id
string | text
string |
|---|---|
0704.0001
|
Title: Calculation of prompt diphoton production cross sections at Tevatron and
LHC energies
Abstract: A fully differential calculation in perturbative quantum chromodynamics is
presented for the production of massive photon pairs at hadron colliders. All
next-to-leading order perturbative contributions from quark-antiquark,
gluon-(anti)quark, and gluon-gluon subprocesses are included, as well as
all-orders resummation of initial-state gluon radiation valid at
next-to-next-to-leading logarithmic accuracy. The region of phase space is
specified in which the calculation is most reliable. Good agreement is
demonstrated with data from the Fermilab Tevatron, and predictions are made for
more detailed tests with CDF and DO data. Predictions are shown for
distributions of diphoton pairs produced at the energy of the Large Hadron
Collider (LHC). Distributions of the diphoton pairs from the decay of a Higgs
boson are contrasted with those produced from QCD processes at the LHC, showing
that enhanced sensitivity to the signal can be obtained with judicious
selection of events.
Body: \newcommand{\barl}{\bar{\lambda}} \newcommand{\barp}{\bar{p}} \preprint{ANL-HEP-PR-07-12, arXiv:0704.0001} \title{Calculation of prompt diphoton production cross sections at Tevatron and LHC energies} \author{C. Bal\'{a}zs$^{1}$} \thanks{balazs@hep.anl.gov; Current address: School of Physics, Monash University, Melbourne VIC 3800, Australia} \author{E.~L.~Berger$^{1}$} \thanks{berger@anl.gov} \author{P. Nadolsky$^{1}$} \thanks{nadolsky@hep.anl.gov} \author{C.-P. Yuan$^{2}$} \thanks{yuan@pa.msu.edu} \affiliation{$^{1}$High Energy Physics Division, Argonne National Laboratory, Argonne, IL 60439 \\ $^{2}$Department of Physics and Astronomy, Michigan State University, East Lansing, MI 48824} \begin{abstract} A fully differential calculation in perturbative quantum chromodynamics is presented for the production of massive photon pairs at hadron colliders. All next-to-leading order perturbative contributions from quark-antiquark, gluon-(anti)quark, and gluon-gluon subprocesses are included, as well as all-orders resummation of initial-state gluon radiation valid at next-to-next-to-leading logarithmic accuracy. The region of phase space is specified in which the calculation is most reliable. Good agreement is demonstrated with data from the Fermilab Tevatron, and predictions are made for more detailed tests with CDF and D\O~data. Predictions are shown for distributions of diphoton pairs produced at the energy of the Large Hadron Collider (LHC). Distributions of the diphoton pairs from the decay of a Higgs boson are contrasted with those produced from QCD processes at the LHC, showing that enhanced sensitivity to the signal can be obtained with judicious selection of events. \end{abstract} \date{May 3, 2007} \pacs{12.15.Ji, 12.38 Cy, 13.85.Qk } \keywords{prompt photons; all-orders resummation; hadron collider phenomenology; Higgs boson; LHC} \maketitle \section{Introduction} The long-sought Higgs boson(s) $h$ of electroweak symmetry breaking in particle physics may soon be observed at the CERN Large Hadron Collider (LHC) through the diphoton decay mode ($h\rightarrow\gamma\gamma$). Purely hadronic standard model processes are a copious source of diphotons, and a narrow Higgs boson signal at relatively low masses will appear as a small peak above this considerable background. A precise theoretical understanding of the kinematic distributions for diphoton production in the standard model could provide valuable guidance in the search for the Higgs boson signal and assist in the important measurement of Higgs boson coupling strengths. In this paper we address the theoretical calculation of the invariant mass, transverse momentum, rapidity, and angular distributions of continuum diphoton production in proton-antiproton and proton-proton interactions at hadron collider energies. We compute all contributions to diphoton production from parton-parton subprocesses through next-to-leading order (NLO) in perturbative quantum chromodynamics (QCD). These higher-order contributions are large at the LHC, and their inclusion is mandatory for quantitatively trustworthy predictions. We resum initial-state soft and collinear logarithmic terms associated with gluon radiation to all orders in the strong coupling strength $\alpha_{s}$. This resummation is essential for physically meaningful predictions of the transverse momentum ($Q_{T}$) distribution of the diphotons at small and intermediate values of $Q_{T}$, where the cross section is large. In addition, we analyze the final-state collinearly-enhanced contributions, also known as `fragmentation' contributions, in which one or both photons are radiated from final-state partonic constituents. We compare the results of our calculations with data on isolated diphoton production from the Fermilab Tevatron~. The good agreement we obtain with the Tevatron data adds confidence to our predictions at the energy of the LHC. The present work expands on our recent abbreviated report~, and it may be read in conjunction with our detailed treatment of the contributions from the gluon-gluon subprocess~. Our attention is focused on the production of isolated photons, \emph{i.e.}, high-energy photons observed at some distance from appreciable hadronic remnants in the particle detector. The rare isolated photons tend to originate directly in hard QCD scattering, in contrast to copiously produced non-isolated photons that arise from nonperturbative processes such as $\pi$ and $\eta$ decays, or from via quasi-collinear radiation off final-state quarks and gluons. We evaluate contributions to continuum diphoton production from the basic short-distance channels for $\gamma\gamma$ production initiated by quark-antiquark and (anti)quark-gluon scattering, as well as by gluon-gluon and gluon-(anti)quark scattering proceeding through a fermion-loop diagram. At lowest order in QCD, a photon pair is produced from $q\bar{q}$ annihilation {[}Fig.~ (a)]. Representative next-to-leading order (NLO) contributions to $q\bar{q}+qg$ scattering are shown in Fig.~ (b)-(e). They are of ${\mathcal{O}}(\alpha_{s})$ in the strong coupling strength~. Production of $\gamma\gamma$ pairs via a box diagram in $gg$ scattering {[}Fig.~ (h)] is suppressed by two powers of $\alpha_{s}$ compared to the lowest-order $q\bar{q}$ contribution, but it is enhanced by a product of two large gluon parton distribution functions (PDFs) if typical momentum fractions $x$ are small~. The ${\mathcal{O}}(\alpha_{s}^{3})$ or NLO corrections to $gg$ scattering include one-loop $gg\rightarrow\gamma\gamma g$ diagrams (i) and (j) derived in Ref.~, as well as 4-leg two-loop diagrams (l) computed in Ref.~. In this study we also include subleading contributions from the process (k), $gq_{S}\rightarrow\gamma\gamma q_{S}$ via the quark loop, where $q_{S}=\sum_{i=u,d,s,...}(q_{i}+\bar{q}_{i})$ denotes the flavor-singlet combination of quark scattering channels. Factorization is a central principle of hadronic calculations in perturbative QCD, in which a high-energy scattering cross section is expressed as a convolution of a perturbative partonic cross section with nonperturbative parton distribution functions (PDFs), thus separating short-distance from long-distance physics. The common factorization is a longitudinal notion, in the sense that the convolution is an integral over longitudinal momentum fractions, even if some partons in the hard-scattering process have transverse momenta that border the nonperturbative regime. Unphysical features may then arise in the transverse momentum ($Q_{T}$) distribution of a color-neutral object with high invariant mass ($Q$), such as a pair of photons produced in hadron-hadron collisions. When calculated in the common factorization approach at any finite order in perturbation theory, this distribution diverges as $Q_{T}\rightarrow0$, signaling that infrared singularities associated with $Q_{T}\rightarrow0$ have not been properly isolated and regulated. These singularities are associated with soft and collinear radiation from initial-state partons shown by the diagrams in Figs.~ (b), (d), and (i). A generalized factorization approach that correctly describes the small-$Q_{T}$ region was developed by Collins, Soper, and Sterman (CSS)~ and applied to photon pair production~. In this approach the hadronic cross section is expressed as an integral over the transverse coordinate (impact parameter). The integrable singular functions present in the finite-order differential distribution as $Q_{T}\rightarrow0$ are resummed, to all orders in the strong coupling $\alpha_{s}$, into a Sudakov exponent, and a well-behaved cross section is obtained for all $Q_{T}$ values. As explained in Sec.~II, our resummed calculation is accurate to next-to-next-to-leading-logarithmic (NNLL) order. It is applicable for values of diphoton transverse momentum that are less than the diphoton mass, i.e., for $Q_{T}<Q$. When $Q_{T}\sim Q$, terms of the form $\ln^{n}(Q_{T}/Q)$ become small. A perturbative expansion with a single hard scale is then applicable, and the cross section can be obtained from finite-order perturbation theory. In addition to the initial-state logarithmic singularities, there is a set of important final-state singularities which arise in the matrix elements when at least one photon's momentum is collinear to the momentum of a final-state parton. They are sometimes referred to as `fragmentation' singularities. At lowest order in $\alpha_{s}$, the final-state singularity appears only in the $qg\rightarrow\gamma\gamma q$ diagrams, as in Fig.~ (e). There are various methods used in the literature to deal with the final-state singularity, including the introduction of explicit fragmentation functions $D_{\gamma}(z)$ for hard photon production, where $z$ is the light-cone fraction of the intermediate parton's momentum carried by the photon. These single-photon ``one-fragmentation'' and ``two-fragmentation'' contributions, corresponding to one or both photons produced in independent fragmentation processes, are illustrated by the diagrams in Figs.~ (f) and (g). In addition, a fragmentation contribution of entirely different nature arises when the $\gamma\gamma$ pair is relatively light and produced from fragmentation of {\it one} parton, as discussed in Secs.~ and . A full and consistent treatment of the final-state logarithms beyond lowest order would require a joint resummation of the initial- and final-state logarithmic singularities. In the work reported here, we are guided by our interest in describing the cross section for \emph{isolated} photons, in which the fragmentation contributions are largely suppressed. A typical isolation condition requires the hadronic activity to be minimal (e.g., comparable to the underlying event) in the immediate neighborhood of each candidate photon. Candidate photons can be rejected by energy deposit nearby in the hadronic calorimeter or the presence of hadronic tracks near the photons. A theory calculation may approximate the experimental isolation by requiring the full energy of the hadronic remnants to be less than a threshold {}``isolation energy'' $E_{T}^{iso}$ in a cone of size $\Delta R$ around each photon. The two photons must be also separated in the plane of the rapidity $\eta$ and azimuthal angle $\varphi$ by an amount exceeding the resolution $\Delta R_{\gamma\gamma}$ of the detector. The values of $E_{T}^{iso},$ $\Delta R$, and $\Delta R_{\gamma\gamma}$ serve as crude characteristics of the actual measurement. The magnitude of the final-state fragmentation contribution depends on the assumed values of $E_{T}^{iso},$ $\Delta R$, and $\Delta R_{\gamma\gamma}$. An additional complication arises when the fragmentation radiation is assumed to be exactly collinear to the photon's momentum, as implied by the photon fragmentation functions $D_{\gamma}(z)$. The collinear approximation constrains from below the values of $z$ accessible to $D_{\gamma}(z)$: $z>z_{min}$. The size of the fragmentation contribution may depend strongly on the values of $E_{T}^{iso}$ and $z_{min}$ as a consequence of rapid variation of $D_{\gamma}(z)$ with $z$. In our work we treat the final-state singularity using a prescription that reproduces desirable features of the isolated cross sections while bypassing some of the technical difficulties alluded to above. For $Q_{T}>E_{T}^{iso}$, we avoid the final-state collinear singularity in the $qg$ scattering channel by applying quasi-experimental isolation. When $Q_{T}<E_{T}^{iso}$, we apply an auxiliary regulator which approximates on average the full NLO rate from direct $qg$ and fragmentation cross sections in this $Q_{T}$ range. Two prescriptions for the auxiliary regulator (subtraction and smooth-cone isolation inside the photon's isolation cone) are considered and lead to similar predictions at the Tevatron and the LHC. We begin with our notation in Sec.~, followed by an overview of the procedure for resummation of initial-state multiple parton radiation in Sec.~. The issue of the final-state fragmentation singularity is discussed in Sec.~. Our approach is compared with that of the DIPHOX calculation~, in which explicit fragmentation function contributions are included at NLO, but all-orders resummation is not performed. Our theoretical framework is summarized in Sec.~. In Sec.~ we compare the predictions of our resummation calculation with Tevatron data. Resummation is shown to be important for the successful description of physical $Q_{T}$ distributions, as well as for stable estimates of the effects of experimental acceptance on distributions in the diphoton invariant mass. We compare our results with the DIPHOX calculation~ and demonstrate that the requirement $Q_{T}<Q$ further suppresses the effects of the final-state fragmentation contribution, beyond the reduction associated with isolation. Next, we present our predictions for distributions of diphoton pairs produced at the energy of the LHC. Various distributions of the diphoton pairs produced from the decay of a Higgs boson are contrasted with those produced from QCD continuum processes at the LHC, showing that enhanced sensitivity to the signal can be obtained with judicious event selection. Our conclusions are presented in Sec.~. \section{Theory overview } \subsection{Notation } We consider the scattering process $h_{1}(P_{1})+h_{2}(P_{2})\rightarrow\gamma(P_{3})+\gamma(P_{4})+X$, where $h_{1}$ and $h_{2}$ are the initial-state hadrons. In terms of the center-of-mass collision energy $\sqrt{S}$, the invariant mass $Q,$ transverse momentum $Q_{T}$, and rapidity $y$ of the $\gamma\gamma$ pair, the laboratory frame momenta $P_{1}^{\mu}$ and $P_{2}^{\mu}$ of the initial hadrons and $q^{\mu}\equiv P_{3}^{\mu}+P_{4}^{\mu}$ of the $\gamma\gamma$ pair are \begin{eqnarray} P_{1}^{\mu} & = & \frac{\sqrt{S}}{2}\left\{ 1,0,0,1\right\} ;\\ P_{2}^{\mu} & = & \frac{\sqrt{S}}{2}\left\{ 1,0,0,-1\right\} ;\\ q^{\mu} & = & \left\{ \sqrt{Q^{2}+Q_{T}^{2}}\cosh y,Q_{T},0,\sqrt{Q^{2}+Q_{T}^{2}}\sinh y\right\}. \end{eqnarray} The light-cone momentum fractions for the boosted $2\rightarrow2$ scattering system are\begin{equation} x_{1,2}\equiv\frac{2(P_{2,1}\cdot q)}{S}=\frac{\sqrt{Q^{2}+Q_{T}^{2}}e^{\pm y}}{\sqrt{S}}.\end{equation} Decay of the $\gamma\gamma$ pairs is described in the hadronic Collins-Soper frame . The Collins-Soper frame is a rest frame of the $\gamma\gamma$ pair (with $q^{\mu}=\left\{ Q,0,0,0\right\} $ in this frame), chosen so that (a) the momenta $\vec{P}_{1}$ and $\vec{P}_{2}$ of the initial hadrons lie in the $Oxz$ plane (with zero azimuthal angle), and (b) the $z$ axis bisects the angle between $\vec{P}_{1}$ and $-\vec{P}_{2}$. The photon momenta are antiparallel in the Collins-Soper frame: \begin{eqnarray} P_{3}^{\mu} & = & \frac{Q}{2}\left\{ 0,\sin\theta_{*}\cos\varphi_{*},\sin\theta_{*}\sin\varphi_{*},\cos\theta_{*}\right\} ,\\ P_{4}^{\mu} & = & \frac{Q}{2}\left\{ 0,-\sin\theta_{*}\cos\varphi_{*},-\sin\theta_{*}\sin\varphi_{*},-\cos\theta_{*}\right\} ,\end{eqnarray} where $\theta_{*}$ and $\varphi_{*}$ are the photon's polar and azimuthal angles. In this section, we derive resummed predictions for the fully differential $\gamma\gamma$ cross section $d\sigma/(dQ^{2}dydQ_{T}^{2}d\Omega_{*}),$ where $d\Omega_{*}=d\cos\theta_{*}d\varphi_{*}$ is a solid angle element around the direction of $\vec{P}_{3}$ in the Collins-Soper frame defined in Eq.~(). The angles in the Collins-Soper frame are denoted by a {}``$*$'' subscript, in contrast to angles in the lab frame, which do not have such a subscript. The parton momenta and helicities are denoted by lowercase $p_{i}$ and $\lambda_{i}$, respectively. \subsection{Resummation of the initial-state QCD radiation } For completeness, we present an overview of the finite-order and resummed contributions associated with the direct production of diphotons. At the lowest order in the strong coupling strength $\alpha_{s}$, photon pairs are produced with zero transverse momentum $Q_{T}$. The Born $q\bar{q}\rightarrow\gamma\gamma$ cross section corresponding to Fig.~ (a) is \begin{equation} \left.\frac{d\sigma_{q\bar{q}}}{dQ^{2}dy\, dQ_{T}^{2}d\Omega_{*}}\right|_{Born}=\delta(\vec{Q}_{T})\sum_{i=u,\bar{u},d,\bar{d},...}\frac{\Sigma_{i}(\theta_{*})}{S}f_{q_{i}/h_{1}}(x_{1},\mu_{F})f_{\bar{q}_{i}/h_{2}}(x_{2},\mu_{F}),\end{equation} where $f_{q_{i}/h}(x,\mu_{F})$ denotes the parton distribution function (PDF) for a quark of a flavor $i$, evaluated at a factorization scale $\mu_{F}$ of order $Q$. The prefactor\begin{equation} \Sigma_{i}(\theta_{*})\equiv\sigma_{i}^{(0)}\frac{1+\cos^{2}\theta_{*}}{1-\cos^{2}\theta_{*}},\end{equation} with\begin{equation} \sigma_{i}^{(0)}\equiv\frac{\alpha^{2}(Q)e_{i}^{4}\pi}{2N_{c}Q^{2}},\end{equation} is composed of the running electromagnetic coupling strength $\alpha\equiv e^{2}/4\pi$ evaluated at the scale $Q$, fractional quark charge $e_{i}=2/3$ or $-1/3$, and number of QCD colors $N_{c}=3$. The lowest-order $gg\rightarrow\gamma\gamma$ scattering proceeds through an amplitude with a virtual quark loop (a box diagram) shown in Fig.~ (h). Its cross section takes the form\begin{equation} \left.\frac{d\sigma_{gg}}{dQ^{2}dy\, dQ_{T}^{2}d\Omega_{*}}\right|_{Born}=\delta(\vec{Q}_{T})\frac{\Sigma_{g}(\theta_{*})}{S}f_{g/h_{1}}(x_{1},\mu_{F})f_{g/h_{2}}(x_{2},\mu_{F}),\end{equation} where the prefactor\begin{equation} \Sigma_{g}(\theta_{*})\equiv\sigma_{g}^{(0)}L_{g}(\theta_{*})\end{equation} depends on the polar angle $\theta_{*}$ $ $ through a function $L_{g}(\theta_{*})$ presented explicitly in Ref.~. The overall normalization coefficient\begin{equation} \sigma_{g}^{(0)}=\frac{\alpha^{2}(Q)\alpha_{s}^{2}(Q)}{32\pi Q^{2}(N_{c}^{2}-1)}\left(\sum_{i}e_{i}^{2}\right)^{2}\end{equation} involves the sum of the squared charges $e_{i}^{2}$ of the quarks circulating in the loop. The NLO direct contributions, represented by Figs.~ (b)-(e), (i)-(l) and denoted as $P(Q,Q_{T},y,\Omega_{*})$, are computed in Refs.~. The NLO $2\rightarrow3$ differential cross section grows logarithmically if the final-state parton is soft or collinear to the initial-state quark or gluon, i.e., when $Q_{T}$ of the $\gamma\gamma$ pair is much smaller than $Q$. These {}``initial-state'' logarithmic contributions are summed to all orders later in this subsection. The NLO $qg$ cross section also contains a large logarithm when one of the photons is produced from a collinear $q\!\!\!\!^{^{(-)}}\rightarrow q\!\!\!\!^{^{(-)}}\gamma$ splitting in the final state. This {}``final-state'' collinear limit is discussed in Section~. With contributions from the initial-state soft or collinear radiation included, the NLO cross section is approximated in the small-$Q_{T}$ asymptotic limit by\begin{equation} A_{q\bar{q}}(Q,Q_{T},y,\Omega_{*})=\sum_{i=u,\bar{u},d,\bar{d},...}\frac{\Sigma_{i}(\theta_{*})}{S}\left\{ \delta(\vec{Q}_{T})F_{i,\delta}(Q,y,\theta_{*})+F_{i,+}(Q,y,Q_{T})\right\} \end{equation} in the $q\bar{q}+qg$ scattering channel, and by\begin{eqnarray} A_{gg}(Q,Q_{T},y,\Omega_{*}) & = & \frac{1}{S}\Biggl\{\Sigma_{g}(\theta_{*})\left[\delta(\vec{Q}_{T})F_{g,\delta}(Q,y,\theta_{*})+F_{g,+}(Q,y,Q_{T})\right]\nonumber \\ & & \hspace{12pt}+\Sigma_{g}^{\prime}(\theta_{*},\varphi_{*})F_{g}^{\prime}(Q,y,Q_{T})\Biggr\}\end{eqnarray} in the $gg+gq_{S}$ scattering channel. The functions $F_{a,\delta}(Q,y,\theta_{*})$ and $F_{a,+}^{(\prime)}(Q,y,Q_{T})$ for relevant parton flavors $a$ are listed in Appendix~. They include `plus function' contributions of the type $\left[Q_{T}^{-2}\ln^{p}\left(Q^{2}/Q_{T}^{2}\right)\right]_{+}$ with $p\geq0$, universal functions describing soft and collinear scattering, and process-dependent corrections from NLO virtual diagrams. The $q\bar{q}+qg$ asymptotic cross section $A_{q\bar{q}}(Q,Q_{T},y,\Omega_{*})$ is proportional to the angular function $\Sigma_{i}(\theta_{*}),$ the same as in the Born $q\bar{q}\rightarrow\gamma\gamma$ cross section, cf. Eq.~(). Similarly, the $gg+gq_{S}$ asymptotic cross section $A_{gg}(Q,Q_{T},y,\Omega_{*})$ includes a term proportional to the Born angular function $\Sigma_{g}(\theta_{*}).$ In addition, $A_{gg}(Q,Q_{T},y,\Omega_{*})$ contains another term proportional to $\Sigma_{g}^{\prime}(\theta_{*},\varphi_{*})\equiv L_{g}^{\prime}(\theta_{*})\cos2\varphi_{*}$, where $L_{g}^{\prime}(\theta_{*})$ is derived in Ref.~. This term arises due to the interference of Born amplitudes with incoming gluons of opposite polarizations and affects the azimuthal angle ($\varphi_{*}$) distribution of the photons in the Collins-Soper frame . The small-$Q_{T}$ representations in Eqs.~() and () can be used to compute fixed-order particle distributions in the phase-space slicing method. In this method, we choose a small $Q_{T}$ value $Q_{T}^{sep}$ in the range of validity of Eqs.~() and (). If the actual $Q_{T}$ in the computation exceeds $Q_{T}^{sep}$, we calculate the differential cross section using the full $2\rightarrow3$ matrix element. When $Q_{T}$ is smaller than $Q_{T}^{sep}$, we calculate the event rate using the small-$Q_{T}$ asymptotic approximation $A(Q,Q_{T},y,\Omega_{*})$ and $2\rightarrow2$ phase space. Hence, the lowest bin of the $Q_{T}$ distribution is approximated in the NLO prediction by its \textit{average} value in the interval $0\leq Q_{T}\leq Q_{T}^{sep}$, computed by integration of the asymptotic approximations. The phase-space slicing procedure is sufficient for predictions of observables inclusive in $Q_{T}$, but not of the shape of $d\sigma/dQ_{T}$ distributions. The latter goal is met by all-orders summation of singular asymptotic contributions with the help of the Collins-Soper-Sterman (CSS) method . The small-$Q_{T}$ resummed cross section is denoted as $W(Q,Q_{T},y,\Omega_{*})$ and given by a two-dimensional Fourier transform of a function $\widetilde{W}(Q,b,y,\Omega_{*})$ that depends on the impact parameter $\vec{b}$:\begin{eqnarray} W(Q,Q_{T},y,\Omega_{*}) & = & \int\frac{d\vec{b}}{(2\pi)^{2}}e^{i\vec{Q}_{T}\cdot\vec{b}}\widetilde{W}(Q,b,y,\Omega_{*})\nonumber \\ & \equiv & \int\frac{d\vec{b}}{(2\pi)^{2}}e^{i\vec{Q}_{T}\cdot\vec{b}}\widetilde{W}_{pert}(Q,b_{*},y,\Omega_{*})e^{-{\cal F}_{NP}(Q,b)}.