betaFeedback

WorksPhysics and Astronomy

TMD phenomenology motivated by nonperturbative structures

Abstract

This talk summarized work done recently to organize the steps for implementing TMD phenomenology in a way optimized for contexts where the extraction and interpretation of hadronic structures and nonperturbative effects is the primary driving motivation.

JLAB-THY-24-4138

7th International Workshop on “Transverse phenomena in hard processes and the transverse structure of the proton (Transversity2024)

3-7 June 2024

Trieste, Italy

© Copyright owned by the author(s) under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0). https://pos.sissa.it/

Introduction

I will structure these proceedings as follows: To establish context, I will briefly review the basic setup of TMD factorization for Drell-Yan scattering. Next, I will review the most common steps used to construct parametrizations of TMD pdfs in phenomenological implementations and discuss some of their advantages and disadvantages. Then, I will summarize our proposed modifications from Refs. [1, 8, 9]. I will organize the remainder of the proceedings around answering questions that arose in the context of the talk.

A familiar example of a TMD factorization formula is the one for unpolarized Drell-Yan scattering [4],

dσd4q dΩ=H(Q0,Q0/μQ0)×∫d2kaT d2kbT fj/ha(xa,kaT;μQ0,Q02)f‾jˉ/hb(xb,kbT;μQ0,Q02)δ(2)(qT−kaT−kbT),(1)\frac{\mathrm{d}\sigma}{\mathrm{d}^{4}q\,\mathrm{d}\Omega} = \mathcal{H}(Q_{0},Q_{0}/\mu_{Q_{0}}) \times\int\mathrm{d}^{2}k_{aT}\,\mathrm{d}^{2}k_{bT}\,f_{j/h_{a}}(x_{a},k_{aT};\mu_{Q_{0}},Q_{0}^{2})\overline{f}_{\bar{j}/h_{b}}(x_{b},k_{bT};\mu_{Q_{0}},Q_{0}^{2})\delta^{(2)}(q_{T}-k_{aT}-k_{bT}) , \tag*{(1)}

with TMD pdfs fj/ha(xa,kaT;μQ0,Q02)f_{j/h_{a}}(x_{a},k_{aT};\mu_{Q_{0}},Q_{0}^{2}) and f‾jˉ/hb(xb,kbT;μQ0,Q02)\overline{f}_{\bar{j}/h_{b}}(x_{b},k_{bT};\mu_{Q_{0}},Q_{0}^{2}) describing longitudinal momentum fractions and transverse momenta for the incoming partons in each hadron hah_{a} and hbh_{b}. There is also a hard factor H\mathcal{H} and a sum over all parton flavors. Apart from the dependence on auxiliary variables associated with evolution, Eq. (1) has exactly the form of a TMD parton model with a clear hadron structure interpretation. (For brevity, power suppressed errors, sums over flavors, and a YY-term correction for qT≈Q0q_{T} \approx Q_{0} are not written explicitly.) The renormalization and Collins-Soper (CS) scales are normally written as generic scales μ\mu and ζ\sqrt{\zeta}, but since we have physical applications in mind I express them with physical input scales μQ0\mu_{Q_{0}} and Q0Q_{0}, fixed to μQ0∝Q0\mu_{Q_{0}} \propto Q_{0} from the outset.

It is often the case that one prefers to express Eq. (1) in terms of transverse coordinate space TMDs where implementing evolution becomes simple,

dσd4q dΩ=H(Q0,Q0/μQ0)∫d2bT(2π)2eiqT⋅bTf~j/ha(xa,bT;μQ0,Q02)f~jˉ/hb(xb,bT;μQ0,Q02).(2)\frac{\mathrm{d}\sigma}{\mathrm{d}^{4}q\,\mathrm{d}\Omega} = \mathcal{H}(Q_{0},Q_{0}/\mu_{Q_{0}}) \int\frac{\mathrm{d}^{2}b_{T}}{(2\pi)^{2}} e^{i q_{T}\cdot b_{T}} \widetilde{f}_{j/h_{a}}(x_{a},b_{T};\mu_{Q_{0}},Q_{0}^{2})\widetilde{f}_{\bar{j}/h_{b}}(x_{b},b_{T};\mu_{Q_{0}},Q_{0}^{2}) . \tag*{(2)}

The formulas above are written for the case of scales near the input scale, Q0Q_{0}, which is understood to be the lowest scale for which TMD factorization is considered reasonable. To relate them to higher QQ requires solving evolution equations [4],

∂ln⁡f~j/h(x,bT;μQ,ζ)∂ln⁡ζ=K~(bT;μ),(3)\frac{\partial\ln\widetilde{f}_{j/h}(x,b_{T};\mu_{Q},\zeta)}{\partial\ln\sqrt{\zeta}} = \widetilde{K}(b_{T};\mu) , \tag*{(3)}
dK~(bT;μ)dln⁡μ=−γK(αs(μ)),(4)\frac{\mathrm{d}\widetilde{K}(b_{T};\mu)}{\mathrm{d}\ln\mu} = -\gamma_{K}(\alpha_{s}(\mu)) , \tag*{(4)}
dln⁡f~j/h(x,bT;μQ,ζ)dln⁡μ=γ(αs(μ);ζ/μ2)=γ(αs(μ);1)−γK(αs(μ))12ln⁡(ζμ2).(5)\frac{\mathrm{d}\ln\widetilde{f}_{j/h}(x,b_{T};\mu_{Q},\zeta)}{\mathrm{d}\ln\mu} = \gamma(\alpha_{s}(\mu);\zeta/\mu^{2}) = \gamma(\alpha_{s}(\mu);1) - \gamma_{K}(\alpha_{s}(\mu))\frac{1}{2}\ln\left(\frac{\zeta}{\mu^{2}}\right) . \tag*{(5)}

Here, γ\gamma and γK\gamma_{K} are perturbatively calculable anomalous dimensions and K~(bT;μ)\widetilde{K}(b_{T};\mu) is the CS kernel, which is perturbatively calculable at small bTb_{T}, but is nonperturbative in the large-bTb_{T} limit. In bTb_{T}-space, the exact solution to Eqs. (3)–(5) for all bTb_{T} is simple, and the factorized cross section at an arbitrary QQ becomes

dσd4q dΩ=H(Q,Q/μQ)∫d2bT(2π)2eiqT⋅bTf~j/ha(xa,bT;μQ0,Q02)f~jˉ/hb(xb,bT;μQ0,Q02)E(μQ/μQ0,Q/Q0,μQ0bT)2.(6)\frac{\mathrm{d}\sigma}{\mathrm{d}^{4}q\,\mathrm{d}\Omega}=\mathcal{H}(Q,Q/\mu_{Q})\int\frac{\mathrm{d}^{2}b_{T}}{(2\pi)^{2}}e^{i\boldsymbol{q}_{T}\cdot\boldsymbol{b}_{T}}\widetilde{f}_{j/h_{a}}(x_{a},b_{T};\mu_{Q_{0}},Q_{0}^{2})\widetilde{f}_{\bar{j}/h_{b}}(x_{b},b_{T};\mu_{Q_{0}},Q_{0}^{2})E(\mu_{Q}/\mu_{Q_{0}},Q/Q_{0},\mu_{Q_{0}}b_{T})^{2}. \tag*{(6)}

with only a multiplicative evolution factor

E(μQ/μQ0,Q/Q0,μQ0bT)2=exp⁡{K~(bT;μQ0)ln⁡(Q2Q02)+∫μQ0μQdμ′μ′[2γ(αs(μ′);1)−ln⁡(Q2μ′2)γK(αs(μ′))]}.(7)\begin{aligned} E(\mu_{Q}/\mu_{Q_{0}},Q/Q_{0},\mu_{Q_{0}}b_{T})^{2} \\ &=\exp\left\{\widetilde{K}(b_{T};\mu_{Q_{0}})\ln\left(\frac{Q^{2}}{Q_{0}^{2}}\right)+\int_{\mu_{Q_{0}}}^{\mu_{Q}}\frac{\mathrm{d}\mu'}{\mu'}\left[2\gamma(\alpha_{s}(\mu');1)-\ln\left(\frac{Q^{2}}{\mu'^{2}}\right)\gamma_{K}(\alpha_{s}(\mu'))\right]\right\}. \tag*{(7)} \end{aligned}

(The square on EE is present because there is an evolution factor for each TMD pdf.) Therefore, knowledge of the TMD pdfs and the CS kernel at an input scale and for all bTb_{T}, along with perturbative calculations of the anomalous dimensions, is enough to determine the TMD pdfs and the cross section at any higher scales.

