Topology of SU(N) gauge theories at T=0 and T=Tc

The new thing here is N=8, plus the ability to sit at T≈Tc and measure the confined and deconfined phases separately on the same lattice because the transition is strongly first order. That is a genuine capability, not a rehash of their earlier work. The confined-phase result — χ_t/σ² essentially equal to its T=0 value at every N — is clean and holds across the whole table. And the trick in Fig. 9, splitting the deconfined fields by net Q and showing that the broad low peaks live in the Q=0 sample while the real charges are narrow, is a nice piece of self-checking. It makes the artifact interpretation credible rather than merely asserted, which is more than most peak-finding papers manage. The paper is also candid throughout: they flag the SU(8) continuum problem, the semi-quantitative D(ρ), and the large 1/N⁴ coefficient possibly contaminated by the bulk transition. Credit for that; it is not hidden anywhere. On the stress-test worry about topology freezing: real, but overstated as stated. The §2.3 freezing argument is about N and lattice spacing, not phase. At N=8, a≈1/5Tc, the confined phase on the same lattice still gives χ_t/σ²=0.0161, so Q is evidently still moving there. That is the control the stress test misses. What is genuinely missing is the diagnostic: no integrated autocorrelation time for Q, no count of Q changes per sequence, no hot-versus-cold start. For the N=8 deconfined point (4/500 fields with Q≠0) that matters, because the physical picture — small instantons suppressed — predicts exactly the slowing that would also produce a spurious zero. The two are hard to separate without the diagnostic. So: a missing check, not a refutation. The N=3 and N=4 points are certainly equilibrated and already show the trend. The other limits are real and acknowledged: one coarse spacing, no continuum extrapolation, and the δ-function claim in eqn 24 is the weakest headline result, resting on the crude peak finder. Who should read it: the large-N topology and U_A(1) crowd, and anyone who needs a reference for phase-separated large-N topology. Recommendation: send it out. Solid, candid, incremental but real. Ask the referees specifically for the Q autocorrelation and tunneling statistics in the deconfined phase at N=6 and N=8.

The headline empirical claim (abstract:s5, eqn 23) rests on the monotone collapse of χ_t/σ² in the deconfined phase: 0.0141 (N=3), 0.0070 (N=4), 0.00171 (N=6), 0.00005 (N=8) in Table 3. The paper's own §2.3 states that at fixed lattice spacing topology suffers critical slowing down in N, Q ceases to change, and ⟨Q²⟩ cannot be calculated in the usual fashion; this is invoked explicitly for the T=0 SU(8) continuum extrapolation. The same argument applies to the deconfined phase at T≈Tc: if Q freezes at its initial value (here Q=0), the measured ⟨Q²⟩ is not the equilibrium susceptibility but a sampling artifact, and the monotone collapse with N is exactly what increasingly complete freezing would produce. The paper reports 4/500 deconfined SU(8) fields with Q≠0 (§4.3.2) but gives no integrated autocorrelation time for Q, no count of Q changes, and no hot-start versus cold-start comparison. Without that, the data cannot distinguish physical suppression from algorithmic freezing. This is distinct from the reader's broad-peak concern: the broad peaks do not contribute to net Q, so their interpretation affects D(ρ) and eqn 24 but not the Table 3 χ_t values that carry the headline claim.

The reader's verdict is careful and largely correct: the deconfinement suppression of χ_t is a real, monotone effect in Table 3, the confined-phase insensitivity to T is consistent across N, and the authors are candid about the single coarse spacing, the five-point 1/N² fit with a large 1/N⁴ coefficient, and the semi-quantitative D(ρ) extraction. The reader's weakest_assumption — that the broad low peaks in Fig. 8/9 are cooling artifacts — is a genuine soft spot, but it is not the most load-bearing concern for the headline claim. Those broad peaks carry zero net Q, so their status affects the size-distribution/delta-function conclusion (eqn 24) rather than the Table 3 χ_t values that support abstract:s5 and eqn 23. The more load-bearing gap is the one the paper itself flags in §2.3 for T=0 SU(8): at fixed a, topology freezes as N grows, and the measured ⟨Q²⟩ is then not the equilibrium susceptibility. The paper does not explicitly connect this to the deconfined-phase measurement, and it provides no autocorrelation time or hot/cold-start check. Because the paper's theoretical argument (small-instanton suppression in §2.3 plus large-instanton suppression in §2.4) is independent of the lattice measurement, the conclusion could still be correct even if the measurement is frozen; the paper is therefore conditionally supported rather than undermined. The verdict label should remain CONDITIONAL, but the condition should include the equilibration check above in addition to the continuum extrapolation and the artifact check the reader already named. I therefore record UNCHANGED rather than a label shift, while noting that the reader's stated condition is incomplete: the artifact check alone would not settle whether the Table 3 collapse is physical or frozen.

If it holds, the result is a useful and fairly clean statement about the large-N phase structure: the deconfined phase loses its topological content entirely, while the confined phase retains a finite susceptibility at all T. The N-range (up to 8) is larger than in earlier lattice work, and the separate treatment of the two phases at a first-order transition is a genuine methodological advantage that the authors exploit well. The qualitative collapse in Table 3 is far too large (a factor of ~300 from N = 3 to N = 8) to be a statistical fluctuation, so the central qualitative conclusion is likely robust. The paper does not ship code or machine-checked material, and the quantitative claims rest on a single coarse lattice spacing; the strength of the paper is the breadth of the N-range and the honesty about systematic uncertainties, not parameter-free derivation.

