---
title: Partial Deconfinement in Large-N Gauge Theories
url: https://www.emergentmind.com/topics/partial-deconfinement
type: topic
---

# Partial Deconfinement in Large-N Gauge Theories

Partial deconfinement is an intermediate large-\(N\) gauge-theory phase in which only a subset of color degrees of freedom is thermally active. Instead of a direct jump from confinement, with thermodynamic quantities of order \(N^0\), to complete deconfinement, with order \(N^2\), the theory can pass through a regime where an \(\mathrm{SU}(M)\subset \mathrm{SU}(N)\) subsector with \(0<M<N\) is deconfined while the complementary sector remains confined. In holographic settings, this phase is identified with the small black hole between thermal AdS or graviton gas and the large AdS black hole, and it has been proposed as a microscopic language for intermediate black-hole phases and for the emergence of spacetime from color degrees of freedom [2210.11216].

## 1. Definition and color-space interpretation

In its standard formulation, partial deconfinement distinguishes three regimes:
\[
M=0 \quad \text{(completely confined)}, \qquad 0<M<N \quad \text{(partially deconfined)}, \qquad M=N \quad \text{(completely deconfined)}.
\]
The defining statement is not merely that the system is “less deconfined” than usual, but that color space itself splits into confined and deconfined sectors. The deconfined sector behaves approximately as an \(\mathrm{SU}(M)\) gauge theory embedded inside the full \(\mathrm{SU}(N)\) theory [2210.11216].

A useful intuition is the block-submatrix picture. For matrix degrees of freedom transforming under \(\mathrm{SU}(N)\), one may think of an \(M\times M\) block as dynamically active, while the remaining matrix entries stay in the confined sector. In weakly coupled constructions, one can explicitly build states by restricting color indices to \(1,\dots,M\), obtaining \(\mathrm{SU}(M)\)-invariant states, and then restoring full \(\mathrm{SU}(N)\) gauge invariance by averaging over gauge transformations. This yields a gauge-invariant realization of an \(\mathrm{SU}(M)\) deconfined block inside the \(\mathrm{SU}(N)\) theory [1909.09118].

This coexistence occurs in internal color space rather than in physical space. That distinction is central. Ordinary phase coexistence separates phases spatially; partial deconfinement separates them among matrix indices and holonomy eigenvalues. In that sense, the phenomenon is closely related to what was later described more generally as partial symmetry breaking in large-rank systems with non-local interactions, with an effective pattern
\[
\mathrm{SU}(N)\to \mathrm{SU}(M)\times \mathrm{SU}(N-M)\times \mathrm{U}(1),
\]
subject to the standard caveat that local gauge symmetry is not literally broken in the Elitzur-theorem sense [1911.06223].

## 2. Thermodynamic structure and phase transitions

The thermodynamic rationale follows large-\(N\) counting. In confinement, entropy and excitation energy are \(O(N^0)\); in complete deconfinement they are \(O(N^2)\). Partial deconfinement fills the intermediate regime because the active degrees of freedom are those of the deconfined \(\mathrm{SU}(M)\) sector, so
\[
E \sim M^2, \qquad S \sim M^2.
\]
If the total excitation energy scales as
\[
E \sim \epsilon N^2, \qquad 0<\epsilon<1,
\]
then the natural estimate is
\[
M \sim \sqrt{\epsilon}\,N.
\]
This scaling is the simplest quantitative expression of the idea that only part of color space has thermalized [1911.11465].

At large \(N\), the standard organization is
\[
\text{confined} \;\longrightarrow\; \text{partially deconfined} \;\longrightarrow\; \text{fully deconfined}.
\]
The lower boundary is associated with a Hagedorn transition, where \(M\) first becomes nonzero, and the upper boundary with a Gross–Witten–Wadia transition, where the deconfined block expands to \(M=N\) [2210.11216]. In this interpretation, the GWW transition is not the onset of deconfinement but its completion.

