Non-perturbative quark mass renormalization in two-flavor QCD

The thing worth taking from this paper is Eq. (3.13): M/m(μ) = 1.297(16) in the SF scheme, continuum extrapolated, flavor- and regularization-independent. That is a reusable number — any lattice action can plug it in and only redo ZA·ZP. The supporting work is careful: six couplings spanning roughly 100 GeV down to 1 GeV, three lattice resolutions (6, 8, 12), continuum fits to a constant with the coarsest-lattice difference carried as a systematic (Table 1, Fig. 2), and a recursion that is flat in k (Table 2). ZA and cA are non-perturbative from their own earlier papers. The comparison with perturbation theory in Fig. 3 is honest — the data do not discriminate, and they say so rather than overselling the agreement. Table 6 also shows the conversion to MS-bar is well behaved, so the perturbative step at 2 GeV is not where the uncertainty lives. The strange mass is the soft part, and the authors know it. Section 4.2 declines a continuum extrapolation of Mref and quotes the β = 5.4 value with the β = 5.2 difference as the systematic: 72(3)(13) MeV. Your stress-test note lands on something real. Table 5 uses L/a = 16, 16, 24 at β = 5.20, 5.29, 5.40, and with a ≈ 0.092, 0.079, 0.071 fm that gives L ≈ 1.47, 1.26, 1.70 fm. The text's claim that at the higher β values they choose approximately matched or larger physical volumes is not quite true for β = 5.29, which sits below the L ≳ 1.5 fm threshold the paper itself sets. So the ±13 MeV band is a mix of cutoff and volume, and the monotone rise 58.7 → 63.5 → 72.0 MeV cannot be read as pure a². I would not push the alternative either, though. Three points over a 30% spacing range with a confounded volume do not support a linear a² extrapolation to ~85 MeV; that is just a different guess dressed as an extrapolation. Taking the finest lattice and inflating the error by the coarsest difference is defensible and clearly labeled conservative. The chiral perturbation theory bridge in Section 4.3 (Ms = 48/25 Mref, using Ms/M̂ = 24.4(1.5)) is an assumption, stated as one, with the Nf = 0 agreement as its only support. Bottom line: the running is a real, citable result; the 97(22) MeV strange mass is a placeholder that will move once they have finer lattices. Worth a referee report, mainly to press on the volume matching at β = 5.29 and to ask for the Mref values with volumes quoted explicitly.

Section 4.2 states the principle explicitly: 'The cutoff effects in mref also depend on the volume. In fact, at a lattice spacing of 0.1 fm this dependence is rather strong [40]. Thus one has to define the PCAC mass for a fixed volume in order to ensure that cutoff effects disappear smoothly as O(a²).' The three β values used to build Mref = 72(3)(13) MeV do not satisfy this. Table 5 uses L/a = 16, 16, 24 at β = 5.20, 5.29, 5.40, and the conclusions give a ≈ 0.092…0.071 fm over that range. With a decreasing in β, the physical volumes are L ≈ 1.47, 1.3, 1.7 fm — the β = 5.29 volume is smaller than β = 5.20's, contradicting the text's claim that 'at the higher β values we choose approximately matched (or larger) physical volumes.' The volume check was performed only at β = 5.20 (against JLQCD 20³×48). Consequently the systematic band, taken as the β = 5.4 minus β = 5.2 difference, is not a clean cutoff effect: it is confounded with the volume mismatch, and the intermediate β = 5.29 point carries an uncontrolled volume contamination. The three Mref values (58.7, 63.5, 72.0 MeV) rise monotonically and are roughly linear in a²; a linear a² extrapolation lands near ~85 MeV, at the top edge of the quoted 72±13 band. So the central value may be biased low and the error band may not represent the true continuum uncertainty. This is the same section the reader flagged as a load-bearing premise, sharpened to a specific, text-grounded mechanism.

The reader's weakest_assumption names two premises: the χPT bridge (Section 4.3) and the Mref continuum estimate (Section 4.2). I agree that the running result itself is robust — six couplings, three resolutions, continuum fits with stated systematics, and a recursion stable across k and loop order (Table 2) — and that the soft spot is the physical strange-mass chain. My concern is squarely the second premise the reader identified, sharpened to a concrete mechanism: the paper's own sentence in Section 4.2 requires a fixed volume for clean O(a²) cutoff behavior, yet the three β values used have L ≈ 1.47, 1.3, 1.7 fm, with β = 5.29 smaller than β = 5.20. The volume check was done only at β = 5.20. This makes the quoted ±13 MeV band a mixture of volume and cutoff effects rather than a clean cutoff systematic, and the monotonic, roughly a²-linear rise of Mref (58.7 → 63.5 → 72.0 MeV) suggests the continuum value may sit near the top of the band (~85 MeV). This does not overturn the paper: the running factor is the reusable, scheme-independent result and stands; the strange mass is already presented as a conservative estimate with a dominant stated systematic. The reader's CONDITIONAL verdict is appropriate, and my sharpening does not move it. The χPT bridge (Ms = 48/25 Mref, using published Ms/Mhat = 24.4(1.5) from Ref. [19]) is a cited, acknowledged assumption and is not a load-bearing defect on its own. The perturbative step in Eq. (3.12) (2/3-loop τ, β for the RGI conversion) is controlled by the stability shown in Table 2 and is not a serious concern. So: no change to the verdict, agreement with the reader, with the volume/cutoff entanglement as the concrete substance behind the reader's general point about the Mref estimate.

The running factor Eq. 3.13 is a genuine non-perturbative continuum quantity, scheme- and flavor-independent, and reusable by any lattice action; this is a real advance over perturbative renormalization and is the paper's central and well-supported result. The paper is careful: ZA, cA, and ZP are all non-perturbative; the continuum extrapolations carry an explicit systematic from the coarsest lattice; the recursion is stable (Table 2); and the assumptions entering the strange-mass conversion are stated explicitly. The main weaknesses are in the secondary Mref/ms application, where the continuum estimate is acknowledged to be uncontrolled and where the error budget has some gaps.

