BESIII recently made the first observation of the decay $D^+ \to \eta' \mu^+ \nu_\mu$, in which a charmed meson turns into the η′ meson, a lepton, and an undetected neutrino. The collaboration reported the first measurement of the decay's normalization, $f_+^{\eta'}(0)|V_{cd}| = (5.92 \pm 0.56 \pm 0.13)\times 10^{-2}$. They also extracted an η–η′ mixing angle of $(39.8 \pm 0.8 \pm 0.3)^\circ$ Observation of $D^+\toη^\primeμ^+ν_μ$…. The only theory they could compare against was light-cone sum rules, which are model-dependent. No first-principles calculation existed for the experiment to test. That gap is not an oversight. The η′ is the one light meson whose mass and structure come largely from the topology of the gluon field (its "winding," the topological charge). Computing it on a lattice means fighting the two problems that dominate lattice gauge theory: noisy short-distance fluctuations, and topological charge that stops changing in the simulation as the grid is refined.
This note proposes to compute the $D^+\to\eta'$ and $D^+\to\eta$ form factors on a spacetime grid, across several grid spacings, and confront them with the BESIII measurement. The plan uses two tools from lattice QCD, the discretized theory of quarks and gluons. The first is gauge field smoothing: gradient flow, a diffusion-like smoothing of the gluon field. The second is continuum extrapolation: fitting results at several grid spacings and taking the spacing to zero. The experimental side, double-tag reconstruction plus missing-energy reconstruction of the neutrino, is the evaluation setting. It determines which observables are clean enough to compare and in which momentum-transfer range. The physics payoff is a direct test of the quark-flavor mixing model behind BESIII's mixing angle, which ignores any gluonic component of the η′.
Core Claim
Gradient flow gives two things at once in this problem. First, it gives a clean valence-quark setup. Smoothing the gluon background lets domain-wall quarks, a discretization that preserves chiral symmetry almost exactly, run cheaply on existing staggered-quark ensembles. Second, it gives a smooth, renormalizable definition of the topological charge density, $q_t(x)$, at flow time $t$. We propose to use $q_t(x)$ as an extra source operator for the flavor-singlet part of the η′ in three-point functions. The three-point function here is the lattice correlator that encodes the $D\to\eta'$ transition.
Our claim has two parts. Adding this gluonic operator to the usual quark-bilinear operators makes the noisy "disconnected" quark-loop contributions tractable. Taking flow time and grid spacing to zero together then yields a continuum $f_+^{\eta'}(q^2)$ with total uncertainty below the experiment's current ~10%. Here $q^2$ is the squared momentum carried off by the lepton and neutrino.
The quantities we target are chosen so the experimental side is as clean as possible. The ratio of η′ to η rates and the normalized shape $f_+(q^2)/f_+(0)$ do not depend on the Cabibbo–Kobayashi–Maskawa matrix element $|V_{cd}|$, which sets the overall strength of the charm-to-down transition. In the double-tag measurement, the shared tags and shared neutrino reconstruction also cancel many systematics in these quantities.
Research Program
The lattice ensembles are the MILC $N_f=2+1+1$ ensembles with highly improved staggered sea quarks. They come with four grid spacings (≈0.15, 0.12, 0.09, 0.06 fm) and pion masses down to near physical. The mixed-action setup follows Möbius domain-wall fermions on gradient-flowed dynamical HISQ ensembles: domain-wall valence quarks run on the gradient-flowed staggered sea. That work showed that flowing the background suppresses residual chiral breaking to a small fraction of the light-quark mass. We reuse this pipeline as-is. Its precision is demonstrated by $F_K/F_\pi$ from Möbius domain-wall fermions solved on gradient-flowed HISQ ensembles, which reached $F_K/F_\pi = 1.1964(44)$ through a continuum and chiral fit with Bayesian model averaging. Those papers compute decay constants of light mesons that have no flavor-singlet component. We extend the same machinery to a heavy-to-singlet transition. That extension is where the difficulty lies.
The new ingredient is a basis of interpolating operators for the η/η′ system. It contains light and strange quark bilinears plus the flowed topological density $q_t$. The bilinears need stochastically estimated disconnected loops. We solve the resulting matrix eigenvalue problem to isolate η and η′ and then build $D\to\eta^{(\prime)}$ three-point functions.
Flow time is a second extrapolation variable. The flow smooths the field over a radius $\sqrt{8t}$. That radius must stay below hadronic sizes, otherwise smoothing destroys real short-distance physics, a known failure mode of over-smearing. We therefore work at several $t$ and extrapolate jointly. The scale is set with the flow scale $w_0$. Its universality across actions was shown in High-precision scale setting in lattice QCD, which also flags the 2 fm minimum box size. All MILC volumes used must satisfy that bound.
Topological freezing at 0.06 fm is a constraint we cannot hide from. The same Borsányi et al. paper reports strong Monte Carlo autocorrelations in flow observables. The flow-model samplers of Sampling QCD field configurations with gauge-equivariant flow models are not a usable fix: they reach only $4^4$ volumes with effective sample size ≈0.1. We will therefore measure the integrated autocorrelation of $Q$ on every ensemble. On ensembles where tunneling between topological sectors is inadequate, we apply fixed-topology finite-volume corrections and carry them as a separate systematic.
The continuum extrapolation follows the Symanzik form, which expands lattice errors in powers of the grid spacing. It includes terms in $a^2$ and in $(a\,|\vec p_{\eta'}|)^2$, the grid spacing times the recoil momentum of the η′. The recoil terms matter most at low $q^2$, where the η′ recoils hardest. We combine $z$-expansion fits, a standard series parameterization of the $q^2$ dependence, by Bayesian model averaging, as Miller et al. did for $F_K/F_\pi$.