To obtain expressions for the TMD pdfs at kT≈Q0k_{T}\approx Q_{0} or higher (or at bT≲1/Q0b_{T}\lesssim1/Q_{0} in coordinate space), one also normally takes advantage of the ability to use collinear factorization to describe the perturbative tail, or in terms of the small-bTb_{T} limit to use an operator product expansion,

f~j/h(x,bT;μQ0,Q02)=∫x1dξξC~j/k(x/ξ,bT;μQ0,Q02,αs(μQ0))fk/p(ξ;μQ0)+O((bTΛQCD)a),(8)\widetilde{f}_{j/h}(x,b_{T};\mu_{Q_{0}},Q_{0}^{2})=\int_{x}^{1}\frac{\mathrm{d}\xi}{\xi}\widetilde{C}_{j/k}(x/\xi,b_{T};\mu_{Q_{0}},Q_{0}^{2},\alpha_{s}(\mu_{Q_{0}}))f_{k/p}(\xi;\mu_{Q_{0}})+O\left((b_{T}\Lambda_{\mathrm{QCD}})^{a}\right), \tag*{(8)}

where C~j/k\widetilde{C}_{j/k} is a hard coefficient and a>0a>0. The only logarithmic behavior in the perturbative expression for C~j/k\widetilde{C}_{j/k} is of the form ln⁡n(bTQ0)\ln^{n}(b_{T}Q_{0}), so using Eq. (8) directly in calculations is already optimal for Q≈Q0Q\approx Q_{0} where the only relevant perturbative region of bTb_{T} occurs at bT≈1/Q0b_{T}\approx1/Q_{0}. By definition, all larger regions of bTb_{T} are to be considered nonperturbative. However, after one evolves to Q≫Q0Q\gg Q_{0}, the bT≪1/Q0b_{T}\ll1/Q_{0} region starts to become important, and the ln⁡n(bTQ0)\ln^{n}(b_{T}Q_{0}) terms grow unacceptably large. To account for this, one normally performs another scale transformation on f~j/h(x,bT;μQ0,Q02)\widetilde{f}_{j/h}(x,b_{T};\mu_{Q_{0}},Q_{0}^{2}) in the bT≪1/Q0b_{T}\ll1/Q_{0} region from (μQ0,Q0)→∼1/bT(\mu_{Q_{0}},Q_{0})\to\sim1/b_{T} via another application of Eqs. (3)–(5).

The abstract set of steps summarized above are made concrete in applications by replacing the various pieces with a combination of perturbative approximations and nonperturbative parametrizations. The details of the different ways this is done in practice come with advantages and disadvantages depending on the specific goals, to be discussed below.

b* method with g-function prescription

A basic task is to fix a method for merging nonperturbative descriptions of transverse momentum dependence with perturbative descriptions. In transverse coordinate space, that amounts to matching large-bTb_{T} nonperturbative descriptions of f~j/h(x,bT;μQ0,Q02)\widetilde{f}_{j/h}(x,b_{T};\mu_{Q_{0}},Q_{0}^{2}) and K~(bT;μ)\widetilde{K}(b_{T};\mu) onto small-bTb_{T} perturbative calculations.

The most common way of organizing the separation is the b∗b_*-prescription [5, 6]. It is motivated by a desire to separate out a factor that one can view as strictly perturbative, in the sense that it is only sensitive to small-bTb_T behavior and has zero or only minimal sensitivity to parameters that are meant to describe truly nonperturbative transverse momentum dependence. In other words, the perturbatively describable transverse coordinate dependence is to be quarantined from explicitly nonperturbative models of transverse coordinate dependence. The way it is implemented is by introducing an arbitrary function that freezes bTb_T when bT→∞b_T \to\infty,

b∗(bT)={bTbT≪bmax⁡bmax⁡bT≫bmax⁡(9)b_*(b_T) = \begin{cases} b_T & b_T \ll b_{\max} \\ b_{\max} & b_T \gg b_{\max} \end{cases} \tag*{(9)}

with bmax⁡b_{\max} understood to mark a boundary between what will be treated perturbatively and what will be considered nonperturbative regions of bTb_T dependence. (Thus, bmax⁡≲1/Q0b_{\max} \lesssim1/Q_0.) The most common realization of Eq. (9) is

b∗(bT)=bT1+bT2/bmax⁡2,(10)b_*(b_T) = \frac{b_T}{\sqrt{1+b_T^2/b_{\max}^2}}, \tag*{(10)}

though any function that satisfies Eq. (9) is acceptable, and others have been used in practice. The TMD pdfs at the input scale are then written as

f~j/h(x,bT;μQ0,Q02)=f~j/h(x,b∗;μQ0,Q02)(f~j/h(x,bT;μQ0,Q02)f~j/h(x,b∗;μQ0,Q02)).(11)\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q_0^2) = \widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2)\left(\frac{\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q_0^2)}{\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2)}\right). \tag*{(11)}

The factor in front is still just the coordinate space TMD pdf, but now it is only ever evaluated at bT<bmax⁡b_T < b_{\max}, so we anticipate using collinear factorization to calculate it. The factor in parentheses vanishes like a power of bTb_T at small bTb_T, so we label it as being the “nonperturbative” contribution. It is conventionally written in the form of an exponential of a function, so that

f~j/h(x,bT;μQ0,Q02)=f~j/h(x,b∗;μQ0,Q02)e−gj/h(x,bT)(12)\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q_0^2) = \widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2)e^{-g_{j/h}(x,b_T)} \tag*{(12)}

with

gj/h(x,bT)≡−ln⁡(f~j/h(x,bT;μQ0,Q02)f~j/h(x,b∗;μQ0,Q02)).(13)g_{j/h}(x,b_T) \equiv-\ln\left(\frac{\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q_0^2)}{\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2)}\right). \tag*{(13)}

The gj/h(x,bT)g_{j/h}(x,b_T) function depends on the choice of input scale Q0Q_0 and on the form of b∗b_*, but it is otherwise scale-independent.

As long as there is negligible sensitivity to the very small bT≪1/Q0b_T \ll1/Q_0 region, the form of Eq. (12) is sufficient to allow one to treat the overall factor f~j/h(x,b∗;μQ0,Q02)\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2) using collinear factorization. However, if we ultimately plan to evolve to such large QQ that the bT≪1/Q0b_T \ll1/Q_0 limit becomes accessible, then we must transform the scales in f~j/h(x,bT;μQ0,Q02)\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q_0^2) once more by writing

f~j/h(x,b∗;μQ0,Q02)=f~j/h(x,b∗;μb∗,μb∗2)E(μQ0/μb∗,Q0/μb∗,μb∗b∗),(14)\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2) = \widetilde{f}_{j/h}(x,b_*;\mu_{b_*},\mu_{b_*}^2)E(\mu_{Q_0}/\mu_{b_*},Q_0/\mu_{b_*},\mu_{b_*}b_*), \tag*{(14)}

where we have defined the scale

μb∗≡C1/b∗.(15)\mu_{b_*} \equiv C_1/b_* . \tag*{(15)}

and C1C_1 is a constant with size comparable to 1. Then, f~j/h(x,b∗(bT);μb∗,μb2)\widetilde{f}_{j/h}(x,b_*(b_T);\mu_{b_*},\mu_b^2) is perturbatively very well behaved as bT→0b_T \to0. The factors in front of e−gj/h(x,bT)e^{-g_{j/h}(x,b_T)} are never evaluated at bTb_T above bmax⁡b_{\max}, so one may choose bmax⁡b_{\max} to be small enough that the O((bmax⁡ΛQCD)a)O\left((b_{\max}\Lambda_{\mathrm{QCD}})^a\right) errors in Eq. (8) are negligible. Thus, we write Eq. (12) as

f~OPE,j/h(x,b∗;μb∗,μb∗2)E(μQ0/μb∗,Q0/μb∗,μb∗b∗)e−gj/h(x,bT)+O((bmax⁡ΛQCD)a)(16)\widetilde{f}_{\mathrm{OPE},j/h}(x,b_*;\mu_{b_*},\mu_{b_*}^2)E(\mu_{Q_0}/\mu_{b_*},Q_0/\mu_{b_*},\mu_{b_*}b_*)e^{-g_{j/h}(x,b_T)}+O\left((b_{\max}\Lambda_{\mathrm{QCD}})^a\right) \tag*{(16)}

where the OPE subscript symbolizes the first term in Eq. (8).

