---
title: 'Hyperlinear: Groups & STM Insights'
url: https://www.emergentmind.com/topics/hyperlinear
type: topic
---

# Hyperlinear: Groups & STM Insights

Searching arXiv for recent papers on hyperlinear groups and related topics.
Hyperlinear is a technical adjective with distinct meanings in contemporary research. In geometric group theory, operator algebras, and quantum groups, it denotes approximability by finite-dimensional unitary groups equipped with the normalized Hilbert–Schmidt metric, equivalently a Connes-embeddability property for the associated von Neumann algebra [1309.2034; 2506.20843]. In scanning tunneling microscopy, it denotes a current dependence that is more than linear and, in the CO/Si(001) system, is associated with multiple vibrational excitation and local vibronic heating [1210.0963]. The term therefore does not name a single unified theory, but two established technical usages.

## 1. Formal mathematical meaning

For a countable discrete group \(G\), the standard unitary model uses the normalized Hilbert–Schmidt norm
\[
\|A\|_2 := \sqrt{\frac{1}{n}\operatorname{Tr}(A^*A)}
\]
on \(U(n)\), with induced bi-invariant metric
\[
d_{HS}(U,V):=\|U-V\|_2.
\]
A group is hyperlinear if there exists a non-principal ultrafilter \(\omega\) and an embedding
\[
G\hookrightarrow \prod_{n\to\omega}(U(n),d_{HS}),
\]
or, equivalently, if every finite portion of the multiplication table can be approximated inside some \(U(n)\) by maps that are approximately multiplicative and uniformly separate nontrivial elements from the identity [2004.05735; 1309.2034].

Several equivalent formulations are used in the literature. One formulation requires maps \(\phi:G\to U(n)\) with \(\phi(1)=I_n\), approximate multiplicativity on a prescribed finite set, and a trace-separation condition \(|\operatorname{Tr}(\phi(g))|<\varepsilon\) for \(g\neq 1\), which is equivalent to \(\phi(g)\) being almost orthogonal to the identity in Hilbert–Schmidt norm [2004.05735]. Another formulation uses asymptotic unitary representations \(\pi_n:\Gamma\to U(d_n)\) satisfying
\[
\lim_{n\to\infty}\|\pi_n(gh)-\pi_n(g)\pi_n(h)\|_{2,d_n}=0
\]
for all \(g,h\), together with
\[
\lim_{n\to\infty}\|\pi_n(g)-\pi_n(h)\|_{2,d_n}=\sqrt{2}
\]
for \(g\neq h\) [2506.20843].

A von Neumann algebraic formulation is also standard: a countable discrete group \(\Gamma\) is hyperlinear, or Connes-embeddable, if its group von Neumann algebra \(L(\Gamma)\) embeds trace-preservingly into a tracial ultraproduct of matrix algebras, equivalently into an ultrapower \(R^\omega\) of the hyperfinite \(\mathrm{II}_1\) factor [2506.20843; 1309.2034]. In continuous logic, this extends to bi-invariant metric groups: hyperlinear metric groups are those that embed isometrically as closed subgroups of ultraproducts of \((U(n),d_{HS})\), and the class is sup-axiomatizable [1604.08446].

## 2. Position in approximation theory

Hyperlinearity is one member of a broader family of metric-approximation properties. Arzhantseva–Paunescu and Thom’s framework, as presented by Brude–Sasyk, treats several classes uniformly [2004.05735].

| Property | Approximating class | Metric |
|---|---|---|
| Weakly sofic | Finite groups | Any bi-invariant metric |
| Sofic | \(\mathrm{Sym}(A)\) | \(d_{Hamm}\) |
| Linear-sofic | \(\mathrm{GL}_n(K)\) | \(d_{rk}\) |
| Hyperlinear | \(U(n)\) | \(d_{HS}\) |

Within this hierarchy, every sofic group is known to be hyperlinear by sending permutations to permutation matrices and comparing Hamming distance with Hilbert–Schmidt distance [1309.2034]. The converse remains open in the cited literature, as does the existence of any non-sofic or non-hyperlinear group [2004.05735; 1309.2034].

