Prepared for submission to JHEP

From Operator Product Expansion to Anomalous Dimensions

arXiv:2410.03283v1 [hep-th] 4 Oct 2024

Rijun Huanga , Qingjun Jinb and Yi Lib

a
Institute of Theoretical Physics, School of Physics and Technology, Nanjing Normal University,
No.1 Wenyuan Road, Nanjing 210046, P.R.China
b
Graduate School of China Academy of Engineering Physics,
No. 10 Xibeiwang East Road, Haidian District, Beijing, 100193, P.R.China

E-mail: [email protected], [email protected], [email protected]

Abstract: We propose a new method for computing renormalization functions, which is based on the
ideas of operator product expansion and large momentum expansion. In this method, the renormalization
Z-factors are determined by the ultraviolet finiteness of Wilson coefficients in the dimensional regularization
scheme. The ultraviolet divergence is extracted solely from two-point functions at the large momentum
limit. We develop this method in scalar field theories and provide a general framework for computing
anomalous dimensions of field, mass, couplings and composite operators. In particular, it is applied to
6-dimensional cubic scalar theory and 4-dimensional quartic scalar theory. We demonstrate this method
by computing the anomalous dimension of the ϕQ operator in cubic theory up to four loops for arbitrary
Q, which is in agreement with the known result in the large N limit. The idea of computing anomalous
dimensions from operator production expansion is general and can be extended beyond scalar theories.

Keywords: OPE, large momentum expansion, renormalization, anomalous dimension, scalar theory
Contents

1 Introduction 1

2 A framework of computing renormalization functions 3

2.1 The operator product expansion 4
2.2 The large momentum expansion 5
2.3 Determining leading order term of large momentum expansion 7
2.4 Analyzing leading order term from OPE 11
2.5 Expansion of correlation functions and computation of Z-factors 13

3 The anomalous dimension of ϕQ operator in O(N ) cubic scalar theory 16

3.1 The O(N ) model and ϕQ operator 16
3.2 Treatments on the Feynman graphs 17
3.3 The anomalous dimensions 22

4 Conclusion 23

A Sample UV divergence computation of two-point function 24

B The 4-loop correction of scaling dimension ∆Q 25

1 Introduction

In quantum field theory, renormalization was originally developed as a method to remove ultraviolet (UV)
divergences and avoid infinities when computing Feynman integrals. It plays a crucial role in defining
effective theories that predict and match experimental observables. As in the early development of Quantum
Electrodynamics (QED) in 1940s, it resolved the disagreement between the theoretical divergent electron
self-energy correction and the experimental finite physical observables [1–4]. A significant achievement of
renormalization in particle physics is the explanation of asymptotic freedom in Quantum Chromodynamics
(QCD). Asymptotic freedom is manifested by the decreasing coupling with increasing energy, which can
be understood through the computation of the beta function. As is well-known, this idea is crucial for
establishing QCD as the theory of strong interaction [5, 6]. Apart from its early interests in particle
physics, the renormalization group soon advanced into statistical physics and found itself highly suitable
for understanding problems of phase transitions and critical phenomena. For example, the critical index in
the scalar ϕ4 theory with O(N ) symmetry [7, 8] can describe the critical phenomena of some systems such

–1–
as superfluid 4 He, ferromagnetism, etc. In condensed matter physics, renormalization has been applied to
understanding the quantum liquids or phases with broken symmetries, such as superconducting states. It
is said that there are also applications in the stoichiometry, crystalline structure and orbital configurations.
Since all of these are outside our research directions, we suggest consulting related specialists if interested.

Initiated from QED, the idea of renormalization was reinforced and formalized by several works in the
following 1950s [9–12] and eventually evolved into the concept of continuum renormalization group. In the
continuum picture, physical quantities are required to be independent of the renormalization conditions,
and equal to observed values. Through the efforts of Callan and Symanzik [13, 14], the continuum renor-
malization group found its extensive applications in particle physics, such as computing the beta function
and the anomalous dimensions of field and mass. A decade later, Kenneth Wilson’s series of works laid the
foundation for the modern understanding of renormalization [15–17], known today as the Wilsonian renor-
malization group. In the Wilsonian picture, the coupling constants in a theory change in such a way that
the observables remain the same when the UV cutoff Λ is changing. Furthermore, the Wilsonian picture
provides a reasonable viewpoint for understanding the physical meaning of non-renormalizable theories.
That is, one can regard a non-renormalizable theory as an effective field theory describing physics at low
energies. Then, Joseph Polchinski’s work in the mid-1980s clarified an approach in which the Wilsonian
renormalization group can be implemented via path integral formalism [18]. This method induces a set of
differential equations on the couplings known as renormalization group equations, and it gradually becomes
a fundamental tool in quantum field theory with broad applications.

Working out the renormalization group equations in practical computation poses a significant challenge
as it essentially involves Feynman graph and Feynman integral computation. It inherits the same difficulties
as most problems based on the Feynman graph method. The complexity escalates rapidly with increasing
loop orders and the number of external momenta. Not only does the number of graphs increase factorially,
but performing integration also becomes an arduous mathematical task that involves an increasing types
and number of special functions. Along the development of renormalization, numerous methods have been
proposed to enhance computation ability and efficiency, enabling one to take on the higher loop (L ≥ 4)
challenge. For example, the R∗ operation method [19, 20], which utilizes the fact that the UV counterterm
of (L+1)-loop Feynman integral can be expressed as L-loop one with massless propagators, has been widely
used in calculating renormalization functions. The beta function in QCD theory has been generalized to
five-loop order by the R∗ operation method [21], although it took around 15-20 workstations and more
than seven months to complete. However, when applied to integrals with more than five loops and/or
non-trivial numerators such as multiple tensor structures of loop momenta, the R∗ operation method can
be inefficient because it involves complicated subtraction of UV and IR sub-divergences. An alternative
approach that circumvents the subtraction of sub-divergences is the massive bubble approach [22, 23],
which reduces the computation of UV divergences to the evaluation of massive vacuum bubbles. It has
been applied to, for instance, the four-loop order renormalization of the Gross-Neveu model [24]. However,
the master integrals of vacuum bubble integrals are more difficult to evaluate than that of the massless

–2–
propagators. Currently, the high precision numerical values of complete five-loop master integrals are still
unavailable [25]. Another promising method for the higher loop challenge is the graphical function method
[26], which evaluates Feynman integrals in coordinate space. It has been applied to the scalar ϕ4 theory,
and obtained the beta function and mass anomalous dimension to seven loops, and the scalar self-energy
to eight loops [27]. Such computations can be performed by a normal desktop in a reasonable computation
time. However, although the graphical function method is well-suited for computing scalar integrals, it is
only getting started for general Feynman integrals with numerators [28]. Despite their achievements, each
of the aforementioned methods has its own limitation, and we need new methods in order to deal with
higher loop integrals with complicated numerators.
In this paper, we proposed a new approach to renormalization functions based on the ideas of operator
product expansion (OPE) and large momentum expansion. The OPE states that the product of two nearby
operators, ObI and O
bII , can be expanded by a list of local operators O
bi ,

x→y X
O
bII (x)O
bI (y) −
−−→ Ci (x − y)O
bi (y) . (1.1)
i

Using the large momentum expansion, the Wilson coefficients Ci can be expressed by a combination
of Z-factors of ObI , O bi , times a propagator-type integral1 . Given that the ϵ expansion of the
bII and O
propagator-type integral is known, the combination of Z-factors can be fixed by the UV finiteness of the
Wilson coefficient. Consequently, the Z-factor of a complicated operator can be expressed by that of a
relatively simple operator. The Z-factor of a given operator can be completely fixed by a proper chosen
set of OPE coefficients. The anomalous dimension are subsequently extracted from the corresponding
Z-factor. The method can also be applied to the computation of beta functions. In this way, we have
reduced the computation of renormalization factors to the evaluation of propagator-type integrals.
This paper is organized as follows. In §2, we start with a brief review of operator product expansion
and large momentum expansion, then we describe the general framework for computing Z-factors from the
large momentum expansion of correlation functions. In §3, this new framework is applied to the calculation
of the anomalous dimension of the ϕQ operator in 6-dimensional O(N ) cubic scalar theory up to four loops,
demonstrating its algorithm and validation. Discussions and extensions is provided in §4. In the appendix,
an example of extracting UV divergence from two-point functions is provided in §A, and the complete
result of four-loop correction of the scaling dimension ∆Q is given in §B.

2 A framework of computing renormalization functions

In this section, a method of computing renormalization functions will be developed, relying on the ideas
of OPE and large momentum expansion. Ultimately, the computation is reduced to the problem of com-
puting propagator-type integrals. The basic strategy for deriving this method can be outlined as follows.
1
A propagator-type integral refers to an integral with two external legs. Without confusion, we will call it a two-point
function in this paper.

–3–
Firstly, the correlation function of operators and fundamental fields is expanded as products of two-point
functions and lower-point correlation functions through large momentum expansion. Then, by arranging
the expansion terms, we can identify the quantity of two-point functions multiplied by a combination of
Z-factors as the Wilson coefficient of the leading-order operator basis in OPE. Finally, the UV finiteness
of the Wilson coefficient in dimensional regularization parameter ϵ ensures that the product of two-point
functions and the combination of Z-factors is also UV finite. This sets up a system of algebraic equations
for Z-factors by requiring the cancelation of ϵ poles, provided the ϵ expansion of two-point functions have
been computed. This introduces a major improvement from our method, that in order to solve Z-factors,
we only need to evaluate the integrals of two-point functions, which are much simpler than the original
correlation functions. By solving algebraic equations, the Z-factors can be determined recursively.
We will elaborate on above strategy in details in the following subsections. After a brief review of
the OPE and large momentum expansion techniques, in §2.3, we discuss the leading order contributions of
large momentum expansion for correlation functions of operators and fundamental fields in 6-dimensional
cubic or 4-dimensional quartic scalar theories. We introduce the minimal-cut prescription2 to identify
graphs contributing to the leading order terms of large momentum expansion. In §2.4 we re-examine
OPE in the viewpoint of large momentum expansion, and present consistent discussions on the leading
order contributions. With the theoretical background clarified, in §2.5 we provide a general proof on our
strategy of computing Z-factors, which would serve as a fundamental framework for designing computation
algorithms.

