betaFeedback

WorksMathematics

Degree and arity of tracial separation for self-adjoint contractions

Abstract

Motivated by the recent explicit counterexample to Connes' embedding conjecture, we study the degree and number of variables needed to separate normalized matrix traces on self-adjoint contractions from traces on finite von Neumann algebras. We prove that the two trace infima coincide for every Hermitian polynomial of degree at most two in at most three variables, with a dimension-explicit approximation bound. We give three explicit separators: an integer-coefficient polynomial of degree six in ten variables with 392 nonzero words, a quadratic polynomial in 214 variables, and an integer-coefficient polynomial in two variables of ordinary and cyclic degree 24. Each has normalized trace at least 3/43/4 on every matrix contraction tuple and trace −1-1 at a finite tracial witness. We provide the source systems, transfer estimates, and exact coefficient recipes for all three constructions.

Introduction

The trace separation problem

Let C⟨X1,…,Xn⟩\mathbb{C}\langle X_1,\ldots,X_n\rangle be the free ∗*-algebra on self-adjoint indeterminates. For a Hermitian polynomial f=f∗f=f^*, we compare normalized traces on self-adjoint matrix tuples A=(A1,…,An)A=(A_1,\ldots,A_n) of common size with traces on finite von Neumann algebras. Requiring contractions bounds the domains independently of matrix size. Set

mmat(f)=inf⁡N≥1Ai=Ai∗∈MN(C)∥Ai∥≤1tr⁡Nf(A),mtr(f)=inf⁡(M,τ) finite tracialai=ai∗, ∥ai∥≤1τf(a).(1)m_{\mathrm{mat}}(f)=\inf_{\substack{N\ge1\\ A_i=A_i^*\in M_N(\mathbb{C})\\ \lVert A_i\rVert\le1}}\operatorname{tr}_N f(A), \qquad m_{\mathrm{tr}}(f)=\inf_{\substack{(M,\tau)\ \text{finite tracial}\\ a_i=a_i^*,\ \lVert a_i\rVert\le1}}\tau f(a). \tag*{(1)}

Here tr⁡N=N−1Tr⁡\operatorname{tr}_N=N^{-1}\operatorname{Tr}, and finite tracial algebras have a normalized faithful normal trace. By including matrices among finite tracial algebras, one always has mtr(f)≤mmat(f)m_{\mathrm{tr}}(f)\le m_{\mathrm{mat}}(f). We call ff a tracial separator when the inequality is strict. The constructions below have the normalization

tr⁡Nf(A)≥34for all matrix contraction tuples,τf(a)=−1for one tracial contraction tuple.(2)\operatorname{tr}_N f(A)\ge\frac{3}{4}\quad\text{for all matrix contraction tuples},\qquad \tau f(a)=-1\quad\text{for one tracial contraction tuple}. \tag*{(2)}

Positivity at each matrix size need not give a uniform positive lower bound.

The operator-algebraic background is Connes’ embedding problem for finite factors [1]. It is now known to have a negative answer: the MIP⁡∗=RE⁡\operatorname{MIP}^*=\operatorname{RE} theorem implies a counterexample to the conjecture [2]. On the polynomial side, unrestricted matrix positivity is characterized by sums of squares [3], whereas trace positivity leads to a distinct tracial moment problem [4]. Klep and Schweighofer connected this trace-positive setting to Hermitian squares, contraction constraints and commutators [5]; the real-algebraic equivalence requires the correction in [6].

Wang and Zhi gave explicit counterexamples to Connes’ embedding conjecture [10]. Their main construction is an integer-coefficient Hermitian polynomial in six variables of degree 7272 with 3333 monomials. It has the form f=−1+ω+ω∗+M∑j=115(2−uj−uj∗)f=-1+\omega+\omega^*+M\sum_{j=1}^{15}(2-u_j-u_j^*), where ω,u1,…,u15\omega,u_1,\ldots,u_{15} are specified words and MM is an explicit positive integer. The relation penalties are nonnegative in normalized matrix trace on