The class includes many standard examples. Finite groups, residually finite groups, and amenable groups are hyperlinear [1309.2034]. The notion is also linked to longstanding structural problems. Capraro–Lupini’s survey emphasizes its role in the Connes embedding framework for group von Neumann algebras [1309.2034], while Klyachko–Thom use hyperlinearity to solve a family of equations over groups: if \(w\in G*F_2\) has content \(E(w)\notin \gamma_2(F_2)\), then \(w=1\) has a solution over a hyperlinear group \(G\), and if \(G\) is finite the solution can be found in a finite extension [1509.01376].

A recurrent misconception is to identify hyperlinearity with arbitrary matrix approximability. The defining feature is not merely approximation by matrices, but approximation by unitaries in the normalized Hilbert–Schmidt geometry, or equivalently embeddability into the corresponding tracial ultraproducts [1309.2034; 1604.08446].

## 3. Permanence under amenability

One of the strongest permanence theorems currently available concerns amenable actions. Brude and Sasyk proved that if \(H\) is amenable and \(G\) is hyperlinear, then the unrestricted wreath product
\[
G\wr H = \left(\prod_H G\right)\rtimes H
\]
is hyperlinear [2004.05735]. The same theorem is proved simultaneously for weakly sofic, sofic, and linear-sofic groups.

The proof proceeds in two stages. First, amenability of \(H\) provides a large finite Følner set \(B\) and a near-action \(\sigma:H\to \mathrm{Sym}(B)\) that is Hamming-multiplicative on a prescribed finite set. Second, the resulting permutational wreath product \(K\wr_B \mathrm{Sym}(B)\) is embedded into a larger unitary group by a block-matrix construction that controls the Hilbert–Schmidt metric [2004.05735]. In the hyperlinear case this yields an \((F,\varepsilon)\)-multiplicative, trace-preserving map of \(G\wr H\) into a finite unitary group, hence hyperlinearity.

Two immediate corollaries are especially important. If
\[
1\to N\to G\to Q\to 1
\]
is exact with \(Q\) amenable and \(N\) hyperlinear, then \(G\) is hyperlinear. Likewise, if \(H<G\) is co-amenable and \(H\) is hyperlinear, then \(G\) is hyperlinear [2004.05735]. These results place hyperlinearity on the same permanence footing as soficity and linear-soficity under amenable extensions.

A complementary amenability result is quantitative rather than closure-theoretic. For amenable groups, every hyperlinear approximation is essentially produced from a sofic approximation: Burton proves that for every finite \(E\subseteq G\) and \(\epsilon>0\), there are \(F\subseteq G\) and \(\delta>0\) such that any \((F,\delta)\)-hyperlinear approximation is close in Hilbert–Schmidt norm to a sofic-induced approximation arising from an \((E,\epsilon)\)-sofic partial action [2311.09202]. This gives an effective version of “hyperlinear \(\Rightarrow\) sofic” in the amenable case.

## 4. Quantitative theory: hyperlinear profile and non-local games

The qualitative definition of hyperlinearity admits a quantitative refinement through the hyperlinear profile. For a finitely presented group \(G=\langle S:R\rangle\), an \(\varepsilon\)-representation in dimension \(d\) is a map from the free group on \(S\) to \(U(d)\) that satisfies each relator up to normalized Frobenius error at most \(\varepsilon\). Given a finite set of nontrivial words \(T\), the quantity \(\Lambda_G(T;\delta,\varepsilon)\) is the smallest dimension permitting an \(\varepsilon\)-representation in which every word in \(T\) stays at least \(\delta\) away from the identity [1711.10676].

Slofstra and Vidick introduced this profile to relate group approximation to the entanglement cost of linear-system non-local games. They exhibit a finitely presented group
\[
K=\langle a,b,c,x,y : a^2=b^2=c^2=e,\ c=ab,\ x y x^{-1}=y^2,\ x c x^{-1}=c,\ y a y^{-1}=b,\ y b y^{-1}=a\rangle
\]
and prove, for the central involution \(c\), the bounds
\[
\Lambda_K(c;2,\varepsilon)=\Omega(1/\varepsilon^{2/3}), \qquad \Lambda_K(c;2,\varepsilon)=O(1/\varepsilon),
\]
for small \(\varepsilon\) [1711.10676]. After embedding into a solution group \(\Gamma(A,b)\), they obtain a fixed non-local game for which the amount of entanglement required to play \(\varepsilon\)-optimally grows polynomially; the advertised lower bound is \(d(\varepsilon)\ge C/\varepsilon^k\) with \(k=1/6\) [1711.10676].

