---
title: Smirnov-Type Decomposition
url: https://www.emergentmind.com/topics/smirnov-type-decomposition
type: topic
---

# Smirnov-Type Decomposition

Smirnov-type decomposition denotes, in the sources considered here, a family of representation theorems associated with S. Smirnov and later analogues. In its classical form, it states that a divergence-free vector-valued Radon measure, equivalently a \(1\)-current of order \(0\), can be written as a superposition of elementary curve-currents with an exact equality of masses [2405.13406]. Related constructions appear for \(N\)-flows on countable hypergraphs, where a finite-mass flow splits uniquely into acyclic, solenoidal, and finite-cycle parts [1903.09817]; for horizontal vector charges in the Heisenberg group, where divergence-free horizontal currents are decomposed into horizontal curves [2605.12716]; and in Hilbert function space theory, where “Smirnov-type” takes the form of factorization \(h=b/a\) with an outer denominator [1806.05270]. This suggests a unifying template: an object with a conservation law or dense-range property is represented by elementary constituents without loss of mass or norm.

## 1. Classical decomposition into elementary curve-currents

Let \(\Omega\subset \mathbb{R}^n\) be open, and let \(u=(u_1,\dots,u_n)\in M(\Omega;\mathbb{R}^n)\) be a finite \(\mathbb{R}^n\)-valued Radon measure. Its action on test fields \(\phi\in C_c^\infty(\Omega)^n\) is
\[
\langle u,\phi\rangle = \sum_{k=1}^n \int_\Omega \phi_k\,d u_k,
\]
its total variation is the positive Radon measure
\[
|u|(E)=\sup\{\langle u,\phi\rangle: \phi\in C_c^\infty(E)^n,\ |\phi(x)|\le 1\},
\]
and its distributional divergence is
\[
\mathrm{Div}\,u=\sum_{k=1}^n \partial_k u_k,
\qquad
\langle \mathrm{Div}\,u,\psi\rangle=-\langle u,\nabla\psi\rangle.
\]
The measure is divergence-free when \(\mathrm{Div}\,u=0\) in \(\mathcal{D}'(\Omega)\) [2405.13406].

If \(I=[a,b]\) and \(\gamma:I\to\Omega\) is Lipschitz, then \(\gamma\) defines the elementary curve-current
\[
\langle [\gamma],\phi\rangle=\int_a^b \langle \phi(\gamma(t)),\dot\gamma(t)\rangle\,dt,
\]
with
\[
\mathrm{Div}\,[\gamma]=\delta_{\gamma(a)}-\delta_{\gamma(b)}.
\]
Its total variation satisfies
\[
|[\gamma]|(E)\le \int_{\gamma^{-1}(E)} |\dot\gamma(t)|\,dt,
\qquad
L(\gamma)=\int_a^b |\dot\gamma(t)|\,dt,
\]
and equality occurs exactly in the no-backtracking case:
\[
|[\gamma]|(E)=\int_{\gamma^{-1}(E)}|\dot\gamma|\,dt
\quad\text{for every Borel }E\subset \Omega.
\]
This identifies the elementary constituents that appear in Smirnov’s theorem [2405.13406].

For fixed \(\ell>0\), let
\[
K_\ell=\{\gamma:[0,\ell]\to\Omega:\gamma \text{ is Lipschitz with }|\dot\gamma(t)|\le 1 \text{ a.e.}\},
\]
equipped with the topology of uniform convergence. The classical theorem states that if \(u\in M(\Omega;\mathbb{R}^n)\) is finite and divergence-free, then for every \(\ell>0\) there exists a finite positive Borel measure \(\eta\) on \(K_\ell\) such that
\[
u=\int_{K_\ell}[\gamma]\,d\eta(\gamma),
\qquad
|u|(\Omega)=\int_{K_\ell} L(\gamma)\,d\eta(\gamma),
\]
and for \(\eta\)-almost every \(\gamma\),
\[
\gamma([0,\ell])\subset \mathrm{supp}\,u,
\qquad
|[\gamma]|(\Omega)=L(\gamma)=\ell.
\]
Here \(\eta\) is the Smirnov measure, and the second identity is the equality of masses [2405.13406].