Isolating the nonperturbative part of the CS kernel is done in a way similar to the treatment of the TMD pdf itself. The evolution factor in the cross section, Eq. (6), is rewritten in the form

E(μQ/μQ0,Q/Q0,μQ0bT)=E(μQ/μQ0,Q/Q0,μQ0b∗)×E(μQ/μQ0,Q/Q0,μQ0bT)E(μQ/μQ0,Q/Q0,μQ0b∗)=E(μQ/μQ0,Q/Q0,μQ0b∗)exp⁡{−[K~(b∗;μQ0)−K~(bT;μQ0)]ln⁡(QQ0)}.(17)\begin{aligned} E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_T) &= E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_*) \\ &\quad{}\times\frac{E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_T)}{E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_*)} \\ &= E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_*)\exp\left\{-\left[\widetilde{K}(b_*;\mu_{Q_0})-\widetilde{K}(b_T;\mu_{Q_0})\right]\ln\left(\frac{Q}{Q_0}\right)\right\}. \tag*{(17)} \end{aligned}

Now one uses the fact (from Eq. (5)) that scale dependence in Eq. (17) is bTb_T-independent to identify and define the quantity in brackets as a scale-independent function called gK(bT)g_K(b_T). So the evolution factor is

E(μQ/μQ0,Q/Q0,μQ0bT)=E(μQ/μQ0,Q/Q0,μQ0b∗)exp⁡{−gK(bT)ln⁡(QQ0)}.(18)E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_T)=E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_*)\exp\left\{-g_K(b_T)\ln\left(\frac{Q}{Q_0}\right)\right\}. \tag*{(18)}

Now E(μQ/μQ0,Q/Q0,μQ0b∗)E(\mu_Q/\mu_{Q_0},Q/Q_0,\mu_{Q_0}b_*) is also never evaluated at bT>bmax⁡b_T>b_{\max}. 2 Substituting Eq. (16) and Eq. (18) into Eq. (6) reproduces the fully evolved cross section expression in the form that it is usually presented.

In the above summary, all symbols are meant to be exact so, apart from possibly dropping the O((bmax⁡ΛQCD)a)O\left((b_{\max}\Lambda_{\mathrm{QCD}})^a\right) in Eq. (16), nothing has been done to Eq. (6) other than to shuffle around factors arbitrarily. An implementation amounts to replacing these various pieces by approximations and parametrizations. The perturbatively calculable parts are normally obtained by truncating beyond some order, while in most approaches the nonperturbative functions gj/h(x,bT)g_{j/h}(x,b_T) and gK(bT)g_K(b_T) are replaced by nonperturbative ansatzes, which are then fit to data.

Organizing the steps in this way has at least the following major advantages:

  1. The final f~j/h(x,b∗;μQ0,Q02)\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2) is restricted to perturbatively small values of bTb_T. Calculating it using perturbative collinear factorization with the OPE involves no sensitivity to genuinely nonperturbative parameters, since changing b∗(bT)b_*(b_T) or the value of bmax⁡b_{\max} only amounts to shifting contributions between f~j/h(x,b∗;μQ0,Q02)\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2) and gj/h(x,bT)g_{j/h}(x,b_T). Analogous statements are true for the connection between K~(b∗;μQ0)\widetilde{K}(b_*;\mu_{Q_0}) and gK(bT)g_K(b_T).

At least formally, therefore, the goal of isolating “purely perturbative” contributions is achieved with this prescription. In the final expression for the cross section, the only sensitivity to b∗(bT)b_*(b_T) or bmax⁡b_{\max} comes from dropping the O((bmax⁡ΛQCD)a)O\left((b_{\max}\Lambda_{\mathrm{QCD}})^a\right) terms in Eq. (16), and these errors can be made arbitrarily small by choosing small enough bmax⁡b_{\max}.

2.) The gg-functions, gj/h(x,bT)g_{j/h}(x,b_T) and gK(bT)g_K(b_T), that describe nonperturbative transverse coordinate dependence are exactly scale independent since they are defined with ratios and differences. The gj/h(x,bT)g_{j/h}(x,b_T) function only depends on the initial choice of Q0Q_0 (through the CS evolution), and both gj/h(x,bT)g_{j/h}(x,b_T) and gK(bT)g_K(b_T) depend on b∗b_*. However, all QQ-dependence is either in perturbatively calculable quantities or is just the ln⁡Q\ln Q multiplying gK(bT)g_K(b_T).

3.) There are formally no modifications to or approximations of the fundamental operator definitions. For dealing with bTb_T becoming large, the only change is to the arguments of f~j/h(x,bT;μQ0,Q2)\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q^2) and K~(bT;μQ0)\widetilde{K}(b_T;\mu_{Q_0}), not to the functions themselves. Thus, the nonperturbative functions gj/h(x,bT)g_{j/h}(x,b_T) and gK(bT)g_K(b_T) have clear and unambiguous relationships to their underlying operator definitions, namely Eq. (13) and gK(bT)=K~(b∗;μQ0)−K~(bT;μQ0)g_K(b_T)=\widetilde{K}(b_*;\mu_{Q_0})-\widetilde{K}(b_T;\mu_{Q_0}).

Issues & complications

The advantages listed in 1.)-3.) above are formally true if the various parts of the TMD factorization expression for the cross section are understood to be exactly known. Of course, however, in practical implementations one always replaces them by approximations or ansatzes, and this is where complications may enter. Because gj/h(x,bT)g_{j/h}(x,b_T) and gK(bT)g_K(b_T) are where all the physics associated with nonperturbative transverse momentum dependence resides, and since detailed access to that information is the main goal in the present context, they will be the focus of the discussion below. The main complications with the standard organizational scheme, from the perspective of hadron structure studies, are the following:

a) In general, if nonperturbative parametrizations for gj/h(x,bT)g_{j/h}(x,b_T) and gK(bT)g_K(b_T) are chosen arbitrarily, then item 1.) above will be at most only approximately true. If violations of 1.) are too large, then f~j/h(x,b∗;μQ0,Q2)\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q^2) is not actually sequestered from the nonperturbative parametrization. The b∗(bT)b_*(b_T) function then effectively becomes an additional nonperturbative model and bmax⁡b_{\max} becomes a model parameter. Therefore, nonperturbative model parameters impact the perturbative/collinear part, thus removing one of the motivating advantages of the b∗b_* and gg-function method.

In applications to nonperturbative hadron structure, one of the tasks is to disentangle any purely nonperturbative transverse momentum dependence associated with bound states and hadronization from perturbative transverse momentum dependence, but violations of b∗b_*-independence indicate a breakdown in the ability to identify separate perturbative and nonperturbative parts.

b) In a parton-model-like picture, where TMD pdfs have a literal number density interpretation, the relationship between collinear and TMD pdfs is via simple integrals like

f(x)=∫d2kT f(x,kT).(19)f(x)=\int\mathrm{d}^{2}k_T\,f(x,k_T). \tag*{(19)}

In full QCD, this type of relation fails because the integral is ultraviolet (UV) divergent, but generalized versions of it [7, 9, 13] do hold in the form

f(x;μQ)=∫reg.d2kT f(x,kT;μQ,μQ2)+Δ+p.s.,(20)f(x;\mu_Q)=\int_{\mathrm{reg.}} \mathrm{d}^{2}k_T\,f(x,k_T;\mu_Q,\mu_Q^2)+\Delta+\mathrm{p.s.}, \tag*{(20)}

where the “reg” subscript symbolizes a UV regulator that suppresses the contribution from kT≳Qk_T \gtrsim Q, Δ\Delta is a higher order correction that is calculable in perturbative collinear factorization which is needed to transform the kTk_T-regulated collinear pdf to the MS scheme, and p.s. means “power-suppressed.” An arbitrarily chosen functional form for gj/h(x,bT)g_{j/h}(x,b_T) will not generally give a f~j/h(x,bT;μQ0,Q2)\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q^2) that satisfies Eq. (20) with previously known collinear pdf extractions for the left-hand side.

