Convolved substructure: analytically decorrelating jet substructure observables

Journal of High Energy Physics, May 2018

Ian Moult, Benjamin Nachman, Duff Neill

A PDF file should load here. If you do not see its contents the file may be temporarily unavailable at the journal website or you do not have a PDF plug-in installed and enabled in your browser.

Alternatively, you can download the file locally and open with any standalone PDF reader:

https://link.springer.com/content/pdf/10.1007%2FJHEP05%282018%29002.pdf

Convolved substructure: analytically decorrelating jet substructure observables

HJE Convolved substructure: analytically decorrelating jet Ian Moult 0 2 4 5 Benjamin Nachman 0 3 5 0 Berkeley , CA 94720 , U.S.A 1 Theoretical Division , MS B283 , Los Alamos National Laboratory 2 Theoretical Physics Group, Lawrence Berkeley National Laboratory 3 Physics Division, Lawrence Berkeley National Laboratory 4 Berkeley Center for Theoretical Physics, University of California , USA 5 Los Alamos , NM 87545 , U.S.A A number of recent applications of jet substructure, in particular searches for light new particles, require substructure observables that are decorrelated with the jet mass. In this paper we introduce the Convolved SubStructure (CSS) approach, which uses a theoretical understanding of the observable to decorrelate the complete shape of its distribution. This decorrelation is performed by convolution with a shape function whose parameters and mass dependence are derived analytically. We consider in detail the case of the D2 observable and perform an illustrative case study using a search for a light hadronically decaying Z0. We nd that the CSS approach completely decorrelates the D2 observable over a wide range of masses. Our approach highlights the importance of improving the theoretical understanding of jet substructure observables to exploit increasingly subtle features for performance. Jets; QCD Phenomenology - Neilld 1 Introduction 2 3 4 5 Correlation with mass for jet substructure observables Model search for H ! bb [19, 20], and to searches for new light Z0 bosons, deriving bounds in a previously unprobed region of parameter space [21{23].1 These searches represent an impressive advance in the sophistication of jet substructure techniques. Unlike for high mass resonance searches, these low mass searches use the mass of the jet itself. This makes it important that the jet substructure observable used for tagging is independent of the mass of the jet. Otherwise, the cut on the tagging observable can signi cantly distort the jet mass spectrum, making it di cult to search for resonances. This was rst highlighted in [25], where a procedure, termed DDT, was introduced to decorrelate the observable from the jet mass and pT . More precisely, the DDT decorrelates the rst moment of the observable. Due to the importance of this problem, several other groups have applied machine learning to develop tagging observables that are decorrelated with the jet mass and pT [26, 27]. In parallel with experimental advances, there have been signi cant advances in the theoretical understanding of jet substructure observables,2 and a large number of calculations from rst principles QCD [29{45]. These calculations provide signi cant insight into the behavior of jet substructure observables, and have enabled advances in their sophistication, 1For other recent bounds on this region see [24]. 2For a review of recent advances in jet substructure, see [28]. { 1 { with many of the most important observables in current use arising out of analytic calculations. Recently an all orders factorization formula [46, 47] was derived for the groomed3 D2 observable [41, 48], which is used extensively by ATLAS [1{10]. It was derived in soft-collinear e ective theory (SCET) [49{53] and its multi-scale extensions [41, 54{59]. This factorization allows for an understanding of the all orders perturbative and nonperturbative behavior of the observable. In this paper, we show how we can use an understanding of substructure observables to completely decorrelate them with the jet mass. In particular, we will show that the standard way of incorporating non-perturbative hadronization e ects, namely convolution with a model shape function, motivates a simple way of performing the decorrelation: convolution with a function that maps the distribution at any mass to the distribution at a reference mass. We will call this approach to decorrelation Convolved SubStructure (CSS).4 The CSS approach naturally preserves the domain and normalization of the tagging observable, and allows a decorrelation of the complete shape of the observable, not just the rst moment. The philosophy of our approach is slightly distinct from [26, 27], namely it attempts to decorrelate a given standard observable, such as D2 [41, 48], or N2 [65], using a theoretical understanding of that particular observable, and as such, is similar in spirit to the original DDT [25]. Indeed, we will show that the rst moment of our approach reproduces the DDT, and therefore the CSS approach should be thought of as a systematic generalization of the DDT beyond the rst moment. A schematic depiction of our approach is shown in gure 1. At a given mass, the distribution predicted by the factorization theorem for an observable such as D2 is given as a convolution of a non-perturbative shape function [66{69] FNP( ; m) which encodes the e ects of hadronization, with the perturbative distribution (here and throughout the text, will denote a dimensionless convolution variable, and m denotes the mass). Both the perturbative distribution, as well as the non-perturbative shape function depend on the jet mass, and therefore both introduce correlations between the observable and the jet mass. However, with an understanding of these di erent functions, we can map the distribution at a given mass to a reference mass if we know both the non-perturbative shape function, FNP( ; m), as well as the mapping between the perturbative distributions, FP( ; m1; m2), which is a perturbatively calculable function. The end result is that we can derive a function, FCSS( ; m1; m2), which completely decorrelates the observable by mapping it to a reference mass point.5 This de nes the CSS decorrelated D2 observable: d CSS dD2 = 1 Z 0 d FCSS( ; m1; m2) dD2 (D2 ) : d (1.1) 3By grooming we mean modi ed mass drop (MMDT) [38, 39] or soft drop [37] groomers, which for = 0 are equivalent. function describing the mapping between the perturbative distributions at the masses m1 and m2 (They are technically de ned as convolutions in as described in the text, which has been suppressed in the gure.). By combining these mappings we can completely decorrelate the observable by mapping it to a reference mass value. Here is a dimensionless convolution variable, and m1 and m2 denotes the masses that the function maps between. The exact function can be determined through an understanding of both the perturbative and non-perturbative aspects of the distribution, namely, FCSS = FNP1 FP FNP ; (1.2) where denotes convolution. This combination of mappings is shown schematically in gure 1. While it is of course trivial that such a function exists, the simple structure of the observable enables us to provide a simple analytic form for the function FCSS, allowing for a fast numerical implementation, as well as an understanding of how it scales with m1 and m2. Furthermore, the function FCSS can be systematically improved starting from this initial function, using an expansion in orthogonal polynomials, as developed in [69].6 An outline of this paper is as follows. In section 2 we discuss the sources of correlation between a two-prong substructure observable such as D2, and the jet mass, treating both the perturbative and non-perturbative aspects of this correlation, and we show that in both cases they can be modeled using shape functions. Furthermore, we analytically derive the mass scaling of the shape function parameters. In section 3 we discuss how we can use 6The perturbative distribution can of course be calculated, while the non-perturbative contribution must currently be modeled. However, due to the structure of the factorization theorem for the tagging observable, one can con dently predict the scaling of the non-perturbative corrections with the jet mass (that is, their contributions to the moments of the distribution) in a systematically improvable manner, thus xing the functional form of the jet mass dependence in the expansion of the shape function with respect to the orthogonal polynomials (speci cally, generalized Laguerre polynomials). { 3 { this understanding to decorrelate jet substructure observables using shape functions, and introduce the CSS approach. We then illustrate concretely how the decorrelation can be done in practice. In section 4 we perform a brief study, illustrating the e ectiveness of the decorrelation procedure for Z0 ! qq. We conclude in section 5. 2 Correlation with mass for jet substructure observables In this section we discuss the sources of correlation between a two-prong observable, such as D2, and the jet mass (for brevity, we will not always explicitly say groomed jet mass, although we always work with groomed observables), to illustrate how these correlations arise. In section 2.1, we discuss the dependence of non-perturbative physics on jet mass, and introduce the modeling of hadronization e ects using shape functions. In section 2.2 we discuss perturbative sources of correlation, and show that they can also be well captured by a simple shape function. In this paper we will consider the concrete example of the D2 observable, for which a factorization formula is known [46, 47]. This allows us to make precise statements about the perturbative and non-perturbative behavior of the observable. The D2 observable is de ned in terms of the energy correlation functions [ 70 ] 2 e( ) = 3 e( ) = 1 1 X 2 pT J i<j2J X 3 pT J i<j<k2J pT ipT j Rij ; pT ipT j pT kRij RikRjk ; D2 ( ) = e( ) 3 (e(2 ))3 Jet substructure observables are sensitive to low scales within a jet, and are therefore naturally susceptible to non-perturbative e ects. Non-perturbative contributions can arise both from the underlying event (UE), as well as from the standard hadronization process within the jet. In [46], it was shown that due to the grooming procedure, non-perturbative e ects from the underlying event are negligible. We will therefore neglect them in what follows. Using the factorization formula for the D2 observable derived in [46, 47], it can be shown that the dominant non-perturbative e ects from hadronization are captured by a collinear-soft function Csi(e3) = trh0jT fYig (e3 E^ 3) SDT fYigj0i : (2.4) { 4 { Here the Yi are products of Wilson lines along the subjet directions, T and T denote time and anti-time ordering respectively. The measurement function and soft drop constraints are implemented by the energy ow operators E^ 3 and SD, whose exact form is not relevant for the current discussion. These operators can be written in terms of the energy-momentum tensor [71{74]. Importantly, due to the application of the grooming algorithm, the collinear-soft function, and hence the non-perturbative hadronization corrections, depend only on the color structure of the jet itself, and not on the color structure of the global event, making them a property of the observable. While the collinear-soft function in eq. (2.4) can be calculated perturbatively, it is currently not possible to calculate it non-perturbatively. Instead, a functional parametrization of the non-perturbative matrix element, which is referred to as a shape function, FNP, is used [66{69]. Shape functions have been used in a variety of contexts in jet physics [41, 45, 46, 75{77]. For the particular case of D2, this allows the non-perturbative D2 distribution to be written as a convolution of the perturbative distribution and the shape function d NP = dD2 1 Z 0 dx F~NP(x) d dD2 D2 x mJ zc3u=t2 ! function in eq. (2.4), and were derived in [46, 47]. We will take our model shape function to have the simple functional form7 F~NP(x; ; D) = D 1 ( ) x 1 e x D : This function has a rst moment D QCD, is normalized to unity, and we may think of this speci c shape function as but the rst term in an orthogonal expansion which speci es the non-perturbtive corrections to all moments of the distribution, where we have truncated to speci cally x only the rst moment. Here is a parameter, which speci es the functional form. We will choose such that the function vanishes as x ! 0. We nd that = 2-3 provides a good description of the non-perturbative correction. Since the dominant e ect is a shift of the rst moment, which is xed, it is only at small value of D2 that there is dependence on . The physical interpretation of this function is that it smears the energies within the jet at the scale QCD. In certain cases universal properties of the rst moment of shape functions can be proven [78, 79]. These moments, as well as higher moments have been extracted from event shape data, for example from the thrust event shape [80]. Ref. [46] studied the non-perturbative shape parameter D, and found D is independent of the quark or gluon nature of the jet. The scaling predicted by eq. (2.4), namely that the non-perturbative shift in the distribution is inversely proportional to the mass, is well respected in parton shower Monte Carlo simulations. 7This functional form is that of a Gamma distribution. Amusingly, we note that these are the maximally entropic distributions with a xed rst moment and rst logarithmic moment. { 5 { = ∈ [ = hadronized distributions as found in Pythia, and as modeled using the non-perturbative shape function described in the text. (b) The dependence of the non-perturbative shift, NP as a function D of the groomed mass, which introduces a source of correlation of the D2 distribution with the groomed jet mass. In gure 2, we show the e ects of hadronization on the D2 observable found in Pythia, and as modeled using the shape function of eq. (2.6). We see that the simple shape function reproduces quite well the e ects of the hadronization. Although it is conventional to work with a shape function parameter that has mass dimension 1, such as D, for our purposes it will be convenient to introduce the dimensionless shift in the rst moment of the D2 distribution, which we denote case of the non-perturbative hadronization corrections, we have the relation NDP. For the NP = D D mJ zc3u=t2 : When using the dimensionless variable, we use the shape function FNP( ; ; D) = D 1 ( ) 1 e D ; which is the same functional form as in eq. (2.6), but we have dropped the tilde to emphasize that the dimension of the argument has changed. The dependence of D NP as extracted from Pythia is shown in gure 2b, as well as a t for the non-perturbative parameter D. To extract this scaling, we have t the shift parameter in the tail region of the distribution, where we expect that a shift of the distribution is valid. The uncertainties represent a conservative estimate due to the fact that the precise region in which one should be performing the t is not always clear. The strong dependence on the mass of the jet is clearly visible, which introduces a non-perturbative correlation between the D2 distribution and the jet mass. It is also important to note that the shift NDP is dependent { 6 { (2.7) (2.8) groomed D2 spectrum as a function of the jet mass, and in (b) we show the decorrelated D2 spectrum. The movement of the distribution in (a) as the mass is varied is largely eliminated by the decorrelation procedure in (b). only on mJ , and not on pT , as can be derived from the factorization formula [46, 47]. This simpli cation is only true for groomed distributions. Inverting the logic of this section, if we are able to transform between the perturbative and non-perturbative distributions using a convolution with a simple function, this also implies that we can perform the deconvolution to obtain the perturbative distribution. Doing this would remove the correlation of the D2 distribution with the jet mass arising from hadronization corrections. However, to completely decorrelate the distribution, we also need to understand how to decorrelate the perturbative distributions, which can also depend on the jet mass. This will be addressed in section 2.2. In section 3 we will then give a numerically simple way of performing the decorrelation via convolution. 2.2 Perturbative e ects In addition to a dependence of the hadronization corrections on the jet mass, there is also a dependence of the perturbative D2 distribution on the jet mass that introduces a further correlation between the D2 distribution and the jet mass. Unlike the hadronization correction, where only the scaling of the hadronization corrections as a function of the jet mass is calculable, the perturbative distribution can be calculated to a given accuracy, and hence the complete dependence of the distribution on the jet mass can be understood. In gure 3a we show a plot of the perturbative D2 distribution at next-to-leading logarithm matched to leading order 1 ! 3 splitting functions in the large D2 region in order to reproduce the correct endpoint behavior. In the gures this accuracy is referred to as NLL+LO. See [47] for a more detailed discussion of the order counting. Here the H ! gg process was used to produce gluon jets. We can see that there is a mild, but non-negligible dependence on the jet mass within the peak region. A more quantitative measure, PD, the { 7 { Δ Δ result, along with the analytic prediction described in the text. (b) The shift in the log mean for both the standard and CSS decorrelated distributions for di erent values of the parameter for the shape function. shift in the mean relative to the distribution at m = 35 GeV is shown in gure 4a. This is only a small e ect for the groomed D2, which has a xed endpoint at 1=(2zcut), independent of the jet mass. It is ultimately this fact that leads to a large degree of stability of the distribution. For the ungroomed D2 distribution the endpoint depends strongly on the jet mass so that the distribution displays a much more complicated dependence on the jet mass. Following the logic of the previous section, if we understand the form of the correlation between the D2 distribution and the jet mass, we can also remove this correlation. Motivated by the implementation of the shape function for the non-perturbative contribution, we can also attempt to decorrelate the perturbative component of the distribution by convolving with a function which takes the perturbative distributions to some reference value. Since the mean of the D2 distribution increases with decreasing mass, to decorrelate by convolution with a simple shape function, we will always use as a reference mass value the lowest mass value of interest. Namely, we write 1 Z 0 d dD2 ( 1; m2) = d FP ( ; m1; m2) dD2 d ( 1 ; m1) ; m2 < m1: (2.9) Here we have made explicit the mass dependence of the functions, which is separated from the argument of the function by a semi-colon. The fact that such a (possibly singular) function exists is trivial, and it can be determined by division in Laplace or Fourier space (i.e. by deconvolution). Furthermore, this function is (in principle) exactly calculable from the factorization theorem, given predictions for the perturbative D2 distribution at any given accuracy at any jet mass. However, to have a reasonable prediction for the D2 distribution requires a matched calculation. This implies that results for the distribution are necessarily numerical instead of analytic, making it di cult to understand the deconvolu{ 8 { HJEP05(218) tion analytically. We would therefore like to nd a simple function that provides a good approximation to the exact result. Although we cannot analytically predict the exact shape function (in a practical way), we can use our analytic NLL+LO result to compute moments of the perturbative distribution. We expect that the dominant e ect of the correlation between the D2 observable and the jet mass will be a shift of the rst moment, as can be seen from gure 3a. The shift in the mean relative to the distribution at m = 35 GeV, The shift in the rst moment of the distribution arises due to the renormalization group evolution of the functions appearing in the factorization theorem of refs. [46, 47]. We can PD, is shown in gure 4a. therefore write the shift in the rst perturbative order as PD = D m1 Z m2 d s( ) + : : : ; (2.10) where D is a constant, which we extract from our calculation of the distribution at two mass points. The prediction from this functional form is shown in the dashed line in gure 4a, which provides an excellent description of the numerical results at many other values of the jet mass, con rming the perturbative evolution of the rst moment. To perform the perturbative decorrelation, we will use as the base decorrelation function the functional form of eq. (2.6). Since we can analytically predict the shift can use this function to exactly decorrelate the mean. However, by tuning the angular exponent, with the mean xed, we can further attempt to decorrelate the complete shape of the distribution. The value of can be extracted by decorrelating the log-mean of the distribution, which can be computed analytically from our NLL+LO calculation. The evolution of the log mean with mass is shown in gure 4, both without decorrelation, and after decorrelation using the function of eq. (2.6) for several values of . We nd that for in the range of = 2-3, we have good decorrelation of the log mean. Furthermore, it is quite insensitive to the exact value of used, which shows that the correlation is dominated by a shift in the mean. The decorrelation of the full distribution for = 2:4 is shown in gure 3b. As compared with gure 3a, we see a good decorrelation of the full shape of the distribution. This shows that the dependence of the D2 observable on the mass is in fact remarkably simple, being driven by a shift in the rst moment captured by eq. (2.10), with deviations from this to account for the behavior at the endpoints being captured by the D P , we simple class of functions in eq. (2.6). We conclude this section by emphasizing that this analysis could be improved by iteratively building up a shape function starting from the base function of eq. (2.6) using an expansion in orthogonal functions, as has been done in [69], requiring all moments to be decorrelated exactly. However, for our purposes we will nd that the simple function of eq. (2.6) works extremely well, as will be illustrated in our case study in section 4. 3 Convolved substructure Motivated by the above observation that both the perturbative and non-perturbative components of the distribution can be decorrelated using simple shape functions, we propose { 9 { that we can use shape functions as an e cient way to completely decorrelate two-prong substructure observables by mapping them to reference mass. This is what we will call the Convolved Substructure, or CSS procedure. Since the shape functions used in hadronization are typically used to shift the distribution to a larger value, for the D2 observable, we will also choose the reference mass to be the lowest mass of interest, ensuring that the shift in the mean required for decorrelation is positive. We de ne the CSS decorrelated D2 observable by Here FCSS is an as of yet unspeci ed function with unit norm. While we have used the speci c example of D2, this approach should apply much more generally, however, we expect that it will only be for IRC safe observables with su ciently favorable factorization properties that analytic scalings for the FCSS function can be derived. Within this subset of observables, we believe that this represents a completely general and e cient way of performing the decorrelation. Unlike previously proposed analytic approaches, it aims to decorrelate all moments of the distribution, and naturally preserves the domain and norm of the distribution. Furthermore, motivated by the success in describing non-perturbative corrections using a simple basis of functions [69], we will show that we can choose a simple analytic form of the function FCSS as the initial approximation. Further improvements can be systematically added, if needed. It is also interesting to see that this approach includes as a special case the standard DDT, which is a shift of the rst moment. Performing a Taylor expansion for a small shift, we have d CSS dD2 1 Z 0 = d FCSS( ) (D2 ) : (3.1) This reproduces (a constrained form of) the DDT, which decorrelates the rst moment. We note that while the DDT procedure was originally introduced as a shift which decorrelates the rst moment of the distribution, it has since been generalized to decorrelate, for example, the background e ciency at a given cut. Nevertheless, it can still only decorrelate a single chosen moment of the distribution. We will re-emphasize this point in our numerical comparisons in section 4. Note that when used for incorporating non-perturbative e ects, the linear shift applies in a particular region of the distribution, but the full shape function is needed at small values. We will see in section 3.1 that this is also true when used for decorrelation, with the full convolution reducing to a linear shift throughout most of the distribution, and the full non-linear nature of the function only becoming relevant near the endpoints of the distribution. m2 < m1, can be written as The exact function FCSS to shift from the mass m1 to a reference mass m2, with FCSS( ; m1; m2) = FNP1( ; m1) FP( ; m1; m2) FNP( ; m2) ; (3.4) as was illustrated in gure 1. Here the denotes convolution in the variable , and the inverse denotes an inverse in the convolutional sense (i.e. a deconvolution). Instead of performing the decorrelation in this form, we will simplify our discussion and use a single e ective function. This can certainly be improved, however, we will already nd that with a single function we will nd an excellent decorrelation. We will use the decorrelation function of the previous section, namely8 FCSS( ; ; D) = D 1 ( ) 1 e D : With this parametrization, we have that the rst moment is D for all values of , but we allow for a general power law behavior as x ! 0, speci ed by . When considering a full example at the LHC, we will nd that a value of slightly larger than two will give an excellent t. Taking the rst moment of eq. (3.4), we nd that D(m1; m2) = NDP(m2) NDP(m1) + PD(m1; m2) : Again, we assume that the reference mass that we are shifting the distributions to, namely m2, satis es m2 < m1. In sections 2.1 and 2.2 we have used the factorization formula for the D2 observable derived in [46, 47] to predict the mass dependence of both the perturbative, PD, and non-perturbative, NDP, moments appearing in eq. (3.6). In principle, the exact values of the moments can be extracted for given processes and observables, by studying the distributions with and without hadronization, as was done above. The decorrelation using this procedure on our NLL+LO calculation is shown in gure 5, which shows both the perturbative and non-perturbative distributions, as well as the nal CSS curve, and can be viewed as an analytic realization of the strategy outlined in gure 1. Good, but not perfect decorrelation is observed, and we will see in section 4 that the decorrelation procedure seems to work even better in Pythia than for the analytic example shown here.9 For ease of applicability, we nd it more convenient to give a formula for D(m1; m2), with two constants that can be directly extracted by tting the decorrelation at several points, as will be demonstrated in a practical example in section 4. Using our 8That the nal convolution in eq. (3.4) can be approximated by a single function of the same form can be understood by looking at the functional form in Laplace space, where these functions take the form of rational functions to the power using the rst term in the expansion for FCSS. Due to the inverse convolution appearing in eq. (3.4), the Laplace transform of the convolution of the three functions has the same polynomial degree as the Laplace transform of a single such function. 9There is also a tradeo between exactly reproducing the mean and accurately capturing other aspects of the shape. (3.5) (3.6) using the rst moment shift determined analytically from eq. (3.6). Perturbative distributions at the di erent mass values are shown in small-dashed, while the full distributions are shown in solid. The CSS result decorrelated to mJ = 35 GeV is shown in dashed blue. It involves decorrelating both the perturbative and non-perturbative evolution, as can be seen from the di erent curves. understanding of the functional dependence on the jet mass for both the perturbative and non-perturbative contributions to the moment discussed in sections 2.1 and 2.2, we have the general form of the moment for the CSS approach as D(m1; m2) = cNP ' cNP 1 m2 1 m2 m1 1 1 m1 m1 Z d m2 + cP + c~P log s( ) ; ; m1 m2 (3.7) where the second line is an approximation that is good for most numerical purposes. Again, we emphasize that the reference mass, m2 is taken to satisfy m2 < m1, so that this shift is positive. Here the cNP, cP and c~P are constants that can be t for numerically, and describe the non-perturbative and perturbative scalings respectively. We note that although it may appear unnatural, the coe cients cNP and c~P have di erent mass dimensions, since cNP is associated with a power-law variation, while c~P is associated with a logarithmic variation. From a practical perspective, the CSS decorrelation function can be constructed by xing the value of appearing in eq. (3.5) using a single value of the mass. For D2, we nd values of 2 [2; 3] work well, with no strong preference for a given value. Using several values of the mass, one can then t for cNP and c~P to give a smooth function that describes the evolution of the moment of the shape function. Knowing the analytic scaling of the function is therefore important, as it allows the shape to be xed using dedicated Monte Carlo at a few speci c mass points, and does not require Monte Carlo at every single value of the mass to determine the form. We will illustrate this for a case study of Z0 ! qq in section 4, where we will nd that this gives a remarkably good (almost perfect) decorrelation of the D2 observable. In practice, the convolution procedure described above needs to be applied jet-by-jet and not at the distribution level. The convolution of two distributions corresponds to the addition of the random variables described by the distributions. Therefore, one possibility for translating the distribution-level results from earlier to event-by-event results is to add to every observed D2 value a random value drawn from the distribution FCSS(x; ; D) from eq. (3.5). This is not ideal because (a) the randomness can introduce features in the classi cation performance for nite statistics and (b) there are various technical reasons like reproducibility that make injecting randomness unattractive. Another way to accomplish the convolution but using a deterministic approach is to use the (inverse) cumulative distribution function (CDF). Given a random variable X with CDF C(x) = Pr(X < x), C(X) is a new random variable that follows a uniform distribution. For any other CDF G, one can then form the random variable G 1(C(X)), which follows the probability distribution (3.8) (3.9) (3.10) c(x) = 1 d dD2 g(x; ; D) = c(x) FCSS(x; ; D): We can now de ne the CDFs C(x) = R0x c(x0)dx0 and G(x; ; D) = R0x g(x0; ; D)dx0. Then, the jet-by-jet transformation is given by D2 7! G 1(C(D2); ; D): This simple mapping allows us to numerically implement the CSS procedure in an e cient manner. An explicit example of the mapping given by eq. (3.10) for the example of Z0 ! qq, which is discussed in detail in section 4, is shown in gure 6. This gure demonstrates the construction of the CSS D2, following the procedure from section 3.1. The CDF for each D2 distribution is computed (C for D2 and G for D2 CSS), as shown in gure 6a and then the transformation in eq. (3.10) is shown in gure 6b. While the CSS curves may look mostly linear, there are important non-linear features at high and low D2. These will be discussed in detail in section 4, and will play an important role in decorrelating the complete D2 distribution, and not just the rst moment. The perturbative expansion of the CSS procedure to its rst moment, as was discussed around eq. (3.2) gives rise to a linear behavior, and the fact that the mapping in gure 6b is mostly linear simply shows that this is a good approximation. Note that the DDT procedure would result in straight lines in gure 6 with a mass-dependent o set. 4 An important and recent application of variable decorrelation is the search for a low mass hadronic resonance, Z0 ! qq [21{23], which we therefore use as a case study. The generic ∈ [ for QCD jets. (b) The mapping between D2 and D2 CSS. Here an angular exponent used for D2. The mapping is linear throughout most of the range of interest, but with important non-linearities at small and large values of D2. angular exponent of suppress the argument. quark and gluon background is too large to observe a dijet resonance directly, but when the Z0 is produced in association with initial state radiation, it can be su ciently boosted for its decay products to be collimated inside a single jet. For our study, both the Z0 and the generic quark and gluon background are simulated with Pythia 8.183 [81, 82]; the former by changing the mass of a Standard Model Z boson and the latter with all hard QCD processes. All stable nal state particles excluding neutrinos and muons are clustered into jets with FastJet 3.1.3 [83] using the anti-kt algorithm [83, 84] with R = 0:8. In order to make sure that the Z0 particles with masses up to 300 GeV are mostly contained inside a single jet, jets are required to have pT > 1 TeV. Jets are then re-clustered using the Cambridge/Aachen algorithm [85{87] and groomed with mMDT/soft drop using zcut = 0:1. From the groomed jet's constituents, the jet mass is calculated along with D2 using the EnergyCorrelator FastJet contrib [83, 88]. Throughout this section we will use an = 2 for the D2 observable, but for notational simplicity, we will To perform the CSS decorrelation, we will shift all distributions to the reference mass of m = 50 GeV, and we will consider jets with masses in the range 50 GeV < m < 250 GeV, namely a factor of 5 variation. This is approximately the mass range used in the current LHC searches [21{23]. In a realistic application, it may be convenient to shift the distributions in di erent mass regions to di erent reference values. For example, for low mass searches, the Z mass provides a natural mass scale where the analysis changes, and therefore it may prove useful to shift jets with mass m > mZ to the reference mass of mZ , and jets with mass m < mZ to the lower mass limit of the search. In this way, the required decorrelation in each mass window is minimized. However, the goal of this section is simply to illustrate that we can completely decorrelate the D2 distribution over a wide range of jet masses. ∈ [ ∈ [ = ∈ [ and (c) show the groomed D2 distribution after the application of the CSS and DDT procedures, respectively. The di erences between the DDT and CSS distributions are shown in gure (d), and grow at small values. The DDT and CSS procedures are applied to both signal and background, where the transformations are de ned by the background distributions. In gure 7 we show the standard D2 distribution, as well as the decorrelated distributions using the CSS and DDT approaches, for ve narrow bins in the groomed jet mass. The DDT is applied by shifting D2 7! D2 hD2jmi + hD2j50 < m=GeV < 55i; where the averages hxjyi (this means the average of x given y) are computed using the QCD background jets. By construction, the average of the resulting DDT distribution is independent of m: hD2 DDTjmi = hD2 hD2jmi + hD2j50 < m=GeV < 55ijmi = hD2jmi hD2jmi + hD2j50 < m=GeV < 55i = hD2j50 < m=GeV < 55i (4.1) (4.2) Δ D, as a function of jet mass as extracted from Pythia, and compared with a t to the analytic form described in eq. (3.7) in the text. The CSS procedure is applied using the shape function, FCSS, of eq. (3.5) with = 2:4 and D as indicated in the gure. The value of was xed for a single value of the mass, however, fortunately, we nd that we are quite insensitive to the precise choice of . The values of D are plotted in gure 8 along with a t to the analytic form, which we see provides an excellent description. The extractions of the shift at these ve mass values can be viewed as xing the coe cients of the analytic mass dependence of the decorrelation procedure of eq. (3.7), and providing a prediction for every other value of the mass, as would be required experimentally. Here we see the advantage of knowing the analytic form, namely that one only needs dedicated Monte Carlo at several speci c mass values. The signal distributions are also shown to give a feeling for the range of interest of the D2 observable for discrimination. A number of features of the di erent decorrelation procedures are clearly evident from these gures. First, we see in gure 7b that the CSS decorrelated observable has essentially no dependence on the jet mass. The complete shape of the distribution is identical for the wide range of masses shown. By contrast, the shape of the DDT version in gure 7c changes with mass even though the mean is xed. This is particularly true on the left side of the peak. The di erence between methods arises from the non-linear nature of the CSS mapping, as was mentioned in gure 6. A zoomed in view of the small D2 region is shown in gure 9, which highlights the di erence between the two approaches. Both the CSS and DDT mappings are e ectively linear to the right of the D2 peak, where we see that both decorrelate the observable very well, but the non-linear mapping is required to perform the decorrelation of the shape of the distribution at small values of D2. It is in this region that the shape of the distribution changes non-trivially with mass, and the di erence between distributions at di erent masses cannot simply be described by a shift. The ability of the CSS approach to correctly reproduce the change in shape of the distribution in this region of the distribution, which is the most important region for HJEP05(218) distribution in (b) at small values. The CSS approach decorrelates the entire shape of the distribution, including at low values of D2, where the shape of the distribution changes non-trivially, and which is the relevant region for discrimination. discrimination, is quite remarkable. The di erences between the two di erent decorrelated distributions are shown in gure 7d, which also highlights that the di erences between the two decorrelation procedures become large at small values of D2. We also note that here we have chosen to decorrelate the background (QCD) distributions, and therefore the signal distributions exhibit some dependence on mass. As a further quantitative comparison between the CSS and DDT approaches, in gure 10 we compare di erent integrals of the distributions, namely the mean, and the probability that D2 0:4 (the lower tail fraction), which we denote Pr(D2 < 0:4). By construction, the mean of the DDT D2 distribution is independent of mass, as seen in gure 10a. However, the shape does change with mass as indicated by the lower tail fraction in gure 10b (lower D2 is more signal-like). On the other hand, since the CSS approach decorrelates the complete shape of the distribution both the mean and the tail fraction are nearly independent of mass. We must also emphasize that the DDT approach could equally well be applied to atten the Pr(D2(2) < 0:4) (or any other given integral of the distribution). However, it would then not decorrelate the mean. In other words, it can be used to decorrelate a single moment at a time. On the other hand, the CSS approach aims to decorrelate all moments. Finally, it is important to check that the CSS procedure does not degrade the tagging performance of the D2 observable. This was shown for the DDT approach in [25]. Applying the mapping shown in the right plot of gure 6 also to Z0 events results in the distributions that were already shown in gure 7. Lower values of D2 are more signal-like so an upperthreshold on the D2 distribution is an e ective two-prong tagger. Figure 11 quanti es the tradeo between signal and background e ciency with and without the CSS procedure. As desired, there is a minimal di erence in the ROC curve after applying CSS. This di erence could be further minimized by performing the CSS decorrelation in narrower mass windows. construction the DDT decorrelates a single moment, chosen here to be the rst moment, but does not decorrelate higher moments. On the other hand, the CSS procedure is designed to decorrelate the entire shape of the distribution. 5 Conclusions In this paper, we have shown how a given jet substructure observable, such as N2 or D2, can be decorrelated with the jet mass using an understanding of its perturbative and non-perturbative behavior. Inspired by the use of shape functions for modeling nonperturbative e ects, we introduced the Convolved SubStructure (CSS) approach, which uses a shape function, convolved with the substructure observable's distribution, to map it to a reference mass. The shape function incorporates e ects due to both perturbative and non-perturbative physics, and we used a recently derived factorization formula to analytically derive the mass dependence of both these contributions. Unlike previous approaches with similar philosophies, the CSS approach completely decorrelates the entire shape of the distribution. Furthermore, it is systematically improvable by expanding the shape function in a basis of orthogonal functions [69], and uses maximally the theoretical understanding of the observable. We have shown in detail how the CSS approach can be practically implemented in an extremely simple manner, and studied its behavior for the example of a light Z0 ! qq search using the D2 observable. We found that using a simple two parameter shape function we were able to obtain an excellent decorrelation of the entire D2 distribution over a wide range of mass values. The shape function parameter de ning the shift of the rst moment of the distribution has a functional dependence on the jet mass that can be understood from rst principles, and is xed by demanding that the rst moment of the mapped distributions are the same as the reference mass distribution. Higher moments can be handled similarly, but since we require the shape function to maintain the domain and norm of the distribution, we nd that already the decorrelation of the rst moment ∈ [ curve quantifying the tradeo between Z' (signal) e ciency and QCD (background) e ciency for various groomed jet mass bins, shown in a linear plot in a) and a log plot in b). The CSS procedure is not found to signi cantly degrade the discrimination power of the observable. e ectively decorrelates the whole spectrum. Furthermore, the discrimination power of the CSS observable was not signi cantly degraded. In real applications, the tradeo between discrimination power and decorrelation must be evaluated, and it may be practical to perform the decorrelation in mass windows. One important aspect that we did not study in this paper is whether an identical mapping applies at detector level. Even if it is not the case, our approach is general, and another simple functional form that performs the decorrelation could be found. It will also be interesting to apply the CSS approach to other observables, such as N2, for which the DDT approach has been applied successfully [21, 23]. Again, a slightly modi ed convolution function may be required, depending on the behavior of the observable. We therefore hope that the CSS approach can be used to decorrelate a variety of substructure observables, improving the reach and performance of searches for low mass particles at the LHC. Acknowledgments We thank Nhan Tran and Simone Marzani for helpful and encouraging discussions, and Andrew Larkoski for collaboration on related topics and for catching several typos in the rst draft of this work. This work is supported by the U.S. Department of Energy (DOE) under cooperative research agreements DE-FG02-05ER-41360, and DE-SC0011090 and by the O ce of High Energy Physics of the U.S. DOE under Contract No. DE-AC02-05CH11231, and the LDRD Program of LBNL. This work was performed in part at the Aspen Center for Physics, which is supported by National Science Foundation grant PHY-1607611. D.N. thanks both the Berkeley Center for Theoretical Physics and LBNL for hospitality while portions of this work were completed, as well as support from the Munich Institute for Astro- and Particle Physics (MIAPP) of the DFG cluster of excellence \Origin and Structure of the Universe", and support from DOE contract DE-AC52-06NA25396 at LANL and through the LANL/LDRD Program. Open Access. This article is distributed under the terms of the Creative Commons Attribution License (CC-BY 4.0), which permits any use, distribution and reproduction in any medium, provided the original author(s) and source are credited. [1] ATLAS collaboration, Performance of jet substructure techniques in early p s = 13 TeV pp HJEP05(218) collisions with the ATLAS detector, ATLAS-CONF-2015-035 [INSPIRE]. [2] ATLAS collaboration, Identi cation of boosted, hadronically decaying W bosons and comparisons with ATLAS data taken at p [arXiv:1510.05821] [INSPIRE]. s = 8 TeV, Eur. Phys. J. C 76 (2016) 154 p p pp collisions at p [arXiv:1607.06363] [INSPIRE]. decaying vector boson in pp collisions at p B 763 (2016) 251 [arXiv:1608.02372] [INSPIRE]. [3] ATLAS collaboration, Searches for heavy diboson resonances in pp collisions at s = 13 TeV with the ATLAS detector, JHEP 09 (2016) 173 [arXiv:1606.04833] [INSPIRE]. [4] ATLAS collaboration, Search for heavy resonances decaying to a Z boson and a photon in s = 13 TeV with the ATLAS detector, Phys. Lett. B 764 (2017) 11 [5] ATLAS collaboration, Search for dark matter produced in association with a hadronically s = 13 TeV with the ATLAS detector, Phys. Lett. at p states in pp collisions at p [arXiv:1708.09638] [INSPIRE]. particle X in the XH ! qqbb 24 [arXiv:1709.06783] [INSPIRE]. Z boson in hadronic 145 [arXiv:1506.01443] [INSPIRE]. [6] ATLAS collaboration, Search for pair production of heavy vector-like quarks decaying to high-pT W bosons and b quarks in the lepton-plus-jets nal state in pp collisions at s = 13 TeV with the ATLAS detector, JHEP 10 (2017) 141 [arXiv:1707.03347] [INSPIRE]. [7] ATLAS collaboration, Search for heavy resonances decaying to a W or Z boson and a Higgs boson in the qq(0)bb nal state in pp collisions at p Phys. Lett. B 774 (2017) 494 [arXiv:1707.06958] [INSPIRE]. s = 13 TeV with the ATLAS detector, [8] ATLAS collaboration, Search for diboson resonances with boson-tagged jets in pp collisions s = 13 TeV with the ATLAS detector, Phys. Lett. B 777 (2018) 91 [arXiv:1708.04445] [9] ATLAS collaboration, Searches for heavy ZZ and ZW resonances in the ``qq and qq nal s = 13 TeV with the ATLAS detector, JHEP 03 (2018) 009 [10] ATLAS collaboration, A search for resonances decaying into a Higgs boson and a new nal state with the ATLAS detector, Phys. Lett. B 779 (2018) [11] CMS collaboration, Search for a massive resonance decaying into a Higgs boson and a W or nal states in proton-proton collisions at p s = 8 TeV, JHEP 02 (2016) [12] CMS collaboration, Search for heavy resonances decaying to two Higgs bosons in nal states containing four b quarks, Eur. Phys. J. C 76 (2016) 371 [arXiv:1602.08762] [INSPIRE]. proton-proton collisions at p [13] CMS collaboration, Search for massive resonances decaying into WW, WZ or ZZ bosons in s = 13 TeV, JHEP 03 (2017) 162 [arXiv:1612.09159] [14] CMS collaboration, Search for massive resonances decaying into W W , W Z, ZZ, qW and s = 13 TeV, Phys. Rev. D 97 (2018) 072006 [15] CMS collaboration, Search for heavy resonances that decay into a vector boson and a Higgs s = 13 TeV, Eur. Phys. J. C 77 (2017) 636 [16] CMS collaboration, Searches for W' bosons decaying to a top quark and a bottom quark in HJEP05(218) proton-proton collisions at 13 TeV, JHEP 08 (2017) 029 [arXiv:1706.04260] [INSPIRE]. top quark at p s = 13 TeV, JHEP 09 (2017) 053 [arXiv:1703.06352] [INSPIRE]. [17] CMS collaboration, Search for a heavy resonance decaying to a top quark and a vector-like nal states in proton-proton collisions at p s = 13 TeV, JHEP 07 (2017) 001 [18] CMS collaboration, Search for tt resonances in highly boosted lepton+jets and fully hadronic [arXiv:1704.03366] [INSPIRE]. collisions at p [19] CMS collaboration, Inclusive search for a highly boosted Higgs boson decaying to a bottom quark-antiquark pair, Phys. Rev. Lett. 120 (2018) 071802 [arXiv:1709.05543] [INSPIRE]. [20] CMS collaboration, Inclusive search for the standard model Higgs boson produced in pp s = 13 TeV using H! bb decays, CMS-PAS-HIG-17-010 [INSPIRE]. [21] CMS collaboration, Search for light vector resonances decaying to a quark pair produced in association with a jet in proton-proton collisions at p s = 13 TeV, CMS-PAS-EXO-17-001 [22] CMS collaboration, Search for Low Mass Vector Resonances Decaying to Quark-Antiquark s = 13 TeV, Phys. Rev. Lett. 119 (2017) 111802 [23] CMS collaboration, Search for low mass vector resonances decaying into quark-antiquark s = 13 TeV, JHEP 01 (2018) 097 [arXiv:1710.00159] [24] A. Mariotti, D. Redigolo, F. Sala and K. Tobioka, New LHC bound on low-mass diphoton [25] J. Dolen, P. Harris, S. Marzani, S. Rappoccio and N. Tran, Thinking outside the ROCs: Designing Decorrelated Taggers (DDT) for jet substructure, JHEP 05 (2016) 156 Pairs in Proton-Proton Collisions at p [arXiv:1705.10532] [INSPIRE]. pairs in proton-proton collisions at p [26] C. Shimmin et al., Decorrelated Jet Substructure Tagging using Adversarial Neural Networks, Phys. Rev. D 96 (2017) 074034 [arXiv:1703.03507] [INSPIRE]. [27] J.A. Aguilar-Saavedra, J.H. Collins and R.K. Mishra, A generic anti-QCD jet tagger, JHEP 11 (2017) 163 [arXiv:1709.01087] [INSPIRE]. [28] A.J. Larkoski, I. Moult and B. Nachman, Jet Substructure at the Large Hadron Collider: A Review of Recent Advances in Theory and Machine Learning, arXiv:1709.04464 [INSPIRE]. [29] S. Marzani, L. Schunk and G. Soyez, A study of jet mass distributions with grooming, JHEP 07 (2017) 132 [arXiv:1704.02210] [INSPIRE]. observables, arXiv:1603.06375 [INSPIRE]. [INSPIRE]. [35] T. Becher and M. Neubert, Factorization and NNLL Resummation for Higgs Production with a Jet Veto, JHEP 07 (2012) 108 [arXiv:1205.3806] [INSPIRE]. [36] I. Feige, M.D. Schwartz, I.W. Stewart and J. Thaler, Precision Jet Substructure from Boosted Event Shapes, Phys. Rev. Lett. 109 (2012) 092001 [arXiv:1204.3898] [INSPIRE]. [arXiv:1402.2657] [INSPIRE]. substructure, JHEP 09 (2013) 029 [arXiv:1307.0007] [INSPIRE]. [39] M. Dasgupta, A. Fregoso, S. Marzani and A. Powling, Jet substructure with analytical methods, Eur. Phys. J. C 73 (2013) 2623 [arXiv:1307.0013] [INSPIRE]. [40] M. Dasgupta, L. Schunk and G. Soyez, Jet shapes for boosted jet two-prong decays from rst-principles, JHEP 04 (2016) 166 [arXiv:1512.00516] [INSPIRE]. [41] A.J. Larkoski, I. Moult and D. Neill, Analytic Boosted Boson Discrimination, JHEP 05 (2016) 117 [arXiv:1507.03018] [INSPIRE]. [42] C. Frye, A.J. Larkoski, J. Thaler and K. Zhou, Casimir Meets Poisson: Improved Quark/Gluon Discrimination with Counting Observables, JHEP 09 (2017) 083 [arXiv:1704.06266] [INSPIRE]. [43] T.T. Jouttenus, I.W. Stewart, F.J. Tackmann and W.J. Waalewijn, Jet mass spectra in Higgs boson plus one jet at next-to-next-to-leading logarithmic order, Phys. Rev. D 88 (2013) 054031 [arXiv:1302.0846] [INSPIRE]. 08 (2015) 079 [arXiv:1503.01088] [INSPIRE]. with Light Grooming, arXiv:1708.02586 [INSPIRE]. Hadron Collider, arXiv:1708.06760 [INSPIRE]. [44] M. Dasgupta, A. Powling and A. Siodmok, On jet substructure methods for signal jets, JHEP [45] A.H. Hoang, S. Mantry, A. Pathak and I.W. Stewart, Extracting a Short Distance Top Mass [46] A.J. Larkoski, I. Moult and D. Neill, Analytic Boosted Boson Discrimination at the Large [47] A.J. Larkoski, I. Moult and D. Neill, Factorization and Resummation for Groomed Multi-Prong Jet Shapes, JHEP 02 (2018) 144 [arXiv:1710.00014] [INSPIRE]. [48] A.J. Larkoski, I. Moult and D. Neill, Power Counting to Better Jet Observables, JHEP 12 (2014) 009 [arXiv:1409.6298] [INSPIRE]. [INSPIRE]. and soft gluons: Heavy to light decays, Phys. Rev. D 63 (2001) 114020 [hep-ph/0011336] [50] C.W. Bauer and I.W. Stewart, Invariant operators in collinear e ective theory, Phys. Lett. B 516 (2001) 134 [hep-ph/0107001] [INSPIRE]. Phys. Rev. D 65 (2002) 054022 [hep-ph/0109045] [INSPIRE]. [53] I.Z. Rothstein and I.W. Stewart, An E ective Field Theory for Forward Scattering and Factorization Violation, JHEP 08 (2016) 025 [arXiv:1601.04695] [INSPIRE]. [54] C.W. Bauer, F.J. Tackmann, J.R. Walsh and S. Zuberi, Factorization and Resummation for Dijet Invariant Mass Spectra, Phys. Rev. D 85 (2012) 074006 [arXiv:1106.6047] [INSPIRE]. [55] A.J. Larkoski, I. Moult and D. Neill, Toward Multi-Di erential Cross Sections: Measuring Two Angularities on a Single Jet, JHEP 09 (2014) 046 [arXiv:1401.4458] [INSPIRE]. [56] M. Procura, W.J. Waalewijn and L. Zeune, Resummation of Double-Di erential Cross Sections and Fully-Unintegrated Parton Distribution Functions, JHEP 02 (2015) 117 [arXiv:1410.6483] [INSPIRE]. [57] A.J. Larkoski, I. Moult and D. Neill, Non-Global Logarithms, Factorization, and the Soft Substructure of Jets, JHEP 09 (2015) 143 [arXiv:1501.04596] [INSPIRE]. [58] P. Pietrulewicz, F.J. Tackmann and W.J. Waalewijn, Factorization and Resummation for Generic Hierarchies between Jets, JHEP 08 (2016) 002 [arXiv:1601.05088] [INSPIRE]. [59] Y.-T. Chien, A. Hornig and C. Lee, Soft-collinear mode for jet cross sections in soft collinear e ective theory, Phys. Rev. D 93 (2016) 014033 [arXiv:1509.04287] [INSPIRE]. [Erratum ibid. B 213 (1983) 545] [INSPIRE]. [61] J.C. Collins, D.E. Soper and G.F. Sterman, Transverse Momentum Distribution in Drell-Yan Pair and W and Z Boson Production, Nucl. Phys. B 250 (1985) 199 [INSPIRE]. [62] J.C. Collins, D.E. Soper and G.F. Sterman, Factorization for Short Distance Hadron Hadron Scattering, Nucl. Phys. B 261 (1985) 104 [INSPIRE]. [63] J.C. Collins, D.E. Soper and G.F. Sterman, Soft Gluons and Factorization, Nucl. Phys. B 308 (1988) 833 [INSPIRE]. [64] J.C. Collins, D.E. Soper and G.F. Sterman, Factorization of Hard Processes in QCD, Adv. Ser. Direct. High Energy Phys. 5 (1989) 1 [hep-ph/0409313] [INSPIRE]. [65] I. Moult, L. Necib and J. Thaler, New Angles on Energy Correlation Functions, JHEP 12 (2016) 153 [arXiv:1609.07483] [INSPIRE]. [66] G.P. Korchemsky and G.F. Sterman, Power corrections to event shapes and factorization, Nucl. Phys. B 555 (1999) 335 [hep-ph/9902341] [INSPIRE]. QCD, JHEP 10 (2000) 010 [hep-ph/0007005] [INSPIRE]. [67] G.P. Korchemsky and S. Tafat, On power corrections to the event shape distributions in [68] A.H. Hoang and I.W. Stewart, Designing gapped soft functions for jet production, Phys. Lett. B 660 (2008) 483 [arXiv:0709.3519] [INSPIRE]. [69] Z. Ligeti, I.W. Stewart and F.J. Tackmann, Treating the b quark distribution function with JHEP 06 (2013) 108 [arXiv:1305.0007] [INSPIRE]. 403 [hep-ph/9512370] [INSPIRE]. HJEP05(218) AIP Conf. Proc. 407 (1997) 988 [hep-ph/9708346] [INSPIRE]. [73] C. Lee and G.F. Sterman, Momentum Flow Correlations from Event Shapes: Factorized Soft Gluons and Soft-Collinear E ective Theory, Phys. Rev. D 75 (2007) 014022 [hep-ph/0611061] [INSPIRE]. Distributions with Hadronic Final States in Soft Collinear E ective Theory, Phys. Rev. D 78 (2008) 034027 [arXiv:0801.4569] [INSPIRE]. Power Corrections and a Precision Global Fit for alphas(mZ), Phys. Rev. D 83 (2011) 074021 [arXiv:1006.3080] [INSPIRE]. [76] I.W. Stewart, F.J. Tackmann and W.J. Waalewijn, Dissecting Soft Radiation with Factorization, Phys. Rev. Lett. 114 (2015) 092001 [arXiv:1405.6722] [INSPIRE]. [77] A.H. Hoang, D.W. Kolodrubetz, V. Mateu and I.W. Stewart, C-parameter distribution at N3LL' including power corrections, Phys. Rev. D 91 (2015) 094017 [arXiv:1411.6633] Mod. Phys. Lett. A 22 (2007) 835 [hep-ph/0703030] [INSPIRE]. 0601121 (2006) A001 [hep-ph/0603066] [INSPIRE]. [79] C. Lee and G.F. Sterman, Universality of nonperturbative e ects in event shapes, eConf C [80] R. Abbate, M. Fickinger, A.H. Hoang, V. Mateu and I.W. Stewart, Precision Thrust Cumulant Moments at N 3LL, Phys. Rev. D 86 (2012) 094002 [arXiv:1204.5746] [INSPIRE]. [81] T. Sjostrand, S. Mrenna and P.Z. Skands, PYTHIA 6.4 Physics and Manual, JHEP 05 (2006) 026 [hep-ph/0603175] [INSPIRE]. 159 [arXiv:1410.3012] [INSPIRE]. 1896 [arXiv:1111.6097] [INSPIRE]. (2008) 063 [arXiv:0802.1189] [INSPIRE]. JHEP 08 (1997) 001 [hep-ph/9707323] [INSPIRE]. [82] T. Sjostrand et al., An introduction to PYTHIA 8.2, Comput. Phys. Commun. 191 (2015) [83] M. Cacciari, G.P. Salam and G. Soyez, FastJet User Manual, Eur. Phys. J. C 72 (2012) [84] M. Cacciari, G.P. Salam and G. Soyez, The anti-k(t) jet clustering algorithm, JHEP 04 [85] Y.L. Dokshitzer, G.D. Leder, S. Moretti and B.R. Webber, Better jet clustering algorithms, scattering, in Monte Carlo generators for HERA physics. Proceedings, Workshop, Hamburg, deep inelastic positron-proton collisions at p [30] C. Frye , A.J. Larkoski , M.D. Schwartz and K. Yan , Factorization for groomed jet substructure beyond the next-to-leading logarithm , JHEP 07 ( 2016 ) 064 [arXiv: 1603 .09338] [INSPIRE]. [31] C. Frye , A.J. Larkoski , M.D. Schwartz and K. Yan , Precision physics with pile-up insensitive [32] A. Ban , H. McAslan , P.F. Monni and G. Zanderighi, The two-jet rate in e+e at next-to-next-to-leading-logarithmic order , Phys. Rev. Lett . 117 ( 2016 ) 172001 [33] A. Ban et al., Jet-vetoed Higgs cross section in gluon fusion at N3LO+NNLL with small-R resummation , JHEP 04 ( 2016 ) 049 [arXiv: 1511 .02886] [INSPIRE]. [34] I.W. Stewart , F.J. Tackmann , J.R. Walsh and S. Zuberi , Jet pT resummation in Higgs production at N N LL0 + N N LO , Phys. Rev. D 89 ( 2014 ) 054001 [arXiv: 1307 . 1808 ] [37] A.J. Larkoski , S. Marzani , G. Soyez and J. Thaler , Soft Drop, JHEP 05 ( 2014 ) 146 [38] M. Dasgupta , A. Fregoso , S. Marzani and G.P. Salam , Towards an understanding of jet [51] C.W. Bauer , D. Pirjol and I.W. Stewart , Soft collinear factorization in e ective eld theory , [52] C.W. Bauer , S. Fleming , D. Pirjol , I.Z. Rothstein and I.W. Stewart , Hard scattering factorization from e ective eld theory , Phys. Rev. D 66 ( 2002 ) 014017 [ hep -ph/0202088] [60] J.C. Collins and D.E. Soper , Back- To-Back Jets in QCD , Nucl. Phys. B 193 ( 1981 ) 381 reliable uncertainties , Phys. Rev. D 78 ( 2008 ) 114014 [arXiv: 0807 . 1926 ] [INSPIRE]. [70] A.J. Larkoski , G.P. Salam and J. Thaler , Energy Correlation Functions for Jet Substructure , [71] N.A. Sveshnikov and F.V. Tkachov , Jets and quantum eld theory , Phys. Lett. B 382 ( 1996 ) [72] G.P. Korchemsky , G. Oderda and G.F. Sterman , Power corrections and nonlocal operators, [74] C.W. Bauer , S.P. Fleming , C. Lee and G.F. Sterman , Factorization of e+e [75] R. Abbate , M. Fickinger , A.H. Hoang , V. Mateu and I.W. Stewart , Thrust at N 3LL with [86] M. Wobisch and T. Wengler , Hadronization corrections to jet cross-sections in deep inelastic [87] H1 collaboration, C. Adlo et al ., Measurement and QCD analysis of jet cross-sections in s = 300 GeV, Eur . Phys. J. C 19 ( 2001 ) 289 [88] Fastjet contrib , http://fastjet.hepforge.org/contrib/.


This is a preview of a remote PDF: https://link.springer.com/content/pdf/10.1007%2FJHEP05%282018%29002.pdf

Ian Moult, Benjamin Nachman, Duff Neill. Convolved substructure: analytically decorrelating jet substructure observables, Journal of High Energy Physics, 2018, 2, DOI: 10.1007/JHEP05(2018)002