\end{eqnarray} In this equation, $\widetilde{W}(Q,b,y,\Omega_{*})$ is written as a product of the perturbative part $\widetilde{W}_{pert}(Q,b_{*},y,\Omega_{*})$ and the nonperturbative exponent $\exp\left(-{\cal F}_{NP}(Q,b)\right),$ which describe the dynamics at small ($b\lesssim1\mbox{ GeV}^{-1}$) and large ($b\gtrsim1\mbox{ GeV}^{-1}$) impact parameters, respectively. The purpose of the variable $b_{*}$ is reviewed below. If $Q$ is large, the perturbative form factor $\widetilde{W}_{pert}$ dominates the integral in Eq.~(). It is computed at small $b$ as \begin{eqnarray} \widetilde{W}_{pert}(Q,b,y,\theta_{*}) & = & \sum_{a}\frac{\Sigma_{a}(\theta_{*})}{S}h_{a}^{2}(Q,\theta_{*})e^{-\mathcal{S}_{a}(Q,b)}\nonumber \\ & \times & \left[\mathcal{C}_{a/a_{1}}\otimes f_{a_{1}/h_{1}}\right](x_{1},b;\mu)\left[\mathcal{C}_{\bar{a}/a_{2}}\otimes f_{a_{2}/h_{2}}\right](x_{2},b;\mu).\end{eqnarray} The {}``hard-vertex'' function $\Sigma_{a}(\theta_{*})h_{a}^{2}(Q,\theta_{*})$ is the normalized cross section for the Born scattering $a\bar{a}\rightarrow\gamma\gamma$, with $a=u,\bar{u},d,\bar{d},...$ in $q\bar{q}\rightarrow\gamma\gamma$, and $a=\bar{a}=g$ in $gg\rightarrow\gamma\gamma$. The Sudakov exponent\begin{eqnarray} \mathcal{S}_{a}(Q,b)=\int_{C_{1}^{2}/b^{2}}^{C_{2}^{2}Q^{2}}\frac{d\bar{\mu}^{2}}{\bar{\mu}^{2}}\left[\mathcal{A}_{a}\left(C_{1},\bar{\mu}\right)\ln\left(\frac{C_{2}^{2}Q^{2}}{\bar{\mu}^{2}}\right)+\mathcal{B}_{a}\left(C_{1},C_{2},\bar{\mu}\right)\right]\end{eqnarray} is an integral of two functions $\mathcal{A}_{a}\left(C_{1},\bar{\mu}\right)$ and $\mathcal{B}_{a}\left(C_{1},C_{2},\bar{\mu}\right)$ between momentum scales $C_{1}/b$ and $C_{2}Q$, and $C_{1}$ and $C_{2}$ are constants of order $c_{0}\equiv2e^{-\gamma_{E}}=1.123...$ and $1$, respectively. The symbol $\left[\mathcal{C}_{a/a_{1}}\otimes f_{a_{1}/h}\right](x,b;\mu)$ stands for a convolution of the $k_{T}-$integrated PDF $f_{a_{1}/h}(x,\mu)$ and Wilson coefficient function $\mathcal{C}_{a/a_{1}}(x,b;C_{1}/C_{2},\mu)$, evaluated at a factorization scale $\mu$ and summed over intermediate parton flavors $a_{1}$: \begin{eqnarray} \left[\mathcal{C}_{a/a_{1}}\otimes f_{a_{1}/h}\right](x,b;\mu) & \equiv & \sum_{a_{1}}\left[\int_{x}^{1}{\frac{d\xi}{\xi}}\mathcal{C}_{a/a_{1}}\left(\frac{x}{\xi},b;\frac{C_{1}}{C_{2}},\mu\right)f_{a_{1}/h}(\xi,\mu)\right].\end{eqnarray} We compute the functions $h_{a},$ $\mathcal{A}_{a}$, $\mathcal{B}_{a}$ and $\mathcal{C}_{a/a_{1}}$ up to orders $\alpha_{s},$ $\alpha_{s}^{3},$ $\alpha_{s}^{2},$ and $\alpha_{s},$ respectively, corresponding to the NNLL accuracy of resummation. The perturbative coefficients at these orders in $\alpha_{s}$ are listed in Appendix~. The subleading contribution from the nonperturbative region $b\gtrsim1\mbox{ GeV}^{-1}$ is included in our calculation using a revised {}``$b_{*}$'' model~, which provides excellent agreement with $p_{T}$-dependent data on Drell-Yan pair and $Z$ boson production. In this model, the perturbative form factor $\widetilde{W}_{pert}(Q,b_{*},y,\Omega_{*})$ in Eq.~() is evaluated as a function of $b_{*}\equiv b/(1+b^{2}/b_{max}^{2})^{1/2},$ with $b_{max}=1.5\mbox{ GeV}^{-1}$. The factorization scale $\mu$ in $\left[{\mathcal{C}}\otimes f\right]$ is set equal to $c_{0}\sqrt{b^{-2}+Q_{ini}^{2}}$ , where $Q_{ini}$ is the initial scale of order 1 GeV in the parameterization employed for $f_{a/h}(x,\mu)$, for instance, 1.3 GeV for the CTEQ6 PDFs~. We have $\widetilde{W}_{pert}(b_{*})=\widetilde{W}_{pert}(b)$ at $b^{2}\ll b_{max}^{2},$ and $\widetilde{W}_{pert}(b_{*})=\widetilde{W}_{pert}(b_{max})$ at $b^{2}\gg b_{max}^{2}$. Hence, this ansatz preserves the exact form of the perturbative form factor $\widetilde{W}_{pert}(Q,b,y,\Omega_{*})$ in the perturbative region of small $b$, while also incorporating the leading nonperturbative contributions (described by a phenomenological function $\mathcal{F}_{NP}(Q,b)$) at large $b$. The form of $\mathcal{F}_{NP}(Q,b)$ found in the global $p_{T}$ fit in Ref.~ suggests approximate independence of $\mathcal{F}_{NP}(Q,b)$ from the type of $q\bar{q}$ scattering process. It is used here to describe the nonperturbative terms in the leading $q\bar{q}\rightarrow\gamma\gamma$ channel. We neglect possible corrections to the nonperturbative contributions arising from the final-state soft radiation in the $qg$ channel and additional $\sqrt{S}$ dependence affecting Drell-Yan-like processes at $x\lesssim10^{-2}$ , as these exceed the accuracy of the present measurements at the Tevatron. The experimentally unknown $\mathcal{F}_{NP}(Q,b)$ in the $gg$ channel is approximated by $\mathcal{F}_{NP}(Q,b)$ for the $q\bar{q}$ channel, multiplied by the ratio $C_{A}/C_{F}=9/4$. This choice is motivated by the fact that the leading Sudakov color factors ${\cal A}_a^{(k)}$ in the $gg$ and $q\bar{q}$ channels are proportional to $C_{A}=3$ and $C_{F}=4/3$, respectively. The uncertainties in the $\gamma\gamma$ cross sections associated with $\mathcal{F}_{NP}(Q,b)$ are investigated numerically in Ref.~. In the region $Q_{T}\sim Q$, collinear QCD factorization at a finite fixed order in $\alpha_{s}$ is applicable. In order to include non-singular contributions important in this region, we add to $W(Q,Q_{T},y,\Omega_{*})$ the regular piece $Y(Q,Q_{T},y,\Omega_{*}),$ defined as the difference between the NLO cross section $P(Q,Q_{T},y,\Omega_{*})$ and its small-$Q_{T}$ asymptotic approximation $A(Q,Q_{T},y,\Omega_{*})$:\begin{eqnarray} \frac{d\sigma(h_{1}h_{2}\rightarrow\gamma\gamma)}{dQ\, dQ_{T}^{2}\, dy\, d\Omega_{*}} & = & W(Q,Q_{T},y,\Omega_{*})+P(Q,Q_{T},y,\Omega_{*})-A(Q,Q_{T},y,\Omega_{*})\nonumber \\ & \equiv & W(Q,Q_{T},y,\Omega_{*})+Y(Q,Q_{T},y,\Omega_{*}).\end{eqnarray} At small $Q_{T},$ subtraction of $A(Q,Q_{T},y,\Omega_{*})$ in Eq.~() cancels large initial-state radiative corrections in $P(Q,Q_{T},y,\Omega_{*}),$ which are incorporated in their resummed form within $W(Q,Q_{T},y,\Omega_{*})$. At $Q_{T}$ comparable to $Q$, $A(Q,Q_{T},y,\Omega_{*})$ cancels the leading terms in $W(Q,Q_{T},y,\Omega_{*})$, but higher-order contributions remain from the infinite tower of logarithmic terms that are resummed in $W$. In this situation the $W+Y$ cross section drops below the finite-order result $P(Q,Q_{T},y,\Omega_{*})$ at some value of $Q_{T}$ (referred to as the {\em crossing point}) in both the $q\bar{q}+qg$ and $gg+gq_{S}$ channels, for each $Q$ and $y$. We use the $W+Y$ cross section as our final prediction at $Q_{T}$ values below the crossing point, and the NLO cross section $P$ at $Q_{T}$ values above the crossing point. A few comments are in order about our resummation calculation. The hard-vertex contribution $\Sigma_{a}(\theta_{*})h_{a}^{2}(Q,\theta_{*})$ and the functions $\mathcal{B}_{a}\left(C_{1},C_{2},\bar{\mu}\right)$ and $\mathcal{C}_{a/a_{1}}(x,b;C/C_{2},\mu)$ can be varied in a mutually compensating way while preserving the same value of the form factor $W$ up to higher-order corrections in $\alpha_{s}$. This ambiguity, or dependence on the chosen {}``resummation scheme'' within the CSS formalism, can be employed to explore the sensitivity of theoretical predictions to further next-to-next-to-next-to-leading logarithmic (NNNLL) effects that are not accounted for explicitly. The perturbative coefficients in Appendix~ are presented in the CSS resummation scheme , our default choice in numerical calculations, and in an alternative scheme by Catani, de Florian and Grazzini (CFG) . In the original CSS resummation scheme, the ${{\cal B}}$ and ${{\cal C}}$ functions contain the finite virtual NLO corrections to the $2\rightarrow2$ scattering process, whereas in the CFG scheme the universal ${\mathcal{B}}$ and ${\mathcal{C}}$ depend only on the type of incident partons, and the process-dependent virtual correction is included in the function $h_{a}$. The difference between the CSS and CFG schemes is numerically small in $\gamma\gamma$ production at both the Tevatron and the LHC . In the $gg+gq_{S}$ scattering channel, the unpolarized resummed cross section includes an additional contribution from elements of $k_{T}$-dependent PDF spin matrices with opposite helicities of outgoing gluons . The NLO expansion of this spin-flip resummed cross section generates the term proportional to $\Sigma_{g}^{\prime}(\theta_{*},\varphi_{*})\propto\cos2\varphi_{*}$ in the small-$Q_{T}$ asymptotic cross section, cf. Eq.~(). Although the logarithmic spin-flip contribution must be resummed in principle to all orders to predict the $\varphi_{*}$ dependence in the $gg+gq_{S}$ channel, it is neglected in the present work in view of its small effect on the full $\gamma\gamma$ cross section. When integrated over $Q_T$ from 0 to scales of order $Q$, the resummed cross section becomes approximately equal to the finite-order (NLO) cross section, augmented typically by a few-percent correction from integrated higher-order terms logarithmic in $Q_{T}$. Inclusive observables that allow such integration (e.g., the large-$Q$ region of the $\gamma\gamma$ invariant mass distribution) are approximated well both by the resummed and NLO calculations. However, the experimental acceptance constrains the range of the integration over $Q_{T}$ in parts of phase space and may break delicate cancellations between integrable singularities present in the finite-order differential distribution. In this situation (e.g., in the vicinity of the kinematic cutoff in $d\sigma/dQ$ discussed in Sec.~) the NLO cross section becomes unstable, while the resummed cross section (free of discontinuities) continues to depend smoothly on kinematic constraints. We see that the resummation is essential not only for the prediction of physical $Q_{T}$ distributions in $\gamma\gamma$ production, but also for credible estimates of the effects of experimental acceptance on distributions in the diphoton invariant mass and other variables. \subsection{Final-state photon fragmentation } \subsubsection{Single-photon fragmentation} In addition to the QCD singularities associated with initial-state radiation {[}described by the asymptotic terms in Eqs.~() and ()], other singularities arise in the ${\mathcal{O}}(\alpha_{s})$ process $q(p_{1})+g(p_{2})\rightarrow\gamma(p_{3})+\gamma(p_{4})+q(p_{5})$ {[}Fig.~ (e)] when a photon is collinear to the final-state quark. In this limit, the $qg\rightarrow q\gamma\gamma$ squared matrix element grows as $1/s_{\gamma5}$, when $s_{\gamma5} \rightarrow 0$, where $s_{\gamma5}$ is the squared invariant mass of the collinear $\gamma q$ pair. In this limit, the squared matrix element factors as \begin{equation} |{\mathcal{M}}(qg\rightarrow q\gamma\gamma)|^{2}\approx{\frac{2e^{2}e_{i}^{2}}{s_{\gamma5}}}P_{\gamma\leftarrow q}({z})|{\mathcal{M}}(qg\rightarrow q\gamma)|^{2}\end{equation} into the product of the squared matrix element $\left|{\mathcal{M}}(qg\rightarrow q\gamma)\right|^{2}$ for the production of a photon and an intermediate quark, and a splitting function $P_{\gamma\leftarrow q}(z)=(1+(1-z)^{2})/z$ for fragmentation of the intermediate quark into a collinear $\gamma q$ pair. In Eq.~() $z$ is the light-cone fraction of the intermediate quark's momentum carried by the fragmentation photon, and $ee_{i}$ is the charge of the intermediate quark. When the photon-quark separation $\Delta r=\sqrt{(\eta_{5}-\eta_{\gamma})^{2}+(\varphi_{5}-\varphi_{\gamma})^{2}}$ in the plane of pseudorapidity $\eta=-\log(\tan(\theta/2))$ and azimuthal angle $\varphi$ in the lab frame is small, as in the collinear limit, $s_{\gamma5}\approx E_{T\gamma}E_{T5}\Delta r^{2},$ where $E_{T\gamma}$ and $E_{T5}$ are the transverse energies of the photon and quark, with $E_{T}\equiv E\sin\theta$. Note that $E_{T5}=Q_{T}$ at the order in $\alpha_{s}$ at which we are working. Therefore, contributions from the final-state collinear, or fragmentation, region are most pronounced at small $\Delta r$ and relatively small $Q_{T}.$ \rightarrow0,$ limit, the final-state collinear contribution is suppressed, reflecting the absence of the soft singularity in the $qg\rightarrow q\gamma\gamma$ cross section. } A fully consistent treatment of the initial- and final-state singularities would require a joint initial- and final-state resummation. In the approaches taken to date, the fragmentation singularity may be subtracted from the direct cross section and replaced by a single-photon {}``one-fragmentation'' contribution $q+g\rightarrow(q\stackrel{frag}{\longrightarrow}\gamma)+\gamma$, where {}``$(\stackrel{frag}{q\longrightarrow\gamma})$'' denotes collinear production of one hard photon from a quark, described by a function $D_{\gamma}(z,\mu)$ at a light-cone momentum fraction $z$ and factorization scale $\mu$. Single-photon {}``two-fragmentation'' contributions arise in processes like $g+g\rightarrow(q\stackrel{frag}{\longrightarrow}\gamma)+(\bar{q}\stackrel{frag}{\longrightarrow}\gamma)$ and involve convolutions with two functions $D_{\gamma}(z,\mu)$ (one per photon). The lowest-order Feynman diagrams for the one- and two-fragmentation contributions are shown in Figs.~(f) and (g), respectively. Parameterizations must be adopted for the nonperturbative functions $D_{\gamma}(z,\mu)$ at an initial scale $\mu=\mu_{0}$. This is the approach followed in the DIPHOX calculation~, in which the sum of real and virtual NLO corrections to direct and single-$\gamma$ fragmentation cross sections is included. When explicit fragmentation function contributions are included, the inclusive rate is increased by higher-order contributions from photon production within hadronic jets. However, much of the enhancement is suppressed by isolation constraints imposed on the inclusive photon cross sections before the comparison with data. Nevertheless, fragmentation contributions surviving isolation may be moderately important in parts of phase space. An infrared-safe procedure can be formulated to apply isolation cuts at each order of $\alpha_{s}$~. This procedure encounters difficulties in reproducing the effects of isolation on fragmentation contributions, because theoretical models reflect only basic features of the experimental isolation and may introduce new logarithmic singularities near the edges of the isolation cones. As mentioned in the Introduction, the magnitude of the fragmentation contribution depends on the values of isolation parameters $E_{T}^{iso},$ $\Delta R$, and $\Delta R_{\gamma\gamma}$, modeled only approximately in a theoretical calculation. The collinear approximation constrains from below the values of $z$ accessible to $D_{\gamma}(z,\mu)$: $z>z_{min}\equiv(1+E_{T5}^{iso}/E_{T\gamma})^{-1}$. If $D_{\gamma}(z,\mu)$ varies rapidly with $z$, the fragmentation cross section is particularly sensitive to the assumed values of $E_{T}^{iso}$ and $z_{min}$. For instance, if $D_{\gamma}(z,\mu)\sim1/z$, the fragmentation cross section is roughly proportional to $E_{T}^{iso}$ under a typical condition $E_{T}^{iso}/E_{T\gamma}\ll1$. Such nearly linear dependence on $E_{T}^{iso}$ of the fragmentation cross section $d\sigma/dQ_{T}$ is indeed observed in the DIPHOX calculation, as reviewed in Sec.~. In reality, some spread of the parton radiation in the direction transverse to the photon's motion is expected. The treatment of kinematics in parton showering programs like PYTHIA results in somewhat different dependence on $z$~ compared to the collinear approximation, hence in a different magnitude of the fragmentation cross section. In this work we adopt a procedure that reproduces desirable features of the isolated cross sections, while bypassing some of the difficulties summarized above. To simulate experimental isolation, we reject an event if (a) the separation $\Delta r$ between the final-state parton and one of the photons is less than $\Delta R$, and (b) $E_{T5}$ of the parton is larger than $E_{T}^{iso}$. This condition is applied to the NLO cross section $P(Q,Q_{T},y,\Omega_{*})$, but not to $W(Q,Q_{T},y,\Omega_{*})$ and $A(Q,Q_{T},y,\theta_{*})$, as these correspond to initial-state QCD radiation and are free of the final-state collinear singularity. This quasi-experimental isolation excludes the singular final-state direct contributions at $E_{T5}>E_{T}^{iso}$ and $\Delta r<\Delta R$ (or $s_{\gamma5}<E_{T\gamma}E_{T5}\Delta R^{2}$). It is effective for $Q_{T}>E_{T}^{iso}$, but the collinear direct contributions survive when $Q_{T}<E_{T}^{iso}$. The integrated (but not the differential) fragmentation rate in the region $Q_{T}<E_{T}^{iso}$ may be estimated from a calculation with explicit fragmentation functions. In our approach, we do not introduce fragmentation functions, but we apply an auxiliary regulator to the direct $qg$ cross section at $Q_{T}<E_{T}^{iso}$ and $\Delta r<\Delta R$. In our numerical study we find that this prescription preserves a continuous differential distribution except for a small finite discontinuity at $Q_{T} = E_{T}^{iso}$. It approximately reproduces the integrated $qg$ rate obtained in the DIPHOX calculation at small $Q_{T}$, for the nominal $E_{T}^{iso}$. Two forms of the auxiliary regulator are considered below, based on subtraction of the leading collinear contribution and smooth-cone isolation . In the first case, we subtract the leading part Eq.~() of the direct $qg$ matrix element when $E_{T5}<E_{T}^{iso}$ and $\Delta r<\Delta R.$ We take $z=1-p_{s}\cdot p_{5}/(p_{s}\cdot p_{f}+p_{s}\cdot p_{5}+p_{f}\cdot p_{5}),$ where $p_{f}^{\mu},$ $p_{5}^{\mu},$ and $p_{s}^{\mu}$ are the four-momenta of the fragmentation photon, fragmentation quark, and spectator photon, respectively~. This prescription is used in most of the numerical results in this paper. In the second case, we suppress fragmentation contributions at $\Delta r<\Delta R$ and $E_{T5}<E_{T}^{iso}$ by rejecting events in the $\Delta R$ cone that satisfy $E_{T5}<\chi(\Delta r)$, where $\chi(\Delta r)$ is a smooth function satisfying $\chi(0)=0,$ $\chi(\Delta R)=E_{T}^{iso}$. This {}``smooth-cone isolation''~ transforms the fragmentation singularity associated with $D_{\gamma}(z,\mu)$ into an integrable singularity, which depends on the assumed functional form of $\chi(\Delta r)$. The cross section for direct contributions is rendered finite by this prescription without explicit introduction of fragmentation functions $D_{\gamma}(z,\mu)$. For our smooth function, we choose $\chi(\Delta r)=E_{T}^{iso}(1-\cos\Delta r)^{2}/(1-\cos\Delta R)^{2}$, which differs from the specific form considered in Ref.~, but still satisfies the condition $\chi(0)=0.$ Our earlier results in Ref.~ are computed with this prescription. Here we employ it only in a few instances for comparison with the subtraction method and obtain similar results. Differences between the two prescriptions can be used to quantify sensitivity of the predictions to the treatment of the $Q_{T}<E_{T}^{iso}$ and $\Delta r<\Delta R$ region. The two prescriptions yield identical predictions outside of this restricted region, notably at $Q_{T}>E_{T}^{iso}$, where our NLO perturbative expression $P(Q,Q_{T},y,\Omega_{*})$ in the $q\bar{q}+qg$ channel is controlled only by quasi-experimental isolation and coincides with the corresponding direct cross section in DIPHOX. The default subtraction prescription predicts a vanishing $d\sigma/dQ_{T}$ in the extreme $Q_{T}\rightarrow0$ limit, while the smooth-cone prescription has an integrable singularity in this limit, avoided by an explicit small-$Q_{T}$ cutoff in the calculation of our $Y$-piece. Both prescriptions are free of the logarithmic singularity at $Q_{T}=E_{T}^{iso}$ arising in the fixed-order (DIPHOX) calculation. \subsubsection{Low-$Q$ diphoton fragmentation } Another class of large radiative corrections arises when the $\gamma\gamma$ invariant mass $Q$ is smaller than the $\gamma\gamma$ transverse momentum $Q_{T}$. In this case, one final-state quark or gluon fragments into a low-mass $\gamma\gamma$ pair, e.g. as $q+g\rightarrow(q\stackrel{frag}{\longrightarrow}\gamma\gamma)+g$. The lowest-order contributions of this kind are shown in Fig.~. The process is described by a $\gamma\gamma$-fragmentation function $D_{\gamma\gamma}(z_{1},z_{2},\mu)$, different from the single-photon fragmentation function $D_{\gamma}(z,\mu)$. This new {}``two-photons from one-fragmentation'' contribution is not included yet in existing calculations, even though similar fragmentation mechanisms have been studied in large-$Q_{T}$ Drell-Yan pair production~. The importance of low-$Q$ $\gamma\gamma$-fragmentation may be elevated in some kinematic regions for typical experimental cuts. They can be removed by adjustments in the experimental cuts, as discussed in Sec.~. \subsection{Summary of the calculation } We conclude this section by summarizing the main features of our calculation. Full direct NLO cross sections, represented by the graphs (a)-(e), (h)-(l) in Fig.~, are computed, and their initial-state soft/collinear logarithmic singularities are resummed at small $Q_{T}$ in both the $q\bar{q}+qg$ and $gg+gq_{S}$ channels. The perturbative Sudakov functions ${\mathcal{A}}$ and ${\mathcal{B}}$ and Wilson coefficient functions ${\mathcal{C}}$ in the resummed cross section $W$ are computed up to orders $\alpha_{s}^{3},$ $\alpha_{s}^{2}$, and $\alpha_{s}$, respectively, corresponding to resummation at NNLL accuracy. Our resummation calculation requires an integration over all values of impact parameter $b$, including the nonperturbative region of large $b$. In our default calculation of the resummed cross section, we adopt the nonperturbative functions introduced in Ref.