Optimized Perturbation Theory (OPT)

• Optimized Perturbation Theory (OPT) is a variational method that reorganizes standard perturbation theory by introducing an auxiliary parameter, which is fixed using optimization criteria like PMS or FAC.
• It employs an interpolation between a solvable zeroth-order problem and the full interacting theory, with implementations such as the linear delta expansion and RGOPT adapting to diverse physical systems.
• RG consistency in OPT reduces scale dependence and enhances predictive accuracy, making it effective in applications ranging from thermal field theory to NJL models and quantum chromodynamics.

Searching arXiv for the cited OPT and RGOPT papers to ground the article in current arXiv records. Optimized Perturbation Theory (OPT) is a family of variational reorganizations of perturbation theory in which an auxiliary parameter—most often a mass—is introduced into a solvable zeroth-order problem, while a bookkeeping parameter interpolates between that problem and the original interacting theory. The resulting truncated series depends on the auxiliary parameter, which is then fixed by an optimization prescription such as the Principle of Minimal Sensitivity (PMS) or the Fastest Apparent Convergence (FAC) condition. In modern usage, OPT encompasses several closely related constructions: the linear delta expansion, renormalization-group-consistent variants such as RGOPT, and domain-specific formulations in thermal field theory, NJL models, molecular fluids, and electronic many-body perturbation theory (Kneur et al., 2015, Tsutsui et al., 2019, Kneur et al., 2010).

1. Origins, scope, and terminological variants

OPT was formulated for strongly interacting statistical systems through trial Green functions with control functions determined by self-consistency conditions, so that the reorganized approximation sequence becomes convergent (Yukalov, 2019). In that formulation, the exact causal Green function $G$ and a model Green function $G_0$ are related by a nonlinear Dyson-type equation,

$G(12) = G_0(12) + \int K(13) G(32)\, d(3),$

with optimization imposed by requiring selected observable averages to remain unchanged between successive iterations (Yukalov, 2019). This construction already contains the characteristic OPT idea: perturbation theory is not expanded around a fixed free theory, but around a trial object carrying control parameters.

A later and more widely used field-theoretic formulation is the linear delta expansion. There one introduces a solvable quadratic Lagrangian $\mathcal{L}_0(\eta)$ and interpolates as

$$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$

then truncates in $\delta$ and sets $\delta \to 1$ at the end (Duarte et al., 2011, Kneur et al., 2012). In scalar theories and NJL-type models this is commonly written as a mass deformation, for example

$$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$

or, in fermionic models,

$$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$

with the auxiliary parameter fixed variationally after truncation (Kneur et al., 2015, Kneur et al., 2012).

The term “OPT” is not completely uniform across subfields. In thermal scalar and NJL applications it usually denotes a variational resummation with an explicit auxiliary mass and PMS (Kneur et al., 2015, Duarte et al., 2011, Kneur et al., 2012). In the theory of colloidal dumbbells, by contrast, “optimized perturbation theory” refers to a fourth-order high-temperature expansion of the Helmholtz free energy whose coefficients are “optimized phenomenological” fits to simulation data, without an explicit variational parameter or stationarity condition (Munaò et al., 2015). This suggests that OPT is best understood as a methodological family rather than a single formalism.

2. Linear delta expansion and optimization criteria

The central structural move in OPT is to introduce an interpolation between a solvable problem and the target theory. For a renormalized observable $O(m,\lambda)$, one convenient scalar-field implementation is

$G_0$0

followed by reexpansion in $G_0$1 to a chosen order and the limit $G_0$2 (Kneur et al., 2015). In the simplest linear delta expansion, one takes $G_0$3 or, more generally, treats $G_0$4 as part of the interpolation; in RGOPT it is fixed by RG constraints, as discussed below (Kneur et al., 2015, Kneur et al., 2010).

The most common optimization prescription is PMS. For a truncated quantity $G_0$5, PMS requires

$G_0$6

so that the approximation is locally insensitive to the arbitrary parameter (Duarte et al., 2011, Kneur et al., 2012). In finite-temperature applications, $G_0$7 is usually the effective potential or free energy; in zero-dimensional benchmark models, PMS can be imposed directly on the effective potential, self-energy, or 1PI four-point function (Rosa et al., 2016).

