---
title: Kirkwood–Salsburg Hierarchy and Equilibrium Correlations
url: https://www.emergentmind.com/topics/kirkwood-salsburg-hierarchy
type: topic
---

# Kirkwood–Salsburg Hierarchy and Equilibrium Correlations

Searching arXiv for recent and foundational papers on the Kirkwood–Salsburg hierarchy.
The Kirkwood–Salsburg hierarchy is the infinite recursive system that governs the correlation functions of grand-canonical equilibrium particle systems. In its classical form it is a coupled family of linear inhomogeneous integral equations for all \(n\)-point functions simultaneously, but it also admits operator, graph-theoretic, and density-side formulations. Across these formulations, the hierarchy serves as a common framework for analyticity of correlation functions, convergence of cluster and virial expansions, equivalence with stationary BBGKY equations, and more recent constructions at negative activity and with multi-body interactions [1611.06912][1205.2788][2009.09211].

## 1. Classical hierarchy for equilibrium correlation functions

For a classical continuous system in a bounded volume \(A\subset \mathbb R^\nu\) in the grand-canonical ensemble, the \(n\)-point correlation functions are defined by
\[
\rho_A(z;(x)_n) = \frac{1}{Z_A(z)} \sum_{m=0}^\infty \frac{z^m}{m!} \int_{A^m} d(y)^m\, e^{-\beta U((x)_n,(y)_m)},
\]
with grand-canonical partition function
\[
Z_A(z) = 1+\sum_{m=1}^\infty \frac{z^m}{m!} \int_{A^m} d(y)^m\, e^{-\beta U(y_m)}.
\]
The standard assumptions in this formulation are stability,
\[
U(x_n)\ge -Bn,
\]
and regularity,
\[
C := \int_{\mathbb{R}^\nu} \bigl|e^{-\beta \varphi(x)}-1\bigr|\,dx < \infty .
\]
Under these assumptions, the correlation functions satisfy the Kirkwood–Salsburg equations [1611.06912].

The hierarchy is written as
\[
\rho_A(z;x_1) = z\, \chi_A(x_1)\,\bigl(1+(K_A\rho_A)(x_1)\bigr),
\]
and for \(n\ge 2\),
\[
\rho_A(z;(x)_n) = z\, \chi_A(x_n)\,(K_A\rho_A)(x_n).
\]
Here \(K_A\) is the Kirkwood–Salsburg operator,
\[
(K_A\rho_A)(x_n) = e^{-\beta W(x_1;(x)'_n)} \sum_{m=0}^\infty \frac{1}{m!}\int_{A^m} d(y)^m\, K(x_1;(y)_m)\,\rho_A((x)'_n,(y)_m),
\]
with
\[
W(x_1;(x)'_n)=U(x_n)-U((x)'_n), \qquad
K(x_1;(y)_m)=\prod_{k=1}^m\left(e^{-\beta\varphi(|x_1-y_k|)}-1\right).
\]
This exhibits the hierarchy as one equation for each particle number, with the \(n\)-point function coupled to higher-order functions through the Mayer kernel [1611.06912].

A closely related formulation appears for grand-canonical Gibbs measures with pairwise interactions in general measurable spaces. If \(v(x,y)\) is the pair potential and
\[
f(x,y)=e^{-v(x,y)}-1,
\]
then graph weights are
\[
w(G;x_1,\dots,x_n)=\prod_{\{i,j\}\in E(G)} f(x_i,x_j).
\]
In this setting the factorial moment measures \(p_s\), also described as \(s\)-point correlation functions, satisfy
\[
p_1(dx_1) = z(dx_1)\left(1+\sum_{n=1}^{\infty}\frac1{n!} \int_{X^n}\prod_{i=1}^n f(x_1,y_i)\, p_n(dy_1,\dots,dy_n)\right),
\]
and for \(s\ge 2\),
\[
p_s(d(x_1,\dots,x_s)) = \prod_{i=2}^s(1+f(x_1,x_i))\,z(dx_1)\,p_{s-1}(d(x_2,\dots,x_s)) + \sum_{n=1}^{\infty}\frac1{n!} \int_{X^n} \prod_{i=1}^n f(x_1,y_i)\, p_{s-1+n}(d(x_2,\dots,x_s,y_1,\dots,y_n)).
\]
These are presented as the classical hierarchy for correlation functions in the Gibbsian setting [2009.09211].