The partially deconfined branch is especially natural in the microcanonical ensemble. In many holographic examples it is stable or meaningful at fixed energy even when it is metastable or unstable in the canonical ensemble. This point is crucial for small black holes, which can have negative specific heat. Canonically, such a branch appears as a local free-energy maximum rather than a minimum; microcanonically, it remains a legitimate family of states [1812.05494].

The possibility of negative specific heat follows from the fact that the number of active degrees of freedom itself depends on energy. If the effective temperature is estimated by energy per active degree of freedom, roughly \(T\sim E/M^2\), then sufficiently rapid growth of \(M(E)\) can cause temperature to decrease as energy increases. This mechanism underlies the identification of unstable partial deconfinement with the small-black-hole branch in AdS/CFT [1812.05494].

## 3. Polyakov loop and holonomy eigenvalue diagnostics

The most widely used diagnostic is the Polyakov loop,
\[
P \equiv \frac{1}{N}\operatorname{Tr}\,\mathcal{P}\exp\!\left[i\int_0^\beta dt\,A_t\right]
     = \frac{1}{N}\sum_{j=1}^N e^{i\theta_j}
     = \int d\theta\, \rho^{(\mathrm P)}(\theta)e^{i\theta},
\]
together with the large-\(N\) eigenvalue density \(\rho^{(\mathrm P)}(\theta)\) [2210.11216].

The phase structure is encoded in the shape of \(\rho^{(\mathrm P)}(\theta)\). In the completely confined phase,
\[
\rho^{(\mathrm P)}(\theta)=\frac{1}{2\pi},
\]
the uniform distribution. In the partially deconfined phase, the distribution is non-uniform but remains positive on the entire interval \([-\pi,\pi)\); it is distorted but ungapped. At the GWW point, a gap opens at \(\theta=\pm \pi\). In the completely deconfined phase, the distribution is gapped [2210.11216].

A particularly useful quantitative relation is
\[
\rho^{(\mathrm P)}_{\min}=\frac{1}{2\pi}\left(1-\frac{M}{N}\right),
\]
so the minimum of the eigenvalue density directly measures the confined fraction \(1-M/N\). This is important because it does not rely on exact center symmetry; the detailed shape of \(\rho(\theta)\), rather than only \(\langle P\rangle\), can diagnose partial deconfinement even in theories with fundamental matter [2005.04103].

In solvable weak-coupling examples, the partially deconfined density takes an explicit mixed form. For weakly coupled Yang–Mills on \(S^3\),
\[
\rho_{\mathrm{p.d.}}(\theta)
=
\left(1-A\right)\frac{1}{2\pi}
+
A\,\frac{1+\cos\theta}{2\pi},
\qquad
A=\frac{M}{N},
\]
so the full density is a convex combination of the confined distribution and the GWW critical distribution [1911.11465]. In the gauged Gaussian matrix model at the critical temperature,
\[
\rho^{(\mathrm P)}(\theta)=\frac{1}{2\pi}(1+2P\cos\theta),
\qquad
P=\frac{M}{2N},
\]
which is the same structural decomposition written in terms of the fundamental Polyakov loop [2005.04103].

## 4. Holography, submatrix deconfinement, and black holes

The holographic interpretation is one of the main motivations for the subject. In AdS/CFT, the standard correspondence identifies the confined gauge phase with thermal AdS or graviton gas and the fully deconfined phase with the large AdS black hole. Partial deconfinement fills the intermediate region and is proposed as the gauge-theory dual of the small black hole [2210.11216].

For \(4d\ \mathcal N=4\) super Yang–Mills on \(S^3\), the partially deconfined phase is identified with the small black hole and with the Hagedorn string regime between thermal AdS and the large black hole. For a small black hole approximated by a ten-dimensional Schwarzschild solution, the energy scales as
\[
E \sim N^2 T^{-7},
\]
which implies negative specific heat. Partial deconfinement supplies the gauge-theory explanation: only an \(\mathrm{SU}(M)\) sector is thermalized into a bound state, rather than the full \(N^2\) adjoint sector [2210.11216].

