observables as long-time averages over the induced real probability distribution on the complexified configuration space [1110.5749][1408.3577]. The method is intended for systems with a sign problem, but its validity depends on holomorphicity, localization of the sampled distribution, and the absence of dangerous boundary terms or singular-drift effects [1303.6425][1701.02322].

## 1. Stochastic formulation and complexification

For a lattice field theory with action \(S[\phi]\), CLD introduces a fictitious Langevin time \(t\) and evolves each degree of freedom according to
\[
d\phi_x(t)=-\frac{\partial S[\phi]}{\partial \phi_x}\,dt+dW_x(t),
\]
with real Wiener increments satisfying \(\langle dW_x(t)\rangle=0\) and \(\langle dW_x(t)\,dW_y(t)\rangle=2\delta_{xy}dt\) [2509.13314]. When the action is complex, the fields must be complexified,
\[
\phi_x\rightarrow \phi_x^R+i\phi_x^I,
\]
and the evolution splits into coupled real equations for \(\phi_x^R\) and \(\phi_x^I\), with drift components given by the real and imaginary parts of \(-\partial S/\partial\phi_x\) evaluated on the complexified fields [1009.5838][2509.13314].

In practice one usually discretizes Langevin time by an Euler–Maruyama update. In the notation used for the three-dimensional XY model,
\[
\phi_x^R(n+1)=\phi_x^R(n)+\epsilon\,K_x^R[\phi(n)]+\sqrt{\epsilon}\,\eta_x(n),\qquad
\phi_x^I(n+1)=\phi_x^I(n)+\epsilon\,K_x^I[\phi(n)],
\]
with independent Gaussian noise of zero mean and variance \(2\) [1009.5838]. Closely related discrete updates appear in chiral random matrix theory, where each real matrix component \(u\in\{a,b,\alpha,\beta\}\) is updated as
\[
u_{ij}(t+\Delta t)=u_{ij}(t)+\Delta t\,K^{(u)}_{ij}+\sqrt{2\Delta t}\,\eta_{ij}(t)
\]
[1309.4335].

The same structure extends to gauge theories and matrix models. For gauge links \(U_{x,\mu}\in SL(N,\mathbb C)\), CLD is written directly on the complexified group manifold, while for unitary matrix models one often evolves the eigenvalue angles \(\theta_i\) into the complex plane, so that the eigenvalues \(z_i=e^{i\theta_i}\) leave the unit circle and explore \(\mathbb C^\times\) [1311.1056][1802.10381]. This makes CLD fundamentally different from reweighting or dual-variable reformulations: it replaces sampling of a complex measure by sampling of a real distribution in a higher-dimensional complexified space.

## 2. Fokker–Planck structure and criteria for correctness

The formal basis of CLD is a real Fokker–Planck equation for the probability density \(P(x,y;t)\) of the complexified process. In the one-variable case the density satisfies
\[
\frac{\partial}{\partial t}P(x,y;t)=L^T P(x,y;t),
\]
and observables are computed as
\[
\langle O\rangle_{P(t)}=\int dx\,dy\,P(x,y;t)\,O(x+iy)
\]
[1110.5749]. A corresponding “complex density” \(\rho(x;t)\) on the original real manifold has stationary solution \(\rho(x;\infty)\propto e^{-S(x)}\), and the formal argument attempts to show equality of \(\langle O\rangle_{P(t)}\) and \(\langle O\rangle_{\rho(t)}\) for holomorphic observables by transferring time evolution from the density to the observable and integrating by parts [1110.5749][2509.13314].

The critical point is that this proof requires vanishing boundary terms. A necessary and sufficient condition for correctness is that all such boundary terms vanish; equivalently, at stationarity one must have
\[
\int dx\,dy\,(L\,O)(x+iy)\,P(x,y)=0
\]
for every holomorphic observable \(O(z)\) [1110.5749]. In the related consistency-condition language, one defines
\[
C_O=\langle \tilde L\,O(z)\rangle,
\qquad
\tilde L=[\partial_z-(\partial_z S)]\partial_z,
\]
and requires \(C_O=0\) for a sufficiently large basis of observables [1306.3075][1309.3191].