## 2. Operator-theoretic structure and spectral interpretation

A central feature of the hierarchy is that it can be encoded by a single linear operator on a Banach space of sequences. One convenient choice is
\[
\mathcal{E}_\xi(A) = \left\{ \delta=(\delta^{(n)})_{n\ge1} : \|\delta\|_\xi = \sup_{n\ge1}\operatorname*{ess\,sup}_{(x)_n\in A^n} |\delta^{(n)}(x_n)|\,\xi^{-n} <\infty \right\}.
\]
In this language the hierarchy becomes
\[
(I-zK_A)\rho_A = z\,\phi_A,
\]
or, with \(\lambda=z^{-1}\),
\[
(K_A-\lambda)\rho_A=-\phi_A.
\]
The source vector \(\phi_A\) has \(\phi_A^{(1)}=1\) and \(\phi_A^{(n)}=0\) for \(n\ge2\), so solving the hierarchy is reduced to the resolvent
\[
R(\lambda):=(K_A-\lambda)^{-1}.
\]
The coefficient-space picture makes the right-shift-like structure explicit:
\[
K_A\{a_m\}_{m\ge1} = \{a_{m-1}\}_{m\ge1}.
\]
This shift-like behavior is the operator-theoretic explanation for the spectral structure emphasized in the literature [1611.06912].

In the settings treated there, \(K_A\) can be realized on Banach subspaces for which the spectrum is purely point spectrum and is identified with the inverse zeros of the partition function:
\[
\lambda\neq 0 \in \sigma(K_A) \quad \Longleftrightarrow \quad z=\lambda^{-1} \text{ is a zero of } Z_A(z).
\]
If \(\lambda_c=z_c^{-1}\) is an eigenvalue of largest modulus, \(|\lambda_c|=r(K_A)\), then the resolvent has a Laurent expansion
\[
R(\lambda) = -\frac{P}{\lambda-\lambda_c} + \sum_{n\ge 0} (\lambda-\lambda_c)^n S^{\,n+1},
\]
with projection operators
\[
P=\frac{1}{2\pi i}\oint_C R(\lambda)\,d\lambda, \qquad S=\frac{1}{2\pi i}\oint_C (\lambda-\lambda_c)^{-1}R(\lambda)\,d\lambda .
\]
The pole order is \(1\), the dominant eigenvalue is simple, and the smallest zero of the partition function is therefore simple as well. Correspondingly, the correlation functions satisfy
\[
\rho_A(z;(x)_n) \sim \frac{M_n z}{1-z/z_c}, \qquad z\to z_c,
\]
so the critical singularity is first order [1611.06912].

The same analysis gives explicit analyticity information. For stable and regular potentials,
\[
r(K_A)\le C^{-1},
\]
hence the activity-domain of analyticity contains \(|z|<C^{-1}\). For positive or hard-core potentials, the one-point function has convergence radius exactly \(C^{-1}\), the singularity occurs at
\[
z_c=-C^{-1},
\]
and the virial expansion radius is at least
\[
(2C)^{-1}.
\]
These results make the hierarchy a direct analytic link between correlation functions, partition-function zeros, and the boundary of convergence of activity expansions [1611.06912].

## 3. Equivalence with stationary BBGKY and Gibbs correlation structures

For infinite classical systems at equilibrium, the hierarchy is also the rigorous integral form of the stationary BBGKY hierarchy. In the setting of a smooth Maxwellian equilibrium state, the spatial correlation functions \(\mathcal P_n\) satisfy
\[
\nabla_{q_1}\mathcal P_n(q_1,\dots,q_n) = -\beta\,\mathcal P_n(q_1,\dots,q_n)\,\nabla_{q_1}W_{q_1}(q_2,\dots,q_n) -\beta\int_{\mathbb R^v} \nabla_{q_1}\varphi(q_1-y)\,\mathcal P_{n+1}(q_1,\dots,q_n,y)\,dy,
\]
with
\[
W_{q_1}(q_2,\dots,q_n)=\sum_{j=2}^n \varphi(q_1-q_j).
\]
Assuming stability, regularity, smoothness, and a weak cluster property, a direct iterative integration along a path in configuration space yields the Kirkwood–Salsburg equations
\[
\mathcal P_n(q_1,\dots,q_n) = z\,e^{-\beta W_{q_1}(q_2,\dots,q_n)} \left[ \mathcal P_{n-1}(q_2,\dots,q_n) + \sum_{m=1}^\infty \frac{(-1)^m}{m!} \int_{\mathbb R^{vm}} \prod_{j=1}^m (1-e^{-\beta\varphi(q_1-y_j)}) \,\mathcal P_{n-1+m}(q_2,\dots,q_n,y_1,\dots,y_m)\,dy \right].
\]
Under these assumptions the stationary BBGKY hierarchy and the Kirkwood–Salsburg equations are equivalent, and at low density the solution is unique. The cited sufficient bound is
\[
|z| < \bigl(I_\beta e^{1+2\beta B}\bigr)^{-1}, \qquad
I_\beta=\int_{\mathbb R^v}|1-e^{-\beta\varphi(x)}|\,dx,
\]
together with the smallness condition
\[
p< \left(2I_\beta e^{1+2\beta B}\right)^{-1}.
\]
This gives a rigorous route from the differential hierarchy to the integral hierarchy without assuming the latter at the outset [1205.2788].

