---
title: Birkhoff Center Depth in Polytopes & Dynamics
url: https://www.emergentmind.com/topics/birkhoff-center-depth
type: topic
---

# Birkhoff Center Depth in Polytopes & Dynamics

Birkhoff center depth is used in recent literature for several distinct but structurally related notions centered on extremal or recurrent cores. In the geometry of the Birkhoff polytope, it denotes the Chebyshev radius of the polytope—the minimal bounding-ball radius, also called “center depth” in that setting—under a specified matrix norm, with the barycenter \(J_n=\frac1n\mathbf1\mathbf1^\top\) as the distinguished center [2310.14041]. In dynamical systems, the Birkhoff center is the closure of recurrent points, \(B(\Phi)=\overline{R(\Phi)}=\{x:x\in\omega(x)\}\), and recent work studies both its geometric stratification in competitive flows and a transfinite ordinal “depth” obtained by iterating Birkhoff-type derivative operators [2311.07038]. Across these settings, the common theme is a reduction to a canonical core fixed by symmetry or recurrence.

## 1. Basic objects and competing meanings of “depth”

In the polyhedral setting, the underlying object is the Birkhoff polytope
\[
\mathcal{B}_n=\Omega_n:=\{A\in\mathbb{R}^{n\times n}:A\mathbf{1}=\mathbf{1},\;A^\top\mathbf{1}=\mathbf{1},\;A\ge 0\},
\]
the compact convex set of all \(n\times n\) doubly stochastic matrices. Its extreme points are the permutation matrices, its dimension is \((n-1)^2\), and every \(D\in\Omega_n\) admits a Birkhoff decomposition \(D=\sum_{i=1}^r\alpha_iP_i\) with permutation matrices \(P_i\), \(\alpha_i\ge0\), and \(\sum_i\alpha_i=1\). The matrix
\[
J_n:=\frac1nJ=\frac1n\mathbf1\mathbf1^\top
\]
is the barycenter of \(\Omega_n\), with every entry equal to \(1/n\), and is the uniform convex combination of all permutation matrices [2310.14041].

For operator \(\ell_n^p\)-norms, the induced metric is
\[
d_p(A,B)=\|A-B\|_{p\to p},\qquad
\|A\|_{p\to p}=\sup_{x\neq0}\frac{\|Ax\|_p}{\|x\|_p}.
\]
The Chebyshev center \(C_p\) and Chebyshev radius \(r_p\) of \(\Omega_n\) are
\[
C_p\in\operatorname*{arg\,min}_{C\in\mathbb{R}^{n\times n}}\sup_{X\in\Omega_n}\|X-C\|_{p\to p},\qquad
r_p=\inf_C\sup_{X\in\Omega_n}\|X-C\|_{p\to p}.
\]
In that context, “center depth” refers precisely to this Chebyshev radius, i.e. the smallest \(r\) such that a ball of radius \(r\) centered at the Chebyshev center contains \(\Omega_n\) [2310.14041].

In dynamical systems, the terminology changes. For a flow \(\Phi:\mathbb{R}\times\mathbb{R}^n\to\mathbb{R}^n\), the Birkhoff center is
\[
B(\Phi):=\overline{R(\Phi)}=\{x\in\mathbb{R}^n:x\in\omega(x)\}.
\]
For a homeomorphism \(f:X\to X\) on a compact metric space, the Birkhoff center is
\[
\xi(X):=\operatorname{cl}(\mathrm{Recur}_f)=\operatorname{cl}\{x\in X:x\in\omega_f(x)\}.
\]
The paper “Realizing Arbitrary Depth” defines a transfinite depth by iterating a Birkhoff derivative \(z\) on closed subsets via
\[
X_{\alpha+1}:=z(X_\alpha),\qquad
X_\lambda:=\bigcap_{\beta<\lambda}X_\beta,
\]
and setting
\[
\theta_z(X):=\min\{\alpha:X_\alpha=X_{\alpha+1}\}.
\]
This depth is ordinal-valued and measures the stage at which the derivation stabilizes at the Birkhoff center [2606.01499].