Equations of the form of (19) are an important part of the hadron structure framework of interpretations, where the nonperturbative parts of TMD pdfs are understood to describe properties of bound state quark and gluon constituents. This is emphasized in the classic “prism” diagram that one frequently finds in literature on nonperturbative parton structure — see Figure 1.

Figure 1. A reproduction of the classic “prism” diagram from Ref. [11 figure 1] showing the connections between different types of parton correlation functions in a nonperturbative structure interpretation. The transverse momentum integrals are typically UV divergent, and so preserving the interpretation in full QCD requires translating them into forms like Eq. (20).

Therefore, allowing the nonperturbative parts contained in gj/h(x,bT)g_{j/h}(x,b_T) to be unconstrained by the generalized version of these integrals in Eq. (20) conflicts with a goal of isolating nonperturbative parts that are to be identified and interpreted with these kinds of nonperturbative structures in mind. In the standard way of implementing the b∗/gb_*/g-function prescription, there is not an obvious or direct way to ensure that gj/h(x,bT)g_{j/h}(x,b_T) is subjected to such constraints.

c) In the usual way that the b∗b_*-method is implemented, the value of bmax⁡b_{\max} actually plays two separate roles simultaneously: i.) It is treated as the bTb_T above which transverse coordinate dependence is understood to become nonperturbative, and ii.) it is the bTb_T below which one transforms scales from Q0Q_0 to μb∗∼1/bT\mu_{b_*} \sim1/b_T. But these two transition points have different physical meanings. Transition i.) deals with the physically meaningful question of where and how the completely nonperturbative transverse coordinate dependence starts to become important. It is part of the nonperturbative physics that one hopes to model or extract in phenomenological applications. Transition ii.), by contrast, is an arbitrary scheme change, and in principle it has no effect on exact quantities. The details of how ii.) gets implemented are instead to be guided by the goal of optimizing convergence in perturbation theory calculations and minimizing higher orders.

There is no reason in principle that these two transition points need to be exactly the same. However, the b∗b_\ast-method, at least in the specific way that it is usually set up, makes them identical. This puts unnecessarily rigid constraints on how one describes the nonperturbative-to-perturbative transition.

One may try to avoid this by introducing two separate b∗b_\ast functions, with two separate bmax⁡b_{\max}'s. One of them can then be used to regulate the bTb_T argument at large bTb_T and the other can be used to construct μb∗\mu_{b_\ast}. But that exacerbates the problem in a) by putting multiple arbitrary functions and parameters into the collinear factorization parts. The problem cannot be completely avoided because it is a consequence of relying on the bTb_T-argument substitution method as a way to simultaneously regulate large bTb_T behavior where nonperturbative physics sets in and to transform to ∼1/bT\sim1/b_T scaling in the small bT≪1/Q0b_T \ll1/Q_0 limit.

These complications are ultimately connected to the fact that the usual b∗b_\ast organization of factors views bTb_T as coming from only two very sharply defined contributions, one that involves entirely perturbative bTb_T-dependence and one that is entirely nonperturbative bTb_T-dependence. However, a more realistic view is that there are three types of bTb_T-dependence:

  1. Totally nonperturbative behavior as bT→∞b_T \to\infty.

  1. Very reliable collinear factorization as bT≪1/Q0b_T \ll1/Q_0.

  1. An intermediate transition region around bT≈1/Q0b_T \approx1/Q_0 where bTb_T-dependence is reasonably well-described by perturbative collinear factorization, but is not as isolated from nonperturbative effects as region #2.

In an ideal TMD parametrization, region 3 above would be described by a physically motivated model that interpolates between the first two. However, the standard way of organizing the b∗b_\ast prescription leads one to assign all bTb_T-dependence only to one of either the first two categories. The practical consequences of this are possibly not very serious if QQ is so large that the gg-functions make a relatively small contribution that can be viewed as a correction to resummation calculations that only involves collinear pdfs. For this situation, it is possible that only the general qualitative features of the gg-functions are important.

However, it becomes more of an issue when extracting the details of nonperturbative transverse momentum dependence and separating them from perturbatively calculable parts is the main target goal. Of course, one way around the problem is simply to construct parametrizations of the gg-functions from the outset such that they explicitly evade a)-c) above, and this is essentially the approach we advocate below. (Although, an outcome is that most of the practical advantages of using a b∗b_\ast prescription and gg-function style of organization to start with are no longer present.)

The identification of the issues summarized above is not new. Our contribution is only to offer one strategy for systematizing their treatment. In particular, the problems that arise from nonperturbative model dependence affecting perturbative calculations have been discussed for a long time. References [10, 12], for example, deal with the problem by imposing continuity and smoothness conditions at the boundary between the perturbative and nonperturbative regions, thus preserving more of the predictive power coming from the collinear factorization parts of calculations.

Proposed modifications

In Refs. [1, 8, 9], we proposed a set of steps designed to evade issues a)-c) above. Since they are meant for applications where extracting the details of nonperturbative transverse momentum (the behavior that is normally contained entirely in the gg-functions) is the main objective, we have been calling it a “hadron structure oriented” (HSO) approach to TMD phenomenology. While Refs. [1, 8, 9] list many details, the overall basic strategy is simple to state:

  1. Construct a smooth and continuous parametrization of f~j/h(x,bT;μQ0,Q02)\widetilde{f}_{j/h}(x,b_{\mathrm{T}};\mu_{Q_0},Q_0^2) that directly interpolates between perturbative and nonperturbative bTb_{\mathrm{T}} descriptions for bT≈1/Q0b_{\mathrm{T}}\approx1/Q_0 and bT≫1/Q0b_{\mathrm{T}}\gg1/Q_0 respectively.3

  1. Impose the constraint in Eq. (20) on this parametrization.

  1. Ensure that the parametrization includes the usual scale transformation for the bT≪1/Q0b_{\mathrm{T}}\ll1/Q_0 region that transforms μQ0,Q0\mu_{Q_0},Q_0 to ∼1/bT\sim1/b_{\mathrm{T}}.

  1. Perform analogous steps for K~(bT;μQ0)\widetilde{K}(b_{\mathrm{T}};\mu_{Q_0}).

  1. Evolve to Q>Q0Q>Q_0 for comparing with higher scales.

Once this is done, the result can be transformed into the usual b∗b_*/gg-function construction if that is desired. The resulting gg-functions will no longer have the complications listed in a)-c) of the previous section.

Questions

The following questions arose in the context of the talk:

Can one use a bmin⁡b_{\min} regulator rather than a transverse momentum cutoff?

References [1, 8, 9] used a direct transverse momentum cutoff to regulate the integral in Eq. (20), with an input parametrization constructed in transverse momentum space. This was for various practical reasons, but it is not essential to the basic procedure, and other regulators are acceptable. Another common approach, especially for parametrizations constructed directly in bTb_{\mathrm{T}}-space, is to freeze bTb_{\mathrm{T}} below some bmin⁡b_{\min} to regulate the integral. It is instructive to repeat steps analogous to Secs. III and IV of Refs. [1], but now in bTb_{\mathrm{T}}-space directly and with a small bTb_{\mathrm{T}} cutoff to regulate the UV divergence rather than a large kTk_{\mathrm{T}} cutoff in transverse momentum space.

