ARTICLE

https://doi.org/10.1038/s42005-021-00669-2 OPEN

Noether’s theorem in statistical mechanics

Sophie Hermann 1✉ & Matthias Schmidt 1✉

Noether’s calculus of invariant variations yields exact identities from functional symmetries.
The standard application to an action integral allows to identify conservation laws. Here we
rather consider generating functionals, such as the free energy and the power functional, for
equilibrium and driven many-body systems. Translational and rotational symmetry opera-
tions yield mechanical laws. These global identities express vanishing of total internal and
total external forces and torques. We show that functional differentiation then leads to
hierarchies of local sum rules that interrelate density correlators as well as static and time
1234567890():,;

direct correlation functions, including memory. For anisotropic particles, orbital and spin
motion become systematically coupled. The theory allows us to shed new light on the spatio-
temporal coupling of correlations in complex systems. As applications we consider active
Brownian particles, where the theory clarifies the role of interfacial forces in motility-induced
phase separation. For active sedimentation, the center-of-mass motion is constrained by an
internal Noether sum rule.

Physik II, Physikalisches Institut, Universität Bayreuth, Bayreuth, Germany. ✉email: Sophie.Hermann@uni-bayreuth.de; Matthias.
1 Theoretische

Schmidt@uni-bayreuth.de

COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys 1

ARTICLE COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2

E
mmy Noether’s 1918 Theorems for Invariant Variation
Problems1,2, as applied to action functionals both in
particle-based and field-theoretic contexts, form a staple of
our fundamental description of nature. The formulation of energy
conservation in general relativity had been the then open and
vexing problem, that triggered Hilbert and Klein to draw Noether Fig. 1 Illustration of the three types of dynamical transformations
into their circle, and she ultimately solved the problem3. Her deep considered. The system is spatially displaced by ϵ = const at all times
insights into the relationship of the emergence and validity of (green dashed), analogously to the operation in equilibrium. The system is
conservation laws with the underlying local and global symme- dynamically displaced by ϵðt0 Þ, such that the spatial displacement vanishes
tries of the system has been exploited for over a century. at the boundaries of the considered time interval, ϵ(0) = ϵ(t) = 0 (cyan
While Noether’s work has been motivated by the then ongoing solid). The system is displaced instantaneously only at the latest time t,
developments in general relativity, being a mathematician, she such that the differential displacement is ϵ_ dt (purple dotted).
has formulated her theory in a much broader setting than given
by the specific structure of the action as a space-time integral over crystal deformations35, and –prominently– interfaces of
a Lagrangian density, as formulated by Hilbert in 1916 for Ein- liquids36–40. A range of further techniques besides DFT was used
stein’s field equations. Her work rather applies to functionals of a in this context, including integral equation theory32,36,39, mode-
much more general nature, with only mild assumptions of ana- coupling theory41, and Mori-Zwanzig equations33,35.
lyticity and careful treatment of boundary conditions of inte- Much of very current attention in Statistical Physics is devoted
gration domains. to nonequilibrium and active systems that are driven in a con-
In Statistical Physics, the use of Noether’s theorems is sig- trolled way out of equilibrium, such as e.g. active Brownian
nificantly more scarce, as opposed to both classical mechanics particles42–44 and magnetically controlled topological transport of
and high energy physics. Notable exceptions include the square- colloids45–47. The power functional (variational) theory48 (PFT)
gradient treatment of the free gas–liquid interface, cf. Rowlinson offers to obtain a unifying perspective on nonequilibrium pro-
and Widom’s enlightening description4 of van der Waals’ pro- blems such as the above. In PFT the (time-dependent) density
totypical solution5. In a striking analogy, the square gradient distribution is complemented by the (time-dependent) current
contribution to the free energy is mapped onto kinetic energy of distribution as a further variational field. A rigorous extremal
an effective particle that traverses in time between two potential principle determines the motion of the system, on the one-body
energy maxima of equal height. Exploiting energy conservation in level of correlation functions. The concept enabled to obtain a
the effective system yields a first integral, which constitutes a fundamental understanding and quantitative description of a
nontrivial identity in the statistical problem. This reasoning has significant array of nonequilibrium phenomena, such as the
been generalized to the delicate problem of the three-phase identification of superadiabatic forces49, the treatment of active
contact line that occurs at a triple point of a fluid mixture6,7. Brownian particles50–53, of viscous54, structural55,56 and flow
While these treatments strongly rely on the square-gradient forces56. Crucially, the DFT remains relevant for the description
approximation, Boiteux and Kerins also developed a method that of nonequilibrium situations, via the adiabatic construction48,49,
they refer to as variation under extension, which permitted them which captures those parts of the dynamics that functionally
to treat more general cases8. depend on the density distribution alone, and do so instanta-
Evans has derived a number of exact sum rules for inhomo- neously. Both equilibrium DFT and nonequilibrium PFT provide
geneous fluids in his pivotal treatment of the field9. While not formally exact variational descriptions of their respective realm of
spelling out any connection to Noether’s work, he carefully Statistical Physics. While action integrals feature in neither for-
examines the effects of spatial displacements on distribution mulation, the relevant functionals do fall into the general class of
functions. This shifting enables him, as well as Lovett et al.10 and functionals that Noether considered in her work.
Wertheim11 in earlier work, to identify systematically the effects Here we apply Noether’s theorem to Statistical Physics. We
that result from the displacement and formulate these as highly first introduce the basic concepts via treating spatial translations
nontrivial interrelations (“sum rules”) between correlation func- for both the partition sum and for the free energy density func-
tions. This approach was subsequently generalized to higher than tional. Considering the symmetries of the partition sum does not
two-body direct12 and density13 correlation functions and the require to engage with density functional concepts; the elemen-
relationship to integral equation theory was addressed14,15. tary definition suffices. We demonstrate that this approach is
Considering also rotations Tarazona and Evans16 have addressed consistent with the earlier work in equilibrium9–16, and that it
the case of anisotropic particles, where their sum rules correct enables one to go, with relative ease, beyond the sum rules that
earlier results by Gubbins17. The exploitation of the fundamental these authors formulated. In nonequilibrium, we apply the same
spatial symmetries9–16 appears to be intimately related to Noe- symmetry operations to the time-dependent case and obtain
ther’s thinking. This is no coincidence, as Evans’ classical density novel exact and nontrivial identities that apply for driven and
functional approach (DFT) is variational as is the general pro- active fluids. The three different types of time-dependent shifting
blem that she addresses. are illustrated in Fig. 1. The resulting sum rules are different from
DFT constitutes a powerful modern framework for the the nonequilibrium Ornstein–Zernike (NOZ) relations57,58, but
description of a broad range of interfacial, adsorption, solvation, they possess an equally fundamental status. We also consider the
and phase phenomenology in complex systems9,18–20. Examples more general case of anisotropic interparticle interactions and
of recent pivotal applications include the treatments of treat rotational invariance both in and out of equilibrium. To
hydrophobicity21–27 and of drying23,24,26, electrolytes near illustrate the theory we apply it to both passive and active phase
surfaces28, dense fluid structuring as revealed in atomic force coexistence as well as to active sedimentation under gravity.
microscopy29, thermal resistance of liquid–vapor interfaces30, and
layered freezing in confined colloids31. Xu and Rice31 have used
the sum rules of Lovett et al.10 and Wertheim11 (LMBW) to carry Results and discussion
out a bifurcation analysis of the confined fluid state. The sum Adiabatic state. We start with an initial illustration of Noether’s
rules were instrumental for investigating a range of topics, such as concept as applied to the grand potential Ω. We consider spatial
precursors to freezing32, nonideal33 and cluster crystals34, liquid translations of the position coordinate r at fixed chemical

2 COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys

COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2 ARTICLE

distribution, and the average is over the equilibrium distribution at

fixed μ and T; the sum runs over all particles i = 1…N.
Comparing the left and right hand sides of (1) and noticing
that ϵ is arbitrary, we conclude
Z
Fext ¼  drρðrÞ∇V ext ðrÞ ¼ 0;
tot
ð2Þ

where we have defined the total external force Ftot ext using the one-
body fields ρ(r) and Vext(r). It is straightforward to show the
equivalence with the more elementary form
Ftot
ext ¼ h∑ i ∇ i V ext ðr i Þieq , where ∇ i indicates the derivative with
respect to ri. Clearly, (2) expresses the vanishing of the total
external force (consider e.g. the gravitational weight of an
equilibrium colloidal sediment being balanced by the force that
the lower container wall exerts on the particles).
Equation (2) was previously obtained by Baus13. Here we have
identified it as a Noether sum rule for the case of spatial
displacement of Ω[Vext]. We can generate local sum rules by
observing that (2) holds for any form of Vext(r) and that hence
Fig. 2 Illustrations of the effects induced by shifting in equilibrium. a In Vext(r) → ρ(r) constitutes a functional map (defined by the grand
the presence of external potential Vext(r), the system develops an canonical average h^ρðrÞieq , which features Vext(r) in the
inhomogeneous density profile ρ(r), where r denotes the position
equilibrium many-body probability distribution). We hence
coordinate. b Shifting the external potential by a displacement vector −ϵ
functionally differentiate (2) by V ext ðr0 Þ, where r0 is a new
(green arrow) induces a local change in external potential δVext(r) (black
position variable. The first and the nth functional derivatives
arrow) between the original (solid line) and the shifted external potential
yield, respectively, the identities
(dashed line); the grand potential is invariant, δΩ = 0. c The displaced Z
density profile (dashed line) implies a local change δρ(r) (black arrow) in ∇ρðrÞ ¼  dr0 βH 2 ðr; r0 Þ∇0 V ext ðr0 Þ; ð3Þ
comparison to the initial density profile (solid line), which leaves the
intrinsic free energy unchanged, δF = 0.
n
Z
∑ ∇α H n ¼  drnþ1 βV ext ðrnþ1 Þ∇nþ1 H nþ1 ; ð4Þ
potential μ and fixed temperature T. The system is under the α¼1
influence of a one-body external potential Vext(r), cf. Fig. 2a. We where β = 1/(kBT), with kB indicating the Boltzmann constant,
take Vext(r) to also describe container walls, such that there is no H 2 ðr; r0 Þ ¼ δρðrÞ=δβV ext ðr0 Þ is the two-body correlation func-
need for the system volume as a further thermodynamic variable. tion of density fluctuations, and Hn = δHn−1/δβVext(rn) is its n-
For the moment we only examine systems completely bounded body version9,18. Here position arguments have been omitted for
by external walls. Systems with open boundaries are considered clarity: Hn ≡ Hn(r1…rn), and ∇α indicates the derivative with
below. Clearly the value of the grand potential Ω is independent respect to rα. The variable names r and r0 have been interchanged
of the global location of a system. Hence spatial shifting by a in (3) and ∇0 indicates the derivative with respect to r0 . The
(global) displacement vector ϵ leaves the value of Ω invariant. To derivation of (3) and (4) requires spatial integration by parts.
exploit this symmetry in a variational setting, note that the value Recall that boundary terms vanish as we only consider systems
of Ω depends on the function Vext(r), hence Vext(r) → Ω con- with impenetrable bounding walls.
stitutes a functional map, at given μ and T. Here the grand The sum rule (3) has been obtained by LMBW10,11 and by
potential is defined by its elementary Statistical Mechanics form Evans9 on the basis of shifting considerations. The present
Ω½V ext  ¼ kB Tln Ξ, with the grand partition sum Ξ depending formulation based on Noether’s more general perspective allows
functionally via the Boltzmann factor on Vext(r). The spatial to reproduce (3) with great ease and to generalize to the hierarchy
displacement amounts to the operation Vext(r) → Vext(r + ϵ), cf. (4), as previously obtained by Baus13. Equation (3) has the
Fig. 2b. For small ϵ we can Taylor expand to linear order: Vext(r + interpretation of the density gradient ∇ ρ(r) being stabilized by
ϵ) = Vext(r) + δVext(r), where δVext(r) = ϵ ⋅ ∇ Vext(r) indicates the the action of the external force field, −∇ Vext(r). The effect is
local change of the external potential that is induced by the shift. mediated by βH 2 ðr; r0 Þ, where the correlation of the density
As Ω[Vext] is invariant under the shift (which can be shown by fluctuations is due to the coupled nature of the interparticle
translating all particle coordinates in Ξ accordingly), we have interactions. Equation (4) is the multi-body generalization of this
Z mechanism. Via multiplying (3) by Vext(r), integrating over r, and
δΩ½V ext  using (2), and iteratively repeating this process for all orders, one
Ω½V ext  ¼ Ω½V ext þ δV ext  ¼ Ω½V ext  þ dr ϵ  ∇V ext ðrÞ:
δV ext ðrÞ obtains a multi-body analog of the vanishing external force (2):
Z Z
ð1Þ
dr1 V ext ðr1 Þ ¼ drn V ext ðrn Þ∇α H n ¼ 0; ð5Þ
Here the second equality constitutes a functional Taylor expansion for α = 1…n.
in δVext(r) to linear order, and δΩ[Vext]/δVext(r) indicates the We turn to intrinsic contributions. As Noether’s theorem
functional derivative of Ω[Vext] with respect to its argument, poses no restriction on the type of physical functional, we
evaluated here at the unshifted function Vext(r), i.e. ϵ = 0. It is a consider the intrinsic Helmholtz free energy F[ρ] as a
straightforward elementary exercise9,18 to show via explicit cal- functional of the density profile as its natural argument. Here
culation that δΩ[Vext]/δVext(r) = ρ(r), where ρðrÞ ¼ a functional Legendre transform9,18 yields F[ρ] = Ω[Vext]
h∑i δðr  ri Þieq is the microscopically resolved one-body density − ∫dr(Vext(r) − μ)ρ(r). Crucially, F[ρ] is independent of
profile. Here ri indicates the position of particle i = 1…N, with N Vext(r), and its excess (over ideal gas) contribution Fexc[ρ] is
being the total number of particles, δ(⋅) indicates the Dirac specific to the form of the interparticle interaction potential u

COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys 3

ARTICLE COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2

(r N); here we use the shorthand r1…rN ≡ rN. The full intrinsic distribution function. The total internal force in such a general
free energy functional consistsRof a sum of ideal gas and excess situation is
contributions, i.e. F½ρ ¼ kB T drρðrÞ½ln ðρðrÞΛD Þ  1 þ F exc ½ρ,
int ¼ h ∑ ∇i uðr Þi ¼ 0;
Ftot N
where Λ is the (irrelevant) thermal de Broglie wavelength and D is i
ð10Þ
the dimensionality of space.
As u(rN) is globally translationally invariant, Fexc[ρ] will not where the average is taken over the nonequilibrium many-body
change its value when evaluated at a spatially displaced density, ρ probability distribution at time t. We have hence proven that the
(r + ϵ) = ρ(r) + δρ(r), where δρ(r) = ϵ ⋅ ∇ ρ(r), cf. Fig. 2(c). total internal force vanishes for all times t. In addition, the
Hence in analogy to (1), we obtain Fexc[ρ] = Fexc[ρ + δρ] = particles can possess additional degrees of freedom ωi, i = 1…N,
Fexc[ρ] + ∫dr(δFexc[ρ]/δρ(r))ϵ ⋅ ∇ ρ(r). Again ϵ is arbitrary. As as is the case for the orientation vectors of active Brownian
boundary terms vanish in the considered systems, integration by particles, to which we return after first laying out the setup in
parts yields nonequilibrium.
Z Z
δF exc ½ρ
Ftot ¼  drρðrÞ∇ ¼ drρðrÞf ad ðrÞ ¼ 0; ð6Þ Nonequilibrium states. To be specific, we consider overdamped
ad
δρðrÞ Brownian motion, at constant temperature T and with no
where the first equality expresses the total internal force Ftot hydrodynamic interactions present18, as described by the Smo-
ad ¼
luchowski (Fokker–Planck) equation. The microscopically
h∑i ∇i uðrN Þieq in DFT language. Hence (6) expresses the fact
resolved local internal force field is fint(r, t) = − 〈∑iδ(r − ri)∇iu
that the total internal force vanishes in equilibrium; the more (rN)〉/ρ(r, t), where the average is over the nonequilibrium dis-
general time-dependent case is treated below. The functional tribution (which evolves in time according to the Smoluchowski
derivatives of Fexc[ρ] constitute direct correlation functions9,18,59, equation) at time
with the lowest order being the one-body direct correlation R t. The total internal force is then the spatial
integral Ftot ¼ drρðr; tÞf int ðr; tÞ. Applying Noether’s theorem to
function c1(r) = − δβFexc[ρ]/δρ(r). The equilibrium ("adiabatic”) int
the nonequilibrium case requires to have a variational descrip-
force field is simply fad(r) = kBT ∇ c1(r). This one-body force field tion, as is provided by PFT48. Here the variational fields are the
arises from the interparticle forces that all other particles exert on time-dependent density profile ρ(r, t) and the time-dependence
the particle that resides at position r. one-body current J(r, t) = 〈∑iδ(r − ri)vi〉, where vi(rN, t) is the
From the global internal Noether sum rule (6), we can obtain configurational velocity of particle i. The microscopically resolved
local sum rules by observing that (6) holds for all ρ(r) and hence average velocity profile is v(r, t) = J(r, t)/ρ(r, t). PFT ascertains the
that its functional derivative with respect to ρ(r) vanishes splitting f int ðr; tÞ ¼ f ad ðr; tÞ þ f sup ðr; tÞ, where the adiabatic force
identically, i.e.,
Z field is that in a corresponding equilibrium (“adiabatic”) system
∇c1 ðrÞ ¼ dr0 c2 ðr; r0 Þ∇0 ρðr0 Þ; ð7Þ with identical instantaneous density profile, fad(r, t) = − ∇
δFexc[ρ]/δρ(r, t) and f sup ðr; tÞ is the superadiabatic internal force
n
Z field, obtained as f sup ðr; tÞ ¼ δPexc t ½ρ; J=δJðr; tÞ, where P t ½ρ; J
exc
48
∑ ∇α cn ¼  drnþ1 ρðrnþ1 Þ∇nþ1 cnþ1 ; ð8Þ is the superadiabatic excess free power functional .
α¼1 Crucially, fad(r, t) is a density functional, independent of the flow
where c2 ðr; r0 Þ is the (inhomogeneous) two-body direct correla- in the system, while f sup ðr; tÞ is a kinematic functional, i.e. with
tion function of liquid state theory18; cn ≡ cn(r1…rn) is the n-body dependence on both ρ(r, t) and J(r, t), including memory, i.e.
direct correlation function, defined recursively via cn+1 = δcn/δρ dependence on the value of the fields at times < t. As the local force
(rn+1). As identified by LMBW10,11 and Evans9, (7) expresses the fields split into adiabatic
R and superadiabatic
R Rcontributions,totso do tot
the
conversion of the density gradient, via the two-body direct total forces: Ftotint ¼ drρf int ¼ drρf ad þ drρf sup  Fad þ F sup .
correlations, to the locally resolved intrinsic force field; recall that We have seen above that Ftot ¼ F tot
¼ 0. Hence also
int ad
fad(r) = kBT ∇ c1(r). Via the Noether formalism the correspond- Z Z
ing hierarchy (8) is obtained straightforwardly from repeated δP exc
½ρ; J
Ftot ¼  drρðr; tÞ t
 drρf sup ¼ 0: ð11Þ
functional differentiation12–14 with respect to ρ(r). Note that sup
δJðr; tÞ
similar to the structure of (4), only consecutive terms of order n
and n + 1 are directly coupled in (8). A multi-body version of (6) While the above reasoning required to rely on the many-body level,
is obtained by multiplying (7) with ρ(r), integrating over r, the same result (11) can be straightforwardly obtained in a pure
exploiting (6), and iterating for all orders. The result is: Noetherian way, by considering an instantaneous shift of
Z Z coordinates at time t, i.e. Jðr; tÞ ! Jðr; tÞ  ϵ_ ρðr; tÞ, cf. Fig. 3a
dr1 ρðr1 Þ ¼ drn ρðrn Þ∇α cn ¼ 0; ð9Þ (Fig. 3 gives an overview of the three different types of shifting).
Here ϵ_ is the corresponding instantaneous change in velocity with
for α = 1…n. In the case α = n = 1 we recover (6). vðr; tÞ ! vðr; tÞ  ϵ_ , as obtained by dividing the current by the
The global sum rule (6) of vanishing total internal force can be density profile. Due to the overdamped nature of the Smoluchowski
straightforwardly obtained by more elementary analysis. We dynamics, the internal interactions are unaffected and the shift
exploit translation invariance in this non-functional setting: constitutes a symmetry operation for the generator of the
u(rN) ≡ u(r1 + ϵ…rN + ϵ). Then the derivative with respect to ϵ superadiabatic forces, Pexc t ½ρ; J. Hence the instantaneous current
vanishes, 0 = ∂u(r1 + ϵ…rN + ϵ)/∂ϵ = ∑i∇iu(rN). The latter perturbation δJðr; tÞ ¼ _ϵρðr; tÞ that is generated by the invariance
leads to Pexc t ½ρ; J ¼ P t ½ρ; J þ δJ ¼ P t ½ρ; J
exc exc
expression follows from the chain rule and constitutes the total transformation
R
internal force (up to a minus sign), which hence vanishes for each drðδPexc
t ½ρ; J=δJðr; tÞÞ  _
ϵ ρðr; tÞ. As _
ϵ is arbitrary, we obtain
microstate rN. The connection to (the many-body version of) (11). Treating the dynamical adiabatic contribution _ ¼
F½ρ
R
Newton’s third law actio equals reactio becomes apparent in drJðr; tÞ  ∇δF exc ½ρ=δρðr; tÞ in the same way, we re-obtain (6).
the rewritten form − ∇αu(rN) = ∑i≠α∇iu(rN), for α = 1…N. The As Noether’s theorem is converse, it allows for alternative
reasoning: the invariance of Pexc t ½ρ; J to the instantaneous shift of
thermal equilibrium average is then trivial and on average
ad ¼ 0. This argument is very general and it remains true if
Ftot the current (or analogously of the velocity) can hence be derived
the average is taken over a nonequilibrium many-body from (11) by simply reversing the above chain of arguments.

