WorksPhysics and Astronomy
Cross-Moment Consistency and the Fifth Adler Coefficient in Hadronic Tau Decays
Abstract
The unknown coefficient controls the first omitted fifth-order term in fixed-order analyses of inclusive hadronic decays. We determine what the four known Adler coefficients can actually test. For any normalized polynomial weight, the fixed-order Adler-to-moment map is triangular with unit diagonal. Hence one common Adler-coefficient error reaches every moment with the same absolute size, and a multi-weight relative test reduces to one absolute tolerance. For three representative weights, a ten-percent test resolves only to about ; the closest contour-improved test is weaker by more than an order of magnitude. Moreover, any continuation that preserves the observed sign of the direct Padé/Shanks defects gives , so values near require additional structure. A calibration-and-falsification protocol applied to a simplified Beneke–Jamin-type pure-power ansatz gives the conditional estimate . At the physical order its central value coincides algebraically with the exact four-parameter fit; the protocol adds a falsifiable admissibility test and a method uncertainty, not new coefficient information. We also clarify that the often-quoted Beneke–Jamin value is a linear extrapolation of total-width FOPT te
Introduction
Inclusive hadronic decay is a low-energy precision laboratory for QCD [1, 2]. The inclusive width and weighted integrals of the measured invariant-mass spectrum can be related, by analyticity, to contour integrals of current correlators [3, 4]. On most of the contour the relevant scale is of order , so perturbative QCD can be combined with the operator product expansion (OPE) [5] to extract the strong coupling. The perturbative contribution is known through for massless three-flavor QCD [6], and this calculation has underpinned a series of determinations of [7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]. Power corrections in the OPE are numerically suppressed for the total width [1, 4, 7, 9, 11], whereas the treatment of quark–hadron duality violations near the timelike axis remains an active issue [14, 15, 16, 20, 21, 22]. The next Adler coefficient, conventionally denoted , has not been computed directly.
The practical significance of is limited but real. It controls the first omitted coefficient in a fifth-order fixed-order expansion. It does not, by itself, decide the all-order difference between fixed-order perturbation theory (FOPT) and contour-improved perturbation theory (CIPT) [23, 24], which has long been the dominant theoretical uncertainty in extractions [9, 25, 26, 27, 28, 29]. That difference originates in the asymptotic nature of the perturbative series, whose large-order behaviour is governed by renormalons [30]; it depends on the running of the coupling along the complex contour and on the singularity structure of the Borel transform [31, 32, 33, 34]. Subtracting the gluon-condensate renormalon reduces it substantially in favor of FOPT [35, 36, 37, 38].
In the normalization used throughout this article, the reduced massless Adler function at a Euclidean scale is
and the four known coefficients for in the scheme [39] are [6]
The size of the unknown higher orders [2, 40] has been estimated with scheme-invariant (effective-charge and minimal-sensitivity) methods [41], explicit renormalon models of the Adler function [9, 29, 42, 43, 44, 45, 46], Padé and Borel–Padé approximants [47], conformal mappings of the Borel plane [48], and nonlinear sequence transformations [49]. More broadly, the perturbative series has been reorganized by renormalon-chain resummation [50, 51, 52, 53, 54, 55, 56], modified Borel and ultraviolet-renormalon studies [57, 58], renormalization-group-summed and conformally mapped expansions [59, 60, 61, 62, 63, 64], series acceleration [65], and scheme or scale variations [66, 67, 68, 69].
Producing a number for is therefore not the difficulty; many methods do so. The question addressed here is what the four known coefficients can actually test. Our conclusions are:
Testing a common Adler forecast on several normalized FOPT moments is exactly one scalar test, because the Adler-to-moment map has unit diagonal (§ 4 and § 5).
At order five, the commonly used ten-percent, three-weight test permits an absolute error of about 24 in ; the corresponding CIPT test is weaker by more than an order of magnitude (§ 6).
A direct geometric Shanks/Padé continuation gives , and every continuation of the Adler sequence that preserves the sign of the observed forecast defects gives , above the usual phenomenological range (§ 7 and § 8).
A value near 280 therefore requires additional structure that reverses the defect sign. A simplified Beneke–Jamin (BJ)-type renormalon ansatz [9] can supply it, provided the distinction between calibration and independent validation is retained together with its model dependence (§ 9 and § 10).
§ 2 and § 3 set up the moments and explain the choice of weights. § 12 explains what the companion long paper [70] adds beyond this self-contained account.
From the measured spectrum to an Adler moment
What the experiment measures
For a vector or axial-vector quark current , we define
Lorentz covariance separates transverse () and longitudinal () scalar correlators [4]. Their discontinuities define spectral functions,
Experimentally, exclusive hadronic final states are reconstructed, detector efficiencies and backgrounds are corrected, and migration between true and reconstructed invariant-mass bins is unfolded. The useful data product is therefore a vector of spectral-bin contents together with its full covariance matrix, not only a plotted curve [17, 19, 71, 72]. A weighted moment can be formed as
or, for bins of width ,
If , linear error propagation gives
We notice that different moments are generally strongly correlated because they use the same events, the same unfolding response, and common normalization and systematic corrections. Thus many polynomial weights do not automatically provide the same number of independent experimental measurements.
The physical transverse kinematic weight in the total hadronic width is
It is fixed by the spin structure and phase space of the decay [4], not selected to optimize perturbation theory. Other weights define additional moments of the differential spectrum; they are not alternative expressions for the unmodified total width.