First, one defines the cutoff regulator prescription by specifying an arbitrary cutoff function,

bc(bT)={bTbT≳bmin⁡,bmin⁡bT≪bmin⁡,(21)b_c(b_T)= \begin{cases} b_T & b_T \gtrsim b_{\min},\\ b_{\min} & b_T \ll b_{\min}, \end{cases} \tag*{(21)}

with bmin⁡≈1/μQ0b_{\min}\approx1/\mu_{Q_0}. It could be any function with this property, but for concreteness let us choose

bc(bT)=bT2+bmin⁡2.(22)b_c(b_T)=\sqrt{b_T^2+b_{\min}^2}. \tag*{(22)}

Then the integral in Eq. (20) is regulated if we replace the bTb_T-space TMD pdf by one with the bTb_T argument replaced by bc(bT)b_c(b_T),

f~j/h(x,bT;μQ0;Q02)⟶f~j/h(x,bc(bT);μQ0;Q02).(23)\widetilde f_{j/h}(x,b_T;\mu_{Q_0};Q_0^2)\longrightarrow\widetilde f_{j/h}(x,b_c(b_T);\mu_{Q_0};Q_0^2). \tag*{(23)}

In bTb_T-space, step 1 of § 4 would be to construct a parametrization that interpolates between the logarithmically diverging perturbative collinear factorization prediction at small bTb_T and the nonperturbative behavior at large bT→∞b_T\to\infty. If we keep only the lowest nontrivial order at small bTb_T, this would mean we need

f~(x,bT;μQ0;Q02)param={Constant−Aln⁡(bTμQ0b0)−B[ln⁡2(bTμQ0b0)+ln⁡(bTμQ0b0)ln⁡(Q02μQ02)]bT≪1/ΛQCD,nonperturbative parametrizationbT→∞,(24)\widetilde f(x,b_T;\mu_{Q_0};Q_0^2)_{\mathrm{param}}= \begin{cases} \text{Constant}-A\ln\left(\dfrac{b_T\mu_{Q_0}}{b_0}\right)-B\left[\ln^2\left(\dfrac{b_T\mu_{Q_0}}{b_0}\right)+\ln\left(\dfrac{b_T\mu_{Q_0}}{b_0}\right)\ln\left(\dfrac{Q_0^2}{\mu_{Q_0}^2}\right)\right] & b_T\ll1/\Lambda_{\mathrm{QCD}},\\ \text{nonperturbative parametrization} & b_T\to\infty, \end{cases} \tag*{(24)}

where b0=2e−γEb_0=2e^{-\gamma_E}, AA and BB are known O(αs)O(\alpha_s) functions of xx calculated in collinear perturbation theory and the constant term is ultimately going to be fixed by imposing Eq. (20), which for the bmin⁡b_{\min} cutoff is

∫d2kT∫d2bT(2π)2eikT⋅bTf~(x,bc(bT);μQ0;Q02)param=f~(x,bmin⁡;μQ0;Q02)param=f(x;μQ0)−Δ.(25)\int d^2k_T\int\frac{d^2b_T}{(2\pi)^2}e^{i\mathbf{k}_T\cdot\mathbf{b}_T}\widetilde f(x,b_c(b_T);\mu_{Q_0};Q_0^2)_{\mathrm{param}}=\widetilde f(x,b_{\min};\mu_{Q_0};Q_0^2)_{\mathrm{param}}=f(x;\mu_{Q_0})-\Delta. \tag*{(25)}

An example of a reasonable parametrization that satisfies the above might be

f~(x,bT;μQ0;Q02)param=Ce−mabT2−Aln⁡(bTμQ0b0+mbbT)−B[ln⁡2(bTμQ0b0+mbbT)+ln⁡(bTμQ0b0+mbbT)ln⁡(Q02μQ02)]+Aln⁡(μQ0mb)+B[ln⁡2(μQ0mb)+ln⁡(μQ0mb)ln⁡(Q02μQ02)].(26)\begin{aligned} \widetilde f(x,b_T;\mu_{Q_0};Q_0^2)_{\mathrm{param}} ={}&Ce^{-m_a b_T^2}-A\ln\left(\frac{b_T\mu_{Q_0}}{b_0+m_b b_T}\right)-B\left[\ln^2\left(\frac{b_T\mu_{Q_0}}{b_0+m_b b_T}\right)+\ln\left(\frac{b_T\mu_{Q_0}}{b_0+m_b b_T}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]\\ &+A\ln\left(\frac{\mu_{Q_0}}{m_b}\right)+B\left[\ln^2\left(\frac{\mu_{Q_0}}{m_b}\right)+\ln\left(\frac{\mu_{Q_0}}{m_b}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]. \tag*{(26)} \end{aligned}

The nonperturbative model parameters are the mass scales mam_a and mbm_b, which are expected to be comparable in size to ΛQCD\Lambda_{\mathrm{QCD}}. The model vanishes as bT→∞b_T\to\infty, so the Fourier-Bessel transform to transverse momentum space exists as an ordinary function. CC is the constant term that will be fixed by imposing Eq. (25):

Ce−mabmin⁡2−Aln⁡(bmin⁡μQ0b0+mbbmin⁡)−B[ln⁡2(bmin⁡μQ0b0+mbbmin⁡)+ln⁡(bmin⁡μQ0b0+mbbmin⁡)ln⁡(Q02μQ02)]+Aln⁡(μQ0mb)+B[ln⁡2(μQ0mb)+ln⁡(μQ0mb)ln⁡(Q02μQ02)]=f(x;μQ0)−Δ.(27)\begin{aligned} Ce^{-m_a b_{\min}^2} &-A\ln\left(\frac{b_{\min}\mu_{Q_0}}{b_0+m_b b_{\min}}\right) -B\left[\ln^2\left(\frac{b_{\min}\mu_{Q_0}}{b_0+m_b b_{\min}}\right)+\ln\left(\frac{b_{\min}\mu_{Q_0}}{b_0+m_b b_{\min}}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]\\ &+A\ln\left(\frac{\mu_{Q_0}}{m_b}\right) +B\left[\ln^2\left(\frac{\mu_{Q_0}}{m_b}\right)+\ln\left(\frac{\mu_{Q_0}}{m_b}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right] =f(x;\mu_{Q_0})-\Delta. \tag*{(27)} \end{aligned}

which fixes

C=f(x;μQ0)−Δ+Aln⁡(bmin⁡μQ0b0)+B[ln⁡2(bmin⁡μQ0b0)+ln⁡(bmin⁡μQ0b0)ln⁡(Q02μQ02)]−Aln⁡(μQ0mb)−B[ln⁡2(μQ0mb)+ln⁡(μQ0mb)ln⁡(Q02μQ02)]+p.s.,(28)\begin{aligned} C={}&f(x;\mu_{Q_0})-\Delta+A\ln\left(\frac{b_{\min}\mu_{Q_0}}{b_0}\right)+B\left[\ln^2\left(\frac{b_{\min}\mu_{Q_0}}{b_0}\right)+\ln\left(\frac{b_{\min}\mu_{Q_0}}{b_0}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]\\ &-A\ln\left(\frac{\mu_{Q_0}}{m_b}\right)-B\left[\ln^2\left(\frac{\mu_{Q_0}}{m_b}\right)+\ln\left(\frac{\mu_{Q_0}}{m_b}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]+\text{p.s.}, \tag*{(28)} \end{aligned}

where the “p.s.” terms are O(bmin⁡ΛQCD)O\left(b_{\min}\Lambda_{\mathrm{QCD}}\right). Fixing CC to the leading power terms in Eq. (28) amounts to step 2 of § 4. Recall that we are considering the limit that ΛQCD/μQ0\Lambda_{\mathrm{QCD}}/\mu_{Q_0} is small, so the terms on the second line in Eq. (28) are not perturbatively suppressed even though the AA and BB factors are proportional to αs(μQ0)\alpha_s(\mu_{Q_0}). Setting bmin⁡=b0/μQ0b_{\min}=b_0/\mu_{Q_0} eliminates the logarithmic terms on the first line. Then, the full parametrization in Eq. (26) is

f~(x,bT;μQ0;Q02)param=e−mabT2{f(x;μQ0)−Δ−Aln⁡(μQ0mb)−B[ln⁡2(μQ0mb)+ln⁡(μQ0mb)ln⁡(Q02μQ02)]−Aln⁡(bTμQ0b0+mbbT)−B[ln⁡2(bTμQ0b0+mbbT)+ln⁡(bTμQ0b0+mbbT)ln⁡(Q02μQ02)]+Aln⁡(μQ0mb)+B[ln⁡2(μQ0mb)+ln⁡(μQ0mb)ln⁡(Q02μQ02)]}.(29)\begin{aligned} \widetilde{f}(x,b_{\mathrm{T}};\mu_{Q_0};Q_0^2)_{\mathrm{param}}={}&e^{-m_a b_{\mathrm{T}}^2}\Biggl\{f(x;\mu_{Q_0})-\Delta-A\ln\left(\frac{\mu_{Q_0}}{m_b}\right)-B\left[\ln^2\left(\frac{\mu_{Q_0}}{m_b}\right)+\ln\left(\frac{\mu_{Q_0}}{m_b}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]\\ &-A\ln\left(\frac{b_{\mathrm{T}}\mu_{Q_0}}{b_0+m_b b_{\mathrm{T}}}\right)-B\left[\ln^2\left(\frac{b_{\mathrm{T}}\mu_{Q_0}}{b_0+m_b b_{\mathrm{T}}}\right)+\ln\left(\frac{b_{\mathrm{T}}\mu_{Q_0}}{b_0+m_b b_{\mathrm{T}}}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]\\ &+A\ln\left(\frac{\mu_{Q_0}}{m_b}\right)+B\left[\ln^2\left(\frac{\mu_{Q_0}}{m_b}\right)+\ln\left(\frac{\mu_{Q_0}}{m_b}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]\Biggr\}. \tag*{(29)} \end{aligned}