The Gibbsian meaning of the hierarchy is made explicit through factorial moment measures and the Georgii–Nguyen–Zessin equation. For a grand-canonical Gibbs measure on a general measure space \((X,\mathcal X)\), the correlation functions can be written as
\[
p_s(d(x_1,\dots,x_s)) = g(x_1,\dots,x_s;p)\,p(dx_1)\cdots p(dx_s),
\]
and, under the hypotheses of the density-side convergence theorem discussed below, the family
\[
p_s(d(x_1,\dots,x_s)) = g(x_1,\dots,x_s;p)\,p(dx_1)\cdots p(dx_s)
\]
satisfies the Kirkwood–Salsburg equations at the activity
\[
z(dq)=p(dq)\exp\!\left(-\sum_{n=1}^{\infty}\frac1{n!}\int_{X^n} D_{n+1}(q,x_1,\dots,x_n)\,p(dx_1)\cdots p(dx_n)\right).
\]
This identifies the hierarchy not merely as a formal recursion but as an exact relation for actual Gibbs correlation functions [2009.09211].

## 4. Density-side reformulation, graph recursions, and virial control

A major reformulation replaces the usual activity expansion by a density expansion. For the factorial moment measures \(p_s\), the paper works with the functions \(g(x_1,\dots,x_s;p)\) defined through
\[
p_s(d(x_1,\dots,x_s)) = g(x_1,\dots,x_s;p)\,p(dx_1)\cdots p(dx_s),
\]
and writes the density expansion
\[
g(x_1,\dots,x_s;p) = \varphi(x_1,\dots,x_s;\varnothing) + \sum_{n=1}^{\infty}\frac1{n!}\int_{X^n} \varphi(x_1,\dots,x_s;x_{s+1},\dots,x_{s+n}) \,p(dx_{s+1})\cdots p(dx_{s+n}).
\]
Here \(\varphi\) is the generating function for the graph class \(D(W,B)\), consisting of graphs in which every black vertex is connected to white vertices and this property survives removal of any vertex. This is presented as the density-side analogue of the usual activity expansion [2009.09211].

The graph-theoretic core is a vertex-removal decomposition of weighted \(2\)-connected graphs. Proposition 7 gives
\[
D(\{1,\dots,n\}) = \sum_{L\subset\{2,\dots,n\}} \prod_{\ell\in L} f(x_1,x_\ell)\, dc(L,\{2,\dots,n\}\setminus L),
\]
followed by the Möbius inversion formula on the partition lattice,
\[
dc(L,J) = \sum_{m\ge 1}(-1)^{m-1}(m-1)! \sum_{\{V_1,\dots,V_m\}\in P(L\cup J)} \prod_{r=1}^{m} D(L\cap V_r, J\cap V_r).
\]
The same inversion principle is recorded abstractly as
\[
a(V)=\sum_{\{V_1,\dots,V_m\}\in P(V)}\prod_{r=1}^m b(V_r)
\quad\Longrightarrow\quad
b(V)=\sum_{m\ge 1}(-1)^{m-1}(m-1)!\sum_{\{V_1,\dots,V_m\}\in P(V)}\prod_{r=1}^m a(V_r).
\]
Combining these identities yields a recurrence for \(\varphi(I;J)\) that is explicitly described as similar to the recurrence relations of the coefficients of the activity expansions of the correlation functions inherited from the well-known Kirkwood–Salsburg equation. The resulting recursion is a density-hierarchy obtained by removing one vertex and its incident edges, with an extra Möbius inversion step that eliminates the activity variable [2009.09211].