## 2. Center depth of the Birkhoff polytope under operator \(\ell^p_n\)-norms

For the operator norms induced by the vector \(\ell^p\)-norms, the central result is a complete characterization: for every \(1\le p\le\infty\), the Chebyshev center of \(\Omega_n\) is unique and equals the barycenter,
\[
C_p=J_n,
\]
and the center depth is
\[
r_p=R_p(\Omega_n)=\|I_n-J_n\|_{p\to p}.
\]
The same work establishes the normalization
\[
\|D\|_{p\to p}=1\qquad\text{for all }D\in\Omega_n,\;1\le p\le\infty,
\]
using the Birkhoff decomposition and permutation invariance of the norm [2310.14041].

A broader structure theorem is proved for any permutation-invariant norm \(\|\cdot\|\) and any convex permutation-invariant constraint set \(\mathcal R\subseteq M_n(\mathbb R)\). If a Chebyshev center \(A\) exists, then
\[
J_nAJ_n=\left(\frac1n\sum_{i,j=1}^n a_{ij}\right)J_n
\]
is also a Chebyshev center, and
\[
R_{\|\cdot\|}(\Omega_n)=\|J_nAJ_n-I_n\|
=\inf_{\substack{\alpha\in\mathbb R\\ \alpha J_n\in\mathcal R}}\|\alpha J_n-I_n\|.
\]
In particular, when \(\mathcal R=\Omega_n\), the only admissible center on the \(J_n\)-ray is \(J_n\) itself, so
\[
R_{\|\cdot\|}(\Omega_n)=\|J_n-I_n\|.
\]
A key corollary is equidistance to all permutations:
\[
\|A-P\|=R_{\|\cdot\|}(\Omega_n)\qquad\text{for every permutation matrix }P.
\]
Thus the farthest points from the center are exactly the extreme points of the polytope [2310.14041].

For specific values of \(p\), the radius is explicit:
\[
r_2=1,\qquad
r_1=r_\infty=2\left(1-\frac1n\right).
\]
For \(p=2\), \(I_n-J_n\) is the orthogonal projector onto \(\mathbf1^\perp\), so its spectral norm is \(1\). For \(p=1\) and \(p=\infty\), one has
\[
\|P-J_n\|_{1\to1}=\|P-J_n\|_{\infty\to\infty}
=2\left(1-\frac1n\right)
\]
for every permutation matrix \(P\). Moreover, for \(n=2\),
\[
r_p=1\qquad\text{for all }1\le p\le\infty.
\]
For \(n=3\) and \(1<p<\infty\), the exact formula is
\[
r_p=\|J_3-I_3\|_{p\to p}
=\frac{\big(2^{p-1}+1\big)^{1/p}}{\big(2^{1/(p-1)}+1\big)^{1/p'}},
\qquad p'=\frac{p}{p-1}.
\]

The paper also gives general bounds for \(1<p<\infty\):
\[
1\le r_p=\|I_n-J_n\|_{p\to p}\le
\left(\frac{2(n-1)}{n}\right)^{\left|\frac2p-1\right|}.
\]
These bounds are tight at \(p=2\). A conjectured exact formula is stated for general \(n\) and \(p\neq2\), based on an optimization over \(m\in\{m_1,m_2\}\) determined by a root \(x_p\in[0,1]\) of an explicit scalar equation [2310.14041].

## 3. Center depth under Schatten \(p\)-norms

For \(1\le p<\infty\), the Schatten norm is
\[
\|A\|_{S_p}=\left(\sum_{i=1}^n s_i(A)^p\right)^{1/p}
=\big(\operatorname{tr}((A^\ast A)^{p/2})\big)^{1/p},
\]
where \(s_i(A)\) are the singular values. These norms are unitarily invariant, hence permutation invariant, monotone in \(p\), and strictly convex for \(1<p<\infty\) [2310.14043].

