Panoscopic non-equilibrium fluctuation identity

Juliana Caspers,1, ∗ Karthika Krishna Kumar,2 Clemens Bechinger,2 and Matthias Krüger1, †
1
Institute for Theoretical Physics, Georg-August-Universität Göttingen, 37073 Göttingen, Germany
2
Fachbereich Physik, Universität Konstanz, 78457 Konstanz, Germany
Quantifying and characterizing fluctuations far away from equilibrium is a challenging task. We
introduce and experimentally confirm an identity for a driven classical system, relating the different
non-equilibrium cumulants of the observable conjugate to the driving protocol. The identity is valid
from micro- to macroscopic length scales, and it encompasses the fluctuation dissipation theorem.
We apply it in experiments of a Brownian probe particle confined and driven by an optical potential
arXiv:2502.10175v1 [cond-mat.stat-mech] 14 Feb 2025

and suspended in a nonlinear and non-Markovian fluid.

The fluctuation dissipation theorem [1, 2], connect- from equilibrium.

ing response and fluctuations of equilibrium systems, is The derivative of U with respect to Xt , Ft :=
of fundamental importance for condensed matter, flu- ∂Xt U (xt , Xt ) is the observable conjugate to Xt . E.g., if
ids, plasmas, or electromagnetic fields [3–6]. One of Xt is a position as in our experiments, Ft is (minus) the
its remarkable properties is the validity at any length corresponding force. Ft can be micro- or macroscopic;
scale, be it the nanoscale, as for electric charges, or the for example, let X couple linearly to an observable A(x),
macroscale, as for the macroscopic magnetization. It is U (xt , Xt ) = U (xt , 0)−Xt A(xt ), i.e., Ft = −A(xt ). Thus,
however restricted to the linear regime, i.e., to situations if A(x) is a macroscopic field, such as the macroscopic
close to equilibrium. Most previous research has been magnetization, Ft is macroscopic. If A(x) is the position
largely devoted to determining similar relations for non- of a molecular particle, Ft is microscopic. The following
equilibrium steady states [7–29] as well as for nonlinear remains valid if Xt enters U nonlinearly.
responses [30–47]. A typical observation in the found re- The statistical properties of Ft in this non-equilibrium
lations is the explicit appearance of microscopic details situation are encoded in its cumulants and its correlations
– sometimes referred to as frenetic components [48, 49] – with another state observable Bt = B(xt , Xt ), which we
often hampering a model independent formulation as well aim to study here. The well known fluctuation dissipa-
as systematic change of length scales such as coarse grain- tion theorem (FDT) connects the covariance in the un-
ing to macroscopic scales [42, 43, 45]. As a consequence, perturbed system and the first moment of B under weak
experimental test and application of such relations has driving [1, 2],
indeed been successful for systems with a small number Z t
of accessible Markovian degrees of freedom [47, 50–53]. β2 ds Ẋs ⟨Bt ; Fs ⟩eq = β [⟨Bt ⟩ − ⟨Bt ⟩eq ] + O(Ẋ 2 ),
In this letter, we introduce and experimentally con- −∞

firm an identity for a driven classical system which relates (1)