4 COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys

COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2 ARTICLE

system develops different (bulk) states, e.g. for x → ±∞ as is

relevant for bulk phase separation (see the section below). We
demonstrate that such cases can be systematically treated in the
current framework, by exemplary considering the total internal
force. Then boundary force contributions arise due to an
imbalance of “outside” particles that exert forces on “inside”
particles. The outside particles are per definition excluded from
the accounting of the total internal force exerted by all particles
inside of V. The sum of all interactions between inside particles
vanishes due to the global internal Noether sum rule (10). For
simplicity we restrict ourselves to systems that interact via short-
ranged pairwise central forces, where Fij indicates the force on
particle i exerted by particle j. So only forces exerted from an
inside to an outside particle contribute. The total internal force
0
that acts on V is hence Ftotint ¼ h∑ij Fij i, where the restricted sum
(prime) runs only over those i ∈ V and j 2 V,  where V  indicates
the complement of V. The total internal force between particles
inside of V, i.e. i ∈ V and j ∈ V, vanishes due to (10). We then
Fig. 3 Illustrations of the effects induced by shifting in nonequilibrium. int via inserting the identity ∫drδ(r) = 1 twice into the
rewrite Ftot
All transformations affect both the density profile ρ(r, t) (blue) and the average. Then the restrictions of the sums can be transferred to
current profile J(r, t) (yellow), while the superadiabatic excess power restrictions on the spatial integration domains. As a result the
functional is invariant, δPexc ¼ 0. Here t indicates the time and r is the total internal force acting on V can be expressed via correlation
t
position coordinate. a An instantaneous spatial shift by ϵ_ at time t induces a functions as
current change δJ ¼ _ϵρ. b A static shift by ϵ = const is applied at all times Z Z
t. c A time-dependent shift ϵðt0 Þ is applied between initial time 0 and final Ftot
int ¼ dr dr0 h ∑ δðr  ri Þδðr0  rj ÞFij i ð14Þ
time t of the considered time interval. V 
V i;j≠i

We can generate nonequilibrium sum rules by differentiating Z Z

the Noether identity (11) with respect to J(r, t), which yields ¼ dr dr0 ρðrÞρðr0 Þgðr; r0 Þ∇ϕðjr  r0 jÞ; ð15Þ
Z 
V V
dr0 M2 ðr; r0 ; tÞρðr0 ; tÞ ¼ 0; ð12Þ
where ϕ(r) indicates the interparticle pair potential as a function
where M2 ðr; r0 ; tÞ ¼ βδ2 Pexc 0
t ½ρ; J=δJðr; tÞδJðr ; tÞ is the tensorial
of interparticle distance r. In order to obtain the form (15) we
two-body equal-time direct correlation function57,58. Its n-body have identified the many-body definition of the radial distribution
version is obtained from Mn+1(r1…rn+1, t) = δMn(r1…rn, t)/δJ(rn+1, function gðr; r0 Þ ¼ h∑i;j≠i δðr  ri Þδðr0  rj Þi=ðρðrÞρðr0 ÞÞ. Recall
t), and it satisfies the hierarchy that the pair distribution function g and the density-density
Z correlation function H2, as used in (3)–(5), are related via
drnþ1 Mnþ1 ρðrnþ1 ; tÞ ¼ 0; ð13Þ H 2 ðr; r0 Þ ¼ ðgðr; r0 Þ  1ÞρðrÞρðr0 Þ þ δðr  r0 ÞρðrÞ. Equation (15)
still holds for non-conservative interparticle forces, when
as obtained by differentiating (12) repeatedly with respect to the ∇ϕðjr  r0 jÞ is replaced by the (nongradient) interparticle force
current. field. We demonstrate in the following section the practical
relevance of these considerations.
Open boundaries. All our considerations have been based on
applying the symmetry operation to the entire system confined by Phase coexistence. We turn to situations of phase coexistence. As
external walls. The effects of these system walls are modeled by a we demonstrate, considering a large, but finite subvolume V of
suitable form of Vext(r). The position integrals formally run over the entire system is useful but it also requires to take boundary
all space, with the cutoff provided by hard (or steeply rising) terms into account. Here we take V to be cuboidal and to contain
external wall potentials. In many practical and relevant situations, the free (planar) interface between two coexisting phases, see
it is more useful to consider a system with open boundaries. Fig. 4a for a graphical illustration. The volume boundaries parallel
Alternatively one can consider only a subvolume V of the entire to the interface are taken to be seated deep inside either bulk
system, and restrict the accounting of force contributions to those phase. The internal force contributions on those faces of V that
particles that reside inside of V at a given time. In doing so, one “cut through” the interface, i.e. have a normal that is perpendi-
needs to take account of boundary effects60, as the boundaries of cular to the interface normal, vanish by symmetry. It
V are open, such that interparticle forces can be transmitted, and remains to evaluate (15) over each of the two faces in the
flow can occur. respective bulk region. Therefore the position dependences
In case that there are no net boundary contributions, all simplify to ρ(r) = ρb = const, where ρb indicates the bulk number
previous derived sum rules still hold. This includes e.g. an density, and the inhomogeneous pair distribution function sim-
effectively one-dimensional system in planar geometry that plifies as gðr; r0 Þ ¼ gðjr  r0 jÞ. Furthermore only force contribu-
evolves to the same bulk state at the left and right boundaries tions colinear with e, the outer interface normal of the considered
or if the boundary conditions are periodic. In both cases left and bulk phase b, contribute. For a single face in bulk phase b, it is
right boundary terms are equal up to a minus sign and hence straightforward to show that the result is the virial pressure
cancel each other. This example can be generalized straightfor- multiplied by the interface area A, i.e. the force Apbint e, where the
wardly to more complex geometries. internal interaction pressure pbint in phase b is e.g. given via the
For nonvanishing net boundary terms additional contributions Rr0
arise in the above sum rules. These contributions occur if the Clausius virial18, pbint ¼  π2 ρ2b dr r 2 gðrÞ dϕ
dr in two spatial
0
COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys 5
ARTICLE COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2

external field. The fully resolved one-body density distribution is

ρ(r, ω, ϖ, t) = 〈∑iδ(r − ri)δ(ω − ωi)δ(ϖ − ϖi)〉.
It is straightforward to ascertain that all the above (force) sum
rules for translation remain valid, as the orientations are unaffected
by translations, upon trivially generalizing from position-only to
position-orientation integration, ∫dr → ∫drdωdϖ etc.
In the following for simplicity of notation we first consider
uniaxial particles. Uniaxial particles depend only on one single
orientation ωi, as the particles are rotationally invariant around
this vector. Hence the ϖ-dependence of both the one- and many-
body quantities vanish and the total integral simplifies to ∫drdω.

Motility-induced phase separation. We use active Brownian

particles as an example for uniaxial particles. For simplicity we
consider spherical particles (discs) in two dimensions. The par-
ticles repel each other and they undergo self-propelled motion
along their orientation vector ω. Hence an additional one-body
force γsω acts on each swimmer, with γ the friction constant and s
Fig. 4 Phase separation into macroscopically distinct phases. the speed of free swimming. This self-propulsion creates char-
a Illustration of the geometry of phase separation of passive or active acteristic trajectories (see Fig. 4b for a schematic) which are also
particles. Shown are the gaseous ρg and liquid ρl plateau values of the affected by thermal diffusion (omitted in the schematic) of the
density profile (indicated by the color gradient) and the direction vector particle position ri and orientation ωi. Experimental realizations
e normal to the interface. The total system volume V þ V  consists of the of active Brownian particles include e.g. Janus colloids driven by
subvolume V and its complement V.  b Illustration of an active Brownian photon nudging61–65. If the density is high enough, motility-
particle (blue disc) with position ri and orientation ωi undergoing induced phase separation (MIPS) into an active gas and active
translational and rotational diffusion. The self propulsion along ωi creates liquid phase occurs for high enough values of the swim speed s, cf.
directed motion as indicated by the trajectory (magenta line). c Schematics Fig. 4c.
of motility-induced phase separation into active gas (left) and active liquid The force density balance (see e.g. the work of Hermann
phases (right). The arrows indicate the orientations ωi of the active et al.52) of such a system in steady state (no time dependence) is
particles. The interface is polarized.
γJðr; ωÞ ¼  kB T∇ρðr; ωÞ þ ρðr; ωÞf int ðr; ωÞ þ γsρðr; ωÞω
ð17Þ
dimensions (the argument remains general). The constant r0 þ ρðr; ωÞf ext ðr; ωÞ:
denotes the range of the interparticle interactions. The (negative) frictional force density on the left hand side is
The total force density balance for equilibrium phase balanced with the ideal gas contribution (first term), the
separation contains thermal diffusion and the internal force interparticle interactions (second term), the self-propulsion (third
density, which cancel each other, term) and the external contribution (fourth term) on the right
hand side. As in case of the equilibrium phase separation we
0 ¼ kB T∇ρðrÞ þ ρðrÞf int ðrÞ: ð16Þ integrate (17) over the volume V (see Fig. 4(a) for an illustration)
and over all orientations ω. In the following we discuss each term
Integration over the volume V yields the total force which is separately. For simplicity we assume planar geometry of the
proportional
R int ¼
to the pressure. Hence the total internal force Ftot system and we assume a vanishing external force, fext(r, ω) = 0.
V dr ρf int on V, cf. (15), amounts to the pressure difference Here, the interaction contribution pintAe is obtained as above in
g
ðpint  plint ÞAe, where e is the (unit vector) normal of the equilibrium via (15), but the virial is averaged over the
interface, pointing from, say, the gas (index g) to the liquid phase nonequilibrium steady state many-body probability distribution.
g
(index l). The total diffusive force is ðpid  plid ÞAe with the The integral over the current ∫drdωJ is assumed to vanish in
pressure of the ideal gas pid = kBTρ, evaluated at the gas (ρg) and steady state.
liquid bulk density (ρl). As there are no external forces (Vext ≡ 0) The total swim force that acts on V contributes to the total
then requesting the volume V to be forcefree amounts to force. In the considered situation, the swim force is entirely due to
g
ptot ¼ pltot , where the total pressure is the sum ptot = pid + pint. the polarization Mtot = ∫drdω ωρ of the free interface in MIPS.
Hence the boundary consideration yields the mechanical Particles at the interface tend to align against the dense phase43 if
equilibrium condition of equality of pressure in the coexisting they interact purely repulsively (cf. Fig. 4(c)), and they align
phases. against the dilute phase if interparticle attraction is present44. No
We conclude that the internal interactions that occur across the such spontaneous polarization occurs in bulk. The interface
free interface do not influence the (bulk) balance of the pressure polarization is a state function of the coexisting phases53 as
at phase coexistence, as the net effect of these interactions verified both experimentally66 and numerically67. The total swim
g
vanishes. At the heart of this argument lies Noether’s theorem for force that acts on V is ðpswim  plswim ÞAe, where
invariance against spatial displacements. pswim ¼ γsJ b =ð2Drot Þ, with Jb the bulk current in the forward
b

direction ω and Drot indicating the rotational diffusion constant.