The Chebyshev center and radius of the Birkhoff polytope under \(\|\cdot\|_{S_p}\) are completely explicit:
\[
C_p^\ast=J_n,\qquad
R_p(B_n)=\|J_n-I_n\|_{S_p}=(n-1)^{1/p},
\]
and the center is unique for every \(1\le p<\infty\). The calculation reduces to the \(J_n\)-ray. Since the singular values of \(\alpha J_n-I_n\) are \(|1-\alpha|\) with multiplicity \(1\) and \(1\) with multiplicity \(n-1\),
\[
\|\alpha J_n-I_n\|_{S_p}
=\big((n-1)\cdot1^p+|1-\alpha|^p\big)^{1/p},
\]
whose minimum is attained at \(\alpha=1\). As in the operator-norm setting, every permutation matrix lies at the same distance from the center:
\[
\|P-J_n\|_{S_p}=(n-1)^{1/p}\qquad\text{for all permutation matrices }P.
\]
The same result holds whether one minimizes over all \(A\in M_n(\mathbb R)\) or restricts the center to \(A\in B_n\) [2310.14043].

The Frobenius case \(p=2\) yields a second, more detailed formula. For any \(A\in M_n(\mathbb R)\),
\[
r_{S_2}(A)=\sup_{D\in B_n}\|A-D\|_F
=\big(\|A\|_F^2+n-2\,\operatorname{tr}_{\min}(A)\big)^{1/2},
\]
where
\[
\operatorname{tr}_{\min}(A):=\min_{P\in\mathcal P}\operatorname{tr}(AP)
\]
is the minimal trace, identified with the assignment problem value for cost matrix \(A\). This exhibits an intrinsic connection between the minimal enclosing Frobenius ball and the assignment problem; the quantity \(\operatorname{tr}_{\min}(A)\) can be computed in \(O(n^3)\) via the Hungarian algorithm. For \(A=J_n\), one has \(\operatorname{tr}_{\min}(J_n)=1\) and \(\|J_n\|_F^2=1\), so \(r_{S_2}(J_n)=(n-1)^{1/2}\) [2310.14043].

The norm profile on \(B_n\) is also explicit:
\[
1\le \|D\|_{S_p}\le n^{1/p},
\]
with equality on the left iff \(D=J_n\), and on the right iff \(D\) is a permutation matrix. Thus \(J_n\) is the unique closest point of \(B_n\) to the origin for all Schatten \(p\), while permutations are the farthest [2310.14043].

A useful contrast emerges with operator norms. For the operator \(\ell^2\to\ell^2\) norm, the radius is \(1\); for the Frobenius norm, the radius is \(\sqrt{n-1}\). Both are centered at \(J_n\), but the scale of the center depth depends strongly on the chosen norm [2310.14041].

## 4. The Birkhoff center in competitive dynamical systems

For strongly competitive flows on \(\mathbb R^n\), the Birkhoff center is studied as a recurrent geometric object rather than a metric minimizer. The setting assumes a solid cone \(C_+\subset\mathbb R^n\) inducing a partial order \(x\le y\iff y-x\in C_+\) and a strong order \(x\ll y\iff y-x\in\operatorname{Int}C_+\). A set is unordered if no two distinct points are related by \(\le\), and strongly ordered if any two distinct points are related by \(\ll\). Under the standing hypotheses of strong competitiveness and dissipation—namely, the existence of a compact global attractor \(I\) uniformly attracting compact sets—the paper proves a sharp order-structure dichotomy for connected components of the Birkhoff center [2311.07038].

The principal statement is:
\[
\textbf{Theorem 4.1.}\ \text{Let }B\text{ be a connected component of }B(\Phi).
\]
Then exactly one of the following holds:

1. \(B\) is unordered; or  
2. \(B\) consists of strongly ordered equilibria.