Finite-energy sum rule and integration by parts
Analyticity of away from its physical cut implies the finite-energy sum rule (FESR) [2, 4]
with a counterclockwise contour. Polynomial weights are convenient because they are analytic within the contour and project OPE terms transparently. An integration by parts replaces the correlator by the Adler function. If and the spectral weight is , we define the corresponding Adler weight
and normalize the weights so that . Applying Eq. (10) to Eq. (8) gives
The distinction matters: weights the measured spectrum, whereas multiplies the Adler function on the complex contour.
The perturbative Adler moment is
For the total nonstrange width, this quantity enters schematically as
where and are electroweak corrections [73, 74, 75], is the massless perturbative term, and denotes mass and power corrections of OPE dimension [4, 76].
Why these three weights are complementary
We use the normalized Adler weights
The label “GC” indicates gluon-condensate sensitivity. The label “P2” is a mnemonic for a twice-pinched polynomial: has a double zero at . Endpoint zeros (“pinching”) suppress the region near the timelike point , where the OPE is least reliable and duality violations can be most visible [14, 20, 21, 22]. More pinching can therefore reduce one source of uncertainty, but a higher-degree polynomial also projects higher-dimensional OPE terms. Weight design is a trade-off, not a rule that more zeros are always better.
The three weights also probe the leading infrared renormalon differently [9, 29]. If , the Borel transform of the moment contains the kinematic kernel
To evaluate it at a positive integer , we set . Then
Multiplication gives the simple selection rule
If , the kernel instead has a zero proportional to . Therefore a simple Adler Borel pole at is cancelled, while a branch point is weakened by one power.
At the gluon-condensate location we have,
The standard total- weight suppresses the leading infrared renormalon, associated with the gluon condensate [30], and is therefore an unusually forgiving extrapolation target [9, 29]. and retain the same singularity with different signs and normalizations. This is the physical reason to compare them: the set contains one renormalon-suppressed moment and two renormalon-sensitive moments. The comparison is informative about cancellations and relative sensitivity, even though, as the next two sections show, it does not turn a common coefficient forecast into three independent constraints.
The triangular FOPT map and the transfer identity
In FOPT the running coupling on the contour is re-expanded in powers of . Renormalization-group invariance fixes every logarithmic term in terms of lower-order Adler coefficients and the beta function [9, 77], so that
Two properties of this expansion are used below. It is triangular, , because starts at ; and , because that leading term carries no contour logarithm. Inserting Eq. (21) in Eq. (12) and integrating term by term with the logarithmic contour moments
gives the FOPT series
The off-diagonal entries , , are combinations of beta-function coefficients and that can be evaluated in closed form [9, 78]. The diagonal follows from and Cauchy’s formula,
since every weight in Eqs. (14)–(16) is normalized to . Separating the diagonal term,
where the RG-generated part is fixed entirely by the lower Adler coefficients. Let a hat denote a forecast. Keep at their exact values and forecast only . Transporting this common forecast through Eq. (23) gives , and subtracting Eq. (25),
This is the transfer identity. It is exact under its hypothesis: one common Adler forecast is transported through the same FOPT map with all lower inputs unchanged. It is not a relation between forecasts obtained independently from each moment series, and it concerns absolute errors. Relative errors differ because the denominators differ; at ,
so the small coefficient magnifies any common discrepancy.
Provenance of the Beneke–Jamin values 52 and 283
Beneke and Jamin (BJ) [9] observed that the convergence rate of the FOPT total-width series, with terms at their input , is nearly uniform. The term ratios are
each close to one half, with a slight order dependence that BJ fitted linearly. Their estimate is therefore a linear extrapolation of the term ratios, not a constant-ratio (geometric) continuation. With and the line passes through two points, . With , , the least-squares line has slope and passes through the mean at , so , consistent with the value quoted in Ref. [9]. Since every is proportional to , so is the fitted line, and cancels in . Writing ,
with , and . Subtracting the RG-generated parts and [Eqs. (25) and (34)] gives
the postdiction of footnote 2 and the estimate of Eq. (3.10) of Ref. [9]. Both numbers extrapolate the FOPT total-width coefficients in moment space and are converted to Adler coefficients only afterwards; neither is a direct continuation of the common Adler sequence. In particular, 283 is an input to the main five-parameter BJ renormalon model, not an output of that fit. When BJ instead drop this input, set the linear polynomial term to zero, and fit a four-parameter model to , that separate construction predicts about 280 (our reconstruction gives 280.12).
The postdiction illustrates the transfer identity. Using the unrounded reconstruction and the exact gives the same absolute error 3.0081 in every moment, by Eq. (26), but the relative errors are
for , and . A statement such as “the forecast works for but fails for ” therefore describes the relative-error convention, not three different underlying coefficient errors. Judged by the three-weight criterion of § 5, the estimator that produces 283 fails at the order where it can be checked: exceeds the tolerance 2.151 by a factor 1.399. Using the published rounded postdiction 52 would instead give 2.9243 and 1.36; mixing those rounded quantities with the unrounded formula is the source of the discrepancy in the earlier version.
Resolving power of a multi-weight test
Let be a set of weights normalized to and impose a relative tolerance on every moment,
By the transfer identity (26) all left-hand sides are equal, so the intersection of these conditions is exactly
A multi-weight FOPT test is therefore a single absolute tolerance on one number. An additional weight tightens it only if its moment coefficient is smaller in magnitude; it cannot probe a new direction, because the forecast-error space is one-dimensional. The collapse relies on holding the lower inputs fixed: if several Adler coefficients or beta-function inputs vary together, the error vector can have rank larger than one, and sufficiently different weights then do provide independent constraints. Likewise, the experimental covariance of the moments, Eq. (7), measures how much independent information the data contain, which is a different question from how one missing perturbative coefficient propagates; a phenomenological fit should retain both.
