How a glueball mass is measured

A lattice mass measurement is a chain of inferences. Gauge configurations produce estimated correlators; spectral analysis produces a lattice energy; scale setting and controlled limits connect it to a continuum physical quantity.

Prerequisites: correlators, operator overlap, and continuum limits.

By the end:

Interpret a continuum-extrapolated mass ratio.
Separate fitting, volume, cutoff, and scale uncertainties.

Build an operator with the right quantum numbers

Start with closed spatial Wilson loops. Combine loop shapes and orientations so that the resulting operator transforms in a chosen representation of the lattice rotation group and has definite parity and charge conjugation.

A zero-momentum projection sums over spatial positions. In a scalar vacuum channel, subtract the expectation value:

\widetilde O(t)=O(t)-\langle O\rangle.

Otherwise a constant vacuum contribution can dominate the long-time signal. The measured connected correlator is

C(t)=\langle\widetilde O^\dagger(t)\widetilde O(0)\rangle.

Smearing the spatial links or using larger loops can improve overlap with low-energy states. It is a measurement choice that needs its own systematic checks.

Estimate correlations with an ensemble

Generate configurations with the lattice probability measure and estimate expectation values by averages. Consecutive configurations in a Markov chain are generally correlated. If $N$ saved measurements have significant autocorrelation, the effective amount of independent information is smaller than $N$ .

Errors must therefore reflect the sampling history. Binning, autocorrelation analysis, and appropriately constructed bootstrap or jackknife procedures are tools for this task. Repeated measurements on the same configurations also correlate different time separations.

This matters because a fit to $C(t)$ usually uses several times from the same ensemble. Treating them as independent can give misleading parameter uncertainties.

Find an energy window

If a single spectral contribution dominates and the time extent is effectively infinite,

C(n)\simeq A e^{-\widehat m n}, \qquad \widehat m_{\mathrm{eff}}(n) =\log\frac{C(n)}{C(n+1)}\simeq\widehat m.

Here $n$ is time in lattice steps. With periodic temporal extent $N_t$ , an appropriate leading model is instead

C(n)\simeq A\left(e^{-\widehat m n} +e^{-\widehat m(N_t-n)}\right).

Use a fit range where excited-state contamination is sufficiently controlled and the data still have useful precision. Compare neighboring fit ranges and richer spectral models where the data support them.

A plateau is an informative diagnostic, but it is not an independent theorem. The correlator lab lets you produce a misleading plateau from known synthetic data.

Use a matrix of probes

With a basis of operators, form $C_{ij}(t)$ and solve a generalized eigenvalue problem. The method separates combinations with better overlap onto different low-lying states. Its finite-basis and finite-time corrections must remain part of the analysis.

The Blossier et al. paper gives an analytical treatment of these corrections. The Athenodorou–Teper calculations provide an extensive example of spectrum determination across gauge groups and lattice symmetry channels.

An error budget has several layers

Source	Question to ask
Sampling	Were autocorrelations and ensemble statistics controlled?
Operator basis	Could a poor overlap conceal or distort a low state?
Fit procedure	Are results stable under justified changes of range and model?
Finite volume	Do several physical volumes give compatible results?
Cutoff	Is a continuum extrapolation supported by sufficiently fine lattices?
Scale setting	Which reference quantity defines the physical units?
State identification	How are lattice representations connected to continuum quantum numbers?

There is no universal checklist that automatically validates every calculation. These questions locate the assumptions that determine what a reported number means.

Compare dimensionless results

Suppose a calculation measures a lattice mass $\widehat m=am$ and a lattice string tension $\widehat\sigma=a^2\sigma$ . Their ratio

\frac{\widehat m}{\sqrt{\widehat\sigma}} =\frac{m}{\sqrt{\sigma}}

is dimensionless and does not require an external GeV calibration. Its continuum limit still needs extrapolation and uncertainty estimates.

Statistical errors in numerator and denominator can be correlated when both come from the same ensembles. Propagating them as if they were independent can misstate the ratio’s error.

Precise spectroscopy strongly constrains our physical picture. It does not by itself construct all continuum correlation functions or prove that no unobserved channel has arbitrarily low energy. The evidence belongs in the Research Atlas with those scope distinctions visible.

Read an actual source table

Athenodorou and Teper, arXiv:2106.00364v3, Table 34 (printed p. 83), quotes the continuum-extrapolated pure $SU(3)$ scalar ground-state ratio

\frac{m_{0^{++}}}{\sqrt{\sigma}}=3.405(21).

The parentheses mean a quoted uncertainty of $0.021$ in this dimensionless ratio. Table 24 (printed p. 74) lists the corresponding $A_1^{++}$ lattice-symmetry channel; §5.5, Eq. (27), describes continuum extrapolation. This is a numerical estimate, not an exact inequality. The error discussion in §2.3 and finite-volume/ditorelon analysis in §5.2 remain essential context; the parentheses should not be read as an independently verified bound on every systematic effect.

The relative quoted uncertainty is $0.021/3.405\simeq0.00617$ , about $0.62\%$ . No lattice spacing appears in that calculation because both numerator and denominator carry one power of energy. Converting to GeV would require a chosen value of $\sqrt{\sigma}$ ; pure Yang–Mills has no experimental scale calibration by itself. Comparing this scalar value with an excited state also requires their channels and any error correlations. It cannot bound all other physical sectors from below.

Source-reading exercise. What extra assumptions enter if someone multiplies this ratio by an experimental QCD string tension and labels the answer a measured glueball mass?

Solution. They introduced a scale-setting convention and an identification between pure gauge theory and a theory with dynamical quarks. Neither follows from Table 34. Keep the dimensionless pure- $SU(3)$ result separate from that phenomenological comparison.

Exercises

1. Convert a dimensionless ratio

Suppose $\widehat m=0.45$ and $\widehat\sigma=0.025$ . Find $m/\sqrt{\sigma}$ .

Hint. The spacing cancels.

Solution. The ratio is $0.45/\sqrt{0.025}\simeq2.85$ . These are illustrative numbers, not a quoted Yang–Mills measurement.

2. An exact plateau with a wrong conclusion

A perfect correlator in one channel has a plateau at mass $m_0$ . Does that prove every physical state has energy at least $m_0$ ?

Hint. Recall the oscillator’s even operator.

Solution. No. The operator sees only the states with its quantum numbers and nonzero overlap. The full physical gap requires broader spectral control.

Continue at your own depth

Glueballs and quantum numbers develops this lesson without adding a prerequisite to the next lesson.

Read further

Previous: The continuum limit · Course overview · Next: Reflection positivity