the different non-equilibrium cumulants of the observable with β = 1/kB T and Boltzmann constant kB . ⟨. . . ; . . . ⟩
conjugate to the driving protocol, up to a certain order in denotes the 2nd cumulant, and similary for higher orders
driving velocity. We demonstrate that (i) this identity is below.
valid from micro- to macroscopic length scales, (ii) that it In an accompanying manuscript [54] we derive identi-
is model independent, and (iii) that it encompasses the ties connecting the non-equilibrium cumulants of Ft and
fluctuation dissipation theorem. We apply it in an ex- Bt to different orders. These give rise to the following
perimental many body system of a Brownian probe par- series involving the mentioned cumulants [54],
ticle interacting with worm-like micelles and confined and
driven by an optical potential. In these experiments, we 2
Z t
demonstrate that the third force cumulant quantifies the β ds Ẋs ⟨Bt ; Fs ⟩ = β [⟨Bt ⟩ − ⟨Bt ⟩eq ]
−∞
deviation from the fluctuation dissipation theorem in sec- t t
β3
Z Z
ond order in driving. Notably, our theoretical predictions + ds ds′ Ẋs Ẋs′ ⟨Bt ; Fs ; Fs′ ⟩
demonstrate that the form of the FDT remains valid for 2 −∞ −∞ (2)
purely Gaussian observables within the displayed order. 4 Z t Z t Z t
β
− ds ds′ ds′′ Ẋs Ẋs′ Ẋs′′
Consider a classical system of stochastic degrees xt at 6 −∞ −∞ −∞
time t, (weakly) coupled to a heat bath at temperature T . × ⟨Bt ; Fs ; Fs′ ; Fs′′ ⟩ + O(Ẋ 4 ),
The system’s potential energy U depends on xt and on a
time dependent deterministic protocol Xt , i.e., U (xt , Xt ). Eq. (2) is, as indicated, correct up to fourth order in
The system is prepared in equilibrium at time t → −∞, driving Ẋt under the assumption of local detailed bal-
with protocol value X−∞ = Xt , for simplicity [54]. The ance [55, 56]. Expanding Eq. (2) to first order yields
time dependence of the protocol drives the system away FDT in Eq. (1), so that it is included in Eq. (2). To
 2

higher orders, first and second cumulants do not fulfill

FDT, and Eq. (2) quantifies their difference in terms
of third and fourth cumulants of F and B. Notably,
the second to fourth lines of Eq. (2) vanish for purely
Gaussian distributed F and B, so that first and second
cumulants obey FDT to the given non-equilibrium or-
der. While Eq. (2) suggests a continuation with higher
order terms [54], we have derived it to the order given
in Eq. (2). It is important to note that in Eq. (2) the
protocol Ẋ appears as prefactors as well as in the non-
equilibrium cumulants themselves.
We exploit Eq. (2) with experiments of Brownian par-
ticles interacting with micellar fluid. Specifically, we
use silica particles of diameter ∼ 1 µm suspended in
a 5 mM equimolar solution of cetylpyridinium chloride
monohydrate (CPyCl) and sodium salicylate (NaSal). At
concentrations above the critical micellar concentration
(≳ 4 mM), this fluid is known to form giant worm-like mi-
celles leading to a viscoelastic nonlinear behavior at am-
bient temperatures [57], see SM. At 5 mM, we determine
the relaxation time of the fluid from microrheological re-
coil experiments to be ∼ 3 s [58]. A small amount of silica