Slofstra later constructed a finitely presented group with at least subexponential hyperlinear profile. In that construction there are constants \(\alpha\in(0,\tfrac12)\), \(C>0\), and \(C'>0\) such that
\[
\operatorname{hlp}(\{w\};\delta,\varepsilon)\ge C' \exp\!\bigl[C(\delta/\varepsilon)^\alpha\bigr]
\]
for a distinguished word \(w\) [1806.05267]. The same paper derives a two-player non-local game \(\mathcal G\) with \(\omega_{co}(\mathcal G)=1\) such that any finite-dimensional strategy achieving success at least \(1-\varepsilon\) must use local Hilbert spaces of dimension at least
\[
C'\exp[C/\varepsilon^\alpha].
\]
This shifted hyperlinearity from a purely qualitative approximation property to a source of explicit lower bounds in quantum information theory [1806.05267].

## 5. Quantum-group generalizations

The term also has a precise quantum-group analogue. For a compact quantum group \(G\) of Kac type, the discrete dual \(\widehat G\) is called hyperlinear if the von Neumann algebra \(L^\infty(G)\) admits a Haar-state-preserving embedding into the ultraproduct \(R^\omega\) of the hyperfinite \(\mathrm{II}_1\) factor [1808.08611]. In this setting one has the implication chain
\[
\text{residual finiteness} \Rightarrow \text{Kirchberg factorization} \Rightarrow \text{hyperlinearity}
\]
for discrete quantum groups [1808.08611; 1904.03974].

Brannan, Chirvasitu, and Freslon prove that the discrete duals \(\widehat{S_N^+}\) are residually finite for all \(N\ge 4\), hence hyperlinear, using topological generation of \(S_N^+\) by \(S_N\) and \(S_{N-1}^+\) [1808.08611]. They also prove that for every \(N\ge 4\) and \(1\le s\le \infty\), the discrete duals \(\widehat{H_N^{s+}}\) are residually finite and hence hyperlinear [1808.08611]. In the same work, the universal flat matrix model
\[
\pi:\Pol(S_N^+)\to M_N(C(X_N))
\]
is shown to be inner faithful for \(N=5\), and by induction for all \(N\le 5\) and all \(N\ge 10\); moreover \(\Pol(S_N^+)\) admits an inner faithful finite-dimensional \(*\)-representation for all \(N\notin\{6,7,8,9\}\) [1808.08611].

A parallel inductive strategy handles free unitary and orthogonal quantum groups. Tarrago and Weber prove that for every \(N\ge 3\),
\[
U_N^+ = \overline{\langle U_N,\ U_{N-1}^+\rangle}, \qquad
O_N^+ = \overline{\langle O_N,\ O_{N-1}^+\rangle},
\]
which yields residual finiteness, factorization, and hyperlinearity for all \(\widehat{U_N^+}\) and \(\widehat{O_N^+}\) with \(N\ge 2\) [1904.03974]. These examples show that hyperlinearity is not confined to classical discrete groups but extends naturally to discrete duals of noncommutative compact quantum groups.

## 6. Stability, rigidity, and current routes to non-hyperlinear groups

Recent work has recast hyperlinearity in terms of stability and amenable near actions. Kahl and Schneider prove that a group is hyperlinear if and only if it admits an essentially free amenable near representation on \(\ell^2(X)\), thereby answering a question of Pestov and Kwiatkowska [2504.10988]. In their formulation, one works with a finitely additive probability charge \(\mu\) on an orthonormal basis \(X\), a map \(T:G\to U(\ell^2(X))\) that is multiplicative \(\mu\)-almost everywhere, and an essential freeness condition
\[
\forall g\neq e,\ \forall \varepsilon>0,\quad
\mu\{x\in X:\ |\langle x,T(g)x\rangle|\le \varepsilon\}=1.
\]
They also obtain analogous characterizations of Kirchberg’s factorization property and the Haagerup property [2504.10988].