~. We consider two resummation schemes, the traditional scheme introduced in the CSS paper as well as an alternative scheme~. The comparison allows us to estimate the magnitude of yet higher-order corrections that are not included. The size of these effects is different in the $q\bar{q}+qg$ and $gg+gq_{S}$ channels but not particularly significant in either . The final-state collinear singularity in the $qg$ scattering channel is avoided by applying quasi-experimental isolation when $Q_{T}>E_{T}^{iso}$ and an auxiliary regulator when $Q_{T}<E_{T}^{iso}$ to approximate on average the full NLO rate from direct $qg$ and fragmentation cross sections in this $Q_{T}$ range. Two prescriptions for the auxiliary regulator (subtraction and smooth isolation inside the photon's isolation cone) are considered and lead to similar predictions at the Tevatron and LHC. The singular logarithmic contributions associated with initial-state radiation are subtracted from the NLO cross section $P$ to form a regular piece $Y,$ which is added to the small-$Q_{T}$ resummed cross section $W$ to predict the production rate for small and intermediate values of $Q_{T}$. In the $gg+gq_{S}$ channel, we also subtract from $P$ a new singular spin-flip contribution that affects azimuthal angle ($\varphi_{*})$ dependence in the Collins-Soper reference frame. We switch our prediction to the fixed-order perturbative result $P$ at the point in $Q_{T}$ where the cross section $W+Y$ drops below $P$. This crossing point is located at $Q_{T}$ of order $Q$ in both $q\bar{q}+qg$ and $gg+gq_{S}$ channels. \section{Comparisons with Data and Predictions } Our calculation of the differential cross section $d\sigma/(dQdQ_{T}dyd\Omega_{*})$ is especially pertinent for the transverse momentum $Q_{T}$ distribution in the region $Q_{T}\lesssim Q$, for fixed values of diphoton mass $Q$ (cf. Section~). It would be best to compare our \emph{multiple} differential distribution with experiment, but published collider data tend to be presented in the form of singly differential distributions in $Q$, $Q_{T}$, and $\Delta\varphi\equiv\varphi_{3}-\varphi_{4}$ in the lab frame, after integration over the other independent kinematic variables. We follow suit in order to make comparisons with Tevatron collider data, but we recommend that more differential studies be made, and we comment on the features that can be explored. We show results at the energy of the Tevatron collider and then make predictions for the Large Hadron Collider. The analytical results of Sec.~ are implemented in our computer code. As a first step, resummed and NLO $\gamma\gamma$ cross sections are computed on a grid of discrete values of $Q$, $Q_{T}$, and $y$ by using the resummation program \textsc{Legacy} described in Refs.~. At the second stage, matching of the resummed and NLO cross sections is performed, and fully differential cross sections are evaluated by Monte-Carlo integration of the matched grids in the latest version of the program \textsc{ResBos}~. The calculation is done for $N_{f}=5$ active quark flavors and the following values of the electroweak and strong interaction parameters~: \begin{eqnarray} & & G_{F}=1.16639\times10^{-5}~\textrm{GeV}^{-2},~~m_{Z}=91.1882~{\textrm{GeV}},\\ & & \alpha(m_{Z})=1/128.937,~~\alpha_{s}(m_{Z})=0.1187.\end{eqnarray} The following choices of the factorization constants are used: $C_{1}=C_{3}=2e^{-\gamma_{E}}\approx1.123...$, and $C_{2}=C_{4}=1.$ The choice $C_{4}=1$ implies that we equate the renormalization and factorization scales to the invariant mass of the photon pair, $\mu_{R}=\mu_{F}=Q$, in the fixed-order and asymptotic contributions $P(Q,Q_{T},y,\Omega_{*})$ and $A(Q,Q_{T},y,\Omega_{*})$. We use two-loop expressions for the running electromagnetic and strong couplings $\alpha(\mu)$ and $\alpha_{S}(\mu)$, as well as the NLO parton distribution function set CTEQ6M~ with $Q_{ini}=1.3$ GeV. For calculations with explicit final-state fragmentation functions included, we use set 1 of the NLO photon fragmentation functions from Ref.~. \subsection{Results for Run 2 at the Tevatron} \subsubsection{Kinematic constraints } In this section, we present our results for the Tevatron $p\bar{p}$ collider operating at $\sqrt{S}=1.96$~TeV. In order to compare with the data from the Collider Detector at Fermilab (CDF) collaboration~, we make the same restrictions on the final-state photons as those used in the experimental measurement (unless stated otherwise): \begin{eqnarray} & & {\textrm{transverse momentum}}~p_{T}^{\gamma}>p_{T\, min}^{\gamma}=14~(13)~{\textrm{GeV for the harder (softer) photon, }}\\ & & {\textrm{and rapidity}}~|y^{\gamma}|<0.9~{\textrm{for each photon}}.\end{eqnarray} We impose isolation conditions described in Section , assuming the nominal isolation energy $E_{T}^{iso}=1$ GeV specified in the CDF publication, along with $\Delta R=0.4,$ and $\Delta R_{\gamma\gamma}=0.3$. We also show predictions for the constraints that approximate event selection conditions used by the Fermilab D\O~Collaboration~: $p_{T}^{\gamma}>p_{T\, min}^{\gamma}=21~(20)$ GeV for the harder (softer) photon, $|y^{\gamma}|<1.1$, and $E_{T}^{iso}/E_{T}^{\gamma}=0.07$ for each photon, for the same $\Delta R$ and $\Delta R_{\gamma\gamma}$ values as in the CDF case. \begin{figure*} \caption{The diphoton event distribution from the theoretical simulation for $\sqrt{S}=1.96$\,GeV, with the selection criteria imposed in the CDF measurement, as a function of the various kinematic variables described in the text, shown for $Q_T < Q$ and $Q_T > Q$ separately. } \end{figure*} A scatter plot of event distributions from our theoretical simulation for CDF kinematic cuts and arbitrary luminosity is shown in Fig.~. The events are plotted versus the invariant mass $Q$, transverse momentum $Q_{T}$, rapidity separation $\left|\Delta y\right|\equiv\left|y_{hard}-y_{soft}\right|$, and azimuthal separation $\Delta\varphi\equiv\left|\varphi_{hard}-\varphi_{soft}\right|$ (with $0\leq\Delta\varphi\leq\pi)$ between the harder and softer photon in the lab frame, as well as the cosine of the polar angle $\theta_{*}$ in the Collins-Soper frame. It can be seen from the figure that $\Delta\varphi$ is correlated with the difference $Q_{T}-Q$. Events with $Q_{T}<Q$ ($Q_{T}>Q$) tend to populate regions with $\Delta\varphi>\pi/2$ ($\Delta\varphi<\pi/2$). The extreme case $Q_{T}=0$ relevant to the Born approximation corresponds to $\Delta\varphi=\pi$. The $p_{T}^{\gamma}$ cuts suppress the mass region $Q\lesssim2\sqrt{p_{Tmin}^{\gamma_{3}}p_{Tmin}^{\gamma_{4}}}\approx27$ GeV at $\Delta\varphi\approx\pi$ and $Q_{T}\lesssim25$ GeV at $\Delta\varphi\approx0$, leading to the appearance of a kinematic cutoff in the invariant mass distribution and a {}``shoulder'' in the transverse momentum distribution, as shown in later sections. Our theoretical framework is applicable in the region $Q_{T}\lesssim Q$ (large $\Delta\varphi$), where the dominant fraction of events occurs. The appearance of singularities in the NLO calculation at $Q_{T}\rightarrow0$ and the fact that there are two different hard scales, $Q_{T}$ and $Q$, relevant for the event distributions in the low-$Q_{T}$ region require that we address and resum large logarithmic terms of the form $\log(Q/Q_{T})$. Different and interesting physics becomes important in the complementary region $Q_{T}>Q$ (small $\Delta\varphi$), a topic we address in Sec.~. \subsubsection{Tevatron cross sections } \begin{figure*} \caption{Invariant mass distributions of photon pairs in $p\bar{p}\rightarrow\gamma\gamma X$ at $\sqrt{S}=1.96$~TeV with QCD contributions calculated in the soft--gluon resummation formalism (red solid) and at NLO (blue dashed). The calculations include the cuts used by the CDF collaboration whose data are shown~. } \end{figure*} We compare our resummed and finite-order predictions for the invariant mass ($Q$) distribution of photon pairs, shown in Fig.~ as solid and dashed lines, respectively. The finite-order cross section is evaluated at $O(\alpha_{s})$ accuracy in the $q\bar{q}+qg$ channel and at $O(\alpha_{s}^{3})$ accuracy in the $gg+gq_{S}$ channel. These finite-order calculations are performed with the phase-space slicing method described in Sec.~. When integrated over all $Q_{T}$, as in the $d\sigma/dQ$ distribution at large $Q$, the resummed logarithmic terms from higher orders in $\alpha_{s}$ produce a relatively small NNLO correction, such that the resummed and finite-order mass distributions in Fig.~ are close to one another in normalization and shape. Both distributions also agree with the CDF data in this $Q$ range within experimental uncertainties. The shape of $d\sigma/dQ$ at small $Q$ is affected by the cuts in Eq.~() on the transverse momenta $p_{T}^{\gamma}$ of the two photons. In addition to being responsible for the characteristic cutoff at $Q\approx27$ GeV explained in the previous subsection, the cuts on the individual transverse momenta $p_{T}^{\gamma}$ also introduce a dependence of the invariant mass distribution on the shape of the $Q_{T}$ spectrum of the $\gamma\gamma$ pairs. Because of this correlation between the $Q$ and $Q_{T}$ distributions, the discontinuities in $d\sigma/dQ_{T}$ as $Q_{T}\rightarrow0$, when computed at finite order, make finite-order predictions for $d\sigma/dQ$ somewhat unstable. \begin{figure*} \begin{centering} \\ (a) \hspace{0.45\columnwidth} (b) \par\end{centering} \caption{Transverse momentum distributions in $p\bar{p}\rightarrow\gamma\gamma X$ at $\sqrt{S}=1.96$~TeV along with the CDF data: (a) the fixed-order prediction $P$ (dashes) and its asymptotic approximation $A$ (dots); (b) the full resummed cross section (solid), obtained by matching the resummed $W+Y$ to the fixed-order prediction $P$ (dashed, same as in (a)) at large $Q_{T}$. } \end{figure*} The finite-order expectation for the transverse momentum distribution $d\sigma/dQ_{T}$ (i.e., the integral of $P(Q,Q_{T},y,\Omega_{*})$ over $Q$, $y$, and $\Omega_{*}$, or $P$ for brevity) is shown as a dashed curve in Fig.~(a). It exhibits an integrable singularity in the small-$Q_{T}$ limit. Terms with inverse power and logarithmic dependence on $Q_{T}$, associated with initial-state radiation as $Q_{T}\rightarrow0$, are extracted from $P$ and form the asymptotic contribution, denoted as $A$ (dotted curve). In the figure, both $P$ and $A$ are truncated at a small value of $Q_{T}$, that is, not drawn all the way to $Q_{T}=0$. The curves for $P$ and $A$ are close at small values of $Q_{T}$, signaling that the initial-state logarithmic singularities dominate the NLO distribution. The difference $Y$ between the $P$ and $A$ distributions includes the finite regular terms not included in $A$ and logarithmic terms from the final-state fragmentation singularities, with the latter subtracted when $Q_{T}<E_{T}^{iso}$, as described in Sec.~. The data clearly disfavor the fixed-order prediction in the region of low $Q_{T}$. Figure~(b) features the resummed $W+Y$ contribution (solid curve). Resummation of the initial-state logarithmic terms renders $W$ finite in the region of small $Q_{T}$. The sum of $W$ and $Y$ includes the resummed initial-state singular contributions plus the remaining relevant terms in $P$. Since $P$ provides a reliable fixed-order estimate at large $Q_{T}$, we present our final resummed prediction by switching from $W+Y$ to $P$ at the point at which the two differential cross sections (as functions of $Q$, $Q_{T}$ and $y$) cross each other. In contrast to the fixed-order (dashed) curve $P$ in Fig.~(b), the agreement with data is improved at the lowest values of $Q_{T}$, where resummation brings the rate down, and for $Q_{T}=12-32$ GeV, where the resummed logarithmic terms increase the rate. The resummed predictions for the Tevatron experiments are practically insensitive to the choice of the resummation scheme and the nonperturbative model~. About 75\ at the Tevatron with CDF cuts imposed comes from the $q{\bar{q}}+qg+{\bar{q}}g$ ($gg+gq_{S}$) initial state. The fractions for the cuts used by D\O~ are 84\ $Q_{T}>22$ GeV, because the gluon PDF decreases rapidly with parton fractional momentum $x$ . \begin{figure*} \caption{The difference $\Delta\varphi$ in the azimuthal angles of the two photons in the laboratory frame predicted by the resummed (solid) and fixed-order (dashed) calculations, compared to the CDF data.} \end{figure*} The distribution in the difference $\Delta\varphi$ of the azimuthal angles of the photons is shown in Fig.~. As is true for the transverse momentum distribution in the limit $Q_{T}\rightarrow0$, the distribution computed at fixed order is ill-defined at $\Delta\varphi=\pi$. The resummed distribution shows a larger cross section near $\Delta\varphi=2.5$~rad, in better agreement with the data. In the region of small $\Delta\varphi\lesssim\pi/2$, the fixed-order and the resummed predictions are the same, a result of our matching of the resummed and fixed-order distributions at mid to high values of $Q_{T}$. Although the cross section is not large in the region $\Delta\varphi<\pi/2$, there is an indication of a difference between our predictions and data in this region, a topic we address below. \subsubsection{The region $Q_{T}>Q$ } It is evident from Fig.~ that the $\Delta\varphi<\pi/2$ region is populated mostly by events with $Q_{T}>Q$. New types of radiative contributions may be present in this region, including various fragmentation contributions described in Sec.~ and enhancements at large $|\cos\theta_{*}|$ in the direct production rate. While experimental isolation generally suppresses long-distance fragmentation, a greater fraction of fragmentation photons are expected to survive isolation when $\Delta\varphi<\pi/2$. Besides single-photon `one-fragmentation' and `two-fragmentation' contributions (with one photon per fragmenting parton), one encounters additional logarithmic singularities of the form $\log(Q/Q_{T})$. We noted in Sec.~ that these logarithms are associated with the fragmentation of a parton carrying large transverse momentum $Q_{T}$ into a system of small invariant mass $Q$ , a light $\gamma\gamma$ pair in our case. Small-$Q$ $\gamma\gamma$ fragmentation of this kind is not implemented yet in theoretical models. Therefore, we are prepared for the possibility that both the fixed-order calculation and our resummed calculation may be deficient in the region $Q_{T}\gg Q$. A detailed experimental study of the region $Q_{T}>Q$ may offer the opportunity to measure the parton to two-photon fragmentation function $D_{\gamma\gamma}(z_{1},z_{2})$, provided that the single-photon `one-fragmentation' function $D_{\gamma}(z)$ is determined by single-photon data, and the low-$Q$ logarithmic terms are properly resummed theoretically. In addition to the low-$Q$ fragmentation, the small-$\Delta\varphi$ region may be sensitive to large higher-order contributions associated with $\widehat{t}$- or $\widehat{u}$-channel exchanges in the $q\bar{q}\rightarrow\gamma\gamma$ and $gg\rightarrow\gamma\gamma$ subprocesses. In the Born processes in Figs.~(a) and (h), the $\widehat{t}$- and $\widehat{u}$-channel singularities arise at $\cos\theta_{*}\approx\pm1$ and $\Delta\varphi\approx\pi$. These singularities are excluded by the $p_{T}^{\gamma}$ cuts in Eq.~(), but related residual enhancements in the NLO contributions may still persist at $|\cos\theta_{*}|\approx1$ and $\Delta\varphi\rightarrow0$, not excluded by the cuts (cf. Fig.~). Because $\left| \cos \theta_*\right|$ is large in such events, they tend to have substantial $|\Delta y|$, so they are retained by the $\Delta R_{\gamma\gamma}>0.3$ cut. In contrast, the low-$Q$ fragmentation contributions tend to be abundant at small $|\Delta y|$. It may be therefore possible to distinguish between the large-$|\cos\theta_{*}|$ and fragmentation events at small $\Delta\varphi$ based on the distribution in $|\Delta y|$. We expect much better agreement of our predictions with data if the selection $Q_{T}<Q$ is made. This selection preserves the bulk of the cross section and assures that a fair comparison is made in the region of phase space where the predictions are most valid. \subsubsection{Fragmentation and comparison with the \textsc{DIPHOX} code} One way to obtain an estimate of theoretical uncertainty is to compare theoretical approaches in various parts of phase space, including small $\Delta\varphi$. We handle the collinear final-state photon singularities in the manner described in Sec.~, without including photon fragmentation functions explicitly. An alternative calculation implemented in the \textsc{DIPHOX} code includes NLO cross sections for single-photon fragmentation processes. Neither code includes a term in which both photons are fragmentation products of the same final-state parton, i.e., the diphoton fragmentation function $D_{\gamma\gamma}(z_{1},z_{2})$. \begin{figure*} \begin{centering} \\ (a) \hspace{0.45\columnwidth} (b) \par\end{centering} \caption{Comparison of our resummed and \textsc{DIPHOX} predictions for (a) the invariant mass and (b) transverse momentum distributions of $\gamma\gamma$ pairs for D\O~kinematic cuts. The solid curves show our resummed distributions with all channels included. The dashed and dotted curves illustrate the resummed and DIPHOX distributions in the $q\bar{q}+qg$ channel. } \end{figure*} In Ref.~ we show comparisons of our predictions with those of \textsc{DIPHOX} along with the CDF data. Here in Fig.~, we show analogous plots of the invariant mass and transverse momentum distributions for D\O~cuts. We note that our fixed-order $q\bar{q}+qg$ contribution agrees well with the direct contribution in \textsc{DIPHOX}. This agreement is particularly impressive in the region of large $Q_{T}$, where both codes use the same fixed-order formalism to handle direct contributions. A contribution from the $gg$ channel is also present in both codes, computed at LO in \textsc{DIPHOX} but at NLO+NNLL in our case. Since the $gg+gq_{S}$ contribution is not dominant (especially in the high $Q_{T}$ region), this difference does not have a significant impact on the comparison. The explicit single-photon fragmentation contributions in \textsc{DIPHOX} (mostly `one-fragmentation' contribution) are quite small for the nominal hadronic energy $E_{T}^{iso} \sim 1$~GeV in the cone around each photon. Exceptions occur in the region $Q_{T}\leq E_{T}^{iso}$, where the fragmentation contributions to $d\sigma/dQ_{T}$ have logarithmic singularities, and in the $\Delta\varphi\rightarrow0$ region, where fragmentation is comparable to the direct contributions. Our isolation prescription reproduces the integrated \textsc{DIPHOX} rate well for $0\leq Q_{T}\leq E_{T}^{iso}$, leading to close agreement between the resummed and \textsc{DIPHOX} inclusive rates for most $Q$ values. Returning to the CDF measurement, we remark that the resummed and \textsc{DIPHOX} cross sections for the same $E_{T}^{iso}=1$ GeV underestimate the data within two standard deviations for $Q\lesssim27$ GeV, $Q_{T}>25$ GeV, and $\Delta\varphi<1$ rad (cf. the relevant figures in Ref.~). The \textsc{DIPHOX} cross section can be raised to agree with data in this {}``shoulder'' region, if a much larger isolation energy ($E_{T}^{iso}=4$ GeV) is chosen, and smaller factorization and renormalization scales are used ($\mu_{F}=\mu_{R}=Q/2$). These are the choices made in the CDF study . Since $E_{T}^{iso}$ is an approximate characteristic of the experimental isolation, one might argue that both $E_{T}^{iso}=1$ and 4 GeV can be appropriate in a calculation to match the conditions of the CDF measurement. The direct contribution is weakly sensitive to $E_{T}^{iso}$, while the one-fragmentation part of $d\sigma/dQ_{T}$ is roughly proportional to $E_{T}^{iso}$ (cf. Section~). The one-fragmentation contribution is enhanced on average by 400\ increased in the calculation from 1 to 4 GeV. The rate in the shoulder region is enhanced further if the factorization scale $\mu_{F}$ is reduced. \begin{figure*} (a) \hspace{0.45\columnwidth} (b) \caption{Predicted cross sections for diphoton production in $p\bar{p}\rightarrow\gamma\gamma X$ at $\sqrt{S}=1.96$~TeV as a function of (a) the $\gamma\gamma$ pair transverse momentum $Q_{T}$ and (b) the difference $\Delta\varphi$ in the azimuthal angles of the two photons. Our resummed predictions (solid) are shown together with \textsc{DIPHOX} predictions for the default isolation energy $E_{T}^{iso}=1$ GeV and factorization scale $\mu_{F}=Q$ (dashed), and for $E_{T}^{iso}=4$ GeV, $\mu_{F}=Q/2$ (dotted). We impose the condition $Q_{T}<Q$ to reduce theoretical uncertainties associated with fragmentation.} \end{figure*} Since the theoretical specifications for isolation and for the fragmentation contribution are admittedly approximate, we question whether great importance should be placed on the agreement of theory and experiment in the region of small $\Delta\varphi$ or in the shoulder region in the $Q_{T}$ distribution. A straightforward way to reduce sensitivity to fragmentation is to require $Q>27$ GeV or $Q_{T}<Q$, as discussed above. The two cuts have similar effects on the event distributions. Figure~ shows the effects of the $Q_{T}<Q$ cut on the $Q_{T}$ and $\Delta\varphi$ distributions. The cut $Q_{T}<Q$ is particularly efficient at suppressing the fragmentation $Q_{T}$ shoulder (and the region of small $\Delta\varphi$ altogether), while only a small portion of the event sample is lost. This cut is especially favorable, since it constrains the comparison with data to a region where the theory is well understood and has a small uncertainty. Furthermore, with the requirement of $Q_{T}<Q$, the dependence of differential cross sections on the choices of isolation energy $E_{T}^{iso}$ and factorization scale $\mu_{F}$ is greatly reduced to the typical size of higher-order corrections. We predict that if a $Q_{T}<Q$ cut, or a $Q>27$ GeV cut, is applied to the Tevatron data, the enhancement at low $\Delta\varphi$ and intermediate $Q_{T}$ associated with the fragmentation contribution will disappear. This is an important conclusion of our study, and we urge the CDF and D\O\, collaborations to apply these cuts in their future analyses of the diphoton data. \subsubsection{Average transverse momentum } \begin{figure*} \begin{centering}(a) \hspace{0.45\columnwidth} (b) \par\end{centering} \caption{(a) Resummed transverse momentum distributions of photon pairs in various invariant mass bins used in the CDF measurement, normalized to the total cross section in each $Q$ bin. (b) The average $Q_{T}$ as a function of the $\gamma\gamma$ invariant mass, computed for $Q_{T}<Q$. } \end{figure*} An important prediction of the resummation formalism is the change of the transverse momentum distribution with the diphoton invariant mass. This dependence comes, in part, from the $\ln Q^{2}$ dependence in the Sudakov exponent, Eq.~(), and it is desirable to identify this feature amid other influences. In Fig.