Localization properties of \(P(x,y)\) are therefore central. In the quartic toy model
\[
Z=\int dx\,\exp\!\Bigl[-\tfrac12\sigma x^2-\tfrac14\lambda x^4\Bigr],
\qquad \sigma=A+iB,\ \lambda>0,
\]
it was shown that for real noise and \(3A^2>B^2\), the equilibrium distribution has strict strip support,
\[
P(x,y)=0 \quad \text{for } |y|\ge y_0,
\]
with an explicit \(y_0\), and in that regime correct results are expected because boundary terms vanish [1309.3191]. By contrast, when no such strip exists, power-law tails appear and the formal justification breaks down [1306.3075][1309.3191].

This framework also clarifies why CLD is not guaranteed merely by numerical stability. A trajectory may remain bounded and still sample a distribution whose tails or singular regions invalidate integration by parts. Conversely, a localized distribution with vanishing \(\langle \tilde L O\rangle\) gives a concrete a posteriori basis for trusting the method [1110.5749][1306.3075].

## 3. Singular drifts, logarithms, and mechanisms of wrong convergence

The most persistent obstructions to correctness are slow decay in imaginary directions, singular drifts generated by zeros of determinants, and multivalued logarithms. In theories with
\[
S(z)=S_g(z)-\ln\det M(z),
\]
the drift contains
\[
\mathrm{Tr}\bigl[M^{-1}(z)\,\partial_i M(z)\bigr],
\]
so zeros of \(\det M\) induce poles in the drift and make it meromorphic rather than holomorphic [1701.02322]. In that case the usual integration-by-parts argument acquires extra boundary terms around small neighborhoods of the poles, and correctness requires that the probability distribution be negligible near the determinant zeros [1701.02322].

A particularly transparent example is chiral random matrix theory at nonzero chemical potential. There the action contains \(-N_f\,\mathrm{Tr}\log[m^2-XY]\), and incorrect convergence occurs for small quark masses when the determinant frequently traces out a path surrounding the origin of the complex plane during the Langevin flow [1309.4335]. The study identifies the failed region through scatter plots of \(\det(m^2-XY)\) and through the quantity \(R_{\det}\), the fraction of configurations with \(\mathrm{Re}\,\det>0\). As soon as \(R_{\det}<1\), the CLD results for the condensate \(\Sigma\) and baryon density \(n_B\) deviate strongly from the exact answer, and numerically they become indistinguishable from the phase-quenched theory [1309.4335]. The immediate cause is an ambiguity in the drift: using \(\partial_z\log z=1/z\) everywhere ignores the branch cut, but if \(\det M\) winds around the origin the phase changes by \(2\pi\), and control over the drift is lost [1309.4335].

Related issues appear in simpler one-link models. In the U(1) one-link model with effective action
\[
S_{\rm eff}(x)=-\beta\cos x-\ln\bigl[1+\kappa\cos(x-i\mu)\bigr],
\]
CL reproduces standard observables for \(\kappa=0.5\) and moderate parameters, but for larger \(\kappa\) or certain \(\mu\)-windows the estimates deviate strongly; the stated culprit is the non-analyticity of the logarithm near its cut, which makes the drift ill-defined in regions explored by the process [1408.3577]. The same paper also reports a subtler phenomenon: all moments \(\langle x^n\rangle\) can be misestimated even when observables such as \(\langle e^{ix}\rangle\) are reproduced correctly by a Taylor reconstruction from those moments [1408.3577]. This indicates that the real distribution \(P(x,y)\) generated by CLD does not encode the original complex measure in a simple moment-by-moment way.

Boundary terms associated with poles have a distinct temporal behavior. In simple one-pole models, pole-induced boundary terms arise after finite Langevin time once the process reaches the pole neighborhood, but vanish again as Langevin time goes to infinity, whereas boundary terms at infinity may survive in the long-time limit [2111.01609]. This distinguishes transient singular-drift effects from the asymptotic delocalization that is usually associated with wrong convergence.