Anisotropic particles. We turn to anisotropic interparticle Apart from the ideal term no further forces act, cf. the force
interactions, where ωi, ϖi are two perpendicular unit vectors that density balance (17). The integral over the ideal term is similar to
g
describes the particle orientation in space. Such systems are the total equilibrium ideal contribution, ðpid  plid ÞAe. Combina-
described by an interparticle interaction potential u(rN, ωN, ϖN), tion of all results from integration yields the total force balance.
which is assumed a priori to be invariant under spatial transla- As the negative integral over the force density defines the
tions. Similarly one-body fields in general depend on position r nonequilibrium pressure, the volume V being force free amounts
g
and orientations ω and ϖ of the particles, e.g. Vext(r, ω, ϖ) for the to ptot ¼ pltot , where ptot ¼ pid þ pint þ pswim . Hermann et al.52

6 COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys

COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2 ARTICLE

internal Noether sum rule (10). The integrated swim force

density is proportional to the total polarization Mtot = ∫drdω ωρ.
This quantity vanishes, Mtot = 0, as there is no net flux through
the boundaries in steady state (see Eq. (10) by Hermann et al.53).
Combination of all integrals yields the relation
Z
Fwall ¼ drdω ρf wall ¼ mgNez :
tot
ð19Þ

Hence the z-component of the total force on the wall F tot wall is equal
Fig. 5 Illustration of sedimentation of active Brownian particles under to the total gravitational force acting on all particles (see Fig. 5 for
gravity g. The active particles with orientation ω (black arrows) are a graphical representation). Equation (19) of course also holds for
confined by a lower wall and periodic boundary conditions on the sides passive colloids (s = 0).
(dashed lines). The total force that the swimming particles exert on the Keeping the translational invariance in the x-direction, we next
bottom wall (left scale pan) is equal to their weight (right scale pan) in turn to time-dependent systems. Therefore all one-body field in
steady states of the system. (17) additionally depend on the time t. Integration of the force
density balance (17) gives identical results for the thermal
diffusion, the internal force density and for the gravitational
demonstrate the splitting of pint into adiabatic and superadiabatic contribution as in the above case of steady state. Even for the
contributions and present results for the phase diagram based on time-dependent dynamics these integrals are independent of time
wall ¼ F wall ez and it only
t. The total wall force density is given as Ftot tot
approximate forms for the interparticle interaction contributions.
The pressure, especially its swim contribution is defined in acts along the unit vector in the z-direction ez, due to the
various different ways in the literature68–73. symmetry of the system. The integral of the self-propulsion term
We conclude that the internal interactions that occur across the is still proportional to the total polarization. However, the total
free interface do not influence the (bulk) balance of the pressure polarization does not vanish in general but decays exponentially
at phase coexistence, as the net effect of these interactions (see Eq. (21) by Hermann et al.53),
vanishes. In essence this argument follows from Noether’s
theorem for invariance against spatial displacements. Mtot ðtÞ ¼ Mtot ð0ÞeDrot t ; ð20Þ
where Mtot(0) indicates the initial polarization at time t = 0 and
Active sedimentation. Sedimentation under the influence of the time constant 1/Drot is the inverse rotational diffusion
gravity is a ubiquitous phenomenon in soft matter that has constant. Similarly, integration of the current still gives the (time-
attracted considerable interest, e.g. for colloidal mixtures74–77 and dependent) center of mass velocity vcm(t).
for active systems78–81. As sedimentation is a force driven phe- Insertion of these results into the spatial and orientational
nomenon the Noether sum rules apply directly, as we show in the integration of (17) leads to the total friction force
following.
We assume that the system is translationally invariant in the x- γNv cm ðtÞ ¼ sγMtot ð0Þ eDrot t þ F tot
wall ðtÞ ez  mgNez ; ð21Þ
direction and that an impenetrable wall at z = 0 acts as a lower which is hence a direct consequence of the Noether sum rule (10).
boundary of the system82 (cf. Fig. 5). We assume the wall-particle We find that the x-component of the center of mass velocity
interaction potential to be short-ranged. Its precise form is decays simultaneously with the total polarization, cf. the first term
irrelevant for the following considerations. The force density on the right hand side of (21). The z-component of vcm(t)
balance for such a system is (17) with the external force field depends on Mtot(t) and additionally on the time-dependent total
chosen as force exerted by the wall and the total graviational force. Hence
f ext ðr; ωÞ ¼ mgez þ f wall ðr; ωÞ; ð18Þ measuring the total force on the wall (i.e. by weighing, cf. Fig. 5)
and knowledge of the total initial polarization and the total
where m denotes the mass of a particle, g is the gravitational particle number allows one to determine the center of mass
acceleration and ez indicates the unit vector in z-direction. Hence velocity. Note that in the limit of long times, t → ∞, the total
the external force field fext consists of gravity and the wall polarization vanishes and this system evolves to a steady state.
contribution fwall. Hence the center of mass velocity vanishes and (19) is recovered.
To proceed we integrate the force density balance over As we have demonstrated both statements (19) and (21)
all positions r in the volume V and over all orientations ω. ultimately follow from the global Noether identity (10).
The total integral of the density distribution (per radiant) ∫drdωρ,
as appears in the gravitational term, gives the total number Rotational invariance. We return to the general case and initially
of particles N. The integral over the total current ∫drdω J consider spatial rotations in systems of spheres, i.e. systems where
is proportional
R Rto the center R of mass velocity u(rN) depends solely on (relative) particle positions, and where it
vcm ðtÞ ¼ drdω ρv= drdω ρ ¼ N1 drdω J. Here vcm(t) is a is invariant under global rotation of all rN around the origin. We
global quantity and hence it is independent of both position parameterize the rotation by a vector n. The direction of n
and orientation. We first only consider steady states, so the indicates the rotation axis and the modulus ∣n∣ is the angle of
center of mass velocity vcm(t) = 0 and hence the total rotation. To lowest nonvanishing order, the rotation amounts to
current vanishes. The total thermal diffusion term vanishes r → r + n × r. One-body functions change accordingly: the
because ∫dr ∇ ρ = ∂V dS ρ = 0 as there is no contribution of ρ external potential undergoes Vext(r) → Vext(r) + δVext(r), with
from the boundaries ∂V of the integration volume V. At the upper δVext(r) = (n × r) ⋅ ∇ Vext(r) and the density profile ρ(r) → ρ(r) +
and lower boundary the density is zero as it vanishes in the wall δρ(r) with δρ(r) = (n × r) ⋅ ∇ ρ(r). Much of the reasoning of the
and also for z → ∞. The left and the right boundary contributions above case of spatial displacement can be applied readily: Ω[Vext]
cancel each other as the density is independent of x due to is invariant under the rotation, and δΩ = ∫dr(δΩ/δVext(r))
translational invariance. The R total internal interaction force δVext(r) = ∫drρ(r)(n × r) ⋅ ∇ Vext(r) = 0. As the rotation vector n
density vanishes, Ftot int ¼ drdω ρf int ¼ 0, using the global is arbitrary, we can conclude that the total external torque T totext

COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys 7

ARTICLE COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2

vanishes in equilibrium13,
Z
T tot
ext   drρðrÞðr ´ ∇V ext ðrÞÞ ¼ 0: ð22Þ

As this holds true for any form of the applied Vext(r), we can
differentiate with respect to V ext ðr0 Þ, and obtain13
Z
r ´ ∇ρðrÞ ¼  dr0 βH 2 ðr; r0 Þðr0 ´ ∇0 V ext ðr0 ÞÞ; ð23Þ
Fig. 6 Illustration of the rotation operation of uniaxial particles. The
n
Z particles (rectangular shapes) at position r and with orientation ω are
∑ ðrα ´ ∇α H n Þ ¼  drnþ1 βV ext ðrnþ1 Þðrnþ1 ´ ∇nþ1 H nþ1 Þ: shown in original (black) and rotated (blue) configuration, where n
α¼1
indicates the rotation axis (direction) and angle (length).
ð24Þ
The excess free energy density functional Fexc[ρ] can be treated
accordingly. It is invariant under rotation, as its sole dependence as is consistent with the result of the above many-body derivation.
is on u(rN), which by assumption is rotationally invariant. Ana- As (30) holds for any (trial) J(r, t), the derivative of (30) with respect
logous to this reasoning, we obtain the result that the total to J(r, t) vanishes. Hence
interparticle adiabatic torque T tot
ad vanishes, Z
Z
drρðr; tÞðr ´ M2 ðr; r0 ; tÞÞ ¼ 0; ð31Þ
ad ¼
T tot drρðrÞðr ´ f ad ðrÞÞ ¼ 0: ð25Þ

Differentiation with respect to the independent field ρ(r) once where the cross product with a tensor is defined via contraction
and n times yields the respective identities13,14: with the Levi-Civita tensor. At nth order we obtain
Z
Z
r ´ ∇c1 ðrÞ ¼ dr0 c2 ðr; r0 Þðr0 ´ ∇0 ρðr0 ÞÞ; ð26Þ
drn ρðrn ; tÞðrn ´ Mn Þ ¼ 0: ð32Þ
n
Z
∑ ðrα ´ ∇α cn Þ ¼  drnþ1 ρðrnþ1 Þðrnþ1 ´ ∇nþ1 cnþ1 Þ: ð27Þ This identity and (31) express the vanishing of the total
α¼1
superadiabatic torque, when resolved on the n-body level of (time
The multi-body versions of the theorems of vanishing total direct) correlation functions.
external (22) and adiabatic internal (25) torques are, respectively,
Z Z
Orbital and spin coupling. The case of rotational symmetry of
dr1 V ext ðr1 Þ ¼ drn V ext ðrn Þðrα ´ ∇α H n Þ ¼ 0; ð28Þ
uniaxial particles is clearly more complex, as both particle coor-
dinates and particle orientations are affected by a global (“rigid”)
and
Z Z operation on the entire system, i.e. both positions and orienta-
tions are rotated consistently. Noether’s theorem ensures though
dr1 ρðr1 Þ ¼ drn ρðrn Þðrα ´ ∇α cn Þ ¼ 0; ð29Þ
that this operation is indeed the fundamental one, and that the
physically expected coupling of orbital and spinning effects will
for α = 1…n. These identities are respectively derived from (23)
naturally and systematically emerge. Here we use (common)
by multiplying with Vext(r), integrating over r and exploiting (22),
terminology for referring to spin as orientation vector rotation
and from (26) by multiplying with ρ(r), integrating over r and
(i.e. particle rotation around its center), as opposed to orbital
exploiting (25), and iteratively repeating for each order n.
rotation (of position vector) around the origin of position space.
On the many-body level, it is straightforward to see that the
Hence all the above considered torques that already occur in
total internal torque − ∑i(ri × ∇iu(rN)) = 0. Hence, as this
systems of spheres are of orbital nature. These of course remain
identity holds for each microstate, its general, nonequilibrium
relevant for anisotropic particles, but the nontrivial orientational
average vanishes, T tot int ¼ 0. The force field splitting f int ¼ f ad þ behavior of the latter will generate additional spin torques. The
f sup induces corresponding additive structure for the internal total global rotation consists of an orbital part, r → r + n × r, and a
torque: T tot
int ¼ T ad R
tot
þ T tot
sup , with the total superadiabatic (inter- spin part, ω → ω + n × ω; see Fig. 6 for a graphical
nal) torque T sup ¼ drρðrÞ½r ´ f sup ðr; tÞ. As T tot
tot
int ¼ T ad ¼ 0, we
tot representation.
For anisotropic systems the external field naturally acquires
conclude T sup ¼ 0, ∀ t.
tot
dependence on position r and orientation ω, i.e. Vext(r, ω), where
To apply Noether’s theorem to the power functional, we
−∇ Vext(r, ω) is the external force field as before, and
consider an instantaneous rotation, with infinitesimal angular
−ω × ∇ωVext(r, ω) is the external torque field, where ∇ω is the
velocity n_ at time t. The effect is a change in current J → J + δJ
derivative with respect to ω in orientation space. Hence the
with δJ ¼ ðn_ ´ rÞρðrÞ. Correspondingly, the velocity field acquires
induced change of external potential is δVext(r, ω) = (n × r) ⋅ ∇
an instantaneous global rotational contribution, according to
Vext + (n × ω) ⋅ ∇ωVext. This change leaves the grand potential Ω
vðr; tÞ ! vðr; tÞ þ n_ ´ r. The superadiabatic excess power func-
[Vext] invariant, hence δΩ = ∫drdω(δΩ/δVext(r, ω))n ⋅ (r × ∇ Vext
tional is invariant under this R operation and hence Pexc t ½ρ; J ¼ + ω × ∇ωVext) = 0, from which we identify the rotational Noether
Pt ½ρ; J þ δJ ¼ Pt ½ρ; J þ drðδPt ½ρ; J=δJðr; tÞÞ  ðn_ ´ rÞρðr; tÞ.
exc exc exc
theorem for the total external torque of anisotropic particles:
As n_ is arbitrary, we can conclude
Z Z
T tot
sup ¼ drρðr; tÞðr ´ f sup ðr; tÞÞ ¼ 0; 8t; ð30Þ T tot
ext ¼  drdωρðr; ωÞðr ´ ∇V ext þ ω ´ ∇ω V ext Þ ¼ 0: ð33Þ