This dichotomy is complemented by a canonical countable pairwise disjoint family \(\mathcal F=\{M_i\}\) of invariant open \((n-1)\)-cells such that every unordered connected component of \(B(\Phi)\) lies on one of these cells. More precisely, if \(B\) is unordered, then there exist \(p\in E\cup\{-\infty\}\) and \(q\in E\cup\{+\infty\}\) such that
\[
B\subset M_+(p)\cap M_-(q),
\]
where \(M_+(p)\) and \(M_-(q)\) are invariant unordered open \((n-1)\)-cells constructed from one-sided boundaries of repulsion basins. For \(q\in E\cup\{+\infty\}\), the relevant lower cell is built from
\[
R_-(q)=\{x\in R(q):\Phi_t(x)\ll q\text{ for all sufficiently large }t>0\},\qquad
M_-(q)=\partial_-R_-(q),
\]
with analogous upper constructions \(R_+(p)\), \(M_+(p)\) [2311.07038].

The central geometric mechanism is the joint cone-boundary intersection principle:
\[
\textbf{Proposition 3.5.}\ \text{Let }x\in R(\Phi).\ \text{Then }B(\Phi)\cap\partial C_x=\{x\}.
\]
Combined with a connecting lemma and an absorbing principle, this prevents connected recurrent sets from crossing the cone boundary except at the base recurrent point. The consequence is that recurrence in strongly competitive systems is forced into either dynamically trivial ordered equilibrium components or codimension-one unordered sheets.

The same paper proves an analogous statement for supports of invariant probability measures. If \(\mu\) is invariant and \(B\) is a connected component of \(\operatorname{supp}(\mu)\), then either \(B\) consists of strongly ordered equilibria or \(B\) lies on one element of \(\mathcal F\). In dimension \(3\), this yields
\[
h_{\mathrm{top}}(\Phi)=0,
\]
because supports of ergodic measures are either equilibria or compact invariant sets lying on invariant \(2\)-cells, and continuous flows on compact \(2\)-manifolds have zero measure-theoretic entropy. The paper explicitly notes that it does not introduce a numeric depth invariant, but its decomposition theorems suggest an “effective depth \(\le1\)” phenomenon: recurrent and statistical behavior is confined either to equilibrium sets or to invariant codimension-one cells [2311.07038].

## 5. Transfinite Birkhoff center depth and ordinal realization

A different notion of Birkhoff center depth is developed for compact metric dynamical systems \((X,f)\) with \(f\) a homeomorphism. The paper introduces several one-step derivative operators on closed subsets:
\[
z_{BN}(X)=\{x\in X:x\text{ is nonwandering in }X\},
\]
\[
z_{BL}(X)=\operatorname{cl}\Big(\bigcup_{x\in X}\omega_f(x)\Big),
\]
\[
z_{BL\pm}(X)=\operatorname{cl}\Big(\bigcup_{x\in X}\big(\alpha_f(x)\cup\omega_f(x)\big)\Big),
\]
and the Cantor–Bendixson derivative
\[
z_{CB}(X)=\{x\in X:x\text{ is not isolated in }X\}.
\]
Assuming \(X\) has no periodic isolated points, the general inclusions are
\[
\xi(X)\subseteq z_{BL}(X)\subseteq z_{BL\pm}(X)\subseteq z_{BN}(X)\subseteq z_{CB}(X).
\]
The transfinite derivation is defined by
\[
X_{\alpha+1}=z(X_\alpha),\qquad
X_\lambda=\bigcap_{\beta<\lambda}X_\beta
\]
for limit ordinals \(\lambda\), and the depth is
\[
\theta_z(X)=\min\{\alpha:X_\alpha=X_{\alpha+1}\}.
\]
For compact metric spaces, all such depths are countable ordinals [2606.01499].

The main theorem states that every countable ordinal is realizable as Birkhoff center depth. Specifically, for every countable ordinal \(\theta\) there exists a dynamical system \((X,f)\) such that, for every ordinal \(0\le\alpha<\theta\), the derived subsystem \((X_\alpha,f|_{X_\alpha})\) is a nontrivial simple system and
\[
X_\theta=\{e\}.
\]
Consequently,
\[
\theta_{BL}(X)=\theta_{BL\pm}(X)=\theta_{BN}(X)=\theta,
\qquad
\theta_{CB}(X)=\theta+1.
\]
Here a simple system means: \(X\) is countable; there is a fixed point \(e\in X\) which is the unique recurrent point; and every non-isolated point is an \(\omega\)-limit point of some isolated point. In any nontrivial simple system,
\[
z_{BL}(X)=z_{BL\pm}(X)=z_{BN}(X)=z_{CB}(X),
\]
and in totally simple systems this equality persists at every derived stage [2606.01499].