## 4. Diagnostics and empirical reliability tests

A practical CLD calculation is usually accompanied by explicit diagnostics that test localization, analyticity, and agreement with sign-problem-free limits. These diagnostics are not interchangeable: some probe the formal assumptions behind the Fokker–Planck argument, while others test thermodynamic consistency or continuity in control parameters.

| Diagnostic | Criterion | Representative source |
|---|---|---|
| Continuity in \(\mu^2\) | Values from \(\mu^2\to0^-\) and \(\mu^2\to0^+\) should coincide | [1005.3468] |
| Hot/cold start at \(\mu=0\) | Complexified and purely real starts should converge to the same result | [1005.3468] |
| Drift-decay test | For large \(u\), \(P(u)\lesssim \exp(-c\,u)\) | [2509.13314] |
| Configurational temperature | \(\beta_{\rm M}=\langle|\nabla S|^2\rangle/\langle\nabla^2 S\rangle\) should match the input \(\beta\) | [2509.13314] |
| Consistency conditions | \(\langle \tilde L O\rangle=0\) for a basis of holomorphic \(O\) | [1306.3075] |
| Determinant winding | Monitor \(\det M(t)\) and require \(R_{\det}=1\) | [1309.4335] |

In the three-dimensional XY model, continuity around \(\mu^2=0\), comparison with the world-line formalism, hot/cold starts at \(\mu=0\), the width
\[
W^2=\Bigl\langle \frac1\Omega\sum_x(\phi_x^I)^2\Bigr\rangle-\Bigl\langle \frac1\Omega\sum_x\phi_x^I\Bigr\rangle^2,
\]
and the distribution of the maximal drift \(K^{\max}\) were all used to diagnose failure [1005.3468]. These tests show that CLD works in the ordered phase at larger \(\beta\), but fails in the disordered phase at smaller \(\beta\), and that the failure is not driven by sign-problem severity; it is already visible at \(\mu=0\), where no sign problem exists [1005.3468][1009.5838].

The configurational-temperature diagnostic adds a thermodynamic consistency test. For the 3D XY model one defines
\[
T_{\rm conf}
=
\frac{\langle |\nabla S[\phi]|^2\rangle}{\langle \nabla^2 S[\phi]\rangle},
\qquad
\beta_{\rm M}=1/T_{\rm conf},
\]
and correct sampling requires \(\beta_{\rm M}=\beta\) [2509.13314]. In the ordered phase, \(\beta_{\rm M}\) matches the input \(\beta\), the action density is continuous across \(\mu^2=0\), and the drift histograms fall exponentially. In the disordered phase at real \(\mu\), \(\beta_{\rm M}\) deviates strongly from \(\beta\), the action density violates analytic continuation, and \(P(u)\) develops power-law tails [2509.13314].

Taken together, these results establish a recurrent empirical pattern: correct CLD is associated with localization, continuity, and exponentially decaying drift histograms, whereas wrong convergence is associated with broad complex excursions, singular regions, or winding of logarithmic arguments [1005.3468][1309.4335][2509.13314].

## 5. Stabilisation, control, and algorithmic modifications

The most basic stabilization device is adaptive step sizing. In the XY-model implementation one monitors
\[
K_n^{\max}=\max_x |K_x^R(n)+iK_x^I(n)|
\]
and chooses
\[
\epsilon_n=\min\{\bar\epsilon,\bar\epsilon\,\langle K^{\max}\rangle/K_n^{\max}\},
\]
which suppresses runaway excursions caused by unbounded drifts in the complexified theory [1009.5838][1005.3468]. Because the steps are unequal, observables are then averaged with Langevin-time weights \(\epsilon_n\) [1009.5838]. This cures the numerical problem of runaway solutions in the XY model, but it does not by itself guarantee correctness [1009.5838].

For non-Abelian gauge theories, gauge cooling is the standard control mechanism. Since CLD evolves \(SU(N)\) links into \(SL(N,\mathbb C)\), one introduces complexified gauge transformations chosen to minimize a distance from unitarity, such as
\[
d(U)=\frac{1}{N}\sum_{x,\mu}\mathrm{Tr}\bigl(U_{x,\mu}U_{x,\mu}^\dagger-\mathbf 1\bigr)\ge0
\]
[1311.1056]. Cooling steps alternate with Langevin updates, and adaptive choices of the cooling parameter,
\[
\alpha_{\rm ad}(U)=\frac{\alpha_0}{D(U)},
\]
accelerate the reduction of the unitarity norm and suppress excursions into non-compact directions [1311.1056]. In toy models a force-based adaptive choice \(D_2\) yields a faster decay \(d(t)\sim t^{-3/2}\), while in full gauge theory adaptive cooling keeps \(d(U)\) orders of magnitude smaller than without cooling [1311.1056].