The convergence theorem is expressed through an operator \(K_p\) and the condition
\[
(K_pm)(x_1,\dots,x_s)\le m(x_1,\dots,x_s),\qquad (R_pm)(x_1)<1,
\]
where
\[
(R_pm)(x_1)= \sum_{j=1}^{\infty}\frac1{j!} \int_{X^j}\prod_{i=1}^j |f(x_1,y_i)|\,m(y_1,\dots,y_j)\,p(dy_1)\cdots p(dy_j).
\]
Under this hypothesis one obtains
\[
g(x_1,\dots,x_s;p)\le m(x_1,\dots,x_s), \qquad d(x_1;p)\le -\log\bigl(1-(R_pm)(x_1)\bigr).
\]
The paper emphasizes that this supplies a constructive bound ensuring that the hierarchy has an absolutely convergent solution in the density domain. For inhomogeneous systems the theorem is formulated on a general measure space and allows weights of the form
\[
m(x_1,\dots,x_s)=e^{b(x_1)+\cdots+b(x_s)}.
\]
A plausible implication is that the hierarchy is not intrinsically tied to translation invariance or even to activity variables; its recursive content survives in a density-based, inhomogeneous setting [2009.09211].

## 5. Absolute convergence and the sign-flipped Kirkwood–Salsburg operator

The activity expansion of correlation functions can also be organized through multi-rooted graph coefficients. In this formulation,
\[
\rho_n(x_1,\ldots,x_n;z) =\sum_{k=0}^\infty \frac{1}{k!} \int_{\mathbb X^k} \psi_{n,n+k}(x_1,\ldots,x_n,y_1,\ldots,y_k)\, z(x_1)\cdots z(x_n)\,\lambda_z^k(dy),
\]
where \(\psi_{n,n+k}\) sums graph weights over graphs on \([n+k]\) such that each non-root vertex is connected to at least one root, and each graph has weight
\[
w(G;x_1,\ldots,x_{n+k})=\prod_{\{i,j\}\in E(G)} f(x_i,x_j), \qquad f(x,y)=e^{-v(x,y)}-1.
\]
Using a selection rule that chooses one current root, the recursion can be rewritten through a sign-flipped Kirkwood–Salsburg operator \(\tilde K_z^s\), obtained from the standard recursion by replacing the relevant factors \(f\) with \(|f|\). The resulting hierarchy is
\[
\tilde\rho(z)\le e_z+\tilde K_z^s \tilde\rho(z)
\]
for general pair potentials, and
\[
\tilde\rho(z)= e_z+\tilde K_z^s \tilde\rho(z)
\]
for non-negative pair potentials, where \(e_z\) has \((e_z)_1(x)=z(x)\) and \((e_z)_n=0\) for \(n\neq 1\). For repulsive or hard-core interactions, \(f\le 0\), so \(|f|=-f\), and the sign flip turns the alternating-sign combinatorics into a positive operator inequality adapted to absolute convergence [2112.13134].

The main criterion is the existence of a measurable symmetric nonnegative supersolution \(\xi=(\xi_n)_{n\ge1}\) such that
\[
z(x_1)\delta_{n,1}+(\tilde K_z^s\xi)_n(x_1,\ldots,x_n)\le \xi_n(x_1,\ldots,x_n)
\]
for all \(n\). This is sufficient for absolute convergence of the activity expansions, and for non-negative pair potentials it is also necessary; in that case one may take \(\xi=\tilde\rho\). The necessity relies on the alternating-sign identity
\[
\psi_{n,n+k}(x_1,\ldots,x_{n+k}) = (-1)^k \bigl|\psi_{n,n+k}(x_1,\ldots,x_{n+k})\bigr|
\]
for non-negative potentials. The operator inequality
\[
e_z+\tilde K_z^s\xi\le \xi
\]
thus becomes an if and only if criterion in the non-negative case [2112.13134].

Classical convergence criteria reappear as special choices of \(\xi\) and of the selection rule. The Kotecký–Preiss condition is recovered from the ansatz
\[
\xi_n(x_1,\ldots,x_n)=z(x_1)\cdots z(x_n)e^{a(x_1)+\cdots+a(x_n)},
\]
while the Fernández–Procacci condition yields the bound
\[
\rho_n(x_1,\ldots,x_n;z)\le \prod_{1\le i<j\le n}(1+f(x_i,x_j))\prod_{i=1}^n \mu(x_i).
\]
The same framework also produces new sufficient conditions for hard-core systems in \(\mathbb R^d\) and \(\mathbb Z^d\), and for abstract polymer systems; for subset polymers on \(\mathbb Z^d\), the cited example gives a bound about \(6\%\) better than the Fernández–Procacci bound for a sample model of hard cubes [2112.13134].

## 6. Negative activity, closure processes, and multi-body extensions

The hierarchy has recently been used to construct point processes from negative-activity solutions. A Kirkwood closure process is defined by exact correlation functions
\[
\rho^{(n)}(\mathbf x_n)= \varsigma^n \prod_{1\le i<j\le n}\phi(x_i-x_j),
\]
and for \(\phi=e^{-\beta u}\),
\[
\rho^{(n)}(\mathbf x_n) = \varsigma^n e^{-\beta H(\mathbf x_n)}.
\]
If \(u\) is a stable and regular pair potential, \(\beta>0\), and
\[
0<z<z_0,\qquad z_0:=\left(e^{2\beta B+1}C_\beta\right)^{-1}, \qquad
C_\beta:=\int_{\mathbb R^d}|f_\beta(x)|\,dx,
\]
then the process exists for \(\varsigma=z\) and \(\phi=e^{-\beta u}\). The construction proceeds by solving finite-volume Kirkwood–Salsburg equations for complex activity, evaluating the solution at negative activity, and proving positivity of the alternating-sign quantities
\[
(-1)^n\theta_\Lambda^{(n)}(-z;\mathbf x_n)\ge0.
\]
A central identity is the Janossy formula
\[
j_\Lambda^{(n)}(z;\mathbf x_n) = (-1)^n \Xi_\Lambda(-z)\,\theta_\Lambda^{(n)}(-z;\mathbf x_n),
\]
which allows one to verify Lenard positivity and thereby realize the prescribed correlations as those of a genuine point process [2506.08242].

Under the stronger assumptions of local stability, regularity, and lower regularity, the closure process is Gibbs and its Papangelou kernel satisfies a KS-type hierarchy. Writing
\[
\iota^{(n)}(z;\mathbf x_n):=(-1)^n\theta^{(n)}(-z;\mathbf x_n),
\]
the infinite-volume Kirkwood–Salsburg equation
\[
(\mathrm I-z\,\Pi\mathcal K)\boldsymbol{\theta}=z\,\mathbf e
\]
translates into a sign-corrected recursion for \(\iota^{(n)}\), while the multivariate GNZ equation is satisfied with a kernel derived from a KS-type equation on configuration space. This shows that negative-activity KS solutions can encode not only correlation functions but also the Gibbsian insertion structure of the resulting process [2506.08242].

A different extension treats finite-volume lattice gases with possibly complex-valued multi-body interactions. There the hierarchy is written for finite-set correlations \(R(X,\Lambda\mid B)\) using the kernel
\[
\gamma(s,N\mid B) := \sum_{M\subset N} (-1)^{|N\setminus M|}\,\kappa(s\mid B\cup M),
\]
which is Möbius-dual to \(\kappa(s\mid B\cup M)=\sum_{N\subset M}\gamma(s,N\mid B)\). The KS equation becomes
\[
R(\{s\}\cup T,\Lambda) = \sum_{N\subset \Lambda\setminus T} z(s)\,\gamma(s,N\mid T)\, R(T\cup N,\Lambda).
\]
A uniqueness proposition shows that any solution \(\rho\) of this hierarchy must be proportional to the finite-volume partition functions:
\[
\rho(X)=Z(X,\Lambda)\,\mu(\varnothing).
\]
Hence a nonzero solution implies \(Z(\Lambda)\neq 0\). The corresponding domination principle uses a positive operator \(\tilde K_\Lambda\) and a supersolution \(\xi\) with \(\tilde K_\Lambda\xi\le \xi\), while the key new estimate comes from a partition scheme for coverings that sharpens the classical Gallavotti–Miracle-Solé bound. In the absolutely summable potential case this yields the sufficient condition
\[
2|z(x)W(x)|\,e^{D_V(x)}\le 1,
\]
presented as an improvement of the Gallavotti–Miracle-Solé bound and of the Gallavotti–Miracle-Solé–Robinson refinement [2508.12078].

Taken together, these developments indicate that the Kirkwood–Salsburg hierarchy is best understood not as a single formula but as a family of equivalent recursive architectures. In the supplied literature it appears as an integral hierarchy for equilibrium correlations, an operator equation with a tractable spectrum, a direct integral form of the stationary BBGKY hierarchy, a density-side recursion derived from \(2\)-connected graphs and Möbius inversion, a positive fixed-point problem for absolute convergence, and a vehicle for negative-activity and multi-body constructions [1611.06912][1205.2788][2009.09211][2112.13134][2506.08242][2508.12078].

Source: https://www.emergentmind.com/topics/kirkwood-salsburg-hierarchy