Hanada, Ishiki, and Watanabe sharpened this picture by interpreting the partially deconfined sector as a D-brane bound state. If \(M\) of the \(N\) D-branes form the bound state, then the relevant open-string modes scale as \(M^2\), and the effective coupling of that sector is
\[
\lambda_{\rm BH}=g_{\rm YM}^2M=\lambda\frac{M}{N}.
\]
From the size of the bound state they argued that
\[
T\sim \lambda_{\rm BH}^{-1/4},
\]
leading to
\[
E\sim N^2T^{-7}, \qquad S\sim N^2T^{-8},
\]
the expected small-Schwarzschild-black-hole scaling [1812.05494].

Berenstein formulated the same idea as submatrix deconfinement in gauged multi-matrix quantum mechanics. In the microcanonical window
\[
1\ll E\ll N^2,
\]
typical states can be interpreted as excitations of an \(M\times M\) submatrix with
\[
M\sim \sqrt{E},
\]
so that \(E\sim M^2\) and \(S\sim M^2\). In that formulation, the same microcanonical states admit both a long-string interpretation and a submatrix or D-brane interpretation, providing a concrete realization of the smooth string/black-hole correspondence [1806.05729].

The operator-language version of this correspondence identifies long traces with very long strings, black holes, or Hagedorn strings, and short traces with gravitons or tiny deconfined blocks. Black-hole growth then corresponds to more colors joining the deconfined sector; evaporation corresponds to the reverse [2210.11216].

## 5. Models, evidence, and extensions toward QCD

The original evidence base combines solvable weak-coupling models, microcanonical state counting, numerical matrix-model studies, and lattice simulations. In weakly coupled \(4d\) Yang–Mills on \(S^3\), the interval between the Hagedorn point and the GWW point is analytically tractable and can be reinterpreted as the growth of a deconfined \(\mathrm{SU}(M)\) block from \(M=0\) to \(M=N\). The same program was carried out explicitly in the gauged Gaussian matrix model and in the \(\mathrm O(N)\) vector model, where state counting shows that the relevant singlet-sector states are exactly those expected from a truncated \(M\)-color subsector [1909.09118].

The microcanonical counting problem was pushed further in the free two-matrix singlet model. There the Hagedorn regime was shown to be controlled by typical Young diagrams of VKLS shape, and the endpoint of partial confinement or partial deconfinement occurs when the diagram depth reaches the maximal allowed depth \(N\). This gives
\[
E_{\text{end}}=\frac{N^2}{4}
\]
up to \(O(N)\) corrections, and the paper argues that this endpoint is independent of the charge \(Q\) [2307.06122]. This suggests a precise representation-theoretic criterion for when the partially deconfined regime terminates.

At strong coupling, direct nonperturbative evidence was obtained in lattice Monte Carlo studies of the bosonic Yang–Mills matrix model. Using static diagonal gauge and suitable constraints on Polyakov phases, Hanada and collaborators reported that gauge-fixed configurations exhibit an approximately deconfined \(M\times M\) block inside the full matrix, with the excess excitation energy localized in that block. In this setting, the Polyakov-loop density, matrix-element histograms, and constrained simulations all support the interpretation of an \(\mathrm{SU}(M)\) subgroup deconfining while the rest remains in a confined background [2005.04103].

Applications to QCD remain more conjectural but have become increasingly concrete. Earlier work emphasized that partial deconfinement does not require exact center symmetry and can already be demonstrated in weakly coupled theories with fundamental matter on \(S^3\) [1911.11465]. A later proposal argued that, in large-\(N\) QCD in the Veneziano limit, a partially deconfined phase must intervene between complete confinement and complete deconfinement unless the theory makes a direct first-order jump. For real-world \(\mathrm{SU}(3)\) QCD, the same work proposed finite-\(N\) diagnostics based on departure from the finite-\(N\) Haar distribution, the earlier rise of the fundamental Polyakov loop, the delayed rise of higher-representation loops, and the disappearance of instanton condensation. In the lattice data examined there, the fundamental Polyakov loop starts to increase at
\[
T\sim 174\ \text{MeV},
\]
whereas adjoint, rank-2 symmetric, and rank-3 symmetric loops grow only for
\[
T\gtrsim 300\ \text{MeV},
\]
with topological-charge peaks at nonzero integer \(Q\) disappearing by
\[
T\gtrsim 348\ \text{MeV}.
\]
Those observations were interpreted as evidence for a finite-\(N\) remnant of partial deconfinement rather than as proof of sharply distinct thermodynamic phases [2312.17136].