## 2. Proof architecture and measure-theoretic identities

A streamlined proof proceeds in three steps. First, one approximates \(u\) by smooth divergence-free fields \(u_\varepsilon=u*k_\varepsilon\). Second, at each \(x\) one considers the flow line \(\Phi(t,x)\) solving
\[
\frac{d}{dt}\Phi(t,x)=\frac{u_\varepsilon(\Phi(t,x))}{|u_\varepsilon(\Phi(t,x))|},
\qquad
\Phi(0,x)=x,
\]
on \([0,\ell]\). Liouville’s theorem for divergence-free flows shows that the push-forward of \(\rho_\varepsilon=|u_\varepsilon|\,dx\) under \(x\mapsto \Phi(\cdot,x)\) is time-independent, yielding
\[
u_\varepsilon
=
\int_\Omega [\,t\mapsto \Phi(t,x)\,]\,d(\rho_\varepsilon/\ell)(x)
=
\int_{K_\ell} [\gamma]\,d\eta_\varepsilon(\gamma).
\]
Third, Prokhorov’s theorem, or Banach–Alaoglu on \(C(K_\ell)^*\), gives a weakly convergent subsequence \(\eta_{\varepsilon_j}\rightharpoonup \eta\); continuity of \([\gamma](\phi)\) in \(\gamma\) allows passage to the limit and recovery of \(u=\int [\gamma]\,d\eta\) [2405.13406].

The mass identity is not auxiliary but structural. Testing against vector fields of unit norm supported in open sets yields
\[
|u|(\Omega)=\int L(\gamma)\,d\eta(\gamma),
\]
from which one deduces both \(L(\gamma)=\ell\) and \(\mathrm{supp}\,\gamma\subset \mathrm{supp}\,u\) for \(\eta\)-almost every \(\gamma\). In the 2024 note, these equalities are presented alongside the curve-current action, the mass formula
\[
|u|(\Omega)=\sup\{\langle u,\varphi\rangle:\varphi\in\mathcal{D}(\Omega)^n,\ |\varphi|\le 1\},
\]
the divergence identity for curves, and the no-backtracking criterion [2405.13406].

The same source also records broader directions: variants for currents of higher dimension, for elementary solenoids parametrized on \(\mathbb{R}\), and in metric spaces; and applications to optimal transport and metric currents through the interpretation of \(\eta\) as a transport plan or as a decomposition in the sense of Ambrosio–Kirchheim [2405.13406].

## 3. Smirnov-type decomposition for \(N\)-flows

In multimarginal transport, the relevant combinatorial object is an \(N\)-flow. Let \(X\) be a Polish space and \(c:X^N\to (-\infty,+\infty]\). One forms the \(N\)-partite directed hypergraph
\[
\widetilde X_N=X\times\{1,\dots,N\},
\qquad
\widetilde E_N=\{(x_1,\dots,x_N)\in X^N:\ c(x_1,\dots,x_N)<+\infty\}.
\]
A signed \(N\)-flow is a finitely supported real-valued function \(m:\widetilde E_N\to\mathbb{R}\), written formally as
\[
[A]=\sum_{\vec x\in\widetilde E_N} m(\vec x)\,[\vec x],
\]
with mass
\[
|[A]|=\sum_{\vec x} |m(\vec x)|<+\infty,
\]
and \(N\)-boundary
\[
\partial^{(N)}[A]=(\mu_1,\dots,\mu_N),
\qquad
\mu_i=\sum_{\vec x=(x_1,\dots,x_N)} m(\vec x)\,\delta_{x_i}.
\]
The flow is closed, or a cycle, when \(\partial^{(N)}[A]=0\) [1903.09817].

The Smirnov-type decomposition theorem for \(N\)-flows states that any finite-mass \(N\)-flow on a countable hypergraph admits unique subflows
\[
[A_1],\ [A_2],\ [A_3]
\]
such that
\[
[A]=[A_1]+[A_2]+[A_3],
\qquad
|[A]|=|[A_1]|+|[A_2]|+|[A_3]|.
\]
Here \([A_1]\) is acyclic, meaning it has no nonzero closed subflows; \([A_3]\) is a finite or countable superposition of finite closed \(N\)-flows; and \([A_2]\) is solenoidal, meaning it is closed but contains no finite cycles [1903.09817].

The proof removes finite closed subflows first, always choosing one of largest mass, then removes acyclic subflows in decreasing order of boundary-mass, and identifies the remainder as closed and cycle-free in the finite sense. A diagonal argument yields convergence of the peeled-off pieces in mass, and uniqueness follows because solenoidal and acyclic flows cannot share any nonzero subflow [1903.09817]. In the case \(N=2\), every finite closed \(2\)-flow is a finite sum of elementary loops, the Euler–Kirchhoff decomposition on a directed graph; for \(N\ge 3\), that simple loop decomposition fails, which is precisely why the more elaborate trichotomy is needed [1903.09817].

