---
title: Edrei Theorem in Total Positivity
url: https://www.emergentmind.com/topics/edrei-theorem
type: topic
---

# Edrei Theorem in Total Positivity

Searching arXiv for the cited work and closely related papers on the Edrei theorem and its modern extensions.
arxiv_search(query="Edrei theorem Toeplitz total nonnegative Schoenberg parameters", max_results=10)
arxiv_search(query="1506.07379 generalized Hurwitz matrices forbidden sectors Edrei theorem", max_results=5)
The Edrei theorem is a classical classification result in total positivity. In its standard one-sided Toeplitz form, it characterizes normalized infinite upper-triangular Toeplitz matrices \(u(\mathbf c)=(c_{i-j})_{i,j\in\mathbb N}\), with \(c_0=1\) and \(c_{-n}=0\) for \(n>0\), whose minors are all nonnegative, by a generating-function factorization
\[
1+c_1x+c_2x^2+c_3x^3+\dotsc
=
e^{\gamma x}\prod_{i=1}^\infty\frac{1+\beta_i x}{1-\alpha_i x},
\]
where \(\gamma\ge 0\) and \(\boldsymbol\alpha,\boldsymbol\beta\) are weakly decreasing summable sequences of nonnegative reals. In this form the theorem is also called the Edrei–Thoma theorem, and the parameters \(\alpha_i,\beta_i\) are called the Schoenberg parameters [2509.25163]. In adjacent polynomial literature, however, the name “Edrei theorem” is also used for the Aissen–Edrei–Schoenberg–Whitney theorem, which characterizes real polynomials with only negative real zeros by total nonnegativity of a Toeplitz matrix of coefficients [1506.07379].

## 1. Classical Toeplitz formulation

A one-sided sequence \(\mathbf c=(c_1,c_2,c_3,\dotsc)\) defines the infinite upper-triangular Toeplitz matrix
\[
u(\mathbf c)=(c_{i-j})_{i,j\in\mathbb N},
\qquad c_0=1,\quad c_{-n}=0\ \text{for }n>0.
\]
The sequence is called totally nonnegative when all minors of \(u(\mathbf c)\) lie in \(\mathbb R_{\ge 0}\); the same objects are also called Pólya frequency sequences. Edrei’s theorem states that this condition is equivalent to the factorization
\[
1+c_1x+c_2x^2+c_3x^3+\dotsc
=
e^{\gamma x}\prod_{i=1}^\infty\frac{1+\beta_i x}{1-\alpha_i x},
\]
for parameters \((\gamma,\boldsymbol\alpha,\boldsymbol\beta)\) in
\[
\Omega_S
=
\{(\gamma,\boldsymbol\alpha,\boldsymbol\beta)\in \mathbb R_{\ge 0}\times \mathbb R_{\ge 0}^{\mathbb N}\times \mathbb R_{\ge 0}^{\mathbb N}
\mid
\boldsymbol\alpha,\boldsymbol\beta\ \text{weakly decreasing with finite sum}\}.
\]
The nonzero \(\alpha_i\) and \(\beta_i\) encode poles and zeros of the generating function, while the residual entire factor is exactly \(e^{\gamma x}\) [2509.25163].

A point emphasized in modern expositions is that the difficult part of the theorem is not the existence of root and pole data, but the identification of the remaining entire factor. Schoenberg had shown that the displayed factorization produces a totally nonnegative sequence, and Aissen–Schoenberg–Whitney had identified the permissible roots and poles; Edrei’s contribution was to prove that the leftover factor must be the simple exponential \(e^{\gamma x}\) [2509.25163]. The theorem therefore parametrizes all normalized infinite totally nonnegative upper-triangular Toeplitz matrices by the Schoenberg parameters.

The theorem is fundamentally about total nonnegativity rather than strict total positivity. Later work distinguishes the totally positive locus, where all nontrivial minors are strictly positive, as an open dense subset of the totally nonnegative part. In the infinite case, refined subsets are obtained by requiring all Schoenberg parameters to be nonzero, or moreover pairwise distinct, but the classical theorem itself is the biconditional classification of the totally nonnegative case [2509.25163].