FAC provides an alternative criterion. For a truncated series $G_0$8, FAC sets the last known term to zero,

$G_0$9

thereby minimizing the apparent size of the highest retained contribution (Tsutsui et al., 2019). In the language of OPT integral representations, this becomes a condition on the zeros of the $G(12) = G_0(12) + \int K(13) G(32)\, d(3),$0th coefficient,

$G(12) = G_0(12) + \int K(13) G(32)\, d(3),$1

with $G(12) = G_0(12) + \int K(13) G(32)\, d(3),$2 defined by the $G(12) = G_0(12) + \int K(13) G(32)\, d(3),$3-expanded integral (Tsutsui et al., 2019). In solvable models, FAC and PMS are closely related asymptotically.

A third criterion sometimes used when PMS and FAC have no nontrivial solution is the turning-point condition,

$G(12) = G_0(12) + \int K(13) G(32)\, d(3),$4

which was examined explicitly in the zero-dimensional $G(12) = G_0(12) + \int K(13) G(32)\, d(3),$5 scalar model (Rosa et al., 2016). That study found that the PMS procedure tends to produce better results, particularly when applied directly to the self-energy, while ordinary perturbation theory has zero radius of convergence and the large-$G(12) = G_0(12) + \int K(13) G(32)\, d(3),$6 expansion converges more slowly at finite $G(12) = G_0(12) + \int K(13) G(32)\, d(3),$7 (Rosa et al., 2016).

The same logic has been adapted outside relativistic field theory. In “Perturbation-Adapted Perturbation Theory” (Knowles, 2021), the zeroth-order Hamiltonian is chosen to be a best fit to the exact Hamiltonian within a specified functional form by enforcing projected similarity conditions on the first-order interacting space. This is not standard OPT in the PMS sense, but it is closely related in spirit: the zeroth-order problem is optimized rather than fixed a priori (Knowles, 2021).

3. Renormalization-group consistency and RGOPT

A recurrent limitation of ordinary OPT is residual renormalization-scale dependence. RGOPT addresses this by requiring the $G(12) = G_0(12) + \int K(13) G(32)\, d(3),$8-deformed truncated series to satisfy RG constraints order by order. For a physical quantity $G(12) = G_0(12) + \int K(13) G(32)\, d(3),$9, the RG operator is

$\mathcal{L}_0(\eta)$0

with

$\mathcal{L}_0(\eta)$1

and RG invariance requires $\mathcal{L}_0(\eta)$2 (Kneur et al., 2015). When PMS eliminates the explicit $\mathcal{L}_0(\eta)$3 derivative, the RG condition reduces to

$\mathcal{L}_0(\eta)$4

which, together with PMS, fixes both the optimized mass and the coupling at a given order (Kneur et al., 2015).

The defining RGOPT modification is that the interpolation exponent is not arbitrary. Requiring RG consistency of the $\mathcal{L}_0(\eta)$5-modified series fixes

$\mathcal{L}_0(\eta)$6

in the scalar thermal formulation of RGOPT (Kneur et al., 2015), and

$\mathcal{L}_0(\eta)$7

in the Gross–Neveu and low-energy QCD RGOPT constructions (Kneur et al., 2010, Kneur et al., 2013). For the $\mathcal{L}_0(\eta)$8 $\mathcal{L}_0(\eta)$9 theory in the $$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$0 scheme,

$$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$1

so that $$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$2 (Kneur et al., 2015). For the $$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$3 model,

$$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$4

hence

$$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$5

which explains the exact recovery of the large-$$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$6 limit at one loop in that setting (Kneur et al., 2015).

In massive theories, RG consistency also requires finite vacuum-energy subtraction terms. For scalar thermodynamics, RGOPT introduces

$$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$7

with coefficients fixed perturbatively by demanding RG invariance of the truncated free energy; for the $$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$8 scalar theory,

$$\mathcal{L}_\delta = \mathcal{L}_0(\eta) + \delta \big[\mathcal{L} - \mathcal{L}_0(\eta)\big],$$9

up to two loops (Kneur et al., 2015). These terms are temperature-independent but necessary for perturbative RG invariance.

The power of RGOPT is most transparent in exact or nearly exact benchmarks. In the $\delta$0 Gross–Neveu model, first-order RGOPT reproduces the exact large-$\delta$1 mass gap, and second-order results achieve controllable percent accuracy or better for arbitrary $\delta$2 when compared with the exact thermodynamic Bethe Ansatz (Kneur et al., 2010). In low-energy QCD, RGOPT has been used to determine $\delta$3 and the QCD coupling, yielding

$\delta$4

from the pion-decay-constant analysis (Kneur et al., 2013). Applied to the spectral density of the Dirac operator and the chiral condensate, it gives

$\delta$5

and

$\delta$6

with a moderate suppression from $\delta$7 to $\delta$8 (Kneur et al., 2015).