Table 1 lists the coefficients for the three weights of § 3. At every order the smallest one, and hence the tolerance, belongs to . By the unit diagonal, the order-five coefficients are exactly affine in the unknown ,
| coefficient tolerance | ||||
| 3 | 26.366 | 34.484 | 14.523 | |
| 4 | 127.079 | 214.140 | 21.509 | |
| 5 | 584.78 | 1433.93 |
Table 1. FOPT moment coefficients and the coefficient tolerance of Eq. (33) at . The row uses . The minimum is set by at every order.
so the fifth-order row is shown at the representative value , the central value of the Borel–Padé estimate of Ref. [47]; another choice shifts all three entries rigidly.
remains the most restrictive weight at fifth order whenever , i.e. for , which covers the phenomenological range. There Eq. (33) reads
ranging from 18 to 29 over the Borel–Padé range . The tolerance shrinks to zero as approaches 512.6, where vanishes and a relative criterion for that moment ceases to be meaningful. Two candidate values of closer than this tolerance cannot be distinguished, however many weights display the same common error. This is a statement about the validation procedure, not a claim that the moments carry no distinct physics.
To set the scale of the fifth-order term, we take [79], as in Ref. [49], and evolve it with RunDec [80, 81] at five loops with decoupling of the bottom and charm quarks at their masses, obtaining , so that and . This input is used only here and in § 6; the reconstruction of Ref. [9] in § 4.1 uses that paper’s own value . Because , a change shifts by the same amount for every weight. For , the total uncertainty of the forecast in Eq. (66), the fifth-order contribution to each FOPT moment changes by
about of (evaluated through fifth order at ). The coefficient thus matters for precision theory, but its uncertainty is only one component of the complete experimental and theoretical error budget.
What changes in contour-improved perturbation theory
FOPT re-expands the running coupling in before performing the contour integral. CIPT [23, 24] instead keeps the running coupling on the contour and truncates the Adler series before integration:
This map is diagonal in the Adler coefficients, but its entries are -dependent contour integrals rather than one. A forecast error at the last retained order therefore enters as
and the simultaneous relative test again collapses to a single inequality,
The unit diagonal of Eq. (24) is lost, but the logical collapse survives: one unknown coefficient multiplies one known factor for each weight.
Table 2 compares the two tests weight by weight. The FOPT entries are from Table 1; they involve only pure numbers and do not depend on . The CIPT entries are the right-hand side of Eq. (39) for each weight separately, obtained by integrating the five-loop renormalization-group equation on the contour at . The entries use in both schemes. The tolerance of the three-weight test is the smallest entry in each row (in bold).
| \multicolumn{3}{c}{FOPT} | \multicolumn{3}{c}{CIPT} | |||||
| 4 | ||||||
| 5 |
Table 2. Single-weight coefficient tolerances at . FOPT: . CIPT: at . The entries use . Bold entries are the three-weight tolerances.
The most restrictive weight differs between the schemes, for different reasons. In FOPT it is the weight with the smallest , which is because its RG-generated part is negative and partly cancels the Adler coefficient: and . In CIPT the partial sums differ only moderately between weights, for , , , whereas the last-order contour integrals differ much more, and . The tightest CIPT bound therefore comes from , which has the largest , and , the most restrictive weight in FOPT, becomes the least restrictive in CIPT. The comparison does not rely on mixing weights: for alone, FOPT is tighter than CIPT by factors of 3.6 at and 5.4 at .
CIPT is less selective because the last retained term is only a small fraction of the contour-improved partial sum. For it is 6.3% of and 3.6% of , so a ten-percent tolerance on the partial sum admits an excursion larger than the coefficient being tested; at it is 2.8 times . Over the CIPT tolerances change by only to . For the present validation question, FOPT is therefore the stronger test.
This comparison should not be confused with the all-order FOPT–CIPT separation. In modern Borel analyses [31, 32, 34], one may define an asymptotic separation as the difference between specified all-order Borel sums associated with the two perturbative organizations. Its value is prescription- and model-dependent, and it receives information from the whole Borel singularity structure [33, 35, 36]. Determining improves the finite-order input but does not set to zero and does not decide which organization best approximates a selected Borel sum.
Direct Shanks/Padé forecasts and balance
The local geometric model
We truncate the Adler series Eq. (1) at the last known order and write its partial sums as
The first Shanks transformation, equivalently Aitken’s process [82], acts on the last three partial sums as
where . It is exact when the remainder is a single geometric mode, and it completes the series with a geometric tail whose term ratio is the last observed one. Expanding in , , so the first unknown coefficient is forecast by extending the last observed coefficient ratio by one order:
The coupling cancels, as in the ratio fits of § 4.1.
The same formula follows from matching a one-pole rational function. Writing and clearing the denominator shows that is the Padé approximant of in the convention of Eq. (47) below: it reproduces all known coefficients, in accordance with the counting rule of Eq. (48). Its denominator generates for , whose first step is Eq. (42). This is why the estimate is called a direct Padé/Shanks forecast: the first Shanks element and the one-pole approximant encode the same local geometric assumption. For this is the approximant, one of the balanced choices of Eq. (49), and
The same algebra can instead be applied to each FOPT moment series (23), with and , giving the moment-space forecast . Converted into an Adler forecast through Eq. (25), it is in general weight-dependent, because the ratio extrapolation does not commute with the linear map. For at , and, by Eq. (34), . This is the constant-ratio counterpart of the BJ linear-ratio estimate 283 in Eq. (30). Both are moment-space forecasts and should not be conflated with the common-Adler value (43); the same holds for the Levin-transform estimate of Ref. [49], which transforms the FOPT series of . The separate four-parameter BJ Borel-model variant, fitted to only, forecasts 280.12 (Section 10). These numbers test different maps and are not contradictory.