2020 Mathematics Subject Classification. 46L10, 13J30, 16S34, 15A18, 90C22.

Key words and phrases. Connes’ embedding problem, trace-positive noncommutative polynomial, tracial moment problem, sums of Hermitian squares, quadratic module, matrix contractions, Horn inequalities. contraction tuples and vanish at the specified group witness. Consequently, ff has normalized trace at least 3/43/4 on every matrix contraction tuple, while the witness has trace −1-1 [10 Theorem 1.1]. This gap also rules out a contraction-module certificate for f+εf + \varepsilon modulo commutators whenever 0≤ε<10 \le\varepsilon< 1 [10 Corollary 5.12]. The same paper constructs a 37-variable quartic with Gaussian-integer coefficients and 387 monomials, with the same trace gap but without norm restrictions on the matrix inputs, and an integer-coefficient two-variable separator of degree at most 48 on contractions [10 Appendices A–B]. Thus degree four is minimal in the unrestricted setting, while two variables suffice and are necessary in the contraction setting. Their group obstruction and quantitative normalization use results of Kun–Thom, Thom, and Alekseev–Liu–Thom [11, 12, 13].

The constructions give further degree–arity tradeoffs on contractions. The degree-six example lowers the degree of the cited six-variable construction using four additional variables. The quadratic construction uses a characteristic-two presentation, whose basic root elements are symmetries, and three-corner projection geometry to convert relation defects into quadratic trace energies. Its characteristic-two source requires a separate proof from the characteristic-five source of [10]. The two-variable construction lowers the cited degree bound from 48 to 24.

Four complementary results

Table 1 summarizes the constructions and the rigidity theorem. Each separator has the normalization (1.2). Degree means ordinary free-word degree; the last row also has cyclic degree 24.

ResultVariablesDegreeFurther information
Quadratic rigidityn≤3n \le3≤2\le2mmat=mtrm_{\mathrm{mat}} = m_{\mathrm{tr}}
Quadratic construction21422,416 ordinary words; symmetry witness
Short-degree construction106392 ordinary words; positive contraction witness
Two-variable construction224Exact cyclic degree; strict contraction witness

Table 1. Dimension-independent tracial separation and the quadratic obstruction.

Write n2n_2 for the smallest number of self-adjoint variables admitting a degree-at-most-two tracial separator, and d2d_2 for the smallest degree of a two-variable separator. Combining the constructions and quadratic rigidity with the cubic obstruction of Corollary 2.3 gives

4≤n2≤214,4≤d2≤24.(3)4 \le n_2 \le214,\qquad4 \le d_2 \le24. \tag*{(3)}

Scalar spectral integration excludes one-variable separators at every degree. Thus two is the minimum arity. In two variables, the cyclically sorted polynomial result of Klep–Schweighofer [5 Proposition 4.2] excludes separators of degree at most three.

Method and proof scope

The constructions convert group-theoretic trace obstructions into low-degree polynomials. For degree six, a weighted graph encodes cyclic products as closed walks; calibration squares and consistency penalties control the decoding on arbitrary matrices. For degree two, projection geometry converts products, relations and spectral constraints into quadratic energies, with a forest of rectangular calibrations. For two variables, a moment encoder realizes auxiliary operators in a small corner, and algebraic corrections control the selector defect. Each construction requires both a uniform finite-matrix lower bound and an exact negative tracial witness.

Section 2 gives the conventions and elementary obstructions; Section 3 proves quadratic rigidity and its quantitative form. After the transfer estimates in Section 4, Sections 5–7 give the three constructions. Section 8 discusses their consequences and limits.

Tracial positivity and elementary reductions

Let AnR=R⟨X1,…,Xn⟩\mathcal{A}^{\mathbb{R}}_n=\mathbb{R}\langle X_1,\ldots,X_n\rangle and An=C⟨X1,…,Xn⟩\mathcal{A}_n=\mathbb{C}\langle X_1,\ldots,X_n\rangle. Involution fixes the variables, conjugates coefficients and reverses words. Write [An,An]lin[\mathcal{A}_n,\mathcal{A}_n]_{\mathrm{lin}} for the C\mathbb{C}-linear span of additive commutators. Two Hermitian polynomials are cyclically equivalent when their difference lies in this span; their tracial evaluations then agree. For f∈Anf\in\mathcal{A}_n, let deg⁡cycf\deg_{\mathrm{cyc}} f be the highest degree of a nonzero coefficient sum on a cyclic word class (if such a class exists).

