Solving DGLAP
We are solving the DGLAP equations [AP77, Dok77, GL72] given in x-space by
with \(\mathbf P\) the Altarelli-Parisi splitting functions (see pQCD ingredients). In Mellin space the DGLAP equations are just differential equations:
(Note the additional minus in the definition for \(\gamma\)).
We change the evolution variable to the (monotonic) Strong Coupling \(a_s(\mu_F^2)\) and the equations to solve become
This assumes the factorization scale \(\mu_F^2\) (the inherit scale of the PDF) and the renormalization scale \(\mu_R^2\) (the inherit scale for the strong coupling) to be equal, but tis constraint can however be lifted (see Scale Variations).
The (formal) solution can then be written in terms of an EKO \(\mathbf E\) [Bon12]
with \(\mathcal P\) the path-ordering operator. In the non-singlet sector the equations decouple and we do not need to worry about neither matrices nor the path-ordering.
Using Interpolation on both the initial and final PDF, we can then discretize the
EKO in x-space and define \({\mathbf{E}}_{k,j}\) (represented by
Operator
) by
Now, we can write the solution to DGLAP in a true matrix operator scheme and find
so the EKO is a rank-4 operator acting both in flavor and momentum fraction space.
The issue of matching conditions when crossing flavor thresholds is discussed in a separate document
Leading Order
Expanding the anomalous dimension \(\gamma(a_s)\) and the beta function \(\beta(a_s)\) to LO we obtain the (exact) EKO:
In LO we always use the exact solution.
LO Non-Singlet Evolution
We find
with \(\gamma_{ns}^{(0)} = \gamma_{ns,+}^{(0)} = \gamma_{ns,-}^{(0)} = \gamma_{ns,v}^{(0)} = \gamma_{qq}^{(0)}\).
The EKO is then given by a simple exponential [Vog05]
LO Singlet Evolution
We find
In order to exponentiate the EKO, we decompose it \(\ln \mathbf{\tilde E}^{(0)}_S = \lambda_+ {\mathbf e}_+ + \lambda_- {\mathbf e}_-\) with the eigenvalues \(\lambda_{\pm}\) and the projectors \(\mathbf e_{\pm}\) given by [Vog05]
with \(\mathbf I\) the 2x2 identity matrix in flavor space and, e.g., \(\ln \tilde E_{qq}^{(0)} = \gamma_{qq}^{(0)}j^{(0,0)}(a_s,a_s^0)\).
The projectors obey the usual properties, i.e.
and thus the exponentiation becomes easier again.
The EKO is then given by
Next-to-Leading Order
NLO Non-Singlet Evolution
We find
with \(\gamma_{ns} \in \{\gamma_{ns,+},\gamma_{ns,-}=\gamma_{ns,v}\}\).
We obtain the (exact) EKO [Bon12, RA98, Vog05]:
Note that we recover the LO solution:
In NLO we provide different strategies to define the EKO:
method in ['iterate-exact', 'decompose-exact', 'perturbative-exact']
: use the exact solution as defined abovemethod in ['iterate-expanded', 'decompose-expanded', 'perturbative-expanded']
: use the exact LO solution and substitute \(j^{(1,1)}(a_s,a_s^0) \to j^{(1,1)}_{exp}(a_s,a_s^0) = \frac 1 {\beta_0}(a_s - a_s^0)\) and \(j^{(0,1)}(a_s,a_s^0) \to j^{(0,1)}_{exp}(a_s,a_s^0) = j^{(0,0)}(a_s,a_s^0) - b_1 j^{(1,1)}_{exp}(a_s,a_s^0)\)method = 'ordered-truncated'
: expanding the argument of the exponential of the new term but keeping the order we obtain:
method = 'truncated'
: expanding the whole exponential of the new term we obtain:
NLO Singlet Evolution
We find
with \(\gamma_{S}^{(0)} \gamma_{S}^{(1)} \neq \gamma_{S}^{(1)} \gamma_{S}^{(0)}\).
Here the strategies are:
for
method in ['iterate-exact', 'iterate-expanded']
we use a discretized path-ordering [Bon12]:
where the order of the product is such that later EKO are to the left and
using the projector algebra from LO to exponentiate the single steps.