The broader control strategy emphasized in reviews is to combine adaptive step size, cooling, localization diagnostics, and consistency checks. Gauge cooling is effective because it localizes the sampled distribution in the complexified gauge manifold, removes broad “skirts,” and correlates with stable observables in Polyakov-chain models and heavy-quark QCD [1303.6425]. However, gauge cooling and adaptive stepping do not remove ambiguities associated with a multi-valued logarithm; the chiral-random-matrix study states explicitly that one still needs a prescription for the logarithmic branch structure [1309.4335].

Other stabilizing modifications are available. The analysis of effective models shows that kernels or coordinate transformations can generate additional restoring drift and mimic the stabilizing role of a nontrivial Haar measure [1212.5231]. In real-time CLD, modern implicit solvers replace explicit Euler updates by \(\theta\)-schemes. For \(\theta\ge1/2\), these schemes are unconditionally stable; backward Euler–Maruyama is L-stable, and Crank–Nicolson–Maruyama is A-stable and preserves free-theory norms exactly [2105.02735]. The same work interprets implicitness as an intrinsic regularization of the real-time path integral and shows that comparatively large Langevin time steps can be used without runaway behavior [2105.02735].

## 6. Model systems, related frameworks, and scope

The three-dimensional XY model remains a canonical test bed for CLD. Two independent studies found that CLD is reliable at larger \(\beta\) and fails at smaller \(\beta\), with the boundary between good and bad behavior tracking the ordered-to-disordered transition rather than the severity of the sign problem [1005.3468][1009.5838]. The later configurational-temperature study reached the same phase-structure conclusion and added a physics-driven reliability test that complements the Nagata–Nishimura–Shimasaki drift-decay criterion [2509.13314].

The three-dimensional SU(3) spin model provides an important contrast. A criteria-for-correctness analysis found strong indications that CLD yields correct results in this theory [1110.5749]. The subsequent stability analysis argues that the difference relative to the XY model is due to the nontrivial Haar measure, which exerts a stabilizing effect on the complexified dynamics [1212.5231]. In this sense, the group manifold is not a passive kinematic background; it directly shapes the localization properties on which CLD correctness depends.

Chiral random matrix theory at nonzero chemical potential isolates the effect of logarithmic determinants and determinant winding in a setting close in spirit to finite-density QCD [1309.4335]. Large-\(N\) unitary matrix models show that CLD can also reproduce analytically known phase structures, including a series of Gross-Witten-Wadia transitions, Polyakov lines, and quark number density, with excellent agreement at large \(N\) and \(N_f\) [1802.10381]. In supersymmetric quantum mechanics with complex actions, CLD together with drift-distribution and Langevin-operator diagnostics suggests that the method can reliably predict the absence or presence of dynamical supersymmetry breaking when those tests are satisfied [2011.08107].

CLD also has a close but non-identical relationship to Lefschetz-thimble methods. In quartic, U(1), and SU(2) examples, the sampled CL distribution is typically related to the contributing thimble or thimbles, but it remains a genuine two-dimensional probability distribution and may avoid repulsive saddle directions or terminate differently near singular points [1407.2090]. A thimble-motivated modification strategy goes further by altering a model so that the modified integral has a single relevant thimble, after which CLD can reproduce the exact result in cases where the naive implementation fails [1508.04231]. This suggests that thimble geometry can be used not only for interpretation but also for algorithm design.

Across these settings, the central lesson is consistent. CLD is neither a purely formal replacement for importance sampling nor a universally reliable black box. It is a stochastic method whose success depends on the geometry of the complexified manifold, the analytic structure of the drift, and the measured localization properties of the sampled distribution [1110.5749][1303.6425]. Where distributions are localized, boundary terms vanish, and diagnostics are satisfied, CLD can reproduce nontrivial finite-density and real-time observables. Where trajectories probe poles, logarithmic branch cuts, or delocalized regions, the method can converge to stable but incorrect limits [1309.4335][1701.02322].

Source: https://www.emergentmind.com/topics/complex-langevin-dynamics