## 6. Analogies, misconceptions, and open questions

Several analogies have been used to clarify the mechanism. One is Bose–Einstein condensation: the confined sector is analogous to the condensed component, while the deconfined colors are analogous to excited particles, so partial deconfinement resembles a state in which part of the system remains condensed while the rest is excited [2210.11216]. Another is the ant-trail model used by Hanada, Ishiki, and Watanabe, where D-branes, open strings, and black holes are mapped to ants, pheromones, and trails. The intended lesson is the positive-feedback structure: once a sufficiently large block has formed, it attracts further degrees of freedom more efficiently [1812.05494].

A recurring misconception is to identify partial deconfinement with ordinary mixed phases in real space. The canonical formulation instead places coexistence in color space, among holonomy eigenvalues and matrix indices. A second misconception is to read the \(\mathrm{SU}(M)\) subsector literally as spontaneous breaking of local gauge symmetry. The large-\(N\) description often uses symmetry-breaking language, but the more careful statement is that gauge-invariant states can be represented by averaging block-structured configurations over the full gauge group, and that at large \(N\) a superselection-like structure makes the block picture meaningful [2210.11216].

The term should also be distinguished from other uses of “partial confinement.” In the spin-1 \(1+1\)-dimensional \(\mathrm U(1)\) quantum link model, “partial confinement” denotes a configuration-dependent phenomenon in which opposite charges are confined for one spatial ordering and deconfined for the opposite ordering. That is a different mechanism from large-\(N\) color-subsector deconfinement and is not a reformulation of the \(\mathrm{SU}(M)\subset \mathrm{SU}(N)\) framework [2404.18095]. Likewise, Kharzeev’s proposal of entropy-driven delocalization near \(T_c\) is conceptually adjacent because it describes a gradual, selective onset of deconfinement across states and length scales, but it does not formulate partial deconfinement in the modern large-\(N\) submatrix sense [1409.2496].

Open problems remain substantial. The cleanest phase structure is a large-\(N\) statement, and at finite \(N\) one expects crossovers or smoothed transitions rather than mathematically sharp Hagedorn and GWW singularities [2210.11216]. The precise holographic dictionary for all intermediate phases is still being refined, especially beyond highly symmetric matrix models. Extensions to realistic QCD remain incomplete. Finally, anomaly arguments constrain but do not by themselves establish partial deconfinement: Shimizu and Yonekura showed that in several classes of gauge theories a chirally restored yet otherwise trivial confined phase is impossible below deconfinement, so any intermediate phase, including a partially deconfined one, must realize the relevant center/chiral or center-flavor/chiral anomalies nontrivially [1706.06104].

Partial deconfinement therefore occupies a distinctive position in gauge theory and holography. It is simultaneously a thermodynamic intermediate phase, a statement about the decomposition of color space, a diagnostic framework based on holonomy eigenvalue distributions, and a microscopic model for small black holes. Across weakly coupled gauge theories on \(S^{d-1}\), free and interacting matrix models, lattice simulations, and QCD-inspired analyses, the central claim remains the same: deconfinement need not liberate all colors at once. Instead, color deconfinement can proceed by progressive activation of an \(\mathrm{SU}(M)\) subsector, with the GWW point marking the boundary between partial and complete deconfinement [1911.11465].

Source: https://www.emergentmind.com/topics/partial-deconfinement