2.1 The operator product expansion

The product of two local operators exhibits asymptotic behavior when their coordinates tend to each
other. Such behavior can be described by OPE method with an expansion into a well-defined basis of local
operators of a given theory. Since product of operators is in general singular when they approach each
other, one expects a singular behavior for expansion coefficients with respect to the difference of coordinates.
The expansion basis would be the tensor products of fields, and possibly also relies on the classical scaling
dimensions ∆, etc. Hence the OPE of two operators O bII (x), O
bI (y) generally takes a schematic form as,

x→y
bµ1 ···µn (y) ,
X
O
bII (x)O
bI (y) −
−−→ Cµ∆,i
1 ···µn
(x − y)O ∆,i (2.1)
i,∆,n

where Obµ1 ···µn belongs to a complete set of expansion basis defined in coordinate y, and Cµ∆,i···µn (x − y)’s
∆,i 1
′
are Wilson coefficients depending only on the difference of coordinates x = x − y. The Lorentz indices of
Wilson coefficients can be carried by x′µ or spacetime measure η µν , while the latter can be contracted with
the operators of the expansion basis to produce lower rank ones. In this case, what remains is the tensor

2
Note that throughout this paper we use the notation cut to represent the splitting of Feynman graphs, which should not
be misunderstood as the famous unitarity-cut method.

–4–
structure of x′µ , and the Wilson coefficient in general behaves as

x′µ1 · · · x′µn
Cµ∆,i
1 ···µn
(x′ ) = c∆,i
n ∆I +∆II −∆+n , (2.2)
(x′2 ) 2

where the singular behavior is determined by scaling dimensions ∆I , ∆II of the original operators O bI , O
bII ,
and the coefficient cn∆,i is a function of logarithms ln(x′2 ). As expected, the powers and logarithms of x′2
enter into Wilson coefficients, showing generic asymptotic behavior of quantum field theory. Accordingly,
we can require the operators to be symmetric traceless tensors, making them irreducible representation of
SO(1, D − 1) group. Then the correlation functions of these operators are also symmetric and traceless,
and we do not need to worry about the mixing among operators of different tensor ranks.
Now let us consider the correlation function of both sides of eqn.(2.1) with fundamental fields ϕ(yi )’s,
and obtain the following relations among correlation functions,
D E x→y X D E
O
bII (x)O
bI (y)ϕ(y1 ) · · · ϕ(ym ) −−−→ Cµ∆,i (x − y) bµ1 ···µn (y)ϕ(y1 ) · · · ϕ(ym ) .
O (2.3)
1 ···µn ∆,i
i,∆,n

The LHS of above equation is a (m + 2)-point correlation function of two operators and m fields, while
the RHS consists of (m + 1)-point correlation functions3 . The Wilson coefficients can be determined by
matching the correlation functions on both sides of the equation. If all operators are renormalized, the
full correlation functions on both sides are UV finite (free of ϵ poles in dimensional regularization), so the
Wilson coefficients are also UV finite.
OPE is closely related to the large momentum expansion. To see this, let us Fourier transfer both
sides of the original OPE to momentum space, and get
Z X
OII (q1 )OI (q2 ) := d4 x d4 y eiq1 ·x e−iq1 ·y eiq1 ·y eiq2 ·y


b b Ci (x − y)O e 1 )O(q
bi (y) = C(q b 1 + q2 ) . (2.4)
i

The limit x → y in coordinate space corresponds to q1 → ∞ in momentum space, corresponding to a large

off-shell field, while (q1 + q2 ) remains finite. Thus we can take advantage of large momentum expansion
techniques to study the asymptotic behavior of correlation functions.

2.2 The large momentum expansion

The large momentum expansion is a technique for evaluating the asymptotic expansion of Feynman integrals
when two or more external momenta are large compared to other scales [29]. For our purpose, we will
focus on the large momentum expansion with two large external momenta, so that the integral contains a
single hard scale. From experience, we know that a Feynman integral FG is expected to have the following
behavior when expanded around small parameters such as, for instance, a ratio m2 /q 2 , with q being the
large momentum [29],
3
If the operators are replaced by fundamental fields, the above OPE also describes expansion of correlation functions of
off-shell fields.

–5–
 ∞ X
2L n
m2 m2

2 2 large q 2 limit 2 ω
X
FG (q , m ) −−−−−−−−→ (q ) cni lnj , (2.5)
n=n0 j=0
q2 q2

where L is the loop order of Feynman graph G and ω is the degree of divergence of the graph. Although
this expansion behavior is not a consequence of rigorous mathematical theorems, it is expected that such
asymptotic expansion is a general property of Feynman integrals. The difference between the original
Feynman integral and the asymptotic expansion to an order n = N is of order O((m2 /q 2 )N ), hence
it is possible to take several expansion terms to ensure that the difference approaches to infinitesimal.
Furthermore, different expansion orders have different q 2 -dependence, which enables us to distinguish
contributions from different expansion series.
In a Feynman graph G, the large momenta could have many ways of flowing in the sub-graph of G.
In each way, graph G can be split into two graphs, one with large momentum flow, and the other with all
soft momenta. Since large momentum expansion is only applied in large momentum flow, the asymptotic
behavior of corresponding Feynman integral under large momentum expansion then takes the form [29]
X
FG → FG/γ ◦ Mγ Fγ , (2.6)
γ

where γ is a sub-graph of G with all large external and internal momenta, and we will call it hard graph.
Mγ is an operation of Taylor expansion on the hard graph γ with respect to small expansion parameters.
G/γ is the graph of G by identifying all lines in γ to a vertex, which effectively becomes an off-shell
operator. We will call the graph G/γ as soft graph. The summation is over all possible configurations of
hard graphs.
The Taylor expansion Mγ defines expansion series, and at the integrand level it is applied to prop-
agators with large momenta around small parameters. For example if a propagator involves a large q as
1/(ℓ − q)2 , in the small region of ℓ, since both |ℓ2 |, |2q · ℓ| ≪ q 2 , this propagator can be Taylor expanded
around q → ∞ as
1 1 2q · ℓ − ℓ2 (2q · ℓ − ℓ2 )2
= + + + ··· (2.7)
(ℓ − q)2 q2 (q 2 )2 (q 2 )3
The leading order contribution is simply given by setting soft momentum ℓ to zero, which corresponds to
replacing the propagator by 1/q 2 , while the higher order contributions contain polynomials of ℓ and q in
numerators. In graph representation, the replacement of 1/q 2 changes the corresponding internal lines,
which effectively splits the original Feynman graph to a hard graph and a soft graph. This provides a
pictorial description of eqn.(2.6). For example, a scalar box integral at the large momentum limit of two
external large momenta q1 , q2 with large momentum flow represented by heavy lines, can be expanded
as a single line of large momenta times an 1-loop form factor, as sketched in Fig.(1). The leading order
term of the integrand is then given by product of hard graph and soft graph. For higher order terms of
asymptotic series, there will be powers of ℓ in the numerators, and eventually we get expansion terms with

–6–
 q2 p1 q2 p1

→
ℓ
ℓ
q1 p2 q1 p2

Figure 1. A scalar box integral at large momentum limit with large momentum flow represented by heavy lines,
where external q1 , q2 are taken to be large momenta. In large momentum expansion, this graph is effectively split
into a hard graph and a soft graph. The former contains only large momentum lines, and the latter is produced by
identifying all large momenta to a vertex while keeping soft momentum lines.

tensor structures of ℓ. The splitting of graph G generally separates integration parameters, and makes
integration simpler than the original one. Although the tensor structures in numerators would make
things complicated, it does not trouble us since in this paper we will only use the information of leading
order contributions. It is sufficient to concentrate on the leading order terms for developing a method of
computing renormalization functions.

2.3 Determining leading order term of large momentum expansion

The asymptotic expansion will express integral as series expansion of 1/q 2 , where q is the external mo-
mentum taken to the large limit. As mentioned, each expansion term has distinct 1/q 2 -dependence, so the
contributions of different order terms will not mix together, and the leading order term is free of tensor
structures.
The leading order term of an integral under large momentum expansion is explicitly computed by sum-
mation of products of hard and soft graphs, as described in eqn.(2.6). Pictorially, we define a terminology
of cutting to describe the splitting of Feynman graph. The hard and soft graphs are generated by cutting
graph G at all soft momentum lines that connecting directly to large momentum flow, splitting graph into
hard graph with only large momentum lines and the soft graph with only soft momentum lines. For the
latter one, all soft momentum lines that being cut are re-connected to a vertex4 . From the viewpoint of
external momenta, we are in fact considering all possible cuttings that can separate the large external
momenta with all other soft external momenta, in the condition that the hard graph are connected graph.
Such cuttings produce all possible large momentum flow of a graph.
Applying the cutting prescription to Feynman graphs of correlation functions with fundamental fields
(or fields with an operator inserted), and setting the momenta of two external leg (or one external leg and
one operator) to the large limit, the splitting of graphs can be represented as in Fig.(2). Note that the
k-cutting generates a two-point hard graph and a soft graph of fields with ϕk operator inserted. If for
instance the original graph describes ϕk+1 → (k + 1)ϕ correlation function, then the k-cutting relates the
4
Pictorially, the soft graph can also be considered as shrinking all lines of large momenta to a vertex.

–7–
 (a)

(b)

Figure 2. Contributions of large momentum expansion by cutting k soft lines. The dots in the middle of graph
represents k small propagators or external legs to be cut. The dashed lines represent null momenta in momentum
space or removed legs in coordinate space. (a) A k-cutting on the graph of correlation function of fundamental
fields, producing two-point functions of hard fields and form factors of soft fields. (b) A k-cutting on the graph of
correlation functions of one operator inserted in fundamental fields, producing two-point functions of a hard field
and an off-shell operator, and form factors of soft fields.