8 COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys

COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2 ARTICLE

Differentiation with respect to Vext(r, ω) yields and δJω ¼ ðn_ ´ ωÞρ, we find
Z Z  
r ´ ∇ρ þ ω ´ ∇ω ρ ¼  dr0 dω0 βH 2 ðr; ω; r0 ; ω0 Þðr0 ´ ∇0 Vext
0
þ ω0 δPexc δPexc
T tot ¼  drdωρðr; ω; tÞ r ´ t
þ ω ´ t
¼ 0: ð41Þ
sup
δJ δJω
´ ∇0 ω Vext
0
Þ;
Local sum rules can be obtained straightforwardly by building
ð34Þ derivatives with respect to J(r, ω, t) and Jω(r, ω, t). The result is:
Z
n
∑ ðrα ´ ∇α H n þ ωα ´ ∇ωα H n Þ dr1 dω1 ρð1Þ½r1 ´ δ=δJð1Þ þ ω1 ´ δ=δJω ð1ÞMn;m ¼ 0: ð42Þ
α¼1
Z ð35Þ
¼  dr0 dω0 βVext 0 ω
ðr0 ´ ∇0 H nþ1 þ ω0 ´ ∇0 H nþ1 Þ; Here the tensorial equal-time direct correlation functions are defined
ω ω
as Mn;m ¼ βδ nþm Pexc t =δJð1Þ ¼ δJðnÞδJ ðn þ 1Þ ¼ J ðn þ mÞ,
0
where we have used Vext ¼ V ext ðr0 ; ω0 Þ as a shorthand, where the roman numerals refer to position, orientation and time t
H 2 ðr; ω; r ; ω Þ ¼ δρðr; ωÞ=δβV ext ðr0 ; ω0 Þ is the density-density
0 0 (no index), e.g. 1 ≡ r1, ω1, t. We have assumed that all functional
correlation function, and its n-body version Hn = Hn(r1, ω1…rn, derivatives commute.
ωn) with Hn+1 = δHn/δβVext(rn+1, ωn+1); in (35) the prime refers So far we have restricted ourselves to uniaxial particles. The
to the n + 1th degrees of freedom. On all levels of n-body derived rotational sum rules can be analogously determined for
correlation functions, the spin and orbital torques remain general anisotropic particles with an additional orientation vector
coupled. ϖ. This can be done by simply replacing ω × ∇ω → ω × ∇ω +
Turning to internal torques, the change of density upon global ϖ × ∇ϖ, ω ´ ∂=∂Jω ! ω ´ ∂=∂Jω þ ϖ ´ ∂=∂Jϖ and ∫drdω → ∫
rotation is δρ(r, ω) = n ⋅ (r × ∇ ρ + ω × ∇ωρ) and the net effect on drdωdϖ in (33)–(42). This replacement also affects the adiabatic
the excess free energy is δFexc = ∫drdω(δFexc[ρ]/δρ(r, ω))n ⋅ spin torque field τad and all one-body field depend additionally to
(r × ∇ ρ + ω × ∇ωρ) = 0. We hence obtain r and ω on the orientation ϖ. For anisotropic particles there exists
Z a more general version of (42) by exchange of Mn,m → Mn,m,l,
T ad ¼ drdωρðr; ωÞðr ´ f ad þ τ ad Þ ¼ 0;
tot
ð36Þ where l denotes the number of functional derivations with respect
to the rotational current Jϖ. The indices n and m belong to the
where the adiabatic spin torque field is τad(r, ω) = number of functional derivatives with respect to J and Jω as
− ω × ∇ωδFexc[ρ]/δρ(r, ω). From differentiation with respect to before.
ρ(r, ω) we obtain
Z Memory invariance. In the above treatment of nonequilibrium
ω
r ´ f ad þ τ ad ¼ dr0 dω0 c2 ðr; ω; r0 ; ω0 Þðr0 ´ ∇0 ρ0 þ ω0 ´ ∇0 ρ0 Þ; situations we have exploited invariance against an instantaneous
transformation applied to the system. As we have shown, the
ð37Þ corresponding Noether identities carry imminent physical
meaning. These sum rules hold for the nonequilibrium effects
n
∑ ðrα ´ ∇α cn þ ωα ´ ∇ωα cn Þ that arise from the interparticle interaction, i.e. they constrain
α¼1
Z superadiabatic forces and torques, as obtained from translation
ð38Þ
and rotation, respectively. Here we exploit that the corresponding
¼  dr0 dω0 ρ0 ðr0 ´ ∇0 cnþ1 þ ω0 ´ ∇0 ω cnþ1 Þ; nonequilibrium functional generator Pexc carries further invar-
t
iances, once one allows the transformation to act also on the
where ρ0 ¼ ρðr0 ; ω0 Þ. Multi-body versions of (33) and (36) read as history of the system. As we demonstrate in the following, the
Z Z
resulting identities constitute exact constraints on the memory
d1V ext ð1Þ ¼ dnV ext ðnÞðrα ´ ∇α þ ωα ´ ∇ωα ÞH n ¼ 0 ð39Þ structure that are induced by the coupled interparticle interac-
tions. Recall that a reduced one-body description of a many-body
and system is generically non-Markovian (i.e. nonlocal in time)83. The
Z Z
study of memory kernels, often carried out in the framework of
d1ρð1Þ ¼ dnρðnÞðrα ´ ∇α þ ωα ´ ∇ωα Þcn ¼ 0: ð40Þ generalized Langevin equations, is a topic of significant current
research activity84–91.
Here we have used the shorthand notation 1 ≡ r1, ω1 etc., and the Our approach differs from these efforts in that no a priori
derivation is analogous to the above rotational case of spherical generic form of a reduced equation of motion is assumed. Rather
particles. our considerations are formally exact and interrelate (and hence
The two-body sum rules (34) and (37) are identical to those constrain) time correlation functions, which are generated from
obtained by Tarazona and Evans16 using rotational invariance the central nonequilibrium object Pexc via functional differentia-
t
arguments applied directly to correlation functions. Our meth- tion. Very little is known about the memory structure of
odology not only allows to naturally re-derive their results, but superadiabatic forces, with exceptions being the NOZ
also to identify the full gamut of adiabatic rotatational sum rules, framework57,58 and the demonstration of the relevance of
from the global statements (33) and (36) to the infinite memory for the observed viscoelasticity of hard sphere liquids92.
hierarchies (35) and (38) (we use notational convention different Both the Ornstein–Zernike (OZ) and NOZ relations are different
from Tarazona and Evans16: our ∇ω is (only) a partial derivative from the Noether identities. The former relations are a direct
with respect to ω, i.e. ∇ω ≡ ∂/∂ω, whereas their ∇ω ≡ ω × ∂/∂ω. consequence of the generality of the variational principle. Per se,
The modulus is fixed, ∣ω∣ = 1, in both versions. Tarazona and neither the OZ nor the NOZ relations reflect the Noether
Evans16 notate H2 as G in their (18) and (19)). symmetries.
Again for each microstate ∑i ðri ´ ∇i u þ ωi ´ ∇ωi uÞ ¼ 0 and Recalling the illustrated overview of the different types of
hence on average T tot
int ¼ 0. From the splitting shifting in Fig. 1, in the following we treat two further types of
T int ¼ T ad þ T sup , we conclude T tot
tot tot tot
sup ¼ 0. From rotational invariance transformations: One is the static transformation. This
ω ω
invariance of Pexc t ½ρ; J; J , where J (r, ω, t) is the rotational operation is formally analogous to the above equilibrium
current50, against an instantaneous angular “kick” δJ ¼ ðn_ ´ rÞρ treatment of the adiabatic state, but it is here carried out in the

COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys 9

ARTICLE COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2

same way at all times. This static transformation contrasts (and Repeated differentiation of (44) with respect to J(r, t) generates
complements) the instantaneous transformation used above for a hierarchy,
the time-dependent case. The corresponding changes to density n1
Z
and current are graphically illustrated in Fig. 3b. The second ∑ ∇α Mn1 ðrn1 ; tÞ ¼ drn ðmn ðrn ; tÞ∇n ρðrn ; tÞ þ Mn ðrn ; tÞ
invariance operation is that of memory shifting, where the α¼1
transformation parameter is taken to be time-dependent, cf.  ∇n Jðrn ; tÞT Þ
Fig. 3c for a graphical representation. For simplicity we restrict Z t Z
ourselves to cases where at both ends of the considered time þ dt 0 drn ðmn ðrn1 ; t; rn ; t 0 Þ∇n ρðrn ; t 0 Þ þ Mn ðrn1 ; t; rn ; t 0 Þ
interval, no shifting occurs (i.e. such that the transformation is 0
T
the identity at the limiting times).  ∇n Jðrn ; t 0 Þ Þ;
The static spatial shift consists of ρðr; t 0 Þ ! ρðr þ ϵ; t 0 Þ and
Jðr; t 0 Þ ! Jðr þ ϵ; t 0 Þ, where the time argument t 0 is arbitrary, and ð45Þ
ϵ = const characterizes magnitude and direction of the transla- where the n-body equal-time direct correlation functions of rank
tion, see the illustration in Fig. 3b. Hence the time derivative n are Mn ðrn ; tÞ ¼ βδ n Pexc t =δJðr1 ; tÞ ¼ δJðrn ; tÞ and mn ðr ; tÞ ¼
n
ϵ_ ¼ 0 such that the current does not acquire any displacement βδn Pexc =δJðr ; tÞ ¼ δJðr ; tÞδρðr ; tÞ, where we have used the
t 1 n1 n
contribution, as also _ϵρ ¼ 0 at all times. The spatial shift applies shorthand rn = r1…rn. Furthermore at unequal times we have:
to all times t 0 considered, and we can restrict ourselves to Mn ðrn1 ; t; rn ; t 0 Þ ¼ βδ n Pexc 0
t =δJðr1 ; tÞ ¼ δJðrn1 ; tÞδJðrn ; t Þ as a
0 ≤ t 0 ≤ t. Hence the changes in kinematic fields are to first order rank n tensor, and also a rank n − 1 tensor
given by δρðr; t 0 Þ ¼ ϵ  ∇ρðr; t 0 Þ and δJðr; t 0 Þ ¼ ϵ  ∇Jðr; t 0 Þ. As
mn ðrn1 ; t; rn ; t 0 Þ ¼ βδ n Pexc t =δJðr1 ; tÞ ¼ δJðrn1 ; tÞδρðrn ; t Þ.
0
Pexct originates solely from the interparticle interaction potential,
In the second case, we consider a more general invariance
the invariance of u(rN) against the global displacement at all transformation that is obtained by letting the transformation
times induces invariance of Pexc
t . Hence parameter be time-dependent. In this case of time-dependent
Pt ½ρ; J ¼ Pt ½ρ þ δρ; J þ δJ ¼ Pt ½ρ; J þ δPt . Here δPexc
exc exc exc exc
t shifting, we prescribe a displacement vector ϵðt 0 Þ for times
indicates the change in superadiabatic free power and due to 0 ≤ t 0 ≤ t, i.e. between the initial time, throughout the past and up
the invariance δPexc t ¼ 0. On the other hand we can express δP t
exc
to the "current” time t. We restrict ourselves to vanishing shift at
via the functional Taylor expansion. To linear order the result the boundaries of the considered time interval, i.e. ϵ(0) = ϵ(t) = 0.
consists of two integrals. One integral comes from the time-slice Due to the overdamped character of the dynamics, its
Rfunctional derivative at fixed (end) time t and is given by interparticle contributions are unaffected by this transformation,
t =δρÞδρ þ ðδP t =δJÞ  δJ. The second integral is from a
dr½ðδPexc exc
and hence Pexc t is invariant. The induced changes that the density
functional
Rt 0 R 0 derivative at (variable) time t 0 , given by and the current acquire arise from shifting their position
0 0 0 0
0 dt dr ½ðδPt =δρ Þδρ þ ðδPexc
exc
t =δJ Þ  δJ , where the prime argument, but the current also acquires an additive shifting
indicates dependence on arguments r and t 0 . We then exploit
0
current contribution. The latter contribution is analogous to the
that the displacement ϵ of the static shift (which parametrizes the (sole) effect that is present in the instantaneous shifting, but here
changes in density and in current) is arbitrary. The result is a applicable at all times (in the considered time interval).
global nonequilibrium Noether theorem, given by Hence the time-dependent shifting, as illustrated in Fig. 3c,
Z δPexc  Z t induces the following changes to the density and the current:
dr t
∇ρ  f sup  ∇J þ T
dt 0 δρðr0 ; t 0 Þ ¼ ϵðt 0 Þ  ∇0 ρðr0 ; t 0 Þ and δJðr0 ; t 0 Þ ¼ ϵðt 0 Þ  ∇Jðr0 ; t 0 Þ
δρ ϵ_ ðt 0 Þρðr0 ; t 0 Þ, where ϵ_ ðt 0 Þ ¼ dϵðt 0 Þ=dt 0 . Next we can regard Pexc
Z 0
ð43Þ t ½ρ þ
δPexc δPexc  δρ; J þ δJ as a functional of ϵðt 0 Þ and ϵ_ ðt 0 Þ. Its invariance
0 0T
dr0 t
∇ 0 0
ρ þ t
 ∇ J ¼ 0; amounts to stationarity, i.e. vanishing first functional derivative,
δρ0 δJ0
with respect to the displacement. This problem, in particular for
where we have used the relationship of the superadiabatic force the present case of fixed boundary values, amounts to one of the
field to its generator, f sup ðr; tÞ ¼ δPexc t ½ρ; J=δJðr; tÞ, the primed most basic problems in the calculus of variations. It is realized,
symbol ∇0 indicates the derivative with respect to r0 , and the e.g., in the determination of catenary curves and indeed, in
superscript T denotes the matrix transpose (in index notation the Hamilton’s principle of classical mechanics. Exploiting the
k-component of the vector a ⋅ ∇ bT is ∑k0 ak0 ∇k bk0 ). corresponding Euler–Lagrange equation leads to
Equation (43) constitutes a global identity that links Z Z Z
density, current, and superadiabatic force field in a nontrivial d 0 0 0 0 0 0 0T
dr m ρ þ dr m  ∇ J þ dr0 m01 ∇0 ρ0 ¼ 0; ð46Þ
spatial and temporal form. As the central variation principle48 dt 0 1 1

allows to vary J(r, t) freely, (43) remains true upon building the where the one-body time direct correlation functions are
functional derivative with respect to J(r, t). The result is a local m01 ðr0 ; t 0 ; tÞ ¼ βδPexc 0 0
and m01 ðr0 ; t 0 ; tÞ ¼ βδPexc
t =δJðr ; t Þ t =
identity δρðr0 ; t 0 Þ. Differentiation with respect to J(r, t) yields again a local
Z
memory identity.
β∇f sup ¼ dr0 ðm2 ðr; r0 ; tÞ∇0 ρðr0 ; tÞ þ M2 ðr; r0 ; tÞ  ∇0 JT ðr0 ; tÞÞ
Z t Z Conclusions. We have demonstrated that Noether’s theorem for
þ dt 0 dr0 ðm2 ðr; t; r0 ; t 0 Þ∇0 ρ0 þ M2 ðr; t; r0 ; t 0 Þ  ∇0 J0 T Þ; exploiting symmetry in a variational context has profound
0 implications for Statistical Physics. Known sum rules can be
ð44Þ derived with ease and powerfully generalized to full infinite
where two-body time direct correlation functions occur in hierarchies, to the rotational case, and to time-dependence in
nonequilibrium. Recall the selected applications31–41 of the
vectorial form: m2 ðr; r0 ; tÞ ¼ βδ 2 Pexc
t =δJðr; tÞδρðr ; tÞ,
0
0 0 0 0
equilibrium sum rules, as we have laid out in the introduction.
m2 ðr; t; r ; t Þ ¼ βδ Pt =δJðr; tÞδρðr ; t Þ, as well as in tensorial
2 exc
For the time-dependent case, we envisage similar insights from
form: M2 ðr; r0 ; tÞ ¼ βδ 2 Pexc 0
t =δJðr; tÞδJðr ; tÞ, M2 ðr; t; r0 ; t 0 Þ ¼ using the newly formulated nonequilibrium sum rules in inves-
0 0
βδ Pt =δJðr; tÞδJðr ; t Þ. Here we have made the (common)
2 exc
tigations of e.g., the dynamics of freezing, of liquid crystal flow,
assumption that the second derivatives can be interchanged. and of driven fluid interfaces. On the conceptual level, Noether’s

10 COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys

COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2 ARTICLE

theorem assigns a clear meaning and physical interpretation to all insights in its memory structure and also generalize the
resulting identities, as being generated from an invariance prop- translational memory relations (43)–(46) to anisotropic particles.
erty of an underlying functional generator. Although the sym- It would be highly interesting to apply (49) to the recently
metry operation that we considered are simplistic, and only their obtained direct correlation function of the hard sphere crystal.
lowest order in a power series expansion needs to be taken into This would allow to investigate whether Triezenberg and
account, a significantly complex body of sum rules naturally Zwanzig’s concept that they originally developed for the free
emerges. Hence the application of Noether’s theorem is sig- gas–liquid interface applies to the also self-sustained density
nificantly deeper than mere exploiting of symmetries of argu- inhomogeneity in a solid. Furthermore, addressing further cases
ments of correlation functions, i.e. that the direct correlation of self motility42–44, including active freezing96,97, as well as
function c2 ðr; r0 Þ in bulk fluids depends solely on jr  r0 j. Rather, further types of time evolution, such as molecular dynamics or
as governed by functional calculus, coupling of different levels of quantum mechanics should be interesting. This is feasible, as the
correlation functions occurs. Noether considerations are not restricted to overdamped classical
The Noether sum rules are different from the variational systems, as (formal) power functional generators exist for
principle, as embodied in the Euler–Lagrange equation. On the quantum98 and classical Hamiltonian99 many-body systems. On
formal level, the difference is that the Euler–Lagrange equation the methodological side, besides power functional theory, our
(both of DFT and of PFT) is a formally closed equation on the framework could be complemented by e.g. mode-coupling theory
one-body level. In contrast, the Noether rules couple n- and (n + and Mori-Zwanzig techniques100, as well as approaches beyond
1)-body correlation functions, hence they are of genuine that101. Given that the equilibrium force sum rules are crucial in
hierarchical nature. They also describe different physics, as the the description of crystal33,34 and liquid crystal35 excitations, the
Euler–Lagrange equation expresses a chemical potential equili- study of such systems under drive is a further exciting prospect.
brium in DFT and the local force balance relationship in PFT. In
contrast, the Noether identities stem from the symmetry proper- Methods
ties of the respective underlying physical system. Relationship to classical results. We give an overview of how the Noether sum
The standard DFT approximations, ranging from simple rules relate to previously known results. The famous LMBW-equation was derived
independently by Lovett et al.10 and by Wertheim11 and reads
local, square-gradient, and mean-field functional to more Z
sophisticated weighted-density-schemes including fundamental ∇ln ρðrÞ þ β∇V ext ðrÞ ¼ dr0 c2 ðr; r0 Þ∇0 ρðr0 Þ: ð47Þ
measure theory satisfy the internal force relationships. This can
be seen straightforwardly by observing that these functionals We can conclude that (47) is a combination of the local internal Noether sum rule
do satisfy global translation invariance (the value of the (7) for translational symmetry and the equilibrium Euler–Lagrange equation
free energy is independent of the choice of coordinate origin). c1 ðrÞ ¼ ln ρðrÞ þ β∇V ext ðrÞ  βμ, where μ indicates the chemical potential. LMBW
All higher-order Noether identities are then automatically also derived a lesser known external relation, which is equivalent to (47) and reads
Z
satisfied, as these inherit the correct symmetry properties from
∇ρðrÞ þ βρðrÞ∇V ext ðrÞ ¼ dr0 ðgðr; r0 Þ  1ÞρðrÞρðr0 Þ∇0 V ext ðr0 Þ: ð48Þ
the generating (excess free energy) functional. Our formalism
hence provides a concrete reason, over mere empirical experience, We find that (48) contains the local external Noether sum rule (3) along with the
why the practitioners’ choices for approximate functionals are relation H 2 ðr; r0 Þ ¼ ðgðr; r0 Þ  1ÞρðrÞρ0 ðr0 Þ þ ρðrÞδðr  r0 Þ18.
sound. The situation for more complex DFT schemes could The Triezenberg–Zwanzig equation102 holds for vanishing external potential
potentially be different though. As soon as, say, self-consistency Vext(r) = 0 and is given by
Z
of some form is imposed, or coupling to auxiliary field comes into
∇ln ρðrÞ ¼ dr0 c2 ðr; r0 Þ∇0 ρðr0 Þ: ð49Þ
play, it is easy to imagine that the Noether identities help in
restricting choices in the construction of such approximation Originally (49) was derived for the free liquid–vapor interface in a parallel
schemes. geometry. We find that the relation consists of the local internal Noether sum rule
The sum rules imposed by the three types of dynamical (7) and the equilibrium Euler–Lagrange equation. The LMBW equation (48)
displacements are satisfied within the velocity gradient form of reduces to the Triezenberg–Zwanzig equation (49) for cases of Vext(r) = 0.
Another related equation is the first member of the Yvon–Born–Green (YBG)
the power functional54,56. It is straightforward to see that the hierarchy103,104,
functional is independent of the coordinate origin (static Z
shifting). For the cases of dynamical shifting, the invariance of ρð∇ln ρðrÞ þ β∇V ext ðrÞÞ ¼ β dr0 gðr; r0 ÞρðrÞρðr0 Þ∇ϕðjr  r0 jÞ; ð50Þ
the functional stems from invariance of the velocity field against
shifting. For both instantaneous and memory shifting, the where ϕ(r) denotes the interparticle pair potential as before. Although (50) has a
velocity gradient remains invariant under the displacement. similar structure as the LMBW equation, it is not based on symmetry or Noether
arguments but arises from integration out of degrees of freedom. If one would like
We envisage that the higher than two-body Noether identities to include the translational symmetry one can simply replace the left hand side of
can facilitate the construction of advanced liquid state/density (50) with the right hand side of the LMBW equation (47), which leads to
functional approximations. Such work should surely be highly Z Z
challenging. In the context of fundamental measure theory (see dr0 c2 ðr; r0 Þ∇0 ρðr0 Þ ¼ β dr0 gðr; r0 Þρðr0 Þ∇ϕðjr  r0 jÞ: ð51Þ
e.g. the work by Roth20 for an enlightening review) it is worth
recalling that in Rosenfeld’s original 1989 paper93, he calculated Some of the here derived sum rules are rederivations of known relations. We
reiterate the relationships. In his overview Baus13 showed that in equilibrium the
the three-body direct correlation function from his then newly total external (2) and internal (6) force vanish and derived the corresponding local
proposed functional. The result for the corresponding three-body hierarchies (3), (4) and (7), (8). Similar sum rules13 hold for the external (22) and
pair correlations compared favorably against simulation data. internal (25) total torques and their corresponding hierarchies (23), (24) and (26),
Furthermore, the recent insights into two-body correlations in (27). Tarazona and Evans16 generalized these equations for uniaxial particles and
derived the first order of the external (34) and internal (37) hierarchies due to
inhomogeneous liquids94 and crystals95 demonstrates that work- rotations.
ing with higher-body correlation functions is feasible. To the best of our knowledge the hierarchies of global Noether sum rules, such
In future work it would be very beneficial to bring together the as (5), (9), (28), and (29), have not been determined previously. Furthermore the
Noether identities with the nonequilibrium Ornstein–Zernike global external (33) and internal (36) Noether sum rules for uniaxial colloids and
their corresponding hierarchies (35) and (38) (with exception of the first order) are
relations57,58, in order to aid construction of new dynamical reported here for the first time. As the considerations in the literature focused on
approximations. One could exploit the rotational invariance of equilibrium, all our nonequilibrium relations, as e.g. (11)–(13) and especially the
the superadiabatic excess power functional Pexc t to gain deeper ones including memory (43)–(46) have not been found before.

COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys 11

ARTICLE COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2

Data availability 28. Martin-Jimenez, D., Chacón, E., Tarazona, P. & Garcia, R. Atomically resolved
Data sharing is not applicable to this study as no datasets were generated or analyzed three-dimensional structures of electrolyte aqueous solutions near a solid
during the current study. surface. Nat. Comms. 7, 12164 (2016).
29. Hernández-Muñoz, J., Chacón, E. & Tarazona, P. Density functional analysis
of atomic force microscopy in a dense fluid. J. Chem. Phys. 151, 034701
Received: 2 March 2021; Accepted: 1 July 2021; (2019).
30. Muscatello, J., Chacón, E., Tarazona, P. & Bresme, F. Deconstructing
temperature gradients across fluid interfaces: the structural origin of the
thermal resistance of liquid-vapor interfaces. Phys. Rev. Lett. 119, 045901
References (2017).
1. Noether, E. Invariante Variationsprobleme. Nachr. d. König. Gesellsch. d. Wiss. 31. Xu, X. & Rice, S. A. A density functional theory of one- and two-layer freezing
zu Göttingen, Math.-Phys. Klasse 235 (1918). English translation by Tavel, M. in a confined colloid system. Proc. R. Soc. A 464, 65 (2008).
A. Invariant variation problems. Transp. Theo. Stat. Phys. 1, 186 (1971); for a 32. Brader, J. M. Structural precursor to freezing: an integral equation study. J.
version in modern typesetting see: Wang, F.Y. arXiv:physics/0503066v3 Chem. Phys. 128, 104503 (2008).
(2018). 33. Walz, C. & Fuchs, M. Displacement field and elastic constants in nonideal
2. Neuenschwander, D. E. Emmy Noether’s Wonderful Theorem (Johns Hopkins crystals. Phys. Rev. B 81, 134110 (2010).
University Press, 2011). For a description of many insightful and pedagogical 34. Häring, J. M., Walz, C., Szamel, G. & Fuchs, M. Coarse-grained density and
examples and applications.. compressibility of nonideal crystals: General theory and an application to
3. Byers, N. E. Noether’s discovery of the deep connection between symmetries cluster crystals. Phys. Rev. B 92, 184103 (2015).
and conservation laws. Preprint at https://arxiv.org/abs/physics/9807044 35. Häring, J. M. Microscopically founded elasticity theory for defect-rich systems
(1998). of anisotropic particles. Ph.D. Thesis, Universität Konstanz (2020).
4. Rowlison, J. S. & Widom, B. Molecular theory of capillarity (Dover, New York, 36. Bryk, P., Henderson, D. & Sokolowski, S. A fluid in contact with a
2002). semipermeable surface: second-order integral equation approach. J. Chem.
5. van der Waals, J. D. The thermodynamic theory of capillarity under the Phys. 107, 3333 (1997).
hypothesis of a continuous variation of density. Z. Phys. Chem. 13, 657 (1894); 37. Henderson, J. R. & van Swol, F. On the interface between a fluid and a planar
English translation by J. S. Rowlinson. J. Stat. Phys. 20, 197 (1979). wall. Mol. Phys. 51, 991 (1984).
6. Kerins, J. & Boiteux, M. Applications of noether’s theorem to inhomogeneous 38. Tejero, C. F. & Baus, M. Viscoelastic surface waves and the surface structure of
fluids. Phys. A 117, 575 (1983). liquids. Mol. Phys. 54, 1307 (1985).
7. Bukman, D. J. Torque balance at a line of contact. Phys. A 319, 151 (2003). 39. Iatsevitch, S. & Forstmann, F. Density profiles at liquid-vapor and
8. Boiteux, M. & Kerins, J. Thermodynamic properties of inhomogeneous fluids. liquid-liquid interfaces: an integral equation study. J. Chem. Phys. 107, 6925
Phys. A 121, 399 (1983). (1997).
9. Evans, R. The nature of the liquid-vapour interface and other topics in the 40. Kasch, M. & Forstmann, F. An orientational instability and the liquid-vapor
statistical mechanics of non-uniform, classical fluids. Adv. Phys. 28, 143 interface of a dipolar hard sphere fluid. J. Chem. Phys. 99, 3037 (1993).
(1979). 41. Mandal, S., Lang, S., Botan, V. & Franosch, T. Nonergodicity parameters of
10. Lovett, R. A., Mou, C. Y. & Buff, F. P. The structure of the liquid-vapor confined hard-sphere glasses. Soft. Matter. 13, 6167 (2017).
interface. J. Chem. Phys. 65, 570 (1976). 42. Farage, T. F. F., Krinninger, P. & Brader, J. M. Effective interactions in active
11. Wertheim, M. S. Correlations in the liquid-vapor interface. J. Chem. Phys. 65, Brownian suspensions. Phys. Rev. E 91, 042310 (2015).
2377 (1976). 43. Paliwal, S., Rodenburg, J., van Roij, R. & Dijkstra, M. Chemical potential in
12. Kayser, R. F. & Raveché, H. J. Emergence of periodic density patterns. Phys. active systems: predicting phase equilibrium from bulk equations of state?
Rev. B 22, 424 (1980). New J. Phys. 20, 015003 (2018).
13. Baus, M. Broken symmetry and invariance properties of classical fluids. Mol. 44. Paliwal, S., Prymidis, V., Filion, L. & Dijkstra, M. Non-equilibrium surface
Phys. 51, 211 (1984). tension of the vapour-liquid interface of active Lennard- Jones particles. J.
14. Lovett, R. & Buff, F. P. Examples of the construction of integral equations in Chem. Phys. 147, 084902 (2017).
equilibrium statistical mechanics from invariance principles. Physica A 172, 45. Loehr, J., Loenne, M., Ernst, A., de las Heras, D. & Fischer, T. M. Topological
147 (1991). protection of multiparticle dissipative transport. Nat. Commun. 7, 11745
15. Baus, M. & Lovett, R. A direct derivation of the profile equations of Buff- (2016).
Lovett-Mou-Wertheim from the Born-Green-Yvon equations for a non- 46. Loehr, J. et al. Colloidal topological insulators. Comms. Phys. 1, 4 (2018).
uniform equilibrium fluid. Physica A 181, 329 (1992). 47. Rossi, A. M. E. B. et al. Hard topological versus soft geometrical magnetic
16. Tarazona, P. & Evans, R. On the validity of certain integro-differential particle transport. Soft Matter 15, 8543 (2019).
equations for the density-orientation profile of molecular fluid interfaces. 48. Schmidt, M. & Brader, J. M. Power functional theory for Brownian dynamics.
Chem. Phys. Lett. 97, 279 (1983). J. Chem. Phys. 138, 214101 (2013).
17. Gubbins, K. E. Structure of nonuniform molecular fluids – integrodifferential 49. Fortini, A., de las Heras, D., Brader, J. M. & Schmidt, M. Superadiabatic forces
equations for the density-orientation profile. Chem. Phys. Lett. 76, 329 (1980). in Brownian many-body dynamics. Phys. Rev. Lett. 113, 167801 (2014).
18. Hansen, J. P. & McDonald, I. R. Theory of Simple Liquids, 4th ed. (Academic 50. Krinninger, P., Schmidt, M. & Brader, J. M. Nonequilibrium phase behaviour
Press, London, 2013). from minimization of free power dissipation. Phys. Rev. Lett. 117, 208003
19. Evans, R., Oettel, M., Roth, R. & Kahl, G. New developments in classical (2016).
density functional theory. J. Phys.: Condens. Matter 28, 240401 (2016). 51. Hermann, S., de las Heras, D. & Schmidt, M. Non-negative interfacial tension
20. Roth, R. Fundamental measure theory for hard-sphere mixtures: a review. J. in phase-separated active Brownian particles. Phys. Rev. Lett. 123, 268002
Phys. Condens. Matt. 22, 063102 (2010). (2019).
21. Levesque, M., Vuilleumier, R. & Borgis, D. Scalar fundamental measure theory 52. Hermann, S., Krinninger, P., de las Heras, D. & Schmidt, M. Phase coexistence
for hard spheres in three dimensions: Application to hydrophobic solvation. J. of active Brownian particles. Phys. Rev. E 100, 052604 (2019).
Chem. Phys. 137, 034115 (2012). 53. Hermann, S. & Schmidt, M. Active interface polarization as a state function.
22. Jeanmairet, G., Levesque, M. & Borgis, D. Molecular density functional theory Phys. Rev. Research 2, 022003(R) (2020).
of water describing hydrophobicity at short and long length scales. J. Chem. 54. de las Heras, D. & Schmidt, M. Velocity gradient power functional for
Phys. 139, 154101 (2013). Brownian dynamics. Phys. Rev. Lett. 120, 028001 (2018).
23. Evans, R. & Wilding, N. B. Quantifying density fluctuations in water at a 55. Stuhlmüller, N. C. X., Eckert, T., de las Heras, D. & Schmidt, M. Structural
hydrophobic surface: evidence for critical drying. Phys. Rev. Lett. 115, 016103 nonequilibrium forces in driven colloidal systems. Phys. Rev. Lett. 121, 098002
(2015). (2018).
24. Evans, R., Stewart, M. C. & Wilding, N. B. A unified description of hydrophilic 56. de las Heras, D. & Schmidt, M. Flow and structure in nonequilibrium
and superhydrophobic surfaces in terms of the wetting and drying transitions Brownian many-body systems. Phys. Rev. Lett. 125, 018001 (2020).
of liquids. Proc. Nat. Acad. Sci. 116, 23901 (2019). 57. Brader, J. M. & Schmidt, M. Nonequilibrium Ornstein-Zernike relation for
25. Remsing, R. C. Commentary: playing the long game wins the cohesion- Brownian many-body dynamics. J. Chem. Phys. 139, 104108 (2013).
adhesion rivalry. Proc. Nat. Acad. Sci. 116, 23874 (2019). 58. Brader, J. M. & Schmidt, M. Dynamic correlations in Brownian many-body
26. Evans, R., Stewart, M. C. & Wilding, N. B. Critical drying of liquids. Phys. Rev. systems. J. Chem. Phys. 140, 034104 (2014).
Lett. 117, 176102 (2016). 59. Ornstein, L. S. & Zernike F. The influence of accidental deviations of
27. Chacko, B., Evans, R. & Archer, A. J. Solvent fluctuations around solvophobic, density on the equation of state. Proc. Acad. Sci. Amsterdam 17, 793 (1914);
solvophilic, and patchy nanostructures and the accompanying solvent this article is reprinted in Frisch, H. & Lebowitz, J. L. The Equilibrium Theory
mediated interactions. J. Chem. Phys. 146, 124703 (2017). of Classical Fluids (Benjamin, New York, 1964).