particles is added to the micellar solution which is con-
tained in a rectangular capillary with 100 µm height and
kept at a temperature of 25 ◦ C. This sample is placed on
a custom-built optical tweezer setup that uses a Gaus-
sian laser beam of wavelength 532 nm and a 100× oil
immersion objective (NA = 1.45). The laser beam yields FIG. 1. (a) Asymmetric optical potential U (x) =
a potential U (xt − Xt ) as shown in Fig. 1(a), centered at −kB T ln P (x) felt by the probe particle (inset sketch), with
P (x) the probability distribution with the trap at rest. (b)
Xt , trapping one of the silica particles with coordinate
Mean force β⟨Ft ⟩, (c) force covariance β 2 X̂ω⟨F̃t , Ft ⟩, and
xt . Ft = ∂Xt U (xt − Xt ) is thus the force acting on the
(d) third cumulant β 2 (X̂ω)2 ⟨F̃t ; F̃t ; Ft ⟩eq /2, as functions of
particle by the trapping potential (or vice versa). As the time, for driving frequency ω = 8.4 rad/s and amplitudes
micellar degrees do not couple to X, they do not enter X̂ = {0.03, 0.06, 0.08, 0.09, 0.14}µm as labeled. T = 2π . (e)
ω
Ft explicitly, and how they enter U is not important. We Force covariance (solid line), mean force (dotted line), and
consider B ≡ F , and made the potential asymmetric, to sum of mean force and third force cumulant (dashed line,
obtain a finite second order response of β⟨Ft ⟩. This al- Eq. (4)) for X̂ = 0.14 µm.
lows to test Eq. (2) to second order in our experiments,
and it is achieved by a controlled lateral displacement of
the vertically incident laser beam from the center of the have passed. Prior to each non-equilibrium protocol, we
objective lens (see SM). recorded particle trajectories for another 1000s with Xt
To apply the driving protocol, the sample cell is moved, at rest. This equilibrium data were used to check that
while the optical trap remains stationary in our exper- the form of U does not vary between measurements, and
iments. This is achieved using a piezo-driven stage on also to obtain the force cumulants under equilibrium con-
which the sample is mounted and translated in an os- ditions.
cillating manner relative to the trap. In the fluid’s rest With the protocol of Eq. (3), Eq. (2) takes the form
frame, this yields a periodic motion of the potential min-
imum Xt , i.e., the protocol, β 3 X̂ 2 ω 2
β 2 X̂ω⟨F̃t ; Ft ⟩ = β⟨Ft ⟩ + ⟨F̃t ; F̃t ; Ft ⟩eq
2 (4)
Xt = X̂ sin(ωt), (3)
+ O((X̂ω)3 ),
with amplitude X̂ and frequency ω. Particle trajecto-
ries are recorded with a frame rate of ∼ 150 Hz using a where
Rt the tilde denotes cosine transform, i.e., F̃t ≡
video camera and particle positions are determined using −∞
ds cos(ωs)Fs . We restrict the analysis to the low-
a custom MATLAB algorithm. To yield sufficient statis- est nontrivial, i.e., second, order and expanded Eq. (2)
tics, each protocol (X̂, ω) was measured over 1400s. We accordingly, also using ⟨Ft ⟩eq = 0. The cumulants in
allowed the system to reach a steady state by record- Eq. (4) depend on time t in a periodic manner, as shown
ing trajectories only after at least 5 oscillation periods in Figs. 1(b)-(d), for ω = 8.4 rad/s [59] and driving ampli-
 3