## 2. The polynomial theorem and the \(M=1\) endpoint

In polynomial theory, the same Toeplitz positivity phenomenon appears in the Aissen–Edrei–Schoenberg–Whitney theorem. For
\[
f(x)=a_0x^n+a_1x^{n-1}+\cdots+a_n,\qquad a_0>0,
\]
the Toeplitz matrix
\[
H_1(f)=(a_{j-i})_{i,j\ge 1}
\]
is totally nonnegative if and only if \(f\) has only real negative zeros. Here “totally nonnegative” again means that all minors are nonnegative [1506.07379].

This polynomial form is the \(M=1\) case of a broader framework based on generalized Hurwitz matrices
\[
H_M(f):=(a_{Mj-i})_{i,j},
\]
where \(M\le n\). The generalized sector theorem states that if \(H_M(f)\) is totally nonnegative, then
\[
f(z)\neq 0
\qquad\text{for}\qquad
|\arg z|<\frac{\pi}{M}.
\]
For \(M=2\), this recovers the total-nonnegativity form of the Hurwitz stability theorem; for \(M=n\), it recovers the Cowling–Thron zero-free sector for polynomials with positive coefficients; and for \(M=1\), it yields the Edrei-type conclusion that zeros lie on the nonpositive real axis [1506.07379].

The \(M=1\) case is stronger than the general sector theorem because it is an exact characterization. For \(M>2\), total nonnegativity of \(H_M(f)\) is only sufficient for zero exclusion from the sector \(|\arg z|<\pi/M\), and the converse fails in general. This distinction is central in the generalized Hurwitz-matrix literature: the Edrei case and the Hurwitz case retain “if and only if” formulations, whereas the higher-\(M\) theory is genuinely one-sided [1506.07379].

## 3. Historical development and representation theory

The historical line emphasized in recent work is Schoenberg \(\to\) Edrei \(\to\) Thoma \(\to\) Vershik–Kerov. Schoenberg conjectured the factorization, Edrei completed the proof in 1952, and Thoma later rediscovered the same parametrization in the language of extremal characters of the infinite symmetric group \(S_\infty\). In Thoma’s normalization \(c_1=1\), one obtains the Thoma simplex and the relation
\[
\gamma=1-\sum_{i=1}^\infty(\alpha_i+\beta_i),
\]
so the Schoenberg parameters become exactly the Thoma parameters [2509.25163].

Vershik–Kerov then gave these parameters an asymptotic representation-theoretic meaning in terms of normalized modified Frobenius coordinates of Young diagrams. For a sequence of partitions \(\lambda^{(n)}\vdash n\) with modified Frobenius coordinates \((a^{(n)}|b^{(n)})\), the limits
\[
\alpha_i=\lim_{n\to\infty} a_i^{(n)}/n,
\qquad
\beta_j=\lim_{n\to\infty} b_j^{(n)}/n
\]
exist exactly when normalized characters converge, and these limits are the Thoma parameters [2509.25163].

A two-sided analogue appears in the Edrei–Voiculescu theorem. In that setting, doubly infinite totally positive Toeplitz sequences are classified by parameters
\[
\omega=(\alpha^+,\beta^+;\alpha^-,\beta^-;\gamma^+,\gamma^-),
\]
through the Laurent generating function
\[
\Phi(u;\omega):=
e^{\gamma^+(u-1)+\gamma^-(u^{-1}-1)}
\prod_{i=1}^\infty
\frac{1+\beta_i^+(u-1)}{1-\alpha_i^+(u-1)}
\frac{1+\beta_i^-(u^{-1}-1)}{1-\alpha_i^-(u^{-1}-1)}.
\]
Borodin and Olshanski identify this theorem with the boundary of the Gelfand–Tsetlin graph, the extreme finite characters of \(U(\infty)\), and the space of doubly infinite totally positive sequences [1109.1412]. Petrov’s later work simplifies the determinantal formula underlying that proof and gives a \(q\)-analogue, but the classification itself remains the Edrei–Voiculescu boundary theorem [1208.3443].