12 COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys

COMMUNICATIONS PHYSICS | https://doi.org/10.1038/s42005-021-00669-2 ARTICLE

60. Requardt, M. & Wagner, H. J. (Infinite) Boundary corrections for the LMBW- 93. Rosenfeld, Y. Free-energy model for the inhomogeneous hard-sphere fluid
Equations and the TZ-Formula of surface tension in the presence of mixture and density-functional theory of freezing. Phys. Rev. Lett. 63, 9
spontaneous symmetry breaking. Phys. A 154, 183 (1988). (1989).
61. Khadka, U., Holubec, V., Yang, H. & Cichos, F. Active particles bound by 94. Tschopp, S. M. & Brader, J. M. Fundamental measure theory of inhomogeneous
information flows. Nat. Commun. 9, 3864 (2018). two-body correlation functions. Phys. Rev. E 103, 042103 (2021).
62. Selmke, M., Khadka, U., Bregulla, A. P., Cichos, F. & Yang, H. Theory for 95. Lin, S.-C., Oettel, M., Häring, J. M., Haussmann, R., Fuchs, M., & Kahl, G.
controlling individual self-propelled micro-swimmers by photon nudging I: Direct correlation function of a crystalline solid. Phys. Rev. Lett. (to appear).
directed transport. Phys. Chem. Chem. Phys. 20, 10502 (2018). Preprint at https://arxiv.org/abs/2104.11558 (2021).
63. Selmke, M., Khadka, U., Bregulla, A. P., Cichos, F. & Yang, H. Theory for 96. Turci, F. & Wilding, N. B. Phase separation and multibody effects in
controlling individual self-propelled micro-swimmers by photon nudging II: three-dimensional active Brownian particles. Phys. Rev. Lett. 126, 038002
confinement. Phys. Chem. Chem. Phys. 20, 10521 (2018). (2021).
64. Bregulla, A. P., Yang, H. & Cichos, F. Stochastic localization of 97. Omar, A. K., Klymko, K., GrandPre, T. & Geissler, P. L. Phase diagram of
microswimmers by photon nudging. ACS Nano 8, 6542 (2014). active Brownian spheres: crystallization and the metastability of motility-
65. Qian, B., Montiel, D., Bregulla, A., Cichos, F. & Yang, H. Harnessing thermal induced phase separation. Phys. Rev. Lett. 126, 188002 (2021).
fluctuations for purposeful activities: the manipulation of single micro- 98. Schmidt, M. Quantum power functional theory for many-body dynamics. J.
swimmers by adaptive photon nudging. Chem. Sci. 4, 1420 (2013). Chem. Phys. 143, 174108 (2015).
66. Söker, N. A., Auschra, S., Holubec, V., Kroy, K. & Cichos, F. How Activity 99. Schmidt, M. Power functional theory for Newtonian many-body dynamics. J.
Landscapes Polarize Microswimmers without Alignment Forces. Phys. Rev. Chem. Phys. 148, 044502 (2018).
Lett. 126, 228001 (2021). 100. Janssen, L. M. C. Mode-coupling theory of the glass transition: A primer.
67. Auschra, S., Holubec, V., Söker, N. A., Cichos, F. & Kroy, K. Polarization- Front. Phys. 6, 97 (2018).
density patterns of active particles in motility gradients. Phys. Rev. E 103, 101. Vogel, F. & Fuchs, M. Stress correlation function and linear response of
062601 (2021). Brownian particles. Eur. Phys. J. E 43, 70 (2020).
68. Speck, T. & Jack, R. L. Ideal bulk pressure of active Brownian particles. Phys. 102. Triezenberg, D. G. & Zwanzig, R. Fluctuation theory of surface tension. Phys.
Rev. E 93, 062605 (2016). Rev. Lett. 28, 1183 (1972).
69. Takatori, S. C., Yan, W. & Brady, J. F. Swim pressure: stress generation in 103. Yvon, J. Actualités Scientifiques et Industrielles (Hermann & Cie., 1935).
active matter. Phys. Rev. Lett. 113, 028103 (2014). 104. Born, M. & Green, H. S. A general kinetic theory of liquids I. The molecular
70. Yang, X., Manning, M. L. & Marchetti, M. C. Aggregation and segregation of distribution functions. Proc. R. Soc. London, Ser. A 188, 10 (1946).
confined active particles. Soft. Matter. 10, 6477 (2014).
71. Winkler, R. G., Wysocki, A. & Gompper, G. Virial pressure in systems of
spherical active Brownian particles. Soft. Matter. 11, 6680 (2015). Acknowledgements
72. Solon, A. P. et al. Pressure and phase equilibria in interacting active Brownian This work is supported by the German Research Foundation (DFG) via project number
spheres. Phys. Rev. Lett. 114, 198301 (2015). 436306241. Open Access funding is provided by project DEAL. We would like to thank
73. Paliwal, S., Rodenburg, J., van Roij, R. & Dijkstra, M. Chemical potential in Matthias Fuchs, Johannes Häring, Bob Evans, Pedro Tarazona, Daniel de las Heras,
active systems: predicting phase equilibrium from bulk equations of state? Matthias Krüger, Jim Lutsko, Hartmut Löwen, and Thomas Fischer for useful discussions
New J. Phys. 20, 015003 (2018). and comments.
74. de las Heras, D. et al. Floating nematic phase in colloidal platelet-sphere
mixtures. Sci. Rep. 2, 789 (2012). Author contributions
75. de las Heras, D. & Schmidt, M. The phase stacking diagram of colloidal S.H. and M.S. have jointly carried out the work and written the paper.
mixtures under gravity. Soft Matter 9, 98636 (2013).
76. de las Heras, D. & Schmidt, M. Sedimentation stacking diagram of binary
colloidal mixtures and bulk phases in the plane of chemical potentials. J. Phys.: Funding
Condens. Matter 27, 194115 (2015). Open Access funding enabled and organized by Projekt DEAL.
77. Geigenfeind, T. & de las Heras, D. The role of sample height in the stacking
diagram of colloidal mixtures under gravity. J. Phys.: Condens. Matter 29,
064006 (2017).
Competing interests
The authors declare no competing interests.
78. Enculescu, M. & Stark, H. Active colloidal suspensions exhibit polar order
under gravity. Phys. Rev. Lett. 107, 058301 (2011).
79. Hermann, S. & Schmidt, M. Active ideal sedimentation: exact two- Additional information
dimensional steady states. Soft Matter 14, 1614 (2018). Supplementary information The online version contains supplementary material
80. Vachier, J. & Mazza, M. G. Dynamics of sedimenting active Brownian available at https://doi.org/10.1038/s42005-021-00669-2.
particles. Eur. Phys. J. E 42, 11 (2019).
81. Solon, A. P., Cates, M. E. & Tailleur, J. Active Brownian particles and run-and- Correspondence and requests for materials should be addressed to S.H. or M.S.
tumble particles: A comparative study. Eur. Phys. J. Spec. Top. 224, 1231 (2015).
82. Lee, C. F. Active particles under confinement: aggregation at the wall and Peer review information Communications Physics thanks Benjamin Goddard, Gyula
gradient formation inside a channel. New J. Phys. 15, 055007 (2013). Toth and the other, anonymous, reviewer(s) for their contribution to the peer review of
83. Zwanzig, R. Nonequilibrium statistical mechanics (Oxford University Press, this work. Peer reviewer reports are available.
2001).
84. Lesnicki, D., Vuilleumier, R., Carof, A. & Rotenberg, B. Molecular Reprints and permission information is available at http://www.nature.com/reprints
hydrodynamics from memory kernels. Phys. Rev. Lett. 116, 147804 (2016).
85. Lesnicki, D. & Vuilleumier, R. Microscopic flow around a diffusing particle. J. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
Chem. Phys. 147, 094502 (2017). published maps and institutional affiliations.
86. Jung, G. & Schmid, F. Computing bulk and shear viscosities from simulations
of fluids with dissipative and stochastic interactions. J. Chem. Phys. 144,
204104 (2016).
Open Access This article is licensed under a Creative Commons
87. Jung, G., Hanke, M. & Schmid, F. Iterative reconstruction of memory kernels.
Attribution 4.0 International License, which permits use, sharing,
J. Chem. Theo. Comput. 13, 2481 (2017).
adaptation, distribution and reproduction in any medium or format, as long as you give
88. Jung, G., Hanke, M. & Schmid, F. Generalized Langevin dynamics:
appropriate credit to the original author(s) and the source, provide a link to the Creative
construction and numerical integration of non-Markovian particle-based
Commons license, and indicate if changes were made. The images or other third party
models. Soft Matter 14, 9368 (2018).
material in this article are included in the article’s Creative Commons license, unless
89. Yeomans-Reyna, L. & Medina-Noyola, M. Overdamped van Hove function of
colloidal suspensions. Phys. Rev. E 62, 3382 (2000). indicated otherwise in a credit line to the material. If material is not included in the
90. Chávez-Rojo, M. A. & Medina-Noyola, M. Van Hove function of colloidal article’s Creative Commons license and your intended use is not permitted by statutory
mixtures: Exact results. Physica A 366, 55 (2006). regulation or exceeds the permitted use, you will need to obtain permission directly from
91. Lázaro-Lázaro, E. et al. Self-consistent generalized Langevin equation theory the copyright holder. To view a copy of this license, visit http://creativecommons.org/
of the dynamics of multicomponent atomic liquids. J. Chem. Phys. 146, licenses/by/4.0/.
184506 (2017).
92. Treffenstädt, L. L. & Schmidt, M. Memory-induced motion reversal in
© The Author(s) 2021
Brownian liquids. Soft Matter 16, 1518 (2020).

COMMUNICATIONS PHYSICS | (2021)4:176 | https://doi.org/10.1038/s42005-021-00669-2 | www.nature.com/commsphys 13