PHYSICAL REVIEW D 103, 046015 (2021)

RG limit cycles and unconventional fixed points in perturbative QFT

Christian B. Jepsen ,1,* Igor R. Klebanov ,2,3 and Fedor K. Popov2
1
Simons Center for Geometry and Physics, Stony Brook University, Stony Brook, New York 11794, USA
2
Department of Physics, Princeton University, Princeton, New Jersey 08544, USA
3
Princeton Center for Theoretical Science, Princeton University, Princeton, New Jersey 08544, USA

(Received 22 December 2020; accepted 27 January 2021; published 23 February 2021)

We study quantum field theories with sextic interactions in 3 − ϵ dimensions, where the scalar fields ϕab
form irreducible representations under the OðNÞ2 or OðNÞ global symmetry group. We calculate the beta
functions up to four-loop order and find the renormalization group (RG) fixed points. In an example of
large N equivalence, the parent OðNÞ2 theory and its antisymmetric projection exhibit identical large N
beta functions that possess real fixed points. However, for projection to the symmetric traceless
representation of OðNÞ, the large N equivalence is violated by the appearance of an additional
double-trace operator not inherited from the parent theory. Among the large N fixed points of this
daughter theory we find complex conformal field theories. The symmetric traceless OðNÞ model also
exhibits very interesting phenomena when it is analytically continued to small noninteger values of N. Here
we find unconventional fixed points, which we call “spooky.” They are located at real values of the
coupling constants gi , but two eigenvalues of the Jacobian matrix ∂βi =∂gj are complex. When these
complex conjugate eigenvalues cross the imaginary axis, a Hopf bifurcation occurs, giving rise to RG limit
cycles. This crossing occurs for N crit ≈ 4.475, and for a small range of N above this value we find RG flows
that lead to limit cycles.

DOI: 10.1103/PhysRevD.103.046015

I. INTRODUCTION AND SUMMARY existence in a simple OðNÞ symmetric model of scalar

The renormalization group (RG) is among the deepest
ideas in modern theoretical physics. There is a variety of
possible RG behaviors, and limit cycles are among the most
exotic and mysterious. Their possibility was mentioned in the
classic review [1] in the context of connections between RG
and dynamical systems (for a recent discussion of these
connections, see [2]). However, there has been relatively little
research on RG limit cycles. They have appeared in quantum
mechanical systems [3–6], in particular, in a description of
the Efimov bound states [7] (for a review, see [8]). The status
of RG limit cycles in quantum field theories (QFT) is less
clear. They have been searched for in unitary four-dimen-
sional QFT [9], but turned out to be impossible [10,11],
essentially due to the constraints imposed by the a theorem
[12–14].1
In this paper we report some progress on RG limit cycles
in the context of perturbative QFT. We demonstrate their
*
cjepsen@scgp.stonybrook.edu
1
See, however, Refs. [15,16], where it is argued that QFTs may
exhibit multivalued c or a functions that do not rule out limit
cycles.
Published by the American Physical Society under the terms of
the Creative Commons Attribution 4.0 International license.
Further distribution of this work must maintain attribution to
the author(s) and the published article’s title, journal citation,
and DOI. Funded by SCOAP3.
fields with sextic interactions in 3 − ϵ dimensions. As
expected, the limit cycles appear when the theory is
continued to a range of parameters where it is nonunitary.
The scalar fields form a symmetric traceless N × N matrix,
and imposition of the OðNÞ symmetry restricts the number
of sextic operators to four. When we consider an analytic
continuation of this model to noninteger real values of
N (a mathematical framework for such a continuation
was presented in [17]), we find a surprise. In the range
4.465 < N < 4.534, as well as in three other small ranges
of N, there are special RG fixed points which we call
“spooky.” These fixed points are located at real values of
the sextic couplings gi , but only two of the eigenvalues of
the Jacobian matrix ∂βi =∂gj are real; the other two are
complex conjugates of each other. This means that a pair
of nearly marginal operators at the spooky fixed points
has complex scaling dimensions.2 At the critical value
N crit ≈ 4.475, the two complex eigenvalues of the Jacobian
become purely imaginary. As a result, for N slightly bigger
2
These special complex dimensions appear in addition to the
complex dimensions of certain evanescent operators that are
typically present in ϵ expansions [18]. The latter dimensions have
large real parts and are easily distinguished from our nearly
marginal operators. Some of the operators with complex dimen-
sions we observe resemble evanescent operators in that they
interpolate to vanishing operators at integer values of N; this is
discussed in Sec. IV.

2470-0010=2021=103(4)=046015(25) 046015-1 Published by the American Physical Society

JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

than N crit , where the real part of the complex eigenvalues
becomes negative, there are RG flows that lead to limit
cycles. In the theory of dynamical systems this phenomenon
is called a Hopf (or Poincarè-Andronov-Hopf) bifurcation
[19]. The possibility of RG limit cycles appearing via a Hopf
bifurcation was generally raised in [2], but no specific
examples were provided. As we demonstrate in Sec. IV,
the symmetric traceless OðNÞ model in 3 − ϵ dimensions
provides a simple perturbative example of this phenomenon.
We show that there is no conflict between the limit cycles
we have found and the F theorem [20–27]. This is because
the analytic continuation to noninteger values of N below 5
violates the unitarity of the symmetric traceless OðNÞ
model, so that the F function is not monotonic. We feel
that the simple perturbative realization of limit cycles we
have found is interesting, and we hope that there are
analogous phenomena in other models and dimensions.
Our paper also sheds new light on the large N behavior
of the matrix models in 3 − ϵ dimensions. Among the
fascinating features of various large N limits (for a recent
brief overview, see [28]) are the “large N equivalences,”
which relate models that are certainly different at finite N.
An incomplete list of the conjectured large N equivalences
includes [29–36]. Some of them appear to be valid, even
nonperturbatively, while others are known to break down
dynamically. For example, in the nonsupersymmetric
orbifolds of the N ¼ 4 supersymmetric Yang-Mills theory
[30–33,37], there are perturbative instabilities in the large
N limit due to the beta functions for certain double-trace
appearance of an additional invariant operator in the latter.
The large N fixed points in this theory occur at a complex
value of the coefficient of this operator. As a result, instead
of the convention conformal field theory (CFT) in the
parent theory, we find a “complex CFT” [42,43] (see also
[44]) in the daughter theory. As discussed above, analytical
continuation of this model to small noninteger N leads to
the appearance of the spooky fixed points and limit cycles.
II. THE BETA FUNCTION MASTER FORMULA
In a general sextic scalar theory with potential
λiklmnp
VðϕÞ ¼ ϕi ϕk ϕl ϕm ϕn ϕp ð1Þ
6!
the beta function receives a two-loop contribution from the
Feynman diagram
In [26,27,45] one can find explicit formulas for the
corresponding two-loop beta function in d ¼ 3 − ϵ dimen-
sions. Equation (6.1) of the latter reference reads
1
βV ðϕÞ ¼ −2ϵVðϕÞ þ V ijk ðϕÞV ijk ðϕÞ; ð2Þ
3ð8πÞ2
∂ ∂
where V ij ðϕÞ ≡ ∂ϕ i    ∂ϕj VðϕÞ. By taking the indices to

couplings having no real zeros [38–41]. stand for doublets of subindices, this formula can be used to
In Sec. III we study the RG flows of three scalar theories compute the beta functions of matrix tensor models. In order
in 3 − ϵ dimensions with sextic interactions: the parent to apply the formula to models of symmetric or antisym-
OðNÞ2 symmetric model of N × N matrices ϕab , and its metric matrices, however, we need to slightly modify it.
two daughter theories which have OðNÞ symmetry. For Letting i and j stand for doublets of indices, we define the
each model, we list all sextic operators marginal in three object Cij via the momentum space propagator:
dimensions, compute the associated beta functions up to
four loops, and determine the fixed points. One of our Cij
hϕ̃i ðkÞϕ̃j ð−kÞi0 ¼ : ð3Þ
motivations for this study is to investigate the large N k2
orbifold equivalence and its violation in the simple context
With this definition in hand, Eq. (2) straightforwardly
of purely scalar theories. We observe evidence of large N
generalizes to
equivalence between the parent OðNÞ2 theory and the
0 0 0
daughter OðNÞ theory of antisymmetric matrices: both Cii Cjj Ckk
theories have three invariant operators, and the large N βV ðϕÞ ¼ −2ϵVðϕÞ þ V ijk ðϕÞV i0 j0 k0 ðϕÞ: ð4Þ
3ð8πÞ2
beta functions are identical. However, the large N equiv-
alence of the parent theory with the daughter OðNÞ At four-loops the following four kinds of Feynman diagrams
theory of symmetric traceless matrices is violated by the contribute to the beta function:

046015-2
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)
Z 
The resulting four-loop beta function can be read off from 1 1
S ¼ d x ð∂ μ ϕab Þ2 þ ðg1 O1 ðxÞ
3−ϵ
Eq. (6.2) of [45], 2 6!



ð4Þ 1 1 4 þ g2 O2 ðxÞ þ g3 O3 ðxÞÞ ; ð9Þ
βV ¼ 4
V ij V iklmn V jklmn − V ijk V ilmn V jklmn
ð8πÞ 6 3

π2 where the dynamical degrees of freedom are scalar matrices
− V ijkl V klmn þ þ ϕi γ ϕij V j ; ð5Þ
12 ϕab which transform under the action of a global OðNÞ ×
OðNÞ symmetry. The three operators in the potential are
where the anomalous dimension γ ϕij is given by
O1 ¼ ϕa1 b1 ϕa2 b1 ϕa2 b2 ϕa3 b2 ϕa3 b3 ϕa1 b3 ¼ tr½ϕϕT 3 ;
1 O2 ¼ ϕab ϕab ϕa1 b1 ϕa2 b1 ϕa2 b2 ϕa1 b2 ¼ tr½ϕϕT tr½ϕϕT 2 ;
γ ϕij ¼ λiklmnp λjklmnp : ð6Þ
90ð8πÞ4 O3 ¼ ðϕab ϕab Þ3 ¼ ðtr½ϕϕT Þ3 : ð10Þ
The above two equations also admit of straightforward They make up all sextic operators that are invariant under
generalizations by contracting indices through the Cij the global symmetry. Later we will also study projections
matrix. of the parent theory that have only a global OðNÞ symmetry
Before proceeding to matrix models, we can review the that rotates first and second indices at the same time. In
beta function obtained by the above formulas in the case of such models it becomes possible to construct singlets via
a sextic OðMÞ vector model described by the action contractions between first and second indices, and therefore
Z
 there is an additional sextic scalar:
3−ϵ 1 j 2 g i i 3
S¼ d x ð∂ μ ϕ Þ þ ðϕ ϕ Þ ; ð7Þ
2 6! O4 ¼ ðϕa1 a2 ϕa2 a3 ϕa3 a1 Þ2 ¼ ðtr½ϕ3 Þ2 : ð11Þ

where the field ϕi is an M-component vector. The four-loop The sextic operators are depicted diagrammatically in
beta function of this vector model is given by [45,46] Fig. 1. We could also introduce an operator containing
tr½ϕ, but since the orbifolds we will study are models of
192ð3M þ 22Þ 2 symmetric traceless and antisymmetric matrices, the trace
βg ¼ −2ϵg þ g is identically zero. In the antisymmetric model, the operator
6!ð8πÞ2
O4 vanishes, but it is nonvanishing in the symmetric
1
− ð9216ð53M 2 þ 858M þ 3304Þ orbifold, and so in this case we will introduce this addi-
ð6!Þ2 ð8πÞ4 tional marginal operator to the Lagrangian and take the
þ 1152π 2 ðM3 þ 34M 2 þ 620M þ 2720ÞÞg3 : ð8Þ potential to be given by

This equation provides a means of checking the beta
functions of the matrix models, which reduce to the vector
model when all couplings are set to zero except for the
coupling, denoted g3 below, associated with the triple trace
operator.
III. SEXTIC MATRIX MODELS
We now turn to matrix models in d ¼ 3 − ϵ dimensions.
The parent theory we consider has the Lagrangian given by
1
VðxÞ ¼ ðg O ðxÞ þ g2 O2 ðxÞ þ g3 O3 ðxÞ þ g4 O4 ðxÞÞ:
6! 1 1
ð12Þ
To study the large N behavior of these matrix models, we
introduce rescaled coupling constants λ1 , λ2 , λ3 , λ4 . To
simplify expressions, it will be convenient to also rescale
the coupling constants by a numerical prefactor. We
therefore define the rescaled couplings by

FIG. 1. The sextic operators in matrix models. The double trace operator O4 exists only in the theory of symmetric matrices.

046015-3
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

λ1 λ2
g1 ¼ 6!ð8πÞ2 ; g2 ¼ 6!ð8πÞ2 ; Each eigenvalue mi corresponds to a nearly marginal
N2 N3 operator with scaling dimension
λ λ
g3 ¼ 6!ð8πÞ2 34 ; g4 ¼ 6!ð8πÞ2 43 : ð13Þ
N N Δi ¼ d þ mi ¼ 3 − ϵ þ mi : ð18Þ
To justify
pffiffiffiffithese powers of N, let us perform a scaling Thus, negative eigenvalues correspond to slightly relevant
ϕab → N ϕab . Then the coefficient of each q-trace term in operators, which cause an instability of the fixed point.
the action scales as N 2−q . This is the standard scaling in the The only unstable direction, corresponding to eigenvalue
’t Hooft limit, which ensures that each term in the action is −2ϵ þ 32 2
9 ϵ , is
of order N 2.



245 4225π 2 − 4188 67π 2 − 28
A. The OðNÞ parent theory
2 þ ϵ λ1 þ 10 þ λ2 þ λ3 :
3 36 6
For the matrix model parent theory, the momentum space
propagator is given by ð19Þ

a0 b0 δaa δbb
0 0
The above comments relate to the OðNÞ2 matrix
hϕ̃ ðkÞϕ̃
ab
ð−kÞi0 ¼ : ð14Þ model at N ¼ ∞. We can also study the model at finite
k2
N. One interesting quantity is N min , the smallest value
Computing the four-loop beta functions and taking the of N at which the fixed point that interpolates to the
large N limit with scalings (13), we find that, up to OðN1 Þ large N solution (16) appears as a solution to the beta
corrections, functions. This fixed point emerges along with another
fixed point, and right at N min these solutions to the beta
βλ1 ¼ − 2λ1 ϵ þ 72λ21 − 288ð17 þ π 2 Þλ31 ; ∂βi
functions are identical, so that the matrix ð∂g Þ is
βλ2 ¼ − 2λ2 ϵ þ 432λ21 þ 96λ1 λ2 − 864ð90 þ 7π 2 Þλ31
j
degenerate. So we arrive at the following system of
− 864ð10 þ π 2 Þλ21 λ2 ; equations:
βλ3 ¼ − 2λ3 ϵ þ 168λ21 þ 192λ1 λ2 þ 32λ22

∂βi
βi ðλi ; NÞ ¼ 0; det ðλi ; NÞ ¼ 0: ð20Þ
− 432ð210 þ 23π 2 Þλ31 − 1152ð39 þ 4π 2 Þλ21 λ2 ∂λj
128 2 3
þ 4608λ21 λ3 − 768ð6 þ π 2 Þλ1 λ22 − π λ2 : ð15Þ This system of equations can easily be solved numeri-
3
cally to zeroth order in ϵ, and with a zeroth order
These beta functions have two nontrivial fixed points, which solution in hand the first order solution can be obtained
are both real. But one of these fixed points, which comes from by linearizing the system of equations. We find that
balancing the two-loop and four-loop contributions, is not N min ¼ 23.2541 − 577.350ϵ, which nicely fits the results
perturbatively reliable in an ϵ expansion around ϵ ¼ 0 of a numerical study where we compute N min at
because all the couplings at this fixed point contain terms different values of ϵ:
of order Oðϵ0 Þ. The other fixed point is given by
ϵ 0 0.001 0.002 0.003 0.004 0.005
ϵ 17 þ π 2 2 ϵ 22 þ 7π 2 2 N min 23.255 22.682 22.124 21.576 21.039 20.511
λ1 ¼ þ ϵ; λ2 ¼ − − ϵ;
36 324 2 36
295 4714 þ 6301π 2 2 These values result in a numerical fit N min ðϵÞ ¼
λ3 ¼ ϵþ ϵ: ð16Þ 23.255 − 553.7ϵ, which coincides with the result stated
108 1944 above.
At this fixed point the matrix
∂βλi
has eigenvalues If we take N to be finite and ϵ ≪ N12 , we can provide
∂λj
some more details about the number and stability of fixed
 
32 2 2ϵ 44 þ 10π 2 2 34 þ 2π 2 2 points for different values of N. For N > 23.2541 −
−2ϵ þ ϵ ; − ϵ ; 2ϵ − ϵ : 577.350ϵ there are three nontrivial, real, perturbatively
9 3 27 9
accessible fixed points, which in the large N limit, to
ð17Þ leading order in ϵ, scale with N as

046015-4
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

6!ð8πÞ2 ϵ
g1 ¼ g2 ¼ 0; g3 ¼ ;
288 N 2
6!ð8πÞ2 ϵ 10 ϵ 6!ð8πÞ2 ϵ
g1 ¼ ; g2 ¼ − · 6!ð8πÞ2 3 ; g3 ¼ ;
36 N 2 36 N 288 N 2
6!ð8πÞ2 ϵ 1 ϵ 295 ϵ
g1 ¼ ; g2 ¼ − · 6!ð8πÞ2 3 ; g3 ¼ · 6!ð8πÞ2 4 : ð21Þ
36 N 2 2 N 108 N

The first of these three fixed points is identical to the vector
model fixed point; that is to say, the symmetry is enhanced
from OðNÞ2 to OðN 2 Þ. This fixed point extends to all N in
the small ϵ regime we are considering:
6!ð8πÞ2
g1 ¼ g2 ¼ 0; g3 ¼ ϵ: ð22Þ
96ð22 þ 3N 2 Þ
The third fixed point in (21) extends to the regime where
N 2 > 1ϵ and becomes the large N solution discussed above.
This fixed point merges with the second fixed point in (21)
at a critical point situated at NðϵÞ ¼ 23.2541 − 577.350ϵ
And so at intermediate values of N, only the vector
model fixed point exists. But as we keep decreasing N
we encounter another critical point at NðϵÞ ¼ 5.01072 þ
14.4537ϵ, from which two new solutions to the vanishing
beta functions emerge. As N further decreases past the
value NðϵÞ ¼ 2.75605 − 0.0161858ϵ, another pair of fixed
points appears, but then at NðϵÞ ¼ 2.72717 − 0.757475ϵ
two of the fixed points merge and become complex. Then at
NðϵÞ ¼ 2.33265 − 0.316279ϵ two new fixed points appear,
but these disappear again at NðϵÞ ¼ 0.827007 þ 8.10374ϵ,
so that for N below this value there are a total of three real
nontrivial fixed points. The behavior of the various fixed
points as a function of N is summarized in more detail in
Figs. 2 and 3.
B. The OðNÞ model of antisymmetric matrices
For the theory of antisymmetric matrices ϕT ¼ −ϕ the
momentum space propagator is given by
0 0 1 aa0 bb0 0 0
hϕ̃ab ðkÞϕ̃a b ð−kÞi0 ¼ ðδ δ − δab δba Þ: ð23Þ
2k2
Performing the large N expansion using the scalings (13)
we get the large N beta functions

βλ1 ¼ − 2λ1 ϵ þ 18λ21 − 18ð17 þ π 2 Þλ31 ;

βλ2 ¼ − 2λ2 ϵ þ 108λ21 þ 24λ1 λ2 − 54ð90 þ 7π 2 Þλ31 − 54ð10 þ π 2 Þλ21 λ2 ;
βλ3 ¼ − 2λ3 ϵ þ 42λ21 þ 48λ1 λ2 þ 8λ22 − 27ð210 þ 23π 2 Þλ31 − 72ð39 þ 4π 2 Þλ21 λ2
8
þ 288λ21 λ3 − 48ð6 þ π 2 Þλ1 λ22 − π 2 λ32 : ð24Þ
3
These beta functions are equivalent to (15) up to a redefinition of the rescaled couplings by a factor of 4, which is
compatible with this daughter theory being equivalent in the large N limit to the parent theory studied in the previous
section.
We can also study the behavior of this model for finite N and ϵ ≪ 1. For N > 35.3546 − 673.428ϵ there are three (real,
perturbatively accessible) fixed points, which in the large N limit (keeping ϵ ≪ N12 ) to leading order in the ϵ scale with N as

6!ð8πÞ2 ϵ
g1 ¼ g2 ¼ 0; g3 ¼ ;
144 N 2
6!ð8πÞ2 ϵ 10 ϵ 6!ð8πÞ2 ϵ
g1 ¼ ; g2 ¼ − · 6!ð8πÞ2 3 ; g3 ¼ ;
9 N2 9 N 144 N 2
6!ð8πÞ2 ϵ ϵ 295 ϵ
g1 ¼ ; g2 ¼ −2 · 6!ð8πÞ2 3 ; g3 ¼ · 6!ð8πÞ2 4 : ð25Þ
9 N2 N 27 N

The first of these three fixed points is the vector model fixed point, and it is present more generally in the small ϵ regime we
are considering:

046015-5
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

FIG. 2. The real perturbative fixed points of the OðNÞ2 matrix model parent theory, the intersection point (marked in brown), and the
critical points at which they merge and disappear (marked in black) as a function of N for small ϵ. Fixed points that are IR unstable in all
three directions are drawn in red, those unstable in two directions are drawn in violet, those unstable in one direction are drawn in blue,
and those that are stable in all three directions are drawn in green. The four-loop corrections to the third point on the list, where two fixed
lines intersect, are undetermined for any Oðϵ2 Þ value of λ2 .

6!ð8πÞ2 a critical point situated at NðϵÞ ¼ 35.3546 − 673.428ϵ. And

g1 ¼ g2 ¼ 0; g3 ¼ ϵ: ð26Þ
48ð44 − 3N þ 3N 2 Þ so at intermediate values of N, only the vector model fixed
point exists. But as we keep decreasing N we encounter
The third fixed point in (25) extends to the regime where another critical point at NðϵÞ ¼ 6.02669 þ 7.37013ϵ,
N 2 > 1ϵ and becomes the large N solution discussed above. from which two new solutions to the vanishing beta func-
This fixed point merges with the second fixed point in (25) at tions emerge. As N further decreases past the value

046015-6
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

N
N

2 2

0 g1 g2
200 400 600 800 − 1000 − 500

−2 −2

−4 −4

0 g3
100 200 300 400 500

−2

−4

FIG. 3. The locations of the real perturbative fixed points of the OðNÞ2 matrix model in the space of coupling constants as a function of
N for small ϵ. The colors indicate the number of stable directions associated with a given fixed point as in Fig. 2.

NðϵÞ ¼ 5.70601 þ 0.540694ϵ, another pair of fixed points
appears, and past NðϵÞ ¼ 5.075310 − 0.0278896ϵ yet another
pair of fixed points appears (in this range of N, all seven
nontrivial solutions to the vanishing beta functions are real).
But already below NðϵÞ ¼ 5.03275 − 0.586724ϵ, two of the
fixed points become complex, and below NðϵÞ ¼ 3.08122 þ
8.26176ϵ two more fixed points become complex, so that for N
below this value there are a total of three real nontrivial fixed
points. The behavior of the various fixed points as a function of
N is summarized in more detail in Figs. 4 and 5.
C. Symmetric traceless matrices and violation
of large N equivalence
There is a projection of the parent theory of general real
matrices ϕab which restricts them to symmetric matrices
ϕ ¼ ϕT . In order to have an irreducible representation of
OðNÞ we should also require them to be traceless tr ϕ ¼ 0.
Then the propagator is given by


a0 b0 1 0 bb0 0 ba0 2 ab a0 b0
hϕ̃ ðkÞϕ̃ ð−kÞi0 ¼ 2 δ δ þ δ δ − δ δ
ab aa ab : constants containing Oðϵ0 Þ terms. Another pair of fixed
2k N
ð27Þ
The operators O1;2;3;4 are actually independent for N > 5,
while for N ¼ 2, 3, 4, 5 there are linear relations between them:
(i) N ¼ 2∶O4 ¼ 0, O3 ¼ 2O2 ¼ 4O1 ,
(ii) N ¼ 3∶O3 ¼ 2O2 , 2O4 ¼ 3O3 þ 6O1 ,
(iii) N ¼ 4; 5∶18O2 þ 8O4 ¼ 24O1 þ 3O3 .
We will see that the existence of these relations for small integer
values of N has interesting implications for the analytic
continuation of the theory from N > 5 to N < 5.
Let us first discuss the large N theory. For the rescaled
couplings λ1 , λ2 , and λ3 , the large N beta functions are the
same as (24) for the antisymmetric model. But now there is
an additional coupling constant, whose large N beta
function is given by
βλ4 ¼ −2ϵλ4 þ 72λ21 þ 36λ1 λ4 þ 6λ24 − 738λ21 λ4
− 18ð180 þ 11π 2 Þλ31 : ð28Þ
Consequently, the RG flow now has five nontrivial fixed
points, two of which are real fixed points but with coupling
points is given by

ϵ 17 þ π 2 2 22 þ 7π 2 2 295 4714 þ 6301π 2 2

λ1 ¼ þ ϵ; λ2 ¼ −2ϵ − ϵ; λ3 ¼ ϵþ ϵ;
9 81 9 27 486
pffiffiffiffiffi pffiffiffiffiffi
−3  i 39 273 − 78π 2  i 39ð67 þ 12π 2 Þ 2
λ4 ¼ ϵþ ϵ: ð29Þ
18 2106

046015-7
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

FIG. 4. The real perturbative fixed points of the antisymmetric matrix model, their intersection point (marked in brown), and the
critical points at which they merge and disappear (marked in black) as a function of N for small ϵ. Fixed points that are IR unstable in all
three directions are drawn in red, those unstable in two directions are drawn in violet, those unstable in one direction are drawn in blue,
and those that are stable in all three directions are drawn in green.

The first three coupling constants assume the same value as for the antisymmetric model, a rescaled version of (16) of the
parent theory, but the additional coupling constant assumes a complex value, thus breaking large N equivalence and
suggesting that the fixed point is unstable and described by a complex CFT [42,43].
∂β
We find that the eigenvalues of ∂λλji at this complex fixed point are

 rffiffiffiffiffi 
32 2 13 67 þ 12π 2 2 2 22 þ 5π 2 2 17 þ π 2 2
−2ϵ þ ϵ ; ∓ 2i ϵ  2i pffiffiffiffiffi ϵ ; ϵ − 2 ϵ ; 2ϵ − 2 ϵ ; ð30Þ
9 3 9 39 3 27 9

046015-8
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

N
N

4
4
2
2

g1 g2
0 −1000 1000
500 1000 1500 2000
−2 −2

−4 −4

0 g3
50 100 150 200 250 300
−2

−4

FIG. 5. The locations of the real perturbative fixed points of the antisymmetric matrix model in the space of coupling constants as a
function of N for small ϵ. The colors indicate the number of stable directions associated with a given fixed point as in Fig. 4.

where the imaginary eigenvalue is associated with a (AdS) description they correspond to fields violating the
complex linear combination of λ1 and λ4 . Thus, there is Breitenlohner-Freedman stability bound.
actually a pair of complex large N fixed points: at one of Let us also note that the symmetric orbifold has a fixed
them there is an operator of complex dimension point where only the twisted sector coupling is nonvanishing:
d þ iA ¼ 3 − ϵ þ iA, qwhile
ffiffiffiffi at the other it has dimension
2 ϵ
3
pffiffiffiffi ϵ2 . Thus, this pair of
d − iA, where A ¼ 2 3 ϵ − 2 67þ12π
13 λ1;2;3 ¼ 0; λ4 ¼ : ð33Þ
9 39 3
complex fixed points satisfies the criteria to be identified as
complex CFTs [42,43]. In our large N theory, the scaling It could be connected to the fact that in the large N limit of the
dimensions d  iA correspond to the double-trace operator parent theory the O4 could not contribute to the beta
O4 , so that the single-trace operator trϕ3 should have functions of the other operators and therefore we can safely
scaling dimension 12 ðd  iAÞ. Indeed, we find that its two- set λ1;2;3 ¼ 0 without setting λ4 ≠ 0.
loop anomalous dimension is, for large N, We can also study the behavior of this model for finite N
rffiffiffiffiffi and ϵ ≪ 1. For N > 13.1802 − 57.5808ϵ there are three
13 (real, perturbatively accessible) fixed points, which in the
γ tr ϕ3 ¼ 6ð3λ1 þ λ2 Þ ¼ ϵ  i ϵ: ð31Þ
3 large N limit (keeping ϵ ≪ N12 ) to leading order in ϵ scale with
Therefore, N as

 rffiffiffiffiffi
d 3−ϵ 13 d  iA 6!ð8πÞ2
Δtr ϕ3 ¼ 3 − 1 þ γ tr ϕ3 ¼ i ϵ¼ : 0 ¼ g1 ¼ g2 ¼ g4 ; g3 ¼ ϵ;
2 2 3 2 144N 2
ð32Þ 6!ð8πÞ2 6!ð8πÞ2
g1 ¼ 144 6
ϵ; g2 ¼ 66 ϵ;
N N5
Scaling dimensions of this form are ubiquitous in large N
complex CFTs [41,44,47,48]. In the dual anti–de Sitter 6!ð8πÞ2 6!ð8πÞ2
g3 ¼ 2
ϵ g4 ¼ ϵ;
144N 3N 3
3
As N is reduced, the two complex conjugate fixed points 6!ð8πÞ2 6!ð8πÞ2
g1 ¼ −144 6
ϵ; g2 ¼ 18 ϵ;
persist down to arbitrarily small N. For finite N, however, the N N5
complex scaling dimensions are no longer of the form d  iA: the
6!ð8πÞ2 6!ð8πÞ2
real part deviates from d, which is consistent with the behavior of g3 ¼ −18 ϵ; g 4 ¼ ϵ: ð34Þ
general complex CFTs [42,43]. N6 3N 3

046015-9
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

FIG. 6. The perturbative real fixed points of the symmetric matrix model, the intersection points (marked in brown), and the critical
points at which they merge and disappear (marked in black) as a function of N for small ϵ. Fixed points that are IR unstable in all four
directions are drawn in red, those unstable in three directions are drawn in violet, those unstable in two directions are drawn in blue,
those unstable in one direction are drawn in cyan, and those that are stable in all four directions are drawn in green. The orange dotted
∂βi
lines denote the segments of spooky fixed points, where two eigenvalues of ∂g j
are complex, and at the orange vertex those eigenvalues
are purely imaginary.

046015-10
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

N
N

g1 g2
500 1000 −2000 −1000 1000

−5 −5

−10 −10

N
N

g4
g3 − 1000 1000
200 400 600 800 1000

−5
−5

−10 − 10

FIG. 7. The locations of the real perturbative fixed points of the symmetric matrix model in the space of coupling constants as a
function of N for small ϵ. The colors indicate the number of stable directions associated with a given fixed point as in Fig. 6, with orange
∂βi
signifying that ∂g j
has complex eigenvalues.

The first of these three fixed points is the vector model fixed
point, which is present generally N in the small ϵ regime:
6!ð8πÞ2
0 ¼ g1 ¼ g2 ¼ g4 ; g3 ¼ ϵ: ð35Þ
48ð38 þ 3N þ 3N 2 Þ
The third fixed point in (34) connects to the large N solution
discussed above. This fixed point merges with the second
fixed point in (34) at a critical point situated at NðϵÞ ¼
13.1802 − 57.5808ϵ And so at intermediate values of N,
only the vector model fixed point exists. But as we keep
decreasing N we encounter another critical point at NðϵÞ ¼
5.41410 þ 13.7204ϵ whence two new fixed points emerge.
As we continue to lower N, new fixed points appear and
disappear as summarized in detail in Figs. 6 and 7.
IV. SPOOKY FIXED POINTS
AND LIMIT CYCLES
As indicated in Fig. 6, in the OðNÞ symmetric traceless
model there exist four segments of real, but spooky, fixed
points as a function of N.4 For these fixed points the
∂βi
Jacobian matrix ð∂g j
Þ has, in addition to one negative and
one positive eigenvalue, a pair of complex conjugate
eigenvalues. Therefore, there are two complex scaling
dimensions (18) at these spooky fixed points, so that they
correspond to nonunitary CFTs. The eigenvectors
4
If we allow negative N, there is a fifth segment of spooky
fixed points at N ∈ ð−3.148; −3.183Þ.

046015-11
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

Im( )
corresponding to the complex eigenvalues have zero norm
(a derivation of this fact is given later in this section). Let us 0.15
note that, in the OðNÞ2 model and the OðNÞ model with an
0.10
antisymmetric matrix, there are no real fixed points with
complex eigenvalues. The symmetric traceless model 0.05
provides a simple setting where they occur. In this section
Re( )
we take a close look at the spooky fixed points and show −0.25 −0.20 −0.15 −0.10 −0.05 0.05 0.10
that they lead to a Hopf bifurcation and RG limit cycles. −0.05
Of the four segments of spooky fixed points with
−0.10
positive N, three, namely those that fall within the ranges
given by N ∈ ð1.094; 2.441Þ, N ∈ ð1.041; 1.175Þ, and −0.15
N ∈ ð0.160; 0.253Þ, share the property that the complex
eigenvalues never become purely imaginary. The number
of stable and unstable directions therefore remain the same FIG. 8. The trajectories of the complex eigenvalues of the
∂βi
Jacobian matrix ð∂g Þ as N is varied from N lower to N upper .
within these intervals. Something special happens, how- j

ever, at the integer value N ¼ 2 that lies within the first

interval. Here the two operators with complex dimensions The Jacobian matrix evaluated at this fixed point is
are given by linear combinations of operators Oi that vanish

by virtue of the linear relations between these operators at ∂βi
N ¼ 2.5 As a result, for N ¼ 2 there are no nearly marginal ∂gj
operators with complex dimensions, as expected. 0 1
The fourth segment of spooky fixed points stands out in −1.65273 −1.58311 1.33984 −1.19641
B 1.0242 0.358518 −3.24194 1.21102 C
that it includes a fixed point with imaginary eigenvalues. B C
This fourth segment lies in the range N ∈ ðN lower ; N upper Þ, ¼B Cϵ
@ 0.128059 0.749009 2.9199 −0.210872 A
where, at the four-loop level,
−0.0618889 0.428409 −0.417582 −1.20064
N upper ≈ 4.5339959143 þ 1.54247ϵ; ð39Þ

N lower ≈ 4.4654144982 þ 0.693698ϵ:
∂βi
As N approaches N upper from above, ð∂g
j
ð36Þ with eigenvalues f2; −1.57495; −0.153965i; 0.153965igϵ.
These quantities are subject to further perturbative correc-
tions in powers of ϵ, for example, after including the four-
Þ has one positive
loop corrections N crit ≈ 4.47507431683 þ 3.12476ϵ. The

and three negative eigenvalues, and two of the negative

eigenvalues converge on the same value. As N dips below
N upper , the two erstwhile identical eigenvalues become
complex and form a pair of complex conjugate values.
As we continue to decrease N, the complex conjugate
eigenvalues traverse mirrored trajectories in the complex
plane until they meet at the same positive value for N equal
to N lower . These trajectories are depicted in Fig. 8. For a
critical value N ¼ N crit with N lower < N < N upper , the
trajectories intersect the imaginary axis such that the two
eigenvalues are purely imaginary. At the two-loop order we
find that
N crit ≈ 4.47507431683;
and the fixed point is located at
g1 ¼ 158.684ϵ; g2 ¼ −211.383ϵ;
g3 ¼ 138.686ϵ; g4 ¼ −49.4564ϵ:
5
This is similar to what happens to evanescent operators when
they are continued to an integer dimension.
existence of a special spooky fixed point with imaginary
eigenvalues is robust under loop corrections that are sup-
pressed by a small expansion parameter, since small
perturbations of the trajectories still result in curves that
intersect the imaginary axis. In light of the negative value of
g4 , one may worry that the potential is unbounded from
below at the spooky fixed points. It is not clear how to
resolve this question for noninteger N, but at the fixed
points at N ¼ 4 and N ¼ 5 that this spooky fixed point
interpolates between, one can explicitly check that the
potential is bounded from below.
The appearance of complex eigenvalues changes the
behavior of the RG flow around the spooky fixed point.
ð37Þ Since the fixed point has one negative eigenvalue for all
N ∈ ðN lower ; N upper Þ, there is an unstable direction in the
space of coupling constants that renders the fixed point IR
unstable. But we can ask the following question: How do
the coupling constants flow in the two-dimensional mani-
fold that is invariant under the RG flow and that is tangent
ð38Þ
to the plane spanned by the eigenvectors of the Jacobian
matrix with complex eigenvalues?
If the real parts of these eigenvalues are nonzero, the
spooky fixed point is a focus and the flow around it is

046015-12
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS …
described by spirals steadily moving inwards or outwards
from the fixed point. For N > N crit, the real parts are
negative and the fixed point is IR unstable, while for N <
N crit the real parts are positive and the fixed point is stable.
By the Hartman-Grobman theorem [49,50], one can locally
change coordinates (redefine the coupling constants) such
that the beta functions near the fixed points are linear.
Furthermore, one can get rid of the imaginary part of the
eigenvalues in this subspace by a suitable field redefini-
tion.6 An analogous statement was given in [10].
When N ¼ N crit , the real parts of the complex eigen-
values are equal to zero. In this case the equilibrium point is
a center, the Hartman-Grobman theorem is not applicable,
and the behavior near the fixed point is controlled by the
higher nonlinear terms in the autonomous equations. If we
consider N as a parameter of the RG flow, N ¼ N crit
corresponds to a bifurcation point, as first introduced by
Poincarè. A standard method of analyzing bifurcations is to
reduce the full system to a set of lower dimensional systems
by the use of the center manifold theorem [51]. Denoting by
λ the eigenvalues of the Jacobian matrix at a given fixed
point, this theorem guarantees the existence of invariant
manifolds tangent to the eigenspaces with Re λ > 0,
Re λ < 0, and Re λ ¼ 0, respectively. The latter manifold
is known as the center manifold, and in general it need be
neither unique nor smooth. But when, as in our case, the
center at g is part of a line of fixed points in the space
ðg; NÞ that vary smoothly with a parameter N, and the
complex eigenvalues satisfy
d
κ¼ Re½λðN crit Þ ≠ 0; ð40Þ
dN
then there exists a unique three-dimensional center mani-
fold in ð⃗g; NÞ passing through ðg ; N crit Þ. On planes of
constant N in this manifold, there exist coordinates ðx; yÞ
such that the third order Taylor expansion can be written in
the form
dx
¼ ðκN þ aðx2 þ y2 ÞÞx − ðω þ cN þ bðx2 þ y2 ÞÞy;
dt
dy
¼ ðω þ cN þ bðx2 þ y2 ÞÞx þ ðκN þ aðx2 þ y2 ÞÞy;
dt
ð41Þ
ðN − 5ÞðN − 4ÞðN − 3Þ2 ðN − 2Þ3 N 2 ðN þ 1Þ3 ðN þ 3ÞðN þ 4Þ3 ðN þ 6Þ2 ðN þ 8ÞðN þ 10Þ
2654208
6
For instance, in two dimensions with z ¼ x þ iy, the equation z_ ¼ ð−α þ iωÞz can via a change of variable z → zei α log jzj be reduced
to z_ ¼ −αz.
7
In [26,27] the terminology a function was used, but we prefer to call it F function instead, since a typically refers to a Weyl anomaly
coefficient in d ¼ 4.
046015-13
PHYS. REV. D 103, 046015 (2021)
where t ¼ ln μ. The constant a in these equations is known as
the Hopf constant. By a theorem due to Hopf [19], there exists
an IR-attractive limit cycle in the center manifold if a > 0,
while if a < 0 there exists an IR-repulsive limit cycle. In the
Appendix C, we present an explicit calculation of a for the
critical point in the symmetric matrix model, and we find that
a is positive. Hence, we conclude that on analytically
continuing in N, the RG flow of this QFT contains a periodic
orbit in the space of coupling constants, an orbit that is
unstable but which in the center manifold constitutes an
attractive limit cycle. This conclusion holds true at all orders
in perturbation theory, since the criteria of Hopf’s theorem,
being topological in nature, are not invalidated by small
perturbative corrections. Figure 9 depicts a numerical plot of
RG trajectories approaching the limit cycle.
Now that we have demonstrated the existence of
limit cycles, we should ask about their consistency with
the known RG monotonicity theorems. In particular, in
three dimensions the F theorem has been conjectured and
established [20,21,23]. Furthermore, in perturbative three-
dimensional QFT, one can make a stronger statement that
the RG flow is a gradient flow, i.e.,
∂F
Gij βj ¼ ; ð42Þ
∂gi
where F and the metric Gij are functions of the coupling
constants that can be calculated perturbatively [22,24–27].7
At leading order, Gij may be read off from the two-point
functions of the nearly marginal operators [24,25]:
Gij
hOi ðxÞOj ðyÞi ¼ : ð43Þ
jx − yj6
The F function satisfies the RG equation
∂ ∂
μ F ¼ F ¼ βi βj Gij : ð44Þ
∂μ ∂t
This shows that, if the metric is positive definite, then F
decreases monotonically as the theory flows toward the IR.
These perturbative statements continue to be applicable in
3 − ϵ dimensions.
At leading order, the metric Gij is exhibited in
Appendix B. Its determinant is given by
: ð45Þ
ω
JEPSEN, KLEBANOV, and POPOV
t4
0.0003
0.0002
0.0001
t3
−0.0004 −0.0003 −0.0002 −0.0001 0.0001
−0.0001
−0.0002
FIG. 9. The RG flow in the invariant manifold tangent to the
plane spanned by the eigenvectors with complex eigenvalues in
the space of coupling constants for N ¼ 4.476. In the IR, the blue
curve whirls inwards toward a limit cycle marked in black, while
the orange curve whirls outwards toward the limit cycle. The
coordinates t3 and t4 are given by linear combinations of
the couplings g1 , g2 , g3 , and g4 and are defined in Appendix C.
The RG flow on the invariant manifold admits of a description in
an infinite expansion in powers of t3 and t4 . This plot is drawn
retaining terms up to cubic order.
This shows that the metric has three zero eigenvalues for
N ¼ 2, two zero eigenvalues for N ¼ 3, and one zero
eigenvalue for N ¼ 4 and 5. This is due to the linear
relations between operators Oi at these integer values of N.
For example, for N ¼ 2 there is only one independent
operator. In the range 4 < N < 5, det Gij < 0, the metric
has one negative and three positive eigenvalues. This is
what explains the possibility of RG limit cycles in the range
N lower < N < N upper . For N > 5, Gij is positive definite,
and for N < −10, Gij is negative definite. This is consistent
with our observing spooky fixed points only outside of
these regimes.8
In general, the norms of vectors computed with this
metric are not positive definite for N < 5. In particular, we
can show that the eigenvectors corresponding to complex
eigenvalues of the Jacobian matrix evaluated at real fixed
8 2
We have also found the metric for the parent OðNÞ theory.
In this case it is positive definite for all N except N ∈ f−4; −2;
1; 2g, where there are zero eigenvalues. We further found the
metric for the antisymmetric matrix model. In certain intervals
within the range N ∈ ð−4; 5Þ it has both positive and negative
eigenvalues, but numerical searches reveal no spooky fixed points
in these intervals.
046015-14
PHYS. REV. D 103, 046015 (2021)
5
t
50 100 150 200
−5
− 10
FIG. 10. The plot of 1012 ðFðtÞ − F0 Þ=ϵ3 , where F0 is the value
at the spooky fixed point, for the cyclic solution found in Sec. IV
for N ¼ 4.476.
points have zero norm. Indeed, let us assume that we have a
complex eigenvalue m ∈ C with eigenvector ui,
∂βi j
u ¼ mui : ð46Þ
∂gj
Now let us differentiate the relation (42) with respect to gK :
∂ K GIJ βJ þ GIJ ∂ K βJ ¼ ∂ I ∂ K F: ð47Þ
At a spooky fixed point we have βJ ðgÞ ¼ 0 for real
couplings g. Contracting the relation (47) with uK and
ūI at a spooky fixed point we get
ūI GIJ ∂ K βJ uK ¼ uK ūI ∂ I ∂ K F: ð48Þ
Using (46) we arrive at the following relations:
mūI uJ GIJ ¼ ūI uJ ∂ I ∂ J F: ð49Þ
Since GIJ and ∂ I ∂ J F are real symmetric matrices, the norm
u2 ¼ GIJ uI ūJ and f ¼ ūI uJ ∂ I ∂ J F are real numbers. If they
are not equal to zero, then we must have m ∈ R, which
contradicts our assumption. Therefore, the norm u2 ¼ 0.
Another consequence of the negative eigenvalues of Gij
is that dF=dt can have either sign, as follows from (44). In
Fig. 10 we plot FðtÞ for the limit cycle of Fig. 9, showing
that it oscillates. This can also be shown analytically for a
small limit cycle surrounding a fixed point. We may expand
around it to find
βi ðtÞ ¼ aðtÞvi þ āðtÞv̄i ; ð50Þ
where vi and v̄i are the eigenvectors corresponding to
the complex eigenvalues of the Jacobian matrix at the
spooky fixed point. While Gij vi v̄j vanishes, Gij vi vj ≠ 0.
Therefore, Eq. (44) implies that dF=dt ≠ 0 for a small
limit cycle.
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

ACKNOWLEDGMENTS Letting μ denote the renormalization scale, we take the beta

function associated with a coupling gi to be given by
We are grateful to Alexander Gorsky and Wenli Zhao for
useful discussions. We also thank the referee for useful
comments about evanescent operators. This research was dgi 1 ð2Þ 1 ð4Þ
β gi ¼ μ ¼ −2ϵgi þ β̃g þ β̃g
supported in part by the U.S. NSF under Grant No. PHY- dμ 6!ð8πÞ2 i ð6!Þ2 ð8πÞ4 i
1914860.
þ Oðg6 Þ; ðA1Þ
APPENDIX A: THE BETA FUNCTIONS UP
ð2Þ
TO FOUR LOOPS where we have separated out the two-loop contribution βgi
ð4Þ
In the main text we presented the large N beta functions and the four-loop contribution βgi . The beta functions have
for the matrix models we have studied. In this Appendix we been computed by the use of the formulas for sextic
list the full beta functions for any N up to four-loops. theories in d ¼ 3 − ϵ dimension listed in Sec. II.

1. Beta functions for the OðNÞ2 matrix model

ð2Þ
β̃g1 ¼ 24ð100 þ 24N þ 3N 2 Þg21 þ 384ð9 þ 4NÞg1 g2 þ 3840g1 g3 þ 64ð32 þ N 2 Þg22 ; ðA2Þ

ð2Þ
β̃g2 ¼ 144ð8 þ 3NÞg21 þ 96ð38 þ 4N þ N 2 Þg1 g2 þ 2304ð1 þ NÞg1 g3 þ 128ð8 þ 7NÞg22 þ 384ð18 þ N 2 Þg2 g3 ; ðA3Þ

ð2Þ
β̃g3 ¼ 168g21 þ 96ð3 þ 2NÞg1 g2 þ 1152g1 g3 þ 32ð21 þ 2N þ N 2 Þg22 þ 768ð1 þ 2NÞg2 g3 þ 192ð22 þ 3N 2 Þg23 ; ðA4Þ

ð4Þ
β̃g1 ¼ −288ð47952 þ 4780π 2 þ N 4 ð17 þ π 2 Þ þ N 3 ð372 þ 25π 2 Þ þ 8Nð3102 þ 277π 2 Þ þ N 2 ð5248 þ 412π 2 ÞÞg31
− 576ð64992 þ 6860π 2 þ 6N 3 ð104 þ 7π 2 Þ þ 8Nð4728 þ 415π 2 Þ þ N 2 ð5928 þ 465π 2 ÞÞg21 g2
− 1152ð48Nð274 þ 27π 2 Þ þ N 2 ð2824 þ 225π 2 Þ þ 4ð7640 þ 891π 2 ÞÞg21 g3
− 384ð3N 4 ð10 þ π 2 Þ þ 18N 3 ð12 þ π 2 Þ þ 48Nð884 þ 83π 2 Þ þ 112ð867 þ 94π 2 Þ þ N 2 ð10836 þ 773π 2 ÞÞg1 g22
− 13824ð3984 þ 448π 2 þ 2N 3 ð20 þ π 2 Þ þ N 2 ð92 þ 7π 2 Þ þ 8Nð292 þ 31π 2 ÞÞg1 g2 g3
512 3
− 4608ð5936 − 8N 4 þ 720π 2 þ N 2 ð372 þ 45π 2 ÞÞg1 g23 − ðN ð960 þ 46π 2 Þ þ 64ð900 þ 97π 2 Þ
3
þ N 2 ð1704 þ 137π 2 Þ þ 16Nð2124 þ 203π 2 ÞÞg32 − 9216ð4N 4 þ 384ð9 þ π 2 Þ þ N 2 ð248 þ 21π 2 ÞÞg22 g3 ; ðA5Þ

ð4Þ
β̃g2 ¼ −432ð20400 þ 2260π 2 þ 2N 3 ð90 þ 7π 2 Þ þ 12Nð940 þ 91π 2 Þ þ N 2 ð1740 þ 151π 2 ÞÞg31
− 288ð3N 4 ð10 þ π 2 Þ þ 6N 3 ð56 þ 5π 2 Þ þ 16ð6408 þ 683π 2 Þ þ N 2 ð11184 þ 995π 2 Þ þ Nð46896 þ 4516π 2 ÞÞg21 g2
− 1728ðN 3 ð248 þ 22π 2 Þ þ N 2 ð1380 þ 109π 2 Þ þ 8Nð1510 þ 127π 2 Þ þ 4ð4132 þ 401π 2 ÞÞg21 g3
− 384ð2N 3 ð534 þ 49π 2 Þ þ N 2 ð5148 þ 443π 2 Þ þ 8ð8922 þ 923π 2 Þ þ Nð48384 þ 4444π 2 ÞÞg1 g22
− 4608ð2N 4 ð6 þ π 2 Þ þ 6N 3 ð8 þ π 2 Þ þ 6Nð948 þ 77π 2 Þ þ N 2 ð2748 þ 197π 2 Þ þ 2ð8112 þ 841π 2 ÞÞg1 g2 g3
− 27648ð1 þ NÞðN 2 ð62 þ 3π 2 Þ þ 2ð532 þ 51π 2 ÞÞg1 g23
− 128ð95152 þ 10024π 2 þ 36N 3 ð6 þ π 2 Þ þ 2N 4 ð36 þ 7π 2 Þ þ 24Nð1264 þ 113π 2 Þ þ N 2 ð14804 þ 1179π 2 ÞÞg32
− 768ð2N 4 π 2 þ N 5 π 2 þ 134N 2 ð12 þ π 2 Þ þ 16N 3 ð102 þ 7π 2 Þ þ 8ð4308 þ 433π 2 Þ þ 8Nð4584 þ 437π 2 ÞÞg22 g3
− 13824ð4816 þ 512π 2 þ N 4 ð18 þ π 2 Þ þ N 2 ð644 þ 57π 2 ÞÞg2 g23 ; ðA6Þ

046015-15
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

ð4Þ
β̃g3 ¼ −432ð2760 þ 380π 2 þ N 2 ð210 þ 23π 2 Þ þ 4Nð270 þ 31π 2 ÞÞg31
− 576ð7308 þ 776π 2 þ N 3 ð78 þ 8π 2 Þ þ N 2 ð483 þ 45π 2 Þ þ 6Nð766 þ 83π 2 ÞÞg21 g2
− 576ð−48N 3 − 8N 4 þ 6Nð836 þ 81π 2 Þ þ 6ð1984 þ 189π 2 Þ þ N 2 ð1676 þ 207π 2 ÞÞg21 g3
− 768ð8772 þ 894π 2 þ N 4 ð6 þ π 2 Þ þ N 3 ð36 þ 5π 2 Þ þ 10Nð336 þ 31π 2 Þ þ N 2 ð1269 þ 140π 2 ÞÞg1 g22
− 2304ð6096 þ 550π 2 þ 2N 3 ð84 þ 17π 2 Þ þ N 2 ð432 þ 59π 2 Þ þ Nð6312 þ 554π 2 ÞÞg1 g2 g3
− 1152ð18N 3 π 2 þ 15N 4 π 2 þ 96Nð35 þ 3π 2 Þ þ 8N 2 ð443 þ 36π 2 Þ þ 8ð1876 þ 177π 2 ÞÞg1 g23
− 384ð41328 þ 4192π 2 þ 2N 4 ð54 þ 19π 2 Þ þ N 3 ð216 þ 38π 2 Þ þ 8Nð1536 þ 125π 2 Þ þ 3N 2 ð3932 þ 323π 2 ÞÞg22 g3
128
− ð49104 þ 4784π 2 þ 4N 4 π 2 þ N 5 π 2 þ 12N 2 ð487 þ 42π 2 Þ þ N 3 ð2136 þ 281π 2 Þ þ 12Nð4552 þ 425π 2 ÞÞg32
3
− 3456ð1 þ 2NÞðN 4 π 2 þ 112ð32 þ 3π 2 Þ þ 4N 2 ð88 þ 7π 2 ÞÞg2 g23
− 1152ðN 6 π 2 þ N 4 ð424 þ 34π 2 Þ þ 32ð826 þ 85π 2 Þ þ N 2 ð6864 þ 620π 2 ÞÞg33 : ðA7Þ

2. Beta functions for the antisymmetric matrix model

ð2Þ
β̃g1 ¼ 6ð112 − 3N þ 3N 2 Þg21 þ 384ð−1 þ 2NÞg1 g2 þ 3840g1 g3 þ 32ð64 − N þ N 2 Þg22 ; ðA8Þ

ð2Þ
β̃g2 ¼ 54ð−1 þ 2NÞg21 þ 24ð68 − N þ N 2 Þg1 g2 þ 576ð−1 þ 2NÞg1 g3 þ 224ð−1 þ 2NÞg22 þ 192ð36 − N þ N 2 Þg2 g3 ;
ð2Þ
β̃g3 ¼ 42g21 þ ð−24 þ 48NÞg1 g2 þ 576g1 g3 þ 8ð40 − N þ N 2 Þg22 þ 384ð−1 þ 2NÞg2 g3 þ 96ð44 − 3N þ 3N 2 Þg23 ;
ð4Þ
β̃g1 ¼ −9ð−4N 3 ð17 þ π 2 Þ þ 2N 4 ð17 þ π 2 Þ þ 32ð3209 þ 293π 2 Þ − Nð10928 þ 861π 2 Þ þ N 2 ð10962 þ 863π 2 ÞÞg31
− 72ð−1 þ 2NÞð−3Nð104 þ 7π 2 Þ þ 3N 2 ð104 þ 7π 2 Þ þ 4ð4896 þ 413π 2 ÞÞg21 g2 − 288ð−Nð2824 þ 225π 2 Þ
þ N 2 ð2824 þ 225π 2 Þ þ 4ð7804 þ 945π 2 ÞÞg21 g3 − 48ð198048 þ 21616π 2 − 6N 3 ð10 þ π 2 Þ þ 3N 4 ð10 þ π 2 Þ
þ 2N 2 ð10479 þ 746π 2 Þ − Nð20928 þ 1489π 2 ÞÞg1 g22 − 3456ð−1 þ 2NÞð−Nð20 þ π 2 Þ þ N 2 ð20 þ π 2 Þ
þ 8ð292 þ 31π 2 ÞÞg1 g2 g3 − 2304ð8N 3 − 4N 4 − 3Nð124 þ 15π 2 Þ þ N 2 ð368 þ 45π 2 Þ þ 32ð371 þ 45π 2 ÞÞg1 g23
128
− ð−1 þ 2NÞð33984 þ 3248π 2 − Nð480 þ 23π 2 Þ þ N 2 ð480 þ 23π 2 ÞÞg32
3
− 4608ð−4N 3 þ 2N 4 þ 768ð9 þ π 2 Þ − Nð248 þ 21π 2 Þ þ N 2 ð250 þ 21π 2 ÞÞg22 g3 ; ðA9Þ

ð4Þ
β̃g2 ¼ −27ð−1 þ 2NÞð5760 þ 557π 2 − Nð90 þ 7π 2 Þ þ N 2 ð90 þ 7π 2 ÞÞg31 − 18ð−6N 3 ð10 þ π 2 Þ
þ 3N 4 ð10 þ π 2 Þ þ 34N 2 ð579 þ 52π 2 Þ − Nð19656 þ 1765π 2 Þ þ 4ð49956 þ 5437π 2 ÞÞg21 g2
− 216ð−1 þ 2NÞð9536 þ 830π 2 − Nð124 þ 11π 2 Þ þ N 2 ð124 þ 11π 2 ÞÞg21 g3 − 48ð−1 þ 2NÞð39744 þ 3739π 2
− Nð534 þ 49π 2 Þ þ N 2 ð534 þ 49π 2 ÞÞg1 g22 − 288ð−8N 3 ð6 þ π 2 Þ þ 4N 4 ð6 þ π 2 Þ þ 3N 2 ð3608 þ 259π 2 Þ
− Nð10800 þ 773π 2 Þ þ 4ð25284 þ 2719π 2 ÞÞg1 g2 g3 − 3456ð−1 þ 2NÞð−Nð62 þ 3π 2 Þ þ N 2 ð62 þ 3π 2 Þ
þ 4ð532 þ 51π 2 ÞÞg1g32 − 32ð−2N 3 ð36 þ 7π 2 Þ þ N 4 ð36 þ 7π 2 Þ − 4Nð3656 þ 291π 2 Þ þ N 2 ð14660 þ 1171π 2 Þ
þ 4ð38180 þ 4109π 2 ÞÞg32 − 48ð−1 þ 2NÞð−2N 3 π 2 þ N 4 π 2 − 32Nð102 þ 7π 2 Þ þ 3N 2 ð1088 þ 75π 2 Þ
þ 32ð4584 þ 437π 2 ÞÞg22 g3 − 3456ð−2N 3 ð18 þ π 2 Þ þ N 4 ð18 þ π 2 Þ þ 64ð301 þ 32π 2 Þ − 2Nð644 þ 57π 2 Þ
þ N 2 ð1306 þ 115π 2 ÞÞg2 g23 ; ðA10Þ

046015-16
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

ð4Þ
β̃g3 ¼ −27ð2760 þ 422π 2 − Nð210 þ 23π 2 Þ þ N 2 ð210 þ 23π 2 ÞÞg31 − 18ð−1 þ 2NÞð−2Nð39 þ 4π 2 Þ
þ N 2 ð78 þ 8π 2 Þ þ 75ð96 þ 11π 2 ÞÞg21 g2 − 72ð16724 þ 8N 3 − 4N 4 þ 1647π 2 þ 3N 2 ð592 þ 69π 2 Þ ðA11Þ

− Nð1780 þ 207π 2 ÞÞg21 g3 − 12ð−8N 3 ð6 þ π 2 Þ þ 4N 4 ð6 þ π 2 Þ þ 256ð402 þ 43π 2 Þ þ 3N 2 ð3184 þ 351π 2 Þ

− Nð9528 þ 1049π 2 ÞÞg1 g22 − 288ð−1 þ 2NÞð5952 þ 518π 2 − Nð84 þ 17π 2 Þ þ N 2 ð84 þ 17π 2 ÞÞg1 g2 g3
− 144ð−30N 3 π 2 þ 15N 4 π 2 þ 224ð253 þ 24π 2 Þ − Nð7088 þ 567π 2 Þ þ N 2 ð7088 þ 582π 2 ÞÞg1 g23
4
− ð−1 þ 2NÞð−2N 3 π 2 þ N 4 π 2 − 24Nð178 þ 23π 2 Þ þ N 2 ð4272 þ 553π 2 Þ þ 8ð24672 þ 2359π 2 ÞÞg32
3
− 96ð75984 þ 7828π 2 − 2N 3 ð54 þ 19π 2 Þ þ N 4 ð54 þ 19π 2 Þ þ 3N 2 ð3914 þ 327π 2 Þ − 2Nð5844 þ 481π 2 ÞÞg22 g3
− 432ð−1 þ 2NÞð−2N 3 π 2 þ N 4 π 2 þ 448ð32 þ 3π 2 Þ − 8Nð88 þ 7π 2 Þ þ N 2 ð704 þ 57π 2 ÞÞg2 g23
− 144ð−3N 5 π 2 þ N 6 π 2 þ N 4 ð848 þ 71π 2 Þ þ 256ð826 þ 85π 2 Þ − N 3 ð1696 þ 137π 2 Þ − 16Nð1716 þ 155π 2 Þ
þ 4N 2 ð7076 þ 637π 2 ÞÞg33 : ðA12Þ

3. Beta functions for the symmetric traceless matrix model

ð2Þ 2400 − 1200N þ 250N 2 þ 51N 3 þ 3N 4 2 2N 2 þ 10N − 35

β̃g1 ¼ 6 g 1 þ 384 g1 g2 þ 3840g1 g3
N2 N
−20 þ 5N þ N 2
þ 864 g1 g4 þ 32ð62 þ N þ N 2 Þg22 þ 4608g2 g4 þ 2592g24 ; ðA13Þ
N

ð2Þ −150 þ 35N þ 6N 2 2 480 − 120N þ 66N 2 þ 9N 3 þ N 4 −10 þ 5N þ 2N 2

β̃g2 ¼ 18 g1 þ 24 g g
1 2 þ 576 g1 g3
N N2 N
80 − 20N þ N 2 −132 þ 39N þ 14N 2 2 2 −40 þ 3N þ N 2
þ 216 g g
1 4 þ 32 g 2 þ 192ð34 þ N þ N Þg g
2 3 þ 288 g2 g4
N2 N N
−24 þ 2N þ N 2 2
þ 3456g3 g4 þ 324 g4 ; ðA14Þ
N

ð2Þ −30 þ 7N þ 2N 2 1080 −6 þ 3N þ 2N 2

β̃g3 ¼ 42g21 þ 576g1 g3 þ 24 g1 g2 − g1 g4 þ 384 g2 g3
N N N
−24 þ 3N þ N 2 2 2 288 − 36N þ 30N 2 þ 5N 3 þ N 4 2 3456
− 288 g g
2 4 þ 96ð38 þ 3N þ 3N Þg þ 8 g2 − gg
N2 3
N2 N 3 4
−16 þ 2N þ N 2 2
− 324 g4 ; ðA15Þ
N2

ð2Þ −200 − 75N 2 þ 15N 3 þ 3N 4 2 10 − 5N þ N 2 160 − 120N þ 34N 2 þ 15N 3 þ 3N 4

β̃g4 ¼ 24 g 1 þ 192 g 1 g 2 þ 12 g1 g4
N3 N2 N2
62 þ N þ N 2 2 −15 þ 3N þ N 2 −704 þ 60N þ 28N 2 þ 3N 3 þ N 4 2
− 32 g2 þ 384 g2 g4 þ 3840g3 g4 þ 6 g4 ; ðA16Þ
N N N

ð4Þ 9
β̃g1 ¼ − ð2N 8 ð17 þ π 2 Þ þ 4N 7 ð389 þ 26π 2 Þ þ 38400ð1252 þ 135π 2 Þ − 19200Nð2159 þ 225π 2 Þ
N4
− 60N 4 ð7338 þ 455π 2 Þ − 1200N 3 ð4896 þ 587π 2 Þ þ N 6 ð38822 þ 3167π 2 Þ þ 800N 2 ð30564 þ 3215π 2 Þ
216
þ N 5 ð279004 þ 28019π 2 ÞÞg31 − ð2N 6 ð104 þ 7π 2 Þ þ 4320Nð522 þ 55π 2 Þ þ 10N 3 ð−1416 þ 131π 2 Þ
N3
þ N 5 ð4264 þ 331π 2 Þ − 960ð3344 þ 375π 2 Þ þ 5N 4 ð7096 þ 681π 2 Þ − 40N 2 ð16616 þ 1977π 2 ÞÞg21 g2
288
− ð1920ð388 þ 45π 2 Þ þ 30N 2 ð1184 þ 171π 2 Þ þ N 4 ð2824 þ 225π 2 Þ − 120Nð3628 þ 405π 2 Þ
N2

046015-17
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

54
þ N 3 ð29128 þ 2817π 2 ÞÞg21 g3 − ð12N 6 ð95 þ 7π 2 Þ − 20N 3 ð9940 þ 471π 2 Þ þ 3N 5 ð6608 þ 561π 2 Þ
N3
− 320N 2 ð9175 þ 1104π 2 Þ − 1280ð13166 þ 1485π 2 Þ þ 640Nð17947 þ 1890π 2 Þ þ N 4 ð137468 þ 13785π 2 ÞÞg21 g4
48
− ð3N 6 ð10 þ π 2 Þ þ 6N 5 ð82 þ 7π 2 Þ þ 9504ð756 þ 85π 2 Þ − 2160Nð1581 þ 172π 2 Þ þ 2N 4 ð10971 þ 763π 2 Þ
N2
3456
þ 2N 2 ð37896 þ 6847π 2 Þ þ N 3 ð177600 þ 16271π 2 ÞÞg1 g22 − ð2N 4 ð20 þ π 2 Þ þ N 3 ð244 þ 17π 2 Þ − 24ð2372
N
1728
þ 275π 2 Þ þ N 2 ð4108 þ 447π 2 Þ þ 2Nð8588 þ 975π 2 ÞÞg1 g2 g3 − 2 ð518304 þ 58320π 2 þ N 2 ð−3072 þ 41π 2 Þ
N
þ 2N 4 ð555 þ 49π 2 Þ − 48Nð4834 þ 525π 2 Þ þ N 3 ð10782 þ 1037π 2 ÞÞg1 g2 g4 − 2304ð−8N 3 − 4N 4
13824
þ N 2 ð384 þ 45π 2 Þ þ Nð388 þ 45π 2 Þ þ 6ð1852 þ 225π 2 ÞÞg1 g23 − ð−47ð388 þ 45π 2 Þ þ N 2 ð1120 þ 117π 2 Þ
N
108
þ Nð4844 þ 540π 2 ÞÞg1 g3 g4 − ð301N 5 þ 41N 6 þ N 4 ð10882 þ 1053π 2 Þ − 4N 2 ð26366 þ 1755π 2 Þ
N2
þ 128ð41992 þ 4725π 2 Þ þ N 3 ð98224 þ 9801π 2 Þ − 16Nð140912 þ 15255π 2 ÞÞg1 g24
128 4
− ðN ð960 þ 46π 2 Þ þ 108Nð2220 þ 241π 2 Þ þ N 3 ð4848 þ 343π 2 Þ − 324ð3352 þ 375π 2 Þ
3N
þ 7N 2 ð7584 þ 749π 2 ÞÞg32 − 4608ð6424 þ 4N 3 þ 2N 4 þ 726π 2 þ 3Nð80 þ 7π 2 Þ þ N 2 ð242 þ 21π 2 ÞÞg22 g3
288 4
− ðN ð120 þ 7π 2 Þ þ N 3 ð660 þ 43π 2 Þ þ 6N 2 ð4426 þ 437π 2 Þ þ 24Nð4957 þ 534π 2 Þ − 48ð12638 þ 1413π 2 ÞÞg22 g4
N
1296 4
− 20736ð3368 þ 378π 2 þ Nð44 þ 3π 2 Þ þ N 2 ð44 þ 3π 2 ÞÞg2 g3 g4 − ðN ð32 þ 3π 2 Þ þ N 3 ð96 þ 9π 2 Þ
N
− 896ð188 þ 21π 2 Þ þ 12Nð2440 þ 261π 2 Þ þ N 2 ð7184 þ 716π 2 ÞÞg2 g24 − 4478976ð9 þ π 2 Þg3 g24
1944
− ð36N 3 þ 12N 4 þ 96Nð75 þ 8π 2 Þ − 192ð242 þ 27π 2 Þ þ N 2 ð2028 þ 203π 2 ÞÞg34 ; ðA17Þ
N

ð4Þ 27
β̃g2 ¼ − ð2N 8 ð90 þ 7π 2 Þ þ 24000Nð20 þ 9π 2 Þ − 96000ð28 þ 9π 2 Þ − 3200N 2 ð1443 þ 170π 2 Þ
N5
− 5N 5 ð8372 þ 209π 2 Þ þ N 7 ð3750 þ 323π 2 Þ − 100N 4 ð6542 þ 785π 2 Þ þ 400N 3 ð7120 þ 797π 2 Þ
18
þ N 6 ð28350 þ 3133π 2 ÞÞg31 − ð3N 8 ð10 þ π 2 Þ − 309600Nð32 þ 3π 2 Þ þ 86400ð212 þ 23π 2 Þ þ N 7 ð732 þ 66π 2 Þ
N4
− 8N 4 ð13443 þ 482π 2 Þ þ N 6 ð25770 þ 2294π 2 Þ − 120N 3 ð29800 þ 3209π 2 Þ þ 240N 2 ð46680 þ 4993π 2 Þ
216
þ N 5 ð200436 þ 19313π 2 ÞÞg21 g2 − ð−7200ð88 þ 7π 2 Þ þ N 6 ð248 þ 22π 2 Þ þ 600Nð832 þ 63π 2 Þ
N3
þ N 5 ð3132 þ 251π 2 Þ þ 10N 3 ð2672 þ 359π 2 Þ − 40N 2 ð6504 þ 677π 2 Þ þ 3N 4 ð8852 þ 717π 2 ÞÞg21 g3
162
− ð32000ð92 þ 9π 2 Þ − 16000Nð104 þ 9π 2 Þ þ N 6 ð1418 þ 155π 2 Þ þ 320N 2 ð4644 þ 521π 2 Þ þ N 5 ð12998
N4
48
þ 1497π 2 Þ − 20N 3 ð19900 þ 2381π 2 Þ − N 4 ð42584 þ 3431π 2 ÞÞg12 g4 − 3 ð2N 6 ð534 þ 49π 2 Þ þ 1620Nð2044
N
þ 193π 2 Þ − 2160ð3060 þ 319π 2 Þ þ N 5 ð11898 þ 1033π 2 Þ − 72N 2 ð14522 þ 1613π 2 Þ þ 3N 3 ð32656 þ 4997π 2 Þ
288
þ N 4 ð91914 þ 8441π 2 ÞÞg1 g22 − ð4N 6 ð6 þ π 2 Þ þ 16N 5 ð15 þ 2π 2 Þ þ 720ð1400 þ 139π 2 Þ þ 5N 4 ð2184 þ 149π 2 Þ
N2
216
− 180Nð2568 þ 229π 2 Þ þ N 3 ð52944 þ 4095π 2 Þ þ N 2 ð47952 þ 6716π 2 ÞÞg1 g2 g3 − 3 ð−36N 3 ð626 þ 7π 2 Þ
N
þ 2N ð69 þ 8π Þ þ 9N ð194 þ 19π Þ − 11520ð358 þ 37π Þ þ 480Nð4018 þ 381π 2 Þ þ N 4 ð15348 þ 1549π 2 Þ
6 2 5 2 2

046015-18
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

3456
− 8N 2 ð53892 þ 6641π 2 ÞÞg1 g2 g4 − ð−10 þ 5N þ 2N 2 ÞðNð62 þ 3π 2 Þ þ N 2 ð62 þ 3π 2 Þ þ 6ð334 þ 33π 2 ÞÞg1 g23
N
2592 4
− ðN ð492 þ 35π 2 Þ þ 160ð988 þ 99π 2 Þ − 2N 2 ð2248 þ 179π 2 Þ þ N 3 ð3084 þ 227π 2 Þ
N2
162
− 20Nð3176 þ 291π 2 ÞÞg1 g3 g4 − 3 ð36N 4 ð163 þ 17π 2 Þ þ N 6 ð172 þ 21π 2 Þ þ 7N 5 ð182 þ 27π 2 Þ þ 1600Nð1028
N
32
þ 99π Þ − 640ð6016 þ 621π Þ − 24N 2 ð9098 þ 1363π 2 Þ − 2N 3 ð17432 þ 1635π 2 ÞÞg1 g24 − 2 ðN 6 ð36 þ 7π 2 Þ
2 2
N
þ N 5 ð288 þ 50π 2 Þ þ 2N 4 ð7186 þ 545π 2 Þ þ 8N 3 ð8863 þ 756π 2 Þ þ 216ð11996 þ 1265π 2 Þ − 54Nð13992 þ 1337π 2 Þ
48
þ N 2 ð30720 þ 5671π 2 ÞÞg32 − ð13N 5 π 2 þ 2N 6 π 2 þ 8N 4 ð816 þ 53π 2 Þ − 720ð2872 þ 295π 2 Þ þ 16N 2 ð15324
N
288
þ 1499π 2 Þ þ N 3 ð22656 þ 1637π 2 Þ þ 36Nð17072 þ 1737π 2 ÞÞg22 g3 − 2 ð2N 4 ð771 þ 59π 2 Þ − 144Nð2041 þ 200π 2 Þ
N
− 4N ð9261 þ 824π Þ þ N ð9918 þ 829π Þ þ 144ð7988 þ 839π ÞÞg2 g4 − 3456ð2N 3 ð18 þ π 2 Þ þ N 4 ð18 þ π 2 Þ
2 2 3 2 2 2

1728
þ 2Nð608 þ 55π 2 Þ þ N 2 ð1234 þ 111π 2 Þ þ 8ð2095 þ 228π 2 ÞÞg2 g23 − ð2N 4 ð24 þ π 2 Þ þ N 3 ð192 þ 11π 2 Þ
N
324
þ 6Nð3168 þ 281π 2 Þ − 48ð3112 þ 319π 2 Þ þ N 2 ð5952 þ 481π 2 ÞÞg2 g3 g4 − 2 ðN 6 ð2 þ π 2 Þ þ 2N 5 ð5 þ 3π 2 Þ
N
þ N 4 ð1628 þ 171π 2 Þ − 16N 2 ð3517 þ 339π 2 Þ þ 192ð7024 þ 735π 2 Þ − 32Nð9160 þ 933π 2 Þ
þ N 3 ð9376 þ 974π 2 ÞÞg2 g24 − 20736ðNð62 þ 3π 2 Þ þ N 2 ð62 þ 3π 2 Þ þ 6ð334 þ 33π 2 ÞÞg23 g4
2592 4
− ðN ð28 þ 3π 2 Þ þ 198Nð32 þ 3π 2 Þ þ N 3 ð84 þ 9π 2 Þ þ N 2 ð2572 þ 249π 2 Þ − 8ð7960 þ 813π 2 ÞÞg3 g24
N
972
− 2 ð20N 5 þ 4N 6 þ N 4 ð146 þ 21π 2 Þ þ 4N 3 ð170 þ 27π 2 Þ − 32Nð1082 þ 117π 2 Þ þ 128ð1526 þ 159π 2 Þ
N
− 2N 2 ð4736 þ 513π 2 ÞÞg34 ; ðA18Þ

ð4Þ 9
β̃g3 ¼ − ð432000π 2 þ 72000N 2 ð8 þ 3π 2 Þ þ 2400N 4 ð214 þ 35π 2 Þ − 600N 5 ð312 þ 41π 2 Þ þ N 8 ð630 þ 69π 2 Þ
N6
18
− 2N 6 ð1200 þ 137π 2 Þ þ N 7 ð7110 þ 813π 2 ÞÞg31 − 5 ð28800Nð2 þ 3π 2 Þ þ 4N 8 ð39 þ 4π 2 Þ − 14400ð112 þ 27π 2 Þ
N
− 7200N ð292 þ 27π Þ þ 6N ð361 þ 34π Þ þ 240N 3 ð4272 þ 427π 2 Þ − 20N 4 ð11754 þ 869π 2 Þ þ 2N 6 ð8379 þ 883π 2 Þ
2 2 7 2

72
þ N 5 ð516 þ 1189π 2 ÞÞg21 g2 þ ð56N 7 þ 4N 8 þ N 4 ð2220 − 519π 2 Þ þ 1200Nð−32 þ 9π 2 Þ − 1200ð136 þ 27π 2 Þ
N4
þ 240N 3 ð614 þ 39π 2 Þ − 300N 2 ð896 þ 75π 2 Þ − N 6 ð1576 þ 207π 2 Þ − N 5 ð12524 þ 1179π 2 ÞÞg21 g3
162 6
þ ðN ð302 þ π 2 Þ − 4800Nð2 þ 3π 2 Þ þ 19200ð14 þ 3π 2 Þ þ 1600N 2 ð215 þ 18π 2 Þ − 480N 3 ð321 þ 31π 2 Þ
N5
12
þ 20N 4 ð587 þ 74π 2 Þ þ N 5 ð5174 þ 314π 2 ÞÞg21 g4 − 4 ð4N 8 ð6 þ π 2 Þ þ 48N 7 ð7 þ π 2 Þ − 8640Nð212 þ 19π 2 Þ
N
þ 12960ð352 þ 35π Þ þ 16N ð−1119 þ 188π Þ − 192N 3 ð3093 þ 241π 2 Þ þ 3N 6 ð3440 þ 367π 2 Þ
2 4 2

288
þ 144N 2 ð16952 þ 1693π 2 Þ þ N 5 ð55992 þ 4967π 2 ÞÞg1 g22 − ð2880Nð35 þ 3π 2 Þ þ 2N 6 ð84 þ 17π 2 Þ
N3
− 360ð440 þ 39π 2 Þ þ 24N 3 ð770 þ 69π 2 Þ þ N 5 ð1116 þ 169π 2 Þ − 6N 2 ð10352 þ 953π 2 Þ þ N 4 ð10956 þ 985π 2 ÞÞg1 g2 g3
216
− ð3N 6 ð40 þ π 2 Þ þ 9N 5 ð78 þ π 2 Þ þ 3840ð191 þ 18π 2 Þ − 960Nð317 þ 27π 2 Þ − 24N 3 ð2503 þ 192π 2 Þ
N4

046015-19
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

144
þ 96N 2 ð3303 þ 352π 2 Þ − N 4 ð15564 þ 995π 2 ÞÞg1 g2 g4 − ð66N 5 π 2 þ 15N 6 π 2 þ 600ð112 þ 9π 2 Þ
N2
þ 16N 4 ð443 þ 33π 2 Þ − 60Nð1120 þ 87π 2 Þ þ 6N 2 ð5216 þ 567π 2 Þ þ N 3 ð20528 þ 1377π 2 ÞÞg1 g23
864
þ ð84N 5 þ 12N 6 þ N 4 ð148 − 27π 2 Þ þ N 3 ð1628 þ 57π 2 Þ þ 80ð956 þ 81π 2 Þ − 20Nð2552 þ 207π 2 Þ
N3
162
þ 2N 2 ð9808 þ 1017π 2 ÞÞg1 g3 g4 − 4 ð3N 5 ð−208 þ 7π 2 Þ þ N 6 ð−94 þ 9π 2 Þ þ 1920ð368 þ 33π 2 Þ
N
− 480Nð632 þ 51π Þ − 24N ð1022 þ 81π 2 Þ − 2N 4 ð5156 þ 357π 2 Þ þ 16N 2 ð14228 þ 1707π 2 ÞÞg1 g24
2 3

4
− ð21N 7 π 2 þ 2N 8 π 2 þ 17280Nð465 þ 41π 2 Þ þ 576N 3 ð433 þ 80π 2 Þ − 432N 2 ð1464 þ 205π 2 Þ
3N 3
þ 4N 6 ð2136 þ 277π 2 Þ − 2592ð9552 þ 971π 2 Þ þ 12N 4 ð28836 þ 2929π 2 Þ þ N 5 ð59568 þ 5281π 2 ÞÞg32
96 6
− ðN ð54 þ 19π 2 Þ þ 4N 5 ð81 þ 19π 2 Þ − 108Nð1264 þ 103π 2 Þ þ 540ð1112 þ 109π 2 Þ þ 15N 2 ð2064 þ 281π 2 Þ
N2
72
þ 6N 3 ð5540 þ 431π 2 Þ þ 2N 4 ð5655 þ 454π 2 ÞÞg22 g3 − 3 ð2N 6 ð24 þ π 2 Þ þ 24N 2 ð492 þ π 2 Þ þ N 5 ð384 þ 19π 2 Þ
N
− N 4 ð2952 þ 35π 2 Þ − 1728ð1150 þ 117π 2 Þ − 6N 3 ð4492 þ 271π 2 Þ þ 144Nð3942 þ 367π 2 ÞÞg22 g4
432
− ð−6 þ 3N þ 2N 2 Þð2N 3 π 2 þ N 4 π 2 þ Nð704 þ 52π 2 Þ þ N 2 ð704 þ 53π 2 Þ þ 4ð3232 þ 309π 2 ÞÞg2 g23
N
1728
− 2 ð27N 3 ð4 þ π 2 Þ þ 4N 4 ð3 þ 2π 2 Þ − 174Nð104 þ 9π 2 Þ þ 168ð580 þ 57π 2 Þ − 3N 2 ð1720 þ 137π 2 ÞÞg2 g3 g4
N
324
− 3 ð4N 6 þ N 5 ð50 þ π 2 Þ − N 4 ð1060 þ 91π 2 Þ þ 16N 2 ð1289 þ 133π 2 Þ − 2N 3 ð3704 þ 339π 2 Þ
N
þ 16Nð9704 þ 975π 2 Þ − 64ð9968 þ 1017π 2 ÞÞg2 g24 − 144ð3N 5 π 2 þ N 6 π 2 þ N 4 ð848 þ 65π 2 Þ þ N 3 ð1696 þ 125π 2 Þ
864
þ 4Nð6016 þ 555π 2 Þ þ 24ð6664 þ 711π 2 Þ þ N 2 ð24912 þ 2282π 2 ÞÞg33 − ð−57056 þ 752N − 368N 2
N
432
− 5472π 2 þ 84Nπ 2 þ 9N 3 π 2 þ 3N 4 π 2 Þg23 g4 þ ð20N 5 þ 4N 6 − N 4 ð52 þ 9π 2 Þ − 2N 3 ð212 þ 27π 2 Þ
N2
486
þ 10N 2 ð1580 þ 153π 2 Þ þ 4Nð10064 þ 963π 2 Þ − 16ð17792 þ 1755π 2 ÞÞg3 g24 − 3 ðN 6 ð−8 þ π 2 Þ − 10N 4 ð34 þ 3π 2 Þ
N
þ N 5 ð−40 þ 6π 2 Þ − 16N 3 ð94 þ 13π 2 Þ − 512ð400 þ 41π 2 Þ þ 64Nð632 þ 71π 2 Þ þ 8N 2 ð1354 þ 163π 2 ÞÞg34 ; ðA19Þ

ð4Þ 9
β̃g4 ¼ − ð−13824000 þ 6912000N þ N 8 ð360 þ 22π 2 Þ − 2400N 4 ð448 þ 43π 2 Þ þ 9600N 3 ð395 þ 49π 2 Þ
N5
− 60N 5 ð1672 þ 65π 2 Þ − 3200N 2 ð2352 þ 275π 2 Þ þ N 7 ð6660 þ 489π 2 Þ þ N 6 ð41220 þ 4241π 2 ÞÞg31
72
− ð172800ð14 þ π 2 Þ − 14400Nð68 þ 3π 2 Þ − 2N 4 ð14844 þ 131π 2 Þ þ N 6 ð1848 þ 191π 2 Þ þ 480N 2 ð2120 þ 243π 2 Þ
N4
10368
þ 5N 5 ð3504 þ 415π 2 Þ − 120N 3 ð3640 þ 423π 2 ÞÞg21 g2 − ð800N þ 20N 3 ð17 þ 3π 2 Þ − 200ð28 þ 3π 2 Þ
N3
18
þ 3N 4 ð32 þ 5π 2 Þ − 5N 2 ð392 þ 51π 2 ÞÞg21 g3 − 4 ð670N 7 þ 41N 8 þ 38400ð320 þ 27π 2 Þ − 9600Nð467 þ 27π 2 Þ
N
− 6N 4 ð32878 þ 417π 2 Þ þ N 6 ð14413 þ 801π 2 Þ − 120N 3 ð17800 þ 1731π 2 Þ þ 240N 2 ð22264 þ 2481π 2 Þ
3456
þ N 5 ð95816 þ 8415π 2 ÞÞg21 g4 − ð3N 3 ð−28 þ π 2 Þ þ N 4 ð−4 þ 3π 2 Þ þ 240ð52 þ 5π 2 Þ þ 2N 2 ð−148 þ 9π 2 Þ
N2

046015-20
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

576
− 40Nð152 þ 15π 2 ÞÞg1 g2 g3 − ð960ð94 þ 9π 2 Þ − 240Nð262 þ 27π 2 Þ þ 6N 2 ð728 þ 153π 2 Þ
N2
48
þ N 4 ð1160 þ 153π 2 Þ þ N 3 ð6392 þ 801π 2 ÞÞg1 g3 g4 − 3 ð51840Nð15 þ π 2 Þ − 34560ð57 þ 5π 2 Þ þ N 6 ð48 þ 7π 2 Þ
N
þ N 5 ð228 þ 43π 2 Þ þ 18N 4 ð346 þ 101π 2 Þ þ 4N 3 ð2526 þ 1579π 2 Þ − 8N 2 ð54528 þ 6997π 2 ÞÞg1 g22
432
− ð7N 6 ð16 þ π 2 Þ þ 10N 5 ð102 þ 7π 2 Þ − 1920ð296 þ 27π 2 Þ þ 480Nð521 þ 39π 2 Þ þ 8N 3 ð437 þ 221π 2 Þ
N3
108
þ N 4 ð5556 þ 727π 2 Þ − 4N 2 ð28812 þ 3317π 2 ÞÞg1 g2 g4 − 3 ð2514N 5 þ 318N 6 þ 88N 3 ð−241 þ 18π 2 Þ
N
− 5120ð289 þ 27π Þ þ 7N ð1084 þ 99π Þ þ 320Nð2350 þ 189π 2 Þ − 16N 2 ð16366 þ 1557π 2 ÞÞg1 g24
2 4 2
ðA20Þ

128
þ ð−63504ð10 þ π 2 Þ þ N 4 ð528 þ π 2 Þ þ 324Nð532 þ 53π 2 Þ þ N 3 ð3120 þ 163π 2 Þ þ 8N 2 ð4314 þ 337π 2 ÞÞg32
3N 2
4608 144
þ ð6424 þ 4N 3 þ 2N 4 þ 726π 2 þ 3Nð80 þ 7π 2 Þ þ N 2 ð242 þ 21π 2 ÞÞg22 g3 − 2 ðN 6 ð2 þ π 2 Þ þ N 5 ð12 þ 7π 2 Þ
N N
þ 9408ð88 þ 9π 2 Þ − 6N 2 ð3476 þ 93π 2 Þ þ N 4 ð2134 þ 239π 2 Þ − 48Nð4495 þ 429π 2 Þ þ N 3 ð12708 þ 1351π 2 ÞÞg22 g4
6912 4
− ðN ð4 þ π 2 Þ þ 4N 3 ð6 þ π 2 Þ − 72ð184 þ 21π 2 Þ þ 3N 2 ð308 þ 39π 2 Þ þ Nð2488 þ 306π 2 ÞÞg2 g3 g4
N
2592
− 2 ð68224 þ 12N 5 þ 2N 6 þ 7104π 2 − 224N 2 ð15 þ π 2 Þ þ N 4 ð126 þ 11π 2 Þ þ N 3 ð644 þ 57π 2 Þ
N
− 16Nð1043 þ 93π 2 ÞÞg2 g24 − 2304ð−8N 3 − 4N 4 þ N 2 ð384 þ 45π 2 Þ þ Nð388 þ 45π 2 Þ þ 6ð1852 þ 225π 2 ÞÞg23 g4
864
− ð27N 3 ð8 þ π 2 Þ þ 9N 4 ð8 þ π 2 Þ þ 4N 2 ð808 þ 99π 2 Þ − 32ð2584 þ 297π 2 Þ þ Nð8096 þ 972π 2 ÞÞg3 g24
N
108
− 2 ð13N 5 þ 5N 6 þ N 4 ð1198 þ 45π 2 Þ þ 4N 3 ð1723 þ 63π 2 Þ − 8N 2 ð6253 þ 414π 2 Þ − 32Nð5692 þ 459π 2 Þ
N
þ 128ð6388 þ 675π 2 ÞÞg34 : ðA21Þ

APPENDIX B: THE F FUNCTION AND METRIC FOR THE SYMMETRIC TRACELESS MODEL
Working up to the two-loop order, we find that the F function that enters the gradient flow expression (42) is given by
F ¼ Fð1Þ þ Fð2Þ , where
ϵ
Fð1Þ ¼ −
576N 3
× ½ð2N 2 ð48g2 ð4g3 N 5 þ ð10g3 þ 3g4 ÞN 4 þ 3ð6g3 þ 5g4 ÞN 3 þ 6ð4g3 − 7g4 ÞN 2
−72ðg3 þ 2g4 ÞN þ 288g4 Þ þ 4g22 ðN 6 þ 6N 5 þ 45N 4 þ 124N 3 − 168N 2 − 720N þ 1296Þ
þ 3ðð16g23 þ 3g24 ÞN 6 þ ð32g23 þ 15g24 ÞN 5 þ 24ð6g23 þ g24 ÞN 4 þ 4ð32g23 þ 48g4 g3 þ 15g24 ÞN 3
þ96ð2g23 þ 4g4 g3 − 5g24 ÞN 2 − 192g4 ð8g3 þ 7g4 ÞN þ 3072g24 ÞÞ
þ 12g1 Nð9g4 N 6 þ ð80g3 þ 63g4 ÞN 5 þ ð272g3 − 42g4 ÞN 4 − 120ð2g3 þ 7g4 ÞN 3
− 240ð4g3 − g4 ÞN 2 þ 4g2 ð2N 6 þ 15N 5 þ 11N 4 − 140N 3 þ 720N − 720Þ þ 960ðg3 þ 4g4 ÞN − 3840g4 Þ
þ 3g21 ðN 8 þ 14N 7 þ 83N 6 þ 46N 5 − 960N 4 þ 4800N 2 − 9600N þ 9600ÞÞ ðB1Þ

and Fð2Þ may be written in terms of the three-point functions in the free theory in d ¼ 3 [24,25]:

Cijk
Fð2Þ ∼ Cijk gi gj gk ; hOi ðxÞOj ðyÞOk ðzÞi ¼ : ðB2Þ
jx − yj jx − zj3 jy − zj3
3

Explicitly, we find

046015-21
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

3
Fð2Þ ¼
13271040N 5 π 2
× ½ð3N 12 þ 93N 11 þ 1717N 10 þ 13103N 9 þ 15072N 8 − 227572N 7 − 326400N 6
þ 2596800N 5 − 758400N 4 − 12288000N 3 þ 29952000N 2 − 40704000N þ 29184000Þg31
16
þ Ng2 ð27ð6N 10 þ 109N 9 þ 878N 8 þ 1885N 7 − 10882N 6 − 28000N 5 þ 122880N 4 þ 28800N 3
3
− 672000N 2 þ 1411200N − 1094400Þg21 þ 9NðN 10 þ 15N 9 þ 405N 8 þ 3493N 7 þ 8634N 6
− 30684N 5 − 102504N 4 þ 351168N 3 þ 408960N 2 − 2194560N þ 1969920Þg2 g1
þ 8N 2 ð26N 8 þ 219N 7 þ 1446N 6 þ 5399N 5 − 714N 4 − 57456N 3 þ 30240N 2 þ 343440N − 443232Þg22 Þ
þ 192N 2 g3 ð2ððN 8 þ 7N 7 þ 181N 6 þ 757N 5 þ 1990N 4 þ 3832N 3 − 7296N 2 − 27504N þ 49248Þg22
þ 12Nð6N 6 þ 21N 5 þ 118N 4 þ 253N 3 þ 270N 2 þ 348N − 1368Þg3 g2 þ 4N 2 ð3N 6 þ 9N 5
þ 71N 4 þ 127N 3 þ 402N 2 þ 340N þ 456Þg23 ÞN 2 þ 12g1 ðð2N 8 þ 17N 7 þ 174N 6 þ 773N 5
þ 162N 4 − 6176N 3 þ 240N 2 þ 28080N − 27360Þg2 þ 2Nð15N 6 þ 66N 5 þ 196N 4 þ 421N 3
− 570N 2 − 2100N þ 2280Þg3 ÞN þ 3ð29N 8 þ 310N 7 þ 997N 6 − 1612N 5 − 10020N 4
þ 15600N 3 þ 38400N 2 − 112800N þ 91200Þg21 Þ − 18Nð−N 2 − 2N þ 8Þg4 ððN 8 þ 6N 7 þ 47N 6
þ 198N 5 þ 1428N 4 þ 7416N 3 − 32512N 2 − 121344N þ 311296Þg24 N 2 þ 32ððN 6 þ 7N 5 þ 113N 4
þ 629N 3 − 1470N 2 − 7920N þ 16416Þg22 þ 24NðN 4 þ 4N 3 þ 41N 2 þ 114N − 456Þg3 g2
þ 48N 2 ð3N 2 þ 3N þ 38Þg23 ÞN 2 þ 96ððN 6 þ 6N 5 þ 46N 4 þ 225N 3 − 728N 2 − 3192N þ 7296Þg2
þ 2Nð5N 4 þ 15N 3 þ 86N 2 þ 228N − 1216Þg3 Þg4 N 2 þ 192g1 ðð7N 6 þ 65N 5 þ 52N 4 − 964N 3 − 650N 2
þ 7680N − 9120Þg2 þ þ2Nð27N 4 þ 141N 3 − 190N 2 − 1140N þ 1520Þg3 ÞN þ 3ð3N 8 þ 24N 7
þ 325N 6 þ 2364N 5 − 100N 4 − 41712N 3 − 10240N 2 þ 318720N − 389120Þg1 g4 N þ 3ð21N 8 þ 294N 7
þ 1599N 6 þ 30N 5 − 27920N 4 þ 209600N 2 − 499200N þ 486400Þg21 : ðB3Þ

The metric Gij is given by

1
G11 ¼ ðN 8 þ 14N 7 þ 83N 6 þ 46N 5 − 960N 4 þ 4800N 2 − 9600N þ 9600Þ;
192N 3
1
G12 ¼ G21 ¼ ð2N 6 þ 15N 5 þ 11N 4 − 140N 3 þ 720N − 720Þ;
24N 2
1
G13 ¼ G23 ¼ ð5N 4 þ 17N 3 − 15N 2 − 60N þ 60Þ;
6N
1
G14 ¼ G41 ¼ ðN − 2ÞðN þ 4Þð3N 4 þ 15N 3 − 20N 2 − 120N þ 160Þ;
32N 2
1
G22 ¼ ðN 6 þ 6N 5 þ 45N 4 þ 124N 3 − 168N 2 − 720N þ 1296Þ;
72N
1
G23 ¼ G32 ¼ ð2N 4 þ 5N 3 þ 9N 2 þ 12N − 36Þ;
6
1
G24 ¼ G42 ¼ ðN − 2ÞðN þ 4ÞðN 2 þ 3N − 12Þ;
4N
1
G33 ¼ N 3 ðN 4 þ 2N 3 þ 9N 2 þ 8N þ 12Þ; G34 ¼ G43 ¼ ðN − 2ÞN 3 ðN þ 4Þ;
6
1
G44 ¼ ðN − 2Þ2 ðN þ 4Þ2 ðN 2 þ N þ 16Þ: ðB4Þ
32N

046015-22
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS … PHYS. REV. D 103, 046015 (2021)

At this order it is independent of the couplings gi and is proportional to the matrix of two-point functions (43) in the free
theory in d ¼ 3.

APPENDIX C: CALCULATING THE HOPF CONSTANT

In this Appendix we compute the Hopf constant a at two loops. Introducing rescaled couplings gi ¼ 720ð8πÞ2 ϵgi , the
beta functions at the critical value N ¼ N crit ¼ 4.475 in units of ϵ become

βg1 ¼ −2g1 þ ð2339.99g1 þ 4273.55g2 þ 3840:g3 þ 4325.08g4 Þg1 þ 2768.04g22 þ 2592:g24 þ 4608:g2 g4 ;
βg2 ¼ −2g2 þ ð509.966g1 þ 2962.93g2 þ 6748.16g3 þ 113.519g4 Þg1 þ ð3456:g3 þ 360.299g4 Þg4
þ ð2308.94g2 þ 11232.3g3 − 421.438g4 Þg2 ;
βg3 ¼ −2g3 þ ð42g1 þ 221.912g2 þ 576:g3 − 241.337g4 Þg1 þ 10704.4g23 − 209.942g24 − 772.278g3 g4
þ ð629.906g2 þ 4074.01g3 − 135.923g4 Þg2 ;
βg4 ¼ −2g4 þ ð226.417g1 þ 73.3524g2 þ 1708.55g4 Þg1 − 618.547g22 þ ð1583.3g2 þ 3840:g3 þ 1066.11g4 Þg4 :

These beta functions have a fixed point at

g ðN crit Þ ¼ 10−4 · ð3.48916; −4.64792; 3.04945; −1.08745Þ: ðC1Þ

∂β
Letting V ¼ ðv1 ; v2 ; v3 ; v̄3 Þ be the matrix of eigenvectors vi of the stability matrix ð ∂ggji Þ evaluated at this fixed point,


∂βgi
−1
V V ¼ diagð2; −1.57495; −0.153965i; 0.153965iÞ: ðC2Þ
∂gj

One can check that these eigenvalues change on varying N. In particular, the real parts of the complex eigenvalues change
linearly with N for N close to N crit . Changing to variables t1 ¼ v1 · g, t2 ¼ v2 · g, t3 ¼ ℜ½v3 · g, t4 ¼ ℑ½v3 · g, we get the
equations

βt1 ¼ 2t1 − 3006.27t21 − 635.361t22 − 4.22379t23 þ 4.22379t24 þ 7.65924t3 t4 ;

βt2 ¼ −1.57495t2 þ ð−638.903t1 þ 1471.36t2 − 96.8862t3 þ 72.0709t4 Þt2
þ 1.0131t23 − 0.34628t24 − 1.37241t3 t4 ;
βt3 ¼ −0.153965t4 þ ð231.430t4 − 3006.27t3 Þt1 þ ð−31746.2t2 þ 1284.37t3 − 347.122t4 Þt2
− 49.5972t23 þ 492.731t24 þ 178.686t3 t4 ;
βt4 ¼ 0.153965t3 þ ð−231.43t3 − 3006.27t4 Þt1 þ ð638.003t2 þ 730.144t4 − 82.7131t3 Þt2
þ 8.73689t23 þ 823.772t24 þ 153.731t3 t4 : ðC3Þ

We wish to study the RG flow in the manifold that is tangent
to the center eigenspace. We cannot simply set t1 and t2 to
zero, since this plane is not invariant under the RG flow: the
t23 , t24 , and t3 t3 terms in βt1 and βt2 generate a flow in t1 and t2 .
But by introducing new variables with t1 and t2 suitably
shifted,
u1 ¼ t1 − 1.77501t23 þ 4.3762t4 t3 þ 1.77501t24 ; ðC4Þ
u2 ¼ t2 − 0.709414t23 þ 0.676770t4 t3 þ 0.286027t24 ;
ðC5Þ
the t23 , t24 , and t3 t3 terms in βu1 and βu2 cancel out. While βu1
and βu2 do couple to t3 and t4 at third order, one can introduce
new variables yet again and shift u1 and u2 by cubic terms in
t3 and t4 to remove this third order coupling. This procedure
may be iterated indefinitely to obtain a coordinate expansion
of the center manifold to arbitrary order, in accordance with
the center manifold theorem. We will content ourselves with
the cubic approximation of the center manifold, which
consists of the surface u1 ¼ u2 ¼ 0, since this approximation
suffices to determine the Hopf constant. Eliminating t1 and t2
in favor of u1 and u2 in the equations for βt3 and βt4 , setting u1
and u2 to zero, and discarding unreliable quartic terms gives

046015-23
JEPSEN, KLEBANOV, and POPOV PHYS. REV. D 103, 046015 (2021)

βt3 ¼ −49.5972t23 þ 178.686t4 t3 þ 492.731t24 − 0.153965t4

− 4425.01ð1:t33 − 2.81386t4 t23 − 0.947101t24 t3 þ 0.0703961t34 Þ;
βt4 ¼ 8.73689t23 þ 153.731t4 t3 þ 0.153965t3 þ 823.772t24
− 469.468ð1:t33 þ 7.98654t4 t23 − 27.8962t24 t3 − 10.9216t34 Þ: ðC6Þ

From these equations the Hopf constant can be directly obtained by the use of equation (3.4.11)* in [51] or by the equivalent
formula in [52]. We find that

a ≈ 6204790 ðC7Þ

so that Hopf’s theorem guarantees the existence of a periodic orbit that is IR attractive in the center manifold, implying that if we
fine-tune the couplings in the vicinity of N crit , there is a cyclic solution to the beta functions that comes back precisely to itself.

[1] K. Wilson and J. B. Kogut, The renormalization group and
the epsilon expansion, Phys. Rep. 12, 75 (1974).
[2] S. Gukov, RG flows and bifurcations, Nucl. Phys. B919,
583 (2017).
[3] S. D. Glazek and K. G. Wilson, Limit Cycles in Quantum
Theories, Phys. Rev. Lett. 89, 230401 (2002); Erratum,
Phys. Rev. Lett.92, 139901 (2004).
[4] E. Braaten and D. Phillips, Renormalization-group limit
cycle for the r12 -potential, Phys. Rev. A 70, 052111 (2004).
[5] A. Gorsky and F. Popov, Atomic collapse in graphene and
cyclic renormalization group flow, Phys. Rev. D 89, 061702
(2014).
[6] S. M. Dawid, R. Gonsior, J. Kwapisz, K. Serafin, M.
Tobolski, and S. Głazek, Renormalization group procedure
for potential −g=r2 , Phys. Lett. B 777, 260 (2018).
[7] V. Efimov, Energy levels arising from resonant two-body
forces in a three-body system, Phys. Lett. 33B, 563 (1970).
[8] K. Bulycheva and A. Gorsky, Limit cycles in renormaliza-
tion group dynamics, Usp. Fiz. Nauk 184, 182 (2014).
[9] J.-F. Fortin, B. Grinstein, and A. Stergiou, Limit cycles in
four dimensions, J. High Energy Phys. 12 (2012) 112.
[10] J.-F. Fortin, B. Grinstein, and A. Stergiou, Limit cycles and
conformal invariance, J. High Energy Phys. 01 (2013) 184.
[11] M. A. Luty, J. Polchinski, and R. Rattazzi, The a-theorem
and the asymptotics of 4D quantum field theory, J. High
Energy Phys. 01 (2013) 152.
[12] J. L. Cardy, Is there a c theorem in four-dimensions?, Phys.
Lett. B 215, 749 (1988).
[13] I. Jack and H. Osborn, Analogs for the c theorem for four-
dimensional renormalizable field theories, Nucl. Phys.
B343, 647 (1990).
[14] Z. Komargodski and A. Schwimmer, On renormalization
group flows in four dimensions, J. High Energy Phys. 12
(2011) 099.
[15] A. Morozov and A. J. Niemi, Can renormalization group
flow end in a big mess?, Nucl. Phys. B666, 311 (2003).
[16] T. L. Curtright, X. Jin, and C. K. Zachos, RG Flows, Cycles,
and c-Theorem Folklore, Phys. Rev. Lett. 108, 131601
(2012).
[17] D. J. Binder and S. Rychkov, Deligne categories in lattice
models and quantum field theory, or making sense of OðNÞ
symmetry with non-integer N, J. High Energy Phys. 04
(2020) 117.
[18] M. Hogervorst, S. Rychkov, and B. C. van Rees, Unitarity
violation at the Wilson-Fisher fixed point in 4-ϵ dimensions,
Phys. Rev. D 93, 125025 (2016).
[19] E. Hopf, Bifurcation of a periodic solution from a stationary
solution of a system of differential equations, Berlin
Mathematische Physics Klasse, Sachsischen Akad. Wissen-
schaften Leipzig 94, 3 (1942).
[20] R. C. Myers and A. Sinha, Seeing a c-theorem with
holography, Phys. Rev. D 82, 046006 (2010).
[21] D. L. Jafferis, I. R. Klebanov, S. S. Pufu, and B. R. Safdi,
Towards the F-theorem: N ¼ 2 field theories on the three-
sphere, J. High Energy Phys. 06 (2011) 102.
[22] I. R. Klebanov, S. S. Pufu, and B. R. Safdi, F-theorem
without supersymmetry, J. High Energy Phys. 10 (2011)
038.
[23] H. Casini and M. Huerta, Renormalization group running of
the entanglement entropy of a circle, Phys. Rev. D 85,
125016 (2012).
[24] S. Giombi and I. R. Klebanov, Interpolating between a and
F, J. High Energy Phys. 03 (2015) 117.
[25] L. Fei, S. Giombi, I. R. Klebanov, and G. Tarnopolsky,
Generalized F-theorem and the ϵ expansion, J. High Energy
Phys. 12 (2015) 155.
[26] I. Jack, D. Jones, and C. Poole, Gradient flows in three
dimensions, J. High Energy Phys. 09 (2015) 061.
[27] I. Jack and C. Poole, a-function in three dimensions:
Beyond the leading order, Phys. Rev. D 95, 025010 (2017).
[28] I. R. Klebanov, F. Popov, and G. Tarnopolsky, TASI lectures
on large N tensor models, Proc. Sci., TASI2017 (2018) 004
[arXiv:1808.09434].
[29] T. Eguchi and H. Kawai, Reduction of Dynamical Degrees
of Freedom in the Large N Gauge Theory, Phys. Rev. Lett.
48, 1063 (1982).
[30] S. Kachru and E. Silverstein, 4-D Conformal Theories and
Strings on Orbifolds, Phys. Rev. Lett. 80, 4855 (1998).

046015-24
RG LIMIT CYCLES AND UNCONVENTIONAL FIXED POINTS …
[31] A. E. Lawrence, N. Nekrasov, and C. Vafa, On conformal
field theories in four-dimensions, Nucl. Phys. B533, 199
(1998).
[32] M. Bershadsky, Z. Kakushadze, and C. Vafa, String ex-
pansion as large N expansion of gauge theories, Nucl. Phys.
B523, 59 (1998).
[33] M. Bershadsky and A. Johansen, Large N limit of orbifold
field theories, Nucl. Phys. B536, 141 (1998).
[34] A. Armoni, M. Shifman, and G. Veneziano, Exact results in
non-supersymmetric large N orientifold field theories, Nucl.
Phys. B667, 170 (2003).
[35] P. Kovtun, M. Unsal, and L. G. Yaffe, Necessary and
sufficient conditions for non-perturbative equivalences of
large N(c) orbifold gauge theories, J. High Energy Phys. 07
(2005) 008.
[36] M. Unsal and L. G. Yaffe, (In)validity of large N orientifold
equivalence, Phys. Rev. D 74, 105019 (2006).
[37] M. R. Douglas and G. W. Moore, D-branes, quivers, and
ALE instantons, arXiv:hep-th/9603167.
[38] A. A. Tseytlin and K. Zarembo, Effective potential in non-
supersymmetric SU(N) x SU(N) gauge theory and inter-
actions of type 0 D3-branes, Phys. Lett. B 457, 77 (1999).
[39] A. Dymarsky, I. Klebanov, and R. Roiban, Perturbative
search for fixed lines in large N gauge theories, J. High
Energy Phys. 08 (2005) 011.
[40] A. Dymarsky, I. Klebanov, and R. Roiban, Perturbative
gauge theory and closed string tachyons, J. High Energy
Phys. 11 (2005) 038.
[41] E. Pomoni and L. Rastelli, Large N field theory and AdS
tachyons, J. High Energy Phys. 04 (2009) 020.
046015-25
PHYS. REV. D 103, 046015 (2021)
[42] V. Gorbenko, S. Rychkov, and B. Zan, Walking, Weak first-
order transitions, and complex CFTs, J. High Energy Phys.
10 (2018) 108.
[43] V. Gorbenko, S. Rychkov, and B. Zan, Walking, Weak first-
order transitions, and complex CFTs II. Two-dimensional
Potts model at Q > 4, SciPost Phys. 5, 050 (2018).
[44] D. B. Kaplan, J.-W. Lee, D. T. Son, and M. A. Stephanov,
Conformality lost, Phys. Rev. D 80, 125005 (2009).
[45] H. Osborn and A. Stergiou, Seeking fixed points in multiple
coupling scalar theories in the ϵ expansion, J. High Energy
Phys. 05 (2018) 051.
[46] J. Hager, Six-loop renormalization group functions of
O(n)-symmetric Φ6 -theory and ϵ-expansions of tricritical
exponents up to ϵ3 , J. Phys. A 35, 2703 (2002).
[47] S. Giombi, I. R. Klebanov, and G. Tarnopolsky, Bosonic
tensor models at large N and small ϵ, Phys. Rev. D 96,
106014 (2017).
[48] S. Giombi, I. R. Klebanov, F. Popov, S. Prakash, and G.
Tarnopolsky, Prismatic large N models for bosonic tensors,
Phys. Rev. D 98, 105005 (2018).
[49] P. Hartman, A lemma in the theory of structural stability of
differential equations, Proc. Am. Math. Soc. 11, 610 (1960).
[50] D. M. Grobman, Homeomorphism of systems of differential
equations, Dokl. Akad. Nauk SSSR 128, 880 (1959).
[51] J. Guckenheimer and P. Holmes, Nonlinear Oscillations,
Dynamical Systems, and Bifurcations of Vector Fields,
Vol. 42 (Springer Science & Business Media, New York,
2013).
[52] S.-N. Chow and J. Mallet-Paret, Integral averaging and
bifurcation, J. Differ. Eqn. 26, 112 (1977).