---
title: Renormalisation Group Invariants (RGIs) Overview
url: https://www.emergentmind.com/topics/renormalisation-group-invariants-rgis
type: topic
---

# Renormalisation Group Invariants (RGIs) Overview

to=arxiv.search  天天中彩票人工_code
{"query":"Renormalization Group Invariants arXiv", "max_results": 10, "sort_by":"relevance"}	RTLU
to=arxiv.search  天天中彩票买json
{"query":"1507.00328 Renormalization Group Evolution of Flavour Invariants", "max_results": 3, "sort_by":"relevance"}
to=arxiv.search  天天中彩票在_code
{"query":"2410.10107 all-loop renormalization group invariants for MSSM", "max_results": 3, "sort_by":"relevance"}
Renormalisation Group Invariants (RGIs) are combinations of running parameters that remain constant along the renormalisation-group flow. If the running parameters are $x_i(\mu)$ and $t$ is the logarithmic scale variable, an invariant $I(x(\mu))$ satisfies
\[
\frac{dI}{dt}=\sum_i \frac{\partial I}{\partial x_i}\,\beta_{x_i}(\{x\})=0.
\]
In practice, RGIs occur in several distinct senses: as exact invariants of truncated perturbative RGEs, as approximate invariants controlled by hierarchies, as basis-independent invariants of flavour structures, and as reparametrisation invariants that are invariant under field redefinitions rather than under scale evolution. Across the literature, RGIs are used to encode UV information in low-energy data, to formulate sum rules, to organize flavour and neutrino sectors without choosing a basis, and to identify cases in which effective potentials or specific operator combinations are exactly scale independent [1507.03757].

## 1. Definition, loop order, and the meaning of “invariant”

The defining equation of an RGI is the vanishing of the total derivative with respect to $\ln\mu$. In a Callan–Symanzik formulation this may include explicit scale dependence and field anomalous dimensions,
\[
\frac{d I}{d\ln\mu}
=\mu\frac{\partial I}{\partial\mu}
+\sum_i \beta_i(\lambda)\frac{\partial I}{\partial\lambda_i}
-\gamma(\lambda)\phi\frac{\partial I}{\partial\phi}=0,
\]
so the relevant notion of invariance depends on the object under study and on the parameter space in which the RG flow is formulated [1703.02079].

Perturbative usage is loop-order specific. For one-loop RGIs, $dI/dt=0$ when the $\beta$-functions are truncated at one loop. For two-loop RGIs, one uses an ansatz
\[
I=I_1+\frac{1}{16\pi^2}I_2
\]
and imposes
\[
I_1^{(1)}=0,\qquad I_1^{(2)}+I_2^{(1)}=0,
\]
with the formal three-loop term neglected at two-loop order [1507.03757]. This distinction is essential: exact at one loop is not the same as exact to all loops.

A second distinction is between basis independence and RG invariance. In flavour physics, quantities built from traces, determinants, adjugates, and commutators of Yukawa spurions are flavour invariants because they are unchanged by flavour-basis transformations. They are not automatically constant under scale evolution. The literature on Standard Model flavour therefore separates flavour invariants from true RG invariants, and explicitly notes that true RG invariants are rare [1507.00328].

A third distinction appears in models with two $U(1)$ gauge factors. There the central quantities
\[
I_{f_if_j}=\vec v_{f_i}^{\,T}[K]^{-1}\vec v_{f_j}
\]
are invariant under generic gauge-field reparametrisations, including rescalings, because $[K]\to R^{-T}[K]R^{-1}$ and $\vec v_f\to R^T\vec v_f$ leave the contraction unchanged. These are reparametrisation invariants, not generally RGIs; they still run with scale through the running of $[K]$ [2605.20058].

## 2. Construction principles and algorithmic searches

One construction strategy is symmetry based. In supersymmetric theories, one-loop scalar-mass $\beta$-functions have tightly constrained structures imposed by gauge and global $U(1)$ symmetries. By contracting scalar masses with suitable linear combinations of preserved charges and combining the result with $S\equiv \mathrm{Tr}(Y m^2)$ and gaugino-mass terms, one obtains invariants whose one-loop derivatives vanish identically. This logic underlies the MSSM invariants built from $Y$, $B$, $L$, and $X$ charges and their analogues in the dMSSM and pMSSM [1507.03470].

A second strategy is exhaustive algebraic search. RGIsearch treats the $\beta$-functions as polynomials in the running parameters with rational coefficients and searches for several classes of invariants. Monomial invariants take the form
\[
M=\prod_{i=1}^n x_i^{a_i},
\]
and the invariance condition reduces to a linear system for the integer exponents $a_i$. Polynomial invariants take the form
\[
P=\sum_{i=1}^m C_i M_i,
\]
with monomials of common “dimensionality”, and again the condition of term-by-term cancellation produces a sparse linear system for the coefficients $C_i$. Factorized polynomial invariants,
\[
P_j(\vec d)=x_j^b\Big(\sum_i C_i M_i\Big),
\]
are treated through a generalized-eigenvalue problem in the integer $b$ [1507.03757].

The same work uses automatically discovered additive dimensionalities to block-diagonalize the search and bounds the combinatorics by restricting exponent size and the number of distinct parameters per monomial. The resulting systems are solved with fraction-free Gaussian elimination with Markowitz pivoting, and two-loop searches impose the coupled conditions $I_1^{(1)}=0$ and $I_1^{(2)}+I_2^{(1)}=0$ on the same sparse-linear-algebra infrastructure [1507.03757].

A more formal route is the method of characteristics. In the SI2PI analysis, the general condition
\[
\sum_i \beta_i(\lambda)\frac{\partial I(\lambda)}{\partial\lambda_i}=0
\]
is identified as the equation to be solved for coupling-space invariants, although the paper emphasizes exact RG-invariant observables rather than a catalogue of closed-form coupling invariants [1703.02079].

## 3. Flavour and neutrino-sector invariants

In Standard Model flavour physics, the basic hermitian spurions are
\[
U\equiv Y_UY_U^\dagger,\qquad D\equiv Y_DY_D^\dagger.
\]
For three generations, a convenient generating set consists of $10$ CP-even and $1$ CP-odd invariants:
\[
I_1=\mathrm{tr}(U),\quad I_2=\mathrm{tr}(D),\quad \tilde I_3=\mathrm{tr}(\tilde U),\quad \tilde I_4=\mathrm{tr}(\tilde D),\quad \tilde I_5=\mathrm{tr}(UD),
\]
\[
\tilde I_6=\mathrm{tr}(\tilde U U)=3\det U,\quad \tilde I_8=\mathrm{tr}(\tilde D D)=3\det D,
\]
\[
\tilde I_7=\mathrm{tr}(\tilde U D),\quad \tilde I_9=\mathrm{tr}(U\tilde D),\quad \tilde I_{10}=\mathrm{tr}(\tilde U\tilde D),
\]
and
\[
I_{11}^-=-\frac{3i}{8}\det[U,D]=\mathrm{tr}(A_8^3),
\qquad A_8\equiv \frac{i}{2}[U,D].
\]
These quantities encode masses, moduli of CKM elements, and CP violation without choosing a flavour basis [1507.00328].

At one loop in the Standard Model with QCD only, the Yukawa RGEs close on a finite octet basis, and the invariant RGEs are explicit. Two exact one-loop RGIs reproduced in this language are the Harrison–Krishnan–Scott invariants,
\[
\frac{d}{dt}\Big[\frac{\tilde I_5}{(\tilde I_6\tilde I_8)^{1/3}}\Big]=0,
\qquad
\frac{d}{dt}\Big[\frac{\tilde I_{10}}{(\tilde I_6\tilde I_8)^{2/3}}\Big]=0,
\]
which hold because the one-loop SM satisfies $a_1=-b_2=-a_2=b_1$. The same analysis also states that beyond one loop, or beyond the SM, these combinations are not generally invariant, and that with generic Yukawa structures there are no nontrivial exact RGIs other than those protected by such accidental one-loop identities [1507.00328].

The same framework yields approximate RGIs under realistic hierarchies. With SM-like Wolfenstein/Froggatt–Nielsen scaling and top-Yukawa dominance, seven approximate RG-invariant combinations can be formed, such as
\[
I_1\tilde I_6/\tilde I_3^2 \simeq \mathrm{const.},
\qquad
I_2\tilde I_8/\tilde I_4^2 \simeq \mathrm{const.},
\qquad
(\tilde I_{10}-\tilde I_3\tilde I_4)/(\tilde I_3\tilde I_4)\simeq \mathrm{const.},
\]
which encode the multiplicative running patterns of CKM parameters under top-Yukawa dominance [1507.00328].

In the neutrino sector, the one-loop RGE for the coefficient $\kappa$ of the Weinberg operator is multiplicative and flavor-separable in the charged-lepton mass basis. This implies that the phases of all elements of the Majorana mass matrix are one-loop RGIs,
\[
\frac{d\phi_{ij}}{dt}=0,
\]
and that simple ratios of matrix elements are protected. The paper highlights
\[
\frac{(M_\nu)_{ee}}{(M_\nu)_{e\mu}},\qquad
\frac{(M_\nu)_{ee}}{(M_\nu)_{\mu\mu}},\qquad
\frac{(M_\nu)_{e\tau}}{(M_\nu)_{\mu\tau}},
\]
as numerically excellent invariants under the approximation $I_e\simeq I_\mu$, and
\[
\frac{(M_\nu)_{e\tau}^2}{(M_\nu)_{ee}(M_\nu)_{\tau\tau}}
\]
as an exact one-loop RGI. Their invariance is independent of neutrino mass ordering and of the parameterization of the lepton mixing matrix [1306.1375].

## 4. Supersymmetric RGIs and high-scale sum rules

In the MSSM and its phenomenological restrictions, one-loop RGIs provide direct probes of high-scale supersymmetry breaking. The most basic examples are
\[
I_{B_a}\equiv \frac{M_a}{g_a^2},\qquad a=1,2,3,
\]
and the gauge-only combinations
\[
I_{g_2}\equiv \frac{1}{g_1^2}-\frac{33}{5}\frac{1}{g_2^2},
\qquad
I_{g_3}\equiv \frac{1}{g_1^2}+\frac{11}{5}\frac{1}{g_3^2}.
\]
The hypercharge-trace invariant is
\[
I_{Y\alpha}\equiv \frac{S}{g_1^2},
\]
with $S=\mathrm{Tr}(Y m^2)$, and the scalar sector admits flavour-sensitive combinations such as $D_{B_{13}}$, $D_{L_{13}}$, $D_{\chi_1}$, $D_{Y_{13H}}$, and $D_Z$ [1204.4336].

These quantities organize unification tests and mediation diagnostics into algebraic sum rules. Gauge coupling unification implies
\[
I_{g_2}+\frac{7}{4}I_{g_3}=0.
\]
Gaugino-mass unification implies
\[
\Big(I_{B_1}-\frac{33}{5}I_{B_2}\Big)I_{g_3}
=
\Big(I_{B_1}+\frac{11}{5}I_{B_3}\Big)I_{g_2}.
\]
Flavour universality gives
\[
D_{B_{13}}=0,\qquad D_{L_{13}}=0.
\]
Minimal gauge mediation, general gauge mediation, anomaly mediation, and minimal anomaly mediation each produce characteristic relations among the same invariants, including the fixed AMSB ratios
\[
I_{B_1}-\frac{33}{5}I_{B_2}=0,\qquad
I_{B_1}+\frac{11}{5}I_{B_3}=0,
\]
and, in mAMSB,
\[
D_{\chi_1}-\frac{40}{33}I_{M_1}=0,\qquad
D_{\chi_1}-8I_{M_2}=0,\qquad
D_{\chi_1}+\frac{16}{3}I_{M_3}=0
\]
[1211.1157].

A complementary pMSSM analysis uses posterior distributions of weak-scale parameters, computes the RGIs point by point, and interprets the resulting distributions as messenger-scale information because the invariants are constant to one-loop accuracy. In that setting, the gauge-only invariants are approximately
\[
I_{g_2}\simeq -10.9,\qquad I_{g_3}\simeq 6.2,
\]
and the gaugino-unification condition can be written as
\[
12\,I_{B_2}-5\,I_{B_1}-7\,I_{B_3}=0.
\]
The same analysis maps RGI measurements into the parameters $A_i$ and $B_i$ of General Gauge Mediation and into the one-parameter structure of Minimal Gauge Mediation [1205.5903].

Deflected mirage mediation shows how threshold effects modify the picture. In that scenario the one-loop RGIs are piecewise invariant: below and above the messenger threshold each invariant obeys the usual one-loop relation, but at $\mu=M_{\rm mess}$ the gaugino invariants jump by
\[
\Delta I_{B_a}
=
-\frac{N M_0}{16\pi^2}\,\alpha_m(1+\alpha_g)\ln\!\frac{M_P}{m_{3/2}},
\]
while specially constructed D-type invariants remain continuous because messenger-induced scalar thresholds cancel in the corresponding linear combinations [1505.03455].

Systematic searches in the MSSM, dMSSM, and pMSSM show that the number of invariants is considerably reduced at two loops. In the full MSSM, a two-loop continuation exists for $M_2/g^2$,
\[
J_1\equiv \frac{11M_2}{g^2}-\frac{1}{16\pi^2}\Big(M_1+209M_2-88M_3+\frac{22b}{\mu}\Big),
\]
whereas the more constrained dMSSM and pMSSM admit additional two-loop continuations for $M_1/g'^2$ and $M_3/g_s^2$ [1507.03470].

## 5. Exact all-loop invariance and scheme dependence

All-loop RGIs arise in special circumstances. In the rigid MSSM, exact all-loop invariants can be constructed from gauge couplings, Yukawa determinants, and the Higgs bilinear parameter $\mu$ by combining the NSVZ $\beta$-functions with the nonrenormalisation of the superpotential. The essential ingredients are
\[
\frac{d\ln\det Y_u}{d\ln p}
=
\mathrm{tr}\,\gamma_Q+\mathrm{tr}\,\gamma_U+\gamma_{H_u},
\]
\[
\frac{d\ln\det Y_d}{d\ln p}
=
\mathrm{tr}\,\gamma_Q+\mathrm{tr}\,\gamma_D+\gamma_{H_d},
\]
\[
\frac{d\ln\det Y_e}{d\ln p}
=
\mathrm{tr}\,\gamma_L+\mathrm{tr}\,\gamma_E+\gamma_{H_d},
\qquad
\frac{d\ln\mu}{d\ln p}
=
\frac12(\gamma_{H_u}+\gamma_{H_d}),
\]
together with the exact NSVZ equations for $\alpha_{1,2,3}$. Eliminating anomalous dimensions yields two independent all-loop RGIs. The paper emphasizes that these invariants hold in the HD+MSL scheme and that in the $\overline{\mathrm{DR}}$ scheme the renormalization group invariance does not take place starting from the approximation where the scheme dependence manifests itself [2410.10107].

A different exact statement appears in the Symmetry Improved 2PI formalism. For the $O(2)$ scalar model, the SI2PI effective potential $V(\phi)$ is proved to be exactly RG invariant in the Hartree–Fock and sunset truncations. The proof combines UV-finite running of the proper 2PI couplings with exact cancellation of the $\delta\ln\mu$ terms in the RG variation of the gap equations, yielding
\[
\delta\Delta_H^{-1}=\delta\Delta_G^{-1}=0
\]
and therefore
\[
\frac{d}{d\ln\mu}V_{\rm SI2PI}(\phi)=0.
\]
This is explicitly contrasted with ordinary 1PI perturbation theory, where the effective potential is RG invariant only up to higher-order terms at a given truncation [1703.02079].

These examples delimit the strongest notion of invariance in the subject. Exact all-loop RGIs typically rely on exact structural inputs: NSVZ relations and nonrenormalisation theorems in supersymmetry, or Ward-identity-based cancellations in SI2PI. Outside such settings, exact invariance is exceptional rather than generic.

## 6. Multi-scalar, gauge-mixing, and conceptual limits

Recent work generalizes RGIs beyond supersymmetry by exploiting the synergy of scaling and non-overlapping global symmetries in renormalisable multi-scalar theories. The central idea is to identify scale-invariant field directions along which the bilinear part of the scalar potential vanishes identically and to assign spurion charges under symmetries such as CP2 and $U(1)_{\rm PQ}$. Under these conditions, the dangerous terms in the $\beta$-functions are forbidden to all orders, so certain bilinear combinations become all-loop RGIs. In the two-scalar $U(1)$ example one obtains
\[
\beta_{m_1^2+m_2^2}=0,\qquad \beta_{\lambda_1-\lambda_2}=0
\]
on the scale-invariant surface $m_2^2=-m_1^2$, and in the 2HDM one obtains
\[
\beta_{m_{11}^2+m_{22}^2}=0,\qquad
\beta_{\lambda_1-\lambda_2}=0,\qquad
\beta_{\lambda_6+\lambda_7}=0
\]
to all loops under the CP2-symmetric conditions quoted in the paper [2605.18341].

The same literature stresses the limitations. If symmetries overlap or spurion assignments admit mixing terms, invariants are broken. CP3 invariants require the custodial limit $g'\to 0$; for $g'\neq 0$, two-loop hypercharge corrections break $\lambda_1=\lambda_2$. Threshold matching must preserve the spurion-charge structure if the invariant is to survive decoupling [2605.18341].

In models with two $U(1)$ gauge symmetries, the non-canonical kinetic-matrix formalism supplies a different class of invariants. The combinations
\[
I_{f_if_j}=\vec v_{f_i}^{\,T}[K]^{-1}\vec v_{f_j}
\]
are invariant under arbitrary $GL(2,\mathbb R)$ field reparametrisations, and the ratio
\[
I_\varepsilon(\mu)
=
\frac{\vec v_\chi^{\,T}[K(\mu)]^{-1}\vec v_\psi}{\vec v_\psi^{\,T}[K(\mu)]^{-1}\vec v_\psi}
\]
maps directly onto the effective millicharge. These quantities are observationally useful, but they are not RGIs in general, because $[K]$ obeys one- and two-loop RGEs and kinetic mixing is radiatively generated whenever bi-charged matter is present [2605.20058].

Several recurrent misconceptions are therefore corrected by the literature itself. Basis-independent quantities are not automatically scale independent. Exact one-loop invariance does not imply all-loop invariance. Reparametrisation invariants need not be RGIs. And in generic theories, nontrivial exact RGIs are scarce; outside special symmetry-protected settings, the dominant structures are one-loop exact relations, threshold-sensitive piecewise invariants, or approximate invariants controlled by hierarchies [1507.00328].

Source: https://www.emergentmind.com/topics/renormalisation-group-invariants-rgis