A different recent direction uses Hilbert–Schmidt stability to seek non-hyperlinear groups. Dogon shows that if \(T\) has property (T), \(A\) is torsion-free abelian, and
\[
1\to A\to G\to T\to 1
\]
is a non-split central extension satisfying additional cohomological hypotheses, then flexible HS-stability or even weak ucp-stability of \(T\) forces \(G\) to be non-hyperlinear [2211.10492]. One corollary is that if \(\mathrm{Sp}_{2g}(\mathbb Z)\) were flexibly HS-stable, then there would exist a non-hyperlinear group [2211.10492].

Dogon and Vigdorovich sharpen this mechanism for higher-rank lattices. If \(\Gamma\) is an irreducible lattice in a center-free semisimple Lie group of real rank at least \(2\), has property \((T;FD)\), and is flexibly HS-stable, then any infinite central extension
\[
1\to A\to \widetilde{\Gamma}\to \Gamma\to 1
\]
with \(A\) abelian and \(\widetilde{\Gamma}\) having finite abelianization is not hyperlinear [2506.20843]. The same paper links hyperfinitely HS-stable behavior to character rigidity: every character is either finite-dimensional or induced from the center [2506.20843]. A positive answer to their stability problem for \(\mathrm{SL}_2(\mathbb Z[1/p])\) would yield an explicit non-hyperlinear central extension by \(\mathbb Z\) [2506.20843]. This suggests that the non-hyperlinear-group problem may be accessible through rigidity and stability rather than through direct obstruction to matrix approximation.

## 7. Hyperlinear dependence in scanning tunneling microscopy

Outside group theory, “hyperlinear” appears in surface science to describe a current dependence that is more than linear. In the STM study of CO adsorbed on Si(001), the irreversible lateral motion of a CO molecule showed a hyperlinear dependence on tunneling current, meaning that the displacement rate \(R\) increased faster than \(I\) but could not be fit by a single power law \(R(I)=kI^n\) over the full current range [1210.0963].

Experimentally, a clean Si(001)–\(c(4\times2)\) surface was prepared at \(90\,\mathrm K\), dosed with approximately \(100\,\mathrm L\) of CO, and scanned repeatedly at \(V_{\text{sample}}=-1.6\,\mathrm V\). Under these conditions, CO initially adsorbed invisibly at the down-dimer site (T-CO) and was gradually converted to a bridge site (B-CO), seen as a bright spot at the center of a Si dimer. The rate \(R(I)\) was obtained by counting new bright-spot events at set-point currents \(100\,\mathrm{pA}\), \(300\,\mathrm{pA}\), \(600\,\mathrm{pA}\), \(900\,\mathrm{pA}\), and \(1.2\,\mathrm{nA}\); the resulting log–log plot was curved rather than linear [1210.0963].

The physical interpretation is a multiple-vibration, or ladder-climbing, mechanism. Each tunneling electron can deposit \(\hbar\omega\approx 150\,\mathrm{meV}\) into a local CO vibration, and the T-CO vibrational lifetime of approximately \(2\,\mathrm{ns}\) is comparable to or longer than the average electron arrival time of approximately \(1\,\mathrm{ns}\) at \(1\,\mathrm{nA}\), allowing successive excitations to accumulate until the activation barrier is crossed [1210.0963]. Instead of a pure power law, the data are described by a local vibronic temperature
\[
T_v=T_0+\gamma I^2,\qquad T_0=90\,\mathrm K,
\]
combined with the Arrhenius form
\[
R=A\exp\!\left[-\frac{E_a}{k_B T_v}\right].
\]
Fits yield
\[
E_a=0.11\pm0.05\,\mathrm{eV},\qquad
\gamma=14.3\pm7.3\,\mathrm{K/A^2},\qquad
A=10^8\text{--}10^9\,\mathrm{s}^{-1},
\]
and first-principles calculations give adiabatic barriers of approximately \(0.17\,\mathrm{eV}\) for the neutral slab and approximately \(0.15\,\mathrm{eV}\) with one extra electron [1210.0963]. The efficient local heating is attributed to a tip-induced mid-gap state at T-CO, which enhances inelastic tunneling; no analogous state exists at B-CO, so local heating there is weak [1210.0963].

In this STM usage, hyperlinear therefore denotes super-linear but non-power-law current dependence associated with local vibrational pumping and atomic-scale heating, rather than the unitary-approximation property studied in group theory.

Source: https://www.emergentmind.com/topics/hyperlinear