The realization theorem is obtained from three constructions with precise ordinal arithmetic:

- **Attachment** \((aX,af)\): if \((X,f)\) is totally simple, then
  \[
  \theta(aX)=1+\theta(X),\qquad z(aX)=X\times\{\ast\}.
  \]

- **Stretched suspension** \((sX,sf)\): if \((X,f)\) is totally simple, then
  \[
  \theta(sX)=\theta(X)+1,\qquad z(sX)=s(z(X)).
  \]

- **Pointed union** \(\bigvee_i X_i\): if each \((X_i,f_i)\) is totally simple, then
  \[
  \theta\Big(\bigvee_i X_i\Big)=\sup_i\theta(X_i),\qquad
  z\Big(\bigvee_i X_i\Big)=\bigvee_i z(X_i).
  \]

The master construction is
\[
X^0=\{e\},\qquad
X^{\alpha+1}=sX^\alpha,\qquad
X^\alpha=\bigvee_{\beta<\alpha}X^\beta\quad(\alpha\text{ limit}),
\]
which yields for each countable \(\theta\),
\[
\theta_{BL}(X^\theta)=\theta_{BL\pm}(X^\theta)=\theta_{BN}(X^\theta)=\theta,\qquad
\theta_{CB}(X^\theta)=\theta+1.
\]
The paper emphasizes that noncommutativity of ordinal addition is essential: if one uses attachment at successor stages instead, then \(1+\alpha=\alpha\) for infinite \(\alpha\), so the construction fails to realize the intended higher depths [2606.01499].

## 6. Comparative perspective and recurrent misconceptions

The literature shows that “Birkhoff center depth” is not a single invariant shared across all contexts. In the geometry of the Birkhoff polytope, it is a metric circumradius: the minimal radius of a ball centered at the Chebyshev center that contains \(\Omega_n\). In competitive dynamics, the relevant result is not a numeric depth but a stratification of \(B(\Phi)\) into ordered equilibrium components and unordered invariant \((n-1)\)-cells. In general topological dynamics, depth is an ordinal stabilization index for transfinite Birkhoff derivatives [2310.14041].

A common misunderstanding is to conflate enclosing-ball geometry with inscribed-ball geometry. The polytope papers study smallest enclosing balls—Chebyshev centers and radii—not incenters or inradii. The Schatten-norm paper explicitly states that no incenter/inradius results are developed there [2310.14043]. Another possible confusion is between the dynamical Birkhoff center and the Birkhoff polytope: they share a historical name but belong to different domains, respectively recurrence theory and matrix convexity.

The available results nevertheless exhibit a common structural pattern. In the polytope setting, permutation symmetry forces the center onto the \(J_n\)-ray and ultimately to \(J_n\). In competitive flows, order-theoretic constraints and the joint cone-boundary principle force recurrent components into a dichotomy between equilibrium order and codimension-one unordered cells. In the ordinal setting, repeated removal of wandering or isolated layers stabilizes at the recurrent core, and arbitrary countable ordinal complexity can occur [2311.07038].

This suggests a broad unifying interpretation: Birkhoff center depth measures, in a norm-dependent, order-theoretic, or transfinite sense, how far a system lies from its canonical recurrent or symmetric core. In the Birkhoff polytope the core is the barycenter \(J_n\); in competitive flows it is the stratified Birkhoff center \(B(\Phi)\); and in topological dynamics it is the fixed stage of the transfinite derivation \(\xi(X)\) [2606.01499].

Source: https://www.emergentmind.com/topics/birkhoff-center-depth