## 4. Asymptotic geometry, finite Toeplitz matrices, and quantum cohomology

A major modern theme is that Edrei’s theorem is not only a classification theorem for infinite Toeplitz matrices, but also the asymptotic target of finite-dimensional constructions. For finite totally positive Toeplitz matrices \(u^{(n+1)}\in UToep_{n+1}(\mathbb R_{>0})\), one has torus parameters \(d_i^{(n+1)}\) and quantum parameters \(q_i^{(n+1)}\), and if \(u^{(n+1)}\) converges uniformly to an infinite totally positive Toeplitz matrix with Schoenberg parameters \((\alpha_i),(\beta_j)\), then
\[
\lim_{n\to\infty}\sqrt[n]{d_i^{(n+1)}}=\alpha_i,
\qquad
\lim_{n\to\infty}\sqrt[n]{(d_{n+2-j}^{(n+1)})^{-1}}=\beta_j,
\]
and
\[
\lim_{n\to\infty}\sqrt[n]{q_i^{(n+1)}}=\frac{\alpha_{i+1}}{\alpha_i},
\qquad
\lim_{n\to\infty}\sqrt[n]{q_{n+1-j}^{(n+1)}}=\frac{\beta_{j+1}}{\beta_j}.
\]
Thus finite quantum-cohomological data recover the Schoenberg parameters asymptotically [2509.25163].

The same paper interprets \(1/\alpha_i\) and \(1/\beta_j\) as asymptotic Chern roots. In particular, certain Chern-class limits are
\[
\lim_{n\to\infty}\mathcal F_k
=
\frac{1}{\alpha_1}+\cdots+\frac{1}{\alpha_k},
\qquad
\lim_{n\to\infty}\mathcal F'_k
=
\frac{1}{\beta_1}+\cdots+\frac{1}{\beta_k},
\]
and asymptotics of Schubert classes are controlled by the same reciprocal parameters. This recasts the roots and poles of Edrei’s generating function as asymptotic geometric data on flag varieties [2509.25163].

The Grassmannian version isolates special one-parameter strata inside Edrei’s parameter space. For convergent sequences \(u^{(n)}\in X_P^{(n)}(\mathbb R_{>0})\), the possible infinite limits in the proved cases are
\[
(1+\beta x)^m
\quad\text{from }Gr(n-m,n),
\]
\[
(1-\alpha x)^{-k}
\quad\text{from }Gr(k,n),
\]
and
\[
e^{\eta x}
\quad\text{from }Gr(n,2n).
\]
The middle-dimensional Grassmannian case is especially significant because it isolates precisely the exponential factor in Edrei’s theorem [2606.16983].

## 5. Tropical analogues and the Lusztig connection

Recent work has produced a tropical counterpart of the infinite Toeplitz parametrization. In the tropical setting, the role of an infinite totally positive Toeplitz matrix is played by an infinite min-ideal filling \((M_{ij})\) satisfying
\[
M_{ij}=\min(M_{i+1,j},M_{i,j+1})\qquad\forall i,j\in\mathbb N.
\]
The tropical parameter space consists of two weakly increasing sequences
\[
\mathbf A=(A_i)_i,\qquad \mathbf B=(B_j)_j,
\]
and the tropical Edrei map is
\[
(\mathbf A,\mathbf B)\longmapsto (\min(A_i,B_j))_{i,j\in\mathbb N}.
\]
The corresponding theorem states that this map is a bijection onto the space of infinite min-ideal fillings, with refined versions for asymptotically real and stable fillings [2509.06944].

A detropicalized version is obtained by working over a valued semifield and forming a Toeplitz matrix with generating function
\[
1+\sum_{k=1}^\infty \mathbf c_k x^k
=
\frac{\prod_{j=1}^{\infty}(1+\beta_j x)}{\prod_{i=1}^{\infty}(1-\alpha_i x)}.
\]
If \((\mathbf m_{ij})\) are the standard coordinates of the resulting infinite Toeplitz matrix, then
\[
Val(\mathbf m_{ij})=\min(A_i,B_j),
\]
where \(A_i=Val(\alpha_i)\) and \(B_j=Val(\beta_j)\). In this sense, the tropical formula \(M_{ij}=\min(A_i,B_j)\) is the valuation shadow of the classical Schoenberg-parameter theory [2509.06944].