Figure 3. If cutting one internal line off the large momentum flow, we will get a tadpole soft graph, which is a
scaleless integral in massless theory and should be excluded.

correlation function to a lower-point one, with two-point function as expansion coefficient. In general, all
allowed cuttings applying on the internal and external lines that splitting a graph into hard graph and
soft graph will contribute to the asymptotic expansions. There is no constraint on the number of lines
being cut. This of course will produce a large amount of splitting possibilities for a graph. However we
will soon show that, the lowest order contribution of 1/q 2 that can be detected by a cutting depends on
the number of lines being cut. By selecting cuttings of a specific k lines, we can extract information for
terms contributing to a certain order of 1/q 2 in the asymptotic expansion. Before a detailed discussion, we
should mention that there is one type of cutting that should be excluded in consideration. If a propagator
of soft momentum is attached directly to the large momentum flow, after cutting it off and connecting two
ends of the internal line to a vertex, we would get a tadpole soft graph, as shown in Fig.(3). In massless
case it contributes to a scaleless integral and should be excluded. In the current paper, we are considering
massless theories, so only the cuttings on different soft lines are considered.

–8–
 Now let us consider the k-cutting of a graph shown in Fig.(2), and remark that the lines being cut
could be internal propagators or external soft legs. The large momentum expansion is determined by hard
graph, since it contains all large momenta. We want to determine the number of k that contributing to
the leading order terms of large momentum expansion. Without loss of generality, let us suppose the hard
graph contains nv vertices, ne propagators, and L = ne − nv + 1 number of loops. The integral of hard
graph scales in large momentum as
Z D D
d ℓ1 d ℓ2 · · · dD ℓL 1
∼ , (2.8)
D1 D2 · · · Dne − DL
(q ) 2 +ne
2

where D is the spacetime dimension and Di ’s are the propagators. Specifically, since we will discuss scalar
theories in dimensional regularization scheme in (4 − 2ϵ) or (6 − 2ϵ) dimension, the scale behavior can be
explicitly written as
1 1
D = 4 − 2ϵ : , D = 6 − 2ϵ : . (2.9)
(q 2 )2nv −ne −2+ϵL (q 2 )3nv −2ne −3+ϵL

The 1/(q 2 )ϵL factor after expansion around small ϵ provides logarithmic dependence of 1/q 2 , while the
other factor provides series expansion of 1/q 2 .
For various theories and different correlation functions, by inspecting structures of hard graphs, we
can compute the relations of nv , ne and the number k of cuttings. These relations set constraints on the
behaviors of 1/q 2 expansion, and consequently the leading order contribution of large momentum expansion
depends on the theories and physics quantities to be computed. If we are considering the two-point function
of hard graph with only cubic vertices in D = 6 − 2ϵ dimension, after cutting k lines of the original graph,
the hard graph has ne internal lines and nv vertices. Then counting number of lines and vertices, we get

3nv − (k + 2) = 2ne → 3nv − 2ne = k + 2 . (2.10)

Similarly, if we consider the two-point function of a hard field and a ϕQ operator5 in above theory, the
hard graph has one vertex of operator, (nv − 1) cubic vertices and ne internal lines. Then the counting
leads to
3(nv − 1) + Q − (k + 1) = 2ne → 3nv − 2ne = k + 4 − Q . (2.11)

In both cases, we get the relation of scale behaviors and number of cuttings as

two-point function 1 two-point function 1

: , : . (2.12)
of hard fields (q 2 )k−1+ϵL of hard field and operator (q 2 )k−(Q−1)+ϵL

The order of 1/q 2 depends on the number k. The minimal number of cuttings is determined by the
symmetry of theories. For correlation function of fundamental fields in cubic theory, the minimal number of
cuttings is k = 1. While for correlation function of fields with ϕQ operator inserted, by charge conservation
5
We postpone the explicit definition of ϕQ operator in §3.

–9–
the minimal number of cuttings is k = Q − 1. So from eqn.(2.12) we conclude that the leading order term
is given by the minimal cuts. Cutting more lines produces higher order contributions of large momentum
expansion.
The same discussion can be applied to theories with quartic vertices in D = 4 − 2ϵ dimension. In such
a theory, counting number of lines and vertices yields

two-point function k two-point function k−Q+5

: 2nv − ne = +1 , : 2nv − ne = . (2.13)
of hard fields 2 of hard field and operator 2

Then the scale behavior of 1/q 2 depends on the number of cuttings as

two-point function 1 two-point function 1

: k , : k−(Q−1)
. (2.14)
of hard fields −1+ϵL of hard field and operator +ϵL
(q 2 ) 2 (q 2 ) 2

By symmetry consideration, the minimal number of cuttings for two-point function of fields in quartic
theory is k = 2. While by charge conservation consideration, the minimal number of cuttings for two-point
function of fields with ϕQ operator inserted is k = Q − 1. From eqn.(2.14) we conclude again the leading
order term is given by minimal cuts of corresponding cases.
Computing k-cuttings of the Feynman graph rather than minimal cuts, we will get higher order terms
of 1/q 2 expansion. Taylor expansion of propagators will also generate higher order terms of 1/q 2 with
tensor structures in numerator. The mixing origins of higher order terms makes it a bit complicated to
compute contributions beyond leading order. However, since each order of 1/q 2 expansion is independent,
we can restrict ourselves within the leading order contributions produced by minimal cuts of graph. In
paper [27, 30], similar asymptotic expansion of graph is discussed, which is in fact the large momentum
expansion in coordinate space. In their treatment, all possible number of cuts are considered, leading to
results involving higher order contributions. Our treatment of minimal cuts greatly reduces the number of
expansion terms, and simplifies the computation.
Since we are planing to evaluate Feynman integrals in coordinate space, we can also analyze the scale
behaviors of hard graph in coordinate space, to get the leading order dependence of 1/q 2 expansion under
k-cutting. In coordinate space, the large momentum limit is realized by the limit xQ → 0, where xQ is the
coordinate of large leg. The asymptotic expansion is then described by series expansion of x2Q . The scale
behavior of two-point function of fields is slightly different from that of fields with ϕQ operator inserted,
since for the latter, the vertex of operator is considered to be external vertex and should not be integrated.
Following the hard graph in Fig.(2), we can compute the scale behavior of x2Q as

dD y1 dD y2 · · · dD ynv
Z
two-point function 2 Dnv
−λ(ne +2)
:  λ ∼ (xQ ) 2 , (2.15)
of hard fields
X12 · · · Xn2e (yi − xQ )2 (yj − x′2
Q)

dD y1 dD y2 · · · dD ynv −1
Z
two-point function 2
D(nv −1)
−λ(ne +1)
:  λ ∼ (xQ ) 2 , (2.16)
of hard field and operator 2 2 2
X1 · · · Xne (yi − xQ )

– 10 –
where λ = D−2 ′ 2
2 , yi ’s are labels of internal vertices, xQ , xQ are labels of external legs, while Xi ’s are
propagators in coordinate space. We still have relation ne − nv = L − 1, and eqn.(2.10), eqn.(2.11),
eqn.(2.13) for cubic and quartic theories. With these relations, we get the scale behaviors of hard graphs
as,

D = 6 − 2ϵ, cubic theory D = 4 − 2ϵ, quartic theory

two-point function k
(x2Q )k−2+ϵ(L+1) (x2Q ) 2 −1+ϵ(L+1)
of hard fields
two-point function k−(Q−1)
−1+ϵ(L+1)
(x2Q )k−(Q−1)−2+ϵ(L+1) (x2Q ) 2
of hard field and operator

Similar as discussed in momentum space, the leading order term of x2Q expansion is determined by minimal
cuts. For cases with ϕQ operator the minimal number is k = Q − 1, while for two-point function of
fundamental fields in cubic theory the minimal number is k = 1, and in quartic theory the minimal
number is k = 2. So the leading order term of hard graphs scales as,

D = 6 − 2ϵ, cubic theory D = 4 − 2ϵ, quartic theory

two-point function 1
x2Q
1
of hard fields
two-point function 1 1
(x2Q )2 x2Q
of hard field and operator

In our method of computing renormalization functions, we will make use of the information of leading
order contribution under large momentum expansion. From above discussion we know that, it can be
extracted by minimal cuts of Feynman graph either in momentum or coordinate space. Fig.(4) is an
example of scalar cubic theory in D = 6 − 2ϵ dimension showing the scale behaviors infected by the
number of cuttings, where the leading order term is given by k = 1 cutting shown in the third graph.
Notice that contributions of the k = 2 cutting and k = 3 cutting both contain scaleless tadpole integrals
in the soft graphs, thus they will not contribute to the final result. This observation makes it trivial for
asymptotic expansion of three-point functions with two large external momenta in cubic theory, since the
only possible way of splitting the hard and soft graphs without scaleless integrals is the k = 1 cutting that
removing the soft external leg from original graph, leaving only two-point functions.

2.4 Analyzing leading order term from OPE

The OPE in momentum space eqn.(2.4) clearly shows that the asymptotic expansion at large momentum
limit is described by series expansion of operator basis defined in the soft region, while the scale behavior of
large momentum is encoded in Wilson coefficients. Acting OPE on fundamental fields as eqn.(2.3), we get
an expansion of correlation functions into lower-point ones. Compared to the large momentum expansion
as represented in Fig.(2), we can infer that the Wilson coefficient corresponds to the two-point hard graph.

– 11 –
 cut 1
cut 2
cut 3
1 1 1
→ ∼ (q 2 )2
∼ (q 2 )1+2ϵ
∼ (q 2 )3ϵ

cut 1

Figure 4. Examples of k-cuttings of a Feynman graph for scalar cubic theory in D = 6 − 2ϵ dimension. Different
number of cutting contributes to different orders of 1/q 2 expansion.