The contraction quadratic module is

Mn={∑αpα∗pα+∑i=1n∑βqiβ∗(1−Xi2)qiβ:pα,qiβ∈An}.(4)\mathcal{M}_n=\left\{\sum_{\alpha}p_\alpha^*p_\alpha+\sum_{i=1}^{n}\sum_{\beta}q_{i\beta}^*(1-X_i^2)q_{i\beta}:p_\alpha,q_{i\beta}\in\mathcal{A}_n\right\}. \tag*{(4)}

All sums are finite. The real version uses AnR\mathcal{A}^{\mathbb{R}}_n in place of An\mathcal{A}_n. The module is positive under every representation sending generators to self-adjoint contractions, not only finite-dimensional ones. The relevant property is Mn+[An,An]lin\mathcal{M}_n+[\mathcal{A}_n,\mathcal{A}_n]_{\mathrm{lin}}, not the unrestricted cone of Hermitian squares alone.

Proposition 2.1. Suppose f=f∗∈Anf=f^*\in\mathcal{A}_n and there exists a self-adjoint contraction tuple aa in a finite tracial algebra (M,τ)(M,\tau) with τf(a)=−η<0\tau f(a)=-\eta<0. Then, for 0≤ε<η0\leq\varepsilon<\eta,

f+ε∉Mn+[An,An]lin.f+\varepsilon\notin\mathcal{M}_n+[\mathcal{A}_n,\mathcal{A}_n]_{\mathrm{lin}}.

The same holds over the real free algebra when ff has real coefficients.

Proof. Every Hermitian square has nonnegative tracial evaluation. Since ∥ai∥≤1\lVert a_i\rVert\leq1, the operators 1−ai21-a_i^2 are positive, so the same is true of each localizing term in (2.1). Traces vanish on commutators. Membership would therefore imply τf(a)+ε≥0\tau f(a)+\varepsilon\geq0, contrary to −η+ε<0-\eta+\varepsilon<0. □\square

Lemma 2.2. For every Hermitian f∈A1f\in\mathcal{A}_1,

mmat(f)=mtr(f)=min⁡−1≤t≤1f(t).m_{\mathrm{mat}}(f)=m_{\mathrm{tr}}(f)=\min_{-1\leq t\leq1}f(t).

Proof. Write f=f∗=p(X)f=f^*=p(X) for a real scalar polynomial pp. If aa is any self-adjoint contraction in a finite tracial algebra, the spectral theorem supplies a probability measure ν\nu supported on [−1,1][-1,1] with

τp(a)=∫p(t) dν(t)≥min⁡p.\tau p(a)=\int p(t)\,\mathrm{d}\nu(t)\geq\min p.

A scalar matrix a=t0a=t_0, with t0t_0 minimizing pp, attains the bound. □\square

Corollary 2.3. For every Hermitian f∈A2f\in\mathcal{A}_2 of degree at most three,

mmat(f)=mtr(f)=min⁡(s,t)∈[−1,1]2f(s,t).m_{\mathrm{mat}}(f)=m_{\mathrm{tr}}(f)=\min_{(s,t)\in[-1,1]^2}f(s,t).

In particular, d2≥4d_2\geq4.

Proof. Every word in two letters of length at most three is cyclically equivalent to XiYjX^iY^j. Thus ff is cyclically sorted in the sense of Klep–Schweighofer [5 Definition 4.1]. Apply their Proposition 4.2 to ff minus its scalar minimum: every positive scalar perturbation belongs to the contraction quadratic module modulo commutators. Evaluation in any finite tracial algebra and passage to the limit give the lower bound. Scalar matrices attain it. □\square

Quadratic rigidity in three variables

The proof eliminates one variable exactly and approximates the spectral laws of the remaining two operators and their sum by a single matrix pair. We use the scalar Horn theorem of Berkovici, Collins, Dykema, Li and Timotin [14].

Theorem 3.1. If f=f∗∈Akf=f^*\in\mathcal{A}_k, k≤3k\leq3, and deg⁡f≤2\deg f\leq2, then

mmat(f)=mtr(f).m_{\mathrm{mat}}(f)=m_{\mathrm{tr}}(f).