4. Thermal field theory and external backgrounds

Finite-temperature scalar theory is a canonical OPT testing ground because ordinary perturbation theory converges poorly and exhibits strong renormalization-scale dependence. In the thermal $\delta$9 model, RGOPT gives an exactly scale-invariant one-loop pressure when the exact one-loop running of $\delta \to 1$0 is used, and the corresponding thermal mass gap is determined by the self-consistent equation

$\delta \to 1$1

with

$\delta \to 1$2

(Kneur et al., 2015). In the high-temperature approximation, this leads to

$\delta \to 1$3

and the one-loop pressure normalized to the free-gas value $\delta \to 1$4 becomes

$\delta \to 1$5

(Kneur et al., 2015).

At two loops, RGOPT maintains much smaller scale variation than standard perturbation theory and screened perturbation theory. For $\delta \to 1$6 and $\delta \to 1$7 varied from $\delta \to 1$8 to $\delta \to 1$9, the reported scale variation is about $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$0 in standard perturbation theory, about $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$1 in two-loop SPT with optimized mass, about $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$2 in two-loop SPT with screening mass, and about $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$3 in two-loop RGOPT, reduced to about $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$4 when the $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$5 vacuum subtraction is included (Kneur et al., 2015). The residual scale dependence reappears only at $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$6 rather than $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$7, which is identified as a consequence of RG invariance (Kneur et al., 2015).

A related but distinct development is the “Variational Renormalization Group” framework, which combines RG improvement of the thermal effective potential with standard OPT rather than changing the interpolation exponent. In the quartic scalar theory at finite temperature, this approach significantly improves the scale stability of the effective potential, critical temperature, and pressure compared with OPT alone (Câmara et al., 8 Sep 2025). Since that work lies outside the date range of the main thermal RGOPT paper and introduces a new named framework, it is best regarded as an extension rather than a replacement.

External-field problems show how OPT handles nontrivial backgrounds. For a self-interacting charged scalar field in a constant magnetic field and at finite temperature, first-order OPT yields a renormalized effective potential in which the variational mass satisfies

$$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$8

and the transition remains second order for all magnetic-field strengths, with the critical temperature increasing with $$m \to m(1-\delta)^a,\qquad \lambda \to \delta \lambda,$$9 (Duarte et al., 2011). The paper also introduces an efficient Euler–Maclaurin treatment of the Landau-level sum in the weak-field regime (Duarte et al., 2011).

For a charged scalar under a constant electric field and finite temperature, first-order OPT combined with the Schwinger proper-time method gives explicit real and imaginary parts of the effective potential (Tavares et al., 2024). In that model, the vacuum expectation value decreases weakly with increasing electric field, a first-order transition at zero or weak field becomes second order at sufficiently strong field, and the critical temperature depends only very weakly on the electric field (Tavares et al., 2024). The vacuum persistence probability rate,

$$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$0

peaks at the critical point, with

$$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$1

independent of the coupling constant (Tavares et al., 2024).

5. Applications in fermionic, hadronic, condensed-matter, and molecular systems

OPT has been extensively used in NJL-type models to incorporate corrections beyond the large-$$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$2 or mean-field approximation. In the Abelian NJL model at $$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$3 and finite chemical potential, the first nontrivial OPT corrections are two-loop contributions that generate a $$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$4-suppressed term proportional to the square of the density integral,

$$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$5

in the free-energy density (Kneur et al., 2012). This contribution vanishes at $$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$6 but survives at finite density, where it mimics a repulsive vector–vector interaction, shifts chiral restoration to higher chemical potential, and stiffens the equation of state (Kneur et al., 2012). Near the transition, the effective vector coupling that reproduces the OPT result is numerically close to the Fierz-induced Hartree–Fock value $$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$7 (Kneur et al., 2012).

In the two-flavor, three-color NJL model with diquark interactions, OPT at order $$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$8 incorporates beyond-large-$$m_c \to m_c + (1-\delta)\eta,\qquad G \to \delta G,$$9 corrections to the thermodynamic potential and modifies the BEC–BCS crossover region (Duarte et al., 2017). The formalism introduces several variational mass-like parameters and fixes them by PMS. Relative to the large-$O(m,\lambda)$0 approximation, OPT lowers the diquark condensate, narrows the BEC window, shifts critical points, and, when color neutrality is imposed, can change the nature of the diquark transition from second to first order except very near the upper stability bound of the diquark coupling (Duarte et al., 2017).