An explicit connection between Wilson coefficients and hard graphs will be presented in the next subsection.
In our setup, we consider asymptotic expansion of two large external momenta, which corresponds to OPE
of two operators. Then Wilson coefficients of OPE can be computed by hard graphs of large momentum
expansion, and the correlation functions of operator basis with fundamental fields is equivalent to the
soft graphs. Different cuttings in large momentum expansion contribute to Wilson coefficients of different
operator basis as well as different orders of asymptotic expansion over x2Q in coordinate space. A complete
operator basis contains tensor operators, which can be identified as higher order terms of Taylor expansion
in large momentum expansion. However, at the leading order, we only need to focus on the lowest order
of operator basis.
The leading order term of asymptotic expansion in large momentum expansion can also be analyzed
by OPE, by counting the mass dimension of Wilson coefficients. Let us consider OPE in coordinate space
where the Wilson coefficient is a function of x2Q . In D = 6 − 2ϵ dimension, the mass dimension of a scalar
field of cubic theory is [ϕ] = 2. When both the two operators are set to be fundamental fields, since the
lowest order of operator basis is a single fundamental field, we get the OPE in leading order term as

ϕ(x)ϕ(0) = c0 (x2Q ) ϕ(0) + · · · , (2.17)

where c0 (x2Q ) is the Wilson coefficient of leading order term. Counting the mass dimension of both sides,
we get mass dimension [c0 ] = 2, so c0 (x2Q ) ∼ x12 . Similarly, if one operator is ϕQ and the other is ϕ̄, the
Q
lowest order of operator basis is ϕQ−1 , then

ϕQ (x)ϕ̄(0) = c0 (x2Q ) ϕQ−1 (0) + · · · . (2.18)

The equality of mass dimension in both sides leads to [c0 ] = 4, so the leading order term is c0 ∼ (x21 )2 . The
Q
same discussion can be applied to the quartic theory in D = 4 − 2ϵ dimension, where the mass dimension
of scalar field is [ϕ] = 1. Again if both the two operators are fundamental fields, and now the lowest order
of operator basis is ϕ2 , then OPE equation leads to

ϕ(x)ϕ(0) = c0 (x2Q ) ϕ2 (0) + · · · → [c0 ] = 0 → c0 ∼ 1 . (2.19)

If instead one operator is ϕQ , the equality of mass dimension in both sides of eqn.(2.18) requires [c0 ] = 2,
and we have c0 ∼ x12 . All the behaviors of leading order terms are in agreement with previous discussion
Q
of large momentum expansion.

– 12 –
2.5 Expansion of correlation functions and computation of Z-factors
From above general discussions we know that, the Wilson coefficient of OPE is proportional to the hard
graphs in large momentum expansion, and the latter are computed by two-point functions. Let us apply
this idea to the computation of Z-factors. Without loss of generality, let us consider the computation of
Z-factor for ϕQ operator in scalar theory. The anomalous dimension of ϕQ operator can be extracted from
the following correlation function, D E
ϕQ
R (x 0 )ϕ(x 1 ) · · · ϕ(x Q ) , (2.20)

where ϕQR is the renormalized operator. In order to compute this correlation function, we need to work
out the corresponding Feynman rules. The renormalized operator is related to the normal operator ϕQ
through bare operator ϕQ
0 as,
ϕQ Q Q
R = Z ϕQ ϕ 0 = Z ϕQ ϕ ,
e (2.21)
Q/2 Q Q/2
where we have used ϕQ0 = Zϕ ϕ and define ZϕQ := ZϕQ Zϕ . Therefore, we can determine the Feynman
e
rule of ϕQ Q
R -Qϕ vertex as ZϕQ . The anomalous dimension of ϕ will be determined by ZϕQ . Furthermore,
e
each loop propagator can be combined with counterterms to form an effective propagator as

i i 2 i
iZϕ−1
+ i(Z ϕ − 1)l + · · · = , (2.22)
l2 l2 l2 l2
which defines the Feynman rule for loop propagators. The Feynman rule for a scalar vertex reads
−iZg g µ̃2ϵ = −ig0 Zϕ2 , where µ is the scale parameter. The dependence of Z-factors in the complete cor-
relation function can be computed by counting all the Z-factors from operator vertex, loop propagators
and scalar vertex. For example, let us consider the quartic scalar model, and suppose a L-loop Feynman
diagram has one ϕQ 4
R -Qϕ vertex, nv ϕ vertices, and ne loop propagators. Then counting the number of
vertices, edges and loops we get the relations

Q + 4nv = 2ne + Q , (nv + 1) − ne + L = 1 → nv = L , ne = 2L . (2.23)

So the diagram comes with the following combination of Z factors,

eϕQ Z −ne (g0 Z 2 )nv = ZeϕQ g0L .

Z (2.24)
ϕ ϕ

For L-loop Feynman diagram, the Z-factor appears as an overall factor ZeϕQ universally, and the loop order
only shows up in couplings. Hence the all-loop correlation function can be written as
D E ∞
(L)
ϕQ
X
(x
R 0 )ϕ(x 1 ) · · · ϕ(x Q ) = Z
e ϕ Q g0L GQ (x0 , x1 , . . . , xQ ) := ZeϕQ GQ (g0 ; x0 , x1 , . . . , xQ ) , (2.25)
L=0

where GQ (g0 ; x0 , · · · , xQ ) is the all-loop bare correlation function.

For general quantum field theory, the correlation function has the same structure as
D E
ObR (x0 )ϕ(x1 ) · · · ϕ(xn ) = Z
eO GO (gi0 ; x0 , x1 , . . . , xn ) , (2.26)

– 13 –
 p1 p1
p2 p2
ϕQ ϕQ ϕQ−1
→
P
L L1 ◦ L2

pQ−1 pQ−1
pQ pQ

Figure 5. Pictorial representation of large momentum expansion for ϕQ → Qϕ form factor, where momentum of
the operator and an external momentum pQ are taken to the large limit. By applying minimal cuts to the graph, we
extract the leading order contribution of expansion. The expansion terms are described by hard graphs of two-point
functions with large momenta, and soft graphs of lower-point correlation functions with on-shell soft momenta.

where gi0 represents one or more bare coupling constants of the theory, and all coupling constants gi ’s are
replaced by bare coupling constants gi0 ’s at the end. In fact, correlation functions of fundamental fields
also have similar structures. For example, in ϕ4 theory, we have
D E D E
ϕ(x1 )ϕ(x2 ) = Zϕ G(g0 ; x1 , x2 ) , ϕ(x1 )ϕ(x2 )ϕ(x3 )ϕ(x4 ) = Zϕ2 G(g0 ; x1 , x2 , x3 , x4 ) , (2.27)

and the Z-factors appear as universal overall factor. We will shortly see that, the UV finiteness of correlation
functions determines the overall Z-factors, and also gives constraints to lower-loop correlations of gi0 .
Now let us apply large momentum expansion to correlation function eqn.(2.25) and extract leading
order contribution of asymptotic expansion. This can be realized by applying minimal cuts to the Feynman
graphs. When taking the momentum of operator and an external momenta pQ to the large limit, and
assuming pQ ≫ pi , ∀i ̸= Q, we can asymptotically expand the correlation function around large pQ , as
sketched in Fig.(5). For ϕQ operator, we should consider all possible minimal (Q−1)-cuttings of the original
Feynman graph. Consequently, the expansion terms are given by products of hard graphs of two-point
functions and soft graphs of ϕQ−1 → (Q−1)ϕ form factors with on-shell soft momenta. In coordinate space,
assuming the coordinate of operator is x0 and that of large momentum pQ is xQ , the large momentum
expansion is realized by taking xQ → x0 limit.
Applying large momentum expansion on the RHS of eqn.(2.25), the L-loop correlation function is
expanded as

L
(L) (L−i) (i)
GQ (x0 , x1 , x2 , . . . , xQ ) ∼
X
= GQ (x0 , x1 , . . . , xQ−1 , xQ )GQ−1 (x0 , x1 , x2 , . . . , xQ−1 ) , (2.28)
i=0

where xi represents a corresponding external leg removed. The summation is over all possible minimal
(Q − 1)-cuttings that split Feynman graph to 1PI sub-graphs, producing leading order terms in the large
momentum expansion. Substituting above expansion back to the RHS of eqn.(2.25) and switching the

– 14 –
order of summations, we get
∞ X
∞
(L−i) (i)
X
ZeϕQ g0L GQ (x0 , x1 , . . . , xQ−1 , xQ )GQ−1 (x0 , x1 , x2 , . . . , xQ−1 ) (2.29)
i=0 L−i=0
∞ ∞
! !
(i) (L−i)
X X
=Z
eϕQ g0i GQ−1 (x0 , x1 , x2 , . . . , xQ−1 ) g0L−i GQ (x0 , x1 , . . . , xQ−1 , xQ ) .
i=0 L−i=0

Expression in the first bracket times ZeϕQ−1 is the full correlation function of ϕQ−1 → (Q − 1)ϕ. Hence we
get the relation
 
∞
∼ ZϕQ ϕQ−1 (x0 )ϕ(x1 ) · · · ϕ(xQ−1 ) 
D E e D E
(j)
ϕQ g0j GQ (x0 , x1 , . . . , xQ−1 , xQ ) .
X
R (x0 )ϕ(x1 ) · · · ϕ(xQ ) = e R
ZϕQ−1 j=0
(2.30)
Since both correlation functions are UV finite, the expansion coefficient should also be UV finite in dimen-
sional regularization parameter ϵ. The finiteness condition constraints the ratio of Z-factors by relation
as,
 
∞
ZeϕQ (j) ZeϕQ
g0j GQ (x0 , x1 , . . . , xQ−1 , xQ ) :=
X
G (g ; x , x , . . . , xQ−1 , xQ ) = UV finite . (2.31)
eϕQ−1 Q 0 0 1

Z
eϕQ−1
j=0 Z

From the UV finiteness condition and the ϵ expansion results of two-point functions, the ratio of Z-factors
can be completed determined. The minimal cuts of large momentum expansion extract the leading order
terms of x2Q expansion, while there is also lnk (x2Q )-dependence in the result. Since each rank of logarithms
is also independent, we can select the coefficients of leading order logarithm ln0 (x2Q ) in the result of leading
x2Q contribution to fit the ratio of Z-factors. The coefficients of higher rank logarithms also generate
system of algebraic equations from UV finiteness condition, and they can solve Z-factors of lower-loop
beta functions. The above relation is valid for any loop order and Q. Once the two-point functions are
computed, the Z-factors can be determined recursively by UV finiteness conditions loop by loop.
Comparing eqn.(2.30) with the OPE expansion eqn.(2.3), we see that the eqn.(2.31) is exactly the
Wilson coefficient. In OPE, the Wilson coefficient is singular in x2Q but finite in dimensional regularization
parameter ϵ, and it is the origin of UV finiteness condition. It is now clear that the leading order term
of x2Q series in large momentum expansion corresponds to the Wilson coefficient of leading order operator
basis in OPE. We want to emphasize again that, in the above framework of determining Z-factors, the only
input is the two-point functions of hard field and ϕQ operator, and the corresponding Feynman integral
is an one scale function. The Feynman graphs of two-point functions can be generated by removing
(Q − 1) external legs from the original Feynman graphs of ϕQ → Qϕ correlation functions. Similarly, when
computing anomalous dimension of field, mass or couplings, we need to deal with three-point and four-
point correlation functions. In the framework eqn.(2.31), what we really need to compute is the two-point
function generated by graphs of three or four-point functions with one or two external legs removed. For

– 15 –
scalar theories, we choose to compute these two-point functions by graphical function method, which can
be evaluated by running TwoPoint command in HyperlogProcedures package [31] with the coordinate of
ϕQ operator locating at point 0 and coordinate of hard field locating at point 1.
(L)
In dimensional regularization, the L-loop two-point function GQ (x0 , x1 , . . . , xQ−1 , xQ ) can be com-
puted as an expansion in dimensional regulator ϵ with leading order pole 1/ϵL . In order to correctly
generate equations of UV finiteness condition, the lower-loop two-point functions should be computed to
appropriate orders of rational ϵ terms. Generally in the HyperlogProcedures package, computing integrals
to one order higher ϵ terms would cause several times greater in time consumption. This is a limit for
pushing the computation of renormalization functions towards the next-loop challenge. The determination
of Z
eϕQ can be done recursively with increasing parameter Q and loop order L. In fact, the finiteness
conditions allow us to completely fix all ratios of Z-factors with different values of Q and L simultaneously.
Hence each ZeϕQ is determined when all ratios are solved. Then we can follow the standard textbook
method to compute anomalous dimensions, etc.
The above method of computing Z-factors applies to correlation functions of fundamental fields with
ϕQ operator inserted in either 6-dimensional cubic scalar theory or 4-dimensional quartic scalar theory.
It also applies to correlation functions of fundamental fields. For 6-dimensional cubic scalar theory, the
minimal number of cuts is 1. This allows us to compute Z-factors from related three-point functions by
removing one external leg. For 4-dimensional quartic scalar theory, the minimal number of cuts is 2, and
it allows us to compute Z-factors from corresponding four-point functions with two external legs removed.

3 The anomalous dimension of ϕQ operator in O(N ) cubic scalar theory

As a demonstration of the method for computing renormalization functions, in this section we will compute
the anomalous dimensions of ϕQ operators in cubic scalar theory to 4-loop, and verify the validation and
effectiveness of our algorithm.

3.1 The O(N ) model and ϕQ operator

The renormalization of scalar ϕ4 theory with O(N ) symmetry has long been the focus of study, because
critical exponents in different statistical systems can be obtained from the scaling dimensions of operators at
the Wilson-Fisher fixed point [16, 17, 32]. The beta function and anomalous dimensions of some operators,
e.g., ϕ and ϕQ , are computed to 6-loop using R∗ operation method [33], and recently to 7-loop using
graphical functions [27]. Physical quantities usually take simpler form when some parameters become very
large. The large N expansion of the scaling dimensions of various operators to the O( N13 ) order can be
obtained by the conformal bootstrap method in an entire range of spacetime dimensions [34]. Meanwhile,
the semiclassical method has been generalized to compute the scaling dimension of ϕQ -type operators in the
large charge limit for various theories to all loops, including theories with scalars and gauge fields [35–42].
But to obtain the complete results of anomalous dimensions, we still need the help of the perturbation
method.

– 16 –
 In [43], it is proposed that in 6 − ϵ dimension, the O(N ) ϕ4 scalar theory at its fixed point is equivalent
to the following O(N ) ϕ3 scalar theory,

1 1 g h
L = (∂ϕi )2 + (∂σ)2 − σϕ2i − σ 3 , (3.1)
2 2 2 6

with two kinds of scalar fields σ and ϕi , i = 1, . . . , N . The equivalence is further verified to 4 and 5-loops
[44–47] for ϕ and ϕ2 operators in large N limit, and it is verified to 3-loop for the anomalous dimensions
of ϕQ operators [48]. In this section, we will compute the anomalous dimensions of ϕQ operators in
six-dimensional cubic theory to 4-loop.
The operator ϕQ is defined as the symmetric traceless operator

ϕQ := Ti1 ···iQ ϕi1 · · · ϕiQ . (3.2)

As discussed in [7], it is convenient to choose

1
Ti1 ···iQ = qi · · · qiQ , (3.3)
Q! 1

so that the operator becomes

1 Q
ϕQ := φ , (3.4)
Q!
in which q is an auxiliary O(N ) zero-norm complex vector satisfying q 2 = 0, q · q̄ = 1, and φ is a redefinition
of scalar field φ := q · ϕ. With this definition, we can strip off O(N ) indices from the correlation function,
and compute a simpler φQ → Qφ correlation function.
As proven in §2.5, the ratio ZeφQ /ZeφQ−1 can be determined from the UV finiteness of the Wilson
coefficient CQ ,
 
L
Z
eφQ (j)
g0j GQ (x0 , x1 , . . . , xQ−1 , xQ ) = UV finite ,
X
CQ (xQ ) =  (3.5)
ZφQ−1
e
j=0

up to the desired loop order L. The above relation requires us to compute the two-point function GQ ,
which is generated from φQ → Qφ correlation functions by removing (Q − 1) external legs, leaving only
the off-shell operator and one external leg as large momenta in large momentum expansion.

3.2 Treatments on the Feynman graphs

When Q and the loop order L are large, it can be difficult to generate all contributing Feynman graphs.
This has already been a problem for 4-loop case when Q takes the value of 4 or 5. Hence it is better to
simplify the problem of generating Feynman graphs as much as possible by using properties of Lagrangian
and graphs. Here are some techniques that we adopted to reformulate Feynman graphs before computing
integrals.

– 17 –
Summing over internal states
One difficulty of the considered cubic theory is that we need to consider two different scalar fields, which
increases the number of Feynman graphs drastically. We circumvent this difficulty by combining two scalar
fields into a single (N + 1) component scalar field Φ as,

ΦI = (Φ0 , Φi , . . . , ΦN ) := (σ, φ1 , . . . , φN ) . (3.6)

Then the cubic potential terms can be written as

1
V (Φ) = λIJK ΦI ΦJ ΦK . (3.7)
6
The coupling λIJK is symmetric, and the original couplings h, g have been defined as components of λIJK ,

λ000 = h , λ0ij = gδij , ∀i, j ̸= 0 , (3.8)

and all other components zero. In this definition, contributions of different σ, φi ’s internal state config-
urations of the same topology can be combined into a coupling factor α(g, h, N ), which can be obtained
from the products of λIJK ’s. For example, let us consider an 1-loop 2-point Feynman graph of Φ field as
shown below,
I3 I3
I1 I2
I4 I4

The coupling factor for this graph is a product of two cubic vertices, which reads
N
X
αI1 I2 (g, h, N ) = λI1 I3 I4 λI2 I4 I3 , (3.9)
I3 ,I4 =0

where I1 and I2 are indices of external legs. If the external legs are σ fields, we set I1 = I2 = 0 and obtain
N
X
α00 (g, h, N ) = λ2000 + λ20ii = h2 + N g 2 . (3.10)
i=1
In the graphs of σ and φi fields, this coupling factor just shows the fact that, the graph with σ fields as
internal propagators and those with φ’s fields as internal propagators can be combined together, with a
scalar integral of the topology as overall factor. While if the external legs are φ fields, the only possible
internal state configuration is a σ propagator and a φi propagator. By setting I1 = i1 , I2 = i2 and
computing the coupling factor, we get
N
X N
X
αi1 i2 (g, h, N ) = λi1 0I4 λi2 I4 0 + λi1 I3 0 λi2 0I3 = 2g 2 δi1 i2 , (3.11)
I4 =1 I3 =1

which is as expected from the graphs of σ, φi fields. The coupling factor of φQ correlation functions can be
evaluated in the same manner. In this treatment, color factors of graphs with different internal states but
the same topology can be organized as a single coupling factor α(g, h, N ), and we only need to compute
the integrals of independent topologies, getting rid of the abundant graphs by assigning internal states.

– 18 –
 (a) irreducible graph (b) reducible graph

Figure 6. (a) For irreducible graph, all external legs are connected to the operator vertex via the same loop sub-
graph. (b) For reducible graph, external legs are distributed in two or more different loop sub-graphs connecting to
the operator vertex. The reducible graph can be expressed as products of several irreducible graphs.

Reducible and irreducible graphs

The Feynman graphs of two-point function GQ (x0 , x1 , . . . , xQ−1 , xQ ) are generated from Feynman graphs
of φQ → Qφ correlation function by removing (Q−1) external legs. We will show that, after the removal of
external legs we only need to compute a part of the original graphs. All the Feynman graphs of φQ → Qφ
correlation function can be classified as irreducible graphs and reducible graphs, according to the way of
external legs connecting to the operator vertex. An external leg can attach to the operator vertex directly
or via a loop sub-graph. The directly connected leg gives an identity in the integrand, which could be
trivially omitted. We can now classify the graphs with external legs connected to operator vertex via loop
sub-graphs. If all external legs are connected to the same loop sub-graph, we define it as an irreducible
graph. While if all external legs are distributed in more than one loop sub-graphs, we call it a reducible
graph, as illustrated in Fig.(6).
From the graphs we can infer that, the integrand of reducible graph can be decomposed into integrands
of several irreducible graphs. The L-loop decomposition formula for arbitrary Q is given in [7], and here
we present an all-loop expression,

[Q/2] k
X X Q! 1 Y 1
GQ (x0 , x1 , . . . , xQ ) = Fn , (3.12)
ni ! i
Pk
2≤n1 ≤···≤nk (Q − i=1 ni )!
Sn1 ···nk
k=0 i=1
n1 +···+nk ≤Q

where k is a label denoting the number of loop sub-graphs, and k = 0 stands for contributions of tree-level
graphs. The Sn1 ···nk is a symmetry factor, with the definition that when there are duplicated ni ’s it gets a
non-trivial factorial value for each duplication. For instance S11444 = 2!3! with 2! from double duplicated

– 19 –
indices 1 and 3! from triple duplicated indices 4. The Fni denotes integrand of irreducible graph with ni
external legs6 .
A reducible graph does not contribute to the Wilson coefficient CQ (xQ ) defined in eqn.(3.5). Because
when external legs x1 , . . . , xQ−1 are removed, as can be inferred from Fig.(6), there would be at most one
sub-graph with external leg xQ , while other sub-graphs will inevitably become a scaleless snail graphs, and
thus vanish. For instance, a reducible graph with four external legs distributing in two loop sub-graphs
after removing three legs becomes

F2 (x1 , x2 )F2 (x3 , x4 ) → F2 (x1 , x2 )F2 (x3 , x4 ) = 0 , (3.14)

since F2 (x1 , x2 ) is a scaleless integral.

An irreducible diagram can contribute to CQ (xQ ) only if xQ is not attached directly to the operator
vertex. Otherwise after removing all the other legs, the loop sub-graph will be connected to external
momenta only via operator vertex and the whole graph becomes a scaleless snail diagram, as
1 1
Fn (x1 , · · · , xn ) → Fn (x1 , · · · , xn ) = 0 . (3.15)
(x2n+1 · · · x2Q )λ x2λ
Q

Among all the terms in the decomposition eqn.(3.12), the irreducible graphs are represented by terms
with k = 1, which is
Q Q
X Q! 1 X
n
Fn = CQ Fn , (3.16)
(Q − n)! n!
n=2 n=2
where CQ n is the binomial coefficient. The summation is over all possible distributions of Q external legs

to the loop sub-graph and the operator vertex. For a given Fn , the factor CQ n just stands for the fact

that because of permutation symmetry, there are CQ n copies of F contributing to the same local UV
n
counterterm. However we should exclude graphs with xQ directly connected to the operator vertex, which
n n − Cn n−1
has CQ−1 copies. Then the remaining CQ Q−1 = CQ−1 copies of Fn have non-vanishing contributions
to the Wilson coefficients. For example,
1
Fn (xn−Q+1 , · · · , xQ ) → Fn (xn−Q+1 , · · · , xQ−1 , xQ ) ̸= 0 , (3.17)
(x21 · · · x2n−Q )λ
6
The graphs of correlation function could have external legs connected directly to the operator vertex besides the irreducible
graph. Since graphs which are related by permutations of external legs produce the same local UV counterterm, we can formally
write the permutation related graphs as copies of one graph. For instance, the following four graphs

x1 x2 x3 x4
x2 x1 x1 x1
x3 x3 x2 x2
x4 x4 x4 x3

can be formally written as

1 1 1 1
4F3 := F3 (x2 , x3 , x4 ) + 2λ F3 (x1 , x3 , x4 ) + 2λ F3 (x1 , x2 , x4 ) 2λ F3 (x1 , x2 , x3 ) , (3.13)
x2λ
1 x2 x 3 x4

where 1/x2λ
i is the Feynman rule for external leg in coordinate space.

– 20 –
 Q=2 Q=3 Q=4 Q=5 Total
L=1 1 - - - 1
L=2 7 4 - - 11
L=3 56 67 19 - 142
L=4 540 999 586 107 2232

Table 1. The number of two-point graphs contributing to the two-point function of irreducible L-loop φQ → Qφ
correlation function up to 4-loop. This is the number of graphs we need to compute by graphical function method.

where the resulting Fn is a two-point function with external leg xQ and operator vertex. Eventually we
obtain the contributions for two-point functions as

Q
1 X
n−1
GQ (x0 , xσ1 , . . . , xσQ−1 , xQ ) = 2λ
+ CQ−1 Fn (xσ1 , · · · , xσn−1 , xQ ) , (3.18)
xQ n=2

where the first term comes from contributions of tree-level graphs, and the xσ1 , . . . , xσn−1 of Fn in the
second term is an arbitrary subset of legs x1 , . . . , xQ−1 , which should be removed. Now the original two-
point function can be computed from all irreducible two-point graphs, which further reduces the number
of graphs we need to compute.
In order to compute Fn (xσ1 , · · · , xσn−1 , xQ ), firstly we generate all 1PI irreducible topologies con-
tributing to the φn → nφ correlation functions Fn (xσ1 , · · · , xσn−1 , xQ ) with n external legs and an operator
vertex located at x0 , which will be denoted by Ti ’s. Then for each Ti , we generate n graphs by treating
one external leg as special and compute the non-isomorphic graphs from them. The resulting graphs will
be denoted by {Ti,1 , . . . , Ti,mi }. In coordinate space, each Ti,a will be computed by removing all external
legs but keeping only the special one, leaving a two-point graph with the operator vertex and the surviving
external leg, and the two-point graph can be evaluated by graphical function method. Then the two-point
function of irreducible graphs can be expressed as,

#.topo. mi
X X Ti,a
Fn (xσ1 , · · · , xσn−1 , xQ ) = (n − 1)! αi (g, h, N ) , (3.19)
Si,a
i=1 a=1

in which the summation of i runs over all 1PI irreducible topologies of Fn , and αi (g, h, N ) is the coupling
factor of topology Ti . The Si,a is the full symmetry factor of graph Ti,a , including the symmetry under per-
mutation of vertices and internal propagators, as well as the symmetry under permutation of φ1 , . . . , φn−1
external legs.
The number of two-point graphs for L-loop φQ → Qφ correlation function is shown in Table.(1), and
these graphs will be computed by graphical function method in dimensional regularization scheme to get
the ϵ series results. A sample computation can be found in Appendix §A.

– 21 –
3.3 The anomalous dimensions
In order to obtain the 4-loop correction of γφQ , we computed the ratio Z eφQ /ZeφQ−1 up to 4-loop for
Q = 2, 3, 4, 5. Using Z
eφ1 = 1, we obtain Z
eφQ for Q = 2, 3, 4, 5, which are suffice to fix Z
eφQ to 4-loop for
arbitrary Q. The anomalous dimension of φQ is given by
Q
eφQ Z − 2

∂ ln ZφQ ∂ ln Z ϕ
γφQ = = . (3.20)
∂ ln µ ∂ ln µ
We computed Zϕ and Zσ to 5-loop by evaluating ⟨ϕϕ⟩ and ⟨σσ⟩ correlation functions using two-point
functions directly. Zg and Zh to 4-loop are obtained as side-products, from the UV finiteness of ln(x2Q )
terms in the two-point functions. Our results of γϕ , γσ , β(g), β(h) are consistent with that of [45, 46].
We also computed Zg and Zh directly with the help of OPE,

φ(x)φ̄(0) ∼ C1 (x)σ(0) + O(|x|−1 ) , σ(x)σ(0) ∼ C2 (x)σ(0) + O(|x|−1 ) , (3.21)

and the combinations Zg /Zσ and Zh /Zσ can be determined by the UV finiteness of C1 and C2 , respectively
as,
Zg ϵ Zh ϵ
C1 (x2 ) = g µ̃ G1 (g0 , h0 ; x0 , x1 , x2 ) , C2 (x2 ) = hµ̃ G2 (g0 , h0 ; x0 , x1 , x2 ) . (3.22)
Zσ Zσ
The corresponding two-point functions are generated by removing one leg from three-point correlation
functions G1 and G2 defined as
D E D E
φ̄(0)φ(x2 )σ(x1 ) = Zg g µ̃ϵ G1 (g0 , h0 ; x0 , x1 , x2 ) , σ(0)σ(x2 )σ(x1 ) = Zh hµ̃ϵ G2 (g0 , h0 ; x0 , x1 , x2 ) .

The results are consistent with previous computation.

The 4-loop correction to γφQ is given in Appendix §B. The scaling dimension of φQ at the Wilson-Fisher
fixed point is given by
△Q = (2 − ϵ)Q + γφQ ∗ ∗
. (3.23)
g=g ,h=h

The leading and sub-leading order terms of △Q in the large N limit were computed in [48], (see also
[49–51]),
Q Q 1
△Q = (2 − ϵ)Q + △1Q + 2 △2Q + O( 3 ) . (3.24)
N N N
Our perturbative computation gives
Q 1 Q
△Q = (2 − ϵ)Q + △Q + 2 △2Q , (3.25)
N N
with
   
−32 56 11Q
△1Q = ϵ(8 − 6Q) + ϵ 2 3
+ 7Q + ϵ − +
3 9 2
  
4 128 19
+ ϵ 16ζ3 − +Q − 12ζ3 , (3.26)
27 4

– 22 –
and
 
−6272
△2Q = ϵ(352 − 264Q) + ϵ2 + 1714Q − 180Q2
3
   
3 16420 −3743 2
+ϵ +Q − 864ζ3 + 576ζ3 + 6Q (31 + 48ζ3 )
9 3
48π 4 −14930 72π 4
  
4 38630
+ϵ + + 1184ζ(3) + Q − + 336ζ3
27 5 9 5
24π 4
 
2
+ Q 334 + − 816ζ3 . (3.27)
5
These expressions are consistent with the large N computation to the order O(ϵ4 ).

4 Conclusion

In this paper, we proposed a method for computing Z-factors and anomalous dimensions of fundamental
fields or composite operators for 6-dimensional cubic and 4-dimensional quartic scalar theories with O(N )
symmetry. This method is based on the ideas of OPE and large momentum expansion. Through careful
analysis of the leading order contribution of expansion, we obtain the UV finiteness conditions for Z-factors
as eqn.(2.31). By requiring the cancelation of ϵ poles, a system of algebraic equations can be generated
that solves Z-factors recursively. The advantages of our method can be summarized as follows. Firstly,
all Z-factors can be determined simultaneously by solving algebraic equations of UV finiteness conditions,
and additional equations from logarithm terms of expansion allow us to determine Z-factors related to
lower-loop higher-point correlation functions. Secondly, the UV divergence is computed from two-point
functions, and we do not need to consider sub-divergence. Compared to the original correlation functions,
two-point functions have fewer Feynman graphs, and the Feynman integrals are much simpler to compute.
The two-point functions are scalar integrals, which are universal and once computed, can be used in other
theories. Thirdly, we can easily transform the integrals of two-point functions into coordinate space, and
integrate them by the graphical function method. The related Maple package HyperlogProcedures has
been proven to be a very powerful tool for computing scalar integrals in the dimensional regularization
scheme with high precision. This new method is demonstrated by computing anomalous dimension of the
ϕQ operator in 6-dimensional cubic scalar theory up to four loops, and its validation and efficiency are
confirmed.
A straightforward extension of this paper would be to compute the anomalous dimension of the ϕQ
operator in cubic scalar theory to five loops. The cubic theory is asymptotically free in 6-dimensions.
Using the graphical function method, the five-loop beta function and the anomalous dimensions of field
and mass have already been computed in [47]. Additionally, the perturbative computation of the five-
loop anomalous dimension of the ϕQ operator for all Q-dependence has also been computed in [7] by
the unitarity-cut method. We will perform this renormalization computation at five loops using the new
method presented in this paper in the forthcoming work and explain the details of the computation.

– 23 –
 Although scalar theories are considered throughout this paper, the idea of expanding a correlation
function as a hard two-point function and a soft lower-point correlation function is applicable to any
theories. The framework of this method can be generalized to theories with fermions and gauge fields.
However, this generalization will introduce non-trivial numerators in the two-point functions. One possible
approach to computing the UV divergence of these two-point functions is to generalize the graphical
function method to integrals with numerator structure [28]. But it is still premature to wait for an
available computation package for this purpose. Alternatively, it is possible to combine this method with
the IBP method to reduce the hard two-point functions into master integrals and extract UV divergence
from known master integrals. Nevertheless, the computation of master integrals will also pose a challenge.
It might be a viable option to seek assistance from available packages such as NeatIBP [52] and the auxiliary
mass flow method [53]. This direction is under investigation.
Another challenging direction lies in applying this method to higher dimensional operators with deriva-
tives. For instance, in [54], the symmetric traceless operators ϕi ∂µ1 ···µn ϕj are considered, and the O( N1 )
and O( N12 ) order scaling dimensions are computed based on conformal field theory. We can carry out a
perturbative computation of the anomalous dimensions and verify their results. When considering higher
dimensional operators, the mixing of operators in the OPE would cause some difficulties. We are attempt-
ing to solve this problem in future work by conducting a more sophisticated analysis of the OPE in hard
fields.

Acknowledgments

We would like to thank Oliver Schnetz for stimulating conversations on the graphical function method and
the usage of HyperlogProcedures package. We would also like to thank Bo Feng and Gang Yang for helpful
discussions. RH is supported by the National Natural Science Foundation of China (NSFC) with Grant
No.11805102.

A Sample UV divergence computation of two-point function

Let us explain the computation strategy by an example of 3-loop φQ → Qφ correlation function with
Q = 4. We can generate 7 independent topologies for this correlation function, labeled by T1 , . . . , T7 .
Taking one of them for further investigation, say T1 , which is sketched below as,

It has four external legs, and if we treat each leg as special, there would be four possible configurations, as
highlighted by red lines below

– 24 –
 T1,1 T1,1 T1,2 T1,2

However, because of the symmetry of graphs, for this topology we find that the first and second graphs
are isomorphic, while the third and last graphs are also isomorphic. So computing non-isomorphic graphs
we get only two independent graphs, which can be labeled by T1,1 , T1,2 . For 3-loop Q = 4 case, the 7
independent topologies generate in total 19 independent graphs with one leg special.
Then we should compute the Feynman integrals of these graphs. Since in coordinate space, all external
legs but the special one are removed, we are in fact computing simplified Feynman graphs with fewer
propagators. For instance, the graph T1,1 becomes

z1

z0

where solid dot in the corner represents vertex without external leg, which produces double propagators.
By graphical function method, this graph can be computed by Maple package HyperlogProcedures, with
command TwoPoint by setting z0 = 0, z1 = 1 in 6-dimension. It produces a result up to order ϵ as7 ,

47 π 2 1 83 13π 2 719 47π 2 13ζ3 13π 4

       
3ϵ 1 13 ζ3
T1,1 = (z z̄) + + − + − − + − − + ϵ .
3ϵ3 6ϵ2 6 24 ϵ 4 48 12 16 48 24 5760

Similarly, computing all Ti,a ’s and summing them over by eqn.(3.19), we obtain the corresponding two-
point function as ϵ expansion series in dimensional regularization scheme. We remark that, computation
of one degree higher in ϵ order might already causes drastic increasing in time consuming. However, for
φQ → Qφ correlation function in 6-dimensional cubic scalar theory, all graphs are computed to sufficient
ϵ order in an acceptable short time by a desktop.

B The 4-loop correction of scaling dimension ∆Q

Here we present our result of scaling dimension ∆Q in 6-dimensional cubic scalar theory, and it has been
compared to the large N expansion results in literatures with agreement. The 4-loop correction of ∆Q is
7
In D = 6 − ϵ convention.

– 25 –
given by
8
(4)
X
∆Q = Q g i h8−i δi4 , (B.1)
i=2

in which
   
−69419 29239Q 35 29Q 1 Q
δ24 = + + − ζ3 + − π4 , (B.2)
839808 559872 1296 1728 6480 8640

−94597 6665Q 487Q2 157Q 79Q2 Q2

   
97 −1 Q
δ34 = + − + − + ζ3 + + − π4 , (B.3)
186624 7776 1728 144 144 216 270 240 1080

3960457 719N 1195859Q 17027N Q 989Q2 107Q3

δ44 = + − − + −
839808 52488 186624 2239488 486 576
2 Q3
 
115 N 163Q NQ 179Q
+ − − + + − ζ3 , (B.4)
24 1296 27 1728 108 6
Q2 −205 175Q 25Q2 5Q3
   
31 43Q
+ − + π4 + + − + ζ5
3240 4320 540 18 12 6 12

−88195 63437N 157607Q 739N Q 94337Q2 31N Q2 199Q3 Q4

δ54 = + + − − + + −
17496 279936 15552 2592 15552 486 144 8
6173Q 25N Q 1381Q2 13N Q2 19Q3
 
−635 31N
+ − + + − − + ζ3
36 144 216 72 108 108 12
, (B.5)
11Q N Q 11Q2 N Q2
 
37 N 4
+ + − − + + π
1080 1080 270 720 1080 2160
65 75Q 115Q2 10Q3
 
+ − + − ζ5
3 2 6 3

13501805 742885N 419N 2 593933Q 1535347N Q 29N 2 Q

δ64 = − − − + −
839808 279936 839808 34992 559872 373248
20201Q 2 17365N Q 2 197Q3 25N Q3 5Q4
+ − + + −
7776 31104 288 576 16
2 2 2 2 19Q3
 
19849 43N N 2890Q 2633N Q N Q 12037Q 13N Q
+ + + − − − + + − ζ3 , (B.6)
324 18 1296 27 864 1728 216 16 2
19N Q 13Q2 N Q2
 
1 5N 13Q
+ − − + + − π4
2592 648 2160 1728 2160 288
−1555 285Q 205Q2
 
3
+ + − + 10Q ζ5
18 2 3

– 26 –
 −53117 735733N 1129N 2 697751Q 578341N Q 1319N 2 Q
δ74 = + + + − −
1944 139968 62208 15552 93312 93312
84665Q2 8753N Q2 59N 2 Q2 613Q3 23N Q3 7Q4
− + + + − −
3888 5184 15552 144 144 16
2 2 2 2 3 3
 
−3631 599N N 9773Q 713N Q N Q 3365Q 43N Q 27Q NQ
+ − − + + + − − + + ζ3 , (B.7)
36 216 108 54 216 144 36 36 2 4
13N Q Q2 N Q2
 
41 N 53Q
+ + − − + + π4
1080 540 1080 2160 60 360
495Q 5N Q 255Q2
 
5N 3
+ 140 − − + + − 20Q ζ5
3 2 3 2
8927569 2504627N 11491N 2 N3 7793573Q 428587N Q
δ84 = − + − − +
419904 839808 419904 4374 279936 139968
2 3
9811N Q 19N Q 3695Q 2 365N Q 2 23N Q2 7Q3 17N Q3 21Q4
2
− + + − − − − −
559872 82944 432 2592 3456 24 96 64
2 3 2

3067 1463N 11N N 18293Q 1001N Q 7N Q
+ + − + − − +
72 324 216 1296 216 216 72
. (B.8)
N 3 Q 1855Q2 179N Q2 17N 2 Q2 77Q3 N Q3

− + + − − − ζ3
1728 36 216 432 8 6
N2 29N Q N 2 Q Q2 7N Q2 N 2 Q2
 
41 23N 37Q
+ − + − + − + − + π4
1620 12960 2160 1080 4320 1440 90 2160 4320
5N Q2 65Q3 5N Q3
 
−640 55N 775Q 35N Q 2
+ − + + − 70Q − + + ζ5
9 18 6 12 6 6 12

References

[1] S. Tomonaga, On a relativistically invariant formulation of the quantum theory of wave fields, Prog. Theor.
Phys. 1 (1946) 27–42.
[2] H. A. Bethe, The Electromagnetic shift of energy levels, Phys. Rev. 72 (1947) 339–341.
[3] J. S. Schwinger, On Quantum electrodynamics and the magnetic moment of the electron, Phys. Rev. 73 (1948)
416–417.
[4] F. J. Dyson, The Radiation theories of Tomonaga, Schwinger, and Feynman, Phys. Rev. 75 (1949) 486–502.
[5] D. J. Gross and F. Wilczek, Ultraviolet Behavior of Nonabelian Gauge Theories, Phys. Rev. Lett. 30 (1973)
1343–1346.
[6] H. D. Politzer, Reliable Perturbative Results for Strong Interactions?, Phys. Rev. Lett. 30 (1973) 1346–1349.
[7] Q. Jin and Y. Li, Five-loop anomalous dimensions of ϕQ operators in a scalar theory with O(N) symmetry,
JHEP 10 (2022) 084, [arXiv:2205.02535].
[8] A. Bednyakov and A. Pikelner, Six-loop anomalous dimension of the ϕQ operator in the O(N) symmetric
model, Phys. Rev. D 106 (2022), no. 7 076015, [arXiv:2208.04612].

– 27 –
 [9] E. C. G. Stueckelberg and A. Petermann, The normalization group in quantum theory, Helv. Phys. Acta 24
(1951) 317–319.

[10] E. C. G. Stueckelberg de Breidenbach and A. Petermann, Normalization of constants in the quanta theory,
Helv. Phys. Acta 26 (1953) 499–520.

[11] M. Gell-Mann and F. E. Low, Quantum electrodynamics at small distances, Phys. Rev. 95 (1954) 1300–1312.

[12] N. N. Bogolyubov and D. V. Shirkov, Introduction to the theory of quantized fields, vol. 3. 1959.

[13] C. G. Callan, Jr., Broken scale invariance in scalar field theory, Phys. Rev. D 2 (1970) 1541–1547.

[14] K. Symanzik, Small distance behavior in field theory and power counting, Commun. Math. Phys. 18 (1970)
227–246.

[15] K. G. Wilson, Renormalization group and critical phenomena. i. renormalization group and the kadanoff
scaling picture, Phys. Rev. B 4 (Nov, 1971) 3174–3183.

[16] K. G. Wilson, Renormalization group and critical phenomena. ii. phase-space cell analysis of critical behavior,
Phys. Rev. B 4 (Nov, 1971) 3184–3205.

[17] K. G. Wilson, The renormalization group and critical phenomena, Rev. Mod. Phys. 55 (1983) 583–600.

[18] J. Polchinski, Renormalization and Effective Lagrangians, Nucl. Phys. B 231 (1984) 269–295.

[19] K. Chetyrkin and V. Smirnov, R*-operation corrected, Physics Letters B 144 (1984), no. 5 419–424.

[20] S. Larin and P. van Nieuwenhuizen, The Infrared R* operation, hep-th/0212315.

[21] P. A. Baikov, K. G. Chetyrkin, and J. H. Kühn, Five-Loop Running of the QCD coupling constant, Phys. Rev.
Lett. 118 (2017), no. 8 082002, [arXiv:1606.08659].

[22] M. Misiak and M. Munz, Two loop mixing of dimension five flavor changing operators, Phys. Lett. B 344
(1995) 308–318, [hep-ph/9409454].

[23] K. G. Chetyrkin, M. Misiak, and M. Munz, Beta functions and anomalous dimensions up to three loops, Nucl.
Phys. B 518 (1998) 473–494, [hep-ph/9711266].

[24] J. A. Gracey, T. Luthe, and Y. Schroder, Four loop renormalization of the Gross-Neveu model, Phys. Rev. D
94 (2016), no. 12 125028, [arXiv:1609.05071].

[25] T. Luthe, Fully massive vacuum integrals at 5 loops. PhD thesis, Bielefeld U., 2015.

[26] O. Schnetz, Graphical functions and single-valued multiple polylogarithms, Commun. Num. Theor. Phys. 08
(2014) 589–675, [arXiv:1302.6445].

[27] O. Schnetz, ϕ4 theory at seven loops, Phys. Rev. D 107 (2023), no. 3 036002, [arXiv:2212.03663].

[28] O. Schnetz and S. Theil, Notes on graphical functions with numerator structure, arXiv:2407.17133.

[29] V. A. Smirnov, Applied asymptotic expansions in momenta and masses, Springer Tracts Mod. Phys. 177
(2002) 1–262.

[30] O. Schnetz, Numbers and Functions in Quantum Field Theory, Phys. Rev. D 97 (2018), no. 8 085018,
[arXiv:1606.08598].

– 28 –
[31] O. Schnetz, HyperlogProcedures ver. 0.7 Maple package available on the homepage of the author,
https://www.math.fau.de/person/oliver-schnetz.

[32] K. G. Wilson and M. E. Fisher, Critical exponents in 3.99 dimensions, Phys. Rev. Lett. 28 (1972) 240–243.

[33] M. V. Kompaniets and E. Panzer, Minimally subtracted six loop renormalization of O(n)-symmetric ϕ4 theory
and critical exponents, Phys. Rev. D 96 (2017), no. 3 036016, [arXiv:1705.06483].

[34] A. N. Vasilev, The field theoretic renormalization group in critical behavior theory and stochastic dynamics.
2004.

[35] G. Badel, G. Cuomo, A. Monin, and R. Rattazzi, The Epsilon Expansion Meets Semiclassics, JHEP 11 (2019)
110, [arXiv:1909.01269].

[36] O. Antipin, J. Bersini, F. Sannino, Z.-W. Wang, and C. Zhang, Charging the O(N ) model, Phys. Rev. D 102
(2020), no. 4 045011, [arXiv:2003.13121].

[37] O. Antipin, J. Bersini, F. Sannino, Z.-W. Wang, and C. Zhang, Charging non-Abelian Higgs theories, Phys.
Rev. D 102 (2020), no. 12 125033, [arXiv:2006.10078].

[38] I. Jack and D. R. T. Jones, Anomalous dimensions at large charge in d=4 O(N) theory, Phys. Rev. D 103
(2021), no. 8 085013, [arXiv:2101.09820].

[39] O. Antipin, J. Bersini, F. Sannino, Z.-W. Wang, and C. Zhang, Untangling scaling dimensions of fixed charge
operators in Higgs theories, Phys. Rev. D 103 (2021), no. 12 125024, [arXiv:2102.04390].

[40] I. Jack and D. R. T. Jones, Anomalous dimensions at large charge for U(N)×U(N) theory in three and four
dimensions, Phys. Rev. D 104 (2021), no. 10 105017, [arXiv:2108.11161].

[41] O. Antipin, J. Bersini, and P. Panopoulos, Yukawa interactions at large charge, JHEP 10 (2022) 183,
[arXiv:2208.05839].

[42] O. Antipin, A. Bednyakov, J. Bersini, P. Panopoulos, and A. Pikelner, Gauge Invariance at Large Charge,
Phys. Rev. Lett. 130 (2023), no. 2 021602, [arXiv:2210.10685].

[43] L. Fei, S. Giombi, and I. R. Klebanov, Critical O(N ) models in 6 − ϵ dimensions, Phys. Rev. D 90 (2014),
no. 2 025018, [arXiv:1404.1094].

[44] L. Fei, S. Giombi, I. R. Klebanov, and G. Tarnopolsky, Three loop analysis of the critical O(N) models in 6-ε
dimensions, Phys. Rev. D 91 (2015), no. 4 045011, [arXiv:1411.1099].

[45] J. A. Gracey, Four loop renormalization of ϕ3 theory in six dimensions, Phys. Rev. D 92 (2015), no. 2 025012,
[arXiv:1506.03357].

[46] M. Kompaniets and A. Pikelner, Critical exponents from five-loop scalar theory renormalization near
six-dimensions, Phys. Lett. B 817 (2021) 136331, [arXiv:2101.10018].

[47] M. Borinsky, J. A. Gracey, M. V. Kompaniets, and O. Schnetz, Five-loop renormalization of ϕ3 theory with
applications to the Lee-Yang edge singularity and percolation theory, Phys. Rev. D 103 (2021), no. 11 116024,
[arXiv:2103.16224].

[48] I. Jack and D. R. T. Jones, Scaling dimensions at large charge for cubic ϕ3 theory in six dimensions, Phys.
Rev. D 105 (2022), no. 4 045021, [arXiv:2112.01196].

– 29 –
[49] A. N. Vasiliev, Y. M. Pismak, and Y. R. Khonkonen, Simple Method of Calculating the Critical Indices in the
1/N Expansion, Theor. Math. Phys. 46 (1981) 104–113.
[50] A. N. Vasiliev, Y. M. Pismak, and Y. R. Khonkonen, 1/N Expansion: Calculation of the Exponents η and Nu
in the Order 1/N 2 for Arbitrary Number of Dimensions, Theor. Math. Phys. 47 (1981) 465–475.
[51] A. N. Vasiliev, Y. M. Pismak, and Y. R. Khonkonen, 1/N expansion: calculation of the exponent η in the
order 1/N 3 by the conformal bootstrap method, Theor. Math. Phys. 50 (1982) 127–134.
[52] Z. Wu, J. Boehm, R. Ma, H. Xu, and Y. Zhang, NeatIBP 1.0, a package generating small-size
integration-by-parts relations for Feynman integrals, Comput. Phys. Commun. 295 (2024) 108999,
[arXiv:2305.08783].
[53] X. Liu and Y.-Q. Ma, AMFlow: A Mathematica package for Feynman integrals computation via auxiliary
mass flow, Comput. Phys. Commun. 283 (2023) 108565, [arXiv:2201.11669].
[54] S. E. Derkachov and A. N. Manashov, The Simple scheme for the calculation of the anomalous dimensions of
composite operators in the 1/N expansion, Nucl. Phys. B 522 (1998) 301–320, [hep-th/9710015].

– 30 –