~(a), we show normalized resummed transverse momentum distributions for various selections of the invariant mass of the photon pairs. Without kinematical constraints on the decay photons, the $Q_{T}$ distribution is expected to broaden with increasing $Q$, and the position of the peak in $d\sigma/dQ_T$ to shift to larger $Q_T$ values. The shift of the peak may or may not be observed in the data depending on the chosen lower cuts on $p_T$ of the photons, which suppress the event rate at low $Q$ and $Q_T$. The interplay of the Sudakov broadening of the $Q_T$ distribution and kinematical suppression by the photon $p_T$ cuts is reflected in the shape of $d\sigma/dQ_{T}$ in different $Q$ bins. According to dimensional analysis, the average $\langle Q_{T}\rangle$ in the interval $Q_{T}\leq Q$ may be expected to behave as \begin{equation} \langle Q_{T}\rangle_{Q_{T}\leq Q}=Qf(Q/\sqrt{S}),\end{equation} where the scaling function $f(Q/\sqrt{S}$) reflects phase space constraints, dependence on the Sudakov logarithm, and the $x$ dependence of the PDFs. Figure~(b) shows our calculated diphoton mass dependence of $\langle Q_{T}\rangle$. The linear increase shown in Eq. ({) is observed over the range $30<Q<80$~GeV. For values of $Q$ below the kinematic cutoff at about 30 GeV, the cuts shown in Fig. 3 suppress diphoton production at small $Q_{T}$, and $\langle Q_{T}/Q\rangle$ grows toward 1 as $Q$ decreases (corresponding to production only at $Q_{T}$ close to $Q$). For $Q\sim80$~GeV and above, we see a saturation of the growth of $\langle Q_{T}\rangle$, a reflection of the influences of the $x$ dependence of the PDFs and other factors. Similar saturation behavior is observed in calculations of $\langle Q_{T}\rangle$ in other processes \textbf{}\textbf{.} It would be interesting to see a comparison of our prediction with data from the CDF and D\O~collaborations. \subsection{Results for the LHC} \subsubsection{Event selection} To obtain predictions for $pp$ collisions at the LHC at $\sqrt{S}=14$~TeV, we employ the cuts on the individual photons used by the ATLAS collaboration in their simulations of Higgs boson decay, $h\rightarrow\gamma\gamma$~. We require \begin{eqnarray} & & {\textrm{transverse momentum}}~p_{T}^{\gamma}>40~(25)~{\textrm{GeV for the harder (softer) photon, }}\\ & & {\textrm{and rapidity}}~|y^{\gamma}|<2.5~{\textrm{for each photon}}.\end{eqnarray} In accord with ATLAS specifications, we impose a looser isolation restriction than for our Tevatron study, requiring less than $E_{T}^{iso}=15$ GeV of hadronic and extra electromagnetic transverse energy inside a $\Delta R=0.4$ cone around each photon. We also require the separation $\Delta R_{\gamma\gamma}$ between the two isolated photons to be above 0.4. The cuts listed above, optimized for the Higgs boson search, may require adjustments in order to test perturbative QCD predictions in the full $\gamma\gamma$ invariant mass range accessible at the LHC. The values of the $p_{T}^{\gamma}$ cuts on the photons in Eq.~() preserve a large fraction of Higgs boson events with $Q>115$ GeV. These cuts may be too restrictive in studies of $\gamma\gamma$ production at smaller $Q$, considering that the two final-state photons most likely originate from a $\gamma\gamma$ pair with small $Q_{T}$ and have similar values of $p_{T}^{\gamma}$ of about $Q/2$. The $p_T$ cuts interfere with the expected Sudakov broadening of $Q_T$ distributions with increasing diphoton invariant mass, as discussed in Section~. We further note that the asymmetry between the $p_{T}^{\gamma}$ cuts on the harder and softer photons is necessary in a fixed-order perturbative QCD calculation, but it is not required in the resummed calculation. At a fixed order of $\alpha_{s}$, asymmetry in the $p_{T}^{\gamma}$ cuts prevents instabilities in $d\sigma/dQ$ caused by logarithmic divergences in $d\sigma/dQ_{T}$ at small $Q_{T}$. Such instabilities are eliminated altogether once the small-$Q_{T}$ logarithmic terms are resummed to all orders of $\alpha_{s}$. Here we do not consider alternative $p_{T}^{\gamma}$ cuts, although experimental collaborations are encouraged to employ relaxed and/or symmetric cuts to increase the $\gamma\gamma$ event sample in their data analysis. \subsubsection{Resummed $Q_{T}$ distributions and average transverse momentum} \begin{figure*} \caption{Resummed transverse momentum distributions of photon pairs in various invariant mass bins at the LHC. The cuts listed in Eqs.~()~and~() are imposed.The $Q_T$ distribution for $70 < Q < 115$ GeV with an additional constraint $Q_T < Q$ is shown as a dotted line. } \end{figure*} Figure~ shows transverse momentum distributions of the photon pairs for various invariant masses. The average $\gamma\gamma$ transverse momentum grows with $Q$, as demonstrated by Fig. . However, the rate of the growth decreases monotonically with $Q,$ for similar reasons as at the Tevatron. The $\gamma\gamma$ distributions in $Q$ and $\Delta\varphi$ for different combinations of scattering subchannels and choices of theoretical parameters are discussed in Refs.~. In all ranges of $Q$, the $\gamma\gamma$ production rate is dominated by a large $qg$ contribution, accounting for about 50\ fixed-order (NLO) rate. Although this number depends on the choice of the factorization scheme and scale, and, on the other hand, separate treatment of the $q\bar q$ and $qg$ cross sections is not meaningful in the resummation calculation , it nonetheless reflects, in a crude way, the increased relative importance of the $qg$ cross section. The $gg+gq_{S}$ channel contributes about 25\ GeV (the location of the cutoff in $d\sigma/dQ$ due to the cuts on $p_{T}^{\gamma}$) and less at larger $Q.$ As at the Tevatron, the dependence of the cross sections on the resummation scheme is small~. The dependence on the nonperturbative model can also be neglected, as long as the nonperturbative function does not vary strongly with $x$~. \subsubsection{Final-state fragmentation and comparison with DIPHOX} The impact of the final-state fragmentation at the LHC can be evaluated if we compare our results with DIPHOX predictions. The transverse momentum and invariant mass distributions in the $q\bar{q}+qg$ channel from the two approaches are shown in Fig.~. In both calculations, quasi-experimental isolation removes direct NLO events with collinear final-state photons and partons when $Q_{T}>E_{T}^{iso}=15$ GeV, but not when $Q_{T}$ is below $E_{T}^{iso}$. Concentrating first on $\gamma\gamma$ events with $Q_{T}>E_{T}^{iso}$, we observe that, at $Q_{T}>80$ GeV\textbf{,} the resummed $q\bar{q}+qg$ cross section reduces to the direct fixed-order cross section, evaluated in the same way as in the DIPHOX code. Our resummed and the direct DIPHOX cross sections, shown as solid and dashed curves, respectively, in Fig.~(a) consequently agree well at large $Q_{T}$. At smaller $Q_{T},$ the resummed cross section is enhanced by towers of higher-order logarithmic contributions. On the other hand, the full $q\bar{q}+qg$ DIPHOX rate (shown as a dotted line) also includes single-photon fragmentation contributions, which add to the direct production cross section. For the nominal isolation parameters, the explicit fragmentation contribution constitutes about 25\ the full DIPHOX rate for $60<Q_{T}<120$ GeV. Its magnitude increases approximately linearly with the assumed $E_{T}^{iso}$ value. For $Q_{T}<E_{T}^{iso}$, the final-state collinear region of the direct contribution is regulated by the collinear subtraction prescription adopted in the resummation calculation, whereas the fragmentation singularity is subtracted from the direct contribution and replaced by photon fragmentation functions in the DIPHOX calculation. Subtraction of singularities in DIPHOX introduces integrable singularities in $d\sigma/dQ_{T}$ at different values of $Q_{T}$ below $E_{T}^{iso}$. The origin of the final-state logarithmic singularities at values of $Q_{T}$ below $E_{T}^{iso}$ is discussed in Refs.~. For $Q_{T}<E_{T}^{iso}$, the DIPHOX curves represent the average over singular contributions in this $Q_{T}$ interval. After the fragmentation singularity is subtracted, the DIPHOX direct contribution (dashed line) is on average below our resummed $q\bar{q}+qg$ rate (solid line) over most of the range of $Q_{T}$ shown in Fig.~(a). After integration over all $Q_{T}$, our resummed and DIPHOX $q\bar{q}+qg$ cross sections agree within 10-20\ with our resummed rate being below the DIPHOX rate at all $Q$. The largest difference occurs at the lowest values of $Q$ (below the cutoff), where the rates can differ by a factor of 2. In this region, corresponding to diphoton events with small $\Delta\varphi$ and $Q_{T}$ larger than $Q$, the photon fragmentation contributions included in the DIPHOX calculation are large in comparison to the direct rate. Finally, we note that the integrated rate in DIPHOX is more stable with respect to variations in $E_{T}^{iso}$ than the differential distributions in DIPHOX, because $E_{T}^{iso}$ dependence for $Q_{T}>E_{T}^{iso}$ is canceled to a good degree by $E_{T}^{iso}$ dependence for $Q_{T}<E_{T}^{iso}.$ \begin{figure*} \\ (a)\hspace{3in}(b) \\ (c) \caption{Invariant mass, transverse momentum, and $\Delta\varphi$ distributions from our resummed calculation and from \textsc{DIPHOX} at the LHC. We show our fixed-order (dashed) and resummed (solid) distributions. All initial states are included in both calculations, and the single-$\gamma$ fragmentation contributions are included in \textsc{DIPHOX}.} \end{figure*} To obtain the final $\gamma\gamma$ production cross sections, after inclusion of all channels, we combine the respective $q\bar{q}+qg$ results with the resummed NLO $gg+gq_{S}$ cross section in our case and with the LO $gg$ cross section in the DIPHOX case. The distributions in the $\gamma\gamma$ invariant mass $Q,$ the transverse momentum $Q_{T}$, and the azimuthal angle separation $\Delta\varphi$ in the lab frame are shown in Fig.~. For the cuts chosen, the LO $gg$ and the resummed $gg+gq_{S}$ total rates constitute about 9\ mass distributions (Fig.~(a)) are brought closer to one another as a result of the inclusion of the $gg+gq_{S}$ contribution in the resummed calculation. For $Q_{T}\neq0$, the full DIPHOX $Q_{T}$ distribution in Fig.~(b) is determined entirely by direct plus fragmentation contributions (the same as in Fig.~(a)), because the LO $gg$ cross section contributes at $Q_{T}=0$ only. In contrast, our resummed $gg+gq_{S}$ contribution modifies the event rate at all $Q_{T}$. The resummed and DIPHOX rates are in a reasonable agreement for $1.5\lesssim\Delta\varphi\lesssim2.5$, as shown in Fig.~(c). In the $\Delta\varphi\rightarrow\pi$ limit, the fixed-order rates in DIPHOX diverge because of the singularities at small $Q_{T}$, while our resummed rate yields a finite value. For $\Delta\varphi<1.5$, the DIPHOX cross section is enhanced by photon fragmentation contributions. As at the energy of the Tevatron, theoretical uncertainties are greater at small $\Delta\varphi$. Predictions are most reliable when $Q_{T}<Q$ (and the angles $\theta_*$ and $\varphi_*$ are away from 0 or $\pi$). With the $Q_{T}<Q$ cut imposed, the uncertain large-$Q_T$ photon fragmentation contributions are suppressed, and the resummed and DIPHOX cross sections agree well at large $Q_T$ (cf. Fig.~(b)). The $Q_T$ distribution in the interval $70 < Q < 115 $ GeV with the $Q_{T}<Q$ constraint is shown in Fig.~ by a dotted curve. Distributions in the other two mass bins in Fig.~ are essentially not affected by this cut in the $Q_T$ range presented. \begin{figure*} \\ (a)\hspace{3in}(b) \caption{Invariant mass and transverse momentum distributions from our resummed, NLO, and DIPHOX calculations at the LHC, with the $Q_T < Q$ constraint imposed.} \end{figure*} Our calculation captures the dominant contributions to $\gamma\gamma$ production at the LHC. However, as we noted, direct $qg$ scattering, evaluated at order ${\mathcal{O}}(\alpha_{s})$ in our calculation, is the leading scattering channel in the region relevant for the Higgs boson search at the LHC. It is important to emphasize that the final-state collinear radiation is not the main reason behind the enhancement of the $qg$ rate, which is increased predominantly by contributions from non-singular phase space regions. Consequently, the $q\bar{q}+qg$ direct rate is only weakly sensitive to adjustments in the isolation parameters $E_{T}^{iso}$ and $\Delta R$ . The unknown ${\mathcal{O}}(\alpha_{s}^{2})$ contributions to $qg$ scattering may be non-negligible, and it would be valuable to compute them in the future when LHC data are available. \begin{figure*} \caption{Comparison of the normalized Higgs boson signal and diphoton background distributions at the LHC, both computed at NNLL accuracy. The Higgs boson mass is taken to be $m_{H}=130$ GeV, and the background is calculated for $128<Q<132$ GeV.} \end{figure*} \subsection{Comparison with Higgs boson signal distributions} We highlight some similarities and differences between the production spectra for the Higgs boson signal and the QCD background discussed in this paper. We focus on the diphoton decay mode of a SM Higgs boson produced from the dominant gluon-fusion mechanism, $gg\rightarrow h^{0}\rightarrow\gamma\gamma$, where the Higgs boson production cross section is calculated at the same order of precision as the QCD continuum background. We include initial-state QCD contributions at $O(\alpha_{s}^{3})$ (NLO) and resummed contributions at NNLL accuracy. These contributions are also coded in \textsc{ResBos}~, and we can apply the same cuts on the momenta of the photons to the signal and background. Our findings should remain broadly applicable after the NNLO corrections to Higgs boson production are included. We compute the background in the range $128<Q<132$ GeV, and the signal at a fixed Higgs boson mass $m_{H}=130$ GeV. We impose the kinematic selection $Q_{T}<Q$, but its influence is not important at the large values of diphoton mass of interest here. The cross section times branching ratio for the Higgs boson signal is substantially smaller than the QCD continuum. To better illustrate their differences, Fig.~ presents distributions normalized to the respective total rates. The top-left panel shows normalized transverse momentum distributions of photon pairs. The signal and background peak at about 12 and 5 GeV, respectively. The average values of $Q_{T}$ are 26 and 23 GeV, computed over the range 0 to 75 GeV. Differences in the shapes of these $Q_{T}$ spectra can be attributed to the different structure of the leading terms in the initial-state Sudakov exponents and to the effects of final-state photon fragmentation. The Higgs boson signal is controlled by the characteristics of the $gg+gq_{S}$ initial state, whereas the continuum is controlled primarily by the $q\bar{q}+qg$ initial state. Because the dominant Sudakov coefficient $\mathcal{A}_{q}^{(k)} \propto C_{F}$ in the $q\bar{q}$ case is smaller than $\mathcal{A}_{g}^{(k)}= (C_{A}/C_{F})\mathcal{A}_{q}^{(k)}$ in the $gg$ case, the resummed $q\bar{q}+qg$ initial-state radiation produces narrower $Q_{T}$ distributions than $gg+gq_{S}$ initial-state radiation. About 80\ channel, implying a narrower $Q_{T}$ distribution of the background, if based on the value of $\mathcal{A}^{(k)}$ alone. The continuum background contribution is also enhanced by final-state radiation in $qg$ scattering. The $Q_{T}$ profile of the final-state collinear terms depends more on the isolation model (including $E_{T}^{iso}$ and $\Delta R$) than on the initial-state Sudakov exponent. For the nominal ATLAS cuts, the final-state collinear contribution in our calculation hardens the background $Q_{T}$ distribution, diminishing its difference from the Higgs boson signal distribution. More effective isolation may reduce the impact of the final-state radiation on $Q_{T}$ distributions. Another difference between the signal and continuum is observed in the distribution in the azimuthal angle of the photons, such as the angle $\varphi_{*}$ in the Collins-Soper frame shown in the top-right panel of Fig.~. This distribution is qualitatively the same if integrated over all $Q_{T}$, as in Fig.~, or integrated above some minimal $Q_{T}$ value, as in an experimental measurement. Without isolation imposed, the spin-$0$ Higgs boson signal must be flat in $\varphi_{*},$ but the QCD background peaks toward $\varphi_{*}=0$ and $\pi$ (i.e., $\sin\varphi_{*}=0$) as a result of the final-state $qg$ singularity. \footnote{By definition, the recoil parton 5 always lies in the $Oxz$ plane (has zero azimuthal angle) in the Collins-Soper frame. For the final-state singularity to occur at NLO, the photons should be in the same plane with 5, i.e., have $\sin\varphi_{*}=0$.}${}^,$ \footnote{One of the resummed structure functions for the $gg$ background is modulated by $\cos2\varphi_{*}$ (see Sec.~), but we neglect this modulation in our present calculation. } Isolation cuts suppress both the signal and the background for $\sin\varphi_{*}<\sin\Delta R$. The result is a signal distribution with a broad peak near $\varphi_{*}=\pi/2$, while the background favors values of $\varphi_{*}$ near $0$ and $\pi$. A selection of events with $\varphi_{*}$ in the vicinity of $\pi/2$, and $Q_{T}$ large enough, helps to reduce the impact of the $qg$ background.In the lab frame, a related distribution is in the variable $\left|\varphi_{3T}-\varphi_{4T}\right|,$where $\varphi_{iT}$ is the azimuthal angle between $\vec{p}_{T}^{\gamma_{i}}$ and $\vec{Q}_{T}$. The signal (background) processes tend to have more events with large (small) magnitude of $\left|\varphi_{3T}-\varphi_{4T}\right|$. A third potential discriminator between the signal and background is the difference in the rapidities $\Delta y=y_{hard}-y_{soft}$ of the photons with harder and softer values of $p_{T}^{\gamma}$ in the lab frame, calculated on an event by event basis. This distribution is displayed in the lower-left frame of Fig.~. The background distribution peaks at the origin, while the signal is almost flat over a wide range of $\Delta y$. Different spin correlations in the decay of a spin-0 Higgs boson from those characteristic of QCD background processes are the source of this distinction. Discrimination based on this difference can improve the statistical significance of the signal~. We note that our resummed calculation does not exhibit the kinematic singularity at $\Delta y\approx2$ present in the finite-order cross section and obvious in Fig. 10 of Ref.~, where the distribution with respect to $y^*\equiv\Delta y/2$ is shown. The discontinuity in $d\sigma/dQ_{T}$ caused by the finite-order approximation is resummed in our calculation, yielding a smooth result. The rapidity difference is related to the scattering angle in the Collins-Soper frame: $\tanh(\Delta y/2)=\cos\theta_{*}$ when $Q_{T}$ is zero. The $\cos\theta_{*}$ distribution is shown in the lower-right frame of Fig.~. The difference between the signal and background rates is even more pronounced in this variable, clearly expressing the difference in the spin correlations of the systems producing the photons. A comparison of $Q_{T}$ distributions in the top-left panel of Fig.~ suggests that the signal versus background ratio would be enhanced if a cut is made to restrict $Q_{T}>10$~GeV. After applying this cut, we may again examine the distributions in the rapidity difference of the two photons, the scattering angle in the Collins-Soper frame, and the azimuthal angle distribution of the photons in the Collins-Soper frame. The results are qualitatively similar to those in Fig.~ and are not shown here. A more efficient procedure to increase the Higgs boson discovery significance is to apply a simultaneous likelihood analysis to several kinematic distributions. Based on the present discussion, we would argue that the resummed $Q_{T}$, $\varphi_{*}$, and $\cos\theta_{*}$ distributions are good discriminators between the Higgs boson signal and background in such an analysis. \section{Conclusions } The theoretical study of continuum diphoton production in hadron collisions is interesting and valuable for several reasons{\bf:} there are data from the CDF and D\O~ collaborations at Fermilab with the promise of larger event samples; there are new theoretical challenges associated with all-orders soft-gluon resummation of two-loop amplitudes; and continuum diphotons are a large standard-model background above which one may observe the products of Higgs boson decay into a pair of photons at the LHC. In this paper and Refs.~, we present our calculation of the fully differential cross section $d\sigma/(dQdQ_{T}dyd\Omega_{*})$ as a function of the mass $Q$, transverse momentum $Q_T$, and rapidity $y$ of the diphoton system, and of the polar and azimuthal angles of the individual photons in the diphoton rest frame. Our basic QCD hard-scattering subprocesses are all computed at next-to-leading order (NLO) in the strong coupling strength $\alpha_s$, and we include the state-of-art resummation of initial-state gluon radiation to all orders in $\alpha_s$, valid to next-to-next-to-leading logarithmic accuracy (NNLL). Resummation is essential for a realistic and reliable calculation of the $Q_T$ dependence in the region of small and intermediate values of $Q_T$, where the cross section is greatest. It is also needed for stable estimates of the effects of experimental acceptance on distributions in the diphoton invariant mass and other variables. Our analytical results are included in a fully updated \textsc{ResBos} code~. This numerical program allows us to impose selections on the transverse momenta and angles of the final photons, in order to match those employed by the CDF and D\O~ collaborations, as well as those anticipated in experiments at the LHC. Our predictions are especially pertinent in the region $Q_{T}\lesssim Q$. We show that our results at the Tevatron and at the LHC are insensitive to the choice of the resummation scheme and of the nonperturbative functions required by the integration into the region of large impact parameter. The published collider data are presented in the form of singly differential distributions. We follow suit in order to make comparisons, and we find excellent agreement with data, as shown in Sec.~. We recommend that more differential studies be made, and, to motivate these, we present predictions for the changes expected in the $Q_T$ distribution as a function of mass $Q$, and for the dependence of the mean transverse momentum on $Q$. We make predictions for continuum diphoton mass, transverse momentum, and angular distributions at the energy of the LHC. Moreover, we contrast in Fig.~ the shapes of some of these distributions with those expected from the decay of a Higgs boson. The distinct features of the signal and background suggest that that the Higgs boson discovery significance can be increased via a simultaneous likelihood analysis of several kinematic distributions, particularly the resummed $Q_{T}$, $\varphi_{*}$, and $\cos\theta_{*}$ distributions. Another approach to the computation of continuum diphoton production is presented by the \textsc{DIPHOX} collaboration~. This calculation includes both the direct production of photons from hard-scattering processes and the photons produced from fragmentation of (anti-)quarks or gluons. It is valid at NLO, except for the $gg$ subprocess, which is included at leading order only. The \textsc{DIPHOX} code is useful for rates integrated over transverse momentum, but it is not designed to predict the transverse momentum distribution or other distributions sensitive to the region in which the transverse momentum of the diphoton pair is small. Compared to a fixed-order calculation, such as direct photon pair calculation in \textsc{DIPHOX}, our calculation improves the theoretical prediction for event distributions which are sensitive to the region of low $Q_{T}$. Furthermore, our calculation includes the NLO contribution from the combined $gg+gq_{S}$ channel, leading to more accurate predictions at the LHC, where the $gg+gq_{S}$ contribution is generally not small. Only {\em isolated}, not inclusive, photons are identified experimentally. While it is straightforward to define an isolated photon in a given experiment, it is challenging to devise a theoretical prescription that can match the experimental definition, short of first understanding the long-distance dynamics of QCD. The isolated diphoton production rate is modeled in the \textsc{DIPHOX} code by including explicit photon fragmentation function contributions at NLO accuracy. A shortcoming of this approach (as well as of our method for treating isolation) is that one cannot accurately represent photon fragmentation without including final-state parton showering in the presence of isolation constraints. There is inevitable ambiguity and uncertainty in the choice of the ``isolation energy'' used to define an isolated photon theoretically for comparison with the isolated photon measured experimentally. As shown in Sec.~, the \textsc{DIPHOX} cross section can vary by a large factor in some regions of phase space at the Tevatron when $E_{T}^{iso}$ is changed from 1\,GeV to 4\,GeV. Our approach is to concentrate on physical observables which are less sensitive to the fragmentation contributions. We apply the {}``collinear subtraction'' prescription or the {}``smooth-cone isolation'' prescription to define an isolated photon in our calculation. We find good agreement with the data, except in the region with small $Q$ and $\Delta\varphi<\pi/2$, consistent with our theoretical expectation that higher-order direct photon production and photon fragmentation contributions can strongly modify the rate of diphoton pairs in this region. We suggest that much better agreement with current and future data will be obtained if an addition requirement of $Q_{T}<Q$ is applied. With this cut, the fragmentation contributions are largely suppressed. With the cut $Q_{T}<Q$ cut applied to the Tevatron data, the enhancement at low $\Delta\varphi$ and intermediate $Q_{T}$ (the shoulder region) should disappear. We urge the CDF and D\O\, collaborations to apply these cuts in future analyses of the diphoton data. In our calculation, we identify an interesting spin-flip contribution (with $\cos2\varphi_{*}$ dependence) in the $gg$ channel, cf. Ref.~, and we suggest that measurements be made of the distribution of $\varphi_{*}$ as a function of $Q_{T}$. All-orders resummation of the gluon spin-flip contribution may be needed when a larger statistical sample of diphoton data is available. The contributions from $qg+{\bar{q}}g$ processes become more important at the LHC than at the Tevatron, and calculations at a higher order of precision may be warranted eventually. To improve the theoretical prediction in the region of phase space with $Q_{T}<E_{T}^{iso}$ and $\varphi_{*}\sim0$ or $\pi$, a joint resummation calculation is needed in which the effects of both the initial- and final-state multiple parton emissions are treated simultaneously. Although we emphasize that better agreement of our predictions with data should be apparent if the selection $Q_{T}<Q$ is made, we also point out that the region $Q_{T} > Q$ should manifest very interesting physics of a different sort. Additional logarithmic singularities of the form $\log(Q/Q_{T})$ are encountered in the region $Q_{T} \gg Q$. These logarithms are associated with the fragmentation of a parton carrying large transverse momentum $Q_{T}$ into a system of small invariant mass $Q$ , a light $\gamma\gamma$ pair in our case. Small-$Q$ $\gamma\gamma$ fragmentation of this kind is not implemented yet in theoretical models. Experimental study of the region $Q_{T} \gg Q$ may offer the opportunity to measure the parton to two-photon fragmentation function $D_{\gamma\gamma}(z_{1},z_{2})$. \section*{Acknowledgments} Research in the High Energy Physics Division at Argonne is supported in part by US Department of Energy, Division of High Energy Physics, Contract DE-AC02-06CH11357. The work of C.-P. Y. is supported by the U. S. National Science Foundation under grant PHY-0555545. C. B. thanks the Fermilab Theoretical Physics Department, where a part of this work was done, for its hospitality and financial support. The diagrams in Figs.~ and~ were drawn with aid of the program \textsc{JaxoDraw} . \appendix \section{Summary of perturbative coefficients } In this appendix we present an overview of the perturbative QCD expressions for the resummed and asymptotic cross sections used in our computation. The functions $\mathcal{A}_{a}(C_{1},\bar{\mu}),$ $\mathcal{B}_{a}(C_{1},C_{2},\bar{\mu}),$ $\mathcal{C}_{a/a_{1}}(x,b;C_{1}/C_{2},\mu)$, and $h_{a}(Q,\theta_{*})$ are introduced in Sec.~. These functions are derived as perturbative expansions in the small parameter $\alpha_{s}/\pi$:\begin{eqnarray*} & & \mathcal{A}_{a}(C_{1},\bar{\mu})=\sum_{n=1}^{\infty}\mathcal{A}_{a}^{(n)}(C_{1})\left(\frac{\alpha_{s}(\bar{\mu})}{\pi}\right)^{n};\,\mathcal{B}_{a}(C_{1},C_{2},\bar{\mu})=\sum_{n=1}^{\infty}\mathcal{B}_{a}^{(n)}(C_{1},C_{2})\left(\frac{\alpha_{s}(\bar{\mu})}{\pi}\right)^{n};\\ & & \mathcal{C}_{a/a_{1}}\left(x,b;\frac{C_{1}}{C_{2}},\mu\right)=\sum_{n=0}^{\infty}\mathcal{C}_{a/a_{1}}^{(n)}(x,b\mu,\frac{C_{1}}{C_{2}})\left(\frac{\alpha_{s}(\mu)}{\pi}\right)^{n};\, h_{a}(Q,\theta_{*})=\sum_{n=0}^{\infty}h_{a}^{(n)}(\theta_{*})\left(\frac{\alpha_{s}(Q)}{\pi}\right)^{n}.\end{eqnarray*} The value of a perturbative coefficient $F^{(n)}$ for a set of scales $C_{1}/b$ and $C_{2}Q$ can be expressed in terms of its value $F^{(n,c)}$ obtained for the {}``canonical'' combination $C_{1}=c_{0}$ and $C_{2}=1.$ Here $c_{0}\equiv2e^{-\gamma_{E}}\approx1.123$, where $\gamma_{E}=0.5772\dots$ is the Euler constant. The relationships between $F^{(n)}$ and $F^{(n,c)}$ take the form \begin{eqnarray} \mathcal{A}_{a}^{(1)}(C_{1}) & = & \mathcal{{\mathcal{A}}}_{a}^{(1,c)};\\ \mathcal{A}_{a}^{(2)}(C_{1}) & = & \mathcal{{\mathcal{A}}}_{a}^{(2,c)}-\mathcal{{\mathcal{A}}}_{a}^{(1,c)}\beta_{0}\ln\frac{c_{0}}{C_{1}};\\ \mathcal{A}_{a}^{(3)}(C_{1}) & = & \mathcal{A}_{a}^{(3,c)}-2\mathcal{A}_{a}^{(2,c)}\beta_{0}\ln\frac{c_{0}}{C_{1}}-\frac{\mathcal{A}_{a}^{(1,c)}}{2}\beta_{1}\ln\frac{c_{0}}{C_{1}}+\mathcal{A}_{a}^{(1,c)}\beta_{0}^{2}\left(\ln\frac{c_{0}}{C_{1}}\right)^{2};\\ \mathcal{B}_{a}^{(1)}(C_{1},C_{2}) & = & \mathcal{B}_{a}^{(1,c)}-\mathcal{A}_{a}^{(1,c)}\ln\frac{c_{0}^{2}C_{2}^{2}}{C_{1}^{2}};\\ \mathcal{B}_{a}^{(2)}(C_{1},C_{2}) & = & \mathcal{B}_{a}^{(2,c)}-\mathcal{{\mathcal{A}}}_{a}^{(2,c)}\ln\frac{c_{0}^{2}C_{2}^{2}}{C_{1}^{2}}\nonumber \\ & + & \beta_{0}\left[\mathcal{{\mathcal{A}}}_{a}^{(1,c)}\ln^{2}\frac{c_{0}}{C_{1}}+\mathcal{B}_{a}^{(1,c)}\ln C_{2}-\mathcal{{\mathcal{A}}}_{a}^{(1,c)}\ln^{2}C_{2}\right];\\ \mathcal{C}_{a/a_{1}}^{(1)}(x,b\mu,\frac{C_{1}}{C_{2}}) & = & \mathcal{C}_{a/a_{1}}^{(1,c)}(x)+\delta_{aa_{1}}\delta(1-x)\left(\frac{\mathcal{B}_{a}^{(1,c)}}{2}\ln\frac{c_{0}^{2}C_{2}^{2}}{C_{1}^{2}}-\frac{\mathcal{A}_{a}^{(1,c)}}{4}\left(\ln\frac{c_{0}^{2}C_{2}^{2}}{C_{1}^{2}}\right)^{2}\right)\nonumber \\ & - & P_{a/a_{1}}(x)\ln\frac{\mu b}{c_{0}}.\end{eqnarray} They depend on the QCD beta-function coefficients $\beta_{0}=(11N_{c}-2N_{f})/6$, $\beta_{1}=(17N_{c}^{2}-5N_{c}N_{f}-3C_{F}N_{f})/6$ for $N_{c}$ colors and $N_{f}$ active quark flavors, with $C_{F}=(N_{c}^{2}-1)/(2N_{c})=4/3$ for $N_{c}=3$. The relevant ${\mathcal{O}}(\alpha_{s})$ splitting functions $P_{a/a_{1}}(x)$ are\begin{eqnarray} & & P_{q/q}=C_{F}\left(\frac{1+z^{2}}{1-x}\right)_{+};\, P_{q/g}=\frac{1}{2}(1+2x+2x^{2});\, P_{g/q_{S}}=C_{F}\frac{(1-x)^{2}+1}{x};\\ & & P_{g/g}=2C_{A}\left[\frac{x}{(1-x)_{+}}+\frac{1-x}{x}+x(1-x)\right]+\beta_{0}\delta(1-x).\end{eqnarray} The coefficients $h^{(1)}(\theta_{*})$, ${\cal B}^{(2)}$, and $\mathcal{C}^{(1)}$ depend on the resummation scheme. The hard-scattering function is \begin{equation} h_{a}(Q,\theta_{*})=1+\delta_{s}\frac{\alpha_{s}(Q)}{\pi}\frac{\mathcal{V}_{a}(\theta_{*})}{4}+...,\end{equation} where $\delta_{s}=0$ in the CSS scheme and $\delta_{s}=1$ in the CFG scheme. The functions $\mathcal{V}_{q}(\theta_{*})$ for $q\bar{q}\rightarrow\gamma\gamma$ scattering and $\mathcal{V}_{g}(\theta_{*})$ for $gg\rightarrow\gamma\gamma$ scattering are derived in Refs.~ and , respectively. For the $q\bar{q}+qg$ initial state, we obtain the following expressions for the coefficients $\mathcal{A}$, $\mathcal{B}$, and ${\mathcal{C}}$: \begin{eqnarray} \mathcal{A}_{q}^{(1,c)} & = & C_{F};\nonumber \\ \mathcal{A}_{q}^{(2,c)} & = & C_{F}\left[\left(\frac{67}{36}-\frac{\pi^{2}}{12}\right)C_{A}-\frac{5}{9}T_{R}N_{f}\right];\\ \mathcal{A}_{q}^{(3,c)} & = & \frac{C_{F}^{2}N_{f}}{2}\left(\zeta(3)-\frac{55}{48}\right)-\frac{C_{F}N_{f}^{2}}{108}+C_{A}^{2}C_{F}\left(\frac{11\zeta(3)}{24}+\frac{11\pi^{4}}{720}-\frac{67\pi^{2}}{216}+\frac{245}{96}\right)\nonumber \\ & + & C_{A}C_{F}N_{f}\left(-\frac{7\zeta(3)}{12}+\frac{5\pi^{2}}{108}-\frac{209}{432}\right);\nonumber \\ \mathcal{B}_{q}^{(1,c)} & = & -\frac{3}{2}C_{F};\nonumber \\ \mathcal{B}_{q}^{(2,c)} & = & -\frac{1}{2}\left[{C_{F}}^{2}\,\left(\frac{3}{8}-\frac{\pi^{2}}{2}+6\zeta(3)\right)\right.+C_{F}C_{A}\left(\frac{17}{24}+\frac{11\pi^{2}}{18}-3\zeta(3)\right)\nonumber \\ & - & \left.C_{F}N_{f}T_{R}\left(\frac{1}{6}+\frac{2\pi^{2}}{9}\right)\right]+\beta_{0}\left[\frac{C_{F}\pi^{2}}{12}+(1-\delta_{s})\frac{\mathcal{V}_{q}(\theta_{*})}{4}\right];\nonumber \\ \mathcal{C}_{j/k}^{(0)}(x) & = & \delta_{jk}\delta(1-x);\,\,\mathcal{C}_{j/g}^{(0)}(x)=0;\nonumber \\ \mathcal{C}_{j/k}^{(1,c)}(x) & = & \delta_{jk}\left\{ \frac{C_{F}}{2}(1-x)+\delta(1-x)(1-\delta_{s})\frac{\mathcal{V}_{q}(\theta_{*})}{4}\right\} ;\nonumber \\ \mathcal{C}_{j/g}^{(1,c)}(x) & = & {\frac{1}{2}}x(1-x).\end{eqnarray} Here $C_{A}=N_{c},$ $T_{R}=1/2$, and the Riemann constant $\zeta(3)=1.202\dots$ . The ${\mathcal{C}}$ functions are given for $j,k=u,\bar{u},d,\bar{d},\dots$. These coefficients are taken from . Similarly, the ${\mathcal{A}}$, ${\mathcal{B}}$, and ${\mathcal{C}}$ coefficients in the $gg+gq_{S}$ channel are \begin{eqnarray} \mathcal{A}_{g}^{(k,c)} & = & (C_{A}/C_{F})\mathcal{A}_{q}^{(k,c)},\mbox{ for }k=1,2,3;\nonumber \\ \mathcal{B}_{g}^{(1,c)} & = & -\beta_{0};\nonumber \\ \mathcal{B}_{g}^{(2,c)} & = & -\frac{1}{2}\Biggl[C_{A}^{2}\left(\frac{8}{3}+3\zeta(3)\right)-C_{F}T_{R}N_{f}-\frac{4}{3}C_{A}T_{R}N_{f}\Biggr]+\beta_{0}\left[\frac{C_{A}\pi^{2}}{12}+(1-\delta_{s})\frac{\mathcal{V}_{g}(\theta_{*})}{4}\right];\nonumber \\ \mathcal{C}_{g/a}^{(0)}\left(x\right) & = & \delta_{ga}\delta(1-x);\,\,\mathcal{C}_{g/g}^{(1,c)}\left(x\right)=\delta(1-x)(1-\delta_{s})\frac{\mathcal{V}_{g}(\theta_{*})}{4};\,\,\mathcal{C}_{g/q_{S}}^{(1,c)}\left(x\right)=\frac{C_{F}}{2}x.\end{eqnarray} These coefficients are taken from Refs.~. \section{Components of the asymptotic cross sections } In Sec.~ we introduce asymptotic small-$Q_{T}$ approximations for the $q\bar{q}+qg$ and $gg+gq_{S}$ NLO cross sections,\begin{equation} A_{q\bar{q}}(Q,Q_{T},y,\Omega_{*})=\sum_{i=u,\bar{u},d,\bar{d},...}\frac{\Sigma_{i}(\theta_{*})}{S}\left\{ \delta(\vec{Q}_{T})F_{i,\delta}(Q,y,\theta_{*})+F_{i,+}(Q,y,Q_{T})\right\} ,\end{equation} and\begin{eqnarray} A_{gg}(Q,Q_{T},y,\Omega_{*}) & = & \frac{1}{S}\Biggl\{\Sigma_{g}(\theta_{*})\left[\delta(\vec{Q}_{T})F_{g,\delta}(Q,y,\theta_{*})+F_{g,+}(Q,y,Q_{T})\right]\nonumber \\ & & \hspace{12pt}+\Sigma_{g}^{\prime}(\theta_{*},\varphi_{*})F_{g}^{\prime}(Q,y,Q_{T})\Biggr\}.\end{eqnarray} The functions $F$ in these equations are defined as\begin{eqnarray} & & F_{i,\delta}(Q,y,\theta_{*})\equiv f_{q_{i}/h_{1}}(x_{1},\mu_{F})f_{\bar{q}_{i}/h_{2}}(x_{2},\mu_{F})\left(1+2\frac{\alpha_{s}}{\pi}h_{q}^{(1)}(\theta_{*})\right)\nonumber \\ & + & \frac{\alpha_{s}}{\pi}\Biggl\{\left(\left[\mathcal{C}_{q_{i}/a}^{(1,c)}\otimes f_{a/h_{1}}\right](x_{1},\mu_{F})-\left[P_{q_{i}/a}\otimes f_{a/h_{1}}\right](x_{1},\mu_{F})\,\ln\frac{\mu_{F}}{Q}\right)\, f_{\bar{q}_{i}/h_{2}}(x_{2},\mu_{F})\nonumber \\ & & \hspace{26pt}+f_{q_{i}/h_{1}}(x_{1},\mu_{F})\left(\left[\mathcal{C}_{\bar{q}_{i}/a}^{(1,c)}\otimes f_{a/h_{2}}\right](x_{2},\mu_{F})-\left[P_{\bar{q}_{i}/a}\otimes f_{a/h_{2}}\right](x_{2},\mu_{F})\,\ln\frac{\mu_{F}}{Q}\right)\Biggr\};\end{eqnarray} \begin{eqnarray} F_{q,+} & = & \frac{1}{2\pi}\frac{\alpha_{s}}{\pi}\Biggl\{ f_{q_{i}/h_{1}}(x_{1},\mu_{F})f_{\bar{q}_{i}/h_{2}}(x_{2},\mu_{F})\left(\mathcal{A}_{q}^{(1,c)}\left[\frac{1}{Q_{T}^{2}}\ln\frac{Q^{2}}{Q_{T}^{2}}\right]_{+}+\mathcal{B}_{q}^{(1,c)}\left[\frac{1}{Q_{T}^{2}}\right]_{+}\right)\nonumber \\ & + & \left[\frac{1}{Q_{T}^{2}}\right]_{+}\Bigl(\left[P_{q_{i}/a}\otimes f_{a/h_{1}}\right](x_{1},\mu_{F})\, f_{\bar{q}_{i}/h_{2}}(x_{2},\mu_{F})\nonumber \\ & & \hspace{47pt}+f_{q_{i}/h_{1}}(x_{1},\mu_{F})\left[P_{\bar{q}_{i}/a}\otimes f_{a/h_{2}}\right](x_{2},\mu_{F})\Bigr)\Biggr\};\end{eqnarray} \begin{eqnarray} & & F_{g,\delta}\equiv f_{g/h_{1}}(x_{1},\mu_{F})f_{g/h_{2}}(x_{2},\mu_{F})\left(1+2\frac{\alpha_{s}}{\pi}h_{g}^{(1)}(\theta_{*})\right)\nonumber \\ & + & \frac{\alpha_{s}}{\pi}\Biggl\{\left(\left[\mathcal{C}_{g/a}^{(1,c)}\otimes f_{a/h_{1}}\right](x_{1},\mu_{F})-\left[P_{g/a}\otimes f_{a/h_{1}}\right](x_{1},\mu_{F})\,\ln\frac{\mu_{F}}{Q}\right)\, f_{g/h_{2}}(x_{2},\mu_{F})\nonumber \\ & & \hspace{27pt}+f_{g/h_{1}}(x_{1},\mu_{F})\left(\left[\mathcal{C}_{g/a}^{(1,c)}\otimes f_{a/h_{2}}\right](x_{2},\mu_{F})-\left[P_{g/a}\otimes f_{a/h_{2}}\right](x_{2},\mu_{F})\,\ln\frac{\mu_{F}}{Q}\right)\Biggr\};\end{eqnarray} \begin{eqnarray} F_{g,+} & = & \frac{1}{2\pi}\frac{\alpha_{s}}{\pi}\Biggl\{ f_{g/h_{1}}(x_{1},\mu_{F})f_{g/h_{2}}(x_{2},\mu_{F})\left(\mathcal{A}_{g}^{(1,c)}\left[\frac{1}{Q_{T}^{2}}\ln\frac{Q^{2}}{Q_{T}^{2}}\right]_{+}+\mathcal{{\mathcal{B}}}_{g}^{(1,c)}\left[\frac{1}{Q_{T}^{2}}\right]_{+}\right)\nonumber \\ & + & \left[\frac{1}{Q_{T}^{2}}\right]_{+}\Bigl(\left[P_{g/a}\otimes f_{a/h_{1}}\right](x_{1},\mu_{F})\, f_{g/h_{2}}(x_{2},\mu_{F})\nonumber \\ & & \hspace{47pt}+f_{g/h_{1}}(x_{1},\mu_{F})\left[P_{g/a}\otimes f_{a/h_{2}}\right](x_{2},\mu_{F})\Bigr)\Biggr\};\end{eqnarray} and\begin{eqnarray} F_{g,+}^{\prime} & = & \frac{1}{2\pi}\frac{\alpha_{s}}{\pi}\left[\frac{1}{Q_{T}^{2}}\right]_{+}\Bigl(\left[P_{g/g}^{\prime}\otimes f_{g/h_{1}}\right](x_{1},\mu_{F})\, f_{g/h_{2}}(x_{2},\mu_{F})\nonumber \\ & & \hspace{77pt}+f_{g/h_{1}}(x_{1},\mu_{F})\left[P_{g/g}^{\prime}\otimes f_{g/h_{2}}\right](x_{2},\mu_{F})\Bigr)\Biggr\}.\end{eqnarray} Expressions for the coefficients $\mathcal{A}_{a}^{(1,c)},$ $\mathcal{B}_{a}^{(1,c)}$, $h_{a}^{(1)}(\theta_{*}),$ $\mathcal{C}_{a/a^{\prime}}^{(1,c)}(x),$ and splitting functions $P_{a/c}(x)$, are listed in Appendix~. Summation over all relevant parton flavors $a^{\prime}=g,u,\bar{u,}d,\bar{d},...$ for $a=q$ and $a^{\prime}=g,q_{S}$ for $a=g$ is assumed. In addition, the $\varphi_{*}$-dependent part $\Sigma_{g}^{\prime}(\theta_{*},\varphi_{*})F_{g}^{\prime}(Q,y,Q_{T})$ of the $gg+gq_{S}$ asymptotic cross section $A_{gg}$ contains a splitting function\begin{equation} P_{gg}^{\prime}(x)=2C_{A}(1-x)/x,\end{equation} contributed by the interference of splitting amplitudes with opposite gluon polarizations in the helicity amplitude formalism . The origin and behavior of this spin-flip function are discussed in Ref.~. } \begin{thebibliography}{49} \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi \expandafter\ifx\csname bibnamefont\endcsname\relax \def\bibnamefont#1{#1}\fi \expandafter\ifx\csname bibfnamefont\endcsname\relax \def\bibfnamefont#1{#1}\fi \expandafter\ifx\csname citenamefont\endcsname\relax \def\citenamefont#1{#1}\fi \expandafter\ifx\csname url\endcsname\relax \def\url#1{\texttt{#1}}\fi \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi \providecommand{\bibinfo}[2]{#2} \providecommand{\eprint}[2][]{} \bibitem[{\citenamefont{Acosta et~al.}(2005)}]{Acosta:2004sn} \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Acosta}} \bibnamefont{et~al.} (\bibinfo{collaboration}{CDF Collaboration}), \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{95}}, \bibinfo{pages}{022003} (\bibinfo{year}{2005}). \bibitem[{\citenamefont{Balazs et~al.}(2006)\citenamefont{Balazs, Berger, Nadolsky, and Yuan}}]{Balazs:2006cc} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Balazs}}, \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{Berger}}, \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Nadolsky}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.~P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B637}}, \bibinfo{pages}{235} (\bibinfo{year}{2006}). \bibitem[{\citenamefont{Nadolsky et~al.}(2007)\citenamefont{Nadolsky, Balazs, Berger, and Yuan}}]{Nadolsky:2007ba} \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Nadolsky}}, \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Balazs}}, \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Berger}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.~P.} \bibnamefont{Yuan}} (\bibinfo{year}{2007}), \eprint{hep-ph/0702003}. \bibitem[{\citenamefont{Aurenche et~al.}(1985)\citenamefont{Aurenche, Douiri, Baier, Fontannaz, and Schiff}}]{Aurenche:1985yk} \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Aurenche}}, \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Douiri}}, \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Baier}}, \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fontannaz}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Schiff}}, \bibinfo{journal}{Z. Phys.} \textbf{\bibinfo{volume}{C29}}, \bibinfo{pages}{459} (\bibinfo{year}{1985}). \bibitem[{\citenamefont{Bailey et~al.}(1992)\citenamefont{Bailey, Owens, and Ohnemus}}]{Bailey:1992br} \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Bailey}}, \bibinfo{author}{\bibfnamefont{J.~F.} \bibnamefont{Owens}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Ohnemus}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D46}}, \bibinfo{pages}{2018} (\bibinfo{year}{1992}). \bibitem[{\citenamefont{Berger et~al.}(1984)\citenamefont{Berger, Braaten, and Field}}]{Berger:1983yi} \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{Berger}}, \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Braaten}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{R.~D.} \bibnamefont{Field}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B239}}, \bibinfo{pages}{52} (\bibinfo{year}{1984}). \bibitem[{\citenamefont{Balazs et~al.}(2000)\citenamefont{Balazs, Nadolsky, Schmidt, and Yuan}}]{Balazs:1999yf} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Balazs}}, \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Nadolsky}}, \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Schmidt}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.-P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B489}}, \bibinfo{pages}{157} (\bibinfo{year}{2000}). \bibitem[{\citenamefont{de~Florian and Kunszt}(1999)}]{deFlorian:1999tp} \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{de~Florian}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{Z.}~\bibnamefont{Kunszt}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B460}}, \bibinfo{pages}{184} (\bibinfo{year}{1999}). \bibitem[{\citenamefont{Bern et~al.}(2001)\citenamefont{Bern, De~Freitas, and Dixon}}]{Bern:2001df} \bibinfo{author}{\bibfnamefont{Z.}~\bibnamefont{Bern}}, \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{De~Freitas}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{L.~J.} \bibnamefont{Dixon}}, \bibinfo{journal}{JHEP} \textbf{\bibinfo{volume}{09}}, \bibinfo{pages}{037} (\bibinfo{year}{2001}). \bibitem[{\citenamefont{Bern et~al.}(2002)\citenamefont{Bern, Dixon, and Schmidt}}]{Bern:2002jx} \bibinfo{author}{\bibfnamefont{Z.}~\bibnamefont{Bern}}, \bibinfo{author}{\bibfnamefont{L.~J.} \bibnamefont{Dixon}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Schmidt}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D66}}, \bibinfo{pages}{074018} (\bibinfo{year}{2002}). \bibitem[{\citenamefont{Collins et~al.}(1985)\citenamefont{Collins, Soper, and Sterman}}]{Collins:1984kg} \bibinfo{author}{\bibfnamefont{J.~C.} \bibnamefont{Collins}}, \bibinfo{author}{\bibfnamefont{D.~E.} \bibnamefont{Soper}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Sterman}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B250}}, \bibinfo{pages}{199} (\bibinfo{year}{1985}). \bibitem[{\citenamefont{Balazs et~al.}(1998)\citenamefont{Balazs, Berger, Mrenna, and Yuan}}]{Balazs:1997hv} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Balazs}}, \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{Berger}}, \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Mrenna}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.-P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D57}}, \bibinfo{pages}{6934} (\bibinfo{year}{1998}). \bibitem[{\citenamefont{Nadolsky and Schmidt}(2003)}]{Nadolsky:2002gj} \bibinfo{author}{\bibfnamefont{P.~M.} \bibnamefont{Nadolsky}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.~R.} \bibnamefont{Schmidt}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B558}}, \bibinfo{pages}{63} (\bibinfo{year}{2003}). \bibitem[{\citenamefont{Binoth et~al.}(2000)\citenamefont{Binoth, Guillet, Pilon, and Werlen}}]{Binoth:1999qq} \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Binoth}}, \bibinfo{author}{\bibfnamefont{J.~P.} \bibnamefont{Guillet}}, \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Pilon}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Werlen}}, \bibinfo{journal}{Eur. Phys. J.} \textbf{\bibinfo{volume}{C16}}, \bibinfo{pages}{311} (\bibinfo{year}{2000}). \bibitem[{\citenamefont{Collins and Soper}(1977)}]{Collins:1977iv} \bibinfo{author}{\bibfnamefont{J.~C.} \bibnamefont{Collins}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~E.} \bibnamefont{Soper}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D16}}, \bibinfo{pages}{2219} (\bibinfo{year}{1977}). \bibitem[{\citenamefont{Collins and Soper}(1982)}]{Collins:1981va} \bibinfo{author}{\bibfnamefont{J.~C.} \bibnamefont{Collins}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~E.} \bibnamefont{Soper}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B197}}, \bibinfo{pages}{446} (\bibinfo{year}{1982}). \bibitem[{\citenamefont{Collins and Soper}(1981)}]{Collins:1981uk} \bibinfo{author}{\bibfnamefont{J.~C.} \bibnamefont{Collins}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~E.} \bibnamefont{Soper}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B193}}, \bibinfo{pages}{381} (\bibinfo{year}{1981}). \bibitem[{\citenamefont{Konychev and Nadolsky}(2006)}]{Konychev:2005iy} \bibinfo{author}{\bibfnamefont{A.~V.} \bibnamefont{Konychev}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{P.~M.} \bibnamefont{Nadolsky}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B633}}, \bibinfo{pages}{710} (\bibinfo{year}{2006}). \bibitem[{\citenamefont{Pumplin et~al.}(2002)}]{Pumplin:2002vw} \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Pumplin}} \bibnamefont{et~al.}, \bibinfo{journal}{JHEP} \textbf{\bibinfo{volume}{07}}, \bibinfo{pages}{012} (\bibinfo{year}{2002}). \bibitem[{\citenamefont{Berge et~al.}(2005)\citenamefont{Berge, Nadolsky, Olness, and Yuan}}]{Berge:2004nt} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Berge}}, \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Nadolsky}}, \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Olness}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.-P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D72}}, \bibinfo{pages}{033015} (\bibinfo{year}{2005}). \bibitem[{\citenamefont{Catani et~al.}(2001)\citenamefont{Catani, de~Florian, and Grazzini}}]{Catani:2000vq} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Catani}}, \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{de~Florian}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Grazzini}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B596}}, \bibinfo{pages}{299} (\bibinfo{year}{2001}). \bibitem[{\citenamefont{Berger et~al.}(1996)\citenamefont{Berger, Guo, and Qiu}}]{Berger:1996vy} \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{Berger}}, \bibinfo{author}{\bibfnamefont{X.}~\bibnamefont{Guo}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Qiu}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D54}}, \bibinfo{pages}{5470} (\bibinfo{year}{1996}). \bibitem[{\citenamefont{Catani et~al.}(1998)\citenamefont{Catani, Fontannaz, and Pilon}}]{Catani:1998yh} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Catani}}, \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fontannaz}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Pilon}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D58}}, \bibinfo{pages}{094025} (\bibinfo{year}{1998}). \bibitem[{\citenamefont{Catani et~al.}(2002)\citenamefont{Catani, Fontannaz, Guillet, and Pilon}}]{Catani:2002ny} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Catani}}, \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fontannaz}}, \bibinfo{author}{\bibfnamefont{J.~P.} \bibnamefont{Guillet}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Pilon}}, \bibinfo{journal}{JHEP} \textbf{\bibinfo{volume}{05}}, \bibinfo{pages}{028} (\bibinfo{year}{2002}). \bibitem[{\citenamefont{Frixione}(1998)}]{Frixione:1998jh} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Frixione}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B429}}, \bibinfo{pages}{369} (\bibinfo{year}{1998}). \bibitem[{\citenamefont{Catani and Seymour}(1997)}]{Catani:1996vz} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Catani}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.~H.} \bibnamefont{Seymour}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B485}}, \bibinfo{pages}{291} (\bibinfo{year}{1997}). \bibitem[{\citenamefont{Berger et~al.}(1998)\citenamefont{Berger, Gordon, and Klasen}}]{Berger:1998ev} \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{Berger}}, \bibinfo{author}{\bibfnamefont{L.~E.} \bibnamefont{Gordon}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Klasen}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D58}}, \bibinfo{pages}{074012} (\bibinfo{year}{1998}). \bibitem[{\citenamefont{Berger et~al.}(2002)\citenamefont{Berger, Qiu, and Zhang}}]{Berger:2001wr} \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{Berger}}, \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Qiu}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{X.}~\bibnamefont{Zhang}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D65}}, \bibinfo{pages}{034006} (\bibinfo{year}{2002}). \bibitem[{\citenamefont{Ladinsky and Yuan}(1994)}]{Ladinsky:1993zn} \bibinfo{author}{\bibfnamefont{G.~A.} \bibnamefont{Ladinsky}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.-P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D50}}, \bibinfo{pages}{4239} (\bibinfo{year}{1994}). \bibitem[{\citenamefont{Landry et~al.}(2003)\citenamefont{Landry, Brock, Nadolsky, and Yuan}}]{Landry:2002ix} \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Landry}}, \bibinfo{author}{\bibfnamefont{R.}~\bibnamefont{Brock}}, \bibinfo{author}{\bibfnamefont{P.~M.} \bibnamefont{Nadolsky}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.~P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D67}}, \bibinfo{pages}{073016} (\bibinfo{year}{2003}). \bibitem[{\citenamefont{Balazs and Yuan}(1997)}]{Balazs:1997xd} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Balazs}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.-P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D56}}, \bibinfo{pages}{5558} (\bibinfo{year}{1997}). \bibitem[{\citenamefont{Balazs}(1999)}]{Balazs:1999gh} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Balazs}} (\bibinfo{year}{1999}), \eprint{hep-ph/9906422}. \bibitem[{\citenamefont{Eidelman et~al.}(2004)}]{Eidelman:2004wy} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Eidelman}} \bibnamefont{et~al.} (\bibinfo{collaboration}{Particle Data Group}), \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B592}}, \bibinfo{pages}{1} (\bibinfo{year}{2004}). \bibitem[{\citenamefont{Bourhis et~al.}(1998)\citenamefont{Bourhis, Fontannaz, and Guillet}}]{Bourhis:1997yu} \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Bourhis}}, \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Fontannaz}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{J.~P.} \bibnamefont{Guillet}}, \bibinfo{journal}{Eur. Phys. J.} \textbf{\bibinfo{volume}{C2}}, \bibinfo{pages}{529} (\bibinfo{year}{1998}). \bibitem[{\citenamefont{Dyer}()}]{Dyer:2006} \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Dyer}}, \bibinfo{note}{private communication}. \bibitem[{\citenamefont{Berger and Qiu}(2003)}]{Berger:2002ut} \bibinfo{author}{\bibfnamefont{E.~L.} \bibnamefont{Berger}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Qiu}}, \bibinfo{journal}{Phys. Rev.} \textbf{\bibinfo{volume}{D67}}, \bibinfo{pages}{034026} (\bibinfo{year}{2003}). \bibitem[{\citenamefont{\protect ATLAS~Collaboration}(1999)}]{ATLAS:1999fr} \bibinfo{author}{\bibnamefont{\protect ATLAS~Collaboration}}, \emph{\bibinfo{title}{\protect{ATLAS detector and physics performance. Technical design report. Vol.~2}}} (\bibinfo{year}{1999}), \bibinfo{note}{\protect{CERN-LHCC-99-15}}. \bibitem[{\citenamefont{Balazs and Yuan}(2000)}]{Balazs:2000wv} \bibinfo{author}{\bibfnamefont{C.}~\bibnamefont{Balazs}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.-P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B478}}, \bibinfo{pages}{192} (\bibinfo{year}{2000}). \bibitem[{\citenamefont{Catani et~al.}(2003)\citenamefont{Catani, de~Florian, Grazzini, and Nason}}]{Catani:2003zt} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Catani}}, \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{de~Florian}}, \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Grazzini}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Nason}}, \bibinfo{journal}{JHEP} \textbf{\bibinfo{volume}{07}}, \bibinfo{pages}{028} (\bibinfo{year}{2003}). \bibitem[{\citenamefont{Bozzi et~al.}(2006)\citenamefont{Bozzi, Catani, de~Florian, and Grazzini}}]{Bozzi:2005wk} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Bozzi}}, \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Catani}}, \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{de~Florian}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Grazzini}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B737}}, \bibinfo{pages}{73} (\bibinfo{year}{2006}). \bibitem[{\citenamefont{Binosi and Theussl}(2004)}]{Binosi:2003yf} \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Binosi}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Theussl}}, \bibinfo{journal}{Comput. Phys. Commun.} \textbf{\bibinfo{volume}{161}}, \bibinfo{pages}{76} (\bibinfo{year}{2004}). \bibitem[{\citenamefont{de~Florian and Grazzini}(2000)}]{deFlorian:2000pr} \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{de~Florian}} \bibnamefont{and} \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Grazzini}}, \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{85}}, \bibinfo{pages}{4678} (\bibinfo{year}{2000}). \bibitem[{\citenamefont{Moch et~al.}(2004)\citenamefont{Moch, Vermaseren, and Vogt}}]{Moch:2004pa} \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Moch}}, \bibinfo{author}{\bibfnamefont{J.~A.~M.} \bibnamefont{Vermaseren}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Vogt}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B688}}, \bibinfo{pages}{101} (\bibinfo{year}{2004}). \bibitem[{\citenamefont{Yuan}(1992)}]{Yuan:1991we} \bibinfo{author}{\bibfnamefont{C.-P.} \bibnamefont{Yuan}}, \bibinfo{journal}{Phys. Lett.} \textbf{\bibinfo{volume}{B283}}, \bibinfo{pages}{395} (\bibinfo{year}{1992}). \bibitem[{\citenamefont{Vogt et~al.}(2004)\citenamefont{Vogt, Moch, and Vermaseren}}]{Vogt:2004mw} \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Vogt}}, \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Moch}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{J.~A.~M.} \bibnamefont{Vermaseren}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B691}}, \bibinfo{pages}{129} (\bibinfo{year}{2004}). \bibitem[{\citenamefont{Bern et~al.}(1993)\citenamefont{Bern, Dixon, and Kosower}}]{Bern:1993mq} \bibinfo{author}{\bibfnamefont{Z.}~\bibnamefont{Bern}}, \bibinfo{author}{\bibfnamefont{L.~J.} \bibnamefont{Dixon}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~A.} \bibnamefont{Kosower}}, \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{70}}, \bibinfo{pages}{2677} (\bibinfo{year}{1993}). \bibitem[{\citenamefont{Bern et~al.}(1994{\natexlab{a}})\citenamefont{Bern, Chalmers, Dixon, and Kosower}}]{Bern:1993qk} \bibinfo{author}{\bibfnamefont{Z.}~\bibnamefont{Bern}}, \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Chalmers}}, \bibinfo{author}{\bibfnamefont{L.~J.} \bibnamefont{Dixon}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~A.} \bibnamefont{Kosower}}, \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{72}}, \bibinfo{pages}{2134} (\bibinfo{year}{1994}{\natexlab{a}}). \bibitem[{\citenamefont{Bern et~al.}(1994{\natexlab{b}})\citenamefont{Bern, Dixon, Dunbar, and Kosower}}]{Bern:1994zx} \bibinfo{author}{\bibfnamefont{Z.}~\bibnamefont{Bern}}, \bibinfo{author}{\bibfnamefont{L.~J.} \bibnamefont{Dixon}}, \bibinfo{author}{\bibfnamefont{D.~C.} \bibnamefont{Dunbar}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~A.} \bibnamefont{Kosower}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B425}}, \bibinfo{pages}{217} (\bibinfo{year}{1994}{\natexlab{b}}). \bibitem[{\citenamefont{Bern et~al.}(1995)\citenamefont{Bern, Dixon, and Kosower}}]{Bern:1994fz} \bibinfo{author}{\bibfnamefont{Z.}~\bibnamefont{Bern}}, \bibinfo{author}{\bibfnamefont{L.~J.} \bibnamefont{Dixon}}, \bibnamefont{and} \bibinfo{author}{\bibfnamefont{D.~A.} \bibnamefont{Kosower}}, \bibinfo{journal}{Nucl. Phys.} \textbf{\bibinfo{volume}{B437}}, \bibinfo{pages}{259} (\bibinfo{year}{1995}). \end{thebibliography}
|
0704.0002
|
Title: Sparsity-certifying Graph Decompositions
Abstract: We describe a new algorithm, the $(k,\ell)$-pebble game with colors, and use
it obtain a characterization of the family of $(k,\ell)$-sparse graphs and
algorithmic solutions to a family of problems concerning tree decompositions of
graphs. Special instances of sparse graphs appear in rigidity theory and have
received increased attention in recent years. In particular, our colored
pebbles generalize and strengthen the previous results of Lee and Streinu and
give a new proof of the Tutte-Nash-Williams characterization of arboricity. We
also present a new decomposition that certifies sparsity based on the
$(k,\ell)$-pebble game with colors. Our work also exposes connections between
pebble game algorithms and previous sparse graph algorithms by Gabow, Gabow and
Westermann and Hendrickson.
Body: \pdfoutput=1 \maketitle \begin{abstract} We describe a new algorithm, the $(k,\ell)$-pebble game with colors, and use it to obtain a characterization of the family of $(k,\ell)$-sparse graphs and algorithmic solutions to a family of problems concerning tree decompositions of graphs. Special instances of sparse graphs appear in rigidity theory and have received increased attention in recent years. In particular, our colored pebbles generalize and strengthen the previous results of Lee and Streinu and give a new proof of the Tutte-Nash-Williams characterization of arboricity. We also present a new decomposition that certifies sparsity based on the $(k,\ell)$-pebble game with colors. Our work also exposes connections between pebble game algorithms and previous sparse graph algorithms by Gabow , Gabow and Westermann and Hendrickson . \end{abstract} \section{Introduction and preliminaries} The focus of this paper is decompositions of $(k,\ell)$-sparse graphs into edge-disjoint subgraphs that certify sparsity. We use {\bf graph} to mean a multigraph, possibly with loops. We say that a graph is {\bf $(k,\ell)$-sparse} if no subset of $n'$ vertices spans more than $kn'-\ell$ edges in the graph; a $(k,\ell)$-sparse graph with $kn'-\ell$ edges is {\bf $(k,\ell)$-tight}. We call the range $k\le \ell\le 2k-1$ the upper range of sparse graphs and $0\le \ell\le k$ the lower range. In this paper, we present efficient algorithms for finding decompositions that certify sparsity in the upper range of $\ell$. Our algorithms also apply in the lower range, which was already addressed by . A decomposition certifies the sparsity of a graph if the sparse graphs and graphs admitting the decomposition coincide. Our algorithms are based on a new characterization of sparse graphs, which we call the {\bf pebble game with colors}. The pebble game with colors is a simple graph construction rule that produces a sparse graph along with a sparsity-certifying decomposition. We define and study a canonical class of pebble game constructions, which correspond to previously studied decompositions of sparse graphs into edge disjoint trees. Our results provide a unifying framework for all the previously known special cases, including Nash-Williams-Tutte and . Indeed, in the lower range, canonical pebble game constructions capture the properties of the augmenting paths used in matroid union and intersection algorithms. Since the sparse graphs in the upper range are not known to be unions or intersections of the matroids for which there are efficient augmenting path algorithms, these do not easily apply in the upper range. Pebble game with colors constructions may thus be considered a strengthening of augmenting paths to the upper range of matroidal sparse graphs. \subsection{Sparse graphs} A graph is {\bf $(k,\ell)$-sparse} if for any non-empty subgraph with $m'$ edges and $n'$ vertices, \( m' \le kn'-\ell. \) We observe that this condition implies that $0\le \ell\le 2k-1$, and from now on in this paper we will make this assumption. A sparse graph that has $n$ vertices and exactly $kn-\ell$ edges is called {\bf tight}. For a graph $G=(V,E)$, and $V'\subset V$, we use the notation $\grsp (V')$ for the number of edges in the subgraph induced by $V'$. In a directed graph, $\out (V')$ is the number of edges with the tail in $V'$ and the head in $V-V'$; for a subgraph induced by $V'$, we call such an edge an {\bf out-edge}. There are two important types of subgraphs of sparse graphs. A {\bf block} is a tight subgraph of a sparse graph. A {\bf component} is a maximal block. \begin{table} \begin{tabular} {|l|l|} \hline {\bf Term} & {\bf Meaning} \\ \hline \hline Sparse graph $G$ & Every non-empty subgraph on $n'$ vertices has $\le kn'-\ell$ edges\\ \hline Tight graph $G$ & $G=(V,E)$ is sparse and $\card{V}=n$, $\card{E}=kn-\ell$ \\ \hline Block $H$ in $G$ & $G$ is sparse, and $H$ is a tight subgraph \\ \hline Component $H$ of $G$ & $G$ is sparse and $H$ is a maximal block \\ \hline Map-graph & Graph that admits an out-degree-exactly-one orientation \\ \hline $(k,\ell)$-maps-and-trees & Edge-disjoint union of $\ell$ trees and $(k-\ell)$ map-grpahs\\ \hline \ellteekay & Union of $\ell$ trees, each vertex is in exactly $k$ of them \\ \hline Set of tree-pieces of an $\ellteekay$ induced on $V'\subset V$ & Pieces of trees in the \ellteekay spanned by $E(V')$ \\ \hline Proper \ellteekay & Every $V'\subset V$ contains $\ge\ell$ pieces of trees from the \ellteekay \\ \hline \end{tabular} \caption{Sparse graph and decomposition terminology used in this paper.} \end{table} Table summarizes the sparse graph terminology used in this paper. \subsection{Sparsity-certifying decompositions} A $k$-arborescence is a graph that admits a decomposition into $k$ edge-disjoint spanning trees. \reffig{colored-3-tree} shows an example of a $3$-arborescence. The $k$-arborescent graphs are described by the well-known theorems of Tutte and Nash-Williams as exactly the $(k,k)$-tight graphs. A {\bf map-graph} is a graph that admits an orientation such that the out-degree of each vertex is exactly one. A $k$-{\bf map-graph} is a graph that admits a decomposition into $k$ edge-disjoint map-graphs. \reffig{colored-2-map} shows an example of a 2-map-graphs; the edges are oriented in one possible configuration certifying that each color forms a map-graph. Map-graphs may be equivalently defined (see, e.g., ) as having exactly one cycle per connected component.}sz in . In the matroid literature map-graphs are sometimes known as bases of the bicycle matroid or spanning pseudoforests.} A {\bf $(k,\ell)$-maps-and-trees} is a graph that admits a decomposition into $k-\ell$ edge-disjoint map-graphs and $\ell$ spanning trees. Another characterization of map-graphs, which we will use extensively in this paper, is as the $(1,0)$-tight graphs . The $k$-map-graphs are evidently $(k,0)$-tight, and show that the converse holds as well. A \ellteekay is a decomposition into $\ell$ edge-disjoint (not necessarily spanning) trees such that each vertex is in exactly $k$ of them. \reffig{2-3-t-a-t} shows an example of a $3{\mathsf T}2$. Given a subgraph $G'$ of a \ellteekay graph $G$, the {\bf set of tree-pieces} in $G'$ is the collection of the components of the trees in $G$ induced by $G'$ (since $G'$ is a subgraph each tree may contribute multiple pieces to the set of tree-pieces in $G'$). We observe that these tree-pieces may come from the same tree or be single-vertex ``empty trees.'' It is also helpful to note that the definition of a tree-piece is {\it relative to a specific subgraph}. An \ellteekay decomposition is \textbf{proper} if the set of tree-pieces in any subgraph $G'$ has size at least $\ell$. \reffig{2-3-t-a-t} shows a graph with a $3\mathsf{T}2$ decomposition; we note that one of the trees is an isolated vertex in the bottom-right corner. The subgraph in \reffig{2-3-t-a-t-b} has three black tree-pieces and one gray tree-piece: an isolated vertex at the top-right corner, and two single edges. These count as three tree-pieces, even though they come from the same back tree when the whole graph in considered. \reffig{2-3-t-a-t-c} shows another subgraph; in this case there are three gray tree-pieces and one black one. Table contains the decomposition terminology used in this paper. \paragraph{The decomposition problem.} We define the {\bf decomposition} problem for sparse graphs as taking a graph as its input and producing as output, a decomposition that can be used to certify sparsity. In this paper, we will study three kinds of outputs: maps-and-trees; proper \ellteekay decompositions; and the pebble-game-with-colors decomposition, which is defined in the next section. \section{Historical background} The well-known theorems of Tutte and Nash-Williams relate the $(k,k)$-tight graphs to the existence of decompositions into edge-disjoint spanning trees. Taking a matroidal viewpoint, Edmonds gave another proof of this result using matroid unions. The equivalence of maps-and-trees graphs and tight graphs in the lower range is shown using matroid unions in , and matroid augmenting paths are the basis of the algorithms for the lower range of . In rigidity theory a foundational theorem of Laman shows that $(2,3)$-tight (Laman) graphs correspond to generically minimally rigid bar-and-joint frameworks in the plane. Tay proved an analogous result for body-bar frameworks in any dimension using $(k,k)$-tight graphs. Rigidity by counts motivated interest in the upper range, and Crapo proved the equivalence of Laman graphs and proper $3\mathsf{T}2$ graphs. Tay used this condition to give a direct proof of Laman's theorem and generalized the $3\mathsf{T}2$ condition to all $\ell\mathsf{T}k$ for $k\le \ell\le 2k-1$. Haas studied \ellteekay decompositions in detail and proved the equivalence of tight graphs and proper \ellteekay graphs for the general upper range. We observe that aside from our new pebble-game-with-colors decomposition, all the combinatorial characterizations of the upper range of sparse graphs, including the counts, have a geometric interpretation . A pebble game algorithm was first proposed in as an elegant alternative to Hendrickson's Laman graph algorithms . Berg and Jordan , provided the formal analysis of the pebble game of and introduced the idea of playing the game on a directed graph. Lee and Streinu generalized the pebble game to the entire range of parameters $0\le \ell\le 2k-1$, and left as an open problem using the pebble game to find sparsity certifying decompositions. \section{The pebble game with colors} Our {\bf pebble game with colors} is a set of rules for constructing graphs indexed by nonnegative integers $k$ and $\ell$. We will use the pebble game with colors as the basis of an efficient algorithm for the decomposition problem later in this paper. Since the phrase ``with colors'' is necessary only for comparison to , we will omit it in the rest of the paper when the context is clear. We now present the pebble game with colors. The game is played by a single player on a fixed finite set of vertices. The player makes a finite sequence of moves; a move consists in the addition and/or orientation of an edge. At any moment of time, the state of the game is captured by a directed graph $H$, with colored pebbles on vertices and edges. The edges of $H$ are colored by the pebbles on them. While playing the pebble game all edges are directed, and we use the notation $vw$ to indicate a directed edge from $v$ to $w$. We describe the pebble game with colors in terms of its initial configuration and the allowed moves. \medskip {\bf Initialization:} In the beginning of the pebble game, $H$ has $n$ vertices and no edges. We start by placing $k$ pebbles on each vertex of $H$, one of each color $c_i$, for $i=1,2,\ldots,k$. {\bf Add-edge-with-colors:} Let $v$ and $w$ be vertices with at least $\ell+1$ pebbles on them. Assume (w.l.o.g.) that $v$ has at least one pebble on it. Pick up a pebble from $v$, add the oriented edge $vw$ to $E(H)$ and put the pebble picked up from $v$ on the new edge. \reffig{colored-add-edge} shows examples of the {\bf add-edge} move. {\bf Pebble-slide:} Let $w$ be a vertex with a pebble $p$ on it, and let $vw$ be an edge in $H$. Replace $vw$ with $wv$ in $E(H)$; put the pebble that was on $vw$ on $v$; and put $p$ on $wv$. Note that the color of an edge can change with a {\bf pebble-slide} move. \reffig{colored-pebble-slide} shows examples. The convention in these figures, and throughout this paper, is that pebbles on vertices are represented as colored dots, and that edges are shown in the color of the pebble on them. From the definition of the {\bf pebble-slide} move, it is easy to see that a particular pebble is always either on the vertex where it started or on an edge that has this vertex as the tail. However, when making a sequence of {\bf pebble-slide} moves that reverse the orientation of a path in $H$, it is sometimes convenient to think of this path reversal sequence as bringing a pebble from the end of the path to the beginning. \medskip The output of playing the pebble game is its complete configuration. {\bf Output:} At the end of the game, we obtain the directed graph $H$, along with the location and colors of the pebbles. Observe that since each edge has exactly one pebble on it, the pebble game configuration colors the edges. We say that the underlying undirected graph $G$ of $H$ is {\bf constructed} by the $(k,\ell)$-pebble game or that $H$ is a {\bf pebble-game graph}. Since each edge of $H$ has exactly one pebble on it, the pebble game's configuration partitions the edges of $H$, and thus $G$, into $k$ different colors. We call this decomposition of $H$ a {\bf pebble-game-with-colors decomposition}. \reffig{k4-not-canonical-colors} shows an example of a $(2,2)$-tight graph with a pebble-game decomposition. \medskip Let $G=(V,E)$ be pebble-game graph with the coloring induced by the pebbles on the edges, and let $G'$ be a subgraph of $G$. Then the coloring of $G$ induces a set of monochromatic connected subgraphs of $G'$ (there may be more than one of the same color). Such a monochromatic subgraph is called a \textbf{map-graph-piece} of $G'$ if it contains a cycle (in $G'$) and a \textbf{tree-piece} of $G'$ otherwise. The \textbf{set of tree-pieces} of $G'$ is the collection of tree-pieces induced by $G'$. As with the corresponding definition for $\ellteekay$s, the set of tree-pieces is defined \emph{relative to a specific subgraph}; in particular a tree-piece may be part of a larger cycle that includes edges not spanned by $G'$. The properties of pebble-game decompositions are studied in Section , and \refthm{non-canonical-decomposition} shows that each color must be $(1,0)$-sparse. The orientation of the edges in \reffig{k4-not-canonical-colors} shows this. For example \reffig{k4-not-canonical-colors} shows a $(2,2)$-tight graph with one possible pebble-game decomposition. The whole graph contains a gray tree-piece and a black tree-piece that is an isolated vertex. The subgraph in \reffig{k4-not-canonical-colors-b} has a black tree and a gray tree, with the edges of the black tree coming from a cycle in the larger graph. In \reffig{k4-not-canonical-colors-c}, however, the black cycle does not contribute a tree-piece. All three tree-pieces in this subgraph are single-vertex gray trees. In the following discussion, we use the notation $\peb(v)$ for the number of pebbles on $v$ and $\peb_i(v)$ to indicate the number of pebbles of colors $i$ on $v$. Table lists the pebble game notation used in this paper. \begin{table} \begin{tabular} {|l|l|} \hline {\bf Notation} & {\bf Meaning} \\ \hline \hline $\grsp (V')$ & Number of edges spanned in $H$ by $V'\subset V$; i.e. $\card{E_{H}(V')}$\\ \hline $\peb (V')$ &Number of pebbles on $V'\subset V$ \\ \hline $\out (V')$ & Number of edges $vw$ in $H$ with $v\in V'$ and $w\in V-V'$ \\ \hline $\peb_{i} (v)$ &Number of pebbles of color $c_{i}$ on $v\in V$ \\ \hline $\out_{i} (v)$ & Number of edges $vw$ colored $c_i$ for $v\in V$ \\ \hline \end{tabular} \caption{Pebble game notation used in this paper.} \end{table} \section{Our Results} We describe our results in this section. The rest of the paper provides the proofs. Our first result is a strengthening of the pebble games of to include colors. It says that sparse graphs are exactly pebble game graphs. Recall that from now on, all pebble games discussed in this paper are our pebble game with colors unless noted explicitly. \begin{theorem}[{\bf Sparse graphs and pebble-game graphs coincide}] \labelthm{sparse-graphs-are-pebble-graphs} A graph $G$ is $(k,\ell)$-sparse with $0\le\ell\le 2k-1$ if and only if $G$ is a pebble-game graph. \end{theorem} Next we consider pebble-game decompositions, showing that they are a generalization of proper \ellteekay decompositions that extend to the entire matroidal range of sparse graphs. \begin{theorem}[{\bf The pebble-game-with-colors decomposition}] \labelthm{non-canonical-decomposition} A graph $G$ is a pebble-game graph if and only if it admits a decomposition into $k$ edge-disjoint subgraphs such that each is $(1,0)$-sparse and every subgraph of $G$ contains at least $\ell$ tree-pieces of the $(1,0)$-sparse graphs in the decomposition. \end{theorem} The $(1,0)$-sparse subgraphs in the statement of \refthm{non-canonical-decomposition} are the colors of the pebbles; thus \refthm{non-canonical-decomposition} gives a characterization of the pebble-game-with-colors decompositions obtained by playing the pebble game defined in the previous section. Notice the similarity between the requirement that the set of tree-pieces have size at least $\ell$ in \refthm{non-canonical-decomposition} and the definition of a proper $\ellteekay$. Our next results show that for {\it any} pebble-game graph, we can specialize its pebble game construction to generate a decomposition that is a maps-and-trees or proper \ellteekay. We call these specialized pebble game constructions {\bf canonical}, and using canonical pebble game constructions, we obtain new {\it direct} proofs of existing arboricity results. We observe \refthm{non-canonical-decomposition} that maps-and-trees are special cases of the pebble-game decomposition: both spanning trees and spanning map-graphs are $(1,0)$-sparse, and each of the spanning trees contributes at least one piece of tree to every subgraph. The case of proper \ellteekay graphs is more subtle; if each color in a pebble-game decomposition is a forest, then we have found a proper \ellteekay, but this class is a subset of all possible proper \ellteekay decompositions of a tight graph. We show that this class of proper \ellteekay decompositions is sufficient to certify sparsity. We now state the main theorem for the upper and lower range. \begin{theorem}[{\bf Main Theorem (Lower Range): Maps-and-trees coincide with pebble-game graphs}] \labelthm{canonical-decomposition-I} Let $0\le \ell\le k$. A graph $G$ is a tight pebble-game graph if and only if $G$ is a $(k,\ell)$-maps-and-trees. \end{theorem} \begin{theorem}[{\bf Main Theorem (Upper Range): Proper \ellteekay graphs coincide with pebble-game graphs}] Let $k\le \ell\le 2k-1$. A graph $G$ is a tight pebble-game graph if and only if it is a proper \ellteekay with $kn-\ell$ edges. \labelthm{canonical-decomposition-II} \end{theorem} As corollaries, we obtain the existing decomposition results for sparse graphs. \begin{corollary} [\textbf{Nash-Williams , Tutte , White and Whiteley }] \labelcor{m-a-t-equals-tight} Let $\ell\le k$. A graph $G$ is tight if and only if has a $(k,\ell)$-maps-and-trees decomposition. \end{corollary} \begin{corollary} [\textbf{Crapo , Haas }] \labelcor{t-a-t-equals-tight} Let $k\le \ell\le 2k-1$. A graph $G$ is tight if and only if it is a proper \ellteekay. \end{corollary} \paragraph{Efficiently finding canonical pebble game constructions.} The proofs of \refthm{canonical-decomposition-I} and \refthm{canonical-decomposition-II} lead to an obvious algorithm with $O(n^3)$ running time for the {\bf decomposition} problem. Our last result improves on this, showing that a canonical pebble game construction, and thus a maps-and-trees or proper \ellteekay decomposition can be found using a pebble game algorithm in $O(n^2)$ time and space. These time and space bounds mean that our algorithm can be combined with those of without any change in complexity. \section{Pebble game graphs} In this section we prove \refthm{sparse-graphs-are-pebble-graphs}, a strengthening of results from to the pebble game with colors. Since many of the relevant properties of the pebble game with colors carry over directly from the pebble games of , we refer the reader there for the proofs. We begin by establishing some invariants that hold during the execution of the pebble game. \begin{lemma}[{\bf Pebble game invariants}] During the execution of the pebble game, the following invariants are maintained in $H$: \begin{enumerate} \item[{\bf (I1)}] There are at least $\ell$ pebbles on $V$. \item[{\bf (I2)}] For each vertex $v$, $\grsp (v) + \out (v) + \peb (v)=k$. \item[{\bf (I3)}] For each $V'\subset V$, $\grsp (V')+\out (V')+\peb (V')=kn'$. \item[{\bf (I4)}] For every vertex $v\in V$, $\out_i (v)+\peb_i (v)=1$. \item[{\bf (I5)}] Every maximal path consisting only of edges with color $c_i$ ends in either the first vertex with a pebble of color $c_i$ or a cycle. \end{enumerate} \labellem{pebble-game-invariants} \end{lemma} \begin{proof} {\bf (I1)}, {\bf (I2)}, and {\bf (I3)} come directly from . {\bf (I4)} This invariant clearly holds at the initialization phase of the pebble game with colors. That {\bf add-edge} and {\bf pebble-slide} moves preserve {\bf (I4)} is clear from inspection. {\bf (I5)} By {\bf (I4)}, a monochromatic path of edges is forced to end only at a vertex with a pebble of the same color on it. If there is no pebble of that color reachable, then the path must eventually visit some vertex twice. \end{proof} From these invariants, we can show that the pebble game constructible graphs are sparse. \begin{lemma}[{\bf Pebble-game graphs are sparse }]\labellem{pebble-graphs-are-sparse} Let $H$ be a graph constructed with the pebble game. Then $H$ is sparse. If there are exactly $\ell$ pebbles on $V(H)$, then $H$ is tight. \end{lemma} The main step in proving that every sparse graph is a pebble-game graph is the following. Recall that by bringing a pebble to $v$ we mean reorienting $H$ with {\bf pebble-slide} moves to reduce the out degree of $v$ by one. \begin{lemma}[{\bf The $\ell+1$ pebble condition} ]\labellem{can-bring-another-pebble} Let $vw$ be an edge such that $H+vw$ is sparse. If $\peb (\{v,w\})<\ell+1$, then a pebble not on $\{v,w\}$ can be brought to either $v$ or $w$. \end{lemma} It follows that any sparse graph has a pebble game construction. \begin{restate}{sparse-graphs-are-pebble-graphs}[{\bf Sparse graphs and pebble-game graphs coincide}] A graph $G$ is $(k,\ell)$-sparse with $0\le\ell\le 2k-1$ if and only if $G$ is a pebble-game graph. \end{restate} \section{The pebble-game-with-colors decomposition} In this section we prove \refthm{non-canonical-decomposition}, which characterizes all pebble-game decompositions. We start with the following lemmas about the structure of monochromatic connected components in $H$, the directed graph maintained during the pebble game. \begin{lemma}[{\bf Monochromatic pebble game subgraphs are $(1,0)$-sparse}]\labellem{each-color-is-map-sparse} Let $H_i$ be the subgraph of $H$ induced by edges with pebbles of color $c_i$ on them. Then $H_i$ is $(1,0)$-sparse, for $i=1,\ldots,k$. \end{lemma} \begin{proof} By {\bf (I4)} $H_i$ is a set of edges with out degree at most one for every vertex. \end{proof} \begin{lemma}[{\bf Tree-pieces in a pebble-game graph}] \labellem{subtrees} Every subgraph of the directed graph $H$ in a pebble game construction contains at least $\ell$ monochromatic tree-pieces, and each of these is rooted at either a vertex with a pebble on it or a vertex that is the tail of an out-edge. \end{lemma} Recall that an out-edge from a subgraph $H'=(V',E')$ is an edge $vw$ with $v\in V'$ and $vw\notin E'$. \begin{proof} Let $H'=(V',E')$ be a non-empty subgraph of $H$, and assume without loss of generality that $H'$ is induced by $V'$. By {\bf (I3)}, $\out (V')+\peb (V')\ge \ell$. We will show that each pebble and out-edge tail is the root of a tree-piece. Consider a vertex $v\in V'$ and a color $c_i$. By {\bf (I4)} there is a unique monochromatic directed path of color $c_i$ starting at $v$. By {\bf (I5)}, if this path ends at a pebble, it does not have a cycle. Similarly, if this path reaches a vertex that is the tail of an out-edge also in color $c_i$ (i.e., if the monochromatic path from $v$ leaves $V'$), then the path cannot have a cycle in $H'$. Since this argument works for any vertex in any color, for each color there is a partitioning of the vertices into those that can reach each pebble, out-edge tail, or cycle. It follows that each pebble and out-edge tail is the root of a monochromatic tree, as desired. \end{proof} Applied to the whole graph \reflem{subtrees} gives us the following. \begin{lemma}[{\bf Pebbles are the roots of trees}]\labellem{roots} In any pebble game configuration, each pebble of color $c_i$ is the root of a (possibly empty) monochromatic tree-piece of color $c_i$. \end{lemma} {\bf Remark:} Haas showed in that in a \ellteekay, a subgraph induced by $n'\ge 2$ vertices with $m'$ edges has exactly $kn'-m'$ tree-pieces in it. \reflem{subtrees} strengthens Haas' result by extending it to the lower range and giving a construction that finds the tree-pieces, showing the connection between the $\ell+1$ pebble condition and the hereditary condition on proper \ellteekay. We conclude our investigation of arbitrary pebble game constructions with a description of the decomposition induced by the pebble game with colors. \begin{restate}{non-canonical-decomposition}[{\bf The pebble-game-with-colors decomposition}] A graph $G$ is a pebble-game graph if and only if it admits a decomposition into $k$ edge-disjoint subgraphs such that each is $(1,0)$-sparse and every subgraph of $G$ contains at least $\ell$ tree-pieces of the $(1,0)$-sparse graphs in the decomposition. \end{restate} \begin{proof} Let $G$ be a pebble-game graph. The existence of the $k$ edge-disjoint $(1,0)$-sparse subgraphs was shown in \reflem{each-color-is-map-sparse}, and \reflem{subtrees} proves the condition on subgraphs. For the other direction, we observe that a color $c_i$ with $t_i$ tree-pieces in a given subgraph can span at most $n-t_i$ edges; summing over all the colors shows that a graph with a pebble-game decomposition must be sparse. Apply \refthm{sparse-graphs-are-pebble-graphs} to complete the proof. \end{proof} {\bf Remark: } We observe that a pebble-game decomposition for a Laman graph may be read out of the bipartite matching used in Hendrickson's Laman graph extraction algorithm . Indeed, pebble game orientations have a natural correspondence with the bipartite matchings used in . Maps-and-trees are a special case of pebble-game decompositions for tight graphs: if there are no cycles in $\ell$ of the colors, then the trees rooted at the corresponding $\ell$ pebbles must be spanning, since they have $n-1$ edges. Also, if each color forms a forest in an upper range pebble-game decomposition, then the tree-pieces condition ensures that the pebble-game decomposition is a proper $\ellteekay$. In the next section, we show that the pebble game can be specialized to correspond to maps-and-trees and proper \ellteekay decompositions. \section{Canonical Pebble Game Constructions} In this section we prove the main theorems (\refthm{canonical-decomposition-I} and \refthm{canonical-decomposition-II}), continuing the investigation of decompositions induced by pebble game constructions by studying the case where a minimum number of monochromatic cycles are created. The main idea, captured in \reflem{kill-m2-moves-locally} and illustrated in \reffig{m2-meta-picture}, is to avoid creating cycles while collecting pebbles. We show that this is always possible, implying that monochromatic map-graphs are created only when we add more than $k(n'-1)$ edges to some set of $n'$ vertices. For the lower range, this implies that every color is a forest. Every decomposition characterization of tight graphs discussed above follows immediately from the main theorem, giving new proofs of the previous results in a unified framework. In the proof, we will use two specializations of the pebble game moves. The first is a modification of the {\bf add-edge} move. {\bf Canonical add-edge:} When performing an {\bf add-edge} move, cover the new edge with a color that is on both vertices if possible. If not, then take the highest numbered color present. The second is a restriction on which {\bf pebble-slide} moves we allow. {\bf Canonical pebble-slide:} A {\bf pebble-slide} move is allowed only when it does not create a monochromatic cycle. We call a pebble game construction that uses only these moves {\bf canonical}. In this section we will show that every pebble-game graph has a canonical pebble game construction (\reflem{can-kill-m1-moves} and \reflem{kill-m2-moves-locally}) and that canonical pebble game constructions correspond to proper \ellteekay and maps-and-trees decompositions (\refthm{canonical-decomposition-I} and \refthm{canonical-decomposition-II}). We begin with a technical lemma that motivates the definition of canonical pebble game constructions. It shows that the situations disallowed by the canonical moves are {\it all} the ways for cycles to form in the lowest $\ell$ colors. \begin{lemma}[{\bf Monochromatic cycle creation}]\labellem{how-maps-form} Let $v\in V$ have a pebble $p$ of color $c_{i}$ on it and let $w$ be a vertex in the same tree of color $c_i$ as $v$. A monochromatic cycle colored $c_{i}$ is created in exactly one of the following ways: \begin{enumerate} \item[{\bf (M1)}] The edge $vw$ is added with an {\bf add-edge} move. \item[{\bf (M2)}] The edge $wv$ is reversed by a {\bf pebble-slide} move and the pebble $p$ is used to cover the reverse edge $vw$. \end{enumerate} \end{lemma} \begin{proof} Observe that the preconditions in the statement of the lemma are implied by \reflem{pebble-game-invariants}. By \reflem{roots} monochromatic cycles form when the last pebble of color $c_{i}$ is removed from a connected monochromatic subgraph. {\bf (M1)} and {\bf (M2)} are the only ways to do this in a pebble game construction, since the color of an edge only changes when it is inserted the first time or a new pebble is put on it by a {\bf pebble-slide} move. \end{proof} \reffig{m1-create-map} and \reffig{m2-create-map} show examples of {\bf (M1)} and {\bf (M2)} map-graph creation moves, respectively, in a $(2,0)$-pebble game construction. We next show that if a graph has a pebble game construction, then it has a canonical pebble game construction. This is done in two steps, considering the cases {\bf (M1)} and {\bf (M2)} separately. The proof gives two constructions that implement the {\bf canonical add-edge} and {\bf canonical pebble-slide} moves. \begin{lemma}[{\bf The canonical add-edge move}] Let $G$ be a graph with a pebble game construction. Cycle creation steps of type {\bf (M1)} can be eliminated in colors $c_{i}$ for $1\le i\le \ell'$, where $\ell'=\min\{k,\ell\}$. \labellem{can-kill-m1-moves} \end{lemma} \begin{proof} For {\bf add-edge} moves, cover the edge with a color present on both $v$ and $w$ if possible. If this is not possible, then there are $\ell+1$ distinct colors present. Use the highest numbered color to cover the new edge. \end{proof} {\bf Remark:} We note that in the upper range, there is always a repeated color, so {\it no} {\bf canonical add-edge} moves create cycles in the upper range. The {\bf canonical pebble-slide} move is defined by a global condition. To prove that we obtain the same class of graphs using only {\bf canonical pebble-slide} moves, we need to extend \reflem{can-bring-another-pebble} to only canonical moves. The main step is to show that if there is {\it any } sequence of moves that reorients a path from $v$ to $w$, then there is a sequence of canonical moves that does the same thing. \begin{lemma}[{\bf The canonical pebble-slide move}]\labellem{kill-m2-moves-locally} Any sequence of {\bf pebble-slide} moves leading to an {\bf add-edge} move can be replaced with one that has no {\bf (M2)} steps and allows the same {\bf add-edge} move. \end{lemma} In other words, if it is possible to collect $\ell+1$ pebbles on the ends of an edge to be added, then it is possible to do this without creating any monochromatic cycles. \reffig{m2-move-eliminate} and \reffig{m2-move-eliminate-2} illustrate the construction used in the proof of \reflem{kill-m2-moves-locally}. We call this the {\bf shortcut construction} by analogy to matroid union and intersection augmenting paths used in previous work on the lower range. \reffig{m2-meta-picture} shows the structure of the proof. The shortcut construction removes an {\bf (M2)} step at the beginning of a sequence that reorients a path from $v$ to $w$ with pebble-slides. Since one application of the shortcut construction reorients a simple path from a vertex $w'$ to $w$, and a path from $v$ to $w'$ is preserved, the shortcut construction can be applied inductively to find the sequence of moves we want. \begin{proof} Without loss of generality, we can assume that our sequence of moves reorients a simple path in $H$, and that the first move (the end of the path) is {\bf (M2)}. The {\bf (M2)} step moves a pebble of color $c_i$ from a vertex $w$ onto the edge $vw$, which is reversed. Because the move is {\bf (M2)}, $v$ and $w$ are contained in a maximal monochromatic tree of color $c_i$. Call this tree $H'_i$, and observe that it is rooted at $w$. Now consider the edges reversed in our sequence of moves. As noted above, before we make any of the moves, these sketch out a simple path in $H$ ending at $w$. Let $z$ be the first vertex on this path in $H'_i$. We modify our sequence of moves as follows: delete, from the beginning, every move before the one that reverses some edge $yz$; prepend onto what is left a sequence of moves that moves the pebble on $w$ to $z$ in $H'_i$. Since no edges change color in the beginning of the new sequence, we have eliminated the {\bf (M2)} move. Because our construction does not change any of the edges involved in the remaining tail of the original sequence, the part of the original path that is left in the new sequence will still be a simple path in $H$, meeting our initial hypothesis. The rest of the lemma follows by induction. \end{proof} Together \reflem{can-kill-m1-moves} and \reflem{kill-m2-moves-locally} prove the following. \begin{lemma} If $G$ is a pebble-game graph, then $G$ has a canonical pebble game construction. \labellem{canonical-constructions-exist} \end{lemma} Using canonical pebble game constructions, we can identify the tight pebble-game graphs with maps-and-trees and \ellteekay\, graphs. \begin{restate}{canonical-decomposition-I}[{\bf Main Theorem (Lower Range): Maps-and-trees coincide with pebble-game graphs}] Let $0\le \ell\le k$. A graph $G$ is a tight pebble-game graph if and only if $G$ is a $(k,\ell)$-maps-and-trees. \end{restate} \begin{proof} As observed above, a maps-and-trees decomposition is a special case of the pebble game decomposition. Applying \refthm{non-canonical-decomposition}, we see that any maps-and-trees must be a pebble-game graph. For the reverse direction, consider a canonical pebble game construction of a tight graph. From \reflem{pebble-graphs-are-sparse}, we see that there are $\ell$ pebbles left on $G$ at the end of the construction. The definition of the {\bf canonical add-edge} move implies that there must be at least one pebble of each $c_i$ for $i=1,2,\ldots,\ell$. It follows that there is exactly one of each of these colors. By \reflem{roots}, each of these pebbles is the root of a monochromatic tree-piece with $n-1$ edges, yielding the required $\ell$ edge-disjoint spanning trees. \end{proof} \begin{restatecor}{m-a-t-equals-tight} [\textbf{Nash-Williams , Tutte , White and Whiteley }] Let $\ell\le k$. A graph $G$ is tight if and only if has a $(k,\ell)$-maps-and-trees decomposition. \end{restatecor} We next consider the decompositions induced by canonical pebble game constructions when $\ell\ge k+1$. \begin{restate}{canonical-decomposition-II}[{\bf Main Theorem (Upper Range): Proper Trees-and-trees coincide with pebble-game graphs}] Let $k\le \ell\le 2k-1$. A graph $G$ is a tight pebble-game graph if and only if it is a proper \ellteekay with $kn-\ell$ edges. \end{restate} \begin{proof} As observed above, a proper \ellteekay decomposition must be sparse. What we need to show is that a canonical pebble game construction of a tight graph produces a proper \ellteekay. By \refthm{non-canonical-decomposition} and \reflem{canonical-constructions-exist}, we already have the condition on tree-pieces and the decomposition into $\ell$ edge-disjoint trees. Finally, an application of {\bf (I4)}, shows that every vertex must in in exactly $k$ of the trees, as required. \end{proof} \begin{restatecor}{t-a-t-equals-tight}[\textbf{Crapo , Haas }] Let $k\le \ell\le 2k-1$. A graph $G$ is tight if and only if it is a proper \ellteekay. \end{restatecor} \section{Pebble game algorithms for finding decompositions} A naïve implementation of the constructions in the previous section leads to an algorithm requiring $\Theta(n^2)$ time to collect each pebble in a canonical construction: in the worst case $\Theta(n)$ applications of the construction in \reflem{kill-m2-moves-locally} requiring $\Theta(n)$ time each, giving a total running time of $\Theta(n^3)$ for the \textbf{decomposition} problem. In this section, we describe algorithms for the {\bf decomposition} problem that run in time $O(n^2)$. We begin with the overall structure of the algorithm. \begin{algorithm} [\textbf{The canonical pebble game with colors}]\labelalg{canonical-pebble-game} \qquad \noindent {\bf Input:} A graph $G$. \\ {\bf Output:} A pebble-game graph $H$. \\ {\bf Method:} \begin{itemize} \item Set $V(H)=V(G)$ and place one pebble of each color on the vertices of $H$. \item For each edge $vw\in E(G)$ try to collect at least $\ell+1$ pebbles on $v$ and $w$ using {\bf pebble-slide} moves as described by \reflem{kill-m2-moves-locally}. \item If at least $\ell+1$ pebbles can be collected, add $vw$ to $H$ using an {\bf add-edge} move as in \reflem{can-kill-m1-moves}, otherwise discard $vw$. \item Finally, return $H$, and the locations of the pebbles. \end{itemize} \end{algorithm} \paragraph{Correctness.} \refthm{sparse-graphs-are-pebble-graphs} and the result from that the sparse graphs are the independent sets of a matroid show that $H$ is a maximum sized sparse subgraph of $G$. Since the construction found is canonical, the main theorem shows that the coloring of the edges in $H$ gives a maps-and-trees or proper \ellteekay decomposition. \paragraph{Complexity.} We start by observing that the running time of \refalg{canonical-pebble-game} is the time taken to process $O(n)$ edges added to $H$ and $O(m)$ edges not added to $H$. We first consider the cost of an edge of $G$ that is added to $H$. Each of the pebble game moves can be implemented in constant time. What remains is to describe an efficient way to find and move the pebbles. We use the following algorithm as a subroutine of \refalg{canonical-pebble-game} to do this. \begin{algorithm} [\textbf{Finding a canonical path to a pebble.}]\labelalg{find-one-pebble} \qquad \\ \noindent{\bf Input:} Vertices $v$ and $w$, and a pebble game configuration on a directed graph $H$. \\ \noindent {\bf Output:} If a pebble was found, `yes', and `no' otherwise. The configuration of $H$ is updated. \\ \noindent {\bf Method:} \begin{itemize} \item Start by doing a depth-first search from from $v$ in $H$. If no pebble not on $w$ is found, stop and return `no.' \item Otherwise a pebble was found. We now have a path $v=v_{1},e_{1},\ldots,e_{p-1},v_{p}=u$, where the $v_{i}$ are vertices and $e_{i}$ is the edge $v_{i}v_{i+1}$. Let $c[e_{i}]$ be the color of the pebble on $e_{i}$. We will use the array $c[]$ to keep track of the colors of pebbles on vertices and edges after we move them and the array $s[]$ to sketch out a canonical path from $v$ to $u$ by finding a successor for each edge. \item Set\, $s[u]=`end'$ and set\, $c[u]$ to the color of an arbitrary pebble on $u$. We walk on the path in reverse order: $v_{p},e_{p-1}, e_{p-2}, \ldots, e_{1},v_{1}$. For each $i$, check to see if $c[v_{i}]$ is set; if so, go on to the next $i$. Otherwise, check to see if $c[v_{i+1}]=c[e_{i}]$. \item If it is, set $s[v_{i}]=e_{i}$ and set $c[v_{i}]=c[e_{i}]$, and go on to the next edge. \item Otherwise $c[v_{i+1}]\neq c[e_{i}]$, try to find a monochromatic path in color $c[v_{i+1}]$ from $v_{i}$ to $v_{i+1}$. If a vertex $x$ is encountered for which $c[x]$ is set, we have a path $v_{i}=x_{1},f_{1},x_{2},\ldots,f_{q-1},x_{q}=x$ that is monochromatic in the color of the edges; set $c[x_{i}]=c[f_{i}]$ and $s[x_{i}]=f_{i}$ for $i=1,2,\ldots,q-1$. If $c[x]=c[f_{q-1}]$, stop. Otherwise, recursively check that there is not a monochromatic $c[x]$ path from $x_{q-1}$ to $x$ using this same procedure. \item Finally, slide pebbles along the path from the original endpoints $v$ to $u$ specified by the successor array $s[v]$, $s[s[v]]$, $\ldots$ \end{itemize} \end{algorithm} The correctness of \refalg{find-one-pebble} comes from the fact that it is implementing the shortcut construction. Efficiency comes from the fact that instead of potentially moving the pebble back and forth, \refalg{find-one-pebble} pre-computes a canonical path crossing each edge of $H$ at most three times: once in the initial depth-first search, and twice while converting the initial path to a canonical one. It follows that each accepted edges takes $O(n)$ time, for a total of $O(n^2)$ time spent processing edges in $H$. Although we have not discussed this explicity, for the algorithm to be efficient we need to maintain components as in . After each accepted edge, the components of $H$ can be updated in time $O(n)$. Finally, the results of show that the rejected edges take an amortized $O(1)$ time each. Summarizing, we have shown that the canonical pebble game with colors solves the {\bf decomposition} problem in time $O(n^2)$. \section{An important special case: Rigidity in dimension $2$ and slider-pinning} In this short section we present a new application for the special case of practical importance, $k=2$, $\ell=3$. As discussed in the introduction, Laman's theorem characterizes minimally rigid graphs as the $(2,3)$-tight graphs. In recent work on slider pinning, developed after the current paper was submitted, we introduced the slider-pinning model of rigidity . Combinatorially, we model the bar-slider frameworks as simple graphs together with some loops placed on their vertices in such a way that there are no more than $2$ loops per vertex, one of each color. We characterize the minimally rigid bar-slider graphs as graphs that are: \begin{enumerate} \item $(2,3)$-sparse for subgraphs containing no loops. \item $(2,0)$-tight when loops are included. \end{enumerate} We call these graphs $(2,0,3)$-graded-tight, and they are a special case of the graded-sparse graphs studied in our paper . The connection with the pebble games in this paper is the following. \begin{corollary}[{\bf Pebble games and slider-pinning}] In any $(2,3)$-pebble game graph, if we replace pebbles by loops, we obtain a $(2,0,3)$-graded-tight graph. \end{corollary} \begin{proof} Follows from invariant {\bf (I3)} of \reflem{pebble-game-invariants}. \end{proof} In , we study a special case of slider pinning where every slider is either vertical or horizontal. We model the sliders as pre-colored loops, with the color indicating $x$ or $y$ direction. For this axis parallel slider case, the minimally rigid graphs are characterized by: \begin{enumerate} \item $(2,3)$-sparse for subgraphs containing no loops. \item Admit a $2$-coloring of the edges so that each color is a forest (i.e., has no cycles), and each monochromatic tree spans exactly one loop of its color. \end{enumerate} This also has an interpretation in terms of colored pebble games. \begin{corollary}[{\bf The pebble game with colors and slider-pinning}] In any canonical $(2,3)$-pebble-game-with-colors graph, if we replace pebbles by loops of the same color, we obtain the graph of a minimally pinned axis-parallel bar-slider framework. \end{corollary} \begin{proof} Follows from \refthm{canonical-decomposition-II}, and \reflem{roots}. \end{proof} \section{Conclusions and open problems} We presented a new characterization of $(k,\ell)$-sparse graphs, the \textbf{pebble game with colors}, and used it to give an efficient algorithm for finding decompositions of sparse graphs into edge-disjoint trees. Our algorithm finds such sparsity-certifying decompositions in the upper range and runs in time $O(n^2)$, which is as fast as the algorithms for recognizing sparse graphs in the upper range from . We also used the pebble game with colors to describe a new sparsity-certifying decomposition that applies to the entire matroidal range of sparse graphs. We defined and studied a class of canonical pebble game constructions that correspond to either a maps-and-trees or proper \ellteekay decomposition. This gives a new proof of the Tutte-Nash-Williams arboricity theorem and a unified proof of the previously studied decomposition certificates of sparsity. Canonical pebble game constructions also show the relationship between the $\ell+1$ pebble condition, which applies to the upper range of $\ell$, to matroid union augmenting paths, which do not apply in the upper range. \paragraph{Algorithmic consequences and open problems.} In , Gabow and Westermann give an $O(n^{3/2})$ algorithm for recognizing sparse graphs in the lower range and extracting sparse subgraphs from dense ones. Their technique is based on efficiently finding matroid union augmenting paths, which extend a maps-and-trees decomposition. The $O(n^{3/2})$ algorithm uses two subroutines to find augmenting paths: {\bf cyclic scanning}, which finds augmenting paths one at a time, and {\bf batch scanning}, which finds groups of disjoint augmenting paths. We observe that \refalg{canonical-pebble-game} can be used to replace cyclic scanning in Gabow and Westermann's algorithm without changing the running time. The data structures used in the implementation of the pebble game, detailed in are simpler and easier to implement than those used to support cyclic scanning. The two major open algorithmic problems related to the pebble game are then: \begin{problem} Develop a pebble game algorithm with the properties of {\bf batch scanning} and obtain an implementable $O(n^{3/2})$ algorithm for the lower range. \end{problem} \begin{problem} Extend {\bf batch scanning} to the $\ell+1$ pebble condition and derive an $O(n^{3/2})$ pebble game algorithm for the upper range. \end{problem} In particular, it would be of practical importance to find an implementable $O(n^{3/2})$ algorithm for decompositions into edge-disjoint spanning trees. \begin{thebibliography}{21} \providecommand{\natexlab}[1]{#1} \providecommand{\url}[1]{#1} \providecommand{\urlprefix}{URL } \expandafter\ifx\csname urlstyle\endcsname\relax \providecommand{\doi}[1]{DOI~\discretionary{}{}{}#1}\else \providecommand{\doi}{DOI~\discretionary{}{}{}\begingroup \urlstyle{rm}\Url}\fi \bibitem[{Berg and Jord\'{a}n(2003)}]{berg:jordan:2003} Berg, A.R., Jord\'{a}n, T.: Algorithms for graph rigidity and scene analysis. \newblock In: Proc. 11th European Symposium on Algorithms (ESA '03), LNCS, vol. 2832, pp. 78--89. (2003) \bibitem[{Crapo(1988)}]{crapo:rigidity:88} Crapo, H.: On the generic rigidity of plane frameworks. \newblock Tech. Rep. 1278, Institut de recherche d'informatique et d'automatique (1988) \bibitem[{Edmonds(1965)}]{Ed65} Edmonds, J.: Minimum partition of a matroid into independent sets. \newblock J. Res. Nat. Bur. Standards Sect. B \textbf{69B}, 67--72 (1965) \bibitem[{Edmonds(2003)}]{edmonds:matroidpolyhedra} Edmonds, J.: Submodular functions, matroids, and certain polyhedra. \newblock In: Combinatorial Optimization---Eureka, You Shrink!, no. 2570 in LNCS, pp. 11--26. Springer (2003) \bibitem[{Gabow(1995)}]{gabow:jcss-1995} Gabow, H.N.: A matroid approach to finding edge connectivity and packing arborescences. \newblock Journal of Computer and System Sciences \textbf{50}, 259--273 (1995) \bibitem[{Gabow and Westermann(1992)}]{gabow:forests:1992} Gabow, H.N., Westermann, H.H.: Forests, frames, and games: Algorithms for matroid sums and applications. \newblock Algorithmica \textbf{7}(1), 465--497 (1992) \bibitem[{Haas(2002)}]{haas:2002} Haas, R.: Characterizations of arboricity of graphs. \newblock Ars Combinatorica \textbf{63}, 129--137 (2002) \bibitem[{Haas et~al.(2007)Haas, Lee, Streinu, and Theran}]{maps} Haas, R., Lee, A., Streinu, I., Theran, L.: Characterizing sparse graphs by map decompositions. \newblock Journal of Combinatorial Mathematics and Combinatorial Computing \textbf{62}, 3--11 (2007) \bibitem[{Hendrickson(1992)}]{hendrickson:uniqueRealizability:1992} Hendrickson, B.: Conditions for unique graph realizations. \newblock SIAM Journal on Computing \textbf{21}(1), 65--84 (1992) \bibitem[{Jacobs and Hendrickson(1997)}]{jacobs:hendrickson:PebbleGame:1997a} Jacobs, D.J., Hendrickson, B.: An algorithm for two-dimensional rigidity percolation: the pebble game. \newblock Journal of Computational Physics \textbf{137}, 346--365 (1997) \bibitem[{Laman(1970)}]{laman} Laman, G.: On graphs and rigidity of plane skeletal structures. \newblock Journal of Engineering Mathematics \textbf{4}, 331--340 (1970) \bibitem[{Lee and Streinu(2008)}]{pebblegame} Lee, A., Streinu, I.: Pebble game algorihms and sparse graphs. \newblock Discrete Mathematics \textbf{308}(8), 1425--1437 (2008) \bibitem[{Lee et~al.(2005)Lee, Streinu, and Theran}]{cccg} Lee, A., Streinu, I., Theran, L.: Finding and maintaining rigid components. \newblock In: Proc. Canadian Conference of Computational Geometry. Windsor, Ontario (2005). \newblock \bibitem[{Lee et~al.(2007{\natexlab{a}})Lee, Streinu, and Theran}]{graded} Lee, A., Streinu, I., Theran, L.: Graded sparse graphs and matroids. \newblock Journal of Universal Computer Science \textbf{13}(10) (2007{\natexlab{a}}) \bibitem[{Lee et~al.(2007{\natexlab{b}})Lee, Streinu, and Theran}]{sliders} Lee, A., Streinu, I., Theran, L.: The slider-pinning problem. \newblock In: Proceedings of the 19th Canadian Conference on Computational Geometry (CCCG'07) (2007{\natexlab{b}}) \bibitem[{Lov{\'{a}}sz(1979)}]{lovasz:combinatorial-problems} Lov{\'{a}}sz, L.: Combinatorial Problems and Exercises. \newblock Akademiai Kiado and North-Holland, Amsterdam (1979) \bibitem[{Nash-Williams(1964)}]{nash-williams:decomposition-into-forests:1964} Nash-Williams, C.S.A.: Decomposition of finite graphs into forests. \newblock Journal of the London Mathematical Society \textbf{39}, 12 (1964) \bibitem[{Oxley(1992)}]{oxley:matroid} Oxley, J.G.: Matroid theory. \newblock The Clarendon Press, Oxford University Press, New York (1992) \bibitem[{Roskind and Tarjan(1985)}]{RoTa85} Roskind, J., Tarjan, R.E.: A note on finding minimum cost edge disjoint spanning trees. \newblock Mathematics of Operations Research \textbf{10}(4), 701--708 (1985) \bibitem[{Streinu and Theran(2008)}]{genericity} Streinu, I., Theran, L.: Combinatorial genericity and minimal rigidity. \newblock In: SCG '08: Proceedings of the twenty-fourth annual Symposium on Computational Geometry, pp. 365--374. ACM, New York, NY, USA (2008). \bibitem[{Tay(1984)}]{tay:rigidityMultigraphs-I:1984} Tay, T.S.: Rigidity of multigraphs {I}: linking rigid bodies in $n$-space. \newblock Journal of Combinatorial Theory, Series B \textbf{26}, 95--112 (1984) \bibitem[{Tay(1993)}]{Tay93} Tay, T.S.: A new proof of {L}aman's theorem. \newblock Graphs and Combinatorics \textbf{9}, 365--370 (1993) \bibitem[{Tutte(1961)}]{tutte:decomposing-graph-in-factors-1961} Tutte, W.T.: On the problem of decomposing a graph into $n$ connected factors. \newblock Journal of the London Mathematical Society \textbf{142}, 221--230 (1961) \bibitem[{Whiteley(1988)}]{whiteley:union-matroids} Whiteley, W.: The union of matroids and the rigidity of frameworks. \newblock SIAM Journal on Discrete Mathematics \textbf{1}(2), 237--255 (1988) \end{thebibliography}
|
0704.0010
| "Title: Partial cubes: structures, characterizations, and constructions\nAbstract: Partial cubes a(...TRUNCATED)
|
0704.0014
| "Title: Iterated integral and the loop product\nAbstract: In this article we discuss a relation be(...TRUNCATED)
|
0704.0016
| "Title: Lifetime of doubly charmed baryons\nAbstract: In this work, we evaluate the lifetimes of t(...TRUNCATED)
|
0704.0021
| "Title: Molecular Synchronization Waves in Arrays of Allosterically Regulated\n Enzymes\nAbstract: (...TRUNCATED)
|
0704.0022
| "Title: Stochastic Lie group integrators\nAbstract: We present Lie group integrators for nonlinear(...TRUNCATED)
|
0704.0032
| "Title: Probing non-standard neutrino interactions with supernova neutrinos\nAbstract: We analyze (...TRUNCATED)
|
0704.0039
| "Title: Scalar radius of the pion and zeros in the form factor\nAbstract: The quadratic pion scala(...TRUNCATED)
|
0704.0044
| "Title: Astrophysical gyrokinetics: kinetic and fluid turbulent cascades in\n magnetized weakly col(...TRUNCATED)
|
End of preview. Expand
in Data Studio
arxiv-tex-corpus-medium (15GB)
Medium-scale LaTeX corpus from arXiv (math, CS, physics, statistics)
📚 Overview
arxiv-tex-corpus-medium (15GB) is a medium-sized version of the arXiv LaTeX corpus, containing structured LaTeX source content extracted from selected arXiv categories.
This dataset is restricted to the following categories:
mathcshep-thhep-phquant-phstat.MLstat.TH
This version (~15GB) is intended for:
- Research experiments
- Model prototyping
- Ablation studies
- Resource-constrained training setups
🔗 Full Dataset
For the complete version (~80GB), see:
The 80GB version contains the full dataset without size reduction.
📦 Files
train.jsonl
val.jsonl
checkpoint.txt
train.jsonl
Training split in JSON Lines format.
val.jsonl
Validation split in JSON Lines format.
checkpoint.txt
Internal processing checkpoint file used during dataset creation. Not required for training.
🧾 Data Format
Each line in train.jsonl and val.jsonl is a JSON object.
Example schema:
{
"paper_id": "xxxx.xxxxx",
"category": "cs",
"latex": "\\documentclass{article} ..."
}
Fields
paper_id— arXiv identifiercategory— one of:mathcshep-thhep-phquant-phstat.MLstat.TH
latex— extracted LaTeX source content
📊 Dataset Characteristics
- Size: ~15GB
- Format: JSONL
- Categories: mathematics, computer science, theoretical physics, quantum physics, and statistics
- Derived from the same pipeline as the 80GB version
- Snapshot date: (add date here)
This version is a reduced subset of the full dataset.
🧹 Processing Pipeline
The dataset was constructed by:
- Downloading arXiv source archives.
- Extracting LaTeX source files.
- Filtering to retain only specified categories.
- Converting documents to structured JSONL format.
- Creating train/validation splits.
- Subsampling to create the 15GB medium version.
⚖️ Licensing
This dataset contains LaTeX source files from papers hosted on arXiv.
Each paper retains its original license as specified by its authors on arXiv.
Users are responsible for complying with the licensing terms of individual papers.
For license information: https://arxiv.org/help/license
🎯 Intended Use
- Medium-scale pretraining
- Fine-tuning on scientific domains
- Research in mathematical reasoning
- LaTeX modeling and generation
🚫 Limitations
- LaTeX sources may contain compilation errors.
- Some source bundles may be incomplete.
- Licensing varies per paper.
- This dataset is a subset of the full corpus.
📌 Citation
If you use this dataset, please cite:
- The original arXiv papers
- The arxiv-latex-corpus dataset
Example:
@dataset{arxiv_latex_corpus_2026,
title = {arxiv-latex-corpus},
year = {2026},
publisher = {Hugging Face},
}
🤝 Acknowledgements This dataset is derived from papers made publicly available by authors via arXiv. We thank the research community for openly sharing their work.
- Downloads last month
- 49