The direct forecast (43) is not a theorem about QCD. Renormalon-dominated coefficients grow factorially [30], approximately as
so their ratio is not constant but grows linearly with . The direct forecast can still be a useful local diagnostic, but its geometric remainder model is too rigid to represent several competing infrared and ultraviolet renormalon modes.
We can divide coefficients by the factorial growth before applying the same idea. In the conventional Borel variable [9, 30] for which the leading infrared renormalons occur at positive integers, define
where for and . If , direct continuation gives
The large shift between 378.03 and 504.04 is evidence that four coefficients do not identify a unique asymptotic representation.
Balanced and unbalanced Padé approximants
Padé approximants have been applied to the series in Ref. [47]. Let be any of the series above, with known coefficients. After factoring the known zero at , approximate by
There are unknown parameters. Matching the known coefficients of therefore requires
To obtain a rational continuation with at least one denominator pole, one also requires . A common near-diagonal selection rule is
The first two conditions follow from parameter counting and the desire to continue beyond a polynomial. The last is a design criterion, not a theorem: it prevents either numerator or denominator from carrying almost all degrees of freedom. When , the balanced choices are [2/1] and [1/2].
A [3/0] approximant is the input cubic itself. Because it has no denominator and no higher Taylor coefficients, it supplies no nontrivial next-order forecast. A [0/3] approximant can forecast, but it is strongly unbalanced: its constant numerator forces all nontrivial structure into the denominator, making the result especially sensitive to low-order noise and spurious poles. Unbalanced approximants are not mathematically forbidden. They are valuable as stability diagnostics, but should not silently be assigned the same status as a near-diagonal family.
The defect-sign obstruction
We define the one-step direct-forecast defect
It measures how far the new coefficient lies from the direct forecast (42) built from the preceding two; the convention defines . The known Adler coefficients give
All three are positive and their magnitudes increase. A direct Padé/Shanks forecast sets the next defect to zero. More generally, call a map defect-sign-preserving if it returns when the supplied defects are positive and increasing. Such a map must satisfy
The three-weight validation at the previous order does not reveal this restriction. At , the ten-percent tolerance in Table 1 accepts
which is equivalent to
The entire accepted interval has positive defect. Hence the validation test can accept a sign-preserving continuation even though that continuation is structurally unable to reach a value around 280 at the next order. A value , for example, corresponds to
a sign reversal not encoded by the local geometric pattern.
This gives a sharp dichotomy:
Either , or every defect-sign-preserving continuation of the common Adler sequence misses , including those that pass the ten-percent withheld-order test at $N=4.
The statement is conditional and falsifiable. It does not prove that is large. It proves that obtaining a smaller number requires a map outside the sign-preserving class, hence additional declared structure.
Figure 1 gives the compact comparison needed here: it separates the three reference values near the phenomenological range from the direct common-Adler sign-preserving boundary. The complete method-by-method classification is deferred to the companion manuscript.
Figure 1. Compact numerical comparison. The blue circle is the BJ–CVRC central value 247.98 248; the light-blue interval is its structural closure window [229.8, 266.2], not the full adopted uncertainty 248 80. The diamond is the BJ moment-space linear-ratio estimate 282.823 283, which is an input to the main five-parameter BJ model, whereas the square is the separate four-coefficient BJ-model forecast 280.12. The gray band is the independent Borel–Padé range 277 51, and the red line is the direct common-Adler sign-preserving boundary .
An ensemble average requires special care. An average over Padé approximants, Borel normalizations, or renormalization schemes need not inherit a nonlinear inequality obeyed by each member. It can lie where no individual map lies. Therefore the Borel–Padé estimate 277 51 of Ref. [47] is not in conflict with Eq. (52); it is an ensemble result, not one defect-sign-preserving transformation of the four common-Adler inputs.
Renormalons, OPE ambiguities, and what resummation cannot remove
Suppose the Borel transform of the perturbative Adler function, equivalently the perturbative Wilson coefficient in the OPE, has a positive-axis singularity
Expanding the leading term about gives
With the Borel convention of Eq. (45), this produces the factorial large-order behavior
The position controls the exponential rate, the exponent controls a power of , and is the residue or normalization of that renormalon mode.
In the same convention, the formal Borel–Laplace reconstruction is
where specifies how the positive-axis singularities are bypassed. Expanding Eq. (45) inside the integral immediately reproduces , which checks all normalization factors. Because the singularity (56) lies on the integration path, one must specify whether the contour passes above or below the cut. The two lateral sums differ by an imaginary ambiguity of the parametric form
Renormalization-group evolution converts the exponential into a power correction of order , associated with OPE dimension [30]. The OPE [5] separates the observable schematically as
where denotes the common scheme and Borel prescription. The ambiguity of the lateral sum of is cancelled by the corresponding ambiguity in the operator matrix element (and, where appropriate, its Wilson coefficient). The physical observable is prescription independent; the separate perturbative and condensate pieces are not. Renormalon-free definitions of the gluon condensate [35, 36] and gradient-flow formulations of the OPE [37, 38] make this compensation explicit.
Padé approximation and sequence transformations can approximate analytic continuation, reveal candidate singularities, and improve numerical estimates within a chosen prescription. They cannot eliminate the physical need for a prescription, because no manipulation of the perturbative coefficients alone defines the compensating nonperturbative matrix element. A quoted principal-value Borel sum is a consistent and useful convention, but it is a convention. This is why the BJ–CVRC uncertainty below must not be interpreted as removing OPE prescription dependence.
A Beneke–Jamin-type ansatz and cross-validated closure
The defect obstruction identifies what a successful continuation near the usual range must contain: a mechanism capable of reversing the defect sign. Ultraviolet renormalons on the negative Borel axis generate sign-alternating contributions and provide such a mechanism. The natural precedent is the renormalon model of BJ [9]; related Adler-function models are discussed in Refs. [29, 42, 43, 44, 45, 46].
