# Yang-Mills mass gap hunt: Wilson lattice measurement at two lattice spacings

Date: 2026-11-20. Lattice: 8^4, periodic in all directions. Gauge group: SU(2), chosen because it is cheaper than SU(3) and shows the same confinement physics at this scale. Updates: Metropolis with checkerboard even/odd ordering, random SU(2) rotation near the identity, eps 0.35, seed 20261120, cold start. Thermalization: 400 sweeps. Measurement: 1500 sweeps per beta. Errors are jackknife errors over 30 blocks of 50 sweeps, except Creutz ratios which use 20 blocks of the every-5-sweeps Wilson loop measurements. Total runtime: 338.6 seconds. Full numbers live in data/yang_mills.json. The code is compute/wilson_lattice.py.

Two couplings were run, beta 2.2 and beta 2.4, so the gap estimate can be tracked as the lattice spacing changes. Both are in the confined phase for N_t equal to 8. Nothing in this file proves the continuum mass gap. A lattice gap estimate at finite spacing is a measurement, not a proof. What follows is what the data says at these two spacings.

## Plaquette and string tension

Beta 2.2: average plaquette 0.568932 plus or minus 0.000566, acceptance rate 0.466. Beta 2.4: average plaquette 0.631483 plus or minus 0.000365, acceptance rate 0.427. The plaquette rises with beta, as expected when the coupling weakens.

Creutz ratios chi(R, T), from Wilson loops measured every 5 sweeps:

Beta 2.2: chi(2,2) 0.4054 plus or minus 0.0027, chi(2,3) 0.3676 plus or minus 0.0050, chi(3,3) 0.3033 plus or minus 0.0103. Beta 2.4: chi(2,2) 0.2444 plus or minus 0.0018, chi(2,3) 0.1964 plus or minus 0.0028, chi(3,3) 0.1275 plus or minus 0.0033.

String tension estimate sigma a^2 from chi(2,2): 0.4054 plus or minus 0.0027 at beta 2.2, and 0.2444 plus or minus 0.0018 at beta 2.4. The Creutz ratios fall with loop size at both betas, which is the expected approach toward the asymptotic string tension from above, so chi(2,2) overestimates sigma somewhat. The tension in lattice units falls sharply with beta, as expected when the lattice spacing shrinks.

## Gap channel 1: 0++ glueball from the plaquette correlator

Operator: O(t), the spatial plaquette average in time slice t. Connected correlator C(tau) equal to the average of O(0) O(tau) minus the square of the average of O. Effective mass m_eff(tau) equal to minus the log of C(tau+1) over C(tau).

Beta 2.2: C(0) 0.00011187, C(1) 0.00001562, C(2) 0.00000296, C(3) 0.00000214, C(4) 0.00000459. Effective masses: m_eff(0) 1.9691 plus or minus 0.1592, m_eff(1) 1.6622 plus or minus 0.8471, m_eff(2) 0.3269 plus or minus 1.4657.

Beta 2.4: C(0) 0.00008738, C(1) 0.00000584, C(2) 0.00000228, C(3) 0.00000260, C(4) 0.00000062. Effective masses: m_eff(0) 2.7052 plus or minus 0.3061, m_eff(1) 0.9395 plus or minus 0.6160, m_eff(2) minus 0.1303 plus or minus 0.8400.

Reading: the correlator decays by an order of magnitude from tau 0 to tau 1 at both betas, which is the gap signal. But m_eff(0) rises from beta 2.2 to beta 2.4, which is the wrong direction for a physical mass in lattice units as the spacing shrinks, while the cleaner m_eff(1) falls from 1.66 to 0.94 with errors too large to settle the direction. The tau 0 estimate carries excited-state contamination, so it is an upper estimate, and the tau 1 estimate is noise dominated. The 0++ channel alone cannot fix the direction of the gap with beta.

## Gap channel 2: 2++ tensor glueball from the orientation-difference correlator

Operator: D(t) equal to P_xy(t) minus P_xz(t), the difference of two spatial plaquette orientations in time slice t. Cubic symmetry forces its vacuum expectation value to zero, so its correlator needs no disconnected subtraction: C_D(tau) equal to the average of D(0) D(tau).

Beta 2.2: C_D(0) 0.00038538, C_D(1) 0.00000436, C_D(2) minus 0.00000709. Effective mass m_eff(0) 4.4826 plus or minus 0.8670. No positive correlator at tau 1, so no m_eff(1).

Beta 2.4: C_D(0) 0.00028809, C_D(1) 0.00001147, C_D(2) 0.00000136. Effective masses: m_eff(0) 3.2237 plus or minus 0.3401, m_eff(1) 2.1307 plus or minus 3.3575.

Reading: the tensor correlator collapses after tau 0 at beta 2.2 and barely survives to tau 1 at beta 2.4. The tau 0 values, 4.48 and 3.22, fall with beta, which is the expected direction, and the tensor mass sits above the scalar mass at both betas, which matches the known ordering of glueball states. But the signal dies too fast to give a clean ground-state mass, so this channel corroborates rather than leads.