Now the limit of bTΛQCD→0b_{\mathrm{T}}\Lambda_{\mathrm{QCD}}\to0 reproduces the usual well-known OPE expression,

f~(x,bT;μQ0;Q02)param∣bTΛQCD→0=f(x;μQ0)−Δ−Aln⁡(bTμQ0b0)−B[ln⁡2(bTμQ0b0)+ln⁡(bTμQ0b0)ln⁡(Q02μQ02)]+O((bTΛQCD)a),(30)\left.\widetilde{f}(x,b_{\mathrm{T}};\mu_{Q_0};Q_0^2)_{\mathrm{param}}\right|_{b_{\mathrm{T}}\Lambda_{\mathrm{QCD}}\to0} =f(x;\mu_{Q_0})-\Delta-A\ln\left(\frac{b_{\mathrm{T}}\mu_{Q_0}}{b_0}\right)-B\left[\ln^2\left(\frac{b_{\mathrm{T}}\mu_{Q_0}}{b_0}\right)+\ln\left(\frac{b_{\mathrm{T}}\mu_{Q_0}}{b_0}\right)\ln\left(\frac{Q_0^2}{\mu_{Q_0}^2}\right)\right]+O\left((b_{\mathrm{T}}\Lambda_{\mathrm{QCD}})^a\right), \tag*{(30)}

and Eq. (20) holds by construction with only O((bmin⁡ΛQCD)a)O\left((b_{\min}\Lambda_{\mathrm{QCD}})^a\right) errors. The large bTb_{\mathrm{T}} behavior in this model is essentially Gaussian. The constant terms in the OPE arise from the Δ\Delta term.

For dealing with the cross section behavior at Q≈Q0Q\approx Q_0 through O(αs(μQ0))O\left(\alpha_s(\mu_{Q_0})\right), this is sufficient. However, if we plan to evolve to Q≫Q0Q\gg Q_0, then the bT≪1/Q0b_{\mathrm{T}}\ll1/Q_0 region also needs to be properly accounted for. To deal with this, it is straightforward to perform another scale transformation from μQ0,Q0\mu_{Q_0},Q_0 to ∼1/bT\sim1/b_{\mathrm{T}} using the evolution equations in Eqs. (3)–(5) again. Since this transformation takes place in the bT≲1/Q0b_{\mathrm{T}}\lesssim1/Q_0 region of coordinate space, it is under control perturbatively and the errors it introduces can be minimized with detailed treatments of higher orders in perturbation theory.

If it is preferred, one may now take this parametrization and construct a gj/h(x,bT)g_{j/h}(x,b_{\mathrm{T}}) function just by substituting it into Eqs. (11)–(13),

f~(x,bT;μQ0;Q02)param=f~j/h(x,b∗;μQ0,Q02)parame−gj/h(x,bT)param(31)\widetilde{f}(x,b_{\mathrm{T}};\mu_{Q_0};Q_0^2)_{\mathrm{param}}=\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2)_{\mathrm{param}}e^{-g_{j/h}(x,b_{\mathrm{T}})_{\mathrm{param}}} \tag*{(31)}

with

gj/h(x,bT)param=−ln⁡(f~j/h(x,bT;μQ0,Q02)paramf~j/h(x,b∗;μQ0,Q02)param).(32)g_{j/h}(x,b_T)_{\mathrm{param}}=-\ln\left(\frac{\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q_0^2)_{\mathrm{param}}}{\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2)_{\mathrm{param}}}\right). \tag*{(32)}

However, now this gj/h(x,bT)paramg_{j/h}(x,b_T)_{\mathrm{param}} evades all the complications listed in a)-c) of § 3. In particular, there is no disadvantage in making bmax⁡b_{\max} conservatively very small, bmax⁡≲1/Q0b_{\max}\lesssim1/Q_0. The consequence of doing so is only that a significant part of the perturbative transverse coordinate dependence is handled by gj/h(x,bT)paramg_{j/h}(x,b_T)_{\mathrm{param}}. Therefore, one may ensure that the power suppressed O(ΛQCDbmax⁡)aO\left(\Lambda_{\mathrm{QCD}}b_{\max}\right)^a terms in Eq. (16) are negligible without sacrificing the predictive power coming from collinear factorization. The b∗b_*-independence in item 1 of § 2 is exactly realized in a gj/h(x,bT)paramg_{j/h}(x,b_T)_{\mathrm{param}} constructed through steps like the above.

It is instructive to contrast this with the kind of result that would be obtained if one followed more typical steps. Let us label such a result as “param,2.” There, one writes Eq. (12) with f~j/h(x,b∗;μQ0,Q02)\widetilde{f}_{j/h}(x,b_*;\mu_{Q_0},Q_0^2) immediately replaced by its OPE expression,

f~j/h(x,bT;μQ0,Q02)param,2=f~OPE,j/h(x,b∗;μQ0,Q02)e−gj/h(x,bT)param,2,(33)\widetilde{f}_{j/h}(x,b_T;\mu_{Q_0},Q_0^2)_{\mathrm{param,2}}=\widetilde{f}_{\mathrm{OPE},j/h}(x,b_*;\mu_{Q_0},Q_0^2)e^{-g_{j/h}(x,b_T)_{\mathrm{param,2}}}, \tag*{(33)}

and gj/h(x,bT)param,2g_{j/h}(x,b_T)_{\mathrm{param,2}} is replaced with a simple ansatz. A typical example might be a quadratic power law,

−gj/h(x,bT)param,2=−ma2bT2.(34)-g_{j/h}(x,b_T)_{\mathrm{param,2}}=-m_a^2b_T^2. \tag*{(34)}

Now the b∗b_*-independence is clearly not preserved to very high accuracy because this would require at least that

e−gj/h(x,bT)param,2≈−ln⁡(f~OPE,j/h(x,bT;μQ0,Q02)f~OPE,j/h(x,b∗;μQ0,Q02))(35)e^{-g_{j/h}(x,b_T)_{\mathrm{param,2}}}\approx-\ln\left(\frac{\widetilde{f}_{\mathrm{OPE},j/h}(x,b_T;\mu_{Q_0},Q_0^2)}{\widetilde{f}_{\mathrm{OPE},j/h}(x,b_*;\mu_{Q_0},Q_0^2)}\right) \tag*{(35)}

in the region of small bT≈bmax⁡b_T\approx b_{\max}. For a power law gj/h(x,bT)g_{j/h}(x,b_T), however, such a correspondence can only be a rough approximation within some narrow range of moderate bTb_T.

Furthermore, applying Eq. (25) to Eq. (33) gives

∫d2kT∫d2bT(2π)2eikT⋅bTf~(x,bc(bT);μQ0;Q02)param,2=f~OPE,j/h(x,b∗(bT=bmin⁡);μQ0,Q02)e−ma2bmin⁡2=f~OPE,j/h(x,bmin⁡;μQ0,Q02)+O(bmin⁡2bmax⁡2)+O(ΛQCD2bmin⁡2)=f(x;μQ0)−Δ+O(bmin⁡2bmax⁡2)+O(ΛQCD2bmin⁡2).(36)\begin{aligned} \int d^2k_T\int\frac{d^2b_T}{(2\pi)^2}e^{i\boldsymbol{k}_T\cdot\boldsymbol{b}_T}\widetilde{f}\left(x,b_c(b_T);\mu_{Q_0};Q_0^2\right)_{\mathrm{param,2}} &=\widetilde{f}_{\mathrm{OPE},j/h}(x,b_*(b_T=b_{\min});\mu_{Q_0},Q_0^2)e^{-m_a^2b_{\min}^2}\\ &=\widetilde{f}_{\mathrm{OPE},j/h}(x,b_{\min};\mu_{Q_0},Q_0^2)+O\left(\frac{b_{\min}^2}{b_{\max}^2}\right)+O\left(\Lambda_{\mathrm{QCD}}^2b_{\min}^2\right)\\ &=f(x;\mu_{Q_0})-\Delta+O\left(\frac{b_{\min}^2}{b_{\max}^2}\right)+O\left(\Lambda_{\mathrm{QCD}}^2b_{\min}^2\right). \tag*{(36)} \end{aligned}

However, near the input scale both bmin⁡b_{\min} and bmax⁡b_{\max} are O(1/Q0)O(1/Q_0), so the O(bmin⁡2/bmax⁡2)O\left(b_{\min}^2/b_{\max}^2\right) term is not a power suppressed correction (and so the expansion in bmin⁡/bmax⁡b_{\min}/b_{\max} is not valid). Therefore, Eq. (25) fails in general for Eq. (33). To be preserved, the gj/h(x,bT)g_{j/h}(x,b_T) function needs to be constructed so as to cancel the O(bmin⁡2/bmax⁡2)O\left(b_{\min}^2/b_{\max}^2\right) behavior.