The same paper places beside this infinite theorem a finite Toeplitz parametrization arising from quantum cohomology. After tropicalization, the finite Toeplitz parameter map becomes exactly Lusztig’s weight map
\[
\mathcal L:U_+(\mathbb R_{\min})\to \mathfrak h^*_{\mathbb R,PSL_{n+1}}.
\]
An asymptotic theorem then shows that normalized finite Lusztig weights recover the tropical Schoenberg parameters \(A_i,B_j\). A plausible implication is that Edrei’s classical infinite parametrization and Lusztig’s canonical-basis parametrization are linked through a finite-to-infinite tropical limit [2509.06944].

## 6. \(K\)-theoretic, integral, and shifted extensions

The Edrei–Thoma product formula also classifies Schur-positive specializations of the ring \(\Lambda\) of symmetric functions. If \(\rho:\Lambda\to\mathbb R\) is Schur-positive, then
\[
\rho(H(z))
=
e^{\gamma z}\prod_{n=1}^\infty \frac{1+\beta_n z}{1-\alpha_n z},
\]
with \(\alpha_n,\beta_n,\gamma\ge 0\) and \(\sum_n(\alpha_n+\beta_n)<\infty\). Yeliussizov’s work on symmetric Grothendieck polynomials constructs, from a \(G\)-positive specialization \(\varphi\), a Toeplitz matrix
\[
H(\varphi)=[H_{i-j}(\varphi)]_{i,j\ge 0},
\qquad
H_0=1+G_{(1)},\quad H_n=G_{(n)}+G_{(n+1)},
\]
and proves that it is totally nonnegative, thereby reducing the \(K\)-theoretic classification to the classical Edrei–Thoma theorem [1907.06985].

There is also an integral form. In the Witt-vector framework, totally positive integral series \(f(t)\in 1+t\mathbb Z[[t]]\) are exactly those of the form
\[
f(t)=\frac{g(t)}{h(t)},
\]
where \(g(t),h(t)\in 1+t\mathbb Z[t]\), all complex roots of \(g(t)\) are negative real numbers, and all those of \(h(t)\) are positive real numbers. The paper presents this as an integral Edrei–Thoma theorem, equivalent to a classification of Schur Witt vectors over \(\mathbb N\) [1311.5031].

A shifted \(K\)-theoretic analogue has now been proved for Ikeda–Naruse \(K\)-theoretic Schur \(P\)- and \(Q\)-functions. In the unsigned case, the positive specializations are classified by a single summable nonnegative sequence \(a=(a_1\ge a_2\ge\cdots\ge 0)\) and \(\gamma\ge 0\), with generating series
\[
\sum_{n\ge 0}\rho(GQ_n+GQ_{n+1})z^n
=
D^2 e^{2\gamma z}\prod_{n=1}^\infty \frac{1-\overline{a_n} z}{1-a_n z},
\qquad
\overline{a_n}=\frac{-a_n}{1+a_n}.
\]
This extends Nazarov’s shifted \(\beta=0\) theorem and shows that the Edrei–Thoma paradigm persists in the shifted \(K\)-theoretic setting, with the ordinary two-sequence parameter set collapsing to one sequence because of shifted symmetry [2512.23944].

The theorem’s contemporary significance lies in this persistence across settings. In one-sided Toeplitz total positivity it is a classification of generating functions; in polynomial theory it becomes the \(M=1\) endpoint of generalized Hurwitz-sector theorems; in representation theory it reappears as Thoma parameters and, in two-sided form, as the Edrei–Voiculescu description of the Gelfand–Tsetlin boundary; in asymptotic geometry it governs quantum parameters, Chern roots, and Schubert limits; and in tropical and \(K\)-theoretic settings it survives as a structural template for new parametrization theorems [2509.25163].

Source: https://www.emergentmind.com/topics/edrei-theorem