Last, we compute the "gluonic fraction," the overlap of the physical η′ with $q_t$. It is a diagnostic, not a headline number, because any split between gluonic and quark content depends on the flow-time scheme. Its use is to tell whether BESIII's quark-flavor mixing angle is a meaningful parameter or an effective one.
Baselines
The naive reference point is the comparison BESIII already made, against light-cone sum rules. It has no first-principles control of uncertainty. The file does not list the sum-rule values, so we report no number for them. Beating this reference means producing a prediction with a quantified, continuum-extrapolated error budget.
The experimental target is the BESIII measurement, $f_+^{\eta'}(0)|V_{cd}| = (5.92 \pm 0.56_{\rm stat} \pm 0.13_{\rm syst})\times 10^{-2}$ Observation of $D^+\toη^\primeμ^+ν_μ$…. That is roughly a 10% uncertainty. The lattice result counts as decisive if its total uncertainty on $f_+^{\eta'}(0)$ is at most about 5%. At that level the combined comparison is dominated by experiment and stays informative as BESIII statistics grow.
The strongest existing lattice method is the unmodified mixed-action pipeline of Miller et al. and Berkowitz et al. That means quark-bilinear interpolators only, disconnected loops estimated stochastically, and the same ensembles. We found no reported result on Lacuna for a lattice $D^+\to\eta'$ form factor with this or any other method. To keep the comparison fair, we give it the same measurement budget in Dirac-operator solves per configuration. The proposal beats it if adding $q_t$ cuts the statistical error on the η′ three-point function at the same cost, by a factor we pre-register as at least 1.5. It must also leave the continuum value unchanged within errors.
Two ablations isolate the key ingredients. The first drops the flowed topological operator but keeps everything else, which tests whether the gluonic operator is what makes the singlet tractable. The second fits only the three coarser spacings, which tests whether the frozen-topology finest ensemble is driving the continuum value.
Why not just…?
Why not just keep using light-cone sum rules? They are what BESIII used, and they carry model uncertainties that cannot be reduced systematically. Their treatment of the η′ also usually inherits the very mixing model we want to test. Using them to interpret the mixing angle is circular. A lattice number is the only way out.
Why not just compute on one fine ensemble with standard smearing? A single spacing cannot bound discretization errors. The continuum-extrapolation literature in the file treats at least three spacings as the minimum. Recoil-momentum artifacts are largest exactly where the η′ is hardest. This is a real threat to cost only, not to correctness. The coarse-three-spacing ablation will show how much the finest ensemble actually matters.
Why not just compute the disconnected loops harder and skip the gluonic operator? In full QCD, the quark loops already contain the anomaly physics, so $q_t$ is in principle redundant. This alternative might work as well as our method. If the bilinear-only baseline matches our error at equal cost, then the gluonic operator is only a diagnostic, and we will report that. The equal-budget baseline comparison decides it.
Why not just extract the mixing angle phenomenologically from other processes? Those extractions assume the same two-angle quark-flavor picture. Agreement among them does not test that picture. Our direct overlap calculation does.
Proposed Experiments
The first check is the pre-registered noise comparison: the η′ correlator error with and without $q_t$, on the 0.12 fm physical-mass ensemble, at equal solve budget.
The second check is the continuum $f_+^{\eta'}(0)$ and the shape $f_+^{\eta'}(q^2)/f_+^{\eta'}(0)$. The shape is compared to BESIII's binned partial rates, using only the $q^2$ range where the neutrino reconstruction gives clean signal peaking at $U_{\rm miss}\approx 0$. Here $U_{\rm miss}$ is missing energy minus missing momentum, which is zero for a single massless neutrino. Resolution from beam spread and initial-state radiation smears the edges of phase space, and those bins are excluded or flagged.
The third check is the CKM-free ratio $\Gamma(\eta')/\Gamma(\eta)$, to be confronted with the BESIII η′ result and an η measurement from the same 20.3 fb⁻¹ dataset. The file reports no η value, so that comparison waits on the published number.
The fourth check is the gluonic overlap, extrapolated in flow time. A value consistent with zero supports BESIII's 39.8° as a physical quark mixing angle. A clearly nonzero value means that angle is an effective parameter absorbing gluonic content. That second outcome would also say the same caution applies when interpreting charmed-baryon and rare-decay double-tag measurements, such as Evidence for the semileptonic decays $Λ_c^{+} \to Σ^{\pm} π^{\mp} e^+ ν_e$, with simple flavor models.
References
- Observation of $D^+\toη^\primeμ^+ν_μ$ and First Study of $D^+\to η^\prime \ell^+ν_\ell$ Decay Dynamics
- Möbius domain-wall fermions on gradient-flowed dynamical HISQ ensembles
- <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:msub><mml:mi>F</mml:mi><mml:mi>K</mml:mi></mml:msub><mml:mo stretchy="false">/</mml:mo><mml:msub><mml:mi>F</mml:mi><mml:mi>π</mml:mi></mml:msub></mml:math> from Möbius domain-wall fermions solved on gradient-flowed HISQ ensembles
- High-precision scale setting in lattice QCD
- Sampling QCD field configurations with gauge-equivariant flow models
- Evidence for the semileptonic decays $Λ_c^{+} \to Σ^{\pm} π^{\mp} e^+ ν_e$