In particular, no quadratic polynomial in at most three self-adjoint contractions can have a uniform positive matrix trace gap and a negative finite-tracial witness.

Modulo commutators, every polynomial under consideration has the form

f=c+∑i=13biXi+∑i=13diXi2+∑1≤i<j≤3rij2(XiXj+XjXi),(5)f=c+\sum_{i=1}^{3}b_iX_i+\sum_{i=1}^{3}d_iX_i^2+\sum_{1\le i<j\le3}\frac{r_{ij}}{2}(X_iX_j+X_jX_i), \tag*{(5)}

All coefficients are real. Absent variables may be adjoined with zero coefficients.

Exact minimization of one operator

For d,t∈Rd,t\in\mathbb{R}, put

φd(t)=min⁡−1≤s≤1(ds2+ts)={−t2/(4d),d>0, ∣t∣≤2d,d−∣t∣,d>0, ∣t∣≥2d,d−∣t∣,d≤0.(6)\varphi_d(t)=\min_{-1\le s\le1}(ds^2+ts)= \begin{cases} -t^2/(4d), & d>0,\ |t|\le2d,\\ d-|t|, & d>0,\ |t|\ge2d,\\ d-|t|, & d\le0. \end{cases} \tag*{(6)}

References

  1. [1]A. Connes, Classification of injective factors. Cases II₁, II∞, IIIλ, λ ≠ 1, Ann. of Math. (2) 104 (1976), no. 1, 73–115.
  2. [2]Z. Ji, A. Natarajan, T. Vidick, J. Wright, and H. Yuen, MIP* = RE, arXiv:2001.04383 (2020).
  3. [3]J. W. Helton, “Positive” noncommutative polynomials are sums of squares, Ann. of Math. (2) 156 (2002), no. 2, 675–694.DOI
  4. [4]S. Burgdorf and I. Klep, The truncated tracial moment problem, J. Operator Theory 68 (2012), no. 1, 141–163.
  5. [5]I. Klep and M. Schweighofer, Connes’ embedding conjecture and sums of Hermitian squares, Adv. Math. 217 (2008), no. 4, 1816–1837.DOI
  6. [6]S. Burgdorf, K. Dykema, I. Klep, and M. Schweighofer, Addendum to “Connes’ embedding conjecture and sums of Hermitian squares”, Adv. Math. 252 (2014), 805–811.
  7. [7]M. Ershov and A. Jaikin-Zapirain, Property (T) for noncommutative universal lattices, Invent. Math. 179 (2010), no. 2, 303–347.DOI
  8. [8]N. Ozawa, Noncommutative real algebraic geometry of Kazhdan’s property (T), J. Inst. Math. Jussieu 15 (2016), no. 1, 85–90.
  9. [9]C. F. Miller III, Combinatorial Group Theory, lecture notes, March 5, 2002, Section 2.3, Theorems 2.5–2.6. https://www.max.hw.ac.uk/~l~c45/Teaching/kggt/miller.pdf.
  10. [10]J. Wang and L. Zhi, An explicit polynomial counterexample to Connes’ embedding conjecture, arXiv:2610.01536 (2026).arxiv.org/abs/2610.01536
  11. [11]G. Kun and A. Thom, Nonsofic wreath products of residually finite groups, arXiv:2608.06222 (2026).arxiv.org/abs/2608.06222
  12. [12]A. Thom, A conditional construction of a nonhyperlinear group and the centralizer problem, arXiv:2609.27795 (2026).arxiv.org/abs/2609.27795
  13. [13]V. Alekseev, J. Liu, and A. Thom, Spectral gap, internality of relative commutants, and non-hyperlinear groups, arXiv:2609.37761 (2026).arxiv.org/abs/2609.37761
  14. [14]H. Bercovici, B. Collins, K. Dykema, W. S. Li, and D. Timotin, Intersections of Schubert varieties and eigenvalue inequalities in an arbitrary finite factor, J. Funct. Anal. 258 (2010), no. 5, 1579–1627.DOI

Paper details

vlnc:W_2e6trxy6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 5

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

GPT-6 Astra

View statement
Looking for related papers…
Contents