## 4. Multimarginal transport and the failure of the \(N=2\) intuition

The same framework is used to analyze \(c\)-cyclical monotonicity in \(N\)-marginal transport. A central combinatorial statement is that a set \(A\subset X^N\) is \(c\)-monotone if and only if every difference \([F]-[F']\) of two finite-mass \(N\)-flows supported in \(A\) with the same boundary has nonpositive \(c\)-weight. Consequently, if the \(N\)-graph of \(c\) admits no nonzero finite closed \(N\)-flow, then every plan supported in \(\{c<+\infty\}\) is trivially \(c\)-monotone, because there is no competitor \([F']\) to test [1903.09817].

This mechanism underlies a counterexample for \(N=3\). In the construction, \(X=\mathbb{N}\) and
\[
c(a,b,c)=
\begin{cases}
1,&(a,b,c)=(1,1,1),\\
f(a),&\text{if }a=b\text{ and }c=a+1\text{ (and symmetrically under permutations)},\\
+\infty,&\text{otherwise},
\end{cases}
\]
for a suitably chosen bounded \(f(a)\). The resulting \(3\)-graph has vertices \(\mathbb{N}\times\{1,2,3\}\), edges \((1,1,1)\), and for each \(k\ge 1\) the three permutations of \((k,k,k+1)\). It has no nontrivial finite \(3\)-cycles, so every finite \(3\)-flow with zero boundary vanishes [1903.09817].

Nevertheless, the paper constructs two symmetric transport plans
\[
\gamma=\sum_{k\ge 1}4^{-k}\bigl(\delta_{2k-1,2k-1,2k}+\mathrm{sym}\bigr),
\qquad
\bar\gamma=\tfrac12\,\delta_{1,1,1}+\tfrac12\sum_{k\ge 1}4^{-k}\bigl(\delta_{2k,2k,2k+1}+\mathrm{sym}\bigr),
\]
with the same one-point marginal \(\mu\) but satisfying
\[
\langle c,\bar\gamma\rangle<\langle c,\gamma\rangle.
\]
Thus \(\gamma\) is nonoptimal despite being \(c\)-cyclically monotone [1903.09817]. The paper identifies \(\gamma-\bar\gamma\) as a closed \(3\)-flow of infinite support, namely a solenoidal flow. The usual \(N=2\) strategy of peeling off finite loops therefore cannot detect the cost difference. This directly addresses the misconception that the two-marginal criterion “\(c\)-cyclical monotonicity implies optimality” extends naively to \(N\ge 3\).

## 5. Horizontal currents in the Heisenberg group

A Heisenberg-group analogue replaces Euclidean divergence-free vector measures by horizontal vector charges. A horizontal vector measure on \(\mathbb{H}^n\) is a \(2n\)-tuple of signed Radon measures
\[
\mu=(\mu_1,\dots,\mu_{2n})\equiv \sum_{i=1}^n \mu_i X_i+\mu_{i+n}Y_i,
\]
acting on compactly supported horizontal test fields \(\Phi=(\phi_1,\dots,\phi_{2n})\in C_c^\infty(\mathbb{H}^n,H\mathbb{H}^n)\) by
\[
\langle \mu,\Phi\rangle=\sum_{k=1}^{2n}\int_{\mathbb{H}^n}\phi_k\,d\mu_k.
\]
Its total variation identifies \(\mu\) with an element of \(C_0(\mathbb{H}^n,H\mathbb{H}^n)^*\), and its horizontal divergence is
\[
\mathrm{div}_H\,\mu : f\mapsto -\langle \mu,\nabla_H f\rangle.
\]
The current is divergence-free when \(\mathrm{div}_H\,\mu=0\) in the sense of distributions [2605.12716].

The key dynamical ingredient is a horizontal Liouville theorem: if \(\phi\in C_H^2(\mathbb{H}^n;H\mathbb{H}^n)\) is a \(C^2\)-smooth horizontal vector field of at most linear growth generating a complete flow \(u(t,x)\), and if a Radon measure \(\rho\) satisfies \(\mathrm{div}_H(\phi\,\rho)=0\), then \(u(t,\cdot)_\#\rho=\rho\) for all \(t\). Fixing \(\ell>0\), one considers the space \(K_\ell\) of absolutely continuous horizontal curves \(\gamma:[0,\ell]\to\mathbb{H}^n\) with \(\|\dot\gamma\|\le 1\) a.e.; it is \(\sigma\)-compact in the uniform metric [2605.12716].