tudes ranging from X̂ = 0.03 µm (light) to X̂ = 0.14 µm

(dark). Fig. 1(b) shows the mean force, which, as ex-
pected for a driven oscillator, is a periodic function with
period T = 2π ω . For the smallest amplitude shown, the
mean force is nearly harmonic with frequency ω, as ex-
pected from linear response. With growing amplitude,
higher harmonics occur, as expected from nonlinear re-
sponse. This asymmetric system shows second order re-
sponse with frequencies of 2ω and 0ω.
Fig. 1(c) shows the force covariance for the same pa-
rameters and color code. For small amplitude X̂, the
curves in Figs. 1(b) and (c) are equal within experimen-
tal accuracy, as analyzed in detail below. For larger driv-
ing amplitude, the force covariance develops higher har-
monics with signatures of second order. Very little is
known about the properties of such non-equilibrium fluc-
tuations, and quantifying these is difficult. It is notable
that the curves in Fig. 1(c), for larger amplitudes, devi- FIG. 2. Coefficients Am corresponding to harmonics with
ate from Fig. 1(b), the deviation which we claim to be frequency mω, as a function of driving amplitude X̂, for
quantified by Eq. (4). ω = 8.4 rad/s. Top panel: m = 1, demonstrating FDT.
Fig. 1(d) shows the third cumulant of force for the same Lower panels: Difference of first and second cumulants (data
parameters and color code. We have here restricted to points), and third cumulant (lines) for m = 0 and m = 2.
(1) (2)
the equilibrium cumulant as it appears in Eq. (4), multi- The agreement confirms Eq. (4). ∆A2 ≡ [(A2 )2 + (A2 )2 −
(1) (2) (1) (2) 1/2
plied by (X̂ω)2 . The curves in Fig. 1(d) thus differ only 2A2 A2 cos(ϕ2 − ϕ2 )] [60].
because of the factor (X̂ω)2 . They thus scale quadrati-
cally in driving velocity and only show frequencies of 2ω
and 0ω. this panel is expected from the fluctuation dissipation
Eq. (4) states that, in the shown range of amplitudes, theorem.
the curves in Fig. 1(c) are given by the sum of the curves The center and lower panels in Fig. 2 show the orders
in Figs. 1(b) and (d). For X̂ = 0.14 µm the respective m = 0 and m = 2, respectively. Specifically, these panels
summed curve is shown as a dashed line together with present the difference of first and second cumulants in
the mean force and the force covariance in Fig. 1(e). The Eq. (4) (data points), i.e., β⟨Ft ⟩−β 2 X̂ω⟨F̃t ; Ft ⟩, together
agreement is convincing and a confirmation of Eq. (4). with the third cumulant (line), −β 3 (X̂ω)2 ⟨F̃t ; F̃t ; Ft ⟩eq /2
To test this prediction systematically, we dissect the [60]. The latter is shown as a parabola with curvature
curves in Figs. 1 (b), (c) and (d) into the contributions obtained from the third force cumulant at equilibrium.
from harmonics with frequencies 0ω, ω, and 2ω, respec- The data points in this graph thus quantify the deviation
tively, i.e., we expand the cumulant of order n into har- from FDT, with the line giving the prediction of Eq. (4)
monics with frequency mω, for this deviation. The agreement is convincing for both
m = 0 and m = 2, supporting the validity of Eq. (4).
∞
X As the data in the top panel of Fig. 2 grow linearly and
β n (X̂ω)n−1 ⟨(F̃t ; )n−1 Ft ⟩ = A(n) (n)
m sin(mωt + ϕm ), the ones in the center and lower panels grow quadrati-
m=0 cally with X̂, we fit a line and a parabola to obtain the
(5) respective slope and curvature for each m. The obtained
values – divided by the respective power in ω – are shown
(n) (n)
where the coefficients Am depend on X̂ω. We set ϕ0 ≡ in Fig. 3 as a function of frequency ω. We observe con-
π/2 for consistency. Eq. (4), projected on the harmonic of vincing agreement for the measured frequencies further
order m, yields relations between coefficients and phases suporting Eq. (4). In tendency, the coefficients decrease
for each m, which we can test. with increasing ω, as expected for an overdamped system.
(n) Noteably, the statistical accuracy of the data points de-
Fig. 2 shows the coefficients Am as a function of driv-
ing amplitude X̂. The top panel gives the order m = 1, creases with decreasing ω; With smaller ω, longer trajec-
which is seen to be linear in X̂ for the range shown, as ex- tories are required as the period of the cosine transform
pected from linear response. The graph shows the mean increases.
force (data points) as well as the force covariance (line), Fig. 4 provides the final test of Eq. (4), namely the
(n)
the latter evaluated from equilibrium trajectories. Non- phases ϕm of Eq. (5). These hardly depend on driv-
equilibrium contributions to the covariance are small for ing amplitude and the shown data are averaged over the
m = 1 and the shown range of X̂. The agreement in measured values of X̂. The top panel of Fig. 4 shows
 4

the order m = 1, i.e., the linear response, with convinc-

ing agreement. The phase angle for m = 1 is small for
the frequencies measured, indicating that the force Ft is
almost in phase with the protocol Xt , as for an elastic
material. The black curve, extracted from equilibrium
data, shows that with smaller ω the phase increases, pre-
sumably reaching π/2 in the limit of ω → 0. The slow
increase with decreasing ω displays the slow nature of the
investigated system.
The lower panel shows the phase for m = 2, i.e., to
second order. While the agreement between the line and
the data confirms Eq. (4), the graph shows that the dif-
ference of phases of first and second cumulants are rather
small. In other words, first and second cumulants deviate
noticeable in amplitude, seen in Fig. 2, but not so much
in phase.
We presented and tested a non-equilibrium fluctuation
FIG. 3. Coefficients Am , normalized as labeled, as a func- identity for a driven classical system, emphasizing the
tion of frequency ω. The agreement in the top panel confirms validity on various length scales. Indeed, such relations
FDT, the agreement in the lower panels confirms Eq. (4). are necessary, e.g., for a systematic coarse graining of
Each data point is obtained from averaging over driving am- non-equilibrium systems. The identity is confirmed for
plitudes taking the respective scaling of Am with X̂ into ac- experiments of a Brownian particle interacting with a
count (see SM). Error bars are obtained from partitioning
complex surrounding. Future work can explore other sys-
trajectories into two pieces (data points) and from the stan-
dard deviation between separate series of measurements (grey tems, aim at the next terms in Eq. (2), or investigate a
area). possible quantum version.
This project was funded by the Deutsche Forschungs-
gemeinschaft (DFG), Grant No. SFB 1432 (Project ID
425217212)—Project C05.