The paper asks what happens to topological fluctuations — instantons, the lumps of gauge field that carry winding number — in SU(N) gauge theory as N becomes large, and separately in the confined and deconfined phases near the deconfining temperature $T_c$. Because the transition is strongly first order for $N$ between 3 and 8, the authors can sit at $T$ very close to $T_c$ and measure the two phases separately on the same lattice, which is not possible for $N=2$. The answer they report is sharp: the confined phase looks topologically the same at all temperatures, with a susceptibility close to its $T=0$ value, while the deconfined phase loses its topological charge at a rate that grows quickly with $N$, so that at $N=8$ almost every deconfined configuration has $Q=0$. If this is right, then in the 't Hooft large-$N$ limit the deconfined phase has no instantons at all, at any temperature, because the high-temperature suppression of large instantons and the large-$N$ suppression of small instantons meet and overlap. The authors also report that the instanton size distribution in the confined phase narrows as $N$ grows, pointing to a single preferred size of order $1/T_c$ in the limit.

What carries the argument. The machinery is the topological susceptibility $chi_t = angle Q^2 angle / V$, measured on the lattice by cooling Monte Carlo configurations until the field is smooth, then evaluating the standard twisted-plaquette lattice charge density $Q_L(x) = rac{1}{32pi^2} arepsilon_{mu u hosigma} mathrm{ReTr}{U_{mu u}(x)U_{ hosigma}(x)}$ and rounding the total to an integer. Individual instantons are located as peaks in $Q_L(x)$ and assigned a size through the classical relation $Q_{peak} = 6pi^2/ho^4$, which yields the size density $D(ho)$. The other piece of machinery is physical rather than numerical: because the transition is first order for $N$ at least 3, a large enough spatial volume traps the system in one phase for the whole run, so the confined and deconfined susceptibilities can be read off at the same $T$ and the same lattice spacing. The theoretical scaffolding the authors test against is the large-$N$ counting argument that $chi_t$ is $O(N^0)$, the semiclassical suppression $D(ho) ppo e^{-8pi^2/lambda(ho) N}$ of small instantons, and the high-temperature factor $exp(-2N/3 {pi ho T}^2)$ from electric screening.

Core claim. The central claim, stated in the abstract and restated as equation (23), is that $N$ equal to infinity the topological susceptibility is exactly zero in the deconfined phase for every $T$, while it stays finite and non-zero in the confined phase for every $T$. The evidence is a set of lattice measurements at a fixed lattice spacing of $a = 1/5T_c$: the dimensionless ratio $a^4_4^4(1/4)/a$ falls monotonically with $N$ in the deconfined phase, from about $0.0141$ at $N=3$ to $0.00005$ at $N=8$, while the confined-phase value stays close to the $T=0$ value at each $N$. The authors combine this with an analysis of the instanton size distribution $D(ho)$ and argue that in the deconfined phase the only surviving peaks are rare, narrow ones at $ho$ between about 2 and 3.5 lattice units, while the broad, low peaks at large $ho$ do not contribute to the net topological charge and are artifacts of incomplete cooling. In the confined phase, by contrast, $D(ho)$ is largely temperature-independent and becomes more peaked as $N$ grows, which the authors read as evidence that $D(ho)$ tends to a delta function at $ho_c$ of order $1/T_c$.

The premise it rests on. The load-bearing premise is that the broad, low peaks seen in the deconfined phase are artifacts of incomplete cooling rather than real large instantons, an identification the authors themselves describe as based on a crude and semi-quantitative peak-finding algorithm.

In the large-$N$ limit the deconfined phase of quenched QCD has no topological fluctuations at any temperature, so $U_A(1)$ is restored there without any need for a temperature above $T_c$.

The topological susceptibility keeps a finite non-zero value as $N$ goes to infinity in the confined phase, which is the behaviour needed to give the $eta'$ its mass in the large-$N$ counting.

Since at $N$ equal to infinity quenched QCD and QCD with massive quarks coincide, the same no-fluctuation statement should hold for the deconfined phase of QCD in that limit.

In the confined phase the instanton size distribution narrows toward a delta function at a size of order $1/T_c$, so a single size scale controls the topological content of the confining vacuum.

Lattice calculations of topology in the deconfined phase at large $N$ will suffer critical slowing down, because $Q$ can only change through instantons whose probability falls exponentially with $N$.

If the deconfined phase truly has no instantons at large $N$, then the axion cosmology of a confining large-$N$ theory and its deconfined early-universe phase would look qualitatively different from the $N=3$ case, since the topological susceptibility would drop discontinuously across the transition rather than merely falling.

The claim that the broad, low peaks are cooling artifacts could be settled independently by measuring $Q$ with overlap fermions that satisfy the index theorem on the same configurations; agreement between the fermionic charge and the peak-based charge in the deconfined phase would remove the main interpretive uncertainty.

The suggested delta-function form of $D(ho)$ at large $N$ implies a single characteristic instanton size scaling as $1/T_c$, which is a testable statement: a finer lattice spacing at $N=6$ or $8$ should show the peak of $D(ho)$ in physical units staying fixed rather than drifting with the cutoff.

Because the authors work at a fixed coarse lattice spacing, a continuum extrapolation of the deconfined susceptibility at $N=8$ — currently blocked by the freezing of $Q$ — would be the cleanest check of whether the suppression is a physical large-$N$ effect or partly a lattice-spacing effect.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review