## Gap channel 3: torelon from the spatial Polyakov loop correlator

Operator: P_s, the Polyakov loop winding around a spatial direction, averaged over the three spatial directions and over transverse sites. Correlator C_T(tau) equal to the average of P_s(0) times the conjugate of P_s(tau), minus the square of the mean loop. It decays as exp(-m_tor tau) with m_tor equal to sigma times L_s in lattice units for a confining theory.

Beta 2.2: C_T(0) 1.0001, C_T(1) 0.0212, C_T(2) 0.0015, C_T(3) 0.0010, C_T(4) minus 0.0012. Effective masses: m_eff(0) 3.8559 plus or minus 0.0593, m_eff(1) 2.6731 plus or minus 0.7993, m_eff(2) 0.4090 plus or minus 0.8868.

Beta 2.4: C_T(0) 0.9957, C_T(1) 0.0490, C_T(2) 0.0086, C_T(3) 0.0052, C_T(4) minus 0.0001. Effective masses: m_eff(0) 3.0113 plus or minus 0.0279, m_eff(1) 1.7393 plus or minus 0.1361, m_eff(2) 0.5041 plus or minus 0.2904.

Reading: this is the cleanest channel. The correlator drops by a factor of about 20 to 50 from tau 0 to tau 1, and the effective mass falls with tau, as a spectral decomposition requires. The m_eff(1) values, 2.67 plus or minus 0.80 at beta 2.2 and 1.74 plus or minus 0.14 at beta 2.4, fall with beta, which is the expected direction as the lattice spacing shrinks. Cross-check against the string tension: sigma times L_s predicts 3.24 at beta 2.2 and 1.95 at beta 2.4, and the measured m_eff(1) values sit below those predictions by 0.57 and 0.21, consistent within errors and in the direction expected from the flux tube Luescher correction and from residual excited-state contamination in m_eff(0). The torelon mass is nonzero at both spacings with clear statistical significance.

## Confinement diagnostic: time-like Polyakov loop correlator

The correlator of time-like Polyakov loops at spatial separation r decays as exp(-N_t V(r)) with V(r) the static quark potential. Beta 2.2: C_P(0) 0.9963, C_P(1) 0.0185, C_P(2) minus 0.0001. Beta 2.4: C_P(0) 0.9852, C_P(1) 0.0421, C_P(2) 0.0075. Static potential at separation 1: V(1) 0.4980 plus or minus 0.0117 at beta 2.2, and 0.3940 plus or minus 0.0067 at beta 2.4. The potential is large and positive at both spacings and falls with beta, consistent with a confining potential that weakens in lattice units as the spacing shrinks.

## Gap estimates and the continuum extrapolation

Primary gap estimate per channel, taking m_eff(1) when its relative jackknife error is under 50 percent and m_eff(0) otherwise:

0++ glueball: 1.9691 plus or minus 0.1592 at beta 2.2, 2.7052 plus or minus 0.3061 at beta 2.4. 2++ tensor glueball: 4.4826 plus or minus 0.8670 at beta 2.2, 3.2237 plus or minus 0.3401 at beta 2.4. Torelon: 2.6731 plus or minus 0.7993 at beta 2.2, 1.7393 plus or minus 0.1361 at beta 2.4.

In units of the string tension scale, m over sqrt(sigma) for the 0++ channel is 3.09 at beta 2.2 and 5.47 at beta 2.4. It rises and moves substantially between the two spacings, so no stable continuum value can be read off. Two lattice spacings cannot fix an extrapolation: no functional form in a^2 can be fit, no third spacing confirms any trend, and the 8^4 volume leaves finite-volume effects uncontrolled. The glueball channel disagrees with itself between tau 0 and tau 1, which is the signature of excited-state contamination at tau 0 and noise at tau 1.

What the data does say: every channel shows a nonzero mass at finite spacing, the torelon channel moves in the expected direction with beta and stays nonzero at both spacings, and the static potential is large and positive. That points in the direction of a nonzero continuum gap. It does not prove the continuum mass gap. A real extrapolation needs at least three beta values, larger volumes, better glueball operators with less excited-state contamination, and a controlled continuum limit.

## Notes on the run itself

Two bugs were found and fixed during development, before the numbers above were produced. First, the Metropolis sweep originally reused one staple sum for both parities of a direction, but updating the first parity changes the links the second parity's staples depend on, so the second parity was accepted against a stale staple sum and the chain did not sample the Wilson distribution. Staples are now recomputed per direction and parity. Second, the time-like Polyakov loop originally multiplied links in the wrong direction, producing a product that is not a closed loop and is not gauge invariant, which showed up as a correlator of exactly zero at all separations. It now multiplies the time-direction links along time. A third error, an inflated jackknife variance formula, was caught from the plaquette error and fixed; the corrected jackknife was verified against synthetic data with known standard errors before the production run. The JSON on disk is from the corrected production run.