∗
j.caspers@theorie.physik.uni-goettingen.de
†
matthias.kruger@uni-goettingen.de
[1] H. B. Callen and T. A. Welton, Phys. Rev. 83, 34 (1951).
[2] R. Kubo, Rep. Prog. Phys. 29, 255 (1966).
[3] L. D. Landau and E. M. Lifshitz, Statistical Physics
(Pergamon, 1980).
[4] A. G. Sitenko, Electromagnetic Fluctuations In Plasma
(Academic Press, New York, 1967).
[5] S. M. Rytov, I. A. Kravtsov, and V. Tatarskii, Principles
of Statistical Radiophysics (Springer-Verlag, 1987).
[6] J. P. Hansen and I. R. McDonald, Theory of Simple Liq-
uids: with Applications to Soft Matter (Elsevier Science,
2013).
[7] L. F. Cugliandolo, D. S. Dean, and J. Kurchan, Phys.
Rev. Lett. 79, 2168 (1997).
[8] D. Ruelle, Phys. Lett. A 245, 220 (1998).
FIG. 4. Phase angles ϕ1 and ϕ2 of the harmonics in Eq. (5), [9] A. Crisanti and F. Ritort, J. Phys. A: Math. Gen. 36,
as functions of ω. Top panel shows phases of linear re- R181 (2003).
sponse. Lower panel shows the phase of second order re- [10] T. Harada and S.-i. Sasa, Phys. Rev. Lett. 95, 130602
sponse, comparing the contributions of the terms in Eq. (4). (2005).
A2
(1) (1) (2)
sin ϕ2 −A2
(2)
sin ϕ2 [11] T. Speck and U. Seifert, EPL 74, 391 (2006).
∆ϕ2 ≡ arctan (1) (1) (2) (2) , [60]. Error bars are [12] J. M. Deutsch and O. Narayan, Phys. Rev. E 74, 026112
A2 cos ϕ2 −A2 cos ϕ2
obtained from partitioning trajectories into two pieces. Grey (2006).
error band is obtained as the standard deviation between sep- [13] R. Chetrite, G. Falkovich, and K. Gawedzki, J. Stat.
arate series of measurements. Mech. 2008, P08005 (2008).
[14] U. M. B. Marconi, A. Puglisi, L. Rondoni, and A. Vulpi-
ani, Phys. Rep. 461, 111 (2008).
 5