The construction used here is not the full BJ model. It retains the leading singularity locations and renormalization-group exponents, replaces full cuts by pure powers, and omits some subleading cut coefficients and Wilson-coefficient information. We call it a Beneke–Jamin-type ansatz. “CVRC” names the cross-validated-renormalon-closure protocol applied to that ansatz; it does not name a new physical Borel model. The physical construction is denoted CVRC-minimal: at each order it contains only the finite set of declared singularity and polynomial columns needed to match the available coefficients. A separate CVRC-complete construction, containing the full singularity tower available in the exactly solvable large- limit, is used only as an all-order validation laboratory and is not the model used for the physical forecast.
The algebra of closure can be seen without committing to every basis function. We write the model as
where are fixed renormalon or polynomial basis functions, are amplitudes determined by known coefficients, and is one deliberately unclosed amplitude. Matching known Borel coefficients gives a linear system
Any next coefficient extracted linearly from the model is therefore affine in ,
with intercept and slope fixed by the declared structure and the known coefficients. Because perturbative coefficients form an ordered sequence, the available fold is a one-step rolling-origin calibration. The lower-order family built from withholds the next known coefficient and fixes through
The calibration target is therefore reproduced exactly after is fixed and is not independent validation. The nontrivial test is whether the same closure amplitude is compatible, within the declared tolerance, with the lower-order fold and all three moment weights. Equivalently, the admissible intervals obtained at the known orders must have a nonempty intersection; an empty intersection rejects the declared singularity structure. At the physical order the calibrated central value coincides algebraically with the exact four-parameter fit of the pure-power ansatz and contains no information beyond those four inputs. CVRC adds the acceptance-or-rejection test, the admissible interval, external benchmarks, and a calibrated method uncertainty; it does not turn the fit into a determination. We use “cross-validated” only in this restricted cross-order consistency sense, not in the statistical sense of an independent validation sample.
For the declared structure the admissible set is nonempty, around , and the surviving pure-power BJ-type construction gives
The two components have different meanings. The structural component combines linearly the admissible closure window, which maps to , and a variation of the unknown exponent between 3.00 and 3.58, which moves the central value over . This gives , symmetrized to . The method component is the error of the transformation itself, and it dominates. The genuinely blind version of the protocol, with fixed one order lower from , postdicts , which is 29.9% below the known value. In the large- laboratory the same minimal tower is first defined at , where its one-order-ahead error is ; blind tests on the four-loop pole-mass coefficient [83] give 2.3–11.9%, but that series is dominated by the nearby renormalon and is an optimistic bound. We therefore set the method component to 30% of the central value, 74, the larger of the 23.7% error at the first defined minimal-tower large- order and the 29.9% magnitude of the strictly recursive physical miss. This choice brackets the only same-observable blind failure instead of assuming that its relative size improves at the next order. Because the laboratory errors are positive while the physical blind miss is negative, no sign-informed correction is justified and the band is symmetric. The quoted 80 is the quadrature combination
rounded to the nearest ten.
We observe that the central value is compatible with the Borel–Padé estimate 277 51 [47], with the Levin-transform estimate [49], and with the four-coefficient forecast 280.12 of the four-parameter BJ predictive variant [9], which lies 0.40 times the total uncertainty above it. Agreement does not, however, establish uniqueness. The 32-unit difference from that variant is entirely subleading structure: replacing pure powers by BJ’s cut corrections adds 9.6, and the gluon-condensate Wilson coefficient a further 22.6. With that structure declared, however, the admissible set is empty (the closure drifts by 0.236 between folds, against 0.051 for pure powers), and the strictly recursive postdiction is , five times the true value. This is a useful warning: adding theoretically correct asymptotic information can worsen low-order prediction if the asymptotic regime has not begun.
An empty admissible set is also what makes CVRC a test of the declared structure. In a misspecification scan the leading infrared singularity is moved from to positions where the Adler function has none, and the leading ultraviolet singularity from to or . A leading singularity at is rejected, as it should be: the massless Adler function has no renormalon, because no gauge-invariant operator of dimension two enters its OPE. Both ultraviolet displacements and are also rejected. The test is not perfectly discriminating: a leading singularity at survives, with , and the ultraviolet exponent is resolved only coarsely.
The large- limit [30, 50, 51] used above retains and reorganizes a subset of fermion-bubble insertions so that renormalon singularities are explicit. It is a controlled laboratory for large-order behavior, not a quantitatively exact replacement for full QCD. In it the complete singularity tower and the minimal tower reach stable accuracy at different orders. The complete tower meets ten percent persistently from , whereas the minimal tower used for the physical construction does so only from ; keeping this distinction prevents a successful asymptotic test from being overinterpreted at physical order four. The companion paper [70] documents all of these tests.
Experimental relevance and future facilities
The coefficient is not an observable: it multiplies in the massless Adler function and, in a specified renormalization scheme, can be fixed unambiguously only by a genuine six-loop calculation. Future data can nevertheless test its phenomenological consequences. Belle II and the proposed Super Tau–Charm Facility (STCF) can improve exclusive branching fractions, inclusive spectral distributions, strange–nonstrange channel separation, detector unfolding, and bin-to-bin covariance matrices [84, 85]. Belle data already enter the recent analysis of Ref. [19]. More precise spectra reduce moment uncertainties and may permit additional weights or tighter validation criteria. They do not, however, change the algebraic resolution in Eq. (35) when the weights and tolerance are held fixed.
The FCC-ee -pole programme would provide decays, together with precise high-scale QCD observables [86, 87]. If sufficiently inclusive hadronic spectra and their covariance matrices can be reconstructed, they would furnish complementary low-scale moment tests; high-scale observables would instead test the running and global consistency of . Neither programme directly measures .
A controlled forecast of is therefore useful for estimating the first omitted FOPT contribution, assigning correlated truncation uncertainties consistently across different weights, and preventing perturbative model dependence from being mistaken for experimental tension.
Scope, limitations, and guide to the companion paper
Each conclusion includes its hypotheses. The transfer identity is exact for one common Adler forecast mapped to normalized FOPT moments with the lower coefficients held fixed; independent moment-space forecasts give weight-dependent Adler values (§ 7). The defect-sign obstruction constrains individual sign-preserving continuations of the common Adler sequence; it does not apply to moment-space estimates such as the BJ linear-ratio value 283, and it need not be inherited by an ensemble average. The CIPT criterion is also one-dimensional, but it lacks the unit diagonal and acts on the contour-improved partial sum, the closest analogue of the coefficient-level FOPT test. Finally, is conditional on the declared pure-power ansatz and the closure rule; the band is nonstatistical and does not remove the OPE prescription dependence (§ 9).
This paper is self-contained for the physics, exact identities, and numerical conclusion. The companion paper [70] provides the full renormalization-group recursion and FOPT map matrices; the direct, Borel-normalized, Padé, Wynn, Levin, and Brezinski transformations [82, 88, 89, 90], including algebraic coincidences among nominally different methods; the singularity-structure and misspecification scans with their closure intervals; the large- and pole-mass benchmarks; and the numerical contour integration for CIPT. The accompanying reproducibility package regenerates the letter’s numerical audit and figure with two independent implementations.
Conclusions
Several spectral weights are physically well motivated, since they differ in pinching, OPE projection, and renormalon sensitivity. A common Adler-coefficient forecast nevertheless carries a single unknown error. Because the FOPT Adler-to-moment map is triangular with unit diagonal, that error is transferred unchanged to every normalized moment, and a simultaneous relative test reduces to one tolerance set by the smallest moment coefficient. For the three weights used here, a ten-percent fifth-order test resolves only to about ; the closest CIPT analogue, which acts on the partial sum, is weaker by more than an order of magnitude.
The known coefficients also expose a structural obstruction. The defects of the direct Padé/Shanks continuation are positive and increasing, so every sign-preserving continuation of the Adler sequence gives , whereas the usual range near 280 requires the next defect to be negative. Reaching that range requires a mechanism for the reversal. A simplified BJ-type renormalon ansatz supplies one, and CVRC rejects the ansatz if no single closure reproduces the two highest known orders. The resulting conditional estimate, (struct) (method) , is broad by design. Its calibrated central value is the exact four-parameter pure-power fit; the evidential content lies in the admissibility tests and in the uncertainty assignment. It is compatible with the usual range, but four coefficients do not determine the fifth precisely.
No Padé approximation, sequence transformation, or renormalon fit removes the OPE prescription dependence, because the perturbative Borel ambiguity cancels only against the matrix-element ambiguity in a common scheme and prescription. The decisive step would be a direct calculation of . Until then, declared structural assumptions, falsifiable validation, and honest uncertainty bands are more informative than precision generated by an underconstrained model.
Acknowledgments
G.A. is grateful to Antonio Pich for crucial discussions during his visit to the Instituto de Física Corpuscular (IFIC), Universitat de València–CSIC, which shaped the core of this work. He also thanks IFIC for its warm hospitality and stimulating research environment.
References
- [1]M. Davier, A. Höcker and Z. Zhang, Rev. Mod. Phys. 78, 1043 (2006), arXiv:hep-ph/0507078.
- [2]A. Pich, Prog. Part. Nucl. Phys. 117, 103846 (2021), arXiv:2012.04716.DOI
- [3]S. Narison and A. Pich, Phys. Lett. B 211, 183 (1988), doi:10.1016/0370-2693(88)90830-1.
- [4]E. Braaten, S. Narison and A. Pich, Nucl. Phys. B 373, 581 (1992), doi:10.1016/0550-3213(92)90267-F.DOI
- [5]K. G. Wilson, Phys. Rev. 179, 1499 (1969), doi:10.1103/PhysRev.179.1499.DOI
- [6]P. A. Baikov, K. G. Chetyrkin and J. H. Kühn, Phys. Rev. Lett. 101, 012002 (2008), arXiv:0801.1821.DOI
- [7]M. Davier, S. Descotes-Genon, A. Höcker, B. Malaescu and Z. Zhang, Eur. Phys. J. C 56, 305 (2008), arXiv:0803.0979.DOI
- [8]K. Maltman and T. Yavin, Phys. Rev. D 78, 094020 (2008), arXiv:0807.0650.DOI
- [9]M. Beneke and M. Jamin, JHEP 09, 044 (2008), arXiv:0806.3156.DOI
- [10]S. Bethke, Eur. Phys. J. C 64, 689 (2009), arXiv:0908.1135.DOI
- [11]A. Pich, Acta Phys. Polon. Supp. 3, 165 (2010), arXiv:1001.0389.arxiv.org/abs/1001.0389
- [12]A. Pich, Nucl. Phys. B Proc. Suppl. 218, 89 (2011), arXiv:1101.2107.DOI
- [13]S. Bethke et al., Workshop on Precision Measurements of αₛ, arXiv:1110.0016; see in particular the contributions of A. Pich (p. 21) and of M. Beneke and M. Jamin (p. 25).
- [14]D. Boito, O. Catà, M. Golterman, M. Jamin, K. Maltman, J. Osborne and S. Peris, Phys. Rev. D 84, 113006 (2011), arXiv:1110.1127.DOI
- [15]D. Boito, M. Golterman, K. Maltman, J. Osborne and S. Peris, Phys. Rev. D 91, 034003 (2015), arXiv:1410.3528.DOI
- [16]A. Pich and A. Rodríguez-Sánchez, Phys. Rev. D 94, 034027 (2016), arXiv:1605.06830.DOI
- [17]D. Boito, M. Golterman, K. Maltman, S. Peris, M. V. Rodrigues and W. Schaaf, Phys. Rev. D 103, 034028 (2021), arXiv:2012.10440.DOI
- [18]C. Ayala, G. Cvetič and D. Teca, J. Phys. G 50, 045004 (2023), arXiv:2206.05631.DOI
- [19]D. Boito, A. Eiben, M. Golterman, K. Maltman, L. M. Mansur and S. Peris, Phys. Rev. D 111, 074010 (2025), arXiv:2502.08147.arxiv.org/abs/2502.08147
- [20]O. Catà, M. Golterman and S. Peris, Phys. Rev. D 77, 093006 (2008), arXiv:0803.0246.DOI
- [21]A. Pich and A. Rodríguez-Sánchez, JHEP 07, 145 (2022), arXiv:2205.07587.DOI
- [22]D. Boito, M. Golterman, K. Maltman and S. Peris, Phys. Rev. D 111, 074019 (2025), arXiv:2402.05588.arxiv.org/abs/2402.05588
- [23]A. A. Pivovarov, Z. Phys. C 53, 461 (1992) [Sov. J. Nucl. Phys. 54, 676 (1991)], arXiv:hep-ph/0302003.
- [24]F. Le Diberder and A. Pich, Phys. Lett. B 286, 147 (1992).
- [25]M. Jamin, JHEP 09, 058 (2005), arXiv:hep-ph/0509001.DOI
- [26]S. Menke, arXiv:0904.1796.arxiv.org/abs/0904.1796
- [27]G. Cvetič, M. Loewe, C. Martinez and C. Valenzuela, Phys. Rev. D 82, 093007 (2010), arXiv:1005.4444.DOI
- [28]S. Groote, J. G. Körner and A. A. Pivovarov, Phys. Part. Nucl. 44, 285 (2013), arXiv:1212.5346.DOI
- [29]M. Beneke, D. Boito and M. Jamin, JHEP 01, 125 (2013), arXiv:1210.8038.DOI
- [30]M. Beneke, Phys. Rept. 317, 1 (1999), arXiv:hep-ph/9807443.DOI
- [31]A. H. Hoang and C. Regner, Phys. Rev. D 105, 096023 (2022), arXiv:2008.00578.arxiv.org/abs/2008.00578
- [32]A. H. Hoang and C. Regner, Eur. Phys. J. ST 230, 2625 (2021), arXiv:2105.11222.DOI
- [33]M. Golterman, K. Maltman and S. Peris, Phys. Rev. D 108, 014007 (2023), arXiv:2305.10386.DOI
- [34]N. G. Gracia, A. H. Hoang and V. Mateu, Phys. Rev. D 108, 034013 (2023), arXiv:2305.10288.DOI
- [35]M. A. Benitez-Rathgeb, D. Boito, A. H. Hoang and M. Jamin, JHEP 07, 016 (2022), arXiv:2202.10957.
- [36]M. A. Benitez-Rathgeb, D. Boito, A. H. Hoang and M. Jamin, JHEP 09, 223 (2022), arXiv:2207.01116.
- [37]M. Beneke and H. Takaura, PoS RADCOR2023, 062 (2024), arXiv:2309.10853.arxiv.org/abs/2309.10853
- [38]M. Beneke and H. Takaura, arXiv:2510.12193.arxiv.org/abs/2510.12193
- [39]W. A. Bardeen, A. J. Buras, D. W. Duke and T. Muta, Phys. Rev. D 18, 3998 (1978).
- [40]A. David and G. Passarino, Phys. Lett. B 726, 266 (2013), arXiv:1307.1843.DOI
- [41]A. L. Kataev and V. V. Starshenko, Mod. Phys. Lett. A 10, 235 (1995), arXiv:hep-ph/9502348.
- [42]I. Caprini and J. Fischer, Eur. Phys. J. C 64, 35 (2009), arXiv:0906.5211.DOI
- [43]S. Descotes-Genon and B. Malaescu, arXiv:1002.2968.
- [44]M. Jamin, Nucl. Phys. B Proc. Suppl. 218, 98 (2011), arXiv:1101.0681.DOI
- [45]I. Caprini and J. Fischer, Nucl. Phys. B Proc. Suppl. 218, 128 (2011), arXiv:1011.6480.DOI
- [46]I. Caprini and J. Fischer, Phys. Rev. D 84, 054019 (2011), arXiv:1106.5336.DOI
- [47]D. Boito, P. Masjuan and F. Oliani, JHEP 08, 075 (2018), arXiv:1807.01567.DOI
- [48]I. Caprini, Phys. Rev. D 100, 056019 (2019), arXiv:1908.06632.DOI
- [49]G. Abbas and V. Singh, Phys. Lett. B 873, 140156 (2026), arXiv:2510.16119.DOI
- [50]C. N. Lovett-Turner and C. J. Maxwell, Nucl. Phys. B 432, 147 (1994), arXiv:hep-ph/9407268.
- [51]P. Ball, M. Beneke and V. M. Braun, Nucl. Phys. B 452, 563 (1995), arXiv:hep-ph/9502300.DOI
- [52]C. N. Lovett-Turner and C. J. Maxwell, Nucl. Phys. B 452, 188 (1995), arXiv:hep-ph/9505224.DOI
- [53]M. Neubert, Nucl. Phys. B 463, 511 (1996), arXiv:hep-ph/9509432.DOI
- [54]C. J. Maxwell and D. G. Tonge, Nucl. Phys. B 481, 681 (1996), arXiv:hep-ph/9606392.
- [55]C. J. Maxwell and D. G. Tonge, Nucl. Phys. B 535, 19 (1998), arXiv:hep-ph/9705314.DOI
- [56]C. J. Maxwell and A. Mirjalili, Nucl. Phys. B 611, 423 (2001), arXiv:hep-ph/0103164.DOI
- [57]G. Altarelli, P. Nason and G. Ridolfi, Z. Phys. C 68, 257 (1995), arXiv:hep-ph/9501240.DOI
- [58]G. Cvetič, C. Dib, T. Lee and I. Schmidt, Phys. Rev. D 64, 093016 (2001), arXiv:hep-ph/0106024.
- [59]G. Abbas, B. Ananthanarayan and I. Caprini, Phys. Rev. D 85, 094018 (2012), arXiv:1202.2672.DOI
- [60]G. Abbas, B. Ananthanarayan, I. Caprini and J. Fischer, Phys. Rev. D 87, 014008 (2013), arXiv:1211.4316.DOI
- [61]G. Abbas, B. Ananthanarayan, I. Caprini and J. Fischer, Phys. Rev. D 88, 034026 (2013), arXiv:1307.6323.DOI
- [62]I. Caprini, Phys. Rev. D 102, 054017 (2020), arXiv:2006.16605.DOI
- [63]I. Caprini, Phys. Rev. D 108, 114031 (2023), arXiv:2312.07143.arxiv.org/abs/2312.07143
- [64]I. Caprini, Phys. Rev. D 109, 074002 (2024), arXiv:2403.10844.arxiv.org/abs/2403.10844
- [65]K. Schilcher, C. Sebu and H. Spiesberger, Phys. Lett. B 852, 138590 (2024), arXiv:2401.17131.arxiv.org/abs/2401.17131
- [66]D. Boito, M. Jamin and R. Miravitllas, Phys. Rev. Lett. 117, 152001 (2016), arXiv:1606.06175.DOI
- [67]X.-G. Wu, J.-M. Shen, B.-L. Du and S. J. Brodsky, Phys. Rev. D 97, 094030 (2018), arXiv:1802.09154.DOI
- [68]I. Caprini, Phys. Rev. D 98, 056016 (2018), arXiv:1806.10325.DOI
- [69]X.-G. Wu, J.-M. Shen, B.-L. Du, X.-D. Huang, S.-Q. Wang and S. J. Brodsky, Prog. Part. Nucl. Phys. 108, 103706 (2019), arXiv:1903.12177.
- [70]G. Abbas, D. Biswas, P. Dadhich, and V. Singh, “Hadronic tau decays at higher orders in QCD,” [arXiv:2601.11277 [hep-ph]].arxiv.org/abs/2601.11277
- [71]S. Schael et al. [ALEPH Collaboration], Phys. Rept. 421, 191 (2005), arXiv:hep-ex/0506072.
- [72]M. Davier, A. Höcker, B. Malaescu, C.-Z. Yuan and Z. Zhang, Eur. Phys. J. C 74, 2803 (2014), arXiv:1312.1501.DOI
- [73]W. J. Marciano and A. Sirlin, Phys. Rev. Lett. 61, 1815 (1988).
- [74]E. Braaten and C. S. Li, Phys. Rev. D 42, 3888 (1990).
- [75]M. Davier, S. Eidelman, A. Höcker and Z. Zhang, Eur. Phys. J. C 27, 497 (2003), arXiv:hep-ph/0208177.DOI
- [76]A. Pich and J. Prades, JHEP 10, 004 (1999), arXiv:hep-ph/9909244.DOI
- [77]P. A. Baikov, K. G. Chetyrkin and J. H. Kühn, Phys. Rev. Lett. 118, 082002 (2017), arXiv:1606.08659.DOI
- [78]J. Peñarrocha and K. Schilcher, Phys. Lett. B 515, 291 (2001), arXiv:hep-ph/0105222.
- [79]M. Dalla Brida, R. Hölli wieser, F. Knechtli, T. Korzec, A. Ramos, S. Sint and R. Sommer, arXiv:2501.06633.arxiv.org/abs/2501.06633
- [80]K. G. Chetyrkin, J. H. Kühn and M. Steinhauser, Comput. Phys. Commun. 133, 43 (2000), arXiv:hep-ph/0004189.DOI
- [81]F. Herren and M. Steinhauser, Comput. Phys. Commun. 224, 333 (2018), arXiv:1703.03751.DOI
- [82]E. J. Weninger, Comput. Phys. Rep. 10, 189 (1989).
- [83]P. Marquard, A. V. Smirnov, V. A. Smirnov and M. Steinhauser, Phys. Rev. Lett. 114, 142002 (2015), arXiv:1502.01030.
- [84]E. Kou et al. [Belle II Collaboration], Prog. Theor. Exp. Phys. 2019, 123C01 (2019); 2020, 029201(E) (2020), arXiv:1808.10567.DOI
- [85]M. Achasov et al., arXiv:2303.15790.DOI
- [86]M. Dam, arXiv:2107.12832.arxiv.org/abs/2107.12832
- [87]M. Benedikt et al. [FCC Collaboration], Eur. Phys. J. C 85, 1468 (2025), arXiv:2505.00272.DOI
- [88]D. Levin, Int. J. Comput. Math. B 3, 371 (1973).
- [89]D. A. Smith and W. F. Ford, SIAM J. Numer. Anal. 16, 223 (1979).
- [90]D. A. Smith and W. F. Ford, Math. Comput. 38, 481 (1982).