Quark masses are not directly observable: what a lattice or an experiment quotes is a mass at some chosen scale, and that number slides with the scale according to a running equation. This paper sets out to compute that running, and the conversion to the scale-independent (renormalization group invariant) mass, entirely on the lattice, with no perturbative expansion anywhere in the chain. Working in two-flavor QCD with massless $O(a)$ improved Wilson quarks in the Schrödinger Functional scheme, the authors determine the step scaling function $sigma_P(u)$ at six values of the renormalized coupling, extrapolate each to the continuum, and chain them together to obtain $M/m(mu) = 1.297(16)$ at $mu = 1/L_{max}$. Because the continuum limit is taken and the scheme is mass-independent, that factor is claimed to be universal, reusable with any lattice action once the cheaper matching factor $Z_P$ is redone for that action. The payoff is a non-perturbative strange quark mass, $m_s^{MSbar}(2~{GeV}) = 97(22)$ MeV, whose error is dominated by the continuum estimate of a reference mass rather than by statistics.

What carries the argument. The load-bearing object is the step scaling function $sigma_P(u) = lim_{a to 0} Z_P(g_0, 2L/a)/Z_P(g_0, L/a)$ at fixed renormalized coupling $u = bar{g}^2(L)$, which is exactly the ratio $m(mu)/m(mu/2)$ for $mu = 1/L$. It converts the problem of running over two orders of magnitude in energy into a sequence of small, individually continuum-extrapolated factors, iterated through a joint recursion on coupling and mass from $u_0 = 4.61$ up to $u_8 = 1.017$. The second piece of machinery is the splitting $Z_M = [M/m(mu)] times Z_A Z_P$, which isolates the universal, regularization-independent factor from the action-specific matching, so that only the second factor must be recomputed for a different discretization.

Core claim. The paper reports a fully non-perturbative determination of the quark mass running in two-flavor QCD, expressed as the ratio of the renormalization group invariant mass to the running mass at a fixed low-energy scale, $M/m(mu) = 1.297(16)$ at $mu = 1/L_{max}$ (Eq. 3.13). The chain is built from the step scaling function $sigma_P(u)$, computed at six values of the renormalized coupling on lattices of size $L/a = 6, 8, 12$, extrapolated to the continuum by fitting the two finest lattices to a constant and adding the difference to the $L/a = 8$ result as a systematic band. The authors note that the data show no significant lattice-spacing dependence at the achieved accuracy, and that the non-perturbative $sigma_P(u)$ agrees with the two-loop $tau$-function / three-loop $beta$-function curve within errors. The second half of the paper computes the action-dependent matching factor $Z_M = Z_A Z_P$ at three bare couplings ($beta = 5.20, 5.29, 5.40$), then tunes a hopping parameter $kappa_{ref}$ so that the pseudoscalar mass satisfies $(r_0 m_{PS})^2 = 1.5736$, the value corresponding to the physical kaon mass at $r_0 = 0.5$ fm. This gives a reference RGI mass $M_{ref} = 72(3)(13)$ MeV, where the second error is the difference between the $beta = 5.4$ value and the $beta = 5.2$ value, adopted because the authors explicitly decline to perform a continuum extrapolation over so short a range of lattice spacings. Combining with the lowest-order chiral perturbation theory relation $M_s = 48/25 M_{ref}$ yields $M_s = 138(5)(26)$ MeV and, after perturbative conversion, $m_s^{MSbar}(2~{GeV}) = 97(22)$ MeV.

The premise it rests on. The step from a two-flavor lattice number to a physical strange mass assumes that the reference mass $M_{ref}$ computed with two degenerate flavors also holds in a three-flavor degenerate theory, and that lowest-order chiral perturbation theory, $M_s = 48/25 M_{ref}$, is accurate enough to bridge the two.

Any lattice collaboration can reuse $M/m(mu) = 1.297(16)$ as a continuum, flavor- and regularization-independent factor and recompute only the action-dependent $Z_A Z_P$ piece, removing perturbative renormalization as one source of the spread among published strange quark masses.

The quoted $m_s^{MSbar}(2~{GeV}) = 97(22)$ MeV sits in the upper part of the previously published 68 to 132 MeV range, and since it agrees within errors with the quenched determination, the data do not support a strong dependence of the strange mass on the number of dynamical flavors.

The conversion table shows that going from two-loop to four-loop truncation of the MS-bar anomalous dimension shifts $m_s$ by only a few percent at 2 GeV, so the residual perturbative step at the end of the chain is small compared with the lattice systematic.

The same analysis extracts $B_{RGI} = 1.70(38)$ GeV, a low-energy constant of the chiral Lagrangian, from lattice QCD rather than from phenomenology.

If the running factor really is universal, the dominant error in future strange-mass quotes should migrate from renormalization to scale setting, which would make replacing $r_0$ by a decay-constant-based scale the highest-leverage single improvement.

The visible split between the two choices of $Z_A$ at the coarsest lattice spacing suggests the $a^2$ behavior of $M_{ref}$ is the real bottleneck; one additional lattice spacing near 0.06 to 0.07 fm would show whether 72 MeV is a continuum value or effectively an upper bound.

The close agreement of non-perturbative and perturbative running down to roughly 1 GeV may be a property of this particular Schrödinger Functional scheme; repeating the step-scaling study in a different mass-independent scheme would test whether that agreement travels.

A direct three-flavor degenerate lattice computation of $M_{ref}$ would turn the chiral perturbation theory bridge from an assumption into a measurement.

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