In the flavor-symmetric $O(m,\lambda)$1 NJL model, OPT reproduces Hartree–Fock results and shows that vector-like contributions dynamically generated beyond the large-$O(m,\lambda)$2 approximation can cancel when three flavors and the ’t Hooft interaction are included (Macías et al., 2018). This is used to revisit the discrepancy between two-flavor model predictions and lattice data for dynamical vector repulsion above the pseudocritical temperature (Macías et al., 2018). In the Polyakov linear-sigma model, first-order OPT modifies the fermion sector through a variational mass and improves agreement with lattice QCD for higher-order cumulants of conserved charges as compared with mean field, while leaving lower-order thermodynamics comparatively close to MFA (Tawfik et al., 2019).

Outside quantum field theory in the narrow sense, OPT has been used as a semi-empirical high-temperature theory of molecular fluids. For anisotropic colloidal dumbbells with square-well attractions, the Helmholtz free energy is written as a fourth-order series in the attractive strength,

$O(m,\lambda)$3

with the coefficients $O(m,\lambda)$4 taken from optimized fits to atomic square-well fluids (Munaò et al., 2015). This formulation predicts the progressive reduction of gas–liquid phase separation with increasing anisotropy and a linear dependence of the critical temperature on interaction strength, but it generally overestimates the coexistence region and performs less well than RISM, especially in the strong-anisotropy regime (Munaò et al., 2015).

6. Mathematical structure, reliability, and common misconceptions

A major mathematical question in OPT is why truncated approximants can become insensitive to arbitrary variational parameters. In the FAC framework, the $O(m,\lambda)$5th coefficient of the $O(m,\lambda)$6-expanded integral admits a steepest-descent representation,

$O(m,\lambda)$7

with

$O(m,\lambda)$8

and Lefschetz-thimble analysis shows that the zeros of $O(m,\lambda)$9 accumulate on anti-Stokes lines as $G_0$00 (Tsutsui et al., 2019). This explains the emergence of broad plateaus in the optimized approximants and the practical insensitivity of OPT predictions to the choice of the artificial parameter (Tsutsui et al., 2019). PMS solutions lie near the same anti-Stokes structures asymptotically, which clarifies the close relationship between FAC and PMS in many problems (Tsutsui et al., 2019).

The zero-dimensional $G_0$01 scalar model provides an exactly solvable benchmark. There, OPT is implemented by the substitutions

$G_0$02

in exact expressions for the partition function, effective potential, self-energy, and four-point function, followed by PMS, FAC, or turning-point optimization (Rosa et al., 2016). The resulting OPT series are stable even at large couplings and outperform both ordinary perturbation theory and the $G_0$03 expansion; in particular, PMS applied directly to the self-energy tends to give the best results (Rosa et al., 2016). This suggests that OPT’s practical success is not limited to asymptotic weak-coupling regimes.

A common misconception is that OPT always means “introduce a mass and apply PMS.” The literature summarized here shows otherwise. The original strong-interaction Green-function formulation is based on control functions in trial propagators (Yukalov, 2019). Some condensed-matter and liquid-state uses of the term refer instead to optimized coefficients in truncated high-temperature expansions (Munaò et al., 2015). RGOPT further modifies the interpolation exponent and adds finite vacuum-energy subtractions to restore RG invariance (Kneur et al., 2015, Kneur et al., 2010). Another misconception is that OPT is intrinsically unrelated to RG ideas; the thermal scalar, Gross–Neveu, and low-energy QCD results show that enforcing RG consistency can be decisive for physical scale stability and for selecting asymptotically free solutions (Kneur et al., 2015, Kneur et al., 2010, Kneur et al., 2013).

Within QCD perturbation theory itself, PMS can also be formulated as optimization over renormalization schemes. In the Banks–Zaks limit, the “optimal” renormalization scheme determined by PMS is generally irregular in the expansion parameter $G_0$04, yet all-orders OPT reproduces the same Banks–Zaks expansion coefficients for physical infrared quantities as regular schemes (Stevenson, 2016). The fixed-point optimization conditions can be summarized by the master relation

$G_0$05

which makes arbitrarily high-order exploration possible in that limit (Stevenson, 2016). This suggests that OPT is as much a framework for optimizing perturbative organization as it is a particular resummation trick.

Taken together, these developments present OPT as a broad variational strategy for reorganizing perturbation theory, extracting nonperturbative information from truncated series, and, in its RG-consistent forms, restoring scale stability that is otherwise lost in naive resummations. Its success depends on the chosen interpolation, the optimization criterion, and the physical constraints imposed on admissible solutions.