After evolution to Q≫Q0Q\gg Q_0, the bmin⁡b_{\min} becomes ∼1/Q\sim1/Q while bmax⁡b_{\max} remains ∼1/Q0\sim1/Q_0. Then, the O(bmin⁡2/bmax⁡2)O\left(b_{\min}^2/b_{\max}^2\right) error terms are O(Q02/Q2)O\left(Q_0^2/Q^2\right), which are eventually negligible as QQ increases. However, much of the data of interest for studies of hadron structure are in the region of Q≈Q0Q\approx Q_0.

This example of a model parametrization, Eq. (29), is only meant for illustration purposes and others are probably much better in practice. The point of this discussion is not the details of the parametrization itself, but rather the general steps for constructing it.

Is Eq. (20) automatically satisfied within the standard organization?

Equation (20) is obviously not guaranteed to be satisfied for arbitrary functions substituted into the integrand, given a fixed extraction for an MS‾\overline{\mathrm{MS}} collinear pdf on the left-hand side. It will be satisfied eventually for large enough QQ for any parametrization that matches to the OPE as bT→0b_{\mathrm{T}} \to0. That includes standard existing parametrizations that use the usual b∗b_{*}-procedure as described in § 2. However, the reason is that at large enough QQ the sensitivity of the kTk_{\mathrm{T}}-integral to the gg-functions eventually vanishes. Specifically, the version of Eq. (20) that arises automatically in a standard b∗b_{*} implementation has the form,

f(x;μQ)=∫reg.d2kT f(x,kT;μQ,μQ2)+Δ+O((Q0Q)a),a>0.(37)f(x;\mu_{Q}) = \int_{\mathrm{reg.}} \mathrm{d}^{2}k_{\mathrm{T}}\,f(x,k_{\mathrm{T}};\mu_{Q},\mu_{Q}^{2})+\Delta+O\left(\left(\frac{Q_{0}}{Q}\right)^{a}\right), \qquad a>0. \tag*{(37)}

In other words, the power suppressed error terms are in the form of ratios of the input scale to the hard scale, so the relation fails for applications at Q≈Q0Q \approx Q_{0}, which is where sensitivity to nonperturbative parts is the greatest. As the illustrative example above shows, this is the result if one does not incorporate the gg-functions along with the perturbative parts in the integral condition.

Intuitively, the reason it eventually holds for Q≫Q0Q \gg Q_{0} is that the integrand in Eq. (37) becomes completely dominated by a perturbative tail whose description only involves collinear pdfs and none of the sensitivity to nonperturbative transverse momentum dependence. Figure 2 provides a schematic visualization of this. The curve is meant to represent a TMD pdf plotted versus kTk_{\mathrm{T}}, and the shaded blue and yellow areas are meant to represent the contributions from nonperturbative and perturbative transverse momentum dependence respectively to the integral over kTk_{\mathrm{T}}, assuming a regulator that removes the kT≳Qk_{\mathrm{T}} \gtrsim Q region. As QQ grows, the contribution from the yellow part increases until there is essentially no non-negligible contribution from the blue part. That is, the area under the curve is completely dominated by behavior that is describable from the outside in terms of collinear pdfs.

Figure 2. Schematic representation of the relative contributions from perturbative (yellow) and nonperturbative (blue) transverse momentum dependence to the kTk_{\mathrm{T}}-integral of a TMD pdf. For large enough QQ, the yellow contribution dominates.

But situations where nonperturbative transverse momentum plays no role are the opposite of what one is seeking in the context of nonperturbative hadron structure applications. While it might turn out that for specific large values of QQ it is the case that Q0/QQ_0/Q is small enough that the error terms in Eq. (37) are negligible, assuming this is true from the outset restricts one to considering only conservatively very large QQ, and it contradicts the goal of extracting the nonperturbative transverse momentum dependence associated with bound states or hadronization. Furthermore, it discards a valuable consistency constraint on the gg-functions, which could be important for the kT≪Qk_T \ll Q region of the differential cross section even at larger QQ. What is needed for applications where the details of the gg-functions are important is a version of Eq. (37) of the form

f(x;μQ)=∫reg.d2kT f(x,kT;μQ,μQ2)+Δ+O((ΛQCDQ0)a),a>0(38)f(x;\mu_Q) = \int_{\mathrm{reg.}} \mathrm{d}^{2}k_T\,f(x,k_T;\mu_Q,\mu_Q^2) + \Delta+ O\left(\left(\frac{\Lambda_{\mathrm{QCD}}}{Q_0}\right)^a\right), \qquad a > 0 \tag*{(38)}

so that the integrals relating TMD and collinear pdfs hold even for Q≈Q0Q \approx Q_0. This is not automatically satisfied for arbitrary parametrizations of gj/h(x,bT)g_{j/h}(x,b_T) but rather needs to be imposed on them directly. For example, it is satisfied by construction in the example in Eq. (29) but not in Eq. (33).

Examples in toy model field theories are also instructive for illustrating the effect. In Ref. [2], for example, illustrative examples with a scalar Yukawa theory found that significant violations of Eq. (38) with a cutoff regulator result if the standard gg-function ansatz method is used.

What is the difference with the b∗b_* approach?

Both the more traditional organization of TMD factorization and the approach we have adopted in § 4 involve a scheme transformation to ∼1/bT\sim1/b_T behavior at bT≪1/Q0b_T \ll1/Q_0. See, for example, the function named Q‾0(bT,a)\overline{Q}_0(b_T,a) in Eq. (77) of Ref. [1], which is similar to Eq. (15). This can give the appearance that there is no real difference between the two methods.

The significant difference is related to issue c) of § 3. Recall that in the example of § 5.1 the nonperturbative transverse momentum dependence is controlled entirely by the model parameters mam_a and mbm_b. Changing the scales to ∼1/bT\sim1/b_T is left as a separate step, and can be chosen to occur at much large values of 1/bT1/b_T than mam_a or mbm_b. The form of the function used in the scale transformation, and the value of bTb_T where it is implemented, do not need to be connected in any way with how one models the nonperturbative transverse momentum dependence as bT→∞b_T \to\infty. This separation is the main difference from the usual approach.

Contrast this with a parametrization like Eqs. (33)–(35). If, in attempting to make O(αs(μb∗)n)O(\alpha_s(\mu_{b_*})^n) and O(ΛQCDbmax⁡)aO(\Lambda_{\mathrm{QCD}}b_{\max})^a errors as small as possible, one makes bmax⁡b_{\max} too small, then one will be forced to describe even the perturbatively describable bTb_T-dependence with a nonperturbative ansatz. If, in attempting to widen the range that is described perturbatively as far as possible, one makes bmax⁡b_{\max} too large, then αs(μb∗)\alpha_s(\mu_{b_*}) becomes a nonperturbative large coupling and ΛQCDbmax⁡\Lambda_{\mathrm{QCD}}b_{\max} errors stop being negligible. Notice the practical difficulties that this trade-off creates for disentangling truly nonperturbative (bT>1/Q0b_T > 1/Q_0) from perturbative (bT≪1/Q0b_T \ll1/Q_0) transverse momentum dependence.

We argue that such a trade-off is not actually necessary and is only an artifact of using an overly restrictive organizational scheme.