for
method in ['decompose-exact', 'decompose-expanded']
: use the exact or the approximate exact integrals from the non-singlet sector and then decompose \(\ln \tilde{\mathbf E}^{(1)}\) - this will neglect the non-commutativity of the singlet matrices.for
method in ['perturbative-exact', 'perturbative-expanded', 'ordered-truncated', 'truncated']
we seek for an perturbative solution around the (exact) leading order operator:
We set [Vog05]
where in NLO we find
and for the higher coefficients
method = 'perturbative-exact'
: \(\mathbf R_k = - b_1 \mathbf R_{k-1}\,\text{for}\,k>1\)method = 'perturbative-expanded'
: \(\mathbf R_k = 0\,\text{for}\,k>1\)
We make an ansatz for the solution
Inserting this ansatz into the differential equation and sorting by powers of \(a_s\), we obtain a recursive set of commutator relations for the evolution operator coefficients \(\mathbf U_k\):
Multiplying these equations with \(\mathbf e_{\pm}\) from left and right and using the identity
we obtain the \(\mathbf U_k\)
with \(r_{\pm} =\frac 1 {2\beta_0} \left( \gamma_{qq}^{(0)} + \gamma_{gg}^{(0)} \pm \sqrt{(\gamma_{qq}^{(0)}-\gamma_{gg}^{(0)})^2 + 4\gamma_{qg}^{(0)}\gamma_{gq}^{(0)}} \right)\).
So the strategies are
method in ['perturbative-exact', 'perturbative-expanded']
: approximate the full evolution operator \(\mathbf U(a_s)\) with an expansion up toev_op_max_order
method in ['ordered-truncated', 'truncated']
: truncate the evolution operator \(\mathbf U(a_s)\) and use
Next-to-Next-to-Leading Order
NNLO Non-Singlet Evolution
We find
with \(\gamma_{ns} \in \{\gamma_{ns,+},\gamma_{ns,-}=\gamma_{ns,v}\}\).
We obtain the (exact) EKO [CCG08, Vog05]:
with:
and:
Note, plugging the numerical values of \(\beta_i\) we find that the \(\Delta \in \mathbb{R}\) if \(n_f < 6\). However you can notice that \(\Delta\) appears always with \(\delta\) and the fraction \(\frac{\delta}{\Delta} \in \mathbb{R}, \forall n_f\).
We can recover the LO solution:
And thus the NLO solution:
In NNLO we provide different strategies to define the EKO:
method in ['iterate-exact', 'decompose-exact', 'perturbative-exact']
: use the exact solution as defined abovemethod in ['iterate-expanded', 'decompose-expanded', 'perturbative-expanded']
: use the exact LO solution and expand all functions \(j^{(n,m)}(a_s,a_s^0)\) to the order \(\mathcal o(a_s^3)\). We find:
This method corresponds to IMODEV=2
of [Vog05].
method = 'ordered-truncated'
: for this method we follow the prescription from [Vog05] and we get:
with the unitary matrices defined consistently with the method pertubative
adopted for NLO singlet evolution:
This method corresponds to IMODEV=3
of [Vog05].
method = 'truncated'
: we expand the whole exponential and keeping terms within \(\mathcal o(a_s^3)\). This method is the fastest among the ones provided by our program. We obtain:
NNLO Singlet Evolution
For the singlet evolution we find:
with \(\gamma_{S}^{(i)} \gamma_{S}^{(j)} \neq \gamma_{S}^{(j)} \gamma_{S}^{(i)}, \quad i,j=0,1,2\).
In analogy to NLO we define the following strategies :
for
method in ['iterate-exact', 'iterate-expanded']
we use a discretized path-ordering [Bon12]:
All the procedure is identical to NLO, simply the beat function is now expanded until \(\mathcal o(a_s^4)\)
for
method in ['decompose-exact', 'decompose-expanded']
: use the exact or the approximate exact integrals from the non-singlet sector and then decompose \(\ln \tilde{\mathbf E}^{(2)}\) - this will neglect the non-commutativity of the singlet matrices.for
method in ['perturbative-exact', 'perturbative-expanded', 'ordered-truncated', 'truncated']
we seek for an perturbative solution around the (exact) leading order operator. We set [Vog05]
Finding one additional term compared to NLO:
and for the higher coefficients
method = 'perturbative-exact'
: \(\mathbf R_k = - b_1 \mathbf R_{k-1} - b_2 \mathbf R_{k-1} \,\text{for}\,k>2\)method = 'perturbative-expanded'
: \(\mathbf R_k = 0\,\text{for}\,k>2\)
The solution ansatz becomes:
with:
So the strategies are:
method in ['perturbative-exact', 'perturbative-expanded']
: approximate the full evolution operator \(\mathbf U(a_s)\) with an expansion up toev_op_max_order
method in ['ordered-truncated', 'truncated']
: truncate the evolution operator \(\mathbf U(a_s)\) and use
Intrinsic evolution
We also consider the evolution of intrinsic heavy PDF. Since these are massive partons they can not split any collinear particles and thus they do not participate in the DGLAP evolution. Instead, their evolution is simply an identity operation: e.g. for an intrinsic charm distribution we get for \(m_c^2 > Q_1^2 > Q_0^2\):
After crossing the mass threshold (charm in this example) the PDF can not be considered intrinsic any longer and hence, they have to be rejoined with their evolution basis elements and take then again part in the ordinary collinear evolution.