The main theorem states that if \(\mu\) is a horizontal vector measure on \(\mathbb{H}^n\) with finite total variation and \(\mathrm{div}_H\,\mu=0\), then for each \(\ell>0\) there exists a positive Borel measure \(\nu\) on \(K_\ell\) such that
\[
\mu=\int_{K_\ell} [\gamma]\,d\nu(\gamma),
\qquad
\|\mu\|_M=\int_{K_\ell}\|[\gamma]\|_M\,d\nu(\gamma),
\]
and \(\nu\)-almost every \(\gamma\) has constant speed \(\|\dot\gamma\|\equiv 1\), hence length \(\ell\), and lies in \(\mathrm{supp}\,\mu\) [2605.12716]. The proof regularizes \(\mu\), applies the horizontal Liouville theorem to the normalized smooth field, pushes the invariant measure forward to curve space, and then passes to the limit by weak-* compactness. For general finite-divergence currents, the argument embeds \(\mathbb{H}^n\) into \(\mathbb{H}^{n+1}\), constructs a divergence-free lift, decomposes there, and projects back [2605.12716].

An illustrative corollary is a horizontal analogue of the Havin–Smirnov theorem: if \(M\subset \mathbb{H}^1\) is compact and contains no nonconstant horizontal rectifiable curve, then for any continuous \(f,f_1,f_2\) on \(M\) and \(\varepsilon>0\) there exists \(\psi\in C_c^\infty(\mathbb{H}^1)\) such that
\[
\sup_{x\in M}|f(x)-\psi(x)|+\sum_{j=1}^2 \sup_{x\in M}|f_j(x)-(\nabla_H\psi)_j(x)|<\varepsilon.
\]
This parallels the Euclidean application to free approximation on purely unrectifiable sets [2605.12716].

## 6. Smirnov factorization in complete Pick and free Fock settings

In Hilbert function spaces, “Smirnov-type decomposition” takes a different but closely related form. If \(\mathcal{H}\) is a reproducing-kernel Hilbert space on \(\Omega\), with multiplier algebra
\[
\mathrm{Mult}(\mathcal{H})=\{\varphi:\Omega\to\mathbb{C}: M_\varphi \text{ bounded on }\mathcal{H}\},
\]
then \(\varphi\in \mathrm{Mult}(\mathcal{H})\) is \(\mathcal{H}\)-outer, or cyclic, when \(\overline{M_\varphi \mathcal{H}}=\mathcal{H}\). The Smirnov class is
\[
\mathcal{N}^+(\mathcal{H})=
\left\{\frac{b}{a}: a,b\in \mathrm{Mult}(\mathcal{H}),\ a \text{ is }\mathcal{H}\text{-outer}\right\},
\]
equivalently the set of ratios arising from a contractive column multiplier \(\Theta=\binom{a}{b}\) with \(a\) cyclic [1806.05270].

For a normalized complete Nevanlinna–Pick space \(\mathcal{H}(k)\), every \(h\in \mathcal{H}\) admits such a factorization:
\[
h=\frac{b}{a},
\]
with \(a,b\in \mathrm{Mult}(\mathcal{H})\), \(a\) \(\mathcal{H}\)-outer, the column \(\binom{a}{b}\) contractive, and
\[
\frac1a\in \mathcal{H},
\qquad
\left\|\frac1a\right\|_{\mathcal{H}}^2\le 1+\|h\|_{\mathcal{H}}^2.
\]
The proof uses the Agler–McCarthy model, a norm-preserving free lift to the full Fock space \(\mathfrak{F}_d^2\), and restriction back to commuting arguments [1806.05270].

The free version is formulated in terms of the left Smirnov class on the free unit ball. A free holomorphic function \(H\) belongs to \(\mathcal{N}^+_{\mathrm{left}}\) when
\[
H(X)=B(X)A(X)^{-1}
\]
for bounded left multipliers \(A,B\in \mathfrak{F}_d^\infty\) with \(A\) left-outer. The associated multiplication operator
\[
T_H:\mathrm{Dom}(T_H)=\{F\in \mathfrak{F}_d^2: HF\in \mathfrak{F}_d^2\}\to \mathfrak{F}_d^2
\]
is closed and densely defined precisely in this case, and then there is a unique inner-outer pair \(A,B\), up to a unimodular constant, with
\[
\begin{pmatrix}A\\ B\end{pmatrix}
\]
an isometric left multiplier. Moreover every \(H\in \mathfrak{F}_d^2\) satisfies these hypotheses, and
\[
A^{-1}\in \mathfrak{F}_d^2,
\qquad
\|A^{-1}\|_{\mathfrak{F}_d^2}^2=1+\|H\|_{\mathfrak{F}_d^2}^2.
\]
Here the decomposition is not into curves or cycles but into a quotient by an outer multiplier with exact norm control [1806.05270].

Source: https://www.emergentmind.com/topics/smirnov-type-decomposition