[15] K. Saito, Europhys. Lett. 83, 50006 (2008). [40] E. Wang and U. Heinz, Phys. Rev. D 66, 025008 (2002).
[16] M. Baiesi, C. Maes, and B. Wynants, Phys. Rev. Lett. [41] D. Andrieux and P. Gaspard, J. Stat. Mech. 2007,
103, 010602 (2009). P02006 (2007).
[17] J. Prost, J.-F. Joanny, and J. M. R. Parrondo, Phys. Rev. [42] M. Colangeli, C. Maes, and B. Wynants, J. Phys. A:
Lett. 103, 090601 (2009). Math. Theor. 44, 095001 (2011).
[18] T. Harada, Phys. Rev. E 79, 030106 (2009). [43] U. Basu, M. Krüger, A. Lazarescu, and C. Maes, Phys.
[19] M. Krüger and M. Fuchs, Phys. Rev. Lett. 102, 135701 Chem. Chem. Phys. 17, 6653 (2015).
(2009). [44] U. Basu, L. Helden, and M. Krüger, Phys. Rev. Lett.
[20] U. Seifert and T. Speck, EPL 89, 10007 (2010). 120, 180604 (2018).
[21] M. Baiesi, C. Maes, and B. Wynants, Proc. R. Soc. A: [45] C. Maes, Front. Phys. 8 (2020).
Math. Phys. Eng. Sci. 467, 2792 (2011). [46] J. Caspers and M. Krüger, J. Chem. Phys. 161, 124109
[22] L. F. Cugliandolo, J. Phys. A: Math. Theor. 44, 483001 (2024).
(2011). [47] L. Helden, U. Basu, M. Krüger, and C. Bechinger, EPL
[23] G. Verley, R. Chétrite, and D. Lacoste, J. Stat. Mech. 116, 60003 (2016).
2011, 10025 (2011). [48] C. Maes, J. Stat. Phys. 154, 705 (2014).
[24] B. Altaner, M. Polettini, and M. Esposito, Phys. Rev. [49] C. Maes, Phys. Rep. 850, 1 (2020).
Lett. 117, 180601 (2016). [50] J. R. Gomez-Solano, A. Petrosyan, S. Ciliberto,
[25] E. Lippiello, M. Baiesi, and A. Sarracino, Phys. Rev. R. Chetrite, and K. Gawedzki, Phys. Rev. Lett. 103,
Lett. 112, 140602 (2014). 040601 (2009).
[26] W. Wu and J. Wang, Front. Phys. 8 (2020). [51] V. Blickle, T. Speck, C. Lutz, U. Seifert, and
[27] L. Caprini, J. Stat. Mech. 2021, 063202 (2021). C. Bechinger, Phys. Rev. Lett. 98, 210601 (2007).
[28] M. Baldovin, L. Caprini, A. Puglisi, A. Sarracino, and [52] J. Mehl, V. Blickle, U. Seifert, and C. Bechinger, Phys.
A. Vulpiani, in Nonequilibrium Thermodynamics and Rev. E 82, 032401 (2010).
Fluctuation Kinetics: Modern Trends and Open Ques- [53] J. R. Gomez-Solano, A. Petrosyan, S. Ciliberto, and
tions (Springer International Publishing, 2022). C. Maes, J. Stat. Mech. 2011, P01008 (2011).
[29] M. K. Johnsrud and R. Golestanian, Generalized Fluc- [54] J. Caspers and M. Krüger, tba (2025), Note: In the
tuation Dissipation Relations for Active Field Theories accompanying manuscript, xt (not Xt ) denotes the po-
(2024), arXiv:2409.14977. sition of the control parameter at time t. The notation
[30] R. Kubo, J. Phys. Soc. Jpn. 12, 570 (1957). chosen here allows easier presentation.
[31] T. Yamada and K. Kawasaki, Prog. Theor. Phys. 38, [55] C. Maes, SciPost Phys. Lect. Notes , 32 (2021).
1031 (1967). [56] The expansion in Eq. (2) R t suggests the dimensionless ex-
[32] D. J. Evans and G. Morriss, Statistical Mechanics of pansion parameter βF t−τ ds ẋs with cumulant relax-
Nonequilibrium Liquids, 2nd ed. (Cambridge University ation time τ .
Press, 2008). [57] M. E. Cates and S. J. Candau, J. Phys.: Condens. Matter
[33] M. Fuchs and M. E. Cates, J. Phys.: Condens. Matter 2, 6869 (1990).
17, S1681 (2005). [58] J. R. Gomez-Solano and C. Bechinger, New J. Phys. 17,
[34] T. Holsten and M. Krüger, Phys. Rev. E 103, 032116 103032 (2015).
(2021). [59] ω is determined from the power spectral density, thus
[35] I. Oppenheim, Prog. Theor. Phys. Supp. 99, 369 (1989). carrying an error depending on the length of the mea-
[36] J.-P. Bouchaud and G. Biroli, Phys. Rev. B 72, 064204 surement.
(2005). [60] As the phases ϕ2
(1)
and ϕ2
(2)
may differ,
[37] E. Lippiello, F. Corberi, A. Sarracino, and M. Zannetti, the coefficient ∆A2 of the difference of first
Phys. Rev. E 78, 041120 (2008). and second cumulants is found via ∆A2 ≡
[38] V. Lucarini and M. Colangeli, J. Stat. Mech. 2012,
r  2  2  
(1) (2) (1) (2) (1) (2)
P05013 (2012). A2 + A2 − 2A2 A2 cos ϕ2 − ϕ2 .
[39] G. Diezemann, Phys. Rev. E 85, 051502 (2012).