Is it useful to keep the TMD pdfs evaluated at a bc(bT)b_c(b_T)?

The TMD pdfs in the MS‾\overline{\mathrm{MS}} scheme and with fixed μ=ζ∼Q0\mu= \sqrt{\zeta} \sim Q_0 are determined by their operator definitions for all bTb_T, and therefore, from the evolution equations, for all higher QQ. In particular, assuming that the MS‾\overline{\mathrm{MS}} pdfs are known, the small bTb_T (or large-kTk_T) parts are uniquely determined by the OPE, and this rigidity is an important part of the predictive power of the overall formalism. If a replacement like Eq. (23) is made and kept, then the TMD pdfs no longer follow the OPE and they no longer match their operator definition at a given value of bTb_T. This can be seen directly in the example above. In Eq. (29), if the replacement bT→bc(bT)=bT2+bmin⁡2b_T \to b_c(b_T) = \sqrt{b_T^2+b_{\min}^2} is made, then a TMD pdf does not transform into its OPE expression as bT→0b_T \to0, but rather freezes at its value around bT≈bmin⁡b_T \approx b_{\min}. Another way to see it directly is by considering the Fourier-Bessel transforms of common functional forms. The logarithms of the OPE in bTb_T-space translate into (1/kT2)×(1/k_T^2)\timeslogarithms in kTk_T-space:

ln⁡(bTQ2e−γE)↦−12πkT2,ln⁡2(bTQ2e−γE)↦−12πkT2ln⁡(Q2kT2)(39)\ln\left(\frac{b_TQ}{2e^{-\gamma_E}}\right) \mapsto-\frac{1}{2\pi k_T^2}, \qquad\ln^2\left(\frac{b_TQ}{2e^{-\gamma_E}}\right) \mapsto-\frac{1}{2\pi k_T^2}\ln\left(\frac{Q^2}{k_T^2}\right) \tag*{(39)}

whereas with the bc(bT)b_c(b_T) in place

ln⁡(Q2e−γEbT2+bmin⁡2)↦−bmin⁡2πkTK1(bmin⁡kT),\ln\left(\frac{Q}{2e^{-\gamma_E}}\sqrt{b_T^2+b_{\min}^2}\right) \mapsto-\frac{b_{\min}}{2\pi k_T}K_1(b_{\min}k_T),
ln⁡2(Q2e−γEbT2+bmin⁡2)↦−12πkT2[K0(bmin⁡kT)+kTbmin⁡ln⁡(bmin⁡Q22e−γEkT)K1(bmin⁡kT)](40)\ln^2\left(\frac{Q}{2e^{-\gamma_E}}\sqrt{b_T^2+b_{\min}^2}\right) \mapsto-\frac{1}{2\pi k_T^2}\left[K_0(b_{\min}k_T)+k_Tb_{\min}\ln\left(\frac{b_{\min}Q^2}{2e^{-\gamma_E}k_T}\right)K_1(b_{\min}k_T)\right] \tag*{(40)}

which only matches Eq. (39) in the limit that kT≪1/bmin⁡k_T \ll1/b_{\min}. For larger QQ, one would typically choose a bmin⁡∼1/Qb_{\min} \sim1/Q. Then, it could be argued that the only physically relevant region of kTk_T is anyway kT≪1/bmin⁡k_T \ll1/b_{\min}, so the bT→bc(bT)b_T \to b_c(b_T) replacement should be irrelevant. However, the replacement nevertheless allows another adjustable arbitrary parameter to influence the small-to-large transverse momentum transition, so our view is that it should be avoided.

In earlier work [3], I advocated a procedure that did involve keeping the bT→bc(bT)b_T \to b_c(b_T) replacement as described above in the hopes that having the freedom to shift contributions between the WW-term and the YY-term would lead to a more refined treatment of large transverse momentum dependence in the cross section. However, because of the increase in adjustable parameters in the large transverse momentum region and the associated loss of predictive power, I no longer view this as an optimal approach.

Acknowledgments

T. Rogers and T. Rainaldi were supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics, under Award Number DE-SC0024715. F. Aslan was suppoprted by the Center for Nuclear Femtography, Southeastern Universities Research Association, Washington, D.C.. T. Rainaldi, T. Rogers, and A. Simonelli were supported by the DOE Contract No. DE-AC05-06OR23177, under which Jefferson Science Associates, LLC operates Jefferson Lab. M. Boglione and J.O. Gonzalez-Hernandez acknowledge funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 824093 (STRONG2020), and from the European Union “Next Generation EU” program through the Italian PRIN 2022 grant n. 20225ZHA7W.

References

  1. [1]F. Aslan, M. Boglione, J. O. Gonzalez-Hernandez, T. Rainaldi, T. C. Rogers, and A. Simonelli. Phenomenology of TMD parton distributions in Drell-Yan and Z⁰ boson production in a hadron structure oriented approach. 1 2024. arXiv:2401.14266.arxiv.org/abs/2401.14266
  2. [2]Fatma Aslan, L. Gamberg, J. O. Gonzalez-Hernandez, T. Rainaldi, and T. C. Rogers. Basics of factorization in a scalar Yukawa field theory. Phys. Rev. D, 107(7):074031, 2023. arXiv:2212.00757, doi:10.1103/PhysRevD.107.074031.DOI
  3. [3]J. Collins, L. Gamberg, A. Prokudin, T. C. Rogers, N. Sato, and B. Wang. Relating Transverse Momentum Dependent and Collinear Factorization Theorems in a Generalized Formalism. Phys. Rev. D, 94(3):034014, 2016. arXiv:1605.00671, doi:10.1103/PhysRevD.94.034014.DOI
  4. [4]J. C. Collins. Foundations of Perturbative QCD. Cambridge University Press, Cambridge, 2011.
  5. [5]J. C. Collins, D. E. Soper, and G. Sterman. Transverse momentum distribution in Drell-Yan pair and W and Z boson production. Nucl. Phys., B250:199–224, 1985. doi:10.1016/0550-3213(85)90479-1.DOI
  6. [6]John C. Collins and Davison E. Soper. Back-to-back jets: Fourier transform from b to k_T. Nucl. Phys., B197:446–476, 1982. doi:10.1016/0550-3213(82)90453-9.DOI
  7. [7]Oscar del Rio, Alexei Prokudin, Ignazio Scimemi, and Alexey Vladimirov. Transverse momentum moments. Phys. Rev. D, 110(1):016003, 2024. arXiv:2402.01836, doi:10.1103/PhysRevD.110.016003.DOI
  8. [8]J. O. Gonzalez-Hernandez, T. Rainaldi, and T. C. Rogers. Resolution to the problem of consistent large transverse momentum in TMDs. Phys. Rev. D, 107(9):094029, 2023. arXiv:2303.04921, doi:10.1103/PhysRevD.107.094029.
  9. [9]J. O. Gonzalez-Hernandez, T. C. Rogers, and N. Sato. Combining nonperturbative transverse momentum dependence with TMD evolution. Phys. Rev. D, 106(3):034002, 2022. arXiv:2205.05750, doi:10.1103/PhysRevD.106.034002.
  10. [10]Manvir Grewal, Zhong-Bo Kang, Jian-Wei Qiu, and Andrea Signori. Predictive power of transverse-momentum-dependent distributions. Phys. Rev. D, 101(11):114023, 2020. arXiv:2003.07453, doi:10.1103/PhysRevD.101.114023.DOI
  11. [11]Cedric Lorce, Barbara Pasquini, and Marc Vanderhaeghen. Unified framework for generalized and transverse-momentum dependent parton distributions within a 3Q light-cone picture of the nucleon. JHEP, 05:041, 2011. arXiv:1102.4704, doi:10.1007/JHEP05(2011)041.DOI
  12. [12]Jianwei Qiu and Xiao-Fei Zhang. QCD prediction for heavy boson transverse momentum distributions. Phys. Rev. Lett., 86:2724–2727, 2001. arXiv:hep-ph/0012058, doi:10.1103/PhysRevLett.86.2724.DOI
  13. [13]Ted Rogers. Transverse moments of TMD parton densities and ultraviolet divergences. Mod. Phys. Lett. A, 35(37):2030021, 2020. arXiv:2008.05351, doi:10.1142/S0217732320300219.DOI

Paper